xref: /freebsd/sys/kern/imgact_elf.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
1 /*-
2  * Copyright (c) 2017 Dell EMC
3  * Copyright (c) 2000 David O'Brien
4  * Copyright (c) 1995-1996 Søren Schmidt
5  * Copyright (c) 1996 Peter Wemm
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer
13  *    in this position and unchanged.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. The name of the author may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_capsicum.h"
36 #include "opt_compat.h"
37 #include "opt_gzio.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/exec.h>
42 #include <sys/fcntl.h>
43 #include <sys/gzio.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
54 #include <sys/proc.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int interp_name_len, int32_t *osrel);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry, size_t pagesize);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
96     size_t pagesize);
97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
98 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99     int32_t *osrel);
100 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
101 static boolean_t __elfN(check_note)(struct image_params *imgp,
102     Elf_Brandnote *checknote, int32_t *osrel);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107     "");
108 
109 #define	CORE_BUF_SIZE	(16 * 1024)
110 
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118     &elf_legacy_coredump, 0,
119     "include all and only RW pages in core dumps");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if __ELF_WORD_SIZE == 32
133 #if defined(__amd64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136     "enable execution from readable segments");
137 #endif
138 #endif
139 
140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
141 
142 #define	trunc_page_ps(va, ps)	rounddown2(va, ps)
143 #define	round_page_ps(va, ps)	roundup2(va, ps)
144 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
145 
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
147 
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
150 	.hdr.n_descsz	= sizeof(int32_t),
151 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
152 	.vendor		= FREEBSD_ABI_VENDOR,
153 	.flags		= BN_TRANSLATE_OSREL,
154 	.trans_osrel	= __elfN(freebsd_trans_osrel)
155 };
156 
157 static boolean_t
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
159 {
160 	uintptr_t p;
161 
162 	p = (uintptr_t)(note + 1);
163 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 	*osrel = *(const int32_t *)(p);
165 
166 	return (TRUE);
167 }
168 
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
171 
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
174 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
175 	.hdr.n_type	= 1,
176 	.vendor		= GNU_ABI_VENDOR,
177 	.flags		= BN_TRANSLATE_OSREL,
178 	.trans_osrel	= kfreebsd_trans_osrel
179 };
180 
181 static boolean_t
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
183 {
184 	const Elf32_Word *desc;
185 	uintptr_t p;
186 
187 	p = (uintptr_t)(note + 1);
188 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
189 
190 	desc = (const Elf32_Word *)p;
191 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
192 		return (FALSE);
193 
194 	/*
195 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
197 	 */
198 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
199 
200 	return (TRUE);
201 }
202 
203 int
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
205 {
206 	int i;
207 
208 	for (i = 0; i < MAX_BRANDS; i++) {
209 		if (elf_brand_list[i] == NULL) {
210 			elf_brand_list[i] = entry;
211 			break;
212 		}
213 	}
214 	if (i == MAX_BRANDS) {
215 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
216 			__func__, entry);
217 		return (-1);
218 	}
219 	return (0);
220 }
221 
222 int
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
224 {
225 	int i;
226 
227 	for (i = 0; i < MAX_BRANDS; i++) {
228 		if (elf_brand_list[i] == entry) {
229 			elf_brand_list[i] = NULL;
230 			break;
231 		}
232 	}
233 	if (i == MAX_BRANDS)
234 		return (-1);
235 	return (0);
236 }
237 
238 int
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
240 {
241 	struct proc *p;
242 	int rval = FALSE;
243 
244 	sx_slock(&allproc_lock);
245 	FOREACH_PROC_IN_SYSTEM(p) {
246 		if (p->p_sysent == entry->sysvec) {
247 			rval = TRUE;
248 			break;
249 		}
250 	}
251 	sx_sunlock(&allproc_lock);
252 
253 	return (rval);
254 }
255 
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258     int interp_name_len, int32_t *osrel)
259 {
260 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
261 	Elf_Brandinfo *bi, *bi_m;
262 	boolean_t ret;
263 	int i;
264 
265 	/*
266 	 * We support four types of branding -- (1) the ELF EI_OSABI field
267 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 	 * branding w/in the ELF header, (3) path of the `interp_path'
269 	 * field, and (4) the ".note.ABI-tag" ELF section.
270 	 */
271 
272 	/* Look for an ".note.ABI-tag" ELF section */
273 	bi_m = NULL;
274 	for (i = 0; i < MAX_BRANDS; i++) {
275 		bi = elf_brand_list[i];
276 		if (bi == NULL)
277 			continue;
278 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
279 			continue;
280 		if (hdr->e_machine == bi->machine && (bi->flags &
281 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
282 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
283 			/* Give brand a chance to veto check_note's guess */
284 			if (ret && bi->header_supported)
285 				ret = bi->header_supported(imgp);
286 			/*
287 			 * If note checker claimed the binary, but the
288 			 * interpreter path in the image does not
289 			 * match default one for the brand, try to
290 			 * search for other brands with the same
291 			 * interpreter.  Either there is better brand
292 			 * with the right interpreter, or, failing
293 			 * this, we return first brand which accepted
294 			 * our note and, optionally, header.
295 			 */
296 			if (ret && bi_m == NULL && interp != NULL &&
297 			    (bi->interp_path == NULL ||
298 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
299 			    strncmp(interp, bi->interp_path, interp_name_len)
300 			    != 0))) {
301 				bi_m = bi;
302 				ret = 0;
303 			}
304 			if (ret)
305 				return (bi);
306 		}
307 	}
308 	if (bi_m != NULL)
309 		return (bi_m);
310 
311 	/* If the executable has a brand, search for it in the brand list. */
312 	for (i = 0; i < MAX_BRANDS; i++) {
313 		bi = elf_brand_list[i];
314 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
315 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
316 			continue;
317 		if (hdr->e_machine == bi->machine &&
318 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
319 		    (bi->compat_3_brand != NULL &&
320 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
321 		    bi->compat_3_brand) == 0))) {
322 			/* Looks good, but give brand a chance to veto */
323 			if (!bi->header_supported ||
324 			    bi->header_supported(imgp)) {
325 				/*
326 				 * Again, prefer strictly matching
327 				 * interpreter path.
328 				 */
329 				if (interp_name_len == 0 &&
330 				    bi->interp_path == NULL)
331 					return (bi);
332 				if (bi->interp_path != NULL &&
333 				    strlen(bi->interp_path) + 1 ==
334 				    interp_name_len && strncmp(interp,
335 				    bi->interp_path, interp_name_len) == 0)
336 					return (bi);
337 				if (bi_m == NULL)
338 					bi_m = bi;
339 			}
340 		}
341 	}
342 	if (bi_m != NULL)
343 		return (bi_m);
344 
345 	/* No known brand, see if the header is recognized by any brand */
346 	for (i = 0; i < MAX_BRANDS; i++) {
347 		bi = elf_brand_list[i];
348 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
349 		    bi->header_supported == NULL)
350 			continue;
351 		if (hdr->e_machine == bi->machine) {
352 			ret = bi->header_supported(imgp);
353 			if (ret)
354 				return (bi);
355 		}
356 	}
357 
358 	/* Lacking a known brand, search for a recognized interpreter. */
359 	if (interp != NULL) {
360 		for (i = 0; i < MAX_BRANDS; i++) {
361 			bi = elf_brand_list[i];
362 			if (bi == NULL || (bi->flags &
363 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
364 			    != 0)
365 				continue;
366 			if (hdr->e_machine == bi->machine &&
367 			    bi->interp_path != NULL &&
368 			    /* ELF image p_filesz includes terminating zero */
369 			    strlen(bi->interp_path) + 1 == interp_name_len &&
370 			    strncmp(interp, bi->interp_path, interp_name_len)
371 			    == 0)
372 				return (bi);
373 		}
374 	}
375 
376 	/* Lacking a recognized interpreter, try the default brand */
377 	for (i = 0; i < MAX_BRANDS; i++) {
378 		bi = elf_brand_list[i];
379 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
380 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
381 			continue;
382 		if (hdr->e_machine == bi->machine &&
383 		    __elfN(fallback_brand) == bi->brand)
384 			return (bi);
385 	}
386 	return (NULL);
387 }
388 
389 static int
390 __elfN(check_header)(const Elf_Ehdr *hdr)
391 {
392 	Elf_Brandinfo *bi;
393 	int i;
394 
395 	if (!IS_ELF(*hdr) ||
396 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
397 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
398 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
399 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
400 	    hdr->e_version != ELF_TARG_VER)
401 		return (ENOEXEC);
402 
403 	/*
404 	 * Make sure we have at least one brand for this machine.
405 	 */
406 
407 	for (i = 0; i < MAX_BRANDS; i++) {
408 		bi = elf_brand_list[i];
409 		if (bi != NULL && bi->machine == hdr->e_machine)
410 			break;
411 	}
412 	if (i == MAX_BRANDS)
413 		return (ENOEXEC);
414 
415 	return (0);
416 }
417 
418 static int
419 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
420     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
421 {
422 	struct sf_buf *sf;
423 	int error;
424 	vm_offset_t off;
425 
426 	/*
427 	 * Create the page if it doesn't exist yet. Ignore errors.
428 	 */
429 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
430 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
431 
432 	/*
433 	 * Find the page from the underlying object.
434 	 */
435 	if (object != NULL) {
436 		sf = vm_imgact_map_page(object, offset);
437 		if (sf == NULL)
438 			return (KERN_FAILURE);
439 		off = offset - trunc_page(offset);
440 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
441 		    end - start);
442 		vm_imgact_unmap_page(sf);
443 		if (error != 0)
444 			return (KERN_FAILURE);
445 	}
446 
447 	return (KERN_SUCCESS);
448 }
449 
450 static int
451 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
452     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
453     int cow)
454 {
455 	struct sf_buf *sf;
456 	vm_offset_t off;
457 	vm_size_t sz;
458 	int error, locked, rv;
459 
460 	if (start != trunc_page(start)) {
461 		rv = __elfN(map_partial)(map, object, offset, start,
462 		    round_page(start), prot);
463 		if (rv != KERN_SUCCESS)
464 			return (rv);
465 		offset += round_page(start) - start;
466 		start = round_page(start);
467 	}
468 	if (end != round_page(end)) {
469 		rv = __elfN(map_partial)(map, object, offset +
470 		    trunc_page(end) - start, trunc_page(end), end, prot);
471 		if (rv != KERN_SUCCESS)
472 			return (rv);
473 		end = trunc_page(end);
474 	}
475 	if (start >= end)
476 		return (KERN_SUCCESS);
477 	if ((offset & PAGE_MASK) != 0) {
478 		/*
479 		 * The mapping is not page aligned.  This means that we have
480 		 * to copy the data.
481 		 */
482 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
483 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
484 		if (rv != KERN_SUCCESS)
485 			return (rv);
486 		if (object == NULL)
487 			return (KERN_SUCCESS);
488 		for (; start < end; start += sz) {
489 			sf = vm_imgact_map_page(object, offset);
490 			if (sf == NULL)
491 				return (KERN_FAILURE);
492 			off = offset - trunc_page(offset);
493 			sz = end - start;
494 			if (sz > PAGE_SIZE - off)
495 				sz = PAGE_SIZE - off;
496 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
497 			    (caddr_t)start, sz);
498 			vm_imgact_unmap_page(sf);
499 			if (error != 0)
500 				return (KERN_FAILURE);
501 			offset += sz;
502 		}
503 	} else {
504 		vm_object_reference(object);
505 		rv = vm_map_fixed(map, object, offset, start, end - start,
506 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
507 		if (rv != KERN_SUCCESS) {
508 			locked = VOP_ISLOCKED(imgp->vp);
509 			VOP_UNLOCK(imgp->vp, 0);
510 			vm_object_deallocate(object);
511 			vn_lock(imgp->vp, locked | LK_RETRY);
512 			return (rv);
513 		}
514 	}
515 	return (KERN_SUCCESS);
516 }
517 
518 static int
519 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
520     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
521     size_t pagesize)
522 {
523 	struct sf_buf *sf;
524 	size_t map_len;
525 	vm_map_t map;
526 	vm_object_t object;
527 	vm_offset_t off, map_addr;
528 	int error, rv, cow;
529 	size_t copy_len;
530 	vm_ooffset_t file_addr;
531 
532 	/*
533 	 * It's necessary to fail if the filsz + offset taken from the
534 	 * header is greater than the actual file pager object's size.
535 	 * If we were to allow this, then the vm_map_find() below would
536 	 * walk right off the end of the file object and into the ether.
537 	 *
538 	 * While I'm here, might as well check for something else that
539 	 * is invalid: filsz cannot be greater than memsz.
540 	 */
541 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
542 	    filsz > memsz) {
543 		uprintf("elf_load_section: truncated ELF file\n");
544 		return (ENOEXEC);
545 	}
546 
547 	object = imgp->object;
548 	map = &imgp->proc->p_vmspace->vm_map;
549 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
550 	file_addr = trunc_page_ps(offset, pagesize);
551 
552 	/*
553 	 * We have two choices.  We can either clear the data in the last page
554 	 * of an oversized mapping, or we can start the anon mapping a page
555 	 * early and copy the initialized data into that first page.  We
556 	 * choose the second.
557 	 */
558 	if (filsz == 0)
559 		map_len = 0;
560 	else if (memsz > filsz)
561 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
562 	else
563 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
564 
565 	if (map_len != 0) {
566 		/* cow flags: don't dump readonly sections in core */
567 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
568 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
569 
570 		rv = __elfN(map_insert)(imgp, map,
571 				      object,
572 				      file_addr,	/* file offset */
573 				      map_addr,		/* virtual start */
574 				      map_addr + map_len,/* virtual end */
575 				      prot,
576 				      cow);
577 		if (rv != KERN_SUCCESS)
578 			return (EINVAL);
579 
580 		/* we can stop now if we've covered it all */
581 		if (memsz == filsz)
582 			return (0);
583 	}
584 
585 
586 	/*
587 	 * We have to get the remaining bit of the file into the first part
588 	 * of the oversized map segment.  This is normally because the .data
589 	 * segment in the file is extended to provide bss.  It's a neat idea
590 	 * to try and save a page, but it's a pain in the behind to implement.
591 	 */
592 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
593 	    filsz, pagesize);
594 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
595 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
596 	    map_addr;
597 
598 	/* This had damn well better be true! */
599 	if (map_len != 0) {
600 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
601 		    map_addr + map_len, prot, 0);
602 		if (rv != KERN_SUCCESS)
603 			return (EINVAL);
604 	}
605 
606 	if (copy_len != 0) {
607 		sf = vm_imgact_map_page(object, offset + filsz);
608 		if (sf == NULL)
609 			return (EIO);
610 
611 		/* send the page fragment to user space */
612 		off = trunc_page_ps(offset + filsz, pagesize) -
613 		    trunc_page(offset + filsz);
614 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
615 		    (caddr_t)map_addr, copy_len);
616 		vm_imgact_unmap_page(sf);
617 		if (error != 0)
618 			return (error);
619 	}
620 
621 	/*
622 	 * Remove write access to the page if it was only granted by map_insert
623 	 * to allow copyout.
624 	 */
625 	if ((prot & VM_PROT_WRITE) == 0)
626 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
627 		    map_len), prot, FALSE);
628 
629 	return (0);
630 }
631 
632 /*
633  * Load the file "file" into memory.  It may be either a shared object
634  * or an executable.
635  *
636  * The "addr" reference parameter is in/out.  On entry, it specifies
637  * the address where a shared object should be loaded.  If the file is
638  * an executable, this value is ignored.  On exit, "addr" specifies
639  * where the file was actually loaded.
640  *
641  * The "entry" reference parameter is out only.  On exit, it specifies
642  * the entry point for the loaded file.
643  */
644 static int
645 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
646 	u_long *entry, size_t pagesize)
647 {
648 	struct {
649 		struct nameidata nd;
650 		struct vattr attr;
651 		struct image_params image_params;
652 	} *tempdata;
653 	const Elf_Ehdr *hdr = NULL;
654 	const Elf_Phdr *phdr = NULL;
655 	struct nameidata *nd;
656 	struct vattr *attr;
657 	struct image_params *imgp;
658 	vm_prot_t prot;
659 	u_long rbase;
660 	u_long base_addr = 0;
661 	int error, i, numsegs;
662 
663 #ifdef CAPABILITY_MODE
664 	/*
665 	 * XXXJA: This check can go away once we are sufficiently confident
666 	 * that the checks in namei() are correct.
667 	 */
668 	if (IN_CAPABILITY_MODE(curthread))
669 		return (ECAPMODE);
670 #endif
671 
672 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
673 	nd = &tempdata->nd;
674 	attr = &tempdata->attr;
675 	imgp = &tempdata->image_params;
676 
677 	/*
678 	 * Initialize part of the common data
679 	 */
680 	imgp->proc = p;
681 	imgp->attr = attr;
682 	imgp->firstpage = NULL;
683 	imgp->image_header = NULL;
684 	imgp->object = NULL;
685 	imgp->execlabel = NULL;
686 
687 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
688 	if ((error = namei(nd)) != 0) {
689 		nd->ni_vp = NULL;
690 		goto fail;
691 	}
692 	NDFREE(nd, NDF_ONLY_PNBUF);
693 	imgp->vp = nd->ni_vp;
694 
695 	/*
696 	 * Check permissions, modes, uid, etc on the file, and "open" it.
697 	 */
698 	error = exec_check_permissions(imgp);
699 	if (error)
700 		goto fail;
701 
702 	error = exec_map_first_page(imgp);
703 	if (error)
704 		goto fail;
705 
706 	/*
707 	 * Also make certain that the interpreter stays the same, so set
708 	 * its VV_TEXT flag, too.
709 	 */
710 	VOP_SET_TEXT(nd->ni_vp);
711 
712 	imgp->object = nd->ni_vp->v_object;
713 
714 	hdr = (const Elf_Ehdr *)imgp->image_header;
715 	if ((error = __elfN(check_header)(hdr)) != 0)
716 		goto fail;
717 	if (hdr->e_type == ET_DYN)
718 		rbase = *addr;
719 	else if (hdr->e_type == ET_EXEC)
720 		rbase = 0;
721 	else {
722 		error = ENOEXEC;
723 		goto fail;
724 	}
725 
726 	/* Only support headers that fit within first page for now      */
727 	if ((hdr->e_phoff > PAGE_SIZE) ||
728 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
729 		error = ENOEXEC;
730 		goto fail;
731 	}
732 
733 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
734 	if (!aligned(phdr, Elf_Addr)) {
735 		error = ENOEXEC;
736 		goto fail;
737 	}
738 
739 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
740 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
741 			/* Loadable segment */
742 			prot = __elfN(trans_prot)(phdr[i].p_flags);
743 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
744 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
745 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
746 			if (error != 0)
747 				goto fail;
748 			/*
749 			 * Establish the base address if this is the
750 			 * first segment.
751 			 */
752 			if (numsegs == 0)
753   				base_addr = trunc_page(phdr[i].p_vaddr +
754 				    rbase);
755 			numsegs++;
756 		}
757 	}
758 	*addr = base_addr;
759 	*entry = (unsigned long)hdr->e_entry + rbase;
760 
761 fail:
762 	if (imgp->firstpage)
763 		exec_unmap_first_page(imgp);
764 
765 	if (nd->ni_vp)
766 		vput(nd->ni_vp);
767 
768 	free(tempdata, M_TEMP);
769 
770 	return (error);
771 }
772 
773 static int
774 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
775 {
776 	struct thread *td;
777 	const Elf_Ehdr *hdr;
778 	const Elf_Phdr *phdr;
779 	Elf_Auxargs *elf_auxargs;
780 	struct vmspace *vmspace;
781 	const char *err_str, *newinterp;
782 	char *interp, *interp_buf, *path;
783 	Elf_Brandinfo *brand_info;
784 	struct sysentvec *sv;
785 	vm_prot_t prot;
786 	u_long text_size, data_size, total_size, text_addr, data_addr;
787 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
788 	int32_t osrel;
789 	int error, i, n, interp_name_len, have_interp;
790 
791 	hdr = (const Elf_Ehdr *)imgp->image_header;
792 
793 	/*
794 	 * Do we have a valid ELF header ?
795 	 *
796 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
797 	 * if particular brand doesn't support it.
798 	 */
799 	if (__elfN(check_header)(hdr) != 0 ||
800 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
801 		return (-1);
802 
803 	/*
804 	 * From here on down, we return an errno, not -1, as we've
805 	 * detected an ELF file.
806 	 */
807 
808 	if ((hdr->e_phoff > PAGE_SIZE) ||
809 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
810 		/* Only support headers in first page for now */
811 		uprintf("Program headers not in the first page\n");
812 		return (ENOEXEC);
813 	}
814 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
815 	if (!aligned(phdr, Elf_Addr)) {
816 		uprintf("Unaligned program headers\n");
817 		return (ENOEXEC);
818 	}
819 
820 	n = error = 0;
821 	baddr = 0;
822 	osrel = 0;
823 	text_size = data_size = total_size = text_addr = data_addr = 0;
824 	entry = proghdr = 0;
825 	interp_name_len = 0;
826 	err_str = newinterp = NULL;
827 	interp = interp_buf = NULL;
828 	td = curthread;
829 
830 	for (i = 0; i < hdr->e_phnum; i++) {
831 		switch (phdr[i].p_type) {
832 		case PT_LOAD:
833 			if (n == 0)
834 				baddr = phdr[i].p_vaddr;
835 			n++;
836 			break;
837 		case PT_INTERP:
838 			/* Path to interpreter */
839 			if (phdr[i].p_filesz > MAXPATHLEN) {
840 				uprintf("Invalid PT_INTERP\n");
841 				error = ENOEXEC;
842 				goto ret;
843 			}
844 			if (interp != NULL) {
845 				uprintf("Multiple PT_INTERP headers\n");
846 				error = ENOEXEC;
847 				goto ret;
848 			}
849 			interp_name_len = phdr[i].p_filesz;
850 			if (phdr[i].p_offset > PAGE_SIZE ||
851 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
852 				VOP_UNLOCK(imgp->vp, 0);
853 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
854 				    M_WAITOK);
855 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
856 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
857 				    interp_name_len, phdr[i].p_offset,
858 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
859 				    NOCRED, NULL, td);
860 				if (error != 0) {
861 					uprintf("i/o error PT_INTERP\n");
862 					goto ret;
863 				}
864 				interp_buf[interp_name_len] = '\0';
865 				interp = interp_buf;
866 			} else {
867 				interp = __DECONST(char *, imgp->image_header) +
868 				    phdr[i].p_offset;
869 			}
870 			break;
871 		case PT_GNU_STACK:
872 			if (__elfN(nxstack))
873 				imgp->stack_prot =
874 				    __elfN(trans_prot)(phdr[i].p_flags);
875 			imgp->stack_sz = phdr[i].p_memsz;
876 			break;
877 		}
878 	}
879 
880 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
881 	    &osrel);
882 	if (brand_info == NULL) {
883 		uprintf("ELF binary type \"%u\" not known.\n",
884 		    hdr->e_ident[EI_OSABI]);
885 		error = ENOEXEC;
886 		goto ret;
887 	}
888 	et_dyn_addr = 0;
889 	if (hdr->e_type == ET_DYN) {
890 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
891 			uprintf("Cannot execute shared object\n");
892 			error = ENOEXEC;
893 			goto ret;
894 		}
895 		/*
896 		 * Honour the base load address from the dso if it is
897 		 * non-zero for some reason.
898 		 */
899 		if (baddr == 0)
900 			et_dyn_addr = ET_DYN_LOAD_ADDR;
901 	}
902 	sv = brand_info->sysvec;
903 	if (interp != NULL && brand_info->interp_newpath != NULL)
904 		newinterp = brand_info->interp_newpath;
905 
906 	/*
907 	 * Avoid a possible deadlock if the current address space is destroyed
908 	 * and that address space maps the locked vnode.  In the common case,
909 	 * the locked vnode's v_usecount is decremented but remains greater
910 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
911 	 * However, in cases where the vnode lock is external, such as nullfs,
912 	 * v_usecount may become zero.
913 	 *
914 	 * The VV_TEXT flag prevents modifications to the executable while
915 	 * the vnode is unlocked.
916 	 */
917 	VOP_UNLOCK(imgp->vp, 0);
918 
919 	error = exec_new_vmspace(imgp, sv);
920 	imgp->proc->p_sysent = sv;
921 
922 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
923 	if (error != 0)
924 		goto ret;
925 
926 	for (i = 0; i < hdr->e_phnum; i++) {
927 		switch (phdr[i].p_type) {
928 		case PT_LOAD:	/* Loadable segment */
929 			if (phdr[i].p_memsz == 0)
930 				break;
931 			prot = __elfN(trans_prot)(phdr[i].p_flags);
932 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
933 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
934 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
935 			    sv->sv_pagesize);
936 			if (error != 0)
937 				goto ret;
938 
939 			/*
940 			 * If this segment contains the program headers,
941 			 * remember their virtual address for the AT_PHDR
942 			 * aux entry. Static binaries don't usually include
943 			 * a PT_PHDR entry.
944 			 */
945 			if (phdr[i].p_offset == 0 &&
946 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
947 				<= phdr[i].p_filesz)
948 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
949 				    et_dyn_addr;
950 
951 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
952 			seg_size = round_page(phdr[i].p_memsz +
953 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
954 
955 			/*
956 			 * Make the largest executable segment the official
957 			 * text segment and all others data.
958 			 *
959 			 * Note that obreak() assumes that data_addr +
960 			 * data_size == end of data load area, and the ELF
961 			 * file format expects segments to be sorted by
962 			 * address.  If multiple data segments exist, the
963 			 * last one will be used.
964 			 */
965 
966 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
967 				text_size = seg_size;
968 				text_addr = seg_addr;
969 			} else {
970 				data_size = seg_size;
971 				data_addr = seg_addr;
972 			}
973 			total_size += seg_size;
974 			break;
975 		case PT_PHDR: 	/* Program header table info */
976 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
977 			break;
978 		default:
979 			break;
980 		}
981 	}
982 
983 	if (data_addr == 0 && data_size == 0) {
984 		data_addr = text_addr;
985 		data_size = text_size;
986 	}
987 
988 	entry = (u_long)hdr->e_entry + et_dyn_addr;
989 
990 	/*
991 	 * Check limits.  It should be safe to check the
992 	 * limits after loading the segments since we do
993 	 * not actually fault in all the segments pages.
994 	 */
995 	PROC_LOCK(imgp->proc);
996 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
997 		err_str = "Data segment size exceeds process limit";
998 	else if (text_size > maxtsiz)
999 		err_str = "Text segment size exceeds system limit";
1000 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1001 		err_str = "Total segment size exceeds process limit";
1002 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1003 		err_str = "Data segment size exceeds resource limit";
1004 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1005 		err_str = "Total segment size exceeds resource limit";
1006 	if (err_str != NULL) {
1007 		PROC_UNLOCK(imgp->proc);
1008 		uprintf("%s\n", err_str);
1009 		error = ENOMEM;
1010 		goto ret;
1011 	}
1012 
1013 	vmspace = imgp->proc->p_vmspace;
1014 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1015 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1016 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1017 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1018 
1019 	/*
1020 	 * We load the dynamic linker where a userland call
1021 	 * to mmap(0, ...) would put it.  The rationale behind this
1022 	 * calculation is that it leaves room for the heap to grow to
1023 	 * its maximum allowed size.
1024 	 */
1025 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1026 	    RLIMIT_DATA));
1027 	PROC_UNLOCK(imgp->proc);
1028 
1029 	imgp->entry_addr = entry;
1030 
1031 	if (interp != NULL) {
1032 		have_interp = FALSE;
1033 		VOP_UNLOCK(imgp->vp, 0);
1034 		if (brand_info->emul_path != NULL &&
1035 		    brand_info->emul_path[0] != '\0') {
1036 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1037 			snprintf(path, MAXPATHLEN, "%s%s",
1038 			    brand_info->emul_path, interp);
1039 			error = __elfN(load_file)(imgp->proc, path, &addr,
1040 			    &imgp->entry_addr, sv->sv_pagesize);
1041 			free(path, M_TEMP);
1042 			if (error == 0)
1043 				have_interp = TRUE;
1044 		}
1045 		if (!have_interp && newinterp != NULL &&
1046 		    (brand_info->interp_path == NULL ||
1047 		    strcmp(interp, brand_info->interp_path) == 0)) {
1048 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1049 			    &imgp->entry_addr, sv->sv_pagesize);
1050 			if (error == 0)
1051 				have_interp = TRUE;
1052 		}
1053 		if (!have_interp) {
1054 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1055 			    &imgp->entry_addr, sv->sv_pagesize);
1056 		}
1057 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1058 		if (error != 0) {
1059 			uprintf("ELF interpreter %s not found, error %d\n",
1060 			    interp, error);
1061 			goto ret;
1062 		}
1063 	} else
1064 		addr = et_dyn_addr;
1065 
1066 	/*
1067 	 * Construct auxargs table (used by the fixup routine)
1068 	 */
1069 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1070 	elf_auxargs->execfd = -1;
1071 	elf_auxargs->phdr = proghdr;
1072 	elf_auxargs->phent = hdr->e_phentsize;
1073 	elf_auxargs->phnum = hdr->e_phnum;
1074 	elf_auxargs->pagesz = PAGE_SIZE;
1075 	elf_auxargs->base = addr;
1076 	elf_auxargs->flags = 0;
1077 	elf_auxargs->entry = entry;
1078 	elf_auxargs->hdr_eflags = hdr->e_flags;
1079 
1080 	imgp->auxargs = elf_auxargs;
1081 	imgp->interpreted = 0;
1082 	imgp->reloc_base = addr;
1083 	imgp->proc->p_osrel = osrel;
1084 	imgp->proc->p_elf_machine = hdr->e_machine;
1085 	imgp->proc->p_elf_flags = hdr->e_flags;
1086 
1087 ret:
1088 	free(interp_buf, M_TEMP);
1089 	return (error);
1090 }
1091 
1092 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1093 
1094 int
1095 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1096 {
1097 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1098 	Elf_Addr *base;
1099 	Elf_Addr *pos;
1100 
1101 	base = (Elf_Addr *)*stack_base;
1102 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1103 
1104 	if (args->execfd != -1)
1105 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1106 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1107 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1108 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1109 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1110 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1111 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1112 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1113 #ifdef AT_EHDRFLAGS
1114 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1115 #endif
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 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1137 
1138 	free(imgp->auxargs, M_TEMP);
1139 	imgp->auxargs = NULL;
1140 
1141 	base--;
1142 	suword(base, (long)imgp->args->argc);
1143 	*stack_base = (register_t *)base;
1144 	return (0);
1145 }
1146 
1147 /*
1148  * Code for generating ELF core dumps.
1149  */
1150 
1151 typedef void (*segment_callback)(vm_map_entry_t, void *);
1152 
1153 /* Closure for cb_put_phdr(). */
1154 struct phdr_closure {
1155 	Elf_Phdr *phdr;		/* Program header to fill in */
1156 	Elf_Off offset;		/* Offset of segment in core file */
1157 };
1158 
1159 /* Closure for cb_size_segment(). */
1160 struct sseg_closure {
1161 	int count;		/* Count of writable segments. */
1162 	size_t size;		/* Total size of all writable segments. */
1163 };
1164 
1165 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1166 
1167 struct note_info {
1168 	int		type;		/* Note type. */
1169 	outfunc_t 	outfunc; 	/* Output function. */
1170 	void		*outarg;	/* Argument for the output function. */
1171 	size_t		outsize;	/* Output size. */
1172 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1173 };
1174 
1175 TAILQ_HEAD(note_info_list, note_info);
1176 
1177 /* Coredump output parameters. */
1178 struct coredump_params {
1179 	off_t		offset;
1180 	struct ucred	*active_cred;
1181 	struct ucred	*file_cred;
1182 	struct thread	*td;
1183 	struct vnode	*vp;
1184 	struct gzio_stream *gzs;
1185 };
1186 
1187 static void cb_put_phdr(vm_map_entry_t, void *);
1188 static void cb_size_segment(vm_map_entry_t, void *);
1189 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1190     enum uio_seg);
1191 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1192 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1193     struct note_info_list *, size_t);
1194 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1195     size_t *);
1196 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1197 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1198 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1199 static int sbuf_drain_core_output(void *, const char *, int);
1200 static int sbuf_drain_count(void *arg, const char *data, int len);
1201 
1202 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1203 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1204 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1205 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1206 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1207 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1208 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1209 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1210 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1211 static void note_procstat_files(void *, struct sbuf *, size_t *);
1212 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1213 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1214 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1215 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1216 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1217 
1218 #ifdef GZIO
1219 extern int compress_user_cores_gzlevel;
1220 
1221 /*
1222  * Write out a core segment to the compression stream.
1223  */
1224 static int
1225 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1226 {
1227 	u_int chunk_len;
1228 	int error;
1229 
1230 	while (len > 0) {
1231 		chunk_len = MIN(len, CORE_BUF_SIZE);
1232 
1233 		/*
1234 		 * We can get EFAULT error here.
1235 		 * In that case zero out the current chunk of the segment.
1236 		 */
1237 		error = copyin(base, buf, chunk_len);
1238 		if (error != 0)
1239 			bzero(buf, chunk_len);
1240 		error = gzio_write(p->gzs, buf, chunk_len);
1241 		if (error != 0)
1242 			break;
1243 		base += chunk_len;
1244 		len -= chunk_len;
1245 	}
1246 	return (error);
1247 }
1248 
1249 static int
1250 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1251 {
1252 
1253 	return (core_write((struct coredump_params *)arg, base, len, offset,
1254 	    UIO_SYSSPACE));
1255 }
1256 #endif /* GZIO */
1257 
1258 static int
1259 core_write(struct coredump_params *p, const void *base, size_t len,
1260     off_t offset, enum uio_seg seg)
1261 {
1262 
1263 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1264 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1265 	    p->active_cred, p->file_cred, NULL, p->td));
1266 }
1267 
1268 static int
1269 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1270     void *tmpbuf)
1271 {
1272 	int error;
1273 
1274 #ifdef GZIO
1275 	if (p->gzs != NULL)
1276 		return (compress_chunk(p, base, tmpbuf, len));
1277 #endif
1278 	/*
1279 	 * EFAULT is a non-fatal error that we can get, for example,
1280 	 * if the segment is backed by a file but extends beyond its
1281 	 * end.
1282 	 */
1283 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1284 	if (error == EFAULT) {
1285 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1286 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1287 		    "for process %s\n", base, len, offset, curproc->p_comm);
1288 
1289 		/*
1290 		 * Write a "real" zero byte at the end of the target region
1291 		 * in the case this is the last segment.
1292 		 * The intermediate space will be implicitly zero-filled.
1293 		 */
1294 		error = core_write(p, zero_region, 1, offset + len - 1,
1295 		    UIO_SYSSPACE);
1296 	}
1297 	return (error);
1298 }
1299 
1300 /*
1301  * Drain into a core file.
1302  */
1303 static int
1304 sbuf_drain_core_output(void *arg, const char *data, int len)
1305 {
1306 	struct coredump_params *p;
1307 	int error, locked;
1308 
1309 	p = (struct coredump_params *)arg;
1310 
1311 	/*
1312 	 * Some kern_proc out routines that print to this sbuf may
1313 	 * call us with the process lock held. Draining with the
1314 	 * non-sleepable lock held is unsafe. The lock is needed for
1315 	 * those routines when dumping a live process. In our case we
1316 	 * can safely release the lock before draining and acquire
1317 	 * again after.
1318 	 */
1319 	locked = PROC_LOCKED(p->td->td_proc);
1320 	if (locked)
1321 		PROC_UNLOCK(p->td->td_proc);
1322 #ifdef GZIO
1323 	if (p->gzs != NULL)
1324 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1325 	else
1326 #endif
1327 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1328 		    UIO_SYSSPACE);
1329 	if (locked)
1330 		PROC_LOCK(p->td->td_proc);
1331 	if (error != 0)
1332 		return (-error);
1333 	p->offset += len;
1334 	return (len);
1335 }
1336 
1337 /*
1338  * Drain into a counter.
1339  */
1340 static int
1341 sbuf_drain_count(void *arg, const char *data __unused, int len)
1342 {
1343 	size_t *sizep;
1344 
1345 	sizep = (size_t *)arg;
1346 	*sizep += len;
1347 	return (len);
1348 }
1349 
1350 int
1351 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1352 {
1353 	struct ucred *cred = td->td_ucred;
1354 	int error = 0;
1355 	struct sseg_closure seginfo;
1356 	struct note_info_list notelst;
1357 	struct coredump_params params;
1358 	struct note_info *ninfo;
1359 	void *hdr, *tmpbuf;
1360 	size_t hdrsize, notesz, coresize;
1361 #ifdef GZIO
1362 	boolean_t compress;
1363 
1364 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1365 #endif
1366 	hdr = NULL;
1367 	tmpbuf = NULL;
1368 	TAILQ_INIT(&notelst);
1369 
1370 	/* Size the program segments. */
1371 	seginfo.count = 0;
1372 	seginfo.size = 0;
1373 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1374 
1375 	/*
1376 	 * Collect info about the core file header area.
1377 	 */
1378 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1379 	if (seginfo.count + 1 >= PN_XNUM)
1380 		hdrsize += sizeof(Elf_Shdr);
1381 	__elfN(prepare_notes)(td, &notelst, &notesz);
1382 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1383 
1384 	/* Set up core dump parameters. */
1385 	params.offset = 0;
1386 	params.active_cred = cred;
1387 	params.file_cred = NOCRED;
1388 	params.td = td;
1389 	params.vp = vp;
1390 	params.gzs = NULL;
1391 
1392 #ifdef RACCT
1393 	if (racct_enable) {
1394 		PROC_LOCK(td->td_proc);
1395 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1396 		PROC_UNLOCK(td->td_proc);
1397 		if (error != 0) {
1398 			error = EFAULT;
1399 			goto done;
1400 		}
1401 	}
1402 #endif
1403 	if (coresize >= limit) {
1404 		error = EFAULT;
1405 		goto done;
1406 	}
1407 
1408 #ifdef GZIO
1409 	/* Create a compression stream if necessary. */
1410 	if (compress) {
1411 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1412 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1413 		if (params.gzs == NULL) {
1414 			error = EFAULT;
1415 			goto done;
1416 		}
1417 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1418         }
1419 #endif
1420 
1421 	/*
1422 	 * Allocate memory for building the header, fill it up,
1423 	 * and write it out following the notes.
1424 	 */
1425 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1426 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1427 	    notesz);
1428 
1429 	/* Write the contents of all of the writable segments. */
1430 	if (error == 0) {
1431 		Elf_Phdr *php;
1432 		off_t offset;
1433 		int i;
1434 
1435 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1436 		offset = round_page(hdrsize + notesz);
1437 		for (i = 0; i < seginfo.count; i++) {
1438 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1439 			    php->p_filesz, offset, &params, tmpbuf);
1440 			if (error != 0)
1441 				break;
1442 			offset += php->p_filesz;
1443 			php++;
1444 		}
1445 #ifdef GZIO
1446 		if (error == 0 && compress)
1447 			error = gzio_flush(params.gzs);
1448 #endif
1449 	}
1450 	if (error) {
1451 		log(LOG_WARNING,
1452 		    "Failed to write core file for process %s (error %d)\n",
1453 		    curproc->p_comm, error);
1454 	}
1455 
1456 done:
1457 #ifdef GZIO
1458 	if (compress) {
1459 		free(tmpbuf, M_TEMP);
1460 		if (params.gzs != NULL)
1461 			gzio_fini(params.gzs);
1462 	}
1463 #endif
1464 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1465 		TAILQ_REMOVE(&notelst, ninfo, link);
1466 		free(ninfo, M_TEMP);
1467 	}
1468 	if (hdr != NULL)
1469 		free(hdr, M_TEMP);
1470 
1471 	return (error);
1472 }
1473 
1474 /*
1475  * A callback for each_dumpable_segment() to write out the segment's
1476  * program header entry.
1477  */
1478 static void
1479 cb_put_phdr(entry, closure)
1480 	vm_map_entry_t entry;
1481 	void *closure;
1482 {
1483 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1484 	Elf_Phdr *phdr = phc->phdr;
1485 
1486 	phc->offset = round_page(phc->offset);
1487 
1488 	phdr->p_type = PT_LOAD;
1489 	phdr->p_offset = phc->offset;
1490 	phdr->p_vaddr = entry->start;
1491 	phdr->p_paddr = 0;
1492 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1493 	phdr->p_align = PAGE_SIZE;
1494 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1495 
1496 	phc->offset += phdr->p_filesz;
1497 	phc->phdr++;
1498 }
1499 
1500 /*
1501  * A callback for each_dumpable_segment() to gather information about
1502  * the number of segments and their total size.
1503  */
1504 static void
1505 cb_size_segment(vm_map_entry_t entry, void *closure)
1506 {
1507 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1508 
1509 	ssc->count++;
1510 	ssc->size += entry->end - entry->start;
1511 }
1512 
1513 /*
1514  * For each writable segment in the process's memory map, call the given
1515  * function with a pointer to the map entry and some arbitrary
1516  * caller-supplied data.
1517  */
1518 static void
1519 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1520 {
1521 	struct proc *p = td->td_proc;
1522 	vm_map_t map = &p->p_vmspace->vm_map;
1523 	vm_map_entry_t entry;
1524 	vm_object_t backing_object, object;
1525 	boolean_t ignore_entry;
1526 
1527 	vm_map_lock_read(map);
1528 	for (entry = map->header.next; entry != &map->header;
1529 	    entry = entry->next) {
1530 		/*
1531 		 * Don't dump inaccessible mappings, deal with legacy
1532 		 * coredump mode.
1533 		 *
1534 		 * Note that read-only segments related to the elf binary
1535 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1536 		 * need to arbitrarily ignore such segments.
1537 		 */
1538 		if (elf_legacy_coredump) {
1539 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1540 				continue;
1541 		} else {
1542 			if ((entry->protection & VM_PROT_ALL) == 0)
1543 				continue;
1544 		}
1545 
1546 		/*
1547 		 * Dont include memory segment in the coredump if
1548 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1549 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1550 		 * kernel map).
1551 		 */
1552 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1553 			continue;
1554 
1555 		if ((object = entry->object.vm_object) == NULL)
1556 			continue;
1557 
1558 		/* Ignore memory-mapped devices and such things. */
1559 		VM_OBJECT_RLOCK(object);
1560 		while ((backing_object = object->backing_object) != NULL) {
1561 			VM_OBJECT_RLOCK(backing_object);
1562 			VM_OBJECT_RUNLOCK(object);
1563 			object = backing_object;
1564 		}
1565 		ignore_entry = object->type != OBJT_DEFAULT &&
1566 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1567 		    object->type != OBJT_PHYS;
1568 		VM_OBJECT_RUNLOCK(object);
1569 		if (ignore_entry)
1570 			continue;
1571 
1572 		(*func)(entry, closure);
1573 	}
1574 	vm_map_unlock_read(map);
1575 }
1576 
1577 /*
1578  * Write the core file header to the file, including padding up to
1579  * the page boundary.
1580  */
1581 static int
1582 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1583     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1584 {
1585 	struct note_info *ninfo;
1586 	struct sbuf *sb;
1587 	int error;
1588 
1589 	/* Fill in the header. */
1590 	bzero(hdr, hdrsize);
1591 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1592 
1593 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1594 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1595 	sbuf_start_section(sb, NULL);
1596 	sbuf_bcat(sb, hdr, hdrsize);
1597 	TAILQ_FOREACH(ninfo, notelst, link)
1598 	    __elfN(putnote)(ninfo, sb);
1599 	/* Align up to a page boundary for the program segments. */
1600 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1601 	error = sbuf_finish(sb);
1602 	sbuf_delete(sb);
1603 
1604 	return (error);
1605 }
1606 
1607 static void
1608 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1609     size_t *sizep)
1610 {
1611 	struct proc *p;
1612 	struct thread *thr;
1613 	size_t size;
1614 
1615 	p = td->td_proc;
1616 	size = 0;
1617 
1618 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1619 
1620 	/*
1621 	 * To have the debugger select the right thread (LWP) as the initial
1622 	 * thread, we dump the state of the thread passed to us in td first.
1623 	 * This is the thread that causes the core dump and thus likely to
1624 	 * be the right thread one wants to have selected in the debugger.
1625 	 */
1626 	thr = td;
1627 	while (thr != NULL) {
1628 		size += register_note(list, NT_PRSTATUS,
1629 		    __elfN(note_prstatus), thr);
1630 		size += register_note(list, NT_FPREGSET,
1631 		    __elfN(note_fpregset), thr);
1632 		size += register_note(list, NT_THRMISC,
1633 		    __elfN(note_thrmisc), thr);
1634 		size += register_note(list, NT_PTLWPINFO,
1635 		    __elfN(note_ptlwpinfo), thr);
1636 		size += register_note(list, -1,
1637 		    __elfN(note_threadmd), thr);
1638 
1639 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1640 		    TAILQ_NEXT(thr, td_plist);
1641 		if (thr == td)
1642 			thr = TAILQ_NEXT(thr, td_plist);
1643 	}
1644 
1645 	size += register_note(list, NT_PROCSTAT_PROC,
1646 	    __elfN(note_procstat_proc), p);
1647 	size += register_note(list, NT_PROCSTAT_FILES,
1648 	    note_procstat_files, p);
1649 	size += register_note(list, NT_PROCSTAT_VMMAP,
1650 	    note_procstat_vmmap, p);
1651 	size += register_note(list, NT_PROCSTAT_GROUPS,
1652 	    note_procstat_groups, p);
1653 	size += register_note(list, NT_PROCSTAT_UMASK,
1654 	    note_procstat_umask, p);
1655 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1656 	    note_procstat_rlimit, p);
1657 	size += register_note(list, NT_PROCSTAT_OSREL,
1658 	    note_procstat_osrel, p);
1659 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1660 	    __elfN(note_procstat_psstrings), p);
1661 	size += register_note(list, NT_PROCSTAT_AUXV,
1662 	    __elfN(note_procstat_auxv), p);
1663 
1664 	*sizep = size;
1665 }
1666 
1667 static void
1668 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1669     size_t notesz)
1670 {
1671 	Elf_Ehdr *ehdr;
1672 	Elf_Phdr *phdr;
1673 	Elf_Shdr *shdr;
1674 	struct phdr_closure phc;
1675 
1676 	ehdr = (Elf_Ehdr *)hdr;
1677 
1678 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1679 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1680 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1681 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1682 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1683 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1684 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1685 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1686 	ehdr->e_ident[EI_ABIVERSION] = 0;
1687 	ehdr->e_ident[EI_PAD] = 0;
1688 	ehdr->e_type = ET_CORE;
1689 	ehdr->e_machine = td->td_proc->p_elf_machine;
1690 	ehdr->e_version = EV_CURRENT;
1691 	ehdr->e_entry = 0;
1692 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1693 	ehdr->e_flags = td->td_proc->p_elf_flags;
1694 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1695 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1696 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1697 	ehdr->e_shstrndx = SHN_UNDEF;
1698 	if (numsegs + 1 < PN_XNUM) {
1699 		ehdr->e_phnum = numsegs + 1;
1700 		ehdr->e_shnum = 0;
1701 	} else {
1702 		ehdr->e_phnum = PN_XNUM;
1703 		ehdr->e_shnum = 1;
1704 
1705 		ehdr->e_shoff = ehdr->e_phoff +
1706 		    (numsegs + 1) * ehdr->e_phentsize;
1707 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1708 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1709 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1710 
1711 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1712 		memset(shdr, 0, sizeof(*shdr));
1713 		/*
1714 		 * A special first section is used to hold large segment and
1715 		 * section counts.  This was proposed by Sun Microsystems in
1716 		 * Solaris and has been adopted by Linux; the standard ELF
1717 		 * tools are already familiar with the technique.
1718 		 *
1719 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1720 		 * (or 12-7 depending on the version of the document) for more
1721 		 * details.
1722 		 */
1723 		shdr->sh_type = SHT_NULL;
1724 		shdr->sh_size = ehdr->e_shnum;
1725 		shdr->sh_link = ehdr->e_shstrndx;
1726 		shdr->sh_info = numsegs + 1;
1727 	}
1728 
1729 	/*
1730 	 * Fill in the program header entries.
1731 	 */
1732 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1733 
1734 	/* The note segement. */
1735 	phdr->p_type = PT_NOTE;
1736 	phdr->p_offset = hdrsize;
1737 	phdr->p_vaddr = 0;
1738 	phdr->p_paddr = 0;
1739 	phdr->p_filesz = notesz;
1740 	phdr->p_memsz = 0;
1741 	phdr->p_flags = PF_R;
1742 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1743 	phdr++;
1744 
1745 	/* All the writable segments from the program. */
1746 	phc.phdr = phdr;
1747 	phc.offset = round_page(hdrsize + notesz);
1748 	each_dumpable_segment(td, cb_put_phdr, &phc);
1749 }
1750 
1751 static size_t
1752 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1753 {
1754 	struct note_info *ninfo;
1755 	size_t size, notesize;
1756 
1757 	size = 0;
1758 	out(arg, NULL, &size);
1759 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1760 	ninfo->type = type;
1761 	ninfo->outfunc = out;
1762 	ninfo->outarg = arg;
1763 	ninfo->outsize = size;
1764 	TAILQ_INSERT_TAIL(list, ninfo, link);
1765 
1766 	if (type == -1)
1767 		return (size);
1768 
1769 	notesize = sizeof(Elf_Note) +		/* note header */
1770 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1771 						/* note name */
1772 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1773 
1774 	return (notesize);
1775 }
1776 
1777 static size_t
1778 append_note_data(const void *src, void *dst, size_t len)
1779 {
1780 	size_t padded_len;
1781 
1782 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1783 	if (dst != NULL) {
1784 		bcopy(src, dst, len);
1785 		bzero((char *)dst + len, padded_len - len);
1786 	}
1787 	return (padded_len);
1788 }
1789 
1790 size_t
1791 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1792 {
1793 	Elf_Note *note;
1794 	char *buf;
1795 	size_t notesize;
1796 
1797 	buf = dst;
1798 	if (buf != NULL) {
1799 		note = (Elf_Note *)buf;
1800 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1801 		note->n_descsz = size;
1802 		note->n_type = type;
1803 		buf += sizeof(*note);
1804 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1805 		    sizeof(FREEBSD_ABI_VENDOR));
1806 		append_note_data(src, buf, size);
1807 		if (descp != NULL)
1808 			*descp = buf;
1809 	}
1810 
1811 	notesize = sizeof(Elf_Note) +		/* note header */
1812 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1813 						/* note name */
1814 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1815 
1816 	return (notesize);
1817 }
1818 
1819 static void
1820 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1821 {
1822 	Elf_Note note;
1823 	ssize_t old_len, sect_len;
1824 	size_t new_len, descsz, i;
1825 
1826 	if (ninfo->type == -1) {
1827 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1828 		return;
1829 	}
1830 
1831 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1832 	note.n_descsz = ninfo->outsize;
1833 	note.n_type = ninfo->type;
1834 
1835 	sbuf_bcat(sb, &note, sizeof(note));
1836 	sbuf_start_section(sb, &old_len);
1837 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1838 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1839 	if (note.n_descsz == 0)
1840 		return;
1841 	sbuf_start_section(sb, &old_len);
1842 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1843 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1844 	if (sect_len < 0)
1845 		return;
1846 
1847 	new_len = (size_t)sect_len;
1848 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1849 	if (new_len < descsz) {
1850 		/*
1851 		 * It is expected that individual note emitters will correctly
1852 		 * predict their expected output size and fill up to that size
1853 		 * themselves, padding in a format-specific way if needed.
1854 		 * However, in case they don't, just do it here with zeros.
1855 		 */
1856 		for (i = 0; i < descsz - new_len; i++)
1857 			sbuf_putc(sb, 0);
1858 	} else if (new_len > descsz) {
1859 		/*
1860 		 * We can't always truncate sb -- we may have drained some
1861 		 * of it already.
1862 		 */
1863 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1864 		    "read it (%zu > %zu).  Since it is longer than "
1865 		    "expected, this coredump's notes are corrupt.  THIS "
1866 		    "IS A BUG in the note_procstat routine for type %u.\n",
1867 		    __func__, (unsigned)note.n_type, new_len, descsz,
1868 		    (unsigned)note.n_type));
1869 	}
1870 }
1871 
1872 /*
1873  * Miscellaneous note out functions.
1874  */
1875 
1876 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1877 #include <compat/freebsd32/freebsd32.h>
1878 
1879 typedef struct prstatus32 elf_prstatus_t;
1880 typedef struct prpsinfo32 elf_prpsinfo_t;
1881 typedef struct fpreg32 elf_prfpregset_t;
1882 typedef struct fpreg32 elf_fpregset_t;
1883 typedef struct reg32 elf_gregset_t;
1884 typedef struct thrmisc32 elf_thrmisc_t;
1885 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1886 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1887 typedef uint32_t elf_ps_strings_t;
1888 #else
1889 typedef prstatus_t elf_prstatus_t;
1890 typedef prpsinfo_t elf_prpsinfo_t;
1891 typedef prfpregset_t elf_prfpregset_t;
1892 typedef prfpregset_t elf_fpregset_t;
1893 typedef gregset_t elf_gregset_t;
1894 typedef thrmisc_t elf_thrmisc_t;
1895 #define ELF_KERN_PROC_MASK	0
1896 typedef struct kinfo_proc elf_kinfo_proc_t;
1897 typedef vm_offset_t elf_ps_strings_t;
1898 #endif
1899 
1900 static void
1901 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1902 {
1903 	struct sbuf sbarg;
1904 	size_t len;
1905 	char *cp, *end;
1906 	struct proc *p;
1907 	elf_prpsinfo_t *psinfo;
1908 	int error;
1909 
1910 	p = (struct proc *)arg;
1911 	if (sb != NULL) {
1912 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1913 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1914 		psinfo->pr_version = PRPSINFO_VERSION;
1915 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1916 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1917 		PROC_LOCK(p);
1918 		if (p->p_args != NULL) {
1919 			len = sizeof(psinfo->pr_psargs) - 1;
1920 			if (len > p->p_args->ar_length)
1921 				len = p->p_args->ar_length;
1922 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1923 			PROC_UNLOCK(p);
1924 			error = 0;
1925 		} else {
1926 			_PHOLD(p);
1927 			PROC_UNLOCK(p);
1928 			sbuf_new(&sbarg, psinfo->pr_psargs,
1929 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1930 			error = proc_getargv(curthread, p, &sbarg);
1931 			PRELE(p);
1932 			if (sbuf_finish(&sbarg) == 0)
1933 				len = sbuf_len(&sbarg) - 1;
1934 			else
1935 				len = sizeof(psinfo->pr_psargs) - 1;
1936 			sbuf_delete(&sbarg);
1937 		}
1938 		if (error || len == 0)
1939 			strlcpy(psinfo->pr_psargs, p->p_comm,
1940 			    sizeof(psinfo->pr_psargs));
1941 		else {
1942 			KASSERT(len < sizeof(psinfo->pr_psargs),
1943 			    ("len is too long: %zu vs %zu", len,
1944 			    sizeof(psinfo->pr_psargs)));
1945 			cp = psinfo->pr_psargs;
1946 			end = cp + len - 1;
1947 			for (;;) {
1948 				cp = memchr(cp, '\0', end - cp);
1949 				if (cp == NULL)
1950 					break;
1951 				*cp = ' ';
1952 			}
1953 		}
1954 		psinfo->pr_pid = p->p_pid;
1955 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1956 		free(psinfo, M_TEMP);
1957 	}
1958 	*sizep = sizeof(*psinfo);
1959 }
1960 
1961 static void
1962 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1963 {
1964 	struct thread *td;
1965 	elf_prstatus_t *status;
1966 
1967 	td = (struct thread *)arg;
1968 	if (sb != NULL) {
1969 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1970 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1971 		status->pr_version = PRSTATUS_VERSION;
1972 		status->pr_statussz = sizeof(elf_prstatus_t);
1973 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1974 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1975 		status->pr_osreldate = osreldate;
1976 		status->pr_cursig = td->td_proc->p_sig;
1977 		status->pr_pid = td->td_tid;
1978 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1979 		fill_regs32(td, &status->pr_reg);
1980 #else
1981 		fill_regs(td, &status->pr_reg);
1982 #endif
1983 		sbuf_bcat(sb, status, sizeof(*status));
1984 		free(status, M_TEMP);
1985 	}
1986 	*sizep = sizeof(*status);
1987 }
1988 
1989 static void
1990 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1991 {
1992 	struct thread *td;
1993 	elf_prfpregset_t *fpregset;
1994 
1995 	td = (struct thread *)arg;
1996 	if (sb != NULL) {
1997 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1998 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1999 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2000 		fill_fpregs32(td, fpregset);
2001 #else
2002 		fill_fpregs(td, fpregset);
2003 #endif
2004 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2005 		free(fpregset, M_TEMP);
2006 	}
2007 	*sizep = sizeof(*fpregset);
2008 }
2009 
2010 static void
2011 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2012 {
2013 	struct thread *td;
2014 	elf_thrmisc_t thrmisc;
2015 
2016 	td = (struct thread *)arg;
2017 	if (sb != NULL) {
2018 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2019 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2020 		strcpy(thrmisc.pr_tname, td->td_name);
2021 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2022 	}
2023 	*sizep = sizeof(thrmisc);
2024 }
2025 
2026 static void
2027 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2028 {
2029 	struct thread *td;
2030 	size_t size;
2031 	int structsize;
2032 	struct ptrace_lwpinfo pl;
2033 
2034 	td = (struct thread *)arg;
2035 	size = sizeof(structsize) + sizeof(struct ptrace_lwpinfo);
2036 	if (sb != NULL) {
2037 		KASSERT(*sizep == size, ("invalid size"));
2038 		structsize = sizeof(struct ptrace_lwpinfo);
2039 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2040 		bzero(&pl, sizeof(pl));
2041 		pl.pl_lwpid = td->td_tid;
2042 		pl.pl_event = PL_EVENT_NONE;
2043 		pl.pl_sigmask = td->td_sigmask;
2044 		pl.pl_siglist = td->td_siglist;
2045 		if (td->td_si.si_signo != 0) {
2046 			pl.pl_event = PL_EVENT_SIGNAL;
2047 			pl.pl_flags |= PL_FLAG_SI;
2048 			pl.pl_siginfo = td->td_si;
2049 		}
2050 		strcpy(pl.pl_tdname, td->td_name);
2051 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2052 		sbuf_bcat(sb, &pl, sizeof(struct ptrace_lwpinfo));
2053 	}
2054 	*sizep = size;
2055 }
2056 
2057 /*
2058  * Allow for MD specific notes, as well as any MD
2059  * specific preparations for writing MI notes.
2060  */
2061 static void
2062 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2063 {
2064 	struct thread *td;
2065 	void *buf;
2066 	size_t size;
2067 
2068 	td = (struct thread *)arg;
2069 	size = *sizep;
2070 	if (size != 0 && sb != NULL)
2071 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2072 	else
2073 		buf = NULL;
2074 	size = 0;
2075 	__elfN(dump_thread)(td, buf, &size);
2076 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2077 	if (size != 0 && sb != NULL)
2078 		sbuf_bcat(sb, buf, size);
2079 	free(buf, M_TEMP);
2080 	*sizep = size;
2081 }
2082 
2083 #ifdef KINFO_PROC_SIZE
2084 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2085 #endif
2086 
2087 static void
2088 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2089 {
2090 	struct proc *p;
2091 	size_t size;
2092 	int structsize;
2093 
2094 	p = (struct proc *)arg;
2095 	size = sizeof(structsize) + p->p_numthreads *
2096 	    sizeof(elf_kinfo_proc_t);
2097 
2098 	if (sb != NULL) {
2099 		KASSERT(*sizep == size, ("invalid size"));
2100 		structsize = sizeof(elf_kinfo_proc_t);
2101 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2102 		sx_slock(&proctree_lock);
2103 		PROC_LOCK(p);
2104 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2105 		sx_sunlock(&proctree_lock);
2106 	}
2107 	*sizep = size;
2108 }
2109 
2110 #ifdef KINFO_FILE_SIZE
2111 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2112 #endif
2113 
2114 static void
2115 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2116 {
2117 	struct proc *p;
2118 	size_t size, sect_sz, i;
2119 	ssize_t start_len, sect_len;
2120 	int structsize, filedesc_flags;
2121 
2122 	if (coredump_pack_fileinfo)
2123 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2124 	else
2125 		filedesc_flags = 0;
2126 
2127 	p = (struct proc *)arg;
2128 	structsize = sizeof(struct kinfo_file);
2129 	if (sb == NULL) {
2130 		size = 0;
2131 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2132 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2133 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2134 		PROC_LOCK(p);
2135 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2136 		sbuf_finish(sb);
2137 		sbuf_delete(sb);
2138 		*sizep = size;
2139 	} else {
2140 		sbuf_start_section(sb, &start_len);
2141 
2142 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2143 		PROC_LOCK(p);
2144 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2145 		    filedesc_flags);
2146 
2147 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2148 		if (sect_len < 0)
2149 			return;
2150 		sect_sz = sect_len;
2151 
2152 		KASSERT(sect_sz <= *sizep,
2153 		    ("kern_proc_filedesc_out did not respect maxlen; "
2154 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2155 		     sect_sz - sizeof(structsize)));
2156 
2157 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2158 			sbuf_putc(sb, 0);
2159 	}
2160 }
2161 
2162 #ifdef KINFO_VMENTRY_SIZE
2163 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2164 #endif
2165 
2166 static void
2167 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2168 {
2169 	struct proc *p;
2170 	size_t size;
2171 	int structsize, vmmap_flags;
2172 
2173 	if (coredump_pack_vmmapinfo)
2174 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2175 	else
2176 		vmmap_flags = 0;
2177 
2178 	p = (struct proc *)arg;
2179 	structsize = sizeof(struct kinfo_vmentry);
2180 	if (sb == NULL) {
2181 		size = 0;
2182 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2183 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2184 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2185 		PROC_LOCK(p);
2186 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2187 		sbuf_finish(sb);
2188 		sbuf_delete(sb);
2189 		*sizep = size;
2190 	} else {
2191 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2192 		PROC_LOCK(p);
2193 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2194 		    vmmap_flags);
2195 	}
2196 }
2197 
2198 static void
2199 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2200 {
2201 	struct proc *p;
2202 	size_t size;
2203 	int structsize;
2204 
2205 	p = (struct proc *)arg;
2206 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2207 	if (sb != NULL) {
2208 		KASSERT(*sizep == size, ("invalid size"));
2209 		structsize = sizeof(gid_t);
2210 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2211 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2212 		    sizeof(gid_t));
2213 	}
2214 	*sizep = size;
2215 }
2216 
2217 static void
2218 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2219 {
2220 	struct proc *p;
2221 	size_t size;
2222 	int structsize;
2223 
2224 	p = (struct proc *)arg;
2225 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2226 	if (sb != NULL) {
2227 		KASSERT(*sizep == size, ("invalid size"));
2228 		structsize = sizeof(p->p_fd->fd_cmask);
2229 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2230 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2231 	}
2232 	*sizep = size;
2233 }
2234 
2235 static void
2236 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2237 {
2238 	struct proc *p;
2239 	struct rlimit rlim[RLIM_NLIMITS];
2240 	size_t size;
2241 	int structsize, i;
2242 
2243 	p = (struct proc *)arg;
2244 	size = sizeof(structsize) + sizeof(rlim);
2245 	if (sb != NULL) {
2246 		KASSERT(*sizep == size, ("invalid size"));
2247 		structsize = sizeof(rlim);
2248 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2249 		PROC_LOCK(p);
2250 		for (i = 0; i < RLIM_NLIMITS; i++)
2251 			lim_rlimit_proc(p, i, &rlim[i]);
2252 		PROC_UNLOCK(p);
2253 		sbuf_bcat(sb, rlim, sizeof(rlim));
2254 	}
2255 	*sizep = size;
2256 }
2257 
2258 static void
2259 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2260 {
2261 	struct proc *p;
2262 	size_t size;
2263 	int structsize;
2264 
2265 	p = (struct proc *)arg;
2266 	size = sizeof(structsize) + sizeof(p->p_osrel);
2267 	if (sb != NULL) {
2268 		KASSERT(*sizep == size, ("invalid size"));
2269 		structsize = sizeof(p->p_osrel);
2270 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2271 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2272 	}
2273 	*sizep = size;
2274 }
2275 
2276 static void
2277 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2278 {
2279 	struct proc *p;
2280 	elf_ps_strings_t ps_strings;
2281 	size_t size;
2282 	int structsize;
2283 
2284 	p = (struct proc *)arg;
2285 	size = sizeof(structsize) + sizeof(ps_strings);
2286 	if (sb != NULL) {
2287 		KASSERT(*sizep == size, ("invalid size"));
2288 		structsize = sizeof(ps_strings);
2289 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2290 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2291 #else
2292 		ps_strings = p->p_sysent->sv_psstrings;
2293 #endif
2294 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2295 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2296 	}
2297 	*sizep = size;
2298 }
2299 
2300 static void
2301 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2302 {
2303 	struct proc *p;
2304 	size_t size;
2305 	int structsize;
2306 
2307 	p = (struct proc *)arg;
2308 	if (sb == NULL) {
2309 		size = 0;
2310 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2311 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2312 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2313 		PHOLD(p);
2314 		proc_getauxv(curthread, p, sb);
2315 		PRELE(p);
2316 		sbuf_finish(sb);
2317 		sbuf_delete(sb);
2318 		*sizep = size;
2319 	} else {
2320 		structsize = sizeof(Elf_Auxinfo);
2321 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2322 		PHOLD(p);
2323 		proc_getauxv(curthread, p, sb);
2324 		PRELE(p);
2325 	}
2326 }
2327 
2328 static boolean_t
2329 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2330     int32_t *osrel, const Elf_Phdr *pnote)
2331 {
2332 	const Elf_Note *note, *note0, *note_end;
2333 	const char *note_name;
2334 	char *buf;
2335 	int i, error;
2336 	boolean_t res;
2337 
2338 	/* We need some limit, might as well use PAGE_SIZE. */
2339 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2340 		return (FALSE);
2341 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2342 	if (pnote->p_offset > PAGE_SIZE ||
2343 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2344 		VOP_UNLOCK(imgp->vp, 0);
2345 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2346 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2347 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2348 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2349 		    curthread->td_ucred, NOCRED, NULL, curthread);
2350 		if (error != 0) {
2351 			uprintf("i/o error PT_NOTE\n");
2352 			res = FALSE;
2353 			goto ret;
2354 		}
2355 		note = note0 = (const Elf_Note *)buf;
2356 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2357 	} else {
2358 		note = note0 = (const Elf_Note *)(imgp->image_header +
2359 		    pnote->p_offset);
2360 		note_end = (const Elf_Note *)(imgp->image_header +
2361 		    pnote->p_offset + pnote->p_filesz);
2362 		buf = NULL;
2363 	}
2364 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2365 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2366 		    (const char *)note < sizeof(Elf_Note)) {
2367 			res = FALSE;
2368 			goto ret;
2369 		}
2370 		if (note->n_namesz != checknote->hdr.n_namesz ||
2371 		    note->n_descsz != checknote->hdr.n_descsz ||
2372 		    note->n_type != checknote->hdr.n_type)
2373 			goto nextnote;
2374 		note_name = (const char *)(note + 1);
2375 		if (note_name + checknote->hdr.n_namesz >=
2376 		    (const char *)note_end || strncmp(checknote->vendor,
2377 		    note_name, checknote->hdr.n_namesz) != 0)
2378 			goto nextnote;
2379 
2380 		/*
2381 		 * Fetch the osreldate for binary
2382 		 * from the ELF OSABI-note if necessary.
2383 		 */
2384 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2385 		    checknote->trans_osrel != NULL) {
2386 			res = checknote->trans_osrel(note, osrel);
2387 			goto ret;
2388 		}
2389 		res = TRUE;
2390 		goto ret;
2391 nextnote:
2392 		note = (const Elf_Note *)((const char *)(note + 1) +
2393 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2394 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2395 	}
2396 	res = FALSE;
2397 ret:
2398 	free(buf, M_TEMP);
2399 	return (res);
2400 }
2401 
2402 /*
2403  * Try to find the appropriate ABI-note section for checknote,
2404  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2405  * first page of the image is searched, the same as for headers.
2406  */
2407 static boolean_t
2408 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2409     int32_t *osrel)
2410 {
2411 	const Elf_Phdr *phdr;
2412 	const Elf_Ehdr *hdr;
2413 	int i;
2414 
2415 	hdr = (const Elf_Ehdr *)imgp->image_header;
2416 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2417 
2418 	for (i = 0; i < hdr->e_phnum; i++) {
2419 		if (phdr[i].p_type == PT_NOTE &&
2420 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2421 			return (TRUE);
2422 	}
2423 	return (FALSE);
2424 
2425 }
2426 
2427 /*
2428  * Tell kern_execve.c about it, with a little help from the linker.
2429  */
2430 static struct execsw __elfN(execsw) = {
2431 	__CONCAT(exec_, __elfN(imgact)),
2432 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2433 };
2434 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2435 
2436 static vm_prot_t
2437 __elfN(trans_prot)(Elf_Word flags)
2438 {
2439 	vm_prot_t prot;
2440 
2441 	prot = 0;
2442 	if (flags & PF_X)
2443 		prot |= VM_PROT_EXECUTE;
2444 	if (flags & PF_W)
2445 		prot |= VM_PROT_WRITE;
2446 	if (flags & PF_R)
2447 		prot |= VM_PROT_READ;
2448 #if __ELF_WORD_SIZE == 32
2449 #if defined(__amd64__)
2450 	if (i386_read_exec && (flags & PF_R))
2451 		prot |= VM_PROT_EXECUTE;
2452 #endif
2453 #endif
2454 	return (prot);
2455 }
2456 
2457 static Elf_Word
2458 __elfN(untrans_prot)(vm_prot_t prot)
2459 {
2460 	Elf_Word flags;
2461 
2462 	flags = 0;
2463 	if (prot & VM_PROT_EXECUTE)
2464 		flags |= PF_X;
2465 	if (prot & VM_PROT_READ)
2466 		flags |= PF_R;
2467 	if (prot & VM_PROT_WRITE)
2468 		flags |= PF_W;
2469 	return (flags);
2470 }
2471