xref: /freebsd/sys/kern/imgact_elf.c (revision 63a938566d524836885917d95bd491aa4400b181)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 #include "opt_compat.h"
39 
40 #include <sys/param.h>
41 #include <sys/capsicum.h>
42 #include <sys/compressor.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/jail.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mman.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
55 #include <sys/proc.h>
56 #include <sys/procfs.h>
57 #include <sys/ptrace.h>
58 #include <sys/racct.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
61 #include <sys/sbuf.h>
62 #include <sys/sf_buf.h>
63 #include <sys/smp.h>
64 #include <sys/systm.h>
65 #include <sys/signalvar.h>
66 #include <sys/stat.h>
67 #include <sys/sx.h>
68 #include <sys/syscall.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/vnode.h>
72 #include <sys/syslog.h>
73 #include <sys/eventhandler.h>
74 #include <sys/user.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_extern.h>
83 
84 #include <machine/elf.h>
85 #include <machine/md_var.h>
86 
87 #define ELF_NOTE_ROUNDSIZE	4
88 #define OLD_EI_BRAND	8
89 
90 static int __elfN(check_header)(const Elf_Ehdr *hdr);
91 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
92     const char *interp, int interp_name_len, int32_t *osrel);
93 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94     u_long *entry, size_t pagesize);
95 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
96     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
97     size_t pagesize);
98 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
99 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
100     int32_t *osrel);
101 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
102 static boolean_t __elfN(check_note)(struct image_params *imgp,
103     Elf_Brandnote *checknote, int32_t *osrel);
104 static vm_prot_t __elfN(trans_prot)(Elf_Word);
105 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
106 
107 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108     "");
109 
110 #define	CORE_BUF_SIZE	(16 * 1024)
111 
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119     &elf_legacy_coredump, 0,
120     "include all and only RW pages in core dumps");
121 
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
124     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
125 	1;
126 #else
127 	0;
128 #endif
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 
133 #if __ELF_WORD_SIZE == 32
134 #if defined(__amd64__)
135 int i386_read_exec = 0;
136 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
137     "enable execution from readable segments");
138 #endif
139 #endif
140 
141 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
142 
143 #define	trunc_page_ps(va, ps)	rounddown2(va, ps)
144 #define	round_page_ps(va, ps)	roundup2(va, ps)
145 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
146 
147 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
148 
149 Elf_Brandnote __elfN(freebsd_brandnote) = {
150 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
151 	.hdr.n_descsz	= sizeof(int32_t),
152 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
153 	.vendor		= FREEBSD_ABI_VENDOR,
154 	.flags		= BN_TRANSLATE_OSREL,
155 	.trans_osrel	= __elfN(freebsd_trans_osrel)
156 };
157 
158 static bool
159 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
160 {
161 	uintptr_t p;
162 
163 	p = (uintptr_t)(note + 1);
164 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
165 	*osrel = *(const int32_t *)(p);
166 
167 	return (true);
168 }
169 
170 static const char GNU_ABI_VENDOR[] = "GNU";
171 static int GNU_KFREEBSD_ABI_DESC = 3;
172 
173 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
174 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
175 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
176 	.hdr.n_type	= 1,
177 	.vendor		= GNU_ABI_VENDOR,
178 	.flags		= BN_TRANSLATE_OSREL,
179 	.trans_osrel	= kfreebsd_trans_osrel
180 };
181 
182 static bool
183 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
184 {
185 	const Elf32_Word *desc;
186 	uintptr_t p;
187 
188 	p = (uintptr_t)(note + 1);
189 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
190 
191 	desc = (const Elf32_Word *)p;
192 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
193 		return (false);
194 
195 	/*
196 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
197 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
198 	 */
199 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
200 
201 	return (true);
202 }
203 
204 int
205 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
206 {
207 	int i;
208 
209 	for (i = 0; i < MAX_BRANDS; i++) {
210 		if (elf_brand_list[i] == NULL) {
211 			elf_brand_list[i] = entry;
212 			break;
213 		}
214 	}
215 	if (i == MAX_BRANDS) {
216 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
217 			__func__, entry);
218 		return (-1);
219 	}
220 	return (0);
221 }
222 
223 int
224 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
225 {
226 	int i;
227 
228 	for (i = 0; i < MAX_BRANDS; i++) {
229 		if (elf_brand_list[i] == entry) {
230 			elf_brand_list[i] = NULL;
231 			break;
232 		}
233 	}
234 	if (i == MAX_BRANDS)
235 		return (-1);
236 	return (0);
237 }
238 
239 int
240 __elfN(brand_inuse)(Elf_Brandinfo *entry)
241 {
242 	struct proc *p;
243 	int rval = FALSE;
244 
245 	sx_slock(&allproc_lock);
246 	FOREACH_PROC_IN_SYSTEM(p) {
247 		if (p->p_sysent == entry->sysvec) {
248 			rval = TRUE;
249 			break;
250 		}
251 	}
252 	sx_sunlock(&allproc_lock);
253 
254 	return (rval);
255 }
256 
257 static Elf_Brandinfo *
258 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
259     int interp_name_len, int32_t *osrel)
260 {
261 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
262 	Elf_Brandinfo *bi, *bi_m;
263 	boolean_t ret;
264 	int i;
265 
266 	/*
267 	 * We support four types of branding -- (1) the ELF EI_OSABI field
268 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
269 	 * branding w/in the ELF header, (3) path of the `interp_path'
270 	 * field, and (4) the ".note.ABI-tag" ELF section.
271 	 */
272 
273 	/* Look for an ".note.ABI-tag" ELF section */
274 	bi_m = NULL;
275 	for (i = 0; i < MAX_BRANDS; i++) {
276 		bi = elf_brand_list[i];
277 		if (bi == NULL)
278 			continue;
279 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
280 			continue;
281 		if (hdr->e_machine == bi->machine && (bi->flags &
282 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
283 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
284 			/* Give brand a chance to veto check_note's guess */
285 			if (ret && bi->header_supported)
286 				ret = bi->header_supported(imgp);
287 			/*
288 			 * If note checker claimed the binary, but the
289 			 * interpreter path in the image does not
290 			 * match default one for the brand, try to
291 			 * search for other brands with the same
292 			 * interpreter.  Either there is better brand
293 			 * with the right interpreter, or, failing
294 			 * this, we return first brand which accepted
295 			 * our note and, optionally, header.
296 			 */
297 			if (ret && bi_m == NULL && interp != NULL &&
298 			    (bi->interp_path == NULL ||
299 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
300 			    strncmp(interp, bi->interp_path, interp_name_len)
301 			    != 0))) {
302 				bi_m = bi;
303 				ret = 0;
304 			}
305 			if (ret)
306 				return (bi);
307 		}
308 	}
309 	if (bi_m != NULL)
310 		return (bi_m);
311 
312 	/* If the executable has a brand, search for it in the brand list. */
313 	for (i = 0; i < MAX_BRANDS; i++) {
314 		bi = elf_brand_list[i];
315 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
316 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
317 			continue;
318 		if (hdr->e_machine == bi->machine &&
319 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
320 		    (bi->compat_3_brand != NULL &&
321 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
322 		    bi->compat_3_brand) == 0))) {
323 			/* Looks good, but give brand a chance to veto */
324 			if (bi->header_supported == NULL ||
325 			    bi->header_supported(imgp)) {
326 				/*
327 				 * Again, prefer strictly matching
328 				 * interpreter path.
329 				 */
330 				if (interp_name_len == 0 &&
331 				    bi->interp_path == NULL)
332 					return (bi);
333 				if (bi->interp_path != NULL &&
334 				    strlen(bi->interp_path) + 1 ==
335 				    interp_name_len && strncmp(interp,
336 				    bi->interp_path, interp_name_len) == 0)
337 					return (bi);
338 				if (bi_m == NULL)
339 					bi_m = bi;
340 			}
341 		}
342 	}
343 	if (bi_m != NULL)
344 		return (bi_m);
345 
346 	/* No known brand, see if the header is recognized by any brand */
347 	for (i = 0; i < MAX_BRANDS; i++) {
348 		bi = elf_brand_list[i];
349 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
350 		    bi->header_supported == NULL)
351 			continue;
352 		if (hdr->e_machine == bi->machine) {
353 			ret = bi->header_supported(imgp);
354 			if (ret)
355 				return (bi);
356 		}
357 	}
358 
359 	/* Lacking a known brand, search for a recognized interpreter. */
360 	if (interp != NULL) {
361 		for (i = 0; i < MAX_BRANDS; i++) {
362 			bi = elf_brand_list[i];
363 			if (bi == NULL || (bi->flags &
364 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
365 			    != 0)
366 				continue;
367 			if (hdr->e_machine == bi->machine &&
368 			    bi->interp_path != NULL &&
369 			    /* ELF image p_filesz includes terminating zero */
370 			    strlen(bi->interp_path) + 1 == interp_name_len &&
371 			    strncmp(interp, bi->interp_path, interp_name_len)
372 			    == 0 && (bi->header_supported == NULL ||
373 			    bi->header_supported(imgp)))
374 				return (bi);
375 		}
376 	}
377 
378 	/* Lacking a recognized interpreter, try the default brand */
379 	for (i = 0; i < MAX_BRANDS; i++) {
380 		bi = elf_brand_list[i];
381 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
382 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
383 			continue;
384 		if (hdr->e_machine == bi->machine &&
385 		    __elfN(fallback_brand) == bi->brand &&
386 		    (bi->header_supported == NULL ||
387 		    bi->header_supported(imgp)))
388 			return (bi);
389 	}
390 	return (NULL);
391 }
392 
393 static int
394 __elfN(check_header)(const Elf_Ehdr *hdr)
395 {
396 	Elf_Brandinfo *bi;
397 	int i;
398 
399 	if (!IS_ELF(*hdr) ||
400 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
401 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
402 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
403 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
404 	    hdr->e_version != ELF_TARG_VER)
405 		return (ENOEXEC);
406 
407 	/*
408 	 * Make sure we have at least one brand for this machine.
409 	 */
410 
411 	for (i = 0; i < MAX_BRANDS; i++) {
412 		bi = elf_brand_list[i];
413 		if (bi != NULL && bi->machine == hdr->e_machine)
414 			break;
415 	}
416 	if (i == MAX_BRANDS)
417 		return (ENOEXEC);
418 
419 	return (0);
420 }
421 
422 static int
423 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
424     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
425 {
426 	struct sf_buf *sf;
427 	int error;
428 	vm_offset_t off;
429 
430 	/*
431 	 * Create the page if it doesn't exist yet. Ignore errors.
432 	 */
433 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
434 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
435 
436 	/*
437 	 * Find the page from the underlying object.
438 	 */
439 	if (object != NULL) {
440 		sf = vm_imgact_map_page(object, offset);
441 		if (sf == NULL)
442 			return (KERN_FAILURE);
443 		off = offset - trunc_page(offset);
444 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
445 		    end - start);
446 		vm_imgact_unmap_page(sf);
447 		if (error != 0)
448 			return (KERN_FAILURE);
449 	}
450 
451 	return (KERN_SUCCESS);
452 }
453 
454 static int
455 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
456     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
457     int cow)
458 {
459 	struct sf_buf *sf;
460 	vm_offset_t off;
461 	vm_size_t sz;
462 	int error, locked, rv;
463 
464 	if (start != trunc_page(start)) {
465 		rv = __elfN(map_partial)(map, object, offset, start,
466 		    round_page(start), prot);
467 		if (rv != KERN_SUCCESS)
468 			return (rv);
469 		offset += round_page(start) - start;
470 		start = round_page(start);
471 	}
472 	if (end != round_page(end)) {
473 		rv = __elfN(map_partial)(map, object, offset +
474 		    trunc_page(end) - start, trunc_page(end), end, prot);
475 		if (rv != KERN_SUCCESS)
476 			return (rv);
477 		end = trunc_page(end);
478 	}
479 	if (start >= end)
480 		return (KERN_SUCCESS);
481 	if ((offset & PAGE_MASK) != 0) {
482 		/*
483 		 * The mapping is not page aligned.  This means that we have
484 		 * to copy the data.
485 		 */
486 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
487 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
488 		if (rv != KERN_SUCCESS)
489 			return (rv);
490 		if (object == NULL)
491 			return (KERN_SUCCESS);
492 		for (; start < end; start += sz) {
493 			sf = vm_imgact_map_page(object, offset);
494 			if (sf == NULL)
495 				return (KERN_FAILURE);
496 			off = offset - trunc_page(offset);
497 			sz = end - start;
498 			if (sz > PAGE_SIZE - off)
499 				sz = PAGE_SIZE - off;
500 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
501 			    (caddr_t)start, sz);
502 			vm_imgact_unmap_page(sf);
503 			if (error != 0)
504 				return (KERN_FAILURE);
505 			offset += sz;
506 		}
507 	} else {
508 		vm_object_reference(object);
509 		rv = vm_map_fixed(map, object, offset, start, end - start,
510 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
511 		if (rv != KERN_SUCCESS) {
512 			locked = VOP_ISLOCKED(imgp->vp);
513 			VOP_UNLOCK(imgp->vp, 0);
514 			vm_object_deallocate(object);
515 			vn_lock(imgp->vp, locked | LK_RETRY);
516 			return (rv);
517 		}
518 	}
519 	return (KERN_SUCCESS);
520 }
521 
522 static int
523 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
524     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
525     size_t pagesize)
526 {
527 	struct sf_buf *sf;
528 	size_t map_len;
529 	vm_map_t map;
530 	vm_object_t object;
531 	vm_offset_t off, map_addr;
532 	int error, rv, cow;
533 	size_t copy_len;
534 	vm_ooffset_t file_addr;
535 
536 	/*
537 	 * It's necessary to fail if the filsz + offset taken from the
538 	 * header is greater than the actual file pager object's size.
539 	 * If we were to allow this, then the vm_map_find() below would
540 	 * walk right off the end of the file object and into the ether.
541 	 *
542 	 * While I'm here, might as well check for something else that
543 	 * is invalid: filsz cannot be greater than memsz.
544 	 */
545 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
546 	    filsz > memsz) {
547 		uprintf("elf_load_section: truncated ELF file\n");
548 		return (ENOEXEC);
549 	}
550 
551 	object = imgp->object;
552 	map = &imgp->proc->p_vmspace->vm_map;
553 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
554 	file_addr = trunc_page_ps(offset, pagesize);
555 
556 	/*
557 	 * We have two choices.  We can either clear the data in the last page
558 	 * of an oversized mapping, or we can start the anon mapping a page
559 	 * early and copy the initialized data into that first page.  We
560 	 * choose the second.
561 	 */
562 	if (filsz == 0)
563 		map_len = 0;
564 	else if (memsz > filsz)
565 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
566 	else
567 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
568 
569 	if (map_len != 0) {
570 		/* cow flags: don't dump readonly sections in core */
571 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
572 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
573 
574 		rv = __elfN(map_insert)(imgp, map,
575 				      object,
576 				      file_addr,	/* file offset */
577 				      map_addr,		/* virtual start */
578 				      map_addr + map_len,/* virtual end */
579 				      prot,
580 				      cow);
581 		if (rv != KERN_SUCCESS)
582 			return (EINVAL);
583 
584 		/* we can stop now if we've covered it all */
585 		if (memsz == filsz)
586 			return (0);
587 	}
588 
589 
590 	/*
591 	 * We have to get the remaining bit of the file into the first part
592 	 * of the oversized map segment.  This is normally because the .data
593 	 * segment in the file is extended to provide bss.  It's a neat idea
594 	 * to try and save a page, but it's a pain in the behind to implement.
595 	 */
596 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
597 	    filsz, pagesize);
598 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
599 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
600 	    map_addr;
601 
602 	/* This had damn well better be true! */
603 	if (map_len != 0) {
604 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
605 		    map_addr + map_len, prot, 0);
606 		if (rv != KERN_SUCCESS)
607 			return (EINVAL);
608 	}
609 
610 	if (copy_len != 0) {
611 		sf = vm_imgact_map_page(object, offset + filsz);
612 		if (sf == NULL)
613 			return (EIO);
614 
615 		/* send the page fragment to user space */
616 		off = trunc_page_ps(offset + filsz, pagesize) -
617 		    trunc_page(offset + filsz);
618 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
619 		    (caddr_t)map_addr, copy_len);
620 		vm_imgact_unmap_page(sf);
621 		if (error != 0)
622 			return (error);
623 	}
624 
625 	/*
626 	 * Remove write access to the page if it was only granted by map_insert
627 	 * to allow copyout.
628 	 */
629 	if ((prot & VM_PROT_WRITE) == 0)
630 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
631 		    map_len), prot, FALSE);
632 
633 	return (0);
634 }
635 
636 /*
637  * Load the file "file" into memory.  It may be either a shared object
638  * or an executable.
639  *
640  * The "addr" reference parameter is in/out.  On entry, it specifies
641  * the address where a shared object should be loaded.  If the file is
642  * an executable, this value is ignored.  On exit, "addr" specifies
643  * where the file was actually loaded.
644  *
645  * The "entry" reference parameter is out only.  On exit, it specifies
646  * the entry point for the loaded file.
647  */
648 static int
649 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
650 	u_long *entry, size_t pagesize)
651 {
652 	struct {
653 		struct nameidata nd;
654 		struct vattr attr;
655 		struct image_params image_params;
656 	} *tempdata;
657 	const Elf_Ehdr *hdr = NULL;
658 	const Elf_Phdr *phdr = NULL;
659 	struct nameidata *nd;
660 	struct vattr *attr;
661 	struct image_params *imgp;
662 	vm_prot_t prot;
663 	u_long rbase;
664 	u_long base_addr = 0;
665 	int error, i, numsegs;
666 
667 #ifdef CAPABILITY_MODE
668 	/*
669 	 * XXXJA: This check can go away once we are sufficiently confident
670 	 * that the checks in namei() are correct.
671 	 */
672 	if (IN_CAPABILITY_MODE(curthread))
673 		return (ECAPMODE);
674 #endif
675 
676 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
677 	nd = &tempdata->nd;
678 	attr = &tempdata->attr;
679 	imgp = &tempdata->image_params;
680 
681 	/*
682 	 * Initialize part of the common data
683 	 */
684 	imgp->proc = p;
685 	imgp->attr = attr;
686 	imgp->firstpage = NULL;
687 	imgp->image_header = NULL;
688 	imgp->object = NULL;
689 	imgp->execlabel = NULL;
690 
691 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
692 	if ((error = namei(nd)) != 0) {
693 		nd->ni_vp = NULL;
694 		goto fail;
695 	}
696 	NDFREE(nd, NDF_ONLY_PNBUF);
697 	imgp->vp = nd->ni_vp;
698 
699 	/*
700 	 * Check permissions, modes, uid, etc on the file, and "open" it.
701 	 */
702 	error = exec_check_permissions(imgp);
703 	if (error)
704 		goto fail;
705 
706 	error = exec_map_first_page(imgp);
707 	if (error)
708 		goto fail;
709 
710 	/*
711 	 * Also make certain that the interpreter stays the same, so set
712 	 * its VV_TEXT flag, too.
713 	 */
714 	VOP_SET_TEXT(nd->ni_vp);
715 
716 	imgp->object = nd->ni_vp->v_object;
717 
718 	hdr = (const Elf_Ehdr *)imgp->image_header;
719 	if ((error = __elfN(check_header)(hdr)) != 0)
720 		goto fail;
721 	if (hdr->e_type == ET_DYN)
722 		rbase = *addr;
723 	else if (hdr->e_type == ET_EXEC)
724 		rbase = 0;
725 	else {
726 		error = ENOEXEC;
727 		goto fail;
728 	}
729 
730 	/* Only support headers that fit within first page for now      */
731 	if ((hdr->e_phoff > PAGE_SIZE) ||
732 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
733 		error = ENOEXEC;
734 		goto fail;
735 	}
736 
737 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
738 	if (!aligned(phdr, Elf_Addr)) {
739 		error = ENOEXEC;
740 		goto fail;
741 	}
742 
743 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
744 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
745 			/* Loadable segment */
746 			prot = __elfN(trans_prot)(phdr[i].p_flags);
747 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
748 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
749 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
750 			if (error != 0)
751 				goto fail;
752 			/*
753 			 * Establish the base address if this is the
754 			 * first segment.
755 			 */
756 			if (numsegs == 0)
757   				base_addr = trunc_page(phdr[i].p_vaddr +
758 				    rbase);
759 			numsegs++;
760 		}
761 	}
762 	*addr = base_addr;
763 	*entry = (unsigned long)hdr->e_entry + rbase;
764 
765 fail:
766 	if (imgp->firstpage)
767 		exec_unmap_first_page(imgp);
768 
769 	if (nd->ni_vp)
770 		vput(nd->ni_vp);
771 
772 	free(tempdata, M_TEMP);
773 
774 	return (error);
775 }
776 
777 static int
778 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
779 {
780 	struct thread *td;
781 	const Elf_Ehdr *hdr;
782 	const Elf_Phdr *phdr;
783 	Elf_Auxargs *elf_auxargs;
784 	struct vmspace *vmspace;
785 	const char *err_str, *newinterp;
786 	char *interp, *interp_buf, *path;
787 	Elf_Brandinfo *brand_info;
788 	struct sysentvec *sv;
789 	vm_prot_t prot;
790 	u_long text_size, data_size, total_size, text_addr, data_addr;
791 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
792 	int32_t osrel;
793 	int error, i, n, interp_name_len, have_interp;
794 
795 	hdr = (const Elf_Ehdr *)imgp->image_header;
796 
797 	/*
798 	 * Do we have a valid ELF header ?
799 	 *
800 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
801 	 * if particular brand doesn't support it.
802 	 */
803 	if (__elfN(check_header)(hdr) != 0 ||
804 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
805 		return (-1);
806 
807 	/*
808 	 * From here on down, we return an errno, not -1, as we've
809 	 * detected an ELF file.
810 	 */
811 
812 	if ((hdr->e_phoff > PAGE_SIZE) ||
813 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
814 		/* Only support headers in first page for now */
815 		uprintf("Program headers not in the first page\n");
816 		return (ENOEXEC);
817 	}
818 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
819 	if (!aligned(phdr, Elf_Addr)) {
820 		uprintf("Unaligned program headers\n");
821 		return (ENOEXEC);
822 	}
823 
824 	n = error = 0;
825 	baddr = 0;
826 	osrel = 0;
827 	text_size = data_size = total_size = text_addr = data_addr = 0;
828 	entry = proghdr = 0;
829 	interp_name_len = 0;
830 	err_str = newinterp = NULL;
831 	interp = interp_buf = NULL;
832 	td = curthread;
833 
834 	for (i = 0; i < hdr->e_phnum; i++) {
835 		switch (phdr[i].p_type) {
836 		case PT_LOAD:
837 			if (n == 0)
838 				baddr = phdr[i].p_vaddr;
839 			n++;
840 			break;
841 		case PT_INTERP:
842 			/* Path to interpreter */
843 			if (phdr[i].p_filesz > 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\n");
866 					goto ret;
867 				}
868 				interp_buf[interp_name_len] = '\0';
869 				interp = interp_buf;
870 			} else {
871 				interp = __DECONST(char *, imgp->image_header) +
872 				    phdr[i].p_offset;
873 			}
874 			break;
875 		case PT_GNU_STACK:
876 			if (__elfN(nxstack))
877 				imgp->stack_prot =
878 				    __elfN(trans_prot)(phdr[i].p_flags);
879 			imgp->stack_sz = phdr[i].p_memsz;
880 			break;
881 		}
882 	}
883 
884 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
885 	    &osrel);
886 	if (brand_info == NULL) {
887 		uprintf("ELF binary type \"%u\" not known.\n",
888 		    hdr->e_ident[EI_OSABI]);
889 		error = ENOEXEC;
890 		goto ret;
891 	}
892 	et_dyn_addr = 0;
893 	if (hdr->e_type == ET_DYN) {
894 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
895 			uprintf("Cannot execute shared object\n");
896 			error = ENOEXEC;
897 			goto ret;
898 		}
899 		/*
900 		 * Honour the base load address from the dso if it is
901 		 * non-zero for some reason.
902 		 */
903 		if (baddr == 0)
904 			et_dyn_addr = ET_DYN_LOAD_ADDR;
905 	}
906 	sv = brand_info->sysvec;
907 	if (interp != NULL && brand_info->interp_newpath != NULL)
908 		newinterp = brand_info->interp_newpath;
909 
910 	/*
911 	 * Avoid a possible deadlock if the current address space is destroyed
912 	 * and that address space maps the locked vnode.  In the common case,
913 	 * the locked vnode's v_usecount is decremented but remains greater
914 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
915 	 * However, in cases where the vnode lock is external, such as nullfs,
916 	 * v_usecount may become zero.
917 	 *
918 	 * The VV_TEXT flag prevents modifications to the executable while
919 	 * the vnode is unlocked.
920 	 */
921 	VOP_UNLOCK(imgp->vp, 0);
922 
923 	error = exec_new_vmspace(imgp, sv);
924 	imgp->proc->p_sysent = sv;
925 
926 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
927 	if (error != 0)
928 		goto ret;
929 
930 	for (i = 0; i < hdr->e_phnum; i++) {
931 		switch (phdr[i].p_type) {
932 		case PT_LOAD:	/* Loadable segment */
933 			if (phdr[i].p_memsz == 0)
934 				break;
935 			prot = __elfN(trans_prot)(phdr[i].p_flags);
936 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
937 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
938 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
939 			    sv->sv_pagesize);
940 			if (error != 0)
941 				goto ret;
942 
943 			/*
944 			 * If this segment contains the program headers,
945 			 * remember their virtual address for the AT_PHDR
946 			 * aux entry. Static binaries don't usually include
947 			 * a PT_PHDR entry.
948 			 */
949 			if (phdr[i].p_offset == 0 &&
950 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
951 				<= phdr[i].p_filesz)
952 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
953 				    et_dyn_addr;
954 
955 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
956 			seg_size = round_page(phdr[i].p_memsz +
957 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
958 
959 			/*
960 			 * Make the largest executable segment the official
961 			 * text segment and all others data.
962 			 *
963 			 * Note that obreak() assumes that data_addr +
964 			 * data_size == end of data load area, and the ELF
965 			 * file format expects segments to be sorted by
966 			 * address.  If multiple data segments exist, the
967 			 * last one will be used.
968 			 */
969 
970 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
971 				text_size = seg_size;
972 				text_addr = seg_addr;
973 			} else {
974 				data_size = seg_size;
975 				data_addr = seg_addr;
976 			}
977 			total_size += seg_size;
978 			break;
979 		case PT_PHDR: 	/* Program header table info */
980 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
981 			break;
982 		default:
983 			break;
984 		}
985 	}
986 
987 	if (data_addr == 0 && data_size == 0) {
988 		data_addr = text_addr;
989 		data_size = text_size;
990 	}
991 
992 	entry = (u_long)hdr->e_entry + et_dyn_addr;
993 
994 	/*
995 	 * Check limits.  It should be safe to check the
996 	 * limits after loading the segments since we do
997 	 * not actually fault in all the segments pages.
998 	 */
999 	PROC_LOCK(imgp->proc);
1000 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1001 		err_str = "Data segment size exceeds process limit";
1002 	else if (text_size > maxtsiz)
1003 		err_str = "Text segment size exceeds system limit";
1004 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1005 		err_str = "Total segment size exceeds process limit";
1006 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1007 		err_str = "Data segment size exceeds resource limit";
1008 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1009 		err_str = "Total segment size exceeds resource limit";
1010 	if (err_str != NULL) {
1011 		PROC_UNLOCK(imgp->proc);
1012 		uprintf("%s\n", err_str);
1013 		error = ENOMEM;
1014 		goto ret;
1015 	}
1016 
1017 	vmspace = imgp->proc->p_vmspace;
1018 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1019 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1020 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1021 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1022 
1023 	/*
1024 	 * We load the dynamic linker where a userland call
1025 	 * to mmap(0, ...) would put it.  The rationale behind this
1026 	 * calculation is that it leaves room for the heap to grow to
1027 	 * its maximum allowed size.
1028 	 */
1029 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1030 	    RLIMIT_DATA));
1031 	PROC_UNLOCK(imgp->proc);
1032 
1033 	imgp->entry_addr = entry;
1034 
1035 	if (interp != NULL) {
1036 		have_interp = FALSE;
1037 		VOP_UNLOCK(imgp->vp, 0);
1038 		if (brand_info->emul_path != NULL &&
1039 		    brand_info->emul_path[0] != '\0') {
1040 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1041 			snprintf(path, MAXPATHLEN, "%s%s",
1042 			    brand_info->emul_path, interp);
1043 			error = __elfN(load_file)(imgp->proc, path, &addr,
1044 			    &imgp->entry_addr, sv->sv_pagesize);
1045 			free(path, M_TEMP);
1046 			if (error == 0)
1047 				have_interp = TRUE;
1048 		}
1049 		if (!have_interp && newinterp != NULL &&
1050 		    (brand_info->interp_path == NULL ||
1051 		    strcmp(interp, brand_info->interp_path) == 0)) {
1052 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1053 			    &imgp->entry_addr, sv->sv_pagesize);
1054 			if (error == 0)
1055 				have_interp = TRUE;
1056 		}
1057 		if (!have_interp) {
1058 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1059 			    &imgp->entry_addr, sv->sv_pagesize);
1060 		}
1061 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1062 		if (error != 0) {
1063 			uprintf("ELF interpreter %s not found, error %d\n",
1064 			    interp, error);
1065 			goto ret;
1066 		}
1067 	} else
1068 		addr = et_dyn_addr;
1069 
1070 	/*
1071 	 * Construct auxargs table (used by the fixup routine)
1072 	 */
1073 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1074 	elf_auxargs->execfd = -1;
1075 	elf_auxargs->phdr = proghdr;
1076 	elf_auxargs->phent = hdr->e_phentsize;
1077 	elf_auxargs->phnum = hdr->e_phnum;
1078 	elf_auxargs->pagesz = PAGE_SIZE;
1079 	elf_auxargs->base = addr;
1080 	elf_auxargs->flags = 0;
1081 	elf_auxargs->entry = entry;
1082 	elf_auxargs->hdr_eflags = hdr->e_flags;
1083 
1084 	imgp->auxargs = elf_auxargs;
1085 	imgp->interpreted = 0;
1086 	imgp->reloc_base = addr;
1087 	imgp->proc->p_osrel = osrel;
1088 	imgp->proc->p_elf_machine = hdr->e_machine;
1089 	imgp->proc->p_elf_flags = hdr->e_flags;
1090 
1091 ret:
1092 	free(interp_buf, M_TEMP);
1093 	return (error);
1094 }
1095 
1096 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1097 
1098 int
1099 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1100 {
1101 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1102 	Elf_Addr *base;
1103 	Elf_Addr *pos;
1104 
1105 	base = (Elf_Addr *)*stack_base;
1106 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1107 
1108 	if (args->execfd != -1)
1109 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1110 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1111 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1112 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1113 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1114 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1115 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1116 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1117 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1118 	if (imgp->execpathp != 0)
1119 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1120 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1121 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1122 	if (imgp->canary != 0) {
1123 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1124 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1125 	}
1126 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1127 	if (imgp->pagesizes != 0) {
1128 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1129 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1130 	}
1131 	if (imgp->sysent->sv_timekeep_base != 0) {
1132 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1133 		    imgp->sysent->sv_timekeep_base);
1134 	}
1135 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1136 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1137 	    imgp->sysent->sv_stackprot);
1138 	if (imgp->sysent->sv_hwcap != NULL)
1139 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1140 	if (imgp->sysent->sv_hwcap2 != NULL)
1141 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1142 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1143 
1144 	free(imgp->auxargs, M_TEMP);
1145 	imgp->auxargs = NULL;
1146 
1147 	base--;
1148 	suword(base, (long)imgp->args->argc);
1149 	*stack_base = (register_t *)base;
1150 	return (0);
1151 }
1152 
1153 /*
1154  * Code for generating ELF core dumps.
1155  */
1156 
1157 typedef void (*segment_callback)(vm_map_entry_t, void *);
1158 
1159 /* Closure for cb_put_phdr(). */
1160 struct phdr_closure {
1161 	Elf_Phdr *phdr;		/* Program header to fill in */
1162 	Elf_Off offset;		/* Offset of segment in core file */
1163 };
1164 
1165 /* Closure for cb_size_segment(). */
1166 struct sseg_closure {
1167 	int count;		/* Count of writable segments. */
1168 	size_t size;		/* Total size of all writable segments. */
1169 };
1170 
1171 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1172 
1173 struct note_info {
1174 	int		type;		/* Note type. */
1175 	outfunc_t 	outfunc; 	/* Output function. */
1176 	void		*outarg;	/* Argument for the output function. */
1177 	size_t		outsize;	/* Output size. */
1178 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1179 };
1180 
1181 TAILQ_HEAD(note_info_list, note_info);
1182 
1183 /* Coredump output parameters. */
1184 struct coredump_params {
1185 	off_t		offset;
1186 	struct ucred	*active_cred;
1187 	struct ucred	*file_cred;
1188 	struct thread	*td;
1189 	struct vnode	*vp;
1190 	struct compressor *comp;
1191 };
1192 
1193 extern int compress_user_cores;
1194 extern int compress_user_cores_level;
1195 
1196 static void cb_put_phdr(vm_map_entry_t, void *);
1197 static void cb_size_segment(vm_map_entry_t, void *);
1198 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1199     enum uio_seg);
1200 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1201 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1202     struct note_info_list *, size_t);
1203 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1204     size_t *);
1205 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1206 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1207 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1208 static int sbuf_drain_core_output(void *, const char *, int);
1209 static int sbuf_drain_count(void *arg, const char *data, int len);
1210 
1211 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1212 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1213 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1214 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1215 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1216 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1217 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1218 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1219 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1220 static void note_procstat_files(void *, struct sbuf *, size_t *);
1221 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1222 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1223 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1224 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1225 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1226 
1227 /*
1228  * Write out a core segment to the compression stream.
1229  */
1230 static int
1231 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1232 {
1233 	u_int chunk_len;
1234 	int error;
1235 
1236 	while (len > 0) {
1237 		chunk_len = MIN(len, CORE_BUF_SIZE);
1238 
1239 		/*
1240 		 * We can get EFAULT error here.
1241 		 * In that case zero out the current chunk of the segment.
1242 		 */
1243 		error = copyin(base, buf, chunk_len);
1244 		if (error != 0)
1245 			bzero(buf, chunk_len);
1246 		error = compressor_write(p->comp, buf, chunk_len);
1247 		if (error != 0)
1248 			break;
1249 		base += chunk_len;
1250 		len -= chunk_len;
1251 	}
1252 	return (error);
1253 }
1254 
1255 static int
1256 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1257 {
1258 
1259 	return (core_write((struct coredump_params *)arg, base, len, offset,
1260 	    UIO_SYSSPACE));
1261 }
1262 
1263 static int
1264 core_write(struct coredump_params *p, const void *base, size_t len,
1265     off_t offset, enum uio_seg seg)
1266 {
1267 
1268 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1269 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1270 	    p->active_cred, p->file_cred, NULL, p->td));
1271 }
1272 
1273 static int
1274 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1275     void *tmpbuf)
1276 {
1277 	int error;
1278 
1279 	if (p->comp != NULL)
1280 		return (compress_chunk(p, base, tmpbuf, len));
1281 
1282 	/*
1283 	 * EFAULT is a non-fatal error that we can get, for example,
1284 	 * if the segment is backed by a file but extends beyond its
1285 	 * end.
1286 	 */
1287 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1288 	if (error == EFAULT) {
1289 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1290 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1291 		    "for process %s\n", base, len, offset, curproc->p_comm);
1292 
1293 		/*
1294 		 * Write a "real" zero byte at the end of the target region
1295 		 * in the case this is the last segment.
1296 		 * The intermediate space will be implicitly zero-filled.
1297 		 */
1298 		error = core_write(p, zero_region, 1, offset + len - 1,
1299 		    UIO_SYSSPACE);
1300 	}
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Drain into a core file.
1306  */
1307 static int
1308 sbuf_drain_core_output(void *arg, const char *data, int len)
1309 {
1310 	struct coredump_params *p;
1311 	int error, locked;
1312 
1313 	p = (struct coredump_params *)arg;
1314 
1315 	/*
1316 	 * Some kern_proc out routines that print to this sbuf may
1317 	 * call us with the process lock held. Draining with the
1318 	 * non-sleepable lock held is unsafe. The lock is needed for
1319 	 * those routines when dumping a live process. In our case we
1320 	 * can safely release the lock before draining and acquire
1321 	 * again after.
1322 	 */
1323 	locked = PROC_LOCKED(p->td->td_proc);
1324 	if (locked)
1325 		PROC_UNLOCK(p->td->td_proc);
1326 	if (p->comp != NULL)
1327 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1328 	else
1329 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1330 		    UIO_SYSSPACE);
1331 	if (locked)
1332 		PROC_LOCK(p->td->td_proc);
1333 	if (error != 0)
1334 		return (-error);
1335 	p->offset += len;
1336 	return (len);
1337 }
1338 
1339 /*
1340  * Drain into a counter.
1341  */
1342 static int
1343 sbuf_drain_count(void *arg, const char *data __unused, int len)
1344 {
1345 	size_t *sizep;
1346 
1347 	sizep = (size_t *)arg;
1348 	*sizep += len;
1349 	return (len);
1350 }
1351 
1352 int
1353 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1354 {
1355 	struct ucred *cred = td->td_ucred;
1356 	int error = 0;
1357 	struct sseg_closure seginfo;
1358 	struct note_info_list notelst;
1359 	struct coredump_params params;
1360 	struct note_info *ninfo;
1361 	void *hdr, *tmpbuf;
1362 	size_t hdrsize, notesz, coresize;
1363 
1364 	hdr = NULL;
1365 	tmpbuf = NULL;
1366 	TAILQ_INIT(&notelst);
1367 
1368 	/* Size the program segments. */
1369 	seginfo.count = 0;
1370 	seginfo.size = 0;
1371 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1372 
1373 	/*
1374 	 * Collect info about the core file header area.
1375 	 */
1376 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1377 	if (seginfo.count + 1 >= PN_XNUM)
1378 		hdrsize += sizeof(Elf_Shdr);
1379 	__elfN(prepare_notes)(td, &notelst, &notesz);
1380 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1381 
1382 	/* Set up core dump parameters. */
1383 	params.offset = 0;
1384 	params.active_cred = cred;
1385 	params.file_cred = NOCRED;
1386 	params.td = td;
1387 	params.vp = vp;
1388 	params.comp = NULL;
1389 
1390 #ifdef RACCT
1391 	if (racct_enable) {
1392 		PROC_LOCK(td->td_proc);
1393 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1394 		PROC_UNLOCK(td->td_proc);
1395 		if (error != 0) {
1396 			error = EFAULT;
1397 			goto done;
1398 		}
1399 	}
1400 #endif
1401 	if (coresize >= limit) {
1402 		error = EFAULT;
1403 		goto done;
1404 	}
1405 
1406 	/* Create a compression stream if necessary. */
1407 	if (compress_user_cores != 0) {
1408 		params.comp = compressor_init(core_compressed_write,
1409 		    compress_user_cores, CORE_BUF_SIZE,
1410 		    compress_user_cores_level, &params);
1411 		if (params.comp == NULL) {
1412 			error = EFAULT;
1413 			goto done;
1414 		}
1415 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1416         }
1417 
1418 	/*
1419 	 * Allocate memory for building the header, fill it up,
1420 	 * and write it out following the notes.
1421 	 */
1422 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1423 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1424 	    notesz);
1425 
1426 	/* Write the contents of all of the writable segments. */
1427 	if (error == 0) {
1428 		Elf_Phdr *php;
1429 		off_t offset;
1430 		int i;
1431 
1432 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1433 		offset = round_page(hdrsize + notesz);
1434 		for (i = 0; i < seginfo.count; i++) {
1435 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1436 			    php->p_filesz, offset, &params, tmpbuf);
1437 			if (error != 0)
1438 				break;
1439 			offset += php->p_filesz;
1440 			php++;
1441 		}
1442 		if (error == 0 && params.comp != NULL)
1443 			error = compressor_flush(params.comp);
1444 	}
1445 	if (error) {
1446 		log(LOG_WARNING,
1447 		    "Failed to write core file for process %s (error %d)\n",
1448 		    curproc->p_comm, error);
1449 	}
1450 
1451 done:
1452 	free(tmpbuf, M_TEMP);
1453 	if (params.comp != NULL)
1454 		compressor_fini(params.comp);
1455 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1456 		TAILQ_REMOVE(&notelst, ninfo, link);
1457 		free(ninfo, M_TEMP);
1458 	}
1459 	if (hdr != NULL)
1460 		free(hdr, M_TEMP);
1461 
1462 	return (error);
1463 }
1464 
1465 /*
1466  * A callback for each_dumpable_segment() to write out the segment's
1467  * program header entry.
1468  */
1469 static void
1470 cb_put_phdr(vm_map_entry_t entry, void *closure)
1471 {
1472 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1473 	Elf_Phdr *phdr = phc->phdr;
1474 
1475 	phc->offset = round_page(phc->offset);
1476 
1477 	phdr->p_type = PT_LOAD;
1478 	phdr->p_offset = phc->offset;
1479 	phdr->p_vaddr = entry->start;
1480 	phdr->p_paddr = 0;
1481 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1482 	phdr->p_align = PAGE_SIZE;
1483 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1484 
1485 	phc->offset += phdr->p_filesz;
1486 	phc->phdr++;
1487 }
1488 
1489 /*
1490  * A callback for each_dumpable_segment() to gather information about
1491  * the number of segments and their total size.
1492  */
1493 static void
1494 cb_size_segment(vm_map_entry_t entry, void *closure)
1495 {
1496 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1497 
1498 	ssc->count++;
1499 	ssc->size += entry->end - entry->start;
1500 }
1501 
1502 /*
1503  * For each writable segment in the process's memory map, call the given
1504  * function with a pointer to the map entry and some arbitrary
1505  * caller-supplied data.
1506  */
1507 static void
1508 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1509 {
1510 	struct proc *p = td->td_proc;
1511 	vm_map_t map = &p->p_vmspace->vm_map;
1512 	vm_map_entry_t entry;
1513 	vm_object_t backing_object, object;
1514 	boolean_t ignore_entry;
1515 
1516 	vm_map_lock_read(map);
1517 	for (entry = map->header.next; entry != &map->header;
1518 	    entry = entry->next) {
1519 		/*
1520 		 * Don't dump inaccessible mappings, deal with legacy
1521 		 * coredump mode.
1522 		 *
1523 		 * Note that read-only segments related to the elf binary
1524 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1525 		 * need to arbitrarily ignore such segments.
1526 		 */
1527 		if (elf_legacy_coredump) {
1528 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1529 				continue;
1530 		} else {
1531 			if ((entry->protection & VM_PROT_ALL) == 0)
1532 				continue;
1533 		}
1534 
1535 		/*
1536 		 * Dont include memory segment in the coredump if
1537 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1538 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1539 		 * kernel map).
1540 		 */
1541 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1542 			continue;
1543 
1544 		if ((object = entry->object.vm_object) == NULL)
1545 			continue;
1546 
1547 		/* Ignore memory-mapped devices and such things. */
1548 		VM_OBJECT_RLOCK(object);
1549 		while ((backing_object = object->backing_object) != NULL) {
1550 			VM_OBJECT_RLOCK(backing_object);
1551 			VM_OBJECT_RUNLOCK(object);
1552 			object = backing_object;
1553 		}
1554 		ignore_entry = object->type != OBJT_DEFAULT &&
1555 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1556 		    object->type != OBJT_PHYS;
1557 		VM_OBJECT_RUNLOCK(object);
1558 		if (ignore_entry)
1559 			continue;
1560 
1561 		(*func)(entry, closure);
1562 	}
1563 	vm_map_unlock_read(map);
1564 }
1565 
1566 /*
1567  * Write the core file header to the file, including padding up to
1568  * the page boundary.
1569  */
1570 static int
1571 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1572     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1573 {
1574 	struct note_info *ninfo;
1575 	struct sbuf *sb;
1576 	int error;
1577 
1578 	/* Fill in the header. */
1579 	bzero(hdr, hdrsize);
1580 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1581 
1582 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1583 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1584 	sbuf_start_section(sb, NULL);
1585 	sbuf_bcat(sb, hdr, hdrsize);
1586 	TAILQ_FOREACH(ninfo, notelst, link)
1587 	    __elfN(putnote)(ninfo, sb);
1588 	/* Align up to a page boundary for the program segments. */
1589 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1590 	error = sbuf_finish(sb);
1591 	sbuf_delete(sb);
1592 
1593 	return (error);
1594 }
1595 
1596 static void
1597 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1598     size_t *sizep)
1599 {
1600 	struct proc *p;
1601 	struct thread *thr;
1602 	size_t size;
1603 
1604 	p = td->td_proc;
1605 	size = 0;
1606 
1607 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1608 
1609 	/*
1610 	 * To have the debugger select the right thread (LWP) as the initial
1611 	 * thread, we dump the state of the thread passed to us in td first.
1612 	 * This is the thread that causes the core dump and thus likely to
1613 	 * be the right thread one wants to have selected in the debugger.
1614 	 */
1615 	thr = td;
1616 	while (thr != NULL) {
1617 		size += register_note(list, NT_PRSTATUS,
1618 		    __elfN(note_prstatus), thr);
1619 		size += register_note(list, NT_FPREGSET,
1620 		    __elfN(note_fpregset), thr);
1621 		size += register_note(list, NT_THRMISC,
1622 		    __elfN(note_thrmisc), thr);
1623 		size += register_note(list, NT_PTLWPINFO,
1624 		    __elfN(note_ptlwpinfo), thr);
1625 		size += register_note(list, -1,
1626 		    __elfN(note_threadmd), thr);
1627 
1628 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1629 		    TAILQ_NEXT(thr, td_plist);
1630 		if (thr == td)
1631 			thr = TAILQ_NEXT(thr, td_plist);
1632 	}
1633 
1634 	size += register_note(list, NT_PROCSTAT_PROC,
1635 	    __elfN(note_procstat_proc), p);
1636 	size += register_note(list, NT_PROCSTAT_FILES,
1637 	    note_procstat_files, p);
1638 	size += register_note(list, NT_PROCSTAT_VMMAP,
1639 	    note_procstat_vmmap, p);
1640 	size += register_note(list, NT_PROCSTAT_GROUPS,
1641 	    note_procstat_groups, p);
1642 	size += register_note(list, NT_PROCSTAT_UMASK,
1643 	    note_procstat_umask, p);
1644 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1645 	    note_procstat_rlimit, p);
1646 	size += register_note(list, NT_PROCSTAT_OSREL,
1647 	    note_procstat_osrel, p);
1648 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1649 	    __elfN(note_procstat_psstrings), p);
1650 	size += register_note(list, NT_PROCSTAT_AUXV,
1651 	    __elfN(note_procstat_auxv), p);
1652 
1653 	*sizep = size;
1654 }
1655 
1656 static void
1657 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1658     size_t notesz)
1659 {
1660 	Elf_Ehdr *ehdr;
1661 	Elf_Phdr *phdr;
1662 	Elf_Shdr *shdr;
1663 	struct phdr_closure phc;
1664 
1665 	ehdr = (Elf_Ehdr *)hdr;
1666 
1667 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1668 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1669 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1670 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1671 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1672 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1673 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1674 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1675 	ehdr->e_ident[EI_ABIVERSION] = 0;
1676 	ehdr->e_ident[EI_PAD] = 0;
1677 	ehdr->e_type = ET_CORE;
1678 	ehdr->e_machine = td->td_proc->p_elf_machine;
1679 	ehdr->e_version = EV_CURRENT;
1680 	ehdr->e_entry = 0;
1681 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1682 	ehdr->e_flags = td->td_proc->p_elf_flags;
1683 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1684 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1685 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1686 	ehdr->e_shstrndx = SHN_UNDEF;
1687 	if (numsegs + 1 < PN_XNUM) {
1688 		ehdr->e_phnum = numsegs + 1;
1689 		ehdr->e_shnum = 0;
1690 	} else {
1691 		ehdr->e_phnum = PN_XNUM;
1692 		ehdr->e_shnum = 1;
1693 
1694 		ehdr->e_shoff = ehdr->e_phoff +
1695 		    (numsegs + 1) * ehdr->e_phentsize;
1696 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1697 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1698 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1699 
1700 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1701 		memset(shdr, 0, sizeof(*shdr));
1702 		/*
1703 		 * A special first section is used to hold large segment and
1704 		 * section counts.  This was proposed by Sun Microsystems in
1705 		 * Solaris and has been adopted by Linux; the standard ELF
1706 		 * tools are already familiar with the technique.
1707 		 *
1708 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1709 		 * (or 12-7 depending on the version of the document) for more
1710 		 * details.
1711 		 */
1712 		shdr->sh_type = SHT_NULL;
1713 		shdr->sh_size = ehdr->e_shnum;
1714 		shdr->sh_link = ehdr->e_shstrndx;
1715 		shdr->sh_info = numsegs + 1;
1716 	}
1717 
1718 	/*
1719 	 * Fill in the program header entries.
1720 	 */
1721 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1722 
1723 	/* The note segement. */
1724 	phdr->p_type = PT_NOTE;
1725 	phdr->p_offset = hdrsize;
1726 	phdr->p_vaddr = 0;
1727 	phdr->p_paddr = 0;
1728 	phdr->p_filesz = notesz;
1729 	phdr->p_memsz = 0;
1730 	phdr->p_flags = PF_R;
1731 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1732 	phdr++;
1733 
1734 	/* All the writable segments from the program. */
1735 	phc.phdr = phdr;
1736 	phc.offset = round_page(hdrsize + notesz);
1737 	each_dumpable_segment(td, cb_put_phdr, &phc);
1738 }
1739 
1740 static size_t
1741 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1742 {
1743 	struct note_info *ninfo;
1744 	size_t size, notesize;
1745 
1746 	size = 0;
1747 	out(arg, NULL, &size);
1748 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1749 	ninfo->type = type;
1750 	ninfo->outfunc = out;
1751 	ninfo->outarg = arg;
1752 	ninfo->outsize = size;
1753 	TAILQ_INSERT_TAIL(list, ninfo, link);
1754 
1755 	if (type == -1)
1756 		return (size);
1757 
1758 	notesize = sizeof(Elf_Note) +		/* note header */
1759 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1760 						/* note name */
1761 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1762 
1763 	return (notesize);
1764 }
1765 
1766 static size_t
1767 append_note_data(const void *src, void *dst, size_t len)
1768 {
1769 	size_t padded_len;
1770 
1771 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1772 	if (dst != NULL) {
1773 		bcopy(src, dst, len);
1774 		bzero((char *)dst + len, padded_len - len);
1775 	}
1776 	return (padded_len);
1777 }
1778 
1779 size_t
1780 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1781 {
1782 	Elf_Note *note;
1783 	char *buf;
1784 	size_t notesize;
1785 
1786 	buf = dst;
1787 	if (buf != NULL) {
1788 		note = (Elf_Note *)buf;
1789 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1790 		note->n_descsz = size;
1791 		note->n_type = type;
1792 		buf += sizeof(*note);
1793 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1794 		    sizeof(FREEBSD_ABI_VENDOR));
1795 		append_note_data(src, buf, size);
1796 		if (descp != NULL)
1797 			*descp = buf;
1798 	}
1799 
1800 	notesize = sizeof(Elf_Note) +		/* note header */
1801 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1802 						/* note name */
1803 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1804 
1805 	return (notesize);
1806 }
1807 
1808 static void
1809 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1810 {
1811 	Elf_Note note;
1812 	ssize_t old_len, sect_len;
1813 	size_t new_len, descsz, i;
1814 
1815 	if (ninfo->type == -1) {
1816 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1817 		return;
1818 	}
1819 
1820 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1821 	note.n_descsz = ninfo->outsize;
1822 	note.n_type = ninfo->type;
1823 
1824 	sbuf_bcat(sb, &note, sizeof(note));
1825 	sbuf_start_section(sb, &old_len);
1826 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1827 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1828 	if (note.n_descsz == 0)
1829 		return;
1830 	sbuf_start_section(sb, &old_len);
1831 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1832 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1833 	if (sect_len < 0)
1834 		return;
1835 
1836 	new_len = (size_t)sect_len;
1837 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1838 	if (new_len < descsz) {
1839 		/*
1840 		 * It is expected that individual note emitters will correctly
1841 		 * predict their expected output size and fill up to that size
1842 		 * themselves, padding in a format-specific way if needed.
1843 		 * However, in case they don't, just do it here with zeros.
1844 		 */
1845 		for (i = 0; i < descsz - new_len; i++)
1846 			sbuf_putc(sb, 0);
1847 	} else if (new_len > descsz) {
1848 		/*
1849 		 * We can't always truncate sb -- we may have drained some
1850 		 * of it already.
1851 		 */
1852 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1853 		    "read it (%zu > %zu).  Since it is longer than "
1854 		    "expected, this coredump's notes are corrupt.  THIS "
1855 		    "IS A BUG in the note_procstat routine for type %u.\n",
1856 		    __func__, (unsigned)note.n_type, new_len, descsz,
1857 		    (unsigned)note.n_type));
1858 	}
1859 }
1860 
1861 /*
1862  * Miscellaneous note out functions.
1863  */
1864 
1865 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1866 #include <compat/freebsd32/freebsd32.h>
1867 #include <compat/freebsd32/freebsd32_signal.h>
1868 
1869 typedef struct prstatus32 elf_prstatus_t;
1870 typedef struct prpsinfo32 elf_prpsinfo_t;
1871 typedef struct fpreg32 elf_prfpregset_t;
1872 typedef struct fpreg32 elf_fpregset_t;
1873 typedef struct reg32 elf_gregset_t;
1874 typedef struct thrmisc32 elf_thrmisc_t;
1875 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1876 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1877 typedef uint32_t elf_ps_strings_t;
1878 #else
1879 typedef prstatus_t elf_prstatus_t;
1880 typedef prpsinfo_t elf_prpsinfo_t;
1881 typedef prfpregset_t elf_prfpregset_t;
1882 typedef prfpregset_t elf_fpregset_t;
1883 typedef gregset_t elf_gregset_t;
1884 typedef thrmisc_t elf_thrmisc_t;
1885 #define ELF_KERN_PROC_MASK	0
1886 typedef struct kinfo_proc elf_kinfo_proc_t;
1887 typedef vm_offset_t elf_ps_strings_t;
1888 #endif
1889 
1890 static void
1891 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1892 {
1893 	struct sbuf sbarg;
1894 	size_t len;
1895 	char *cp, *end;
1896 	struct proc *p;
1897 	elf_prpsinfo_t *psinfo;
1898 	int error;
1899 
1900 	p = (struct proc *)arg;
1901 	if (sb != NULL) {
1902 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1903 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1904 		psinfo->pr_version = PRPSINFO_VERSION;
1905 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1906 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1907 		PROC_LOCK(p);
1908 		if (p->p_args != NULL) {
1909 			len = sizeof(psinfo->pr_psargs) - 1;
1910 			if (len > p->p_args->ar_length)
1911 				len = p->p_args->ar_length;
1912 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1913 			PROC_UNLOCK(p);
1914 			error = 0;
1915 		} else {
1916 			_PHOLD(p);
1917 			PROC_UNLOCK(p);
1918 			sbuf_new(&sbarg, psinfo->pr_psargs,
1919 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1920 			error = proc_getargv(curthread, p, &sbarg);
1921 			PRELE(p);
1922 			if (sbuf_finish(&sbarg) == 0)
1923 				len = sbuf_len(&sbarg) - 1;
1924 			else
1925 				len = sizeof(psinfo->pr_psargs) - 1;
1926 			sbuf_delete(&sbarg);
1927 		}
1928 		if (error || len == 0)
1929 			strlcpy(psinfo->pr_psargs, p->p_comm,
1930 			    sizeof(psinfo->pr_psargs));
1931 		else {
1932 			KASSERT(len < sizeof(psinfo->pr_psargs),
1933 			    ("len is too long: %zu vs %zu", len,
1934 			    sizeof(psinfo->pr_psargs)));
1935 			cp = psinfo->pr_psargs;
1936 			end = cp + len - 1;
1937 			for (;;) {
1938 				cp = memchr(cp, '\0', end - cp);
1939 				if (cp == NULL)
1940 					break;
1941 				*cp = ' ';
1942 			}
1943 		}
1944 		psinfo->pr_pid = p->p_pid;
1945 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1946 		free(psinfo, M_TEMP);
1947 	}
1948 	*sizep = sizeof(*psinfo);
1949 }
1950 
1951 static void
1952 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1953 {
1954 	struct thread *td;
1955 	elf_prstatus_t *status;
1956 
1957 	td = (struct thread *)arg;
1958 	if (sb != NULL) {
1959 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1960 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1961 		status->pr_version = PRSTATUS_VERSION;
1962 		status->pr_statussz = sizeof(elf_prstatus_t);
1963 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1964 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1965 		status->pr_osreldate = osreldate;
1966 		status->pr_cursig = td->td_proc->p_sig;
1967 		status->pr_pid = td->td_tid;
1968 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1969 		fill_regs32(td, &status->pr_reg);
1970 #else
1971 		fill_regs(td, &status->pr_reg);
1972 #endif
1973 		sbuf_bcat(sb, status, sizeof(*status));
1974 		free(status, M_TEMP);
1975 	}
1976 	*sizep = sizeof(*status);
1977 }
1978 
1979 static void
1980 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1981 {
1982 	struct thread *td;
1983 	elf_prfpregset_t *fpregset;
1984 
1985 	td = (struct thread *)arg;
1986 	if (sb != NULL) {
1987 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1988 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1989 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1990 		fill_fpregs32(td, fpregset);
1991 #else
1992 		fill_fpregs(td, fpregset);
1993 #endif
1994 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1995 		free(fpregset, M_TEMP);
1996 	}
1997 	*sizep = sizeof(*fpregset);
1998 }
1999 
2000 static void
2001 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2002 {
2003 	struct thread *td;
2004 	elf_thrmisc_t thrmisc;
2005 
2006 	td = (struct thread *)arg;
2007 	if (sb != NULL) {
2008 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2009 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2010 		strcpy(thrmisc.pr_tname, td->td_name);
2011 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2012 	}
2013 	*sizep = sizeof(thrmisc);
2014 }
2015 
2016 static void
2017 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2018 {
2019 	struct thread *td;
2020 	size_t size;
2021 	int structsize;
2022 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2023 	struct ptrace_lwpinfo32 pl;
2024 #else
2025 	struct ptrace_lwpinfo pl;
2026 #endif
2027 
2028 	td = (struct thread *)arg;
2029 	size = sizeof(structsize) + sizeof(pl);
2030 	if (sb != NULL) {
2031 		KASSERT(*sizep == size, ("invalid size"));
2032 		structsize = sizeof(pl);
2033 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2034 		bzero(&pl, sizeof(pl));
2035 		pl.pl_lwpid = td->td_tid;
2036 		pl.pl_event = PL_EVENT_NONE;
2037 		pl.pl_sigmask = td->td_sigmask;
2038 		pl.pl_siglist = td->td_siglist;
2039 		if (td->td_si.si_signo != 0) {
2040 			pl.pl_event = PL_EVENT_SIGNAL;
2041 			pl.pl_flags |= PL_FLAG_SI;
2042 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2043 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2044 #else
2045 			pl.pl_siginfo = td->td_si;
2046 #endif
2047 		}
2048 		strcpy(pl.pl_tdname, td->td_name);
2049 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2050 		sbuf_bcat(sb, &pl, sizeof(pl));
2051 	}
2052 	*sizep = size;
2053 }
2054 
2055 /*
2056  * Allow for MD specific notes, as well as any MD
2057  * specific preparations for writing MI notes.
2058  */
2059 static void
2060 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2061 {
2062 	struct thread *td;
2063 	void *buf;
2064 	size_t size;
2065 
2066 	td = (struct thread *)arg;
2067 	size = *sizep;
2068 	if (size != 0 && sb != NULL)
2069 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2070 	else
2071 		buf = NULL;
2072 	size = 0;
2073 	__elfN(dump_thread)(td, buf, &size);
2074 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2075 	if (size != 0 && sb != NULL)
2076 		sbuf_bcat(sb, buf, size);
2077 	free(buf, M_TEMP);
2078 	*sizep = size;
2079 }
2080 
2081 #ifdef KINFO_PROC_SIZE
2082 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2083 #endif
2084 
2085 static void
2086 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2087 {
2088 	struct proc *p;
2089 	size_t size;
2090 	int structsize;
2091 
2092 	p = (struct proc *)arg;
2093 	size = sizeof(structsize) + p->p_numthreads *
2094 	    sizeof(elf_kinfo_proc_t);
2095 
2096 	if (sb != NULL) {
2097 		KASSERT(*sizep == size, ("invalid size"));
2098 		structsize = sizeof(elf_kinfo_proc_t);
2099 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2100 		sx_slock(&proctree_lock);
2101 		PROC_LOCK(p);
2102 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2103 		sx_sunlock(&proctree_lock);
2104 	}
2105 	*sizep = size;
2106 }
2107 
2108 #ifdef KINFO_FILE_SIZE
2109 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2110 #endif
2111 
2112 static void
2113 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2114 {
2115 	struct proc *p;
2116 	size_t size, sect_sz, i;
2117 	ssize_t start_len, sect_len;
2118 	int structsize, filedesc_flags;
2119 
2120 	if (coredump_pack_fileinfo)
2121 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2122 	else
2123 		filedesc_flags = 0;
2124 
2125 	p = (struct proc *)arg;
2126 	structsize = sizeof(struct kinfo_file);
2127 	if (sb == NULL) {
2128 		size = 0;
2129 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2130 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2131 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2132 		PROC_LOCK(p);
2133 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2134 		sbuf_finish(sb);
2135 		sbuf_delete(sb);
2136 		*sizep = size;
2137 	} else {
2138 		sbuf_start_section(sb, &start_len);
2139 
2140 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2141 		PROC_LOCK(p);
2142 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2143 		    filedesc_flags);
2144 
2145 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2146 		if (sect_len < 0)
2147 			return;
2148 		sect_sz = sect_len;
2149 
2150 		KASSERT(sect_sz <= *sizep,
2151 		    ("kern_proc_filedesc_out did not respect maxlen; "
2152 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2153 		     sect_sz - sizeof(structsize)));
2154 
2155 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2156 			sbuf_putc(sb, 0);
2157 	}
2158 }
2159 
2160 #ifdef KINFO_VMENTRY_SIZE
2161 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2162 #endif
2163 
2164 static void
2165 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2166 {
2167 	struct proc *p;
2168 	size_t size;
2169 	int structsize, vmmap_flags;
2170 
2171 	if (coredump_pack_vmmapinfo)
2172 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2173 	else
2174 		vmmap_flags = 0;
2175 
2176 	p = (struct proc *)arg;
2177 	structsize = sizeof(struct kinfo_vmentry);
2178 	if (sb == NULL) {
2179 		size = 0;
2180 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2181 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2182 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2183 		PROC_LOCK(p);
2184 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2185 		sbuf_finish(sb);
2186 		sbuf_delete(sb);
2187 		*sizep = size;
2188 	} else {
2189 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2190 		PROC_LOCK(p);
2191 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2192 		    vmmap_flags);
2193 	}
2194 }
2195 
2196 static void
2197 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2198 {
2199 	struct proc *p;
2200 	size_t size;
2201 	int structsize;
2202 
2203 	p = (struct proc *)arg;
2204 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2205 	if (sb != NULL) {
2206 		KASSERT(*sizep == size, ("invalid size"));
2207 		structsize = sizeof(gid_t);
2208 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2209 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2210 		    sizeof(gid_t));
2211 	}
2212 	*sizep = size;
2213 }
2214 
2215 static void
2216 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2217 {
2218 	struct proc *p;
2219 	size_t size;
2220 	int structsize;
2221 
2222 	p = (struct proc *)arg;
2223 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2224 	if (sb != NULL) {
2225 		KASSERT(*sizep == size, ("invalid size"));
2226 		structsize = sizeof(p->p_fd->fd_cmask);
2227 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2228 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2229 	}
2230 	*sizep = size;
2231 }
2232 
2233 static void
2234 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2235 {
2236 	struct proc *p;
2237 	struct rlimit rlim[RLIM_NLIMITS];
2238 	size_t size;
2239 	int structsize, i;
2240 
2241 	p = (struct proc *)arg;
2242 	size = sizeof(structsize) + sizeof(rlim);
2243 	if (sb != NULL) {
2244 		KASSERT(*sizep == size, ("invalid size"));
2245 		structsize = sizeof(rlim);
2246 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2247 		PROC_LOCK(p);
2248 		for (i = 0; i < RLIM_NLIMITS; i++)
2249 			lim_rlimit_proc(p, i, &rlim[i]);
2250 		PROC_UNLOCK(p);
2251 		sbuf_bcat(sb, rlim, sizeof(rlim));
2252 	}
2253 	*sizep = size;
2254 }
2255 
2256 static void
2257 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2258 {
2259 	struct proc *p;
2260 	size_t size;
2261 	int structsize;
2262 
2263 	p = (struct proc *)arg;
2264 	size = sizeof(structsize) + sizeof(p->p_osrel);
2265 	if (sb != NULL) {
2266 		KASSERT(*sizep == size, ("invalid size"));
2267 		structsize = sizeof(p->p_osrel);
2268 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2269 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2270 	}
2271 	*sizep = size;
2272 }
2273 
2274 static void
2275 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2276 {
2277 	struct proc *p;
2278 	elf_ps_strings_t ps_strings;
2279 	size_t size;
2280 	int structsize;
2281 
2282 	p = (struct proc *)arg;
2283 	size = sizeof(structsize) + sizeof(ps_strings);
2284 	if (sb != NULL) {
2285 		KASSERT(*sizep == size, ("invalid size"));
2286 		structsize = sizeof(ps_strings);
2287 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2288 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2289 #else
2290 		ps_strings = p->p_sysent->sv_psstrings;
2291 #endif
2292 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2293 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2294 	}
2295 	*sizep = size;
2296 }
2297 
2298 static void
2299 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2300 {
2301 	struct proc *p;
2302 	size_t size;
2303 	int structsize;
2304 
2305 	p = (struct proc *)arg;
2306 	if (sb == NULL) {
2307 		size = 0;
2308 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2309 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2310 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2311 		PHOLD(p);
2312 		proc_getauxv(curthread, p, sb);
2313 		PRELE(p);
2314 		sbuf_finish(sb);
2315 		sbuf_delete(sb);
2316 		*sizep = size;
2317 	} else {
2318 		structsize = sizeof(Elf_Auxinfo);
2319 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2320 		PHOLD(p);
2321 		proc_getauxv(curthread, p, sb);
2322 		PRELE(p);
2323 	}
2324 }
2325 
2326 static boolean_t
2327 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2328     int32_t *osrel, const Elf_Phdr *pnote)
2329 {
2330 	const Elf_Note *note, *note0, *note_end;
2331 	const char *note_name;
2332 	char *buf;
2333 	int i, error;
2334 	boolean_t res;
2335 
2336 	/* We need some limit, might as well use PAGE_SIZE. */
2337 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2338 		return (FALSE);
2339 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2340 	if (pnote->p_offset > PAGE_SIZE ||
2341 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2342 		VOP_UNLOCK(imgp->vp, 0);
2343 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2344 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2345 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2346 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2347 		    curthread->td_ucred, NOCRED, NULL, curthread);
2348 		if (error != 0) {
2349 			uprintf("i/o error PT_NOTE\n");
2350 			res = FALSE;
2351 			goto ret;
2352 		}
2353 		note = note0 = (const Elf_Note *)buf;
2354 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2355 	} else {
2356 		note = note0 = (const Elf_Note *)(imgp->image_header +
2357 		    pnote->p_offset);
2358 		note_end = (const Elf_Note *)(imgp->image_header +
2359 		    pnote->p_offset + pnote->p_filesz);
2360 		buf = NULL;
2361 	}
2362 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2363 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2364 		    (const char *)note < sizeof(Elf_Note)) {
2365 			res = FALSE;
2366 			goto ret;
2367 		}
2368 		if (note->n_namesz != checknote->hdr.n_namesz ||
2369 		    note->n_descsz != checknote->hdr.n_descsz ||
2370 		    note->n_type != checknote->hdr.n_type)
2371 			goto nextnote;
2372 		note_name = (const char *)(note + 1);
2373 		if (note_name + checknote->hdr.n_namesz >=
2374 		    (const char *)note_end || strncmp(checknote->vendor,
2375 		    note_name, checknote->hdr.n_namesz) != 0)
2376 			goto nextnote;
2377 
2378 		/*
2379 		 * Fetch the osreldate for binary
2380 		 * from the ELF OSABI-note if necessary.
2381 		 */
2382 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2383 		    checknote->trans_osrel != NULL) {
2384 			res = checknote->trans_osrel(note, osrel);
2385 			goto ret;
2386 		}
2387 		res = TRUE;
2388 		goto ret;
2389 nextnote:
2390 		note = (const Elf_Note *)((const char *)(note + 1) +
2391 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2392 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2393 	}
2394 	res = FALSE;
2395 ret:
2396 	free(buf, M_TEMP);
2397 	return (res);
2398 }
2399 
2400 /*
2401  * Try to find the appropriate ABI-note section for checknote,
2402  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2403  * first page of the image is searched, the same as for headers.
2404  */
2405 static boolean_t
2406 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2407     int32_t *osrel)
2408 {
2409 	const Elf_Phdr *phdr;
2410 	const Elf_Ehdr *hdr;
2411 	int i;
2412 
2413 	hdr = (const Elf_Ehdr *)imgp->image_header;
2414 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2415 
2416 	for (i = 0; i < hdr->e_phnum; i++) {
2417 		if (phdr[i].p_type == PT_NOTE &&
2418 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2419 			return (TRUE);
2420 	}
2421 	return (FALSE);
2422 
2423 }
2424 
2425 /*
2426  * Tell kern_execve.c about it, with a little help from the linker.
2427  */
2428 static struct execsw __elfN(execsw) = {
2429 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2430 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2431 };
2432 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2433 
2434 static vm_prot_t
2435 __elfN(trans_prot)(Elf_Word flags)
2436 {
2437 	vm_prot_t prot;
2438 
2439 	prot = 0;
2440 	if (flags & PF_X)
2441 		prot |= VM_PROT_EXECUTE;
2442 	if (flags & PF_W)
2443 		prot |= VM_PROT_WRITE;
2444 	if (flags & PF_R)
2445 		prot |= VM_PROT_READ;
2446 #if __ELF_WORD_SIZE == 32
2447 #if defined(__amd64__)
2448 	if (i386_read_exec && (flags & PF_R))
2449 		prot |= VM_PROT_EXECUTE;
2450 #endif
2451 #endif
2452 	return (prot);
2453 }
2454 
2455 static Elf_Word
2456 __elfN(untrans_prot)(vm_prot_t prot)
2457 {
2458 	Elf_Word flags;
2459 
2460 	flags = 0;
2461 	if (prot & VM_PROT_EXECUTE)
2462 		flags |= PF_X;
2463 	if (prot & VM_PROT_READ)
2464 		flags |= PF_R;
2465 	if (prot & VM_PROT_WRITE)
2466 		flags |= PF_W;
2467 	return (flags);
2468 }
2469