xref: /freebsd/sys/kern/imgact_elf.c (revision 0a48773f8c6108ad8755065c5cdc13fcdc0728b2)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000-2001, 2003 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
54 #include <sys/proc.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int interp_name_len, int32_t *osrel);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry, size_t pagesize);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
96     size_t pagesize);
97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
98 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99     int32_t *osrel);
100 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
101 static boolean_t __elfN(check_note)(struct image_params *imgp,
102     Elf_Brandnote *checknote, int32_t *osrel);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107     "");
108 
109 #define	CORE_BUF_SIZE	(16 * 1024)
110 
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118     &elf_legacy_coredump, 0,
119     "include all and only RW pages in core dumps");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
124 	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 bool
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 bool
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 == NULL ||
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 && (bi->header_supported == NULL ||
372 			    bi->header_supported(imgp)))
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 		    (bi->header_supported == NULL ||
386 		    bi->header_supported(imgp)))
387 			return (bi);
388 	}
389 	return (NULL);
390 }
391 
392 static int
393 __elfN(check_header)(const Elf_Ehdr *hdr)
394 {
395 	Elf_Brandinfo *bi;
396 	int i;
397 
398 	if (!IS_ELF(*hdr) ||
399 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
400 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
401 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
402 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
403 	    hdr->e_version != ELF_TARG_VER)
404 		return (ENOEXEC);
405 
406 	/*
407 	 * Make sure we have at least one brand for this machine.
408 	 */
409 
410 	for (i = 0; i < MAX_BRANDS; i++) {
411 		bi = elf_brand_list[i];
412 		if (bi != NULL && bi->machine == hdr->e_machine)
413 			break;
414 	}
415 	if (i == MAX_BRANDS)
416 		return (ENOEXEC);
417 
418 	return (0);
419 }
420 
421 static int
422 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
423     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
424 {
425 	struct sf_buf *sf;
426 	int error;
427 	vm_offset_t off;
428 
429 	/*
430 	 * Create the page if it doesn't exist yet. Ignore errors.
431 	 */
432 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
433 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
434 
435 	/*
436 	 * Find the page from the underlying object.
437 	 */
438 	if (object != NULL) {
439 		sf = vm_imgact_map_page(object, offset);
440 		if (sf == NULL)
441 			return (KERN_FAILURE);
442 		off = offset - trunc_page(offset);
443 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
444 		    end - start);
445 		vm_imgact_unmap_page(sf);
446 		if (error != 0)
447 			return (KERN_FAILURE);
448 	}
449 
450 	return (KERN_SUCCESS);
451 }
452 
453 static int
454 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
455     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
456     int cow)
457 {
458 	struct sf_buf *sf;
459 	vm_offset_t off;
460 	vm_size_t sz;
461 	int error, locked, rv;
462 
463 	if (start != trunc_page(start)) {
464 		rv = __elfN(map_partial)(map, object, offset, start,
465 		    round_page(start), prot);
466 		if (rv != KERN_SUCCESS)
467 			return (rv);
468 		offset += round_page(start) - start;
469 		start = round_page(start);
470 	}
471 	if (end != round_page(end)) {
472 		rv = __elfN(map_partial)(map, object, offset +
473 		    trunc_page(end) - start, trunc_page(end), end, prot);
474 		if (rv != KERN_SUCCESS)
475 			return (rv);
476 		end = trunc_page(end);
477 	}
478 	if (start >= end)
479 		return (KERN_SUCCESS);
480 	if ((offset & PAGE_MASK) != 0) {
481 		/*
482 		 * The mapping is not page aligned.  This means that we have
483 		 * to copy the data.
484 		 */
485 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
486 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
487 		if (rv != KERN_SUCCESS)
488 			return (rv);
489 		if (object == NULL)
490 			return (KERN_SUCCESS);
491 		for (; start < end; start += sz) {
492 			sf = vm_imgact_map_page(object, offset);
493 			if (sf == NULL)
494 				return (KERN_FAILURE);
495 			off = offset - trunc_page(offset);
496 			sz = end - start;
497 			if (sz > PAGE_SIZE - off)
498 				sz = PAGE_SIZE - off;
499 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
500 			    (caddr_t)start, sz);
501 			vm_imgact_unmap_page(sf);
502 			if (error != 0)
503 				return (KERN_FAILURE);
504 			offset += sz;
505 		}
506 	} else {
507 		vm_object_reference(object);
508 		rv = vm_map_fixed(map, object, offset, start, end - start,
509 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
510 		if (rv != KERN_SUCCESS) {
511 			locked = VOP_ISLOCKED(imgp->vp);
512 			VOP_UNLOCK(imgp->vp, 0);
513 			vm_object_deallocate(object);
514 			vn_lock(imgp->vp, locked | LK_RETRY);
515 			return (rv);
516 		}
517 	}
518 	return (KERN_SUCCESS);
519 }
520 
521 static int
522 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
523     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
524     size_t pagesize)
525 {
526 	struct sf_buf *sf;
527 	size_t map_len;
528 	vm_map_t map;
529 	vm_object_t object;
530 	vm_offset_t off, map_addr;
531 	int error, rv, cow;
532 	size_t copy_len;
533 	vm_ooffset_t file_addr;
534 
535 	/*
536 	 * It's necessary to fail if the filsz + offset taken from the
537 	 * header is greater than the actual file pager object's size.
538 	 * If we were to allow this, then the vm_map_find() below would
539 	 * walk right off the end of the file object and into the ether.
540 	 *
541 	 * While I'm here, might as well check for something else that
542 	 * is invalid: filsz cannot be greater than memsz.
543 	 */
544 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
545 	    filsz > memsz) {
546 		uprintf("elf_load_section: truncated ELF file\n");
547 		return (ENOEXEC);
548 	}
549 
550 	object = imgp->object;
551 	map = &imgp->proc->p_vmspace->vm_map;
552 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
553 	file_addr = trunc_page_ps(offset, pagesize);
554 
555 	/*
556 	 * We have two choices.  We can either clear the data in the last page
557 	 * of an oversized mapping, or we can start the anon mapping a page
558 	 * early and copy the initialized data into that first page.  We
559 	 * choose the second.
560 	 */
561 	if (filsz == 0)
562 		map_len = 0;
563 	else if (memsz > filsz)
564 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
565 	else
566 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
567 
568 	if (map_len != 0) {
569 		/* cow flags: don't dump readonly sections in core */
570 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
571 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
572 
573 		rv = __elfN(map_insert)(imgp, map,
574 				      object,
575 				      file_addr,	/* file offset */
576 				      map_addr,		/* virtual start */
577 				      map_addr + map_len,/* virtual end */
578 				      prot,
579 				      cow);
580 		if (rv != KERN_SUCCESS)
581 			return (EINVAL);
582 
583 		/* we can stop now if we've covered it all */
584 		if (memsz == filsz)
585 			return (0);
586 	}
587 
588 
589 	/*
590 	 * We have to get the remaining bit of the file into the first part
591 	 * of the oversized map segment.  This is normally because the .data
592 	 * segment in the file is extended to provide bss.  It's a neat idea
593 	 * to try and save a page, but it's a pain in the behind to implement.
594 	 */
595 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
596 	    filsz, pagesize);
597 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
598 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
599 	    map_addr;
600 
601 	/* This had damn well better be true! */
602 	if (map_len != 0) {
603 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
604 		    map_addr + map_len, prot, 0);
605 		if (rv != KERN_SUCCESS)
606 			return (EINVAL);
607 	}
608 
609 	if (copy_len != 0) {
610 		sf = vm_imgact_map_page(object, offset + filsz);
611 		if (sf == NULL)
612 			return (EIO);
613 
614 		/* send the page fragment to user space */
615 		off = trunc_page_ps(offset + filsz, pagesize) -
616 		    trunc_page(offset + filsz);
617 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
618 		    (caddr_t)map_addr, copy_len);
619 		vm_imgact_unmap_page(sf);
620 		if (error != 0)
621 			return (error);
622 	}
623 
624 	/*
625 	 * Remove write access to the page if it was only granted by map_insert
626 	 * to allow copyout.
627 	 */
628 	if ((prot & VM_PROT_WRITE) == 0)
629 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
630 		    map_len), prot, FALSE);
631 
632 	return (0);
633 }
634 
635 /*
636  * Load the file "file" into memory.  It may be either a shared object
637  * or an executable.
638  *
639  * The "addr" reference parameter is in/out.  On entry, it specifies
640  * the address where a shared object should be loaded.  If the file is
641  * an executable, this value is ignored.  On exit, "addr" specifies
642  * where the file was actually loaded.
643  *
644  * The "entry" reference parameter is out only.  On exit, it specifies
645  * the entry point for the loaded file.
646  */
647 static int
648 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
649 	u_long *entry, size_t pagesize)
650 {
651 	struct {
652 		struct nameidata nd;
653 		struct vattr attr;
654 		struct image_params image_params;
655 	} *tempdata;
656 	const Elf_Ehdr *hdr = NULL;
657 	const Elf_Phdr *phdr = NULL;
658 	struct nameidata *nd;
659 	struct vattr *attr;
660 	struct image_params *imgp;
661 	vm_prot_t prot;
662 	u_long rbase;
663 	u_long base_addr = 0;
664 	int error, i, numsegs;
665 
666 #ifdef CAPABILITY_MODE
667 	/*
668 	 * XXXJA: This check can go away once we are sufficiently confident
669 	 * that the checks in namei() are correct.
670 	 */
671 	if (IN_CAPABILITY_MODE(curthread))
672 		return (ECAPMODE);
673 #endif
674 
675 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
676 	nd = &tempdata->nd;
677 	attr = &tempdata->attr;
678 	imgp = &tempdata->image_params;
679 
680 	/*
681 	 * Initialize part of the common data
682 	 */
683 	imgp->proc = p;
684 	imgp->attr = attr;
685 	imgp->firstpage = NULL;
686 	imgp->image_header = NULL;
687 	imgp->object = NULL;
688 	imgp->execlabel = NULL;
689 
690 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
691 	if ((error = namei(nd)) != 0) {
692 		nd->ni_vp = NULL;
693 		goto fail;
694 	}
695 	NDFREE(nd, NDF_ONLY_PNBUF);
696 	imgp->vp = nd->ni_vp;
697 
698 	/*
699 	 * Check permissions, modes, uid, etc on the file, and "open" it.
700 	 */
701 	error = exec_check_permissions(imgp);
702 	if (error)
703 		goto fail;
704 
705 	error = exec_map_first_page(imgp);
706 	if (error)
707 		goto fail;
708 
709 	/*
710 	 * Also make certain that the interpreter stays the same, so set
711 	 * its VV_TEXT flag, too.
712 	 */
713 	VOP_SET_TEXT(nd->ni_vp);
714 
715 	imgp->object = nd->ni_vp->v_object;
716 
717 	hdr = (const Elf_Ehdr *)imgp->image_header;
718 	if ((error = __elfN(check_header)(hdr)) != 0)
719 		goto fail;
720 	if (hdr->e_type == ET_DYN)
721 		rbase = *addr;
722 	else if (hdr->e_type == ET_EXEC)
723 		rbase = 0;
724 	else {
725 		error = ENOEXEC;
726 		goto fail;
727 	}
728 
729 	/* Only support headers that fit within first page for now      */
730 	if ((hdr->e_phoff > PAGE_SIZE) ||
731 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
732 		error = ENOEXEC;
733 		goto fail;
734 	}
735 
736 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
737 	if (!aligned(phdr, Elf_Addr)) {
738 		error = ENOEXEC;
739 		goto fail;
740 	}
741 
742 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
743 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
744 			/* Loadable segment */
745 			prot = __elfN(trans_prot)(phdr[i].p_flags);
746 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
747 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
748 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
749 			if (error != 0)
750 				goto fail;
751 			/*
752 			 * Establish the base address if this is the
753 			 * first segment.
754 			 */
755 			if (numsegs == 0)
756   				base_addr = trunc_page(phdr[i].p_vaddr +
757 				    rbase);
758 			numsegs++;
759 		}
760 	}
761 	*addr = base_addr;
762 	*entry = (unsigned long)hdr->e_entry + rbase;
763 
764 fail:
765 	if (imgp->firstpage)
766 		exec_unmap_first_page(imgp);
767 
768 	if (nd->ni_vp)
769 		vput(nd->ni_vp);
770 
771 	free(tempdata, M_TEMP);
772 
773 	return (error);
774 }
775 
776 static int
777 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
778 {
779 	struct thread *td;
780 	const Elf_Ehdr *hdr;
781 	const Elf_Phdr *phdr;
782 	Elf_Auxargs *elf_auxargs;
783 	struct vmspace *vmspace;
784 	const char *err_str, *newinterp;
785 	char *interp, *interp_buf, *path;
786 	Elf_Brandinfo *brand_info;
787 	struct sysentvec *sv;
788 	vm_prot_t prot;
789 	u_long text_size, data_size, total_size, text_addr, data_addr;
790 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
791 	int32_t osrel;
792 	int error, i, n, interp_name_len, have_interp;
793 
794 	hdr = (const Elf_Ehdr *)imgp->image_header;
795 
796 	/*
797 	 * Do we have a valid ELF header ?
798 	 *
799 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
800 	 * if particular brand doesn't support it.
801 	 */
802 	if (__elfN(check_header)(hdr) != 0 ||
803 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
804 		return (-1);
805 
806 	/*
807 	 * From here on down, we return an errno, not -1, as we've
808 	 * detected an ELF file.
809 	 */
810 
811 	if ((hdr->e_phoff > PAGE_SIZE) ||
812 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
813 		/* Only support headers in first page for now */
814 		uprintf("Program headers not in the first page\n");
815 		return (ENOEXEC);
816 	}
817 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
818 	if (!aligned(phdr, Elf_Addr)) {
819 		uprintf("Unaligned program headers\n");
820 		return (ENOEXEC);
821 	}
822 
823 	n = error = 0;
824 	baddr = 0;
825 	osrel = 0;
826 	text_size = data_size = total_size = text_addr = data_addr = 0;
827 	entry = proghdr = 0;
828 	interp_name_len = 0;
829 	err_str = newinterp = NULL;
830 	interp = interp_buf = NULL;
831 	td = curthread;
832 
833 	for (i = 0; i < hdr->e_phnum; i++) {
834 		switch (phdr[i].p_type) {
835 		case PT_LOAD:
836 			if (n == 0)
837 				baddr = phdr[i].p_vaddr;
838 			n++;
839 			break;
840 		case PT_INTERP:
841 			/* Path to interpreter */
842 			if (phdr[i].p_filesz < 2 ||
843 			    phdr[i].p_filesz > MAXPATHLEN) {
844 				uprintf("Invalid PT_INTERP\n");
845 				error = ENOEXEC;
846 				goto ret;
847 			}
848 			if (interp != NULL) {
849 				uprintf("Multiple PT_INTERP headers\n");
850 				error = ENOEXEC;
851 				goto ret;
852 			}
853 			interp_name_len = phdr[i].p_filesz;
854 			if (phdr[i].p_offset > PAGE_SIZE ||
855 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
856 				VOP_UNLOCK(imgp->vp, 0);
857 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
858 				    M_WAITOK);
859 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
860 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
861 				    interp_name_len, phdr[i].p_offset,
862 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
863 				    NOCRED, NULL, td);
864 				if (error != 0) {
865 					uprintf("i/o error PT_INTERP %d\n",
866 					    error);
867 					goto ret;
868 				}
869 				interp_buf[interp_name_len] = '\0';
870 				interp = interp_buf;
871 			} else {
872 				interp = __DECONST(char *, imgp->image_header) +
873 				    phdr[i].p_offset;
874 				if (interp[interp_name_len - 1] != '\0') {
875 					uprintf("Invalid PT_INTERP\n");
876 					error = ENOEXEC;
877 					goto ret;
878 				}
879 			}
880 			break;
881 		case PT_GNU_STACK:
882 			if (__elfN(nxstack))
883 				imgp->stack_prot =
884 				    __elfN(trans_prot)(phdr[i].p_flags);
885 			imgp->stack_sz = phdr[i].p_memsz;
886 			break;
887 		}
888 	}
889 
890 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
891 	    &osrel);
892 	if (brand_info == NULL) {
893 		uprintf("ELF binary type \"%u\" not known.\n",
894 		    hdr->e_ident[EI_OSABI]);
895 		error = ENOEXEC;
896 		goto ret;
897 	}
898 	et_dyn_addr = 0;
899 	if (hdr->e_type == ET_DYN) {
900 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
901 			uprintf("Cannot execute shared object\n");
902 			error = ENOEXEC;
903 			goto ret;
904 		}
905 		/*
906 		 * Honour the base load address from the dso if it is
907 		 * non-zero for some reason.
908 		 */
909 		if (baddr == 0)
910 			et_dyn_addr = ET_DYN_LOAD_ADDR;
911 	}
912 	sv = brand_info->sysvec;
913 	if (interp != NULL && brand_info->interp_newpath != NULL)
914 		newinterp = brand_info->interp_newpath;
915 
916 	/*
917 	 * Avoid a possible deadlock if the current address space is destroyed
918 	 * and that address space maps the locked vnode.  In the common case,
919 	 * the locked vnode's v_usecount is decremented but remains greater
920 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
921 	 * However, in cases where the vnode lock is external, such as nullfs,
922 	 * v_usecount may become zero.
923 	 *
924 	 * The VV_TEXT flag prevents modifications to the executable while
925 	 * the vnode is unlocked.
926 	 */
927 	VOP_UNLOCK(imgp->vp, 0);
928 
929 	error = exec_new_vmspace(imgp, sv);
930 	imgp->proc->p_sysent = sv;
931 
932 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
933 	if (error != 0)
934 		goto ret;
935 
936 	for (i = 0; i < hdr->e_phnum; i++) {
937 		switch (phdr[i].p_type) {
938 		case PT_LOAD:	/* Loadable segment */
939 			if (phdr[i].p_memsz == 0)
940 				break;
941 			prot = __elfN(trans_prot)(phdr[i].p_flags);
942 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
943 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
944 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
945 			    sv->sv_pagesize);
946 			if (error != 0)
947 				goto ret;
948 
949 			/*
950 			 * If this segment contains the program headers,
951 			 * remember their virtual address for the AT_PHDR
952 			 * aux entry. Static binaries don't usually include
953 			 * a PT_PHDR entry.
954 			 */
955 			if (phdr[i].p_offset == 0 &&
956 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
957 				<= phdr[i].p_filesz)
958 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
959 				    et_dyn_addr;
960 
961 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
962 			seg_size = round_page(phdr[i].p_memsz +
963 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
964 
965 			/*
966 			 * Make the largest executable segment the official
967 			 * text segment and all others data.
968 			 *
969 			 * Note that obreak() assumes that data_addr +
970 			 * data_size == end of data load area, and the ELF
971 			 * file format expects segments to be sorted by
972 			 * address.  If multiple data segments exist, the
973 			 * last one will be used.
974 			 */
975 
976 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
977 				text_size = seg_size;
978 				text_addr = seg_addr;
979 			} else {
980 				data_size = seg_size;
981 				data_addr = seg_addr;
982 			}
983 			total_size += seg_size;
984 			break;
985 		case PT_PHDR: 	/* Program header table info */
986 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
987 			break;
988 		default:
989 			break;
990 		}
991 	}
992 
993 	if (data_addr == 0 && data_size == 0) {
994 		data_addr = text_addr;
995 		data_size = text_size;
996 	}
997 
998 	entry = (u_long)hdr->e_entry + et_dyn_addr;
999 
1000 	/*
1001 	 * Check limits.  It should be safe to check the
1002 	 * limits after loading the segments since we do
1003 	 * not actually fault in all the segments pages.
1004 	 */
1005 	PROC_LOCK(imgp->proc);
1006 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1007 		err_str = "Data segment size exceeds process limit";
1008 	else if (text_size > maxtsiz)
1009 		err_str = "Text segment size exceeds system limit";
1010 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1011 		err_str = "Total segment size exceeds process limit";
1012 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1013 		err_str = "Data segment size exceeds resource limit";
1014 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1015 		err_str = "Total segment size exceeds resource limit";
1016 	if (err_str != NULL) {
1017 		PROC_UNLOCK(imgp->proc);
1018 		uprintf("%s\n", err_str);
1019 		error = ENOMEM;
1020 		goto ret;
1021 	}
1022 
1023 	vmspace = imgp->proc->p_vmspace;
1024 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1025 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1026 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1027 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1028 
1029 	/*
1030 	 * We load the dynamic linker where a userland call
1031 	 * to mmap(0, ...) would put it.  The rationale behind this
1032 	 * calculation is that it leaves room for the heap to grow to
1033 	 * its maximum allowed size.
1034 	 */
1035 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1036 	    RLIMIT_DATA));
1037 	PROC_UNLOCK(imgp->proc);
1038 
1039 	imgp->entry_addr = entry;
1040 
1041 	if (interp != NULL) {
1042 		have_interp = FALSE;
1043 		VOP_UNLOCK(imgp->vp, 0);
1044 		if (brand_info->emul_path != NULL &&
1045 		    brand_info->emul_path[0] != '\0') {
1046 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1047 			snprintf(path, MAXPATHLEN, "%s%s",
1048 			    brand_info->emul_path, interp);
1049 			error = __elfN(load_file)(imgp->proc, path, &addr,
1050 			    &imgp->entry_addr, sv->sv_pagesize);
1051 			free(path, M_TEMP);
1052 			if (error == 0)
1053 				have_interp = TRUE;
1054 		}
1055 		if (!have_interp && newinterp != NULL &&
1056 		    (brand_info->interp_path == NULL ||
1057 		    strcmp(interp, brand_info->interp_path) == 0)) {
1058 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1059 			    &imgp->entry_addr, sv->sv_pagesize);
1060 			if (error == 0)
1061 				have_interp = TRUE;
1062 		}
1063 		if (!have_interp) {
1064 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1065 			    &imgp->entry_addr, sv->sv_pagesize);
1066 		}
1067 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1068 		if (error != 0) {
1069 			uprintf("ELF interpreter %s not found, error %d\n",
1070 			    interp, error);
1071 			goto ret;
1072 		}
1073 	} else
1074 		addr = et_dyn_addr;
1075 
1076 	/*
1077 	 * Construct auxargs table (used by the fixup routine)
1078 	 */
1079 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1080 	elf_auxargs->execfd = -1;
1081 	elf_auxargs->phdr = proghdr;
1082 	elf_auxargs->phent = hdr->e_phentsize;
1083 	elf_auxargs->phnum = hdr->e_phnum;
1084 	elf_auxargs->pagesz = PAGE_SIZE;
1085 	elf_auxargs->base = addr;
1086 	elf_auxargs->flags = 0;
1087 	elf_auxargs->entry = entry;
1088 	elf_auxargs->hdr_eflags = hdr->e_flags;
1089 
1090 	imgp->auxargs = elf_auxargs;
1091 	imgp->interpreted = 0;
1092 	imgp->reloc_base = addr;
1093 	imgp->proc->p_osrel = osrel;
1094 	imgp->proc->p_elf_machine = hdr->e_machine;
1095 	imgp->proc->p_elf_flags = hdr->e_flags;
1096 
1097 ret:
1098 	free(interp_buf, M_TEMP);
1099 	return (error);
1100 }
1101 
1102 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1103 
1104 int
1105 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1106 {
1107 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1108 	Elf_Auxinfo *argarray, *pos;
1109 	Elf_Addr *base, *auxbase;
1110 	int error;
1111 
1112 	base = (Elf_Addr *)*stack_base;
1113 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1114 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1115 	    M_WAITOK | M_ZERO);
1116 
1117 	if (args->execfd != -1)
1118 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1119 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1120 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1121 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1122 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1123 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1124 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1125 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1126 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1127 	if (imgp->execpathp != 0)
1128 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1129 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1130 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1131 	if (imgp->canary != 0) {
1132 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1133 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1134 	}
1135 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1136 	if (imgp->pagesizes != 0) {
1137 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1138 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1139 	}
1140 	if (imgp->sysent->sv_timekeep_base != 0) {
1141 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1142 		    imgp->sysent->sv_timekeep_base);
1143 	}
1144 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1145 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1146 	    imgp->sysent->sv_stackprot);
1147 	if (imgp->sysent->sv_hwcap != NULL)
1148 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1149 	if (imgp->sysent->sv_hwcap2 != NULL)
1150 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1151 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1152 
1153 	free(imgp->auxargs, M_TEMP);
1154 	imgp->auxargs = NULL;
1155 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1156 
1157 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1158 	free(argarray, M_TEMP);
1159 	if (error != 0)
1160 		return (error);
1161 
1162 	base--;
1163 	if (suword(base, imgp->args->argc) == -1)
1164 		return (EFAULT);
1165 	*stack_base = (register_t *)base;
1166 	return (0);
1167 }
1168 
1169 /*
1170  * Code for generating ELF core dumps.
1171  */
1172 
1173 typedef void (*segment_callback)(vm_map_entry_t, void *);
1174 
1175 /* Closure for cb_put_phdr(). */
1176 struct phdr_closure {
1177 	Elf_Phdr *phdr;		/* Program header to fill in */
1178 	Elf_Off offset;		/* Offset of segment in core file */
1179 };
1180 
1181 /* Closure for cb_size_segment(). */
1182 struct sseg_closure {
1183 	int count;		/* Count of writable segments. */
1184 	size_t size;		/* Total size of all writable segments. */
1185 };
1186 
1187 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1188 
1189 struct note_info {
1190 	int		type;		/* Note type. */
1191 	outfunc_t 	outfunc; 	/* Output function. */
1192 	void		*outarg;	/* Argument for the output function. */
1193 	size_t		outsize;	/* Output size. */
1194 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1195 };
1196 
1197 TAILQ_HEAD(note_info_list, note_info);
1198 
1199 /* Coredump output parameters. */
1200 struct coredump_params {
1201 	off_t		offset;
1202 	struct ucred	*active_cred;
1203 	struct ucred	*file_cred;
1204 	struct thread	*td;
1205 	struct vnode	*vp;
1206 	struct compressor *comp;
1207 };
1208 
1209 extern int compress_user_cores;
1210 extern int compress_user_cores_level;
1211 
1212 static void cb_put_phdr(vm_map_entry_t, void *);
1213 static void cb_size_segment(vm_map_entry_t, void *);
1214 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1215     enum uio_seg);
1216 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1217 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1218     struct note_info_list *, size_t);
1219 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1220     size_t *);
1221 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1222 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1223 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1224 static int sbuf_drain_core_output(void *, const char *, int);
1225 static int sbuf_drain_count(void *arg, const char *data, int len);
1226 
1227 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1228 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1229 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1230 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1231 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1232 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1233 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1234 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1235 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1236 static void note_procstat_files(void *, struct sbuf *, size_t *);
1237 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1238 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1239 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1240 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1241 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1242 
1243 /*
1244  * Write out a core segment to the compression stream.
1245  */
1246 static int
1247 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1248 {
1249 	u_int chunk_len;
1250 	int error;
1251 
1252 	while (len > 0) {
1253 		chunk_len = MIN(len, CORE_BUF_SIZE);
1254 
1255 		/*
1256 		 * We can get EFAULT error here.
1257 		 * In that case zero out the current chunk of the segment.
1258 		 */
1259 		error = copyin(base, buf, chunk_len);
1260 		if (error != 0)
1261 			bzero(buf, chunk_len);
1262 		error = compressor_write(p->comp, buf, chunk_len);
1263 		if (error != 0)
1264 			break;
1265 		base += chunk_len;
1266 		len -= chunk_len;
1267 	}
1268 	return (error);
1269 }
1270 
1271 static int
1272 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1273 {
1274 
1275 	return (core_write((struct coredump_params *)arg, base, len, offset,
1276 	    UIO_SYSSPACE));
1277 }
1278 
1279 static int
1280 core_write(struct coredump_params *p, const void *base, size_t len,
1281     off_t offset, enum uio_seg seg)
1282 {
1283 
1284 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1285 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1286 	    p->active_cred, p->file_cred, NULL, p->td));
1287 }
1288 
1289 static int
1290 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1291     void *tmpbuf)
1292 {
1293 	int error;
1294 
1295 	if (p->comp != NULL)
1296 		return (compress_chunk(p, base, tmpbuf, len));
1297 
1298 	/*
1299 	 * EFAULT is a non-fatal error that we can get, for example,
1300 	 * if the segment is backed by a file but extends beyond its
1301 	 * end.
1302 	 */
1303 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1304 	if (error == EFAULT) {
1305 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1306 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1307 		    "for process %s\n", base, len, offset, curproc->p_comm);
1308 
1309 		/*
1310 		 * Write a "real" zero byte at the end of the target region
1311 		 * in the case this is the last segment.
1312 		 * The intermediate space will be implicitly zero-filled.
1313 		 */
1314 		error = core_write(p, zero_region, 1, offset + len - 1,
1315 		    UIO_SYSSPACE);
1316 	}
1317 	return (error);
1318 }
1319 
1320 /*
1321  * Drain into a core file.
1322  */
1323 static int
1324 sbuf_drain_core_output(void *arg, const char *data, int len)
1325 {
1326 	struct coredump_params *p;
1327 	int error, locked;
1328 
1329 	p = (struct coredump_params *)arg;
1330 
1331 	/*
1332 	 * Some kern_proc out routines that print to this sbuf may
1333 	 * call us with the process lock held. Draining with the
1334 	 * non-sleepable lock held is unsafe. The lock is needed for
1335 	 * those routines when dumping a live process. In our case we
1336 	 * can safely release the lock before draining and acquire
1337 	 * again after.
1338 	 */
1339 	locked = PROC_LOCKED(p->td->td_proc);
1340 	if (locked)
1341 		PROC_UNLOCK(p->td->td_proc);
1342 	if (p->comp != NULL)
1343 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1344 	else
1345 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1346 		    UIO_SYSSPACE);
1347 	if (locked)
1348 		PROC_LOCK(p->td->td_proc);
1349 	if (error != 0)
1350 		return (-error);
1351 	p->offset += len;
1352 	return (len);
1353 }
1354 
1355 /*
1356  * Drain into a counter.
1357  */
1358 static int
1359 sbuf_drain_count(void *arg, const char *data __unused, int len)
1360 {
1361 	size_t *sizep;
1362 
1363 	sizep = (size_t *)arg;
1364 	*sizep += len;
1365 	return (len);
1366 }
1367 
1368 int
1369 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1370 {
1371 	struct ucred *cred = td->td_ucred;
1372 	int error = 0;
1373 	struct sseg_closure seginfo;
1374 	struct note_info_list notelst;
1375 	struct coredump_params params;
1376 	struct note_info *ninfo;
1377 	void *hdr, *tmpbuf;
1378 	size_t hdrsize, notesz, coresize;
1379 
1380 	hdr = NULL;
1381 	tmpbuf = NULL;
1382 	TAILQ_INIT(&notelst);
1383 
1384 	/* Size the program segments. */
1385 	seginfo.count = 0;
1386 	seginfo.size = 0;
1387 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1388 
1389 	/*
1390 	 * Collect info about the core file header area.
1391 	 */
1392 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1393 	if (seginfo.count + 1 >= PN_XNUM)
1394 		hdrsize += sizeof(Elf_Shdr);
1395 	__elfN(prepare_notes)(td, &notelst, &notesz);
1396 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1397 
1398 	/* Set up core dump parameters. */
1399 	params.offset = 0;
1400 	params.active_cred = cred;
1401 	params.file_cred = NOCRED;
1402 	params.td = td;
1403 	params.vp = vp;
1404 	params.comp = NULL;
1405 
1406 #ifdef RACCT
1407 	if (racct_enable) {
1408 		PROC_LOCK(td->td_proc);
1409 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1410 		PROC_UNLOCK(td->td_proc);
1411 		if (error != 0) {
1412 			error = EFAULT;
1413 			goto done;
1414 		}
1415 	}
1416 #endif
1417 	if (coresize >= limit) {
1418 		error = EFAULT;
1419 		goto done;
1420 	}
1421 
1422 	/* Create a compression stream if necessary. */
1423 	if (compress_user_cores != 0) {
1424 		params.comp = compressor_init(core_compressed_write,
1425 		    compress_user_cores, CORE_BUF_SIZE,
1426 		    compress_user_cores_level, &params);
1427 		if (params.comp == NULL) {
1428 			error = EFAULT;
1429 			goto done;
1430 		}
1431 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1432         }
1433 
1434 	/*
1435 	 * Allocate memory for building the header, fill it up,
1436 	 * and write it out following the notes.
1437 	 */
1438 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1439 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1440 	    notesz);
1441 
1442 	/* Write the contents of all of the writable segments. */
1443 	if (error == 0) {
1444 		Elf_Phdr *php;
1445 		off_t offset;
1446 		int i;
1447 
1448 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1449 		offset = round_page(hdrsize + notesz);
1450 		for (i = 0; i < seginfo.count; i++) {
1451 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1452 			    php->p_filesz, offset, &params, tmpbuf);
1453 			if (error != 0)
1454 				break;
1455 			offset += php->p_filesz;
1456 			php++;
1457 		}
1458 		if (error == 0 && params.comp != NULL)
1459 			error = compressor_flush(params.comp);
1460 	}
1461 	if (error) {
1462 		log(LOG_WARNING,
1463 		    "Failed to write core file for process %s (error %d)\n",
1464 		    curproc->p_comm, error);
1465 	}
1466 
1467 done:
1468 	free(tmpbuf, M_TEMP);
1469 	if (params.comp != NULL)
1470 		compressor_fini(params.comp);
1471 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1472 		TAILQ_REMOVE(&notelst, ninfo, link);
1473 		free(ninfo, M_TEMP);
1474 	}
1475 	if (hdr != NULL)
1476 		free(hdr, M_TEMP);
1477 
1478 	return (error);
1479 }
1480 
1481 /*
1482  * A callback for each_dumpable_segment() to write out the segment's
1483  * program header entry.
1484  */
1485 static void
1486 cb_put_phdr(vm_map_entry_t entry, void *closure)
1487 {
1488 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1489 	Elf_Phdr *phdr = phc->phdr;
1490 
1491 	phc->offset = round_page(phc->offset);
1492 
1493 	phdr->p_type = PT_LOAD;
1494 	phdr->p_offset = phc->offset;
1495 	phdr->p_vaddr = entry->start;
1496 	phdr->p_paddr = 0;
1497 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1498 	phdr->p_align = PAGE_SIZE;
1499 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1500 
1501 	phc->offset += phdr->p_filesz;
1502 	phc->phdr++;
1503 }
1504 
1505 /*
1506  * A callback for each_dumpable_segment() to gather information about
1507  * the number of segments and their total size.
1508  */
1509 static void
1510 cb_size_segment(vm_map_entry_t entry, void *closure)
1511 {
1512 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1513 
1514 	ssc->count++;
1515 	ssc->size += entry->end - entry->start;
1516 }
1517 
1518 /*
1519  * For each writable segment in the process's memory map, call the given
1520  * function with a pointer to the map entry and some arbitrary
1521  * caller-supplied data.
1522  */
1523 static void
1524 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1525 {
1526 	struct proc *p = td->td_proc;
1527 	vm_map_t map = &p->p_vmspace->vm_map;
1528 	vm_map_entry_t entry;
1529 	vm_object_t backing_object, object;
1530 	boolean_t ignore_entry;
1531 
1532 	vm_map_lock_read(map);
1533 	for (entry = map->header.next; entry != &map->header;
1534 	    entry = entry->next) {
1535 		/*
1536 		 * Don't dump inaccessible mappings, deal with legacy
1537 		 * coredump mode.
1538 		 *
1539 		 * Note that read-only segments related to the elf binary
1540 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1541 		 * need to arbitrarily ignore such segments.
1542 		 */
1543 		if (elf_legacy_coredump) {
1544 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1545 				continue;
1546 		} else {
1547 			if ((entry->protection & VM_PROT_ALL) == 0)
1548 				continue;
1549 		}
1550 
1551 		/*
1552 		 * Dont include memory segment in the coredump if
1553 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1554 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1555 		 * kernel map).
1556 		 */
1557 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1558 			continue;
1559 
1560 		if ((object = entry->object.vm_object) == NULL)
1561 			continue;
1562 
1563 		/* Ignore memory-mapped devices and such things. */
1564 		VM_OBJECT_RLOCK(object);
1565 		while ((backing_object = object->backing_object) != NULL) {
1566 			VM_OBJECT_RLOCK(backing_object);
1567 			VM_OBJECT_RUNLOCK(object);
1568 			object = backing_object;
1569 		}
1570 		ignore_entry = object->type != OBJT_DEFAULT &&
1571 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1572 		    object->type != OBJT_PHYS;
1573 		VM_OBJECT_RUNLOCK(object);
1574 		if (ignore_entry)
1575 			continue;
1576 
1577 		(*func)(entry, closure);
1578 	}
1579 	vm_map_unlock_read(map);
1580 }
1581 
1582 /*
1583  * Write the core file header to the file, including padding up to
1584  * the page boundary.
1585  */
1586 static int
1587 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1588     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1589 {
1590 	struct note_info *ninfo;
1591 	struct sbuf *sb;
1592 	int error;
1593 
1594 	/* Fill in the header. */
1595 	bzero(hdr, hdrsize);
1596 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1597 
1598 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1599 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1600 	sbuf_start_section(sb, NULL);
1601 	sbuf_bcat(sb, hdr, hdrsize);
1602 	TAILQ_FOREACH(ninfo, notelst, link)
1603 	    __elfN(putnote)(ninfo, sb);
1604 	/* Align up to a page boundary for the program segments. */
1605 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1606 	error = sbuf_finish(sb);
1607 	sbuf_delete(sb);
1608 
1609 	return (error);
1610 }
1611 
1612 static void
1613 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1614     size_t *sizep)
1615 {
1616 	struct proc *p;
1617 	struct thread *thr;
1618 	size_t size;
1619 
1620 	p = td->td_proc;
1621 	size = 0;
1622 
1623 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1624 
1625 	/*
1626 	 * To have the debugger select the right thread (LWP) as the initial
1627 	 * thread, we dump the state of the thread passed to us in td first.
1628 	 * This is the thread that causes the core dump and thus likely to
1629 	 * be the right thread one wants to have selected in the debugger.
1630 	 */
1631 	thr = td;
1632 	while (thr != NULL) {
1633 		size += register_note(list, NT_PRSTATUS,
1634 		    __elfN(note_prstatus), thr);
1635 		size += register_note(list, NT_FPREGSET,
1636 		    __elfN(note_fpregset), thr);
1637 		size += register_note(list, NT_THRMISC,
1638 		    __elfN(note_thrmisc), thr);
1639 		size += register_note(list, NT_PTLWPINFO,
1640 		    __elfN(note_ptlwpinfo), thr);
1641 		size += register_note(list, -1,
1642 		    __elfN(note_threadmd), thr);
1643 
1644 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1645 		    TAILQ_NEXT(thr, td_plist);
1646 		if (thr == td)
1647 			thr = TAILQ_NEXT(thr, td_plist);
1648 	}
1649 
1650 	size += register_note(list, NT_PROCSTAT_PROC,
1651 	    __elfN(note_procstat_proc), p);
1652 	size += register_note(list, NT_PROCSTAT_FILES,
1653 	    note_procstat_files, p);
1654 	size += register_note(list, NT_PROCSTAT_VMMAP,
1655 	    note_procstat_vmmap, p);
1656 	size += register_note(list, NT_PROCSTAT_GROUPS,
1657 	    note_procstat_groups, p);
1658 	size += register_note(list, NT_PROCSTAT_UMASK,
1659 	    note_procstat_umask, p);
1660 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1661 	    note_procstat_rlimit, p);
1662 	size += register_note(list, NT_PROCSTAT_OSREL,
1663 	    note_procstat_osrel, p);
1664 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1665 	    __elfN(note_procstat_psstrings), p);
1666 	size += register_note(list, NT_PROCSTAT_AUXV,
1667 	    __elfN(note_procstat_auxv), p);
1668 
1669 	*sizep = size;
1670 }
1671 
1672 static void
1673 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1674     size_t notesz)
1675 {
1676 	Elf_Ehdr *ehdr;
1677 	Elf_Phdr *phdr;
1678 	Elf_Shdr *shdr;
1679 	struct phdr_closure phc;
1680 
1681 	ehdr = (Elf_Ehdr *)hdr;
1682 
1683 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1684 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1685 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1686 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1687 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1688 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1689 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1690 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1691 	ehdr->e_ident[EI_ABIVERSION] = 0;
1692 	ehdr->e_ident[EI_PAD] = 0;
1693 	ehdr->e_type = ET_CORE;
1694 	ehdr->e_machine = td->td_proc->p_elf_machine;
1695 	ehdr->e_version = EV_CURRENT;
1696 	ehdr->e_entry = 0;
1697 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1698 	ehdr->e_flags = td->td_proc->p_elf_flags;
1699 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1700 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1701 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1702 	ehdr->e_shstrndx = SHN_UNDEF;
1703 	if (numsegs + 1 < PN_XNUM) {
1704 		ehdr->e_phnum = numsegs + 1;
1705 		ehdr->e_shnum = 0;
1706 	} else {
1707 		ehdr->e_phnum = PN_XNUM;
1708 		ehdr->e_shnum = 1;
1709 
1710 		ehdr->e_shoff = ehdr->e_phoff +
1711 		    (numsegs + 1) * ehdr->e_phentsize;
1712 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1713 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1714 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1715 
1716 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1717 		memset(shdr, 0, sizeof(*shdr));
1718 		/*
1719 		 * A special first section is used to hold large segment and
1720 		 * section counts.  This was proposed by Sun Microsystems in
1721 		 * Solaris and has been adopted by Linux; the standard ELF
1722 		 * tools are already familiar with the technique.
1723 		 *
1724 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1725 		 * (or 12-7 depending on the version of the document) for more
1726 		 * details.
1727 		 */
1728 		shdr->sh_type = SHT_NULL;
1729 		shdr->sh_size = ehdr->e_shnum;
1730 		shdr->sh_link = ehdr->e_shstrndx;
1731 		shdr->sh_info = numsegs + 1;
1732 	}
1733 
1734 	/*
1735 	 * Fill in the program header entries.
1736 	 */
1737 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1738 
1739 	/* The note segement. */
1740 	phdr->p_type = PT_NOTE;
1741 	phdr->p_offset = hdrsize;
1742 	phdr->p_vaddr = 0;
1743 	phdr->p_paddr = 0;
1744 	phdr->p_filesz = notesz;
1745 	phdr->p_memsz = 0;
1746 	phdr->p_flags = PF_R;
1747 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1748 	phdr++;
1749 
1750 	/* All the writable segments from the program. */
1751 	phc.phdr = phdr;
1752 	phc.offset = round_page(hdrsize + notesz);
1753 	each_dumpable_segment(td, cb_put_phdr, &phc);
1754 }
1755 
1756 static size_t
1757 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1758 {
1759 	struct note_info *ninfo;
1760 	size_t size, notesize;
1761 
1762 	size = 0;
1763 	out(arg, NULL, &size);
1764 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1765 	ninfo->type = type;
1766 	ninfo->outfunc = out;
1767 	ninfo->outarg = arg;
1768 	ninfo->outsize = size;
1769 	TAILQ_INSERT_TAIL(list, ninfo, link);
1770 
1771 	if (type == -1)
1772 		return (size);
1773 
1774 	notesize = sizeof(Elf_Note) +		/* note header */
1775 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1776 						/* note name */
1777 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1778 
1779 	return (notesize);
1780 }
1781 
1782 static size_t
1783 append_note_data(const void *src, void *dst, size_t len)
1784 {
1785 	size_t padded_len;
1786 
1787 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1788 	if (dst != NULL) {
1789 		bcopy(src, dst, len);
1790 		bzero((char *)dst + len, padded_len - len);
1791 	}
1792 	return (padded_len);
1793 }
1794 
1795 size_t
1796 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1797 {
1798 	Elf_Note *note;
1799 	char *buf;
1800 	size_t notesize;
1801 
1802 	buf = dst;
1803 	if (buf != NULL) {
1804 		note = (Elf_Note *)buf;
1805 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1806 		note->n_descsz = size;
1807 		note->n_type = type;
1808 		buf += sizeof(*note);
1809 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1810 		    sizeof(FREEBSD_ABI_VENDOR));
1811 		append_note_data(src, buf, size);
1812 		if (descp != NULL)
1813 			*descp = buf;
1814 	}
1815 
1816 	notesize = sizeof(Elf_Note) +		/* note header */
1817 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1818 						/* note name */
1819 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1820 
1821 	return (notesize);
1822 }
1823 
1824 static void
1825 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1826 {
1827 	Elf_Note note;
1828 	ssize_t old_len, sect_len;
1829 	size_t new_len, descsz, i;
1830 
1831 	if (ninfo->type == -1) {
1832 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1833 		return;
1834 	}
1835 
1836 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1837 	note.n_descsz = ninfo->outsize;
1838 	note.n_type = ninfo->type;
1839 
1840 	sbuf_bcat(sb, &note, sizeof(note));
1841 	sbuf_start_section(sb, &old_len);
1842 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1843 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1844 	if (note.n_descsz == 0)
1845 		return;
1846 	sbuf_start_section(sb, &old_len);
1847 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1848 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1849 	if (sect_len < 0)
1850 		return;
1851 
1852 	new_len = (size_t)sect_len;
1853 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1854 	if (new_len < descsz) {
1855 		/*
1856 		 * It is expected that individual note emitters will correctly
1857 		 * predict their expected output size and fill up to that size
1858 		 * themselves, padding in a format-specific way if needed.
1859 		 * However, in case they don't, just do it here with zeros.
1860 		 */
1861 		for (i = 0; i < descsz - new_len; i++)
1862 			sbuf_putc(sb, 0);
1863 	} else if (new_len > descsz) {
1864 		/*
1865 		 * We can't always truncate sb -- we may have drained some
1866 		 * of it already.
1867 		 */
1868 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1869 		    "read it (%zu > %zu).  Since it is longer than "
1870 		    "expected, this coredump's notes are corrupt.  THIS "
1871 		    "IS A BUG in the note_procstat routine for type %u.\n",
1872 		    __func__, (unsigned)note.n_type, new_len, descsz,
1873 		    (unsigned)note.n_type));
1874 	}
1875 }
1876 
1877 /*
1878  * Miscellaneous note out functions.
1879  */
1880 
1881 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1882 #include <compat/freebsd32/freebsd32.h>
1883 #include <compat/freebsd32/freebsd32_signal.h>
1884 
1885 typedef struct prstatus32 elf_prstatus_t;
1886 typedef struct prpsinfo32 elf_prpsinfo_t;
1887 typedef struct fpreg32 elf_prfpregset_t;
1888 typedef struct fpreg32 elf_fpregset_t;
1889 typedef struct reg32 elf_gregset_t;
1890 typedef struct thrmisc32 elf_thrmisc_t;
1891 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1892 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1893 typedef uint32_t elf_ps_strings_t;
1894 #else
1895 typedef prstatus_t elf_prstatus_t;
1896 typedef prpsinfo_t elf_prpsinfo_t;
1897 typedef prfpregset_t elf_prfpregset_t;
1898 typedef prfpregset_t elf_fpregset_t;
1899 typedef gregset_t elf_gregset_t;
1900 typedef thrmisc_t elf_thrmisc_t;
1901 #define ELF_KERN_PROC_MASK	0
1902 typedef struct kinfo_proc elf_kinfo_proc_t;
1903 typedef vm_offset_t elf_ps_strings_t;
1904 #endif
1905 
1906 static void
1907 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1908 {
1909 	struct sbuf sbarg;
1910 	size_t len;
1911 	char *cp, *end;
1912 	struct proc *p;
1913 	elf_prpsinfo_t *psinfo;
1914 	int error;
1915 
1916 	p = (struct proc *)arg;
1917 	if (sb != NULL) {
1918 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1919 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1920 		psinfo->pr_version = PRPSINFO_VERSION;
1921 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1922 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1923 		PROC_LOCK(p);
1924 		if (p->p_args != NULL) {
1925 			len = sizeof(psinfo->pr_psargs) - 1;
1926 			if (len > p->p_args->ar_length)
1927 				len = p->p_args->ar_length;
1928 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1929 			PROC_UNLOCK(p);
1930 			error = 0;
1931 		} else {
1932 			_PHOLD(p);
1933 			PROC_UNLOCK(p);
1934 			sbuf_new(&sbarg, psinfo->pr_psargs,
1935 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1936 			error = proc_getargv(curthread, p, &sbarg);
1937 			PRELE(p);
1938 			if (sbuf_finish(&sbarg) == 0)
1939 				len = sbuf_len(&sbarg) - 1;
1940 			else
1941 				len = sizeof(psinfo->pr_psargs) - 1;
1942 			sbuf_delete(&sbarg);
1943 		}
1944 		if (error || len == 0)
1945 			strlcpy(psinfo->pr_psargs, p->p_comm,
1946 			    sizeof(psinfo->pr_psargs));
1947 		else {
1948 			KASSERT(len < sizeof(psinfo->pr_psargs),
1949 			    ("len is too long: %zu vs %zu", len,
1950 			    sizeof(psinfo->pr_psargs)));
1951 			cp = psinfo->pr_psargs;
1952 			end = cp + len - 1;
1953 			for (;;) {
1954 				cp = memchr(cp, '\0', end - cp);
1955 				if (cp == NULL)
1956 					break;
1957 				*cp = ' ';
1958 			}
1959 		}
1960 		psinfo->pr_pid = p->p_pid;
1961 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1962 		free(psinfo, M_TEMP);
1963 	}
1964 	*sizep = sizeof(*psinfo);
1965 }
1966 
1967 static void
1968 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1969 {
1970 	struct thread *td;
1971 	elf_prstatus_t *status;
1972 
1973 	td = (struct thread *)arg;
1974 	if (sb != NULL) {
1975 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1976 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1977 		status->pr_version = PRSTATUS_VERSION;
1978 		status->pr_statussz = sizeof(elf_prstatus_t);
1979 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1980 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1981 		status->pr_osreldate = osreldate;
1982 		status->pr_cursig = td->td_proc->p_sig;
1983 		status->pr_pid = td->td_tid;
1984 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1985 		fill_regs32(td, &status->pr_reg);
1986 #else
1987 		fill_regs(td, &status->pr_reg);
1988 #endif
1989 		sbuf_bcat(sb, status, sizeof(*status));
1990 		free(status, M_TEMP);
1991 	}
1992 	*sizep = sizeof(*status);
1993 }
1994 
1995 static void
1996 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1997 {
1998 	struct thread *td;
1999 	elf_prfpregset_t *fpregset;
2000 
2001 	td = (struct thread *)arg;
2002 	if (sb != NULL) {
2003 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2004 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2005 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2006 		fill_fpregs32(td, fpregset);
2007 #else
2008 		fill_fpregs(td, fpregset);
2009 #endif
2010 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2011 		free(fpregset, M_TEMP);
2012 	}
2013 	*sizep = sizeof(*fpregset);
2014 }
2015 
2016 static void
2017 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2018 {
2019 	struct thread *td;
2020 	elf_thrmisc_t thrmisc;
2021 
2022 	td = (struct thread *)arg;
2023 	if (sb != NULL) {
2024 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2025 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2026 		strcpy(thrmisc.pr_tname, td->td_name);
2027 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2028 	}
2029 	*sizep = sizeof(thrmisc);
2030 }
2031 
2032 static void
2033 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2034 {
2035 	struct thread *td;
2036 	size_t size;
2037 	int structsize;
2038 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2039 	struct ptrace_lwpinfo32 pl;
2040 #else
2041 	struct ptrace_lwpinfo pl;
2042 #endif
2043 
2044 	td = (struct thread *)arg;
2045 	size = sizeof(structsize) + sizeof(pl);
2046 	if (sb != NULL) {
2047 		KASSERT(*sizep == size, ("invalid size"));
2048 		structsize = sizeof(pl);
2049 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2050 		bzero(&pl, sizeof(pl));
2051 		pl.pl_lwpid = td->td_tid;
2052 		pl.pl_event = PL_EVENT_NONE;
2053 		pl.pl_sigmask = td->td_sigmask;
2054 		pl.pl_siglist = td->td_siglist;
2055 		if (td->td_si.si_signo != 0) {
2056 			pl.pl_event = PL_EVENT_SIGNAL;
2057 			pl.pl_flags |= PL_FLAG_SI;
2058 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2059 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2060 #else
2061 			pl.pl_siginfo = td->td_si;
2062 #endif
2063 		}
2064 		strcpy(pl.pl_tdname, td->td_name);
2065 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2066 		sbuf_bcat(sb, &pl, sizeof(pl));
2067 	}
2068 	*sizep = size;
2069 }
2070 
2071 /*
2072  * Allow for MD specific notes, as well as any MD
2073  * specific preparations for writing MI notes.
2074  */
2075 static void
2076 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2077 {
2078 	struct thread *td;
2079 	void *buf;
2080 	size_t size;
2081 
2082 	td = (struct thread *)arg;
2083 	size = *sizep;
2084 	if (size != 0 && sb != NULL)
2085 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2086 	else
2087 		buf = NULL;
2088 	size = 0;
2089 	__elfN(dump_thread)(td, buf, &size);
2090 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2091 	if (size != 0 && sb != NULL)
2092 		sbuf_bcat(sb, buf, size);
2093 	free(buf, M_TEMP);
2094 	*sizep = size;
2095 }
2096 
2097 #ifdef KINFO_PROC_SIZE
2098 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2099 #endif
2100 
2101 static void
2102 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2103 {
2104 	struct proc *p;
2105 	size_t size;
2106 	int structsize;
2107 
2108 	p = (struct proc *)arg;
2109 	size = sizeof(structsize) + p->p_numthreads *
2110 	    sizeof(elf_kinfo_proc_t);
2111 
2112 	if (sb != NULL) {
2113 		KASSERT(*sizep == size, ("invalid size"));
2114 		structsize = sizeof(elf_kinfo_proc_t);
2115 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2116 		sx_slock(&proctree_lock);
2117 		PROC_LOCK(p);
2118 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2119 		sx_sunlock(&proctree_lock);
2120 	}
2121 	*sizep = size;
2122 }
2123 
2124 #ifdef KINFO_FILE_SIZE
2125 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2126 #endif
2127 
2128 static void
2129 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2130 {
2131 	struct proc *p;
2132 	size_t size, sect_sz, i;
2133 	ssize_t start_len, sect_len;
2134 	int structsize, filedesc_flags;
2135 
2136 	if (coredump_pack_fileinfo)
2137 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2138 	else
2139 		filedesc_flags = 0;
2140 
2141 	p = (struct proc *)arg;
2142 	structsize = sizeof(struct kinfo_file);
2143 	if (sb == NULL) {
2144 		size = 0;
2145 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2146 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2147 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2148 		PROC_LOCK(p);
2149 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2150 		sbuf_finish(sb);
2151 		sbuf_delete(sb);
2152 		*sizep = size;
2153 	} else {
2154 		sbuf_start_section(sb, &start_len);
2155 
2156 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2157 		PROC_LOCK(p);
2158 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2159 		    filedesc_flags);
2160 
2161 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2162 		if (sect_len < 0)
2163 			return;
2164 		sect_sz = sect_len;
2165 
2166 		KASSERT(sect_sz <= *sizep,
2167 		    ("kern_proc_filedesc_out did not respect maxlen; "
2168 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2169 		     sect_sz - sizeof(structsize)));
2170 
2171 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2172 			sbuf_putc(sb, 0);
2173 	}
2174 }
2175 
2176 #ifdef KINFO_VMENTRY_SIZE
2177 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2178 #endif
2179 
2180 static void
2181 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2182 {
2183 	struct proc *p;
2184 	size_t size;
2185 	int structsize, vmmap_flags;
2186 
2187 	if (coredump_pack_vmmapinfo)
2188 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2189 	else
2190 		vmmap_flags = 0;
2191 
2192 	p = (struct proc *)arg;
2193 	structsize = sizeof(struct kinfo_vmentry);
2194 	if (sb == NULL) {
2195 		size = 0;
2196 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2197 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2198 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2199 		PROC_LOCK(p);
2200 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2201 		sbuf_finish(sb);
2202 		sbuf_delete(sb);
2203 		*sizep = size;
2204 	} else {
2205 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2206 		PROC_LOCK(p);
2207 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2208 		    vmmap_flags);
2209 	}
2210 }
2211 
2212 static void
2213 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2214 {
2215 	struct proc *p;
2216 	size_t size;
2217 	int structsize;
2218 
2219 	p = (struct proc *)arg;
2220 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2221 	if (sb != NULL) {
2222 		KASSERT(*sizep == size, ("invalid size"));
2223 		structsize = sizeof(gid_t);
2224 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2225 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2226 		    sizeof(gid_t));
2227 	}
2228 	*sizep = size;
2229 }
2230 
2231 static void
2232 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2233 {
2234 	struct proc *p;
2235 	size_t size;
2236 	int structsize;
2237 
2238 	p = (struct proc *)arg;
2239 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2240 	if (sb != NULL) {
2241 		KASSERT(*sizep == size, ("invalid size"));
2242 		structsize = sizeof(p->p_fd->fd_cmask);
2243 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2244 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2245 	}
2246 	*sizep = size;
2247 }
2248 
2249 static void
2250 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2251 {
2252 	struct proc *p;
2253 	struct rlimit rlim[RLIM_NLIMITS];
2254 	size_t size;
2255 	int structsize, i;
2256 
2257 	p = (struct proc *)arg;
2258 	size = sizeof(structsize) + sizeof(rlim);
2259 	if (sb != NULL) {
2260 		KASSERT(*sizep == size, ("invalid size"));
2261 		structsize = sizeof(rlim);
2262 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2263 		PROC_LOCK(p);
2264 		for (i = 0; i < RLIM_NLIMITS; i++)
2265 			lim_rlimit_proc(p, i, &rlim[i]);
2266 		PROC_UNLOCK(p);
2267 		sbuf_bcat(sb, rlim, sizeof(rlim));
2268 	}
2269 	*sizep = size;
2270 }
2271 
2272 static void
2273 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2274 {
2275 	struct proc *p;
2276 	size_t size;
2277 	int structsize;
2278 
2279 	p = (struct proc *)arg;
2280 	size = sizeof(structsize) + sizeof(p->p_osrel);
2281 	if (sb != NULL) {
2282 		KASSERT(*sizep == size, ("invalid size"));
2283 		structsize = sizeof(p->p_osrel);
2284 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2285 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2286 	}
2287 	*sizep = size;
2288 }
2289 
2290 static void
2291 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2292 {
2293 	struct proc *p;
2294 	elf_ps_strings_t ps_strings;
2295 	size_t size;
2296 	int structsize;
2297 
2298 	p = (struct proc *)arg;
2299 	size = sizeof(structsize) + sizeof(ps_strings);
2300 	if (sb != NULL) {
2301 		KASSERT(*sizep == size, ("invalid size"));
2302 		structsize = sizeof(ps_strings);
2303 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2304 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2305 #else
2306 		ps_strings = p->p_sysent->sv_psstrings;
2307 #endif
2308 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2309 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2310 	}
2311 	*sizep = size;
2312 }
2313 
2314 static void
2315 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2316 {
2317 	struct proc *p;
2318 	size_t size;
2319 	int structsize;
2320 
2321 	p = (struct proc *)arg;
2322 	if (sb == NULL) {
2323 		size = 0;
2324 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2325 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2326 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2327 		PHOLD(p);
2328 		proc_getauxv(curthread, p, sb);
2329 		PRELE(p);
2330 		sbuf_finish(sb);
2331 		sbuf_delete(sb);
2332 		*sizep = size;
2333 	} else {
2334 		structsize = sizeof(Elf_Auxinfo);
2335 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2336 		PHOLD(p);
2337 		proc_getauxv(curthread, p, sb);
2338 		PRELE(p);
2339 	}
2340 }
2341 
2342 static boolean_t
2343 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2344     int32_t *osrel, const Elf_Phdr *pnote)
2345 {
2346 	const Elf_Note *note, *note0, *note_end;
2347 	const char *note_name;
2348 	char *buf;
2349 	int i, error;
2350 	boolean_t res;
2351 
2352 	/* We need some limit, might as well use PAGE_SIZE. */
2353 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2354 		return (FALSE);
2355 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2356 	if (pnote->p_offset > PAGE_SIZE ||
2357 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2358 		VOP_UNLOCK(imgp->vp, 0);
2359 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2360 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2361 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2362 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2363 		    curthread->td_ucred, NOCRED, NULL, curthread);
2364 		if (error != 0) {
2365 			uprintf("i/o error PT_NOTE\n");
2366 			res = FALSE;
2367 			goto ret;
2368 		}
2369 		note = note0 = (const Elf_Note *)buf;
2370 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2371 	} else {
2372 		note = note0 = (const Elf_Note *)(imgp->image_header +
2373 		    pnote->p_offset);
2374 		note_end = (const Elf_Note *)(imgp->image_header +
2375 		    pnote->p_offset + pnote->p_filesz);
2376 		buf = NULL;
2377 	}
2378 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2379 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2380 		    (const char *)note < sizeof(Elf_Note)) {
2381 			res = FALSE;
2382 			goto ret;
2383 		}
2384 		if (note->n_namesz != checknote->hdr.n_namesz ||
2385 		    note->n_descsz != checknote->hdr.n_descsz ||
2386 		    note->n_type != checknote->hdr.n_type)
2387 			goto nextnote;
2388 		note_name = (const char *)(note + 1);
2389 		if (note_name + checknote->hdr.n_namesz >=
2390 		    (const char *)note_end || strncmp(checknote->vendor,
2391 		    note_name, checknote->hdr.n_namesz) != 0)
2392 			goto nextnote;
2393 
2394 		/*
2395 		 * Fetch the osreldate for binary
2396 		 * from the ELF OSABI-note if necessary.
2397 		 */
2398 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2399 		    checknote->trans_osrel != NULL) {
2400 			res = checknote->trans_osrel(note, osrel);
2401 			goto ret;
2402 		}
2403 		res = TRUE;
2404 		goto ret;
2405 nextnote:
2406 		note = (const Elf_Note *)((const char *)(note + 1) +
2407 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2408 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2409 	}
2410 	res = FALSE;
2411 ret:
2412 	free(buf, M_TEMP);
2413 	return (res);
2414 }
2415 
2416 /*
2417  * Try to find the appropriate ABI-note section for checknote,
2418  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2419  * first page of the image is searched, the same as for headers.
2420  */
2421 static boolean_t
2422 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2423     int32_t *osrel)
2424 {
2425 	const Elf_Phdr *phdr;
2426 	const Elf_Ehdr *hdr;
2427 	int i;
2428 
2429 	hdr = (const Elf_Ehdr *)imgp->image_header;
2430 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2431 
2432 	for (i = 0; i < hdr->e_phnum; i++) {
2433 		if (phdr[i].p_type == PT_NOTE &&
2434 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2435 			return (TRUE);
2436 	}
2437 	return (FALSE);
2438 
2439 }
2440 
2441 /*
2442  * Tell kern_execve.c about it, with a little help from the linker.
2443  */
2444 static struct execsw __elfN(execsw) = {
2445 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2446 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2447 };
2448 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2449 
2450 static vm_prot_t
2451 __elfN(trans_prot)(Elf_Word flags)
2452 {
2453 	vm_prot_t prot;
2454 
2455 	prot = 0;
2456 	if (flags & PF_X)
2457 		prot |= VM_PROT_EXECUTE;
2458 	if (flags & PF_W)
2459 		prot |= VM_PROT_WRITE;
2460 	if (flags & PF_R)
2461 		prot |= VM_PROT_READ;
2462 #if __ELF_WORD_SIZE == 32
2463 #if defined(__amd64__)
2464 	if (i386_read_exec && (flags & PF_R))
2465 		prot |= VM_PROT_EXECUTE;
2466 #endif
2467 #endif
2468 	return (prot);
2469 }
2470 
2471 static Elf_Word
2472 __elfN(untrans_prot)(vm_prot_t prot)
2473 {
2474 	Elf_Word flags;
2475 
2476 	flags = 0;
2477 	if (prot & VM_PROT_EXECUTE)
2478 		flags |= PF_X;
2479 	if (prot & VM_PROT_READ)
2480 		flags |= PF_R;
2481 	if (prot & VM_PROT_WRITE)
2482 		flags |= PF_W;
2483 	return (flags);
2484 }
2485