xref: /freebsd/sys/kern/imgact_elf.c (revision f39bffc62c1395bde25d152c7f68fdf7cbaab414)
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 
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 > MAXPATHLEN) {
843 				uprintf("Invalid PT_INTERP\n");
844 				error = ENOEXEC;
845 				goto ret;
846 			}
847 			if (interp != NULL) {
848 				uprintf("Multiple PT_INTERP headers\n");
849 				error = ENOEXEC;
850 				goto ret;
851 			}
852 			interp_name_len = phdr[i].p_filesz;
853 			if (phdr[i].p_offset > PAGE_SIZE ||
854 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
855 				VOP_UNLOCK(imgp->vp, 0);
856 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
857 				    M_WAITOK);
858 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
859 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
860 				    interp_name_len, phdr[i].p_offset,
861 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
862 				    NOCRED, NULL, td);
863 				if (error != 0) {
864 					uprintf("i/o error PT_INTERP\n");
865 					goto ret;
866 				}
867 				interp_buf[interp_name_len] = '\0';
868 				interp = interp_buf;
869 			} else {
870 				interp = __DECONST(char *, imgp->image_header) +
871 				    phdr[i].p_offset;
872 			}
873 			break;
874 		case PT_GNU_STACK:
875 			if (__elfN(nxstack))
876 				imgp->stack_prot =
877 				    __elfN(trans_prot)(phdr[i].p_flags);
878 			imgp->stack_sz = phdr[i].p_memsz;
879 			break;
880 		}
881 	}
882 
883 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
884 	    &osrel);
885 	if (brand_info == NULL) {
886 		uprintf("ELF binary type \"%u\" not known.\n",
887 		    hdr->e_ident[EI_OSABI]);
888 		error = ENOEXEC;
889 		goto ret;
890 	}
891 	et_dyn_addr = 0;
892 	if (hdr->e_type == ET_DYN) {
893 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
894 			uprintf("Cannot execute shared object\n");
895 			error = ENOEXEC;
896 			goto ret;
897 		}
898 		/*
899 		 * Honour the base load address from the dso if it is
900 		 * non-zero for some reason.
901 		 */
902 		if (baddr == 0)
903 			et_dyn_addr = ET_DYN_LOAD_ADDR;
904 	}
905 	sv = brand_info->sysvec;
906 	if (interp != NULL && brand_info->interp_newpath != NULL)
907 		newinterp = brand_info->interp_newpath;
908 
909 	/*
910 	 * Avoid a possible deadlock if the current address space is destroyed
911 	 * and that address space maps the locked vnode.  In the common case,
912 	 * the locked vnode's v_usecount is decremented but remains greater
913 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
914 	 * However, in cases where the vnode lock is external, such as nullfs,
915 	 * v_usecount may become zero.
916 	 *
917 	 * The VV_TEXT flag prevents modifications to the executable while
918 	 * the vnode is unlocked.
919 	 */
920 	VOP_UNLOCK(imgp->vp, 0);
921 
922 	error = exec_new_vmspace(imgp, sv);
923 	imgp->proc->p_sysent = sv;
924 
925 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
926 	if (error != 0)
927 		goto ret;
928 
929 	for (i = 0; i < hdr->e_phnum; i++) {
930 		switch (phdr[i].p_type) {
931 		case PT_LOAD:	/* Loadable segment */
932 			if (phdr[i].p_memsz == 0)
933 				break;
934 			prot = __elfN(trans_prot)(phdr[i].p_flags);
935 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
936 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
937 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
938 			    sv->sv_pagesize);
939 			if (error != 0)
940 				goto ret;
941 
942 			/*
943 			 * If this segment contains the program headers,
944 			 * remember their virtual address for the AT_PHDR
945 			 * aux entry. Static binaries don't usually include
946 			 * a PT_PHDR entry.
947 			 */
948 			if (phdr[i].p_offset == 0 &&
949 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
950 				<= phdr[i].p_filesz)
951 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
952 				    et_dyn_addr;
953 
954 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
955 			seg_size = round_page(phdr[i].p_memsz +
956 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
957 
958 			/*
959 			 * Make the largest executable segment the official
960 			 * text segment and all others data.
961 			 *
962 			 * Note that obreak() assumes that data_addr +
963 			 * data_size == end of data load area, and the ELF
964 			 * file format expects segments to be sorted by
965 			 * address.  If multiple data segments exist, the
966 			 * last one will be used.
967 			 */
968 
969 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
970 				text_size = seg_size;
971 				text_addr = seg_addr;
972 			} else {
973 				data_size = seg_size;
974 				data_addr = seg_addr;
975 			}
976 			total_size += seg_size;
977 			break;
978 		case PT_PHDR: 	/* Program header table info */
979 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
980 			break;
981 		default:
982 			break;
983 		}
984 	}
985 
986 	if (data_addr == 0 && data_size == 0) {
987 		data_addr = text_addr;
988 		data_size = text_size;
989 	}
990 
991 	entry = (u_long)hdr->e_entry + et_dyn_addr;
992 
993 	/*
994 	 * Check limits.  It should be safe to check the
995 	 * limits after loading the segments since we do
996 	 * not actually fault in all the segments pages.
997 	 */
998 	PROC_LOCK(imgp->proc);
999 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1000 		err_str = "Data segment size exceeds process limit";
1001 	else if (text_size > maxtsiz)
1002 		err_str = "Text segment size exceeds system limit";
1003 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1004 		err_str = "Total segment size exceeds process limit";
1005 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1006 		err_str = "Data segment size exceeds resource limit";
1007 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1008 		err_str = "Total segment size exceeds resource limit";
1009 	if (err_str != NULL) {
1010 		PROC_UNLOCK(imgp->proc);
1011 		uprintf("%s\n", err_str);
1012 		error = ENOMEM;
1013 		goto ret;
1014 	}
1015 
1016 	vmspace = imgp->proc->p_vmspace;
1017 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1018 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1019 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1020 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1021 
1022 	/*
1023 	 * We load the dynamic linker where a userland call
1024 	 * to mmap(0, ...) would put it.  The rationale behind this
1025 	 * calculation is that it leaves room for the heap to grow to
1026 	 * its maximum allowed size.
1027 	 */
1028 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1029 	    RLIMIT_DATA));
1030 	PROC_UNLOCK(imgp->proc);
1031 
1032 	imgp->entry_addr = entry;
1033 
1034 	if (interp != NULL) {
1035 		have_interp = FALSE;
1036 		VOP_UNLOCK(imgp->vp, 0);
1037 		if (brand_info->emul_path != NULL &&
1038 		    brand_info->emul_path[0] != '\0') {
1039 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1040 			snprintf(path, MAXPATHLEN, "%s%s",
1041 			    brand_info->emul_path, interp);
1042 			error = __elfN(load_file)(imgp->proc, path, &addr,
1043 			    &imgp->entry_addr, sv->sv_pagesize);
1044 			free(path, M_TEMP);
1045 			if (error == 0)
1046 				have_interp = TRUE;
1047 		}
1048 		if (!have_interp && newinterp != NULL &&
1049 		    (brand_info->interp_path == NULL ||
1050 		    strcmp(interp, brand_info->interp_path) == 0)) {
1051 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1052 			    &imgp->entry_addr, sv->sv_pagesize);
1053 			if (error == 0)
1054 				have_interp = TRUE;
1055 		}
1056 		if (!have_interp) {
1057 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1058 			    &imgp->entry_addr, sv->sv_pagesize);
1059 		}
1060 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1061 		if (error != 0) {
1062 			uprintf("ELF interpreter %s not found, error %d\n",
1063 			    interp, error);
1064 			goto ret;
1065 		}
1066 	} else
1067 		addr = et_dyn_addr;
1068 
1069 	/*
1070 	 * Construct auxargs table (used by the fixup routine)
1071 	 */
1072 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1073 	elf_auxargs->execfd = -1;
1074 	elf_auxargs->phdr = proghdr;
1075 	elf_auxargs->phent = hdr->e_phentsize;
1076 	elf_auxargs->phnum = hdr->e_phnum;
1077 	elf_auxargs->pagesz = PAGE_SIZE;
1078 	elf_auxargs->base = addr;
1079 	elf_auxargs->flags = 0;
1080 	elf_auxargs->entry = entry;
1081 	elf_auxargs->hdr_eflags = hdr->e_flags;
1082 
1083 	imgp->auxargs = elf_auxargs;
1084 	imgp->interpreted = 0;
1085 	imgp->reloc_base = addr;
1086 	imgp->proc->p_osrel = osrel;
1087 	imgp->proc->p_elf_machine = hdr->e_machine;
1088 	imgp->proc->p_elf_flags = hdr->e_flags;
1089 
1090 ret:
1091 	free(interp_buf, M_TEMP);
1092 	return (error);
1093 }
1094 
1095 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1096 
1097 int
1098 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1099 {
1100 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1101 	Elf_Auxinfo *argarray, *pos;
1102 	Elf_Addr *base, *auxbase;
1103 	int error;
1104 
1105 	base = (Elf_Addr *)*stack_base;
1106 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1107 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1108 	    M_WAITOK | M_ZERO);
1109 
1110 	if (args->execfd != -1)
1111 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1112 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1113 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1114 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1115 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1116 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1117 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1118 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1119 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1120 	if (imgp->execpathp != 0)
1121 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1122 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1123 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1124 	if (imgp->canary != 0) {
1125 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1126 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1127 	}
1128 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1129 	if (imgp->pagesizes != 0) {
1130 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1131 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1132 	}
1133 	if (imgp->sysent->sv_timekeep_base != 0) {
1134 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1135 		    imgp->sysent->sv_timekeep_base);
1136 	}
1137 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1138 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1139 	    imgp->sysent->sv_stackprot);
1140 	if (imgp->sysent->sv_hwcap != NULL)
1141 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1142 	if (imgp->sysent->sv_hwcap2 != NULL)
1143 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1144 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1145 
1146 	free(imgp->auxargs, M_TEMP);
1147 	imgp->auxargs = NULL;
1148 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1149 
1150 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1151 	free(argarray, M_TEMP);
1152 	if (error != 0)
1153 		return (error);
1154 
1155 	base--;
1156 	if (suword(base, imgp->args->argc) == -1)
1157 		return (EFAULT);
1158 	*stack_base = (register_t *)base;
1159 	return (0);
1160 }
1161 
1162 /*
1163  * Code for generating ELF core dumps.
1164  */
1165 
1166 typedef void (*segment_callback)(vm_map_entry_t, void *);
1167 
1168 /* Closure for cb_put_phdr(). */
1169 struct phdr_closure {
1170 	Elf_Phdr *phdr;		/* Program header to fill in */
1171 	Elf_Off offset;		/* Offset of segment in core file */
1172 };
1173 
1174 /* Closure for cb_size_segment(). */
1175 struct sseg_closure {
1176 	int count;		/* Count of writable segments. */
1177 	size_t size;		/* Total size of all writable segments. */
1178 };
1179 
1180 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1181 
1182 struct note_info {
1183 	int		type;		/* Note type. */
1184 	outfunc_t 	outfunc; 	/* Output function. */
1185 	void		*outarg;	/* Argument for the output function. */
1186 	size_t		outsize;	/* Output size. */
1187 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1188 };
1189 
1190 TAILQ_HEAD(note_info_list, note_info);
1191 
1192 /* Coredump output parameters. */
1193 struct coredump_params {
1194 	off_t		offset;
1195 	struct ucred	*active_cred;
1196 	struct ucred	*file_cred;
1197 	struct thread	*td;
1198 	struct vnode	*vp;
1199 	struct compressor *comp;
1200 };
1201 
1202 extern int compress_user_cores;
1203 extern int compress_user_cores_level;
1204 
1205 static void cb_put_phdr(vm_map_entry_t, void *);
1206 static void cb_size_segment(vm_map_entry_t, void *);
1207 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1208     enum uio_seg);
1209 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1210 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1211     struct note_info_list *, size_t);
1212 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1213     size_t *);
1214 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1215 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1216 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1217 static int sbuf_drain_core_output(void *, const char *, int);
1218 static int sbuf_drain_count(void *arg, const char *data, int len);
1219 
1220 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1221 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1222 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1223 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1224 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1225 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1226 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1227 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1228 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1229 static void note_procstat_files(void *, struct sbuf *, size_t *);
1230 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1231 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1232 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1233 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1234 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1235 
1236 /*
1237  * Write out a core segment to the compression stream.
1238  */
1239 static int
1240 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1241 {
1242 	u_int chunk_len;
1243 	int error;
1244 
1245 	while (len > 0) {
1246 		chunk_len = MIN(len, CORE_BUF_SIZE);
1247 
1248 		/*
1249 		 * We can get EFAULT error here.
1250 		 * In that case zero out the current chunk of the segment.
1251 		 */
1252 		error = copyin(base, buf, chunk_len);
1253 		if (error != 0)
1254 			bzero(buf, chunk_len);
1255 		error = compressor_write(p->comp, buf, chunk_len);
1256 		if (error != 0)
1257 			break;
1258 		base += chunk_len;
1259 		len -= chunk_len;
1260 	}
1261 	return (error);
1262 }
1263 
1264 static int
1265 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1266 {
1267 
1268 	return (core_write((struct coredump_params *)arg, base, len, offset,
1269 	    UIO_SYSSPACE));
1270 }
1271 
1272 static int
1273 core_write(struct coredump_params *p, const void *base, size_t len,
1274     off_t offset, enum uio_seg seg)
1275 {
1276 
1277 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1278 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1279 	    p->active_cred, p->file_cred, NULL, p->td));
1280 }
1281 
1282 static int
1283 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1284     void *tmpbuf)
1285 {
1286 	int error;
1287 
1288 	if (p->comp != NULL)
1289 		return (compress_chunk(p, base, tmpbuf, len));
1290 
1291 	/*
1292 	 * EFAULT is a non-fatal error that we can get, for example,
1293 	 * if the segment is backed by a file but extends beyond its
1294 	 * end.
1295 	 */
1296 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1297 	if (error == EFAULT) {
1298 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1299 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1300 		    "for process %s\n", base, len, offset, curproc->p_comm);
1301 
1302 		/*
1303 		 * Write a "real" zero byte at the end of the target region
1304 		 * in the case this is the last segment.
1305 		 * The intermediate space will be implicitly zero-filled.
1306 		 */
1307 		error = core_write(p, zero_region, 1, offset + len - 1,
1308 		    UIO_SYSSPACE);
1309 	}
1310 	return (error);
1311 }
1312 
1313 /*
1314  * Drain into a core file.
1315  */
1316 static int
1317 sbuf_drain_core_output(void *arg, const char *data, int len)
1318 {
1319 	struct coredump_params *p;
1320 	int error, locked;
1321 
1322 	p = (struct coredump_params *)arg;
1323 
1324 	/*
1325 	 * Some kern_proc out routines that print to this sbuf may
1326 	 * call us with the process lock held. Draining with the
1327 	 * non-sleepable lock held is unsafe. The lock is needed for
1328 	 * those routines when dumping a live process. In our case we
1329 	 * can safely release the lock before draining and acquire
1330 	 * again after.
1331 	 */
1332 	locked = PROC_LOCKED(p->td->td_proc);
1333 	if (locked)
1334 		PROC_UNLOCK(p->td->td_proc);
1335 	if (p->comp != NULL)
1336 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1337 	else
1338 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1339 		    UIO_SYSSPACE);
1340 	if (locked)
1341 		PROC_LOCK(p->td->td_proc);
1342 	if (error != 0)
1343 		return (-error);
1344 	p->offset += len;
1345 	return (len);
1346 }
1347 
1348 /*
1349  * Drain into a counter.
1350  */
1351 static int
1352 sbuf_drain_count(void *arg, const char *data __unused, int len)
1353 {
1354 	size_t *sizep;
1355 
1356 	sizep = (size_t *)arg;
1357 	*sizep += len;
1358 	return (len);
1359 }
1360 
1361 int
1362 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1363 {
1364 	struct ucred *cred = td->td_ucred;
1365 	int error = 0;
1366 	struct sseg_closure seginfo;
1367 	struct note_info_list notelst;
1368 	struct coredump_params params;
1369 	struct note_info *ninfo;
1370 	void *hdr, *tmpbuf;
1371 	size_t hdrsize, notesz, coresize;
1372 
1373 	hdr = NULL;
1374 	tmpbuf = NULL;
1375 	TAILQ_INIT(&notelst);
1376 
1377 	/* Size the program segments. */
1378 	seginfo.count = 0;
1379 	seginfo.size = 0;
1380 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1381 
1382 	/*
1383 	 * Collect info about the core file header area.
1384 	 */
1385 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1386 	if (seginfo.count + 1 >= PN_XNUM)
1387 		hdrsize += sizeof(Elf_Shdr);
1388 	__elfN(prepare_notes)(td, &notelst, &notesz);
1389 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1390 
1391 	/* Set up core dump parameters. */
1392 	params.offset = 0;
1393 	params.active_cred = cred;
1394 	params.file_cred = NOCRED;
1395 	params.td = td;
1396 	params.vp = vp;
1397 	params.comp = NULL;
1398 
1399 #ifdef RACCT
1400 	if (racct_enable) {
1401 		PROC_LOCK(td->td_proc);
1402 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1403 		PROC_UNLOCK(td->td_proc);
1404 		if (error != 0) {
1405 			error = EFAULT;
1406 			goto done;
1407 		}
1408 	}
1409 #endif
1410 	if (coresize >= limit) {
1411 		error = EFAULT;
1412 		goto done;
1413 	}
1414 
1415 	/* Create a compression stream if necessary. */
1416 	if (compress_user_cores != 0) {
1417 		params.comp = compressor_init(core_compressed_write,
1418 		    compress_user_cores, CORE_BUF_SIZE,
1419 		    compress_user_cores_level, &params);
1420 		if (params.comp == NULL) {
1421 			error = EFAULT;
1422 			goto done;
1423 		}
1424 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1425         }
1426 
1427 	/*
1428 	 * Allocate memory for building the header, fill it up,
1429 	 * and write it out following the notes.
1430 	 */
1431 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1432 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1433 	    notesz);
1434 
1435 	/* Write the contents of all of the writable segments. */
1436 	if (error == 0) {
1437 		Elf_Phdr *php;
1438 		off_t offset;
1439 		int i;
1440 
1441 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1442 		offset = round_page(hdrsize + notesz);
1443 		for (i = 0; i < seginfo.count; i++) {
1444 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1445 			    php->p_filesz, offset, &params, tmpbuf);
1446 			if (error != 0)
1447 				break;
1448 			offset += php->p_filesz;
1449 			php++;
1450 		}
1451 		if (error == 0 && params.comp != NULL)
1452 			error = compressor_flush(params.comp);
1453 	}
1454 	if (error) {
1455 		log(LOG_WARNING,
1456 		    "Failed to write core file for process %s (error %d)\n",
1457 		    curproc->p_comm, error);
1458 	}
1459 
1460 done:
1461 	free(tmpbuf, M_TEMP);
1462 	if (params.comp != NULL)
1463 		compressor_fini(params.comp);
1464 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1465 		TAILQ_REMOVE(&notelst, ninfo, link);
1466 		free(ninfo, M_TEMP);
1467 	}
1468 	if (hdr != NULL)
1469 		free(hdr, M_TEMP);
1470 
1471 	return (error);
1472 }
1473 
1474 /*
1475  * A callback for each_dumpable_segment() to write out the segment's
1476  * program header entry.
1477  */
1478 static void
1479 cb_put_phdr(vm_map_entry_t entry, void *closure)
1480 {
1481 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1482 	Elf_Phdr *phdr = phc->phdr;
1483 
1484 	phc->offset = round_page(phc->offset);
1485 
1486 	phdr->p_type = PT_LOAD;
1487 	phdr->p_offset = phc->offset;
1488 	phdr->p_vaddr = entry->start;
1489 	phdr->p_paddr = 0;
1490 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1491 	phdr->p_align = PAGE_SIZE;
1492 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1493 
1494 	phc->offset += phdr->p_filesz;
1495 	phc->phdr++;
1496 }
1497 
1498 /*
1499  * A callback for each_dumpable_segment() to gather information about
1500  * the number of segments and their total size.
1501  */
1502 static void
1503 cb_size_segment(vm_map_entry_t entry, void *closure)
1504 {
1505 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1506 
1507 	ssc->count++;
1508 	ssc->size += entry->end - entry->start;
1509 }
1510 
1511 /*
1512  * For each writable segment in the process's memory map, call the given
1513  * function with a pointer to the map entry and some arbitrary
1514  * caller-supplied data.
1515  */
1516 static void
1517 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1518 {
1519 	struct proc *p = td->td_proc;
1520 	vm_map_t map = &p->p_vmspace->vm_map;
1521 	vm_map_entry_t entry;
1522 	vm_object_t backing_object, object;
1523 	boolean_t ignore_entry;
1524 
1525 	vm_map_lock_read(map);
1526 	for (entry = map->header.next; entry != &map->header;
1527 	    entry = entry->next) {
1528 		/*
1529 		 * Don't dump inaccessible mappings, deal with legacy
1530 		 * coredump mode.
1531 		 *
1532 		 * Note that read-only segments related to the elf binary
1533 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1534 		 * need to arbitrarily ignore such segments.
1535 		 */
1536 		if (elf_legacy_coredump) {
1537 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1538 				continue;
1539 		} else {
1540 			if ((entry->protection & VM_PROT_ALL) == 0)
1541 				continue;
1542 		}
1543 
1544 		/*
1545 		 * Dont include memory segment in the coredump if
1546 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1547 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1548 		 * kernel map).
1549 		 */
1550 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1551 			continue;
1552 
1553 		if ((object = entry->object.vm_object) == NULL)
1554 			continue;
1555 
1556 		/* Ignore memory-mapped devices and such things. */
1557 		VM_OBJECT_RLOCK(object);
1558 		while ((backing_object = object->backing_object) != NULL) {
1559 			VM_OBJECT_RLOCK(backing_object);
1560 			VM_OBJECT_RUNLOCK(object);
1561 			object = backing_object;
1562 		}
1563 		ignore_entry = object->type != OBJT_DEFAULT &&
1564 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1565 		    object->type != OBJT_PHYS;
1566 		VM_OBJECT_RUNLOCK(object);
1567 		if (ignore_entry)
1568 			continue;
1569 
1570 		(*func)(entry, closure);
1571 	}
1572 	vm_map_unlock_read(map);
1573 }
1574 
1575 /*
1576  * Write the core file header to the file, including padding up to
1577  * the page boundary.
1578  */
1579 static int
1580 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1581     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1582 {
1583 	struct note_info *ninfo;
1584 	struct sbuf *sb;
1585 	int error;
1586 
1587 	/* Fill in the header. */
1588 	bzero(hdr, hdrsize);
1589 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1590 
1591 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1592 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1593 	sbuf_start_section(sb, NULL);
1594 	sbuf_bcat(sb, hdr, hdrsize);
1595 	TAILQ_FOREACH(ninfo, notelst, link)
1596 	    __elfN(putnote)(ninfo, sb);
1597 	/* Align up to a page boundary for the program segments. */
1598 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1599 	error = sbuf_finish(sb);
1600 	sbuf_delete(sb);
1601 
1602 	return (error);
1603 }
1604 
1605 static void
1606 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1607     size_t *sizep)
1608 {
1609 	struct proc *p;
1610 	struct thread *thr;
1611 	size_t size;
1612 
1613 	p = td->td_proc;
1614 	size = 0;
1615 
1616 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1617 
1618 	/*
1619 	 * To have the debugger select the right thread (LWP) as the initial
1620 	 * thread, we dump the state of the thread passed to us in td first.
1621 	 * This is the thread that causes the core dump and thus likely to
1622 	 * be the right thread one wants to have selected in the debugger.
1623 	 */
1624 	thr = td;
1625 	while (thr != NULL) {
1626 		size += register_note(list, NT_PRSTATUS,
1627 		    __elfN(note_prstatus), thr);
1628 		size += register_note(list, NT_FPREGSET,
1629 		    __elfN(note_fpregset), thr);
1630 		size += register_note(list, NT_THRMISC,
1631 		    __elfN(note_thrmisc), thr);
1632 		size += register_note(list, NT_PTLWPINFO,
1633 		    __elfN(note_ptlwpinfo), thr);
1634 		size += register_note(list, -1,
1635 		    __elfN(note_threadmd), thr);
1636 
1637 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1638 		    TAILQ_NEXT(thr, td_plist);
1639 		if (thr == td)
1640 			thr = TAILQ_NEXT(thr, td_plist);
1641 	}
1642 
1643 	size += register_note(list, NT_PROCSTAT_PROC,
1644 	    __elfN(note_procstat_proc), p);
1645 	size += register_note(list, NT_PROCSTAT_FILES,
1646 	    note_procstat_files, p);
1647 	size += register_note(list, NT_PROCSTAT_VMMAP,
1648 	    note_procstat_vmmap, p);
1649 	size += register_note(list, NT_PROCSTAT_GROUPS,
1650 	    note_procstat_groups, p);
1651 	size += register_note(list, NT_PROCSTAT_UMASK,
1652 	    note_procstat_umask, p);
1653 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1654 	    note_procstat_rlimit, p);
1655 	size += register_note(list, NT_PROCSTAT_OSREL,
1656 	    note_procstat_osrel, p);
1657 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1658 	    __elfN(note_procstat_psstrings), p);
1659 	size += register_note(list, NT_PROCSTAT_AUXV,
1660 	    __elfN(note_procstat_auxv), p);
1661 
1662 	*sizep = size;
1663 }
1664 
1665 static void
1666 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1667     size_t notesz)
1668 {
1669 	Elf_Ehdr *ehdr;
1670 	Elf_Phdr *phdr;
1671 	Elf_Shdr *shdr;
1672 	struct phdr_closure phc;
1673 
1674 	ehdr = (Elf_Ehdr *)hdr;
1675 
1676 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1677 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1678 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1679 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1680 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1681 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1682 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1683 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1684 	ehdr->e_ident[EI_ABIVERSION] = 0;
1685 	ehdr->e_ident[EI_PAD] = 0;
1686 	ehdr->e_type = ET_CORE;
1687 	ehdr->e_machine = td->td_proc->p_elf_machine;
1688 	ehdr->e_version = EV_CURRENT;
1689 	ehdr->e_entry = 0;
1690 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1691 	ehdr->e_flags = td->td_proc->p_elf_flags;
1692 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1693 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1694 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1695 	ehdr->e_shstrndx = SHN_UNDEF;
1696 	if (numsegs + 1 < PN_XNUM) {
1697 		ehdr->e_phnum = numsegs + 1;
1698 		ehdr->e_shnum = 0;
1699 	} else {
1700 		ehdr->e_phnum = PN_XNUM;
1701 		ehdr->e_shnum = 1;
1702 
1703 		ehdr->e_shoff = ehdr->e_phoff +
1704 		    (numsegs + 1) * ehdr->e_phentsize;
1705 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1706 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1707 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1708 
1709 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1710 		memset(shdr, 0, sizeof(*shdr));
1711 		/*
1712 		 * A special first section is used to hold large segment and
1713 		 * section counts.  This was proposed by Sun Microsystems in
1714 		 * Solaris and has been adopted by Linux; the standard ELF
1715 		 * tools are already familiar with the technique.
1716 		 *
1717 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1718 		 * (or 12-7 depending on the version of the document) for more
1719 		 * details.
1720 		 */
1721 		shdr->sh_type = SHT_NULL;
1722 		shdr->sh_size = ehdr->e_shnum;
1723 		shdr->sh_link = ehdr->e_shstrndx;
1724 		shdr->sh_info = numsegs + 1;
1725 	}
1726 
1727 	/*
1728 	 * Fill in the program header entries.
1729 	 */
1730 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1731 
1732 	/* The note segement. */
1733 	phdr->p_type = PT_NOTE;
1734 	phdr->p_offset = hdrsize;
1735 	phdr->p_vaddr = 0;
1736 	phdr->p_paddr = 0;
1737 	phdr->p_filesz = notesz;
1738 	phdr->p_memsz = 0;
1739 	phdr->p_flags = PF_R;
1740 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1741 	phdr++;
1742 
1743 	/* All the writable segments from the program. */
1744 	phc.phdr = phdr;
1745 	phc.offset = round_page(hdrsize + notesz);
1746 	each_dumpable_segment(td, cb_put_phdr, &phc);
1747 }
1748 
1749 static size_t
1750 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1751 {
1752 	struct note_info *ninfo;
1753 	size_t size, notesize;
1754 
1755 	size = 0;
1756 	out(arg, NULL, &size);
1757 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1758 	ninfo->type = type;
1759 	ninfo->outfunc = out;
1760 	ninfo->outarg = arg;
1761 	ninfo->outsize = size;
1762 	TAILQ_INSERT_TAIL(list, ninfo, link);
1763 
1764 	if (type == -1)
1765 		return (size);
1766 
1767 	notesize = sizeof(Elf_Note) +		/* note header */
1768 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1769 						/* note name */
1770 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1771 
1772 	return (notesize);
1773 }
1774 
1775 static size_t
1776 append_note_data(const void *src, void *dst, size_t len)
1777 {
1778 	size_t padded_len;
1779 
1780 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1781 	if (dst != NULL) {
1782 		bcopy(src, dst, len);
1783 		bzero((char *)dst + len, padded_len - len);
1784 	}
1785 	return (padded_len);
1786 }
1787 
1788 size_t
1789 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1790 {
1791 	Elf_Note *note;
1792 	char *buf;
1793 	size_t notesize;
1794 
1795 	buf = dst;
1796 	if (buf != NULL) {
1797 		note = (Elf_Note *)buf;
1798 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1799 		note->n_descsz = size;
1800 		note->n_type = type;
1801 		buf += sizeof(*note);
1802 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1803 		    sizeof(FREEBSD_ABI_VENDOR));
1804 		append_note_data(src, buf, size);
1805 		if (descp != NULL)
1806 			*descp = buf;
1807 	}
1808 
1809 	notesize = sizeof(Elf_Note) +		/* note header */
1810 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1811 						/* note name */
1812 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1813 
1814 	return (notesize);
1815 }
1816 
1817 static void
1818 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1819 {
1820 	Elf_Note note;
1821 	ssize_t old_len, sect_len;
1822 	size_t new_len, descsz, i;
1823 
1824 	if (ninfo->type == -1) {
1825 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1826 		return;
1827 	}
1828 
1829 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1830 	note.n_descsz = ninfo->outsize;
1831 	note.n_type = ninfo->type;
1832 
1833 	sbuf_bcat(sb, &note, sizeof(note));
1834 	sbuf_start_section(sb, &old_len);
1835 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1836 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1837 	if (note.n_descsz == 0)
1838 		return;
1839 	sbuf_start_section(sb, &old_len);
1840 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1841 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1842 	if (sect_len < 0)
1843 		return;
1844 
1845 	new_len = (size_t)sect_len;
1846 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1847 	if (new_len < descsz) {
1848 		/*
1849 		 * It is expected that individual note emitters will correctly
1850 		 * predict their expected output size and fill up to that size
1851 		 * themselves, padding in a format-specific way if needed.
1852 		 * However, in case they don't, just do it here with zeros.
1853 		 */
1854 		for (i = 0; i < descsz - new_len; i++)
1855 			sbuf_putc(sb, 0);
1856 	} else if (new_len > descsz) {
1857 		/*
1858 		 * We can't always truncate sb -- we may have drained some
1859 		 * of it already.
1860 		 */
1861 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1862 		    "read it (%zu > %zu).  Since it is longer than "
1863 		    "expected, this coredump's notes are corrupt.  THIS "
1864 		    "IS A BUG in the note_procstat routine for type %u.\n",
1865 		    __func__, (unsigned)note.n_type, new_len, descsz,
1866 		    (unsigned)note.n_type));
1867 	}
1868 }
1869 
1870 /*
1871  * Miscellaneous note out functions.
1872  */
1873 
1874 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1875 #include <compat/freebsd32/freebsd32.h>
1876 #include <compat/freebsd32/freebsd32_signal.h>
1877 
1878 typedef struct prstatus32 elf_prstatus_t;
1879 typedef struct prpsinfo32 elf_prpsinfo_t;
1880 typedef struct fpreg32 elf_prfpregset_t;
1881 typedef struct fpreg32 elf_fpregset_t;
1882 typedef struct reg32 elf_gregset_t;
1883 typedef struct thrmisc32 elf_thrmisc_t;
1884 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1885 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1886 typedef uint32_t elf_ps_strings_t;
1887 #else
1888 typedef prstatus_t elf_prstatus_t;
1889 typedef prpsinfo_t elf_prpsinfo_t;
1890 typedef prfpregset_t elf_prfpregset_t;
1891 typedef prfpregset_t elf_fpregset_t;
1892 typedef gregset_t elf_gregset_t;
1893 typedef thrmisc_t elf_thrmisc_t;
1894 #define ELF_KERN_PROC_MASK	0
1895 typedef struct kinfo_proc elf_kinfo_proc_t;
1896 typedef vm_offset_t elf_ps_strings_t;
1897 #endif
1898 
1899 static void
1900 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1901 {
1902 	struct sbuf sbarg;
1903 	size_t len;
1904 	char *cp, *end;
1905 	struct proc *p;
1906 	elf_prpsinfo_t *psinfo;
1907 	int error;
1908 
1909 	p = (struct proc *)arg;
1910 	if (sb != NULL) {
1911 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1912 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1913 		psinfo->pr_version = PRPSINFO_VERSION;
1914 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1915 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1916 		PROC_LOCK(p);
1917 		if (p->p_args != NULL) {
1918 			len = sizeof(psinfo->pr_psargs) - 1;
1919 			if (len > p->p_args->ar_length)
1920 				len = p->p_args->ar_length;
1921 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1922 			PROC_UNLOCK(p);
1923 			error = 0;
1924 		} else {
1925 			_PHOLD(p);
1926 			PROC_UNLOCK(p);
1927 			sbuf_new(&sbarg, psinfo->pr_psargs,
1928 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1929 			error = proc_getargv(curthread, p, &sbarg);
1930 			PRELE(p);
1931 			if (sbuf_finish(&sbarg) == 0)
1932 				len = sbuf_len(&sbarg) - 1;
1933 			else
1934 				len = sizeof(psinfo->pr_psargs) - 1;
1935 			sbuf_delete(&sbarg);
1936 		}
1937 		if (error || len == 0)
1938 			strlcpy(psinfo->pr_psargs, p->p_comm,
1939 			    sizeof(psinfo->pr_psargs));
1940 		else {
1941 			KASSERT(len < sizeof(psinfo->pr_psargs),
1942 			    ("len is too long: %zu vs %zu", len,
1943 			    sizeof(psinfo->pr_psargs)));
1944 			cp = psinfo->pr_psargs;
1945 			end = cp + len - 1;
1946 			for (;;) {
1947 				cp = memchr(cp, '\0', end - cp);
1948 				if (cp == NULL)
1949 					break;
1950 				*cp = ' ';
1951 			}
1952 		}
1953 		psinfo->pr_pid = p->p_pid;
1954 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1955 		free(psinfo, M_TEMP);
1956 	}
1957 	*sizep = sizeof(*psinfo);
1958 }
1959 
1960 static void
1961 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1962 {
1963 	struct thread *td;
1964 	elf_prstatus_t *status;
1965 
1966 	td = (struct thread *)arg;
1967 	if (sb != NULL) {
1968 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1969 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1970 		status->pr_version = PRSTATUS_VERSION;
1971 		status->pr_statussz = sizeof(elf_prstatus_t);
1972 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1973 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1974 		status->pr_osreldate = osreldate;
1975 		status->pr_cursig = td->td_proc->p_sig;
1976 		status->pr_pid = td->td_tid;
1977 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1978 		fill_regs32(td, &status->pr_reg);
1979 #else
1980 		fill_regs(td, &status->pr_reg);
1981 #endif
1982 		sbuf_bcat(sb, status, sizeof(*status));
1983 		free(status, M_TEMP);
1984 	}
1985 	*sizep = sizeof(*status);
1986 }
1987 
1988 static void
1989 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1990 {
1991 	struct thread *td;
1992 	elf_prfpregset_t *fpregset;
1993 
1994 	td = (struct thread *)arg;
1995 	if (sb != NULL) {
1996 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1997 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1998 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1999 		fill_fpregs32(td, fpregset);
2000 #else
2001 		fill_fpregs(td, fpregset);
2002 #endif
2003 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2004 		free(fpregset, M_TEMP);
2005 	}
2006 	*sizep = sizeof(*fpregset);
2007 }
2008 
2009 static void
2010 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2011 {
2012 	struct thread *td;
2013 	elf_thrmisc_t thrmisc;
2014 
2015 	td = (struct thread *)arg;
2016 	if (sb != NULL) {
2017 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2018 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2019 		strcpy(thrmisc.pr_tname, td->td_name);
2020 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2021 	}
2022 	*sizep = sizeof(thrmisc);
2023 }
2024 
2025 static void
2026 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2027 {
2028 	struct thread *td;
2029 	size_t size;
2030 	int structsize;
2031 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2032 	struct ptrace_lwpinfo32 pl;
2033 #else
2034 	struct ptrace_lwpinfo pl;
2035 #endif
2036 
2037 	td = (struct thread *)arg;
2038 	size = sizeof(structsize) + sizeof(pl);
2039 	if (sb != NULL) {
2040 		KASSERT(*sizep == size, ("invalid size"));
2041 		structsize = sizeof(pl);
2042 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2043 		bzero(&pl, sizeof(pl));
2044 		pl.pl_lwpid = td->td_tid;
2045 		pl.pl_event = PL_EVENT_NONE;
2046 		pl.pl_sigmask = td->td_sigmask;
2047 		pl.pl_siglist = td->td_siglist;
2048 		if (td->td_si.si_signo != 0) {
2049 			pl.pl_event = PL_EVENT_SIGNAL;
2050 			pl.pl_flags |= PL_FLAG_SI;
2051 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2052 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2053 #else
2054 			pl.pl_siginfo = td->td_si;
2055 #endif
2056 		}
2057 		strcpy(pl.pl_tdname, td->td_name);
2058 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2059 		sbuf_bcat(sb, &pl, sizeof(pl));
2060 	}
2061 	*sizep = size;
2062 }
2063 
2064 /*
2065  * Allow for MD specific notes, as well as any MD
2066  * specific preparations for writing MI notes.
2067  */
2068 static void
2069 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2070 {
2071 	struct thread *td;
2072 	void *buf;
2073 	size_t size;
2074 
2075 	td = (struct thread *)arg;
2076 	size = *sizep;
2077 	if (size != 0 && sb != NULL)
2078 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2079 	else
2080 		buf = NULL;
2081 	size = 0;
2082 	__elfN(dump_thread)(td, buf, &size);
2083 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2084 	if (size != 0 && sb != NULL)
2085 		sbuf_bcat(sb, buf, size);
2086 	free(buf, M_TEMP);
2087 	*sizep = size;
2088 }
2089 
2090 #ifdef KINFO_PROC_SIZE
2091 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2092 #endif
2093 
2094 static void
2095 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2096 {
2097 	struct proc *p;
2098 	size_t size;
2099 	int structsize;
2100 
2101 	p = (struct proc *)arg;
2102 	size = sizeof(structsize) + p->p_numthreads *
2103 	    sizeof(elf_kinfo_proc_t);
2104 
2105 	if (sb != NULL) {
2106 		KASSERT(*sizep == size, ("invalid size"));
2107 		structsize = sizeof(elf_kinfo_proc_t);
2108 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2109 		sx_slock(&proctree_lock);
2110 		PROC_LOCK(p);
2111 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2112 		sx_sunlock(&proctree_lock);
2113 	}
2114 	*sizep = size;
2115 }
2116 
2117 #ifdef KINFO_FILE_SIZE
2118 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2119 #endif
2120 
2121 static void
2122 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2123 {
2124 	struct proc *p;
2125 	size_t size, sect_sz, i;
2126 	ssize_t start_len, sect_len;
2127 	int structsize, filedesc_flags;
2128 
2129 	if (coredump_pack_fileinfo)
2130 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2131 	else
2132 		filedesc_flags = 0;
2133 
2134 	p = (struct proc *)arg;
2135 	structsize = sizeof(struct kinfo_file);
2136 	if (sb == NULL) {
2137 		size = 0;
2138 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2139 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2140 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2141 		PROC_LOCK(p);
2142 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2143 		sbuf_finish(sb);
2144 		sbuf_delete(sb);
2145 		*sizep = size;
2146 	} else {
2147 		sbuf_start_section(sb, &start_len);
2148 
2149 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2150 		PROC_LOCK(p);
2151 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2152 		    filedesc_flags);
2153 
2154 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2155 		if (sect_len < 0)
2156 			return;
2157 		sect_sz = sect_len;
2158 
2159 		KASSERT(sect_sz <= *sizep,
2160 		    ("kern_proc_filedesc_out did not respect maxlen; "
2161 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2162 		     sect_sz - sizeof(structsize)));
2163 
2164 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2165 			sbuf_putc(sb, 0);
2166 	}
2167 }
2168 
2169 #ifdef KINFO_VMENTRY_SIZE
2170 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2171 #endif
2172 
2173 static void
2174 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2175 {
2176 	struct proc *p;
2177 	size_t size;
2178 	int structsize, vmmap_flags;
2179 
2180 	if (coredump_pack_vmmapinfo)
2181 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2182 	else
2183 		vmmap_flags = 0;
2184 
2185 	p = (struct proc *)arg;
2186 	structsize = sizeof(struct kinfo_vmentry);
2187 	if (sb == NULL) {
2188 		size = 0;
2189 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2190 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2191 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2192 		PROC_LOCK(p);
2193 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2194 		sbuf_finish(sb);
2195 		sbuf_delete(sb);
2196 		*sizep = size;
2197 	} else {
2198 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2199 		PROC_LOCK(p);
2200 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2201 		    vmmap_flags);
2202 	}
2203 }
2204 
2205 static void
2206 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2207 {
2208 	struct proc *p;
2209 	size_t size;
2210 	int structsize;
2211 
2212 	p = (struct proc *)arg;
2213 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2214 	if (sb != NULL) {
2215 		KASSERT(*sizep == size, ("invalid size"));
2216 		structsize = sizeof(gid_t);
2217 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2218 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2219 		    sizeof(gid_t));
2220 	}
2221 	*sizep = size;
2222 }
2223 
2224 static void
2225 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2226 {
2227 	struct proc *p;
2228 	size_t size;
2229 	int structsize;
2230 
2231 	p = (struct proc *)arg;
2232 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2233 	if (sb != NULL) {
2234 		KASSERT(*sizep == size, ("invalid size"));
2235 		structsize = sizeof(p->p_fd->fd_cmask);
2236 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2237 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2238 	}
2239 	*sizep = size;
2240 }
2241 
2242 static void
2243 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2244 {
2245 	struct proc *p;
2246 	struct rlimit rlim[RLIM_NLIMITS];
2247 	size_t size;
2248 	int structsize, i;
2249 
2250 	p = (struct proc *)arg;
2251 	size = sizeof(structsize) + sizeof(rlim);
2252 	if (sb != NULL) {
2253 		KASSERT(*sizep == size, ("invalid size"));
2254 		structsize = sizeof(rlim);
2255 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2256 		PROC_LOCK(p);
2257 		for (i = 0; i < RLIM_NLIMITS; i++)
2258 			lim_rlimit_proc(p, i, &rlim[i]);
2259 		PROC_UNLOCK(p);
2260 		sbuf_bcat(sb, rlim, sizeof(rlim));
2261 	}
2262 	*sizep = size;
2263 }
2264 
2265 static void
2266 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2267 {
2268 	struct proc *p;
2269 	size_t size;
2270 	int structsize;
2271 
2272 	p = (struct proc *)arg;
2273 	size = sizeof(structsize) + sizeof(p->p_osrel);
2274 	if (sb != NULL) {
2275 		KASSERT(*sizep == size, ("invalid size"));
2276 		structsize = sizeof(p->p_osrel);
2277 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2278 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2279 	}
2280 	*sizep = size;
2281 }
2282 
2283 static void
2284 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2285 {
2286 	struct proc *p;
2287 	elf_ps_strings_t ps_strings;
2288 	size_t size;
2289 	int structsize;
2290 
2291 	p = (struct proc *)arg;
2292 	size = sizeof(structsize) + sizeof(ps_strings);
2293 	if (sb != NULL) {
2294 		KASSERT(*sizep == size, ("invalid size"));
2295 		structsize = sizeof(ps_strings);
2296 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2297 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2298 #else
2299 		ps_strings = p->p_sysent->sv_psstrings;
2300 #endif
2301 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2302 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2303 	}
2304 	*sizep = size;
2305 }
2306 
2307 static void
2308 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2309 {
2310 	struct proc *p;
2311 	size_t size;
2312 	int structsize;
2313 
2314 	p = (struct proc *)arg;
2315 	if (sb == NULL) {
2316 		size = 0;
2317 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2318 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2319 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2320 		PHOLD(p);
2321 		proc_getauxv(curthread, p, sb);
2322 		PRELE(p);
2323 		sbuf_finish(sb);
2324 		sbuf_delete(sb);
2325 		*sizep = size;
2326 	} else {
2327 		structsize = sizeof(Elf_Auxinfo);
2328 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2329 		PHOLD(p);
2330 		proc_getauxv(curthread, p, sb);
2331 		PRELE(p);
2332 	}
2333 }
2334 
2335 static boolean_t
2336 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2337     int32_t *osrel, const Elf_Phdr *pnote)
2338 {
2339 	const Elf_Note *note, *note0, *note_end;
2340 	const char *note_name;
2341 	char *buf;
2342 	int i, error;
2343 	boolean_t res;
2344 
2345 	/* We need some limit, might as well use PAGE_SIZE. */
2346 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2347 		return (FALSE);
2348 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2349 	if (pnote->p_offset > PAGE_SIZE ||
2350 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2351 		VOP_UNLOCK(imgp->vp, 0);
2352 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2353 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2354 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2355 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2356 		    curthread->td_ucred, NOCRED, NULL, curthread);
2357 		if (error != 0) {
2358 			uprintf("i/o error PT_NOTE\n");
2359 			res = FALSE;
2360 			goto ret;
2361 		}
2362 		note = note0 = (const Elf_Note *)buf;
2363 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2364 	} else {
2365 		note = note0 = (const Elf_Note *)(imgp->image_header +
2366 		    pnote->p_offset);
2367 		note_end = (const Elf_Note *)(imgp->image_header +
2368 		    pnote->p_offset + pnote->p_filesz);
2369 		buf = NULL;
2370 	}
2371 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2372 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2373 		    (const char *)note < sizeof(Elf_Note)) {
2374 			res = FALSE;
2375 			goto ret;
2376 		}
2377 		if (note->n_namesz != checknote->hdr.n_namesz ||
2378 		    note->n_descsz != checknote->hdr.n_descsz ||
2379 		    note->n_type != checknote->hdr.n_type)
2380 			goto nextnote;
2381 		note_name = (const char *)(note + 1);
2382 		if (note_name + checknote->hdr.n_namesz >=
2383 		    (const char *)note_end || strncmp(checknote->vendor,
2384 		    note_name, checknote->hdr.n_namesz) != 0)
2385 			goto nextnote;
2386 
2387 		/*
2388 		 * Fetch the osreldate for binary
2389 		 * from the ELF OSABI-note if necessary.
2390 		 */
2391 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2392 		    checknote->trans_osrel != NULL) {
2393 			res = checknote->trans_osrel(note, osrel);
2394 			goto ret;
2395 		}
2396 		res = TRUE;
2397 		goto ret;
2398 nextnote:
2399 		note = (const Elf_Note *)((const char *)(note + 1) +
2400 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2401 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2402 	}
2403 	res = FALSE;
2404 ret:
2405 	free(buf, M_TEMP);
2406 	return (res);
2407 }
2408 
2409 /*
2410  * Try to find the appropriate ABI-note section for checknote,
2411  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2412  * first page of the image is searched, the same as for headers.
2413  */
2414 static boolean_t
2415 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2416     int32_t *osrel)
2417 {
2418 	const Elf_Phdr *phdr;
2419 	const Elf_Ehdr *hdr;
2420 	int i;
2421 
2422 	hdr = (const Elf_Ehdr *)imgp->image_header;
2423 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2424 
2425 	for (i = 0; i < hdr->e_phnum; i++) {
2426 		if (phdr[i].p_type == PT_NOTE &&
2427 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2428 			return (TRUE);
2429 	}
2430 	return (FALSE);
2431 
2432 }
2433 
2434 /*
2435  * Tell kern_execve.c about it, with a little help from the linker.
2436  */
2437 static struct execsw __elfN(execsw) = {
2438 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2439 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2440 };
2441 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2442 
2443 static vm_prot_t
2444 __elfN(trans_prot)(Elf_Word flags)
2445 {
2446 	vm_prot_t prot;
2447 
2448 	prot = 0;
2449 	if (flags & PF_X)
2450 		prot |= VM_PROT_EXECUTE;
2451 	if (flags & PF_W)
2452 		prot |= VM_PROT_WRITE;
2453 	if (flags & PF_R)
2454 		prot |= VM_PROT_READ;
2455 #if __ELF_WORD_SIZE == 32
2456 #if defined(__amd64__)
2457 	if (i386_read_exec && (flags & PF_R))
2458 		prot |= VM_PROT_EXECUTE;
2459 #endif
2460 #endif
2461 	return (prot);
2462 }
2463 
2464 static Elf_Word
2465 __elfN(untrans_prot)(vm_prot_t prot)
2466 {
2467 	Elf_Word flags;
2468 
2469 	flags = 0;
2470 	if (prot & VM_PROT_EXECUTE)
2471 		flags |= PF_X;
2472 	if (prot & VM_PROT_READ)
2473 		flags |= PF_R;
2474 	if (prot & VM_PROT_WRITE)
2475 		flags |= PF_W;
2476 	return (flags);
2477 }
2478