xref: /freebsd/sys/kern/imgact_elf.c (revision 78f57a9cdef729f94669fe1eb494666f403f9235)
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 
40 #include <sys/param.h>
41 #include <sys/capsicum.h>
42 #include <sys/compressor.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/jail.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mman.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
55 #include <sys/proc.h>
56 #include <sys/procfs.h>
57 #include <sys/ptrace.h>
58 #include <sys/racct.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
61 #include <sys/sbuf.h>
62 #include <sys/sf_buf.h>
63 #include <sys/smp.h>
64 #include <sys/systm.h>
65 #include <sys/signalvar.h>
66 #include <sys/stat.h>
67 #include <sys/sx.h>
68 #include <sys/syscall.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/vnode.h>
72 #include <sys/syslog.h>
73 #include <sys/eventhandler.h>
74 #include <sys/user.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_extern.h>
83 
84 #include <machine/elf.h>
85 #include <machine/md_var.h>
86 
87 #define ELF_NOTE_ROUNDSIZE	4
88 #define OLD_EI_BRAND	8
89 
90 static int __elfN(check_header)(const Elf_Ehdr *hdr);
91 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
92     const char *interp, int interp_name_len, int32_t *osrel);
93 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94     u_long *entry, size_t pagesize);
95 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
96     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
97     size_t pagesize);
98 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
99 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
100     int32_t *osrel);
101 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
102 static boolean_t __elfN(check_note)(struct image_params *imgp,
103     Elf_Brandnote *checknote, int32_t *osrel);
104 static vm_prot_t __elfN(trans_prot)(Elf_Word);
105 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
106 
107 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108     "");
109 
110 #define	CORE_BUF_SIZE	(16 * 1024)
111 
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119     &elf_legacy_coredump, 0,
120     "include all and only RW pages in core dumps");
121 
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
124     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
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 boolean_t
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 boolean_t
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 ||
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)
373 				return (bi);
374 		}
375 	}
376 
377 	/* Lacking a recognized interpreter, try the default brand */
378 	for (i = 0; i < MAX_BRANDS; i++) {
379 		bi = elf_brand_list[i];
380 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
381 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
382 			continue;
383 		if (hdr->e_machine == bi->machine &&
384 		    __elfN(fallback_brand) == bi->brand)
385 			return (bi);
386 	}
387 	return (NULL);
388 }
389 
390 static int
391 __elfN(check_header)(const Elf_Ehdr *hdr)
392 {
393 	Elf_Brandinfo *bi;
394 	int i;
395 
396 	if (!IS_ELF(*hdr) ||
397 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
398 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
399 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
400 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
401 	    hdr->e_version != ELF_TARG_VER)
402 		return (ENOEXEC);
403 
404 	/*
405 	 * Make sure we have at least one brand for this machine.
406 	 */
407 
408 	for (i = 0; i < MAX_BRANDS; i++) {
409 		bi = elf_brand_list[i];
410 		if (bi != NULL && bi->machine == hdr->e_machine)
411 			break;
412 	}
413 	if (i == MAX_BRANDS)
414 		return (ENOEXEC);
415 
416 	return (0);
417 }
418 
419 static int
420 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
421     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
422 {
423 	struct sf_buf *sf;
424 	int error;
425 	vm_offset_t off;
426 
427 	/*
428 	 * Create the page if it doesn't exist yet. Ignore errors.
429 	 */
430 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
431 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
432 
433 	/*
434 	 * Find the page from the underlying object.
435 	 */
436 	if (object != NULL) {
437 		sf = vm_imgact_map_page(object, offset);
438 		if (sf == NULL)
439 			return (KERN_FAILURE);
440 		off = offset - trunc_page(offset);
441 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
442 		    end - start);
443 		vm_imgact_unmap_page(sf);
444 		if (error != 0)
445 			return (KERN_FAILURE);
446 	}
447 
448 	return (KERN_SUCCESS);
449 }
450 
451 static int
452 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
453     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
454     int cow)
455 {
456 	struct sf_buf *sf;
457 	vm_offset_t off;
458 	vm_size_t sz;
459 	int error, locked, rv;
460 
461 	if (start != trunc_page(start)) {
462 		rv = __elfN(map_partial)(map, object, offset, start,
463 		    round_page(start), prot);
464 		if (rv != KERN_SUCCESS)
465 			return (rv);
466 		offset += round_page(start) - start;
467 		start = round_page(start);
468 	}
469 	if (end != round_page(end)) {
470 		rv = __elfN(map_partial)(map, object, offset +
471 		    trunc_page(end) - start, trunc_page(end), end, prot);
472 		if (rv != KERN_SUCCESS)
473 			return (rv);
474 		end = trunc_page(end);
475 	}
476 	if (start >= end)
477 		return (KERN_SUCCESS);
478 	if ((offset & PAGE_MASK) != 0) {
479 		/*
480 		 * The mapping is not page aligned.  This means that we have
481 		 * to copy the data.
482 		 */
483 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
484 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
485 		if (rv != KERN_SUCCESS)
486 			return (rv);
487 		if (object == NULL)
488 			return (KERN_SUCCESS);
489 		for (; start < end; start += sz) {
490 			sf = vm_imgact_map_page(object, offset);
491 			if (sf == NULL)
492 				return (KERN_FAILURE);
493 			off = offset - trunc_page(offset);
494 			sz = end - start;
495 			if (sz > PAGE_SIZE - off)
496 				sz = PAGE_SIZE - off;
497 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
498 			    (caddr_t)start, sz);
499 			vm_imgact_unmap_page(sf);
500 			if (error != 0)
501 				return (KERN_FAILURE);
502 			offset += sz;
503 		}
504 	} else {
505 		vm_object_reference(object);
506 		rv = vm_map_fixed(map, object, offset, start, end - start,
507 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
508 		if (rv != KERN_SUCCESS) {
509 			locked = VOP_ISLOCKED(imgp->vp);
510 			VOP_UNLOCK(imgp->vp, 0);
511 			vm_object_deallocate(object);
512 			vn_lock(imgp->vp, locked | LK_RETRY);
513 			return (rv);
514 		}
515 	}
516 	return (KERN_SUCCESS);
517 }
518 
519 static int
520 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
521     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
522     size_t pagesize)
523 {
524 	struct sf_buf *sf;
525 	size_t map_len;
526 	vm_map_t map;
527 	vm_object_t object;
528 	vm_offset_t off, map_addr;
529 	int error, rv, cow;
530 	size_t copy_len;
531 	vm_ooffset_t file_addr;
532 
533 	/*
534 	 * It's necessary to fail if the filsz + offset taken from the
535 	 * header is greater than the actual file pager object's size.
536 	 * If we were to allow this, then the vm_map_find() below would
537 	 * walk right off the end of the file object and into the ether.
538 	 *
539 	 * While I'm here, might as well check for something else that
540 	 * is invalid: filsz cannot be greater than memsz.
541 	 */
542 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
543 	    filsz > memsz) {
544 		uprintf("elf_load_section: truncated ELF file\n");
545 		return (ENOEXEC);
546 	}
547 
548 	object = imgp->object;
549 	map = &imgp->proc->p_vmspace->vm_map;
550 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
551 	file_addr = trunc_page_ps(offset, pagesize);
552 
553 	/*
554 	 * We have two choices.  We can either clear the data in the last page
555 	 * of an oversized mapping, or we can start the anon mapping a page
556 	 * early and copy the initialized data into that first page.  We
557 	 * choose the second.
558 	 */
559 	if (filsz == 0)
560 		map_len = 0;
561 	else if (memsz > filsz)
562 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
563 	else
564 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
565 
566 	if (map_len != 0) {
567 		/* cow flags: don't dump readonly sections in core */
568 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
569 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
570 
571 		rv = __elfN(map_insert)(imgp, map,
572 				      object,
573 				      file_addr,	/* file offset */
574 				      map_addr,		/* virtual start */
575 				      map_addr + map_len,/* virtual end */
576 				      prot,
577 				      cow);
578 		if (rv != KERN_SUCCESS)
579 			return (EINVAL);
580 
581 		/* we can stop now if we've covered it all */
582 		if (memsz == filsz)
583 			return (0);
584 	}
585 
586 
587 	/*
588 	 * We have to get the remaining bit of the file into the first part
589 	 * of the oversized map segment.  This is normally because the .data
590 	 * segment in the file is extended to provide bss.  It's a neat idea
591 	 * to try and save a page, but it's a pain in the behind to implement.
592 	 */
593 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
594 	    filsz, pagesize);
595 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
596 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
597 	    map_addr;
598 
599 	/* This had damn well better be true! */
600 	if (map_len != 0) {
601 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
602 		    map_addr + map_len, prot, 0);
603 		if (rv != KERN_SUCCESS)
604 			return (EINVAL);
605 	}
606 
607 	if (copy_len != 0) {
608 		sf = vm_imgact_map_page(object, offset + filsz);
609 		if (sf == NULL)
610 			return (EIO);
611 
612 		/* send the page fragment to user space */
613 		off = trunc_page_ps(offset + filsz, pagesize) -
614 		    trunc_page(offset + filsz);
615 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
616 		    (caddr_t)map_addr, copy_len);
617 		vm_imgact_unmap_page(sf);
618 		if (error != 0)
619 			return (error);
620 	}
621 
622 	/*
623 	 * Remove write access to the page if it was only granted by map_insert
624 	 * to allow copyout.
625 	 */
626 	if ((prot & VM_PROT_WRITE) == 0)
627 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
628 		    map_len), prot, FALSE);
629 
630 	return (0);
631 }
632 
633 /*
634  * Load the file "file" into memory.  It may be either a shared object
635  * or an executable.
636  *
637  * The "addr" reference parameter is in/out.  On entry, it specifies
638  * the address where a shared object should be loaded.  If the file is
639  * an executable, this value is ignored.  On exit, "addr" specifies
640  * where the file was actually loaded.
641  *
642  * The "entry" reference parameter is out only.  On exit, it specifies
643  * the entry point for the loaded file.
644  */
645 static int
646 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
647 	u_long *entry, size_t pagesize)
648 {
649 	struct {
650 		struct nameidata nd;
651 		struct vattr attr;
652 		struct image_params image_params;
653 	} *tempdata;
654 	const Elf_Ehdr *hdr = NULL;
655 	const Elf_Phdr *phdr = NULL;
656 	struct nameidata *nd;
657 	struct vattr *attr;
658 	struct image_params *imgp;
659 	vm_prot_t prot;
660 	u_long rbase;
661 	u_long base_addr = 0;
662 	int error, i, numsegs;
663 
664 #ifdef CAPABILITY_MODE
665 	/*
666 	 * XXXJA: This check can go away once we are sufficiently confident
667 	 * that the checks in namei() are correct.
668 	 */
669 	if (IN_CAPABILITY_MODE(curthread))
670 		return (ECAPMODE);
671 #endif
672 
673 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
674 	nd = &tempdata->nd;
675 	attr = &tempdata->attr;
676 	imgp = &tempdata->image_params;
677 
678 	/*
679 	 * Initialize part of the common data
680 	 */
681 	imgp->proc = p;
682 	imgp->attr = attr;
683 	imgp->firstpage = NULL;
684 	imgp->image_header = NULL;
685 	imgp->object = NULL;
686 	imgp->execlabel = NULL;
687 
688 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
689 	if ((error = namei(nd)) != 0) {
690 		nd->ni_vp = NULL;
691 		goto fail;
692 	}
693 	NDFREE(nd, NDF_ONLY_PNBUF);
694 	imgp->vp = nd->ni_vp;
695 
696 	/*
697 	 * Check permissions, modes, uid, etc on the file, and "open" it.
698 	 */
699 	error = exec_check_permissions(imgp);
700 	if (error)
701 		goto fail;
702 
703 	error = exec_map_first_page(imgp);
704 	if (error)
705 		goto fail;
706 
707 	/*
708 	 * Also make certain that the interpreter stays the same, so set
709 	 * its VV_TEXT flag, too.
710 	 */
711 	VOP_SET_TEXT(nd->ni_vp);
712 
713 	imgp->object = nd->ni_vp->v_object;
714 
715 	hdr = (const Elf_Ehdr *)imgp->image_header;
716 	if ((error = __elfN(check_header)(hdr)) != 0)
717 		goto fail;
718 	if (hdr->e_type == ET_DYN)
719 		rbase = *addr;
720 	else if (hdr->e_type == ET_EXEC)
721 		rbase = 0;
722 	else {
723 		error = ENOEXEC;
724 		goto fail;
725 	}
726 
727 	/* Only support headers that fit within first page for now      */
728 	if ((hdr->e_phoff > PAGE_SIZE) ||
729 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
730 		error = ENOEXEC;
731 		goto fail;
732 	}
733 
734 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
735 	if (!aligned(phdr, Elf_Addr)) {
736 		error = ENOEXEC;
737 		goto fail;
738 	}
739 
740 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
741 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
742 			/* Loadable segment */
743 			prot = __elfN(trans_prot)(phdr[i].p_flags);
744 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
745 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
746 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
747 			if (error != 0)
748 				goto fail;
749 			/*
750 			 * Establish the base address if this is the
751 			 * first segment.
752 			 */
753 			if (numsegs == 0)
754   				base_addr = trunc_page(phdr[i].p_vaddr +
755 				    rbase);
756 			numsegs++;
757 		}
758 	}
759 	*addr = base_addr;
760 	*entry = (unsigned long)hdr->e_entry + rbase;
761 
762 fail:
763 	if (imgp->firstpage)
764 		exec_unmap_first_page(imgp);
765 
766 	if (nd->ni_vp)
767 		vput(nd->ni_vp);
768 
769 	free(tempdata, M_TEMP);
770 
771 	return (error);
772 }
773 
774 static int
775 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
776 {
777 	struct thread *td;
778 	const Elf_Ehdr *hdr;
779 	const Elf_Phdr *phdr;
780 	Elf_Auxargs *elf_auxargs;
781 	struct vmspace *vmspace;
782 	const char *err_str, *newinterp;
783 	char *interp, *interp_buf, *path;
784 	Elf_Brandinfo *brand_info;
785 	struct sysentvec *sv;
786 	vm_prot_t prot;
787 	u_long text_size, data_size, total_size, text_addr, data_addr;
788 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
789 	int32_t osrel;
790 	int error, i, n, interp_name_len, have_interp;
791 
792 	hdr = (const Elf_Ehdr *)imgp->image_header;
793 
794 	/*
795 	 * Do we have a valid ELF header ?
796 	 *
797 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
798 	 * if particular brand doesn't support it.
799 	 */
800 	if (__elfN(check_header)(hdr) != 0 ||
801 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
802 		return (-1);
803 
804 	/*
805 	 * From here on down, we return an errno, not -1, as we've
806 	 * detected an ELF file.
807 	 */
808 
809 	if ((hdr->e_phoff > PAGE_SIZE) ||
810 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
811 		/* Only support headers in first page for now */
812 		uprintf("Program headers not in the first page\n");
813 		return (ENOEXEC);
814 	}
815 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
816 	if (!aligned(phdr, Elf_Addr)) {
817 		uprintf("Unaligned program headers\n");
818 		return (ENOEXEC);
819 	}
820 
821 	n = error = 0;
822 	baddr = 0;
823 	osrel = 0;
824 	text_size = data_size = total_size = text_addr = data_addr = 0;
825 	entry = proghdr = 0;
826 	interp_name_len = 0;
827 	err_str = newinterp = NULL;
828 	interp = interp_buf = NULL;
829 	td = curthread;
830 
831 	for (i = 0; i < hdr->e_phnum; i++) {
832 		switch (phdr[i].p_type) {
833 		case PT_LOAD:
834 			if (n == 0)
835 				baddr = phdr[i].p_vaddr;
836 			n++;
837 			break;
838 		case PT_INTERP:
839 			/* Path to interpreter */
840 			if (phdr[i].p_filesz > MAXPATHLEN) {
841 				uprintf("Invalid PT_INTERP\n");
842 				error = ENOEXEC;
843 				goto ret;
844 			}
845 			if (interp != NULL) {
846 				uprintf("Multiple PT_INTERP headers\n");
847 				error = ENOEXEC;
848 				goto ret;
849 			}
850 			interp_name_len = phdr[i].p_filesz;
851 			if (phdr[i].p_offset > PAGE_SIZE ||
852 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
853 				VOP_UNLOCK(imgp->vp, 0);
854 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
855 				    M_WAITOK);
856 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
857 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
858 				    interp_name_len, phdr[i].p_offset,
859 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
860 				    NOCRED, NULL, td);
861 				if (error != 0) {
862 					uprintf("i/o error PT_INTERP\n");
863 					goto ret;
864 				}
865 				interp_buf[interp_name_len] = '\0';
866 				interp = interp_buf;
867 			} else {
868 				interp = __DECONST(char *, imgp->image_header) +
869 				    phdr[i].p_offset;
870 			}
871 			break;
872 		case PT_GNU_STACK:
873 			if (__elfN(nxstack))
874 				imgp->stack_prot =
875 				    __elfN(trans_prot)(phdr[i].p_flags);
876 			imgp->stack_sz = phdr[i].p_memsz;
877 			break;
878 		}
879 	}
880 
881 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
882 	    &osrel);
883 	if (brand_info == NULL) {
884 		uprintf("ELF binary type \"%u\" not known.\n",
885 		    hdr->e_ident[EI_OSABI]);
886 		error = ENOEXEC;
887 		goto ret;
888 	}
889 	et_dyn_addr = 0;
890 	if (hdr->e_type == ET_DYN) {
891 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
892 			uprintf("Cannot execute shared object\n");
893 			error = ENOEXEC;
894 			goto ret;
895 		}
896 		/*
897 		 * Honour the base load address from the dso if it is
898 		 * non-zero for some reason.
899 		 */
900 		if (baddr == 0)
901 			et_dyn_addr = ET_DYN_LOAD_ADDR;
902 	}
903 	sv = brand_info->sysvec;
904 	if (interp != NULL && brand_info->interp_newpath != NULL)
905 		newinterp = brand_info->interp_newpath;
906 
907 	/*
908 	 * Avoid a possible deadlock if the current address space is destroyed
909 	 * and that address space maps the locked vnode.  In the common case,
910 	 * the locked vnode's v_usecount is decremented but remains greater
911 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
912 	 * However, in cases where the vnode lock is external, such as nullfs,
913 	 * v_usecount may become zero.
914 	 *
915 	 * The VV_TEXT flag prevents modifications to the executable while
916 	 * the vnode is unlocked.
917 	 */
918 	VOP_UNLOCK(imgp->vp, 0);
919 
920 	error = exec_new_vmspace(imgp, sv);
921 	imgp->proc->p_sysent = sv;
922 
923 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
924 	if (error != 0)
925 		goto ret;
926 
927 	for (i = 0; i < hdr->e_phnum; i++) {
928 		switch (phdr[i].p_type) {
929 		case PT_LOAD:	/* Loadable segment */
930 			if (phdr[i].p_memsz == 0)
931 				break;
932 			prot = __elfN(trans_prot)(phdr[i].p_flags);
933 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
934 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
935 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
936 			    sv->sv_pagesize);
937 			if (error != 0)
938 				goto ret;
939 
940 			/*
941 			 * If this segment contains the program headers,
942 			 * remember their virtual address for the AT_PHDR
943 			 * aux entry. Static binaries don't usually include
944 			 * a PT_PHDR entry.
945 			 */
946 			if (phdr[i].p_offset == 0 &&
947 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
948 				<= phdr[i].p_filesz)
949 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
950 				    et_dyn_addr;
951 
952 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
953 			seg_size = round_page(phdr[i].p_memsz +
954 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
955 
956 			/*
957 			 * Make the largest executable segment the official
958 			 * text segment and all others data.
959 			 *
960 			 * Note that obreak() assumes that data_addr +
961 			 * data_size == end of data load area, and the ELF
962 			 * file format expects segments to be sorted by
963 			 * address.  If multiple data segments exist, the
964 			 * last one will be used.
965 			 */
966 
967 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
968 				text_size = seg_size;
969 				text_addr = seg_addr;
970 			} else {
971 				data_size = seg_size;
972 				data_addr = seg_addr;
973 			}
974 			total_size += seg_size;
975 			break;
976 		case PT_PHDR: 	/* Program header table info */
977 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
978 			break;
979 		default:
980 			break;
981 		}
982 	}
983 
984 	if (data_addr == 0 && data_size == 0) {
985 		data_addr = text_addr;
986 		data_size = text_size;
987 	}
988 
989 	entry = (u_long)hdr->e_entry + et_dyn_addr;
990 
991 	/*
992 	 * Check limits.  It should be safe to check the
993 	 * limits after loading the segments since we do
994 	 * not actually fault in all the segments pages.
995 	 */
996 	PROC_LOCK(imgp->proc);
997 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
998 		err_str = "Data segment size exceeds process limit";
999 	else if (text_size > maxtsiz)
1000 		err_str = "Text segment size exceeds system limit";
1001 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1002 		err_str = "Total segment size exceeds process limit";
1003 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1004 		err_str = "Data segment size exceeds resource limit";
1005 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1006 		err_str = "Total segment size exceeds resource limit";
1007 	if (err_str != NULL) {
1008 		PROC_UNLOCK(imgp->proc);
1009 		uprintf("%s\n", err_str);
1010 		error = ENOMEM;
1011 		goto ret;
1012 	}
1013 
1014 	vmspace = imgp->proc->p_vmspace;
1015 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1016 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1017 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1018 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1019 
1020 	/*
1021 	 * We load the dynamic linker where a userland call
1022 	 * to mmap(0, ...) would put it.  The rationale behind this
1023 	 * calculation is that it leaves room for the heap to grow to
1024 	 * its maximum allowed size.
1025 	 */
1026 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1027 	    RLIMIT_DATA));
1028 	PROC_UNLOCK(imgp->proc);
1029 
1030 	imgp->entry_addr = entry;
1031 
1032 	if (interp != NULL) {
1033 		have_interp = FALSE;
1034 		VOP_UNLOCK(imgp->vp, 0);
1035 		if (brand_info->emul_path != NULL &&
1036 		    brand_info->emul_path[0] != '\0') {
1037 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1038 			snprintf(path, MAXPATHLEN, "%s%s",
1039 			    brand_info->emul_path, interp);
1040 			error = __elfN(load_file)(imgp->proc, path, &addr,
1041 			    &imgp->entry_addr, sv->sv_pagesize);
1042 			free(path, M_TEMP);
1043 			if (error == 0)
1044 				have_interp = TRUE;
1045 		}
1046 		if (!have_interp && newinterp != NULL &&
1047 		    (brand_info->interp_path == NULL ||
1048 		    strcmp(interp, brand_info->interp_path) == 0)) {
1049 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1050 			    &imgp->entry_addr, sv->sv_pagesize);
1051 			if (error == 0)
1052 				have_interp = TRUE;
1053 		}
1054 		if (!have_interp) {
1055 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1056 			    &imgp->entry_addr, sv->sv_pagesize);
1057 		}
1058 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1059 		if (error != 0) {
1060 			uprintf("ELF interpreter %s not found, error %d\n",
1061 			    interp, error);
1062 			goto ret;
1063 		}
1064 	} else
1065 		addr = et_dyn_addr;
1066 
1067 	/*
1068 	 * Construct auxargs table (used by the fixup routine)
1069 	 */
1070 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1071 	elf_auxargs->execfd = -1;
1072 	elf_auxargs->phdr = proghdr;
1073 	elf_auxargs->phent = hdr->e_phentsize;
1074 	elf_auxargs->phnum = hdr->e_phnum;
1075 	elf_auxargs->pagesz = PAGE_SIZE;
1076 	elf_auxargs->base = addr;
1077 	elf_auxargs->flags = 0;
1078 	elf_auxargs->entry = entry;
1079 	elf_auxargs->hdr_eflags = hdr->e_flags;
1080 
1081 	imgp->auxargs = elf_auxargs;
1082 	imgp->interpreted = 0;
1083 	imgp->reloc_base = addr;
1084 	imgp->proc->p_osrel = osrel;
1085 	imgp->proc->p_elf_machine = hdr->e_machine;
1086 	imgp->proc->p_elf_flags = hdr->e_flags;
1087 
1088 ret:
1089 	free(interp_buf, M_TEMP);
1090 	return (error);
1091 }
1092 
1093 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1094 
1095 int
1096 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1097 {
1098 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1099 	Elf_Addr *base;
1100 	Elf_Addr *pos;
1101 
1102 	base = (Elf_Addr *)*stack_base;
1103 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1104 
1105 	if (args->execfd != -1)
1106 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1107 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1108 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1109 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1110 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1111 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1112 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1113 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1114 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1115 	if (imgp->execpathp != 0)
1116 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1117 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1118 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1119 	if (imgp->canary != 0) {
1120 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1121 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1122 	}
1123 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1124 	if (imgp->pagesizes != 0) {
1125 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1126 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1127 	}
1128 	if (imgp->sysent->sv_timekeep_base != 0) {
1129 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1130 		    imgp->sysent->sv_timekeep_base);
1131 	}
1132 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1133 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1134 	    imgp->sysent->sv_stackprot);
1135 	if (imgp->sysent->sv_hwcap != NULL)
1136 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1137 	if (imgp->sysent->sv_hwcap2 != NULL)
1138 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1139 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1140 
1141 	free(imgp->auxargs, M_TEMP);
1142 	imgp->auxargs = NULL;
1143 
1144 	base--;
1145 	suword(base, (long)imgp->args->argc);
1146 	*stack_base = (register_t *)base;
1147 	return (0);
1148 }
1149 
1150 /*
1151  * Code for generating ELF core dumps.
1152  */
1153 
1154 typedef void (*segment_callback)(vm_map_entry_t, void *);
1155 
1156 /* Closure for cb_put_phdr(). */
1157 struct phdr_closure {
1158 	Elf_Phdr *phdr;		/* Program header to fill in */
1159 	Elf_Off offset;		/* Offset of segment in core file */
1160 };
1161 
1162 /* Closure for cb_size_segment(). */
1163 struct sseg_closure {
1164 	int count;		/* Count of writable segments. */
1165 	size_t size;		/* Total size of all writable segments. */
1166 };
1167 
1168 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1169 
1170 struct note_info {
1171 	int		type;		/* Note type. */
1172 	outfunc_t 	outfunc; 	/* Output function. */
1173 	void		*outarg;	/* Argument for the output function. */
1174 	size_t		outsize;	/* Output size. */
1175 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1176 };
1177 
1178 TAILQ_HEAD(note_info_list, note_info);
1179 
1180 /* Coredump output parameters. */
1181 struct coredump_params {
1182 	off_t		offset;
1183 	struct ucred	*active_cred;
1184 	struct ucred	*file_cred;
1185 	struct thread	*td;
1186 	struct vnode	*vp;
1187 	struct compressor *comp;
1188 };
1189 
1190 extern int compress_user_cores;
1191 extern int compress_user_cores_level;
1192 
1193 static void cb_put_phdr(vm_map_entry_t, void *);
1194 static void cb_size_segment(vm_map_entry_t, void *);
1195 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1196     enum uio_seg);
1197 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1198 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1199     struct note_info_list *, size_t);
1200 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1201     size_t *);
1202 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1203 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1204 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1205 static int sbuf_drain_core_output(void *, const char *, int);
1206 static int sbuf_drain_count(void *arg, const char *data, int len);
1207 
1208 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1209 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1210 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1211 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1212 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1213 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1214 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1215 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1216 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1217 static void note_procstat_files(void *, struct sbuf *, size_t *);
1218 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1219 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1220 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1221 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1222 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1223 
1224 /*
1225  * Write out a core segment to the compression stream.
1226  */
1227 static int
1228 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1229 {
1230 	u_int chunk_len;
1231 	int error;
1232 
1233 	while (len > 0) {
1234 		chunk_len = MIN(len, CORE_BUF_SIZE);
1235 
1236 		/*
1237 		 * We can get EFAULT error here.
1238 		 * In that case zero out the current chunk of the segment.
1239 		 */
1240 		error = copyin(base, buf, chunk_len);
1241 		if (error != 0)
1242 			bzero(buf, chunk_len);
1243 		error = compressor_write(p->comp, buf, chunk_len);
1244 		if (error != 0)
1245 			break;
1246 		base += chunk_len;
1247 		len -= chunk_len;
1248 	}
1249 	return (error);
1250 }
1251 
1252 static int
1253 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1254 {
1255 
1256 	return (core_write((struct coredump_params *)arg, base, len, offset,
1257 	    UIO_SYSSPACE));
1258 }
1259 
1260 static int
1261 core_write(struct coredump_params *p, const void *base, size_t len,
1262     off_t offset, enum uio_seg seg)
1263 {
1264 
1265 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1266 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1267 	    p->active_cred, p->file_cred, NULL, p->td));
1268 }
1269 
1270 static int
1271 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1272     void *tmpbuf)
1273 {
1274 	int error;
1275 
1276 	if (p->comp != NULL)
1277 		return (compress_chunk(p, base, tmpbuf, len));
1278 
1279 	/*
1280 	 * EFAULT is a non-fatal error that we can get, for example,
1281 	 * if the segment is backed by a file but extends beyond its
1282 	 * end.
1283 	 */
1284 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1285 	if (error == EFAULT) {
1286 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1287 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1288 		    "for process %s\n", base, len, offset, curproc->p_comm);
1289 
1290 		/*
1291 		 * Write a "real" zero byte at the end of the target region
1292 		 * in the case this is the last segment.
1293 		 * The intermediate space will be implicitly zero-filled.
1294 		 */
1295 		error = core_write(p, zero_region, 1, offset + len - 1,
1296 		    UIO_SYSSPACE);
1297 	}
1298 	return (error);
1299 }
1300 
1301 /*
1302  * Drain into a core file.
1303  */
1304 static int
1305 sbuf_drain_core_output(void *arg, const char *data, int len)
1306 {
1307 	struct coredump_params *p;
1308 	int error, locked;
1309 
1310 	p = (struct coredump_params *)arg;
1311 
1312 	/*
1313 	 * Some kern_proc out routines that print to this sbuf may
1314 	 * call us with the process lock held. Draining with the
1315 	 * non-sleepable lock held is unsafe. The lock is needed for
1316 	 * those routines when dumping a live process. In our case we
1317 	 * can safely release the lock before draining and acquire
1318 	 * again after.
1319 	 */
1320 	locked = PROC_LOCKED(p->td->td_proc);
1321 	if (locked)
1322 		PROC_UNLOCK(p->td->td_proc);
1323 	if (p->comp != NULL)
1324 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1325 	else
1326 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1327 		    UIO_SYSSPACE);
1328 	if (locked)
1329 		PROC_LOCK(p->td->td_proc);
1330 	if (error != 0)
1331 		return (-error);
1332 	p->offset += len;
1333 	return (len);
1334 }
1335 
1336 /*
1337  * Drain into a counter.
1338  */
1339 static int
1340 sbuf_drain_count(void *arg, const char *data __unused, int len)
1341 {
1342 	size_t *sizep;
1343 
1344 	sizep = (size_t *)arg;
1345 	*sizep += len;
1346 	return (len);
1347 }
1348 
1349 int
1350 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1351 {
1352 	struct ucred *cred = td->td_ucred;
1353 	int error = 0;
1354 	struct sseg_closure seginfo;
1355 	struct note_info_list notelst;
1356 	struct coredump_params params;
1357 	struct note_info *ninfo;
1358 	void *hdr, *tmpbuf;
1359 	size_t hdrsize, notesz, coresize;
1360 
1361 	hdr = NULL;
1362 	tmpbuf = NULL;
1363 	TAILQ_INIT(&notelst);
1364 
1365 	/* Size the program segments. */
1366 	seginfo.count = 0;
1367 	seginfo.size = 0;
1368 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1369 
1370 	/*
1371 	 * Collect info about the core file header area.
1372 	 */
1373 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1374 	if (seginfo.count + 1 >= PN_XNUM)
1375 		hdrsize += sizeof(Elf_Shdr);
1376 	__elfN(prepare_notes)(td, &notelst, &notesz);
1377 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1378 
1379 	/* Set up core dump parameters. */
1380 	params.offset = 0;
1381 	params.active_cred = cred;
1382 	params.file_cred = NOCRED;
1383 	params.td = td;
1384 	params.vp = vp;
1385 	params.comp = NULL;
1386 
1387 #ifdef RACCT
1388 	if (racct_enable) {
1389 		PROC_LOCK(td->td_proc);
1390 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1391 		PROC_UNLOCK(td->td_proc);
1392 		if (error != 0) {
1393 			error = EFAULT;
1394 			goto done;
1395 		}
1396 	}
1397 #endif
1398 	if (coresize >= limit) {
1399 		error = EFAULT;
1400 		goto done;
1401 	}
1402 
1403 	/* Create a compression stream if necessary. */
1404 	if (compress_user_cores != 0) {
1405 		params.comp = compressor_init(core_compressed_write,
1406 		    compress_user_cores, CORE_BUF_SIZE,
1407 		    compress_user_cores_level, &params);
1408 		if (params.comp == NULL) {
1409 			error = EFAULT;
1410 			goto done;
1411 		}
1412 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1413         }
1414 
1415 	/*
1416 	 * Allocate memory for building the header, fill it up,
1417 	 * and write it out following the notes.
1418 	 */
1419 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1420 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1421 	    notesz);
1422 
1423 	/* Write the contents of all of the writable segments. */
1424 	if (error == 0) {
1425 		Elf_Phdr *php;
1426 		off_t offset;
1427 		int i;
1428 
1429 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1430 		offset = round_page(hdrsize + notesz);
1431 		for (i = 0; i < seginfo.count; i++) {
1432 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1433 			    php->p_filesz, offset, &params, tmpbuf);
1434 			if (error != 0)
1435 				break;
1436 			offset += php->p_filesz;
1437 			php++;
1438 		}
1439 		if (error == 0 && params.comp != NULL)
1440 			error = compressor_flush(params.comp);
1441 	}
1442 	if (error) {
1443 		log(LOG_WARNING,
1444 		    "Failed to write core file for process %s (error %d)\n",
1445 		    curproc->p_comm, error);
1446 	}
1447 
1448 done:
1449 	free(tmpbuf, M_TEMP);
1450 	if (params.comp != NULL)
1451 		compressor_fini(params.comp);
1452 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1453 		TAILQ_REMOVE(&notelst, ninfo, link);
1454 		free(ninfo, M_TEMP);
1455 	}
1456 	if (hdr != NULL)
1457 		free(hdr, M_TEMP);
1458 
1459 	return (error);
1460 }
1461 
1462 /*
1463  * A callback for each_dumpable_segment() to write out the segment's
1464  * program header entry.
1465  */
1466 static void
1467 cb_put_phdr(entry, closure)
1468 	vm_map_entry_t entry;
1469 	void *closure;
1470 {
1471 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1472 	Elf_Phdr *phdr = phc->phdr;
1473 
1474 	phc->offset = round_page(phc->offset);
1475 
1476 	phdr->p_type = PT_LOAD;
1477 	phdr->p_offset = phc->offset;
1478 	phdr->p_vaddr = entry->start;
1479 	phdr->p_paddr = 0;
1480 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1481 	phdr->p_align = PAGE_SIZE;
1482 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1483 
1484 	phc->offset += phdr->p_filesz;
1485 	phc->phdr++;
1486 }
1487 
1488 /*
1489  * A callback for each_dumpable_segment() to gather information about
1490  * the number of segments and their total size.
1491  */
1492 static void
1493 cb_size_segment(vm_map_entry_t entry, void *closure)
1494 {
1495 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1496 
1497 	ssc->count++;
1498 	ssc->size += entry->end - entry->start;
1499 }
1500 
1501 /*
1502  * For each writable segment in the process's memory map, call the given
1503  * function with a pointer to the map entry and some arbitrary
1504  * caller-supplied data.
1505  */
1506 static void
1507 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1508 {
1509 	struct proc *p = td->td_proc;
1510 	vm_map_t map = &p->p_vmspace->vm_map;
1511 	vm_map_entry_t entry;
1512 	vm_object_t backing_object, object;
1513 	boolean_t ignore_entry;
1514 
1515 	vm_map_lock_read(map);
1516 	for (entry = map->header.next; entry != &map->header;
1517 	    entry = entry->next) {
1518 		/*
1519 		 * Don't dump inaccessible mappings, deal with legacy
1520 		 * coredump mode.
1521 		 *
1522 		 * Note that read-only segments related to the elf binary
1523 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1524 		 * need to arbitrarily ignore such segments.
1525 		 */
1526 		if (elf_legacy_coredump) {
1527 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1528 				continue;
1529 		} else {
1530 			if ((entry->protection & VM_PROT_ALL) == 0)
1531 				continue;
1532 		}
1533 
1534 		/*
1535 		 * Dont include memory segment in the coredump if
1536 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1537 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1538 		 * kernel map).
1539 		 */
1540 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1541 			continue;
1542 
1543 		if ((object = entry->object.vm_object) == NULL)
1544 			continue;
1545 
1546 		/* Ignore memory-mapped devices and such things. */
1547 		VM_OBJECT_RLOCK(object);
1548 		while ((backing_object = object->backing_object) != NULL) {
1549 			VM_OBJECT_RLOCK(backing_object);
1550 			VM_OBJECT_RUNLOCK(object);
1551 			object = backing_object;
1552 		}
1553 		ignore_entry = object->type != OBJT_DEFAULT &&
1554 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1555 		    object->type != OBJT_PHYS;
1556 		VM_OBJECT_RUNLOCK(object);
1557 		if (ignore_entry)
1558 			continue;
1559 
1560 		(*func)(entry, closure);
1561 	}
1562 	vm_map_unlock_read(map);
1563 }
1564 
1565 /*
1566  * Write the core file header to the file, including padding up to
1567  * the page boundary.
1568  */
1569 static int
1570 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1571     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1572 {
1573 	struct note_info *ninfo;
1574 	struct sbuf *sb;
1575 	int error;
1576 
1577 	/* Fill in the header. */
1578 	bzero(hdr, hdrsize);
1579 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1580 
1581 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1582 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1583 	sbuf_start_section(sb, NULL);
1584 	sbuf_bcat(sb, hdr, hdrsize);
1585 	TAILQ_FOREACH(ninfo, notelst, link)
1586 	    __elfN(putnote)(ninfo, sb);
1587 	/* Align up to a page boundary for the program segments. */
1588 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1589 	error = sbuf_finish(sb);
1590 	sbuf_delete(sb);
1591 
1592 	return (error);
1593 }
1594 
1595 static void
1596 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1597     size_t *sizep)
1598 {
1599 	struct proc *p;
1600 	struct thread *thr;
1601 	size_t size;
1602 
1603 	p = td->td_proc;
1604 	size = 0;
1605 
1606 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1607 
1608 	/*
1609 	 * To have the debugger select the right thread (LWP) as the initial
1610 	 * thread, we dump the state of the thread passed to us in td first.
1611 	 * This is the thread that causes the core dump and thus likely to
1612 	 * be the right thread one wants to have selected in the debugger.
1613 	 */
1614 	thr = td;
1615 	while (thr != NULL) {
1616 		size += register_note(list, NT_PRSTATUS,
1617 		    __elfN(note_prstatus), thr);
1618 		size += register_note(list, NT_FPREGSET,
1619 		    __elfN(note_fpregset), thr);
1620 		size += register_note(list, NT_THRMISC,
1621 		    __elfN(note_thrmisc), thr);
1622 		size += register_note(list, NT_PTLWPINFO,
1623 		    __elfN(note_ptlwpinfo), thr);
1624 		size += register_note(list, -1,
1625 		    __elfN(note_threadmd), thr);
1626 
1627 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1628 		    TAILQ_NEXT(thr, td_plist);
1629 		if (thr == td)
1630 			thr = TAILQ_NEXT(thr, td_plist);
1631 	}
1632 
1633 	size += register_note(list, NT_PROCSTAT_PROC,
1634 	    __elfN(note_procstat_proc), p);
1635 	size += register_note(list, NT_PROCSTAT_FILES,
1636 	    note_procstat_files, p);
1637 	size += register_note(list, NT_PROCSTAT_VMMAP,
1638 	    note_procstat_vmmap, p);
1639 	size += register_note(list, NT_PROCSTAT_GROUPS,
1640 	    note_procstat_groups, p);
1641 	size += register_note(list, NT_PROCSTAT_UMASK,
1642 	    note_procstat_umask, p);
1643 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1644 	    note_procstat_rlimit, p);
1645 	size += register_note(list, NT_PROCSTAT_OSREL,
1646 	    note_procstat_osrel, p);
1647 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1648 	    __elfN(note_procstat_psstrings), p);
1649 	size += register_note(list, NT_PROCSTAT_AUXV,
1650 	    __elfN(note_procstat_auxv), p);
1651 
1652 	*sizep = size;
1653 }
1654 
1655 static void
1656 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1657     size_t notesz)
1658 {
1659 	Elf_Ehdr *ehdr;
1660 	Elf_Phdr *phdr;
1661 	Elf_Shdr *shdr;
1662 	struct phdr_closure phc;
1663 
1664 	ehdr = (Elf_Ehdr *)hdr;
1665 
1666 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1667 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1668 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1669 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1670 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1671 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1672 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1673 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1674 	ehdr->e_ident[EI_ABIVERSION] = 0;
1675 	ehdr->e_ident[EI_PAD] = 0;
1676 	ehdr->e_type = ET_CORE;
1677 	ehdr->e_machine = td->td_proc->p_elf_machine;
1678 	ehdr->e_version = EV_CURRENT;
1679 	ehdr->e_entry = 0;
1680 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1681 	ehdr->e_flags = td->td_proc->p_elf_flags;
1682 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1683 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1684 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1685 	ehdr->e_shstrndx = SHN_UNDEF;
1686 	if (numsegs + 1 < PN_XNUM) {
1687 		ehdr->e_phnum = numsegs + 1;
1688 		ehdr->e_shnum = 0;
1689 	} else {
1690 		ehdr->e_phnum = PN_XNUM;
1691 		ehdr->e_shnum = 1;
1692 
1693 		ehdr->e_shoff = ehdr->e_phoff +
1694 		    (numsegs + 1) * ehdr->e_phentsize;
1695 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1696 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1697 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1698 
1699 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1700 		memset(shdr, 0, sizeof(*shdr));
1701 		/*
1702 		 * A special first section is used to hold large segment and
1703 		 * section counts.  This was proposed by Sun Microsystems in
1704 		 * Solaris and has been adopted by Linux; the standard ELF
1705 		 * tools are already familiar with the technique.
1706 		 *
1707 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1708 		 * (or 12-7 depending on the version of the document) for more
1709 		 * details.
1710 		 */
1711 		shdr->sh_type = SHT_NULL;
1712 		shdr->sh_size = ehdr->e_shnum;
1713 		shdr->sh_link = ehdr->e_shstrndx;
1714 		shdr->sh_info = numsegs + 1;
1715 	}
1716 
1717 	/*
1718 	 * Fill in the program header entries.
1719 	 */
1720 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1721 
1722 	/* The note segement. */
1723 	phdr->p_type = PT_NOTE;
1724 	phdr->p_offset = hdrsize;
1725 	phdr->p_vaddr = 0;
1726 	phdr->p_paddr = 0;
1727 	phdr->p_filesz = notesz;
1728 	phdr->p_memsz = 0;
1729 	phdr->p_flags = PF_R;
1730 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1731 	phdr++;
1732 
1733 	/* All the writable segments from the program. */
1734 	phc.phdr = phdr;
1735 	phc.offset = round_page(hdrsize + notesz);
1736 	each_dumpable_segment(td, cb_put_phdr, &phc);
1737 }
1738 
1739 static size_t
1740 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1741 {
1742 	struct note_info *ninfo;
1743 	size_t size, notesize;
1744 
1745 	size = 0;
1746 	out(arg, NULL, &size);
1747 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1748 	ninfo->type = type;
1749 	ninfo->outfunc = out;
1750 	ninfo->outarg = arg;
1751 	ninfo->outsize = size;
1752 	TAILQ_INSERT_TAIL(list, ninfo, link);
1753 
1754 	if (type == -1)
1755 		return (size);
1756 
1757 	notesize = sizeof(Elf_Note) +		/* note header */
1758 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1759 						/* note name */
1760 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1761 
1762 	return (notesize);
1763 }
1764 
1765 static size_t
1766 append_note_data(const void *src, void *dst, size_t len)
1767 {
1768 	size_t padded_len;
1769 
1770 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1771 	if (dst != NULL) {
1772 		bcopy(src, dst, len);
1773 		bzero((char *)dst + len, padded_len - len);
1774 	}
1775 	return (padded_len);
1776 }
1777 
1778 size_t
1779 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1780 {
1781 	Elf_Note *note;
1782 	char *buf;
1783 	size_t notesize;
1784 
1785 	buf = dst;
1786 	if (buf != NULL) {
1787 		note = (Elf_Note *)buf;
1788 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1789 		note->n_descsz = size;
1790 		note->n_type = type;
1791 		buf += sizeof(*note);
1792 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1793 		    sizeof(FREEBSD_ABI_VENDOR));
1794 		append_note_data(src, buf, size);
1795 		if (descp != NULL)
1796 			*descp = buf;
1797 	}
1798 
1799 	notesize = sizeof(Elf_Note) +		/* note header */
1800 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1801 						/* note name */
1802 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1803 
1804 	return (notesize);
1805 }
1806 
1807 static void
1808 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1809 {
1810 	Elf_Note note;
1811 	ssize_t old_len, sect_len;
1812 	size_t new_len, descsz, i;
1813 
1814 	if (ninfo->type == -1) {
1815 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1816 		return;
1817 	}
1818 
1819 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1820 	note.n_descsz = ninfo->outsize;
1821 	note.n_type = ninfo->type;
1822 
1823 	sbuf_bcat(sb, &note, sizeof(note));
1824 	sbuf_start_section(sb, &old_len);
1825 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1826 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1827 	if (note.n_descsz == 0)
1828 		return;
1829 	sbuf_start_section(sb, &old_len);
1830 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1831 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1832 	if (sect_len < 0)
1833 		return;
1834 
1835 	new_len = (size_t)sect_len;
1836 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1837 	if (new_len < descsz) {
1838 		/*
1839 		 * It is expected that individual note emitters will correctly
1840 		 * predict their expected output size and fill up to that size
1841 		 * themselves, padding in a format-specific way if needed.
1842 		 * However, in case they don't, just do it here with zeros.
1843 		 */
1844 		for (i = 0; i < descsz - new_len; i++)
1845 			sbuf_putc(sb, 0);
1846 	} else if (new_len > descsz) {
1847 		/*
1848 		 * We can't always truncate sb -- we may have drained some
1849 		 * of it already.
1850 		 */
1851 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1852 		    "read it (%zu > %zu).  Since it is longer than "
1853 		    "expected, this coredump's notes are corrupt.  THIS "
1854 		    "IS A BUG in the note_procstat routine for type %u.\n",
1855 		    __func__, (unsigned)note.n_type, new_len, descsz,
1856 		    (unsigned)note.n_type));
1857 	}
1858 }
1859 
1860 /*
1861  * Miscellaneous note out functions.
1862  */
1863 
1864 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1865 #include <compat/freebsd32/freebsd32.h>
1866 #include <compat/freebsd32/freebsd32_signal.h>
1867 
1868 typedef struct prstatus32 elf_prstatus_t;
1869 typedef struct prpsinfo32 elf_prpsinfo_t;
1870 typedef struct fpreg32 elf_prfpregset_t;
1871 typedef struct fpreg32 elf_fpregset_t;
1872 typedef struct reg32 elf_gregset_t;
1873 typedef struct thrmisc32 elf_thrmisc_t;
1874 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1875 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1876 typedef uint32_t elf_ps_strings_t;
1877 #else
1878 typedef prstatus_t elf_prstatus_t;
1879 typedef prpsinfo_t elf_prpsinfo_t;
1880 typedef prfpregset_t elf_prfpregset_t;
1881 typedef prfpregset_t elf_fpregset_t;
1882 typedef gregset_t elf_gregset_t;
1883 typedef thrmisc_t elf_thrmisc_t;
1884 #define ELF_KERN_PROC_MASK	0
1885 typedef struct kinfo_proc elf_kinfo_proc_t;
1886 typedef vm_offset_t elf_ps_strings_t;
1887 #endif
1888 
1889 static void
1890 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1891 {
1892 	struct sbuf sbarg;
1893 	size_t len;
1894 	char *cp, *end;
1895 	struct proc *p;
1896 	elf_prpsinfo_t *psinfo;
1897 	int error;
1898 
1899 	p = (struct proc *)arg;
1900 	if (sb != NULL) {
1901 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1902 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1903 		psinfo->pr_version = PRPSINFO_VERSION;
1904 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1905 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1906 		PROC_LOCK(p);
1907 		if (p->p_args != NULL) {
1908 			len = sizeof(psinfo->pr_psargs) - 1;
1909 			if (len > p->p_args->ar_length)
1910 				len = p->p_args->ar_length;
1911 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1912 			PROC_UNLOCK(p);
1913 			error = 0;
1914 		} else {
1915 			_PHOLD(p);
1916 			PROC_UNLOCK(p);
1917 			sbuf_new(&sbarg, psinfo->pr_psargs,
1918 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1919 			error = proc_getargv(curthread, p, &sbarg);
1920 			PRELE(p);
1921 			if (sbuf_finish(&sbarg) == 0)
1922 				len = sbuf_len(&sbarg) - 1;
1923 			else
1924 				len = sizeof(psinfo->pr_psargs) - 1;
1925 			sbuf_delete(&sbarg);
1926 		}
1927 		if (error || len == 0)
1928 			strlcpy(psinfo->pr_psargs, p->p_comm,
1929 			    sizeof(psinfo->pr_psargs));
1930 		else {
1931 			KASSERT(len < sizeof(psinfo->pr_psargs),
1932 			    ("len is too long: %zu vs %zu", len,
1933 			    sizeof(psinfo->pr_psargs)));
1934 			cp = psinfo->pr_psargs;
1935 			end = cp + len - 1;
1936 			for (;;) {
1937 				cp = memchr(cp, '\0', end - cp);
1938 				if (cp == NULL)
1939 					break;
1940 				*cp = ' ';
1941 			}
1942 		}
1943 		psinfo->pr_pid = p->p_pid;
1944 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1945 		free(psinfo, M_TEMP);
1946 	}
1947 	*sizep = sizeof(*psinfo);
1948 }
1949 
1950 static void
1951 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1952 {
1953 	struct thread *td;
1954 	elf_prstatus_t *status;
1955 
1956 	td = (struct thread *)arg;
1957 	if (sb != NULL) {
1958 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1959 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1960 		status->pr_version = PRSTATUS_VERSION;
1961 		status->pr_statussz = sizeof(elf_prstatus_t);
1962 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1963 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1964 		status->pr_osreldate = osreldate;
1965 		status->pr_cursig = td->td_proc->p_sig;
1966 		status->pr_pid = td->td_tid;
1967 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1968 		fill_regs32(td, &status->pr_reg);
1969 #else
1970 		fill_regs(td, &status->pr_reg);
1971 #endif
1972 		sbuf_bcat(sb, status, sizeof(*status));
1973 		free(status, M_TEMP);
1974 	}
1975 	*sizep = sizeof(*status);
1976 }
1977 
1978 static void
1979 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1980 {
1981 	struct thread *td;
1982 	elf_prfpregset_t *fpregset;
1983 
1984 	td = (struct thread *)arg;
1985 	if (sb != NULL) {
1986 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1987 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1988 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1989 		fill_fpregs32(td, fpregset);
1990 #else
1991 		fill_fpregs(td, fpregset);
1992 #endif
1993 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1994 		free(fpregset, M_TEMP);
1995 	}
1996 	*sizep = sizeof(*fpregset);
1997 }
1998 
1999 static void
2000 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2001 {
2002 	struct thread *td;
2003 	elf_thrmisc_t thrmisc;
2004 
2005 	td = (struct thread *)arg;
2006 	if (sb != NULL) {
2007 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2008 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2009 		strcpy(thrmisc.pr_tname, td->td_name);
2010 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2011 	}
2012 	*sizep = sizeof(thrmisc);
2013 }
2014 
2015 static void
2016 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2017 {
2018 	struct thread *td;
2019 	size_t size;
2020 	int structsize;
2021 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2022 	struct ptrace_lwpinfo32 pl;
2023 #else
2024 	struct ptrace_lwpinfo pl;
2025 #endif
2026 
2027 	td = (struct thread *)arg;
2028 	size = sizeof(structsize) + sizeof(pl);
2029 	if (sb != NULL) {
2030 		KASSERT(*sizep == size, ("invalid size"));
2031 		structsize = sizeof(pl);
2032 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2033 		bzero(&pl, sizeof(pl));
2034 		pl.pl_lwpid = td->td_tid;
2035 		pl.pl_event = PL_EVENT_NONE;
2036 		pl.pl_sigmask = td->td_sigmask;
2037 		pl.pl_siglist = td->td_siglist;
2038 		if (td->td_si.si_signo != 0) {
2039 			pl.pl_event = PL_EVENT_SIGNAL;
2040 			pl.pl_flags |= PL_FLAG_SI;
2041 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2042 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2043 #else
2044 			pl.pl_siginfo = td->td_si;
2045 #endif
2046 		}
2047 		strcpy(pl.pl_tdname, td->td_name);
2048 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2049 		sbuf_bcat(sb, &pl, sizeof(pl));
2050 	}
2051 	*sizep = size;
2052 }
2053 
2054 /*
2055  * Allow for MD specific notes, as well as any MD
2056  * specific preparations for writing MI notes.
2057  */
2058 static void
2059 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2060 {
2061 	struct thread *td;
2062 	void *buf;
2063 	size_t size;
2064 
2065 	td = (struct thread *)arg;
2066 	size = *sizep;
2067 	if (size != 0 && sb != NULL)
2068 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2069 	else
2070 		buf = NULL;
2071 	size = 0;
2072 	__elfN(dump_thread)(td, buf, &size);
2073 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2074 	if (size != 0 && sb != NULL)
2075 		sbuf_bcat(sb, buf, size);
2076 	free(buf, M_TEMP);
2077 	*sizep = size;
2078 }
2079 
2080 #ifdef KINFO_PROC_SIZE
2081 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2082 #endif
2083 
2084 static void
2085 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2086 {
2087 	struct proc *p;
2088 	size_t size;
2089 	int structsize;
2090 
2091 	p = (struct proc *)arg;
2092 	size = sizeof(structsize) + p->p_numthreads *
2093 	    sizeof(elf_kinfo_proc_t);
2094 
2095 	if (sb != NULL) {
2096 		KASSERT(*sizep == size, ("invalid size"));
2097 		structsize = sizeof(elf_kinfo_proc_t);
2098 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2099 		sx_slock(&proctree_lock);
2100 		PROC_LOCK(p);
2101 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2102 		sx_sunlock(&proctree_lock);
2103 	}
2104 	*sizep = size;
2105 }
2106 
2107 #ifdef KINFO_FILE_SIZE
2108 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2109 #endif
2110 
2111 static void
2112 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2113 {
2114 	struct proc *p;
2115 	size_t size, sect_sz, i;
2116 	ssize_t start_len, sect_len;
2117 	int structsize, filedesc_flags;
2118 
2119 	if (coredump_pack_fileinfo)
2120 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2121 	else
2122 		filedesc_flags = 0;
2123 
2124 	p = (struct proc *)arg;
2125 	structsize = sizeof(struct kinfo_file);
2126 	if (sb == NULL) {
2127 		size = 0;
2128 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2129 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2130 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2131 		PROC_LOCK(p);
2132 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2133 		sbuf_finish(sb);
2134 		sbuf_delete(sb);
2135 		*sizep = size;
2136 	} else {
2137 		sbuf_start_section(sb, &start_len);
2138 
2139 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2140 		PROC_LOCK(p);
2141 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2142 		    filedesc_flags);
2143 
2144 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2145 		if (sect_len < 0)
2146 			return;
2147 		sect_sz = sect_len;
2148 
2149 		KASSERT(sect_sz <= *sizep,
2150 		    ("kern_proc_filedesc_out did not respect maxlen; "
2151 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2152 		     sect_sz - sizeof(structsize)));
2153 
2154 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2155 			sbuf_putc(sb, 0);
2156 	}
2157 }
2158 
2159 #ifdef KINFO_VMENTRY_SIZE
2160 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2161 #endif
2162 
2163 static void
2164 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2165 {
2166 	struct proc *p;
2167 	size_t size;
2168 	int structsize, vmmap_flags;
2169 
2170 	if (coredump_pack_vmmapinfo)
2171 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2172 	else
2173 		vmmap_flags = 0;
2174 
2175 	p = (struct proc *)arg;
2176 	structsize = sizeof(struct kinfo_vmentry);
2177 	if (sb == NULL) {
2178 		size = 0;
2179 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2180 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2181 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2182 		PROC_LOCK(p);
2183 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2184 		sbuf_finish(sb);
2185 		sbuf_delete(sb);
2186 		*sizep = size;
2187 	} else {
2188 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2189 		PROC_LOCK(p);
2190 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2191 		    vmmap_flags);
2192 	}
2193 }
2194 
2195 static void
2196 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2197 {
2198 	struct proc *p;
2199 	size_t size;
2200 	int structsize;
2201 
2202 	p = (struct proc *)arg;
2203 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2204 	if (sb != NULL) {
2205 		KASSERT(*sizep == size, ("invalid size"));
2206 		structsize = sizeof(gid_t);
2207 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2208 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2209 		    sizeof(gid_t));
2210 	}
2211 	*sizep = size;
2212 }
2213 
2214 static void
2215 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2216 {
2217 	struct proc *p;
2218 	size_t size;
2219 	int structsize;
2220 
2221 	p = (struct proc *)arg;
2222 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2223 	if (sb != NULL) {
2224 		KASSERT(*sizep == size, ("invalid size"));
2225 		structsize = sizeof(p->p_fd->fd_cmask);
2226 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2227 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2228 	}
2229 	*sizep = size;
2230 }
2231 
2232 static void
2233 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2234 {
2235 	struct proc *p;
2236 	struct rlimit rlim[RLIM_NLIMITS];
2237 	size_t size;
2238 	int structsize, i;
2239 
2240 	p = (struct proc *)arg;
2241 	size = sizeof(structsize) + sizeof(rlim);
2242 	if (sb != NULL) {
2243 		KASSERT(*sizep == size, ("invalid size"));
2244 		structsize = sizeof(rlim);
2245 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2246 		PROC_LOCK(p);
2247 		for (i = 0; i < RLIM_NLIMITS; i++)
2248 			lim_rlimit_proc(p, i, &rlim[i]);
2249 		PROC_UNLOCK(p);
2250 		sbuf_bcat(sb, rlim, sizeof(rlim));
2251 	}
2252 	*sizep = size;
2253 }
2254 
2255 static void
2256 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2257 {
2258 	struct proc *p;
2259 	size_t size;
2260 	int structsize;
2261 
2262 	p = (struct proc *)arg;
2263 	size = sizeof(structsize) + sizeof(p->p_osrel);
2264 	if (sb != NULL) {
2265 		KASSERT(*sizep == size, ("invalid size"));
2266 		structsize = sizeof(p->p_osrel);
2267 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2268 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2269 	}
2270 	*sizep = size;
2271 }
2272 
2273 static void
2274 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2275 {
2276 	struct proc *p;
2277 	elf_ps_strings_t ps_strings;
2278 	size_t size;
2279 	int structsize;
2280 
2281 	p = (struct proc *)arg;
2282 	size = sizeof(structsize) + sizeof(ps_strings);
2283 	if (sb != NULL) {
2284 		KASSERT(*sizep == size, ("invalid size"));
2285 		structsize = sizeof(ps_strings);
2286 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2287 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2288 #else
2289 		ps_strings = p->p_sysent->sv_psstrings;
2290 #endif
2291 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2292 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2293 	}
2294 	*sizep = size;
2295 }
2296 
2297 static void
2298 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2299 {
2300 	struct proc *p;
2301 	size_t size;
2302 	int structsize;
2303 
2304 	p = (struct proc *)arg;
2305 	if (sb == NULL) {
2306 		size = 0;
2307 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2308 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2309 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2310 		PHOLD(p);
2311 		proc_getauxv(curthread, p, sb);
2312 		PRELE(p);
2313 		sbuf_finish(sb);
2314 		sbuf_delete(sb);
2315 		*sizep = size;
2316 	} else {
2317 		structsize = sizeof(Elf_Auxinfo);
2318 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2319 		PHOLD(p);
2320 		proc_getauxv(curthread, p, sb);
2321 		PRELE(p);
2322 	}
2323 }
2324 
2325 static boolean_t
2326 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2327     int32_t *osrel, const Elf_Phdr *pnote)
2328 {
2329 	const Elf_Note *note, *note0, *note_end;
2330 	const char *note_name;
2331 	char *buf;
2332 	int i, error;
2333 	boolean_t res;
2334 
2335 	/* We need some limit, might as well use PAGE_SIZE. */
2336 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2337 		return (FALSE);
2338 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2339 	if (pnote->p_offset > PAGE_SIZE ||
2340 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2341 		VOP_UNLOCK(imgp->vp, 0);
2342 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2343 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2344 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2345 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2346 		    curthread->td_ucred, NOCRED, NULL, curthread);
2347 		if (error != 0) {
2348 			uprintf("i/o error PT_NOTE\n");
2349 			res = FALSE;
2350 			goto ret;
2351 		}
2352 		note = note0 = (const Elf_Note *)buf;
2353 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2354 	} else {
2355 		note = note0 = (const Elf_Note *)(imgp->image_header +
2356 		    pnote->p_offset);
2357 		note_end = (const Elf_Note *)(imgp->image_header +
2358 		    pnote->p_offset + pnote->p_filesz);
2359 		buf = NULL;
2360 	}
2361 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2362 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2363 		    (const char *)note < sizeof(Elf_Note)) {
2364 			res = FALSE;
2365 			goto ret;
2366 		}
2367 		if (note->n_namesz != checknote->hdr.n_namesz ||
2368 		    note->n_descsz != checknote->hdr.n_descsz ||
2369 		    note->n_type != checknote->hdr.n_type)
2370 			goto nextnote;
2371 		note_name = (const char *)(note + 1);
2372 		if (note_name + checknote->hdr.n_namesz >=
2373 		    (const char *)note_end || strncmp(checknote->vendor,
2374 		    note_name, checknote->hdr.n_namesz) != 0)
2375 			goto nextnote;
2376 
2377 		/*
2378 		 * Fetch the osreldate for binary
2379 		 * from the ELF OSABI-note if necessary.
2380 		 */
2381 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2382 		    checknote->trans_osrel != NULL) {
2383 			res = checknote->trans_osrel(note, osrel);
2384 			goto ret;
2385 		}
2386 		res = TRUE;
2387 		goto ret;
2388 nextnote:
2389 		note = (const Elf_Note *)((const char *)(note + 1) +
2390 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2391 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2392 	}
2393 	res = FALSE;
2394 ret:
2395 	free(buf, M_TEMP);
2396 	return (res);
2397 }
2398 
2399 /*
2400  * Try to find the appropriate ABI-note section for checknote,
2401  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2402  * first page of the image is searched, the same as for headers.
2403  */
2404 static boolean_t
2405 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2406     int32_t *osrel)
2407 {
2408 	const Elf_Phdr *phdr;
2409 	const Elf_Ehdr *hdr;
2410 	int i;
2411 
2412 	hdr = (const Elf_Ehdr *)imgp->image_header;
2413 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2414 
2415 	for (i = 0; i < hdr->e_phnum; i++) {
2416 		if (phdr[i].p_type == PT_NOTE &&
2417 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2418 			return (TRUE);
2419 	}
2420 	return (FALSE);
2421 
2422 }
2423 
2424 /*
2425  * Tell kern_execve.c about it, with a little help from the linker.
2426  */
2427 static struct execsw __elfN(execsw) = {
2428 	__CONCAT(exec_, __elfN(imgact)),
2429 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2430 };
2431 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2432 
2433 static vm_prot_t
2434 __elfN(trans_prot)(Elf_Word flags)
2435 {
2436 	vm_prot_t prot;
2437 
2438 	prot = 0;
2439 	if (flags & PF_X)
2440 		prot |= VM_PROT_EXECUTE;
2441 	if (flags & PF_W)
2442 		prot |= VM_PROT_WRITE;
2443 	if (flags & PF_R)
2444 		prot |= VM_PROT_READ;
2445 #if __ELF_WORD_SIZE == 32
2446 #if defined(__amd64__)
2447 	if (i386_read_exec && (flags & PF_R))
2448 		prot |= VM_PROT_EXECUTE;
2449 #endif
2450 #endif
2451 	return (prot);
2452 }
2453 
2454 static Elf_Word
2455 __elfN(untrans_prot)(vm_prot_t prot)
2456 {
2457 	Elf_Word flags;
2458 
2459 	flags = 0;
2460 	if (prot & VM_PROT_EXECUTE)
2461 		flags |= PF_X;
2462 	if (prot & VM_PROT_READ)
2463 		flags |= PF_R;
2464 	if (prot & VM_PROT_WRITE)
2465 		flags |= PF_W;
2466 	return (flags);
2467 }
2468