xref: /freebsd/sys/kern/imgact_elf.c (revision 8ef24a0d4b28fe230e20637f56869cc4148cd2ca)
1 /*-
2  * Copyright (c) 2000 David O'Brien
3  * Copyright (c) 1995-1996 Søren Schmidt
4  * Copyright (c) 1996 Peter Wemm
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_gzio.h"
37 
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
40 #include <sys/exec.h>
41 #include <sys/fcntl.h>
42 #include <sys/gzio.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mman.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
58 #include <sys/sbuf.h>
59 #include <sys/sf_buf.h>
60 #include <sys/smp.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
63 #include <sys/stat.h>
64 #include <sys/sx.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
71 #include <sys/user.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
80 
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
83 #ifdef __arm__
84 #include <machine/acle-compat.h>
85 #endif
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_offset_t offset,
96     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
97     size_t pagesize);
98 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
99 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
100     int32_t *osrel);
101 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
102 static boolean_t __elfN(check_note)(struct image_params *imgp,
103     Elf_Brandnote *checknote, int32_t *osrel);
104 static vm_prot_t __elfN(trans_prot)(Elf_Word);
105 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
106 
107 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108     "");
109 
110 #define	CORE_BUF_SIZE	(16 * 1024)
111 
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119     &elf_legacy_coredump, 0, "");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if __ELF_WORD_SIZE == 32
133 #if defined(__amd64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136     "enable execution from readable segments");
137 #endif
138 #endif
139 
140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
141 
142 #define	trunc_page_ps(va, ps)	rounddown2(va, ps)
143 #define	round_page_ps(va, ps)	roundup2(va, ps)
144 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
145 
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
147 
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
150 	.hdr.n_descsz	= sizeof(int32_t),
151 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
152 	.vendor		= FREEBSD_ABI_VENDOR,
153 	.flags		= BN_TRANSLATE_OSREL,
154 	.trans_osrel	= __elfN(freebsd_trans_osrel)
155 };
156 
157 static boolean_t
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
159 {
160 	uintptr_t p;
161 
162 	p = (uintptr_t)(note + 1);
163 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 	*osrel = *(const int32_t *)(p);
165 
166 	return (TRUE);
167 }
168 
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
171 
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
174 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
175 	.hdr.n_type	= 1,
176 	.vendor		= GNU_ABI_VENDOR,
177 	.flags		= BN_TRANSLATE_OSREL,
178 	.trans_osrel	= kfreebsd_trans_osrel
179 };
180 
181 static boolean_t
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
183 {
184 	const Elf32_Word *desc;
185 	uintptr_t p;
186 
187 	p = (uintptr_t)(note + 1);
188 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
189 
190 	desc = (const Elf32_Word *)p;
191 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
192 		return (FALSE);
193 
194 	/*
195 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
197 	 */
198 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
199 
200 	return (TRUE);
201 }
202 
203 int
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
205 {
206 	int i;
207 
208 	for (i = 0; i < MAX_BRANDS; i++) {
209 		if (elf_brand_list[i] == NULL) {
210 			elf_brand_list[i] = entry;
211 			break;
212 		}
213 	}
214 	if (i == MAX_BRANDS) {
215 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
216 			__func__, entry);
217 		return (-1);
218 	}
219 	return (0);
220 }
221 
222 int
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
224 {
225 	int i;
226 
227 	for (i = 0; i < MAX_BRANDS; i++) {
228 		if (elf_brand_list[i] == entry) {
229 			elf_brand_list[i] = NULL;
230 			break;
231 		}
232 	}
233 	if (i == MAX_BRANDS)
234 		return (-1);
235 	return (0);
236 }
237 
238 int
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
240 {
241 	struct proc *p;
242 	int rval = FALSE;
243 
244 	sx_slock(&allproc_lock);
245 	FOREACH_PROC_IN_SYSTEM(p) {
246 		if (p->p_sysent == entry->sysvec) {
247 			rval = TRUE;
248 			break;
249 		}
250 	}
251 	sx_sunlock(&allproc_lock);
252 
253 	return (rval);
254 }
255 
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258     int interp_name_len, int32_t *osrel)
259 {
260 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
261 	Elf_Brandinfo *bi, *bi_m;
262 	boolean_t ret;
263 	int i;
264 
265 	/*
266 	 * We support four types of branding -- (1) the ELF EI_OSABI field
267 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 	 * branding w/in the ELF header, (3) path of the `interp_path'
269 	 * field, and (4) the ".note.ABI-tag" ELF section.
270 	 */
271 
272 	/* Look for an ".note.ABI-tag" ELF section */
273 	bi_m = NULL;
274 	for (i = 0; i < MAX_BRANDS; i++) {
275 		bi = elf_brand_list[i];
276 		if (bi == NULL)
277 			continue;
278 		if (hdr->e_machine == bi->machine && (bi->flags &
279 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
280 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
281 			/* Give brand a chance to veto check_note's guess */
282 			if (ret && bi->header_supported)
283 				ret = bi->header_supported(imgp);
284 			/*
285 			 * If note checker claimed the binary, but the
286 			 * interpreter path in the image does not
287 			 * match default one for the brand, try to
288 			 * search for other brands with the same
289 			 * interpreter.  Either there is better brand
290 			 * with the right interpreter, or, failing
291 			 * this, we return first brand which accepted
292 			 * our note and, optionally, header.
293 			 */
294 			if (ret && bi_m == NULL && (strlen(bi->interp_path) +
295 			    1 != interp_name_len || strncmp(interp,
296 			    bi->interp_path, interp_name_len) != 0)) {
297 				bi_m = bi;
298 				ret = 0;
299 			}
300 			if (ret)
301 				return (bi);
302 		}
303 	}
304 	if (bi_m != NULL)
305 		return (bi_m);
306 
307 	/* If the executable has a brand, search for it in the brand list. */
308 	for (i = 0; i < MAX_BRANDS; i++) {
309 		bi = elf_brand_list[i];
310 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
311 			continue;
312 		if (hdr->e_machine == bi->machine &&
313 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
314 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
315 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
316 			/* Looks good, but give brand a chance to veto */
317 			if (!bi->header_supported || bi->header_supported(imgp))
318 				return (bi);
319 		}
320 	}
321 
322 	/* No known brand, see if the header is recognized by any brand */
323 	for (i = 0; i < MAX_BRANDS; i++) {
324 		bi = elf_brand_list[i];
325 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
326 		    bi->header_supported == NULL)
327 			continue;
328 		if (hdr->e_machine == bi->machine) {
329 			ret = bi->header_supported(imgp);
330 			if (ret)
331 				return (bi);
332 		}
333 	}
334 
335 	/* Lacking a known brand, search for a recognized interpreter. */
336 	if (interp != NULL) {
337 		for (i = 0; i < MAX_BRANDS; i++) {
338 			bi = elf_brand_list[i];
339 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
340 				continue;
341 			if (hdr->e_machine == bi->machine &&
342 			    /* ELF image p_filesz includes terminating zero */
343 			    strlen(bi->interp_path) + 1 == interp_name_len &&
344 			    strncmp(interp, bi->interp_path, interp_name_len)
345 			    == 0)
346 				return (bi);
347 		}
348 	}
349 
350 	/* Lacking a recognized interpreter, try the default brand */
351 	for (i = 0; i < MAX_BRANDS; i++) {
352 		bi = elf_brand_list[i];
353 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
354 			continue;
355 		if (hdr->e_machine == bi->machine &&
356 		    __elfN(fallback_brand) == bi->brand)
357 			return (bi);
358 	}
359 	return (NULL);
360 }
361 
362 static int
363 __elfN(check_header)(const Elf_Ehdr *hdr)
364 {
365 	Elf_Brandinfo *bi;
366 	int i;
367 
368 	if (!IS_ELF(*hdr) ||
369 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
370 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
371 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
372 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
373 	    hdr->e_version != ELF_TARG_VER)
374 		return (ENOEXEC);
375 
376 	/*
377 	 * Make sure we have at least one brand for this machine.
378 	 */
379 
380 	for (i = 0; i < MAX_BRANDS; i++) {
381 		bi = elf_brand_list[i];
382 		if (bi != NULL && bi->machine == hdr->e_machine)
383 			break;
384 	}
385 	if (i == MAX_BRANDS)
386 		return (ENOEXEC);
387 
388 	return (0);
389 }
390 
391 static int
392 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
393     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
394 {
395 	struct sf_buf *sf;
396 	int error;
397 	vm_offset_t off;
398 
399 	/*
400 	 * Create the page if it doesn't exist yet. Ignore errors.
401 	 */
402 	vm_map_lock(map);
403 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
404 	    VM_PROT_ALL, VM_PROT_ALL, 0);
405 	vm_map_unlock(map);
406 
407 	/*
408 	 * Find the page from the underlying object.
409 	 */
410 	if (object) {
411 		sf = vm_imgact_map_page(object, offset);
412 		if (sf == NULL)
413 			return (KERN_FAILURE);
414 		off = offset - trunc_page(offset);
415 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
416 		    end - start);
417 		vm_imgact_unmap_page(sf);
418 		if (error) {
419 			return (KERN_FAILURE);
420 		}
421 	}
422 
423 	return (KERN_SUCCESS);
424 }
425 
426 static int
427 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
428     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
429 {
430 	struct sf_buf *sf;
431 	vm_offset_t off;
432 	vm_size_t sz;
433 	int error, rv;
434 
435 	if (start != trunc_page(start)) {
436 		rv = __elfN(map_partial)(map, object, offset, start,
437 		    round_page(start), prot);
438 		if (rv)
439 			return (rv);
440 		offset += round_page(start) - start;
441 		start = round_page(start);
442 	}
443 	if (end != round_page(end)) {
444 		rv = __elfN(map_partial)(map, object, offset +
445 		    trunc_page(end) - start, trunc_page(end), end, prot);
446 		if (rv)
447 			return (rv);
448 		end = trunc_page(end);
449 	}
450 	if (end > start) {
451 		if (offset & PAGE_MASK) {
452 			/*
453 			 * The mapping is not page aligned. This means we have
454 			 * to copy the data. Sigh.
455 			 */
456 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
457 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
458 			    0);
459 			if (rv)
460 				return (rv);
461 			if (object == NULL)
462 				return (KERN_SUCCESS);
463 			for (; start < end; start += sz) {
464 				sf = vm_imgact_map_page(object, offset);
465 				if (sf == NULL)
466 					return (KERN_FAILURE);
467 				off = offset - trunc_page(offset);
468 				sz = end - start;
469 				if (sz > PAGE_SIZE - off)
470 					sz = PAGE_SIZE - off;
471 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
472 				    (caddr_t)start, sz);
473 				vm_imgact_unmap_page(sf);
474 				if (error) {
475 					return (KERN_FAILURE);
476 				}
477 				offset += sz;
478 			}
479 			rv = KERN_SUCCESS;
480 		} else {
481 			vm_object_reference(object);
482 			vm_map_lock(map);
483 			rv = vm_map_insert(map, object, offset, start, end,
484 			    prot, VM_PROT_ALL, cow);
485 			vm_map_unlock(map);
486 			if (rv != KERN_SUCCESS)
487 				vm_object_deallocate(object);
488 		}
489 		return (rv);
490 	} else {
491 		return (KERN_SUCCESS);
492 	}
493 }
494 
495 static int
496 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
497     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
498     size_t pagesize)
499 {
500 	struct sf_buf *sf;
501 	size_t map_len;
502 	vm_map_t map;
503 	vm_object_t object;
504 	vm_offset_t map_addr;
505 	int error, rv, cow;
506 	size_t copy_len;
507 	vm_offset_t file_addr;
508 
509 	/*
510 	 * It's necessary to fail if the filsz + offset taken from the
511 	 * header is greater than the actual file pager object's size.
512 	 * If we were to allow this, then the vm_map_find() below would
513 	 * walk right off the end of the file object and into the ether.
514 	 *
515 	 * While I'm here, might as well check for something else that
516 	 * is invalid: filsz cannot be greater than memsz.
517 	 */
518 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
519 		uprintf("elf_load_section: truncated ELF file\n");
520 		return (ENOEXEC);
521 	}
522 
523 	object = imgp->object;
524 	map = &imgp->proc->p_vmspace->vm_map;
525 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
526 	file_addr = trunc_page_ps(offset, pagesize);
527 
528 	/*
529 	 * We have two choices.  We can either clear the data in the last page
530 	 * of an oversized mapping, or we can start the anon mapping a page
531 	 * early and copy the initialized data into that first page.  We
532 	 * choose the second..
533 	 */
534 	if (memsz > filsz)
535 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
536 	else
537 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
538 
539 	if (map_len != 0) {
540 		/* cow flags: don't dump readonly sections in core */
541 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
542 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
543 
544 		rv = __elfN(map_insert)(map,
545 				      object,
546 				      file_addr,	/* file offset */
547 				      map_addr,		/* virtual start */
548 				      map_addr + map_len,/* virtual end */
549 				      prot,
550 				      cow);
551 		if (rv != KERN_SUCCESS)
552 			return (EINVAL);
553 
554 		/* we can stop now if we've covered it all */
555 		if (memsz == filsz) {
556 			return (0);
557 		}
558 	}
559 
560 
561 	/*
562 	 * We have to get the remaining bit of the file into the first part
563 	 * of the oversized map segment.  This is normally because the .data
564 	 * segment in the file is extended to provide bss.  It's a neat idea
565 	 * to try and save a page, but it's a pain in the behind to implement.
566 	 */
567 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
568 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
569 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
570 	    map_addr;
571 
572 	/* This had damn well better be true! */
573 	if (map_len != 0) {
574 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
575 		    map_len, VM_PROT_ALL, 0);
576 		if (rv != KERN_SUCCESS) {
577 			return (EINVAL);
578 		}
579 	}
580 
581 	if (copy_len != 0) {
582 		vm_offset_t off;
583 
584 		sf = vm_imgact_map_page(object, offset + filsz);
585 		if (sf == NULL)
586 			return (EIO);
587 
588 		/* send the page fragment to user space */
589 		off = trunc_page_ps(offset + filsz, pagesize) -
590 		    trunc_page(offset + filsz);
591 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
592 		    (caddr_t)map_addr, copy_len);
593 		vm_imgact_unmap_page(sf);
594 		if (error) {
595 			return (error);
596 		}
597 	}
598 
599 	/*
600 	 * set it to the specified protection.
601 	 * XXX had better undo the damage from pasting over the cracks here!
602 	 */
603 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
604 	    map_len), prot, FALSE);
605 
606 	return (0);
607 }
608 
609 /*
610  * Load the file "file" into memory.  It may be either a shared object
611  * or an executable.
612  *
613  * The "addr" reference parameter is in/out.  On entry, it specifies
614  * the address where a shared object should be loaded.  If the file is
615  * an executable, this value is ignored.  On exit, "addr" specifies
616  * where the file was actually loaded.
617  *
618  * The "entry" reference parameter is out only.  On exit, it specifies
619  * the entry point for the loaded file.
620  */
621 static int
622 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
623 	u_long *entry, size_t pagesize)
624 {
625 	struct {
626 		struct nameidata nd;
627 		struct vattr attr;
628 		struct image_params image_params;
629 	} *tempdata;
630 	const Elf_Ehdr *hdr = NULL;
631 	const Elf_Phdr *phdr = NULL;
632 	struct nameidata *nd;
633 	struct vattr *attr;
634 	struct image_params *imgp;
635 	vm_prot_t prot;
636 	u_long rbase;
637 	u_long base_addr = 0;
638 	int error, i, numsegs;
639 
640 #ifdef CAPABILITY_MODE
641 	/*
642 	 * XXXJA: This check can go away once we are sufficiently confident
643 	 * that the checks in namei() are correct.
644 	 */
645 	if (IN_CAPABILITY_MODE(curthread))
646 		return (ECAPMODE);
647 #endif
648 
649 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
650 	nd = &tempdata->nd;
651 	attr = &tempdata->attr;
652 	imgp = &tempdata->image_params;
653 
654 	/*
655 	 * Initialize part of the common data
656 	 */
657 	imgp->proc = p;
658 	imgp->attr = attr;
659 	imgp->firstpage = NULL;
660 	imgp->image_header = NULL;
661 	imgp->object = NULL;
662 	imgp->execlabel = NULL;
663 
664 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
665 	if ((error = namei(nd)) != 0) {
666 		nd->ni_vp = NULL;
667 		goto fail;
668 	}
669 	NDFREE(nd, NDF_ONLY_PNBUF);
670 	imgp->vp = nd->ni_vp;
671 
672 	/*
673 	 * Check permissions, modes, uid, etc on the file, and "open" it.
674 	 */
675 	error = exec_check_permissions(imgp);
676 	if (error)
677 		goto fail;
678 
679 	error = exec_map_first_page(imgp);
680 	if (error)
681 		goto fail;
682 
683 	/*
684 	 * Also make certain that the interpreter stays the same, so set
685 	 * its VV_TEXT flag, too.
686 	 */
687 	VOP_SET_TEXT(nd->ni_vp);
688 
689 	imgp->object = nd->ni_vp->v_object;
690 
691 	hdr = (const Elf_Ehdr *)imgp->image_header;
692 	if ((error = __elfN(check_header)(hdr)) != 0)
693 		goto fail;
694 	if (hdr->e_type == ET_DYN)
695 		rbase = *addr;
696 	else if (hdr->e_type == ET_EXEC)
697 		rbase = 0;
698 	else {
699 		error = ENOEXEC;
700 		goto fail;
701 	}
702 
703 	/* Only support headers that fit within first page for now      */
704 	if ((hdr->e_phoff > PAGE_SIZE) ||
705 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
706 		error = ENOEXEC;
707 		goto fail;
708 	}
709 
710 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
711 	if (!aligned(phdr, Elf_Addr)) {
712 		error = ENOEXEC;
713 		goto fail;
714 	}
715 
716 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
717 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
718 			/* Loadable segment */
719 			prot = __elfN(trans_prot)(phdr[i].p_flags);
720 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
721 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
722 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
723 			if (error != 0)
724 				goto fail;
725 			/*
726 			 * Establish the base address if this is the
727 			 * first segment.
728 			 */
729 			if (numsegs == 0)
730   				base_addr = trunc_page(phdr[i].p_vaddr +
731 				    rbase);
732 			numsegs++;
733 		}
734 	}
735 	*addr = base_addr;
736 	*entry = (unsigned long)hdr->e_entry + rbase;
737 
738 fail:
739 	if (imgp->firstpage)
740 		exec_unmap_first_page(imgp);
741 
742 	if (nd->ni_vp)
743 		vput(nd->ni_vp);
744 
745 	free(tempdata, M_TEMP);
746 
747 	return (error);
748 }
749 
750 static int
751 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
752 {
753 	struct thread *td;
754 	const Elf_Ehdr *hdr;
755 	const Elf_Phdr *phdr;
756 	Elf_Auxargs *elf_auxargs;
757 	struct vmspace *vmspace;
758 	const char *err_str, *newinterp;
759 	char *interp, *interp_buf, *path;
760 	Elf_Brandinfo *brand_info;
761 	struct sysentvec *sv;
762 	vm_prot_t prot;
763 	u_long text_size, data_size, total_size, text_addr, data_addr;
764 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
765 	int32_t osrel;
766 	int error, i, n, interp_name_len, have_interp;
767 
768 	hdr = (const Elf_Ehdr *)imgp->image_header;
769 
770 	/*
771 	 * Do we have a valid ELF header ?
772 	 *
773 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
774 	 * if particular brand doesn't support it.
775 	 */
776 	if (__elfN(check_header)(hdr) != 0 ||
777 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
778 		return (-1);
779 
780 	/*
781 	 * From here on down, we return an errno, not -1, as we've
782 	 * detected an ELF file.
783 	 */
784 
785 	if ((hdr->e_phoff > PAGE_SIZE) ||
786 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
787 		/* Only support headers in first page for now */
788 		uprintf("Program headers not in the first page\n");
789 		return (ENOEXEC);
790 	}
791 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
792 	if (!aligned(phdr, Elf_Addr)) {
793 		uprintf("Unaligned program headers\n");
794 		return (ENOEXEC);
795 	}
796 
797 	n = error = 0;
798 	baddr = 0;
799 	osrel = 0;
800 	text_size = data_size = total_size = text_addr = data_addr = 0;
801 	entry = proghdr = 0;
802 	interp_name_len = 0;
803 	err_str = newinterp = NULL;
804 	interp = interp_buf = NULL;
805 	td = curthread;
806 
807 	for (i = 0; i < hdr->e_phnum; i++) {
808 		switch (phdr[i].p_type) {
809 		case PT_LOAD:
810 			if (n == 0)
811 				baddr = phdr[i].p_vaddr;
812 			n++;
813 			break;
814 		case PT_INTERP:
815 			/* Path to interpreter */
816 			if (phdr[i].p_filesz > MAXPATHLEN) {
817 				uprintf("Invalid PT_INTERP\n");
818 				error = ENOEXEC;
819 				goto ret;
820 			}
821 			if (interp != NULL) {
822 				uprintf("Multiple PT_INTERP headers\n");
823 				error = ENOEXEC;
824 				goto ret;
825 			}
826 			interp_name_len = phdr[i].p_filesz;
827 			if (phdr[i].p_offset > PAGE_SIZE ||
828 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
829 				VOP_UNLOCK(imgp->vp, 0);
830 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
831 				    M_WAITOK);
832 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
833 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
834 				    interp_name_len, phdr[i].p_offset,
835 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
836 				    NOCRED, NULL, td);
837 				if (error != 0) {
838 					uprintf("i/o error PT_INTERP\n");
839 					goto ret;
840 				}
841 				interp_buf[interp_name_len] = '\0';
842 				interp = interp_buf;
843 			} else {
844 				interp = __DECONST(char *, imgp->image_header) +
845 				    phdr[i].p_offset;
846 			}
847 			break;
848 		case PT_GNU_STACK:
849 			if (__elfN(nxstack))
850 				imgp->stack_prot =
851 				    __elfN(trans_prot)(phdr[i].p_flags);
852 			imgp->stack_sz = phdr[i].p_memsz;
853 			break;
854 		}
855 	}
856 
857 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
858 	    &osrel);
859 	if (brand_info == NULL) {
860 		uprintf("ELF binary type \"%u\" not known.\n",
861 		    hdr->e_ident[EI_OSABI]);
862 		error = ENOEXEC;
863 		goto ret;
864 	}
865 	if (hdr->e_type == ET_DYN) {
866 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
867 			uprintf("Cannot execute shared object\n");
868 			error = ENOEXEC;
869 			goto ret;
870 		}
871 		/*
872 		 * Honour the base load address from the dso if it is
873 		 * non-zero for some reason.
874 		 */
875 		if (baddr == 0)
876 			et_dyn_addr = ET_DYN_LOAD_ADDR;
877 		else
878 			et_dyn_addr = 0;
879 	} else
880 		et_dyn_addr = 0;
881 	sv = brand_info->sysvec;
882 	if (interp != NULL && brand_info->interp_newpath != NULL)
883 		newinterp = brand_info->interp_newpath;
884 
885 	/*
886 	 * Avoid a possible deadlock if the current address space is destroyed
887 	 * and that address space maps the locked vnode.  In the common case,
888 	 * the locked vnode's v_usecount is decremented but remains greater
889 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
890 	 * However, in cases where the vnode lock is external, such as nullfs,
891 	 * v_usecount may become zero.
892 	 *
893 	 * The VV_TEXT flag prevents modifications to the executable while
894 	 * the vnode is unlocked.
895 	 */
896 	VOP_UNLOCK(imgp->vp, 0);
897 
898 	error = exec_new_vmspace(imgp, sv);
899 	imgp->proc->p_sysent = sv;
900 
901 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
902 	if (error != 0)
903 		goto ret;
904 
905 	for (i = 0; i < hdr->e_phnum; i++) {
906 		switch (phdr[i].p_type) {
907 		case PT_LOAD:	/* Loadable segment */
908 			if (phdr[i].p_memsz == 0)
909 				break;
910 			prot = __elfN(trans_prot)(phdr[i].p_flags);
911 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
912 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
913 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
914 			    sv->sv_pagesize);
915 			if (error != 0)
916 				goto ret;
917 
918 			/*
919 			 * If this segment contains the program headers,
920 			 * remember their virtual address for the AT_PHDR
921 			 * aux entry. Static binaries don't usually include
922 			 * a PT_PHDR entry.
923 			 */
924 			if (phdr[i].p_offset == 0 &&
925 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
926 				<= phdr[i].p_filesz)
927 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
928 				    et_dyn_addr;
929 
930 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
931 			seg_size = round_page(phdr[i].p_memsz +
932 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
933 
934 			/*
935 			 * Make the largest executable segment the official
936 			 * text segment and all others data.
937 			 *
938 			 * Note that obreak() assumes that data_addr +
939 			 * data_size == end of data load area, and the ELF
940 			 * file format expects segments to be sorted by
941 			 * address.  If multiple data segments exist, the
942 			 * last one will be used.
943 			 */
944 
945 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
946 				text_size = seg_size;
947 				text_addr = seg_addr;
948 			} else {
949 				data_size = seg_size;
950 				data_addr = seg_addr;
951 			}
952 			total_size += seg_size;
953 			break;
954 		case PT_PHDR: 	/* Program header table info */
955 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
956 			break;
957 		default:
958 			break;
959 		}
960 	}
961 
962 	if (data_addr == 0 && data_size == 0) {
963 		data_addr = text_addr;
964 		data_size = text_size;
965 	}
966 
967 	entry = (u_long)hdr->e_entry + et_dyn_addr;
968 
969 	/*
970 	 * Check limits.  It should be safe to check the
971 	 * limits after loading the segments since we do
972 	 * not actually fault in all the segments pages.
973 	 */
974 	PROC_LOCK(imgp->proc);
975 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
976 		err_str = "Data segment size exceeds process limit";
977 	else if (text_size > maxtsiz)
978 		err_str = "Text segment size exceeds system limit";
979 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
980 		err_str = "Total segment size exceeds process limit";
981 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
982 		err_str = "Data segment size exceeds resource limit";
983 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
984 		err_str = "Total segment size exceeds resource limit";
985 	if (err_str != NULL) {
986 		PROC_UNLOCK(imgp->proc);
987 		uprintf("%s\n", err_str);
988 		error = ENOMEM;
989 		goto ret;
990 	}
991 
992 	vmspace = imgp->proc->p_vmspace;
993 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
994 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
995 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
996 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
997 
998 	/*
999 	 * We load the dynamic linker where a userland call
1000 	 * to mmap(0, ...) would put it.  The rationale behind this
1001 	 * calculation is that it leaves room for the heap to grow to
1002 	 * its maximum allowed size.
1003 	 */
1004 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1005 	    RLIMIT_DATA));
1006 	PROC_UNLOCK(imgp->proc);
1007 
1008 	imgp->entry_addr = entry;
1009 
1010 	if (interp != NULL) {
1011 		have_interp = FALSE;
1012 		VOP_UNLOCK(imgp->vp, 0);
1013 		if (brand_info->emul_path != NULL &&
1014 		    brand_info->emul_path[0] != '\0') {
1015 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1016 			snprintf(path, MAXPATHLEN, "%s%s",
1017 			    brand_info->emul_path, interp);
1018 			error = __elfN(load_file)(imgp->proc, path, &addr,
1019 			    &imgp->entry_addr, sv->sv_pagesize);
1020 			free(path, M_TEMP);
1021 			if (error == 0)
1022 				have_interp = TRUE;
1023 		}
1024 		if (!have_interp && newinterp != NULL &&
1025 		    (brand_info->interp_path == NULL ||
1026 		    strcmp(interp, brand_info->interp_path) == 0)) {
1027 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1028 			    &imgp->entry_addr, sv->sv_pagesize);
1029 			if (error == 0)
1030 				have_interp = TRUE;
1031 		}
1032 		if (!have_interp) {
1033 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1034 			    &imgp->entry_addr, sv->sv_pagesize);
1035 		}
1036 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1037 		if (error != 0) {
1038 			uprintf("ELF interpreter %s not found, error %d\n",
1039 			    interp, error);
1040 			goto ret;
1041 		}
1042 	} else
1043 		addr = et_dyn_addr;
1044 
1045 	/*
1046 	 * Construct auxargs table (used by the fixup routine)
1047 	 */
1048 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1049 	elf_auxargs->execfd = -1;
1050 	elf_auxargs->phdr = proghdr;
1051 	elf_auxargs->phent = hdr->e_phentsize;
1052 	elf_auxargs->phnum = hdr->e_phnum;
1053 	elf_auxargs->pagesz = PAGE_SIZE;
1054 	elf_auxargs->base = addr;
1055 	elf_auxargs->flags = 0;
1056 	elf_auxargs->entry = entry;
1057 	elf_auxargs->hdr_eflags = hdr->e_flags;
1058 
1059 	imgp->auxargs = elf_auxargs;
1060 	imgp->interpreted = 0;
1061 	imgp->reloc_base = addr;
1062 	imgp->proc->p_osrel = osrel;
1063 
1064  ret:
1065 	free(interp_buf, M_TEMP);
1066 	return (error);
1067 }
1068 
1069 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1070 
1071 int
1072 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1073 {
1074 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1075 	Elf_Addr *base;
1076 	Elf_Addr *pos;
1077 
1078 	base = (Elf_Addr *)*stack_base;
1079 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1080 
1081 	if (args->execfd != -1)
1082 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1083 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1084 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1085 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1086 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1087 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1088 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1089 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1090 #ifdef AT_EHDRFLAGS
1091 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1092 #endif
1093 	if (imgp->execpathp != 0)
1094 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1095 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1096 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1097 	if (imgp->canary != 0) {
1098 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1099 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1100 	}
1101 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1102 	if (imgp->pagesizes != 0) {
1103 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1104 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1105 	}
1106 	if (imgp->sysent->sv_timekeep_base != 0) {
1107 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1108 		    imgp->sysent->sv_timekeep_base);
1109 	}
1110 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1111 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1112 	    imgp->sysent->sv_stackprot);
1113 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1114 
1115 	free(imgp->auxargs, M_TEMP);
1116 	imgp->auxargs = NULL;
1117 
1118 	base--;
1119 	suword(base, (long)imgp->args->argc);
1120 	*stack_base = (register_t *)base;
1121 	return (0);
1122 }
1123 
1124 /*
1125  * Code for generating ELF core dumps.
1126  */
1127 
1128 typedef void (*segment_callback)(vm_map_entry_t, void *);
1129 
1130 /* Closure for cb_put_phdr(). */
1131 struct phdr_closure {
1132 	Elf_Phdr *phdr;		/* Program header to fill in */
1133 	Elf_Off offset;		/* Offset of segment in core file */
1134 };
1135 
1136 /* Closure for cb_size_segment(). */
1137 struct sseg_closure {
1138 	int count;		/* Count of writable segments. */
1139 	size_t size;		/* Total size of all writable segments. */
1140 };
1141 
1142 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1143 
1144 struct note_info {
1145 	int		type;		/* Note type. */
1146 	outfunc_t 	outfunc; 	/* Output function. */
1147 	void		*outarg;	/* Argument for the output function. */
1148 	size_t		outsize;	/* Output size. */
1149 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1150 };
1151 
1152 TAILQ_HEAD(note_info_list, note_info);
1153 
1154 /* Coredump output parameters. */
1155 struct coredump_params {
1156 	off_t		offset;
1157 	struct ucred	*active_cred;
1158 	struct ucred	*file_cred;
1159 	struct thread	*td;
1160 	struct vnode	*vp;
1161 	struct gzio_stream *gzs;
1162 };
1163 
1164 static void cb_put_phdr(vm_map_entry_t, void *);
1165 static void cb_size_segment(vm_map_entry_t, void *);
1166 static int core_write(struct coredump_params *, void *, size_t, off_t,
1167     enum uio_seg);
1168 static void each_writable_segment(struct thread *, segment_callback, void *);
1169 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1170     struct note_info_list *, size_t);
1171 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1172     size_t *);
1173 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1174 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1175 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1176 static int sbuf_drain_core_output(void *, const char *, int);
1177 static int sbuf_drain_count(void *arg, const char *data, int len);
1178 
1179 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1180 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1181 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1182 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1183 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1184 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1185 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1186 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1187 static void note_procstat_files(void *, struct sbuf *, size_t *);
1188 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1189 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1190 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1191 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1192 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1193 
1194 #ifdef GZIO
1195 extern int compress_user_cores_gzlevel;
1196 
1197 /*
1198  * Write out a core segment to the compression stream.
1199  */
1200 static int
1201 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1202 {
1203 	u_int chunk_len;
1204 	int error;
1205 
1206 	while (len > 0) {
1207 		chunk_len = MIN(len, CORE_BUF_SIZE);
1208 		copyin(base, buf, chunk_len);
1209 		error = gzio_write(p->gzs, buf, chunk_len);
1210 		if (error != 0)
1211 			break;
1212 		base += chunk_len;
1213 		len -= chunk_len;
1214 	}
1215 	return (error);
1216 }
1217 
1218 static int
1219 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1220 {
1221 
1222 	return (core_write((struct coredump_params *)arg, base, len, offset,
1223 	    UIO_SYSSPACE));
1224 }
1225 #endif /* GZIO */
1226 
1227 static int
1228 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1229     enum uio_seg seg)
1230 {
1231 
1232 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1233 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1234 	    p->active_cred, p->file_cred, NULL, p->td));
1235 }
1236 
1237 static int
1238 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1239     void *tmpbuf)
1240 {
1241 
1242 #ifdef GZIO
1243 	if (p->gzs != NULL)
1244 		return (compress_chunk(p, base, tmpbuf, len));
1245 #endif
1246 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1247 }
1248 
1249 /*
1250  * Drain into a core file.
1251  */
1252 static int
1253 sbuf_drain_core_output(void *arg, const char *data, int len)
1254 {
1255 	struct coredump_params *p;
1256 	int error, locked;
1257 
1258 	p = (struct coredump_params *)arg;
1259 
1260 	/*
1261 	 * Some kern_proc out routines that print to this sbuf may
1262 	 * call us with the process lock held. Draining with the
1263 	 * non-sleepable lock held is unsafe. The lock is needed for
1264 	 * those routines when dumping a live process. In our case we
1265 	 * can safely release the lock before draining and acquire
1266 	 * again after.
1267 	 */
1268 	locked = PROC_LOCKED(p->td->td_proc);
1269 	if (locked)
1270 		PROC_UNLOCK(p->td->td_proc);
1271 #ifdef GZIO
1272 	if (p->gzs != NULL)
1273 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1274 	else
1275 #endif
1276 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1277 		    UIO_SYSSPACE);
1278 	if (locked)
1279 		PROC_LOCK(p->td->td_proc);
1280 	if (error != 0)
1281 		return (-error);
1282 	p->offset += len;
1283 	return (len);
1284 }
1285 
1286 /*
1287  * Drain into a counter.
1288  */
1289 static int
1290 sbuf_drain_count(void *arg, const char *data __unused, int len)
1291 {
1292 	size_t *sizep;
1293 
1294 	sizep = (size_t *)arg;
1295 	*sizep += len;
1296 	return (len);
1297 }
1298 
1299 int
1300 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1301 {
1302 	struct ucred *cred = td->td_ucred;
1303 	int error = 0;
1304 	struct sseg_closure seginfo;
1305 	struct note_info_list notelst;
1306 	struct coredump_params params;
1307 	struct note_info *ninfo;
1308 	void *hdr, *tmpbuf;
1309 	size_t hdrsize, notesz, coresize;
1310 #ifdef GZIO
1311 	boolean_t compress;
1312 
1313 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1314 #endif
1315 	hdr = NULL;
1316 	tmpbuf = NULL;
1317 	TAILQ_INIT(&notelst);
1318 
1319 	/* Size the program segments. */
1320 	seginfo.count = 0;
1321 	seginfo.size = 0;
1322 	each_writable_segment(td, cb_size_segment, &seginfo);
1323 
1324 	/*
1325 	 * Collect info about the core file header area.
1326 	 */
1327 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1328 	__elfN(prepare_notes)(td, &notelst, &notesz);
1329 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1330 
1331 	/* Set up core dump parameters. */
1332 	params.offset = 0;
1333 	params.active_cred = cred;
1334 	params.file_cred = NOCRED;
1335 	params.td = td;
1336 	params.vp = vp;
1337 	params.gzs = NULL;
1338 
1339 #ifdef RACCT
1340 	if (racct_enable) {
1341 		PROC_LOCK(td->td_proc);
1342 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1343 		PROC_UNLOCK(td->td_proc);
1344 		if (error != 0) {
1345 			error = EFAULT;
1346 			goto done;
1347 		}
1348 	}
1349 #endif
1350 	if (coresize >= limit) {
1351 		error = EFAULT;
1352 		goto done;
1353 	}
1354 
1355 #ifdef GZIO
1356 	/* Create a compression stream if necessary. */
1357 	if (compress) {
1358 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1359 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1360 		if (params.gzs == NULL) {
1361 			error = EFAULT;
1362 			goto done;
1363 		}
1364 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1365         }
1366 #endif
1367 
1368 	/*
1369 	 * Allocate memory for building the header, fill it up,
1370 	 * and write it out following the notes.
1371 	 */
1372 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1373 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1374 	    notesz);
1375 
1376 	/* Write the contents of all of the writable segments. */
1377 	if (error == 0) {
1378 		Elf_Phdr *php;
1379 		off_t offset;
1380 		int i;
1381 
1382 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1383 		offset = round_page(hdrsize + notesz);
1384 		for (i = 0; i < seginfo.count; i++) {
1385 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1386 			    php->p_filesz, offset, &params, tmpbuf);
1387 			if (error != 0)
1388 				break;
1389 			offset += php->p_filesz;
1390 			php++;
1391 		}
1392 #ifdef GZIO
1393 		if (error == 0 && compress)
1394 			error = gzio_flush(params.gzs);
1395 #endif
1396 	}
1397 	if (error) {
1398 		log(LOG_WARNING,
1399 		    "Failed to write core file for process %s (error %d)\n",
1400 		    curproc->p_comm, error);
1401 	}
1402 
1403 done:
1404 #ifdef GZIO
1405 	if (compress) {
1406 		free(tmpbuf, M_TEMP);
1407 		if (params.gzs != NULL)
1408 			gzio_fini(params.gzs);
1409 	}
1410 #endif
1411 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1412 		TAILQ_REMOVE(&notelst, ninfo, link);
1413 		free(ninfo, M_TEMP);
1414 	}
1415 	if (hdr != NULL)
1416 		free(hdr, M_TEMP);
1417 
1418 	return (error);
1419 }
1420 
1421 /*
1422  * A callback for each_writable_segment() to write out the segment's
1423  * program header entry.
1424  */
1425 static void
1426 cb_put_phdr(entry, closure)
1427 	vm_map_entry_t entry;
1428 	void *closure;
1429 {
1430 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1431 	Elf_Phdr *phdr = phc->phdr;
1432 
1433 	phc->offset = round_page(phc->offset);
1434 
1435 	phdr->p_type = PT_LOAD;
1436 	phdr->p_offset = phc->offset;
1437 	phdr->p_vaddr = entry->start;
1438 	phdr->p_paddr = 0;
1439 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1440 	phdr->p_align = PAGE_SIZE;
1441 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1442 
1443 	phc->offset += phdr->p_filesz;
1444 	phc->phdr++;
1445 }
1446 
1447 /*
1448  * A callback for each_writable_segment() to gather information about
1449  * the number of segments and their total size.
1450  */
1451 static void
1452 cb_size_segment(entry, closure)
1453 	vm_map_entry_t entry;
1454 	void *closure;
1455 {
1456 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1457 
1458 	ssc->count++;
1459 	ssc->size += entry->end - entry->start;
1460 }
1461 
1462 /*
1463  * For each writable segment in the process's memory map, call the given
1464  * function with a pointer to the map entry and some arbitrary
1465  * caller-supplied data.
1466  */
1467 static void
1468 each_writable_segment(td, func, closure)
1469 	struct thread *td;
1470 	segment_callback func;
1471 	void *closure;
1472 {
1473 	struct proc *p = td->td_proc;
1474 	vm_map_t map = &p->p_vmspace->vm_map;
1475 	vm_map_entry_t entry;
1476 	vm_object_t backing_object, object;
1477 	boolean_t ignore_entry;
1478 
1479 	vm_map_lock_read(map);
1480 	for (entry = map->header.next; entry != &map->header;
1481 	    entry = entry->next) {
1482 		/*
1483 		 * Don't dump inaccessible mappings, deal with legacy
1484 		 * coredump mode.
1485 		 *
1486 		 * Note that read-only segments related to the elf binary
1487 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1488 		 * need to arbitrarily ignore such segments.
1489 		 */
1490 		if (elf_legacy_coredump) {
1491 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1492 				continue;
1493 		} else {
1494 			if ((entry->protection & VM_PROT_ALL) == 0)
1495 				continue;
1496 		}
1497 
1498 		/*
1499 		 * Dont include memory segment in the coredump if
1500 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1501 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1502 		 * kernel map).
1503 		 */
1504 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1505 			continue;
1506 
1507 		if ((object = entry->object.vm_object) == NULL)
1508 			continue;
1509 
1510 		/* Ignore memory-mapped devices and such things. */
1511 		VM_OBJECT_RLOCK(object);
1512 		while ((backing_object = object->backing_object) != NULL) {
1513 			VM_OBJECT_RLOCK(backing_object);
1514 			VM_OBJECT_RUNLOCK(object);
1515 			object = backing_object;
1516 		}
1517 		ignore_entry = object->type != OBJT_DEFAULT &&
1518 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1519 		    object->type != OBJT_PHYS;
1520 		VM_OBJECT_RUNLOCK(object);
1521 		if (ignore_entry)
1522 			continue;
1523 
1524 		(*func)(entry, closure);
1525 	}
1526 	vm_map_unlock_read(map);
1527 }
1528 
1529 /*
1530  * Write the core file header to the file, including padding up to
1531  * the page boundary.
1532  */
1533 static int
1534 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1535     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1536 {
1537 	struct note_info *ninfo;
1538 	struct sbuf *sb;
1539 	int error;
1540 
1541 	/* Fill in the header. */
1542 	bzero(hdr, hdrsize);
1543 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1544 
1545 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1546 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1547 	sbuf_start_section(sb, NULL);
1548 	sbuf_bcat(sb, hdr, hdrsize);
1549 	TAILQ_FOREACH(ninfo, notelst, link)
1550 	    __elfN(putnote)(ninfo, sb);
1551 	/* Align up to a page boundary for the program segments. */
1552 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1553 	error = sbuf_finish(sb);
1554 	sbuf_delete(sb);
1555 
1556 	return (error);
1557 }
1558 
1559 static void
1560 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1561     size_t *sizep)
1562 {
1563 	struct proc *p;
1564 	struct thread *thr;
1565 	size_t size;
1566 
1567 	p = td->td_proc;
1568 	size = 0;
1569 
1570 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1571 
1572 	/*
1573 	 * To have the debugger select the right thread (LWP) as the initial
1574 	 * thread, we dump the state of the thread passed to us in td first.
1575 	 * This is the thread that causes the core dump and thus likely to
1576 	 * be the right thread one wants to have selected in the debugger.
1577 	 */
1578 	thr = td;
1579 	while (thr != NULL) {
1580 		size += register_note(list, NT_PRSTATUS,
1581 		    __elfN(note_prstatus), thr);
1582 		size += register_note(list, NT_FPREGSET,
1583 		    __elfN(note_fpregset), thr);
1584 		size += register_note(list, NT_THRMISC,
1585 		    __elfN(note_thrmisc), thr);
1586 		size += register_note(list, -1,
1587 		    __elfN(note_threadmd), thr);
1588 
1589 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1590 		    TAILQ_NEXT(thr, td_plist);
1591 		if (thr == td)
1592 			thr = TAILQ_NEXT(thr, td_plist);
1593 	}
1594 
1595 	size += register_note(list, NT_PROCSTAT_PROC,
1596 	    __elfN(note_procstat_proc), p);
1597 	size += register_note(list, NT_PROCSTAT_FILES,
1598 	    note_procstat_files, p);
1599 	size += register_note(list, NT_PROCSTAT_VMMAP,
1600 	    note_procstat_vmmap, p);
1601 	size += register_note(list, NT_PROCSTAT_GROUPS,
1602 	    note_procstat_groups, p);
1603 	size += register_note(list, NT_PROCSTAT_UMASK,
1604 	    note_procstat_umask, p);
1605 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1606 	    note_procstat_rlimit, p);
1607 	size += register_note(list, NT_PROCSTAT_OSREL,
1608 	    note_procstat_osrel, p);
1609 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1610 	    __elfN(note_procstat_psstrings), p);
1611 	size += register_note(list, NT_PROCSTAT_AUXV,
1612 	    __elfN(note_procstat_auxv), p);
1613 
1614 	*sizep = size;
1615 }
1616 
1617 static void
1618 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1619     size_t notesz)
1620 {
1621 	Elf_Ehdr *ehdr;
1622 	Elf_Phdr *phdr;
1623 	struct phdr_closure phc;
1624 
1625 	ehdr = (Elf_Ehdr *)hdr;
1626 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1627 
1628 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1629 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1630 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1631 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1632 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1633 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1634 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1635 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1636 	ehdr->e_ident[EI_ABIVERSION] = 0;
1637 	ehdr->e_ident[EI_PAD] = 0;
1638 	ehdr->e_type = ET_CORE;
1639 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1640 	ehdr->e_machine = ELF_ARCH32;
1641 #else
1642 	ehdr->e_machine = ELF_ARCH;
1643 #endif
1644 	ehdr->e_version = EV_CURRENT;
1645 	ehdr->e_entry = 0;
1646 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1647 	ehdr->e_flags = 0;
1648 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1649 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1650 	ehdr->e_phnum = numsegs + 1;
1651 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1652 	ehdr->e_shnum = 0;
1653 	ehdr->e_shstrndx = SHN_UNDEF;
1654 
1655 	/*
1656 	 * Fill in the program header entries.
1657 	 */
1658 
1659 	/* The note segement. */
1660 	phdr->p_type = PT_NOTE;
1661 	phdr->p_offset = hdrsize;
1662 	phdr->p_vaddr = 0;
1663 	phdr->p_paddr = 0;
1664 	phdr->p_filesz = notesz;
1665 	phdr->p_memsz = 0;
1666 	phdr->p_flags = PF_R;
1667 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1668 	phdr++;
1669 
1670 	/* All the writable segments from the program. */
1671 	phc.phdr = phdr;
1672 	phc.offset = round_page(hdrsize + notesz);
1673 	each_writable_segment(td, cb_put_phdr, &phc);
1674 }
1675 
1676 static size_t
1677 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1678 {
1679 	struct note_info *ninfo;
1680 	size_t size, notesize;
1681 
1682 	size = 0;
1683 	out(arg, NULL, &size);
1684 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1685 	ninfo->type = type;
1686 	ninfo->outfunc = out;
1687 	ninfo->outarg = arg;
1688 	ninfo->outsize = size;
1689 	TAILQ_INSERT_TAIL(list, ninfo, link);
1690 
1691 	if (type == -1)
1692 		return (size);
1693 
1694 	notesize = sizeof(Elf_Note) +		/* note header */
1695 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1696 						/* note name */
1697 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1698 
1699 	return (notesize);
1700 }
1701 
1702 static size_t
1703 append_note_data(const void *src, void *dst, size_t len)
1704 {
1705 	size_t padded_len;
1706 
1707 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1708 	if (dst != NULL) {
1709 		bcopy(src, dst, len);
1710 		bzero((char *)dst + len, padded_len - len);
1711 	}
1712 	return (padded_len);
1713 }
1714 
1715 size_t
1716 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1717 {
1718 	Elf_Note *note;
1719 	char *buf;
1720 	size_t notesize;
1721 
1722 	buf = dst;
1723 	if (buf != NULL) {
1724 		note = (Elf_Note *)buf;
1725 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1726 		note->n_descsz = size;
1727 		note->n_type = type;
1728 		buf += sizeof(*note);
1729 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1730 		    sizeof(FREEBSD_ABI_VENDOR));
1731 		append_note_data(src, buf, size);
1732 		if (descp != NULL)
1733 			*descp = buf;
1734 	}
1735 
1736 	notesize = sizeof(Elf_Note) +		/* note header */
1737 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1738 						/* note name */
1739 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1740 
1741 	return (notesize);
1742 }
1743 
1744 static void
1745 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1746 {
1747 	Elf_Note note;
1748 	ssize_t old_len, sect_len;
1749 	size_t new_len, descsz, i;
1750 
1751 	if (ninfo->type == -1) {
1752 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1753 		return;
1754 	}
1755 
1756 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1757 	note.n_descsz = ninfo->outsize;
1758 	note.n_type = ninfo->type;
1759 
1760 	sbuf_bcat(sb, &note, sizeof(note));
1761 	sbuf_start_section(sb, &old_len);
1762 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1763 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1764 	if (note.n_descsz == 0)
1765 		return;
1766 	sbuf_start_section(sb, &old_len);
1767 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1768 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1769 	if (sect_len < 0)
1770 		return;
1771 
1772 	new_len = (size_t)sect_len;
1773 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1774 	if (new_len < descsz) {
1775 		/*
1776 		 * It is expected that individual note emitters will correctly
1777 		 * predict their expected output size and fill up to that size
1778 		 * themselves, padding in a format-specific way if needed.
1779 		 * However, in case they don't, just do it here with zeros.
1780 		 */
1781 		for (i = 0; i < descsz - new_len; i++)
1782 			sbuf_putc(sb, 0);
1783 	} else if (new_len > descsz) {
1784 		/*
1785 		 * We can't always truncate sb -- we may have drained some
1786 		 * of it already.
1787 		 */
1788 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1789 		    "read it (%zu > %zu).  Since it is longer than "
1790 		    "expected, this coredump's notes are corrupt.  THIS "
1791 		    "IS A BUG in the note_procstat routine for type %u.\n",
1792 		    __func__, (unsigned)note.n_type, new_len, descsz,
1793 		    (unsigned)note.n_type));
1794 	}
1795 }
1796 
1797 /*
1798  * Miscellaneous note out functions.
1799  */
1800 
1801 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1802 #include <compat/freebsd32/freebsd32.h>
1803 
1804 typedef struct prstatus32 elf_prstatus_t;
1805 typedef struct prpsinfo32 elf_prpsinfo_t;
1806 typedef struct fpreg32 elf_prfpregset_t;
1807 typedef struct fpreg32 elf_fpregset_t;
1808 typedef struct reg32 elf_gregset_t;
1809 typedef struct thrmisc32 elf_thrmisc_t;
1810 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1811 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1812 typedef uint32_t elf_ps_strings_t;
1813 #else
1814 typedef prstatus_t elf_prstatus_t;
1815 typedef prpsinfo_t elf_prpsinfo_t;
1816 typedef prfpregset_t elf_prfpregset_t;
1817 typedef prfpregset_t elf_fpregset_t;
1818 typedef gregset_t elf_gregset_t;
1819 typedef thrmisc_t elf_thrmisc_t;
1820 #define ELF_KERN_PROC_MASK	0
1821 typedef struct kinfo_proc elf_kinfo_proc_t;
1822 typedef vm_offset_t elf_ps_strings_t;
1823 #endif
1824 
1825 static void
1826 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1827 {
1828 	struct proc *p;
1829 	elf_prpsinfo_t *psinfo;
1830 
1831 	p = (struct proc *)arg;
1832 	if (sb != NULL) {
1833 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1834 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1835 		psinfo->pr_version = PRPSINFO_VERSION;
1836 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1837 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1838 		/*
1839 		 * XXX - We don't fill in the command line arguments properly
1840 		 * yet.
1841 		 */
1842 		strlcpy(psinfo->pr_psargs, p->p_comm,
1843 		    sizeof(psinfo->pr_psargs));
1844 
1845 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1846 		free(psinfo, M_TEMP);
1847 	}
1848 	*sizep = sizeof(*psinfo);
1849 }
1850 
1851 static void
1852 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1853 {
1854 	struct thread *td;
1855 	elf_prstatus_t *status;
1856 
1857 	td = (struct thread *)arg;
1858 	if (sb != NULL) {
1859 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1860 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1861 		status->pr_version = PRSTATUS_VERSION;
1862 		status->pr_statussz = sizeof(elf_prstatus_t);
1863 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1864 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1865 		status->pr_osreldate = osreldate;
1866 		status->pr_cursig = td->td_proc->p_sig;
1867 		status->pr_pid = td->td_tid;
1868 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1869 		fill_regs32(td, &status->pr_reg);
1870 #else
1871 		fill_regs(td, &status->pr_reg);
1872 #endif
1873 		sbuf_bcat(sb, status, sizeof(*status));
1874 		free(status, M_TEMP);
1875 	}
1876 	*sizep = sizeof(*status);
1877 }
1878 
1879 static void
1880 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1881 {
1882 	struct thread *td;
1883 	elf_prfpregset_t *fpregset;
1884 
1885 	td = (struct thread *)arg;
1886 	if (sb != NULL) {
1887 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1888 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1889 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1890 		fill_fpregs32(td, fpregset);
1891 #else
1892 		fill_fpregs(td, fpregset);
1893 #endif
1894 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1895 		free(fpregset, M_TEMP);
1896 	}
1897 	*sizep = sizeof(*fpregset);
1898 }
1899 
1900 static void
1901 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1902 {
1903 	struct thread *td;
1904 	elf_thrmisc_t thrmisc;
1905 
1906 	td = (struct thread *)arg;
1907 	if (sb != NULL) {
1908 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1909 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1910 		strcpy(thrmisc.pr_tname, td->td_name);
1911 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1912 	}
1913 	*sizep = sizeof(thrmisc);
1914 }
1915 
1916 /*
1917  * Allow for MD specific notes, as well as any MD
1918  * specific preparations for writing MI notes.
1919  */
1920 static void
1921 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1922 {
1923 	struct thread *td;
1924 	void *buf;
1925 	size_t size;
1926 
1927 	td = (struct thread *)arg;
1928 	size = *sizep;
1929 	if (size != 0 && sb != NULL)
1930 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1931 	else
1932 		buf = NULL;
1933 	size = 0;
1934 	__elfN(dump_thread)(td, buf, &size);
1935 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1936 	if (size != 0 && sb != NULL)
1937 		sbuf_bcat(sb, buf, size);
1938 	free(buf, M_TEMP);
1939 	*sizep = size;
1940 }
1941 
1942 #ifdef KINFO_PROC_SIZE
1943 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1944 #endif
1945 
1946 static void
1947 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1948 {
1949 	struct proc *p;
1950 	size_t size;
1951 	int structsize;
1952 
1953 	p = (struct proc *)arg;
1954 	size = sizeof(structsize) + p->p_numthreads *
1955 	    sizeof(elf_kinfo_proc_t);
1956 
1957 	if (sb != NULL) {
1958 		KASSERT(*sizep == size, ("invalid size"));
1959 		structsize = sizeof(elf_kinfo_proc_t);
1960 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1961 		sx_slock(&proctree_lock);
1962 		PROC_LOCK(p);
1963 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1964 		sx_sunlock(&proctree_lock);
1965 	}
1966 	*sizep = size;
1967 }
1968 
1969 #ifdef KINFO_FILE_SIZE
1970 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1971 #endif
1972 
1973 static void
1974 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1975 {
1976 	struct proc *p;
1977 	size_t size, sect_sz, i;
1978 	ssize_t start_len, sect_len;
1979 	int structsize, filedesc_flags;
1980 
1981 	if (coredump_pack_fileinfo)
1982 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1983 	else
1984 		filedesc_flags = 0;
1985 
1986 	p = (struct proc *)arg;
1987 	structsize = sizeof(struct kinfo_file);
1988 	if (sb == NULL) {
1989 		size = 0;
1990 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1991 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1992 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1993 		PROC_LOCK(p);
1994 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1995 		sbuf_finish(sb);
1996 		sbuf_delete(sb);
1997 		*sizep = size;
1998 	} else {
1999 		sbuf_start_section(sb, &start_len);
2000 
2001 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2002 		PROC_LOCK(p);
2003 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2004 		    filedesc_flags);
2005 
2006 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2007 		if (sect_len < 0)
2008 			return;
2009 		sect_sz = sect_len;
2010 
2011 		KASSERT(sect_sz <= *sizep,
2012 		    ("kern_proc_filedesc_out did not respect maxlen; "
2013 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2014 		     sect_sz - sizeof(structsize)));
2015 
2016 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2017 			sbuf_putc(sb, 0);
2018 	}
2019 }
2020 
2021 #ifdef KINFO_VMENTRY_SIZE
2022 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2023 #endif
2024 
2025 static void
2026 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2027 {
2028 	struct proc *p;
2029 	size_t size;
2030 	int structsize, vmmap_flags;
2031 
2032 	if (coredump_pack_vmmapinfo)
2033 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2034 	else
2035 		vmmap_flags = 0;
2036 
2037 	p = (struct proc *)arg;
2038 	structsize = sizeof(struct kinfo_vmentry);
2039 	if (sb == NULL) {
2040 		size = 0;
2041 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2042 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2043 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2044 		PROC_LOCK(p);
2045 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2046 		sbuf_finish(sb);
2047 		sbuf_delete(sb);
2048 		*sizep = size;
2049 	} else {
2050 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2051 		PROC_LOCK(p);
2052 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2053 		    vmmap_flags);
2054 	}
2055 }
2056 
2057 static void
2058 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2059 {
2060 	struct proc *p;
2061 	size_t size;
2062 	int structsize;
2063 
2064 	p = (struct proc *)arg;
2065 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2066 	if (sb != NULL) {
2067 		KASSERT(*sizep == size, ("invalid size"));
2068 		structsize = sizeof(gid_t);
2069 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2070 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2071 		    sizeof(gid_t));
2072 	}
2073 	*sizep = size;
2074 }
2075 
2076 static void
2077 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2078 {
2079 	struct proc *p;
2080 	size_t size;
2081 	int structsize;
2082 
2083 	p = (struct proc *)arg;
2084 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2085 	if (sb != NULL) {
2086 		KASSERT(*sizep == size, ("invalid size"));
2087 		structsize = sizeof(p->p_fd->fd_cmask);
2088 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2089 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2090 	}
2091 	*sizep = size;
2092 }
2093 
2094 static void
2095 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2096 {
2097 	struct proc *p;
2098 	struct rlimit rlim[RLIM_NLIMITS];
2099 	size_t size;
2100 	int structsize, i;
2101 
2102 	p = (struct proc *)arg;
2103 	size = sizeof(structsize) + sizeof(rlim);
2104 	if (sb != NULL) {
2105 		KASSERT(*sizep == size, ("invalid size"));
2106 		structsize = sizeof(rlim);
2107 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2108 		PROC_LOCK(p);
2109 		for (i = 0; i < RLIM_NLIMITS; i++)
2110 			lim_rlimit_proc(p, i, &rlim[i]);
2111 		PROC_UNLOCK(p);
2112 		sbuf_bcat(sb, rlim, sizeof(rlim));
2113 	}
2114 	*sizep = size;
2115 }
2116 
2117 static void
2118 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2119 {
2120 	struct proc *p;
2121 	size_t size;
2122 	int structsize;
2123 
2124 	p = (struct proc *)arg;
2125 	size = sizeof(structsize) + sizeof(p->p_osrel);
2126 	if (sb != NULL) {
2127 		KASSERT(*sizep == size, ("invalid size"));
2128 		structsize = sizeof(p->p_osrel);
2129 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2130 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2131 	}
2132 	*sizep = size;
2133 }
2134 
2135 static void
2136 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2137 {
2138 	struct proc *p;
2139 	elf_ps_strings_t ps_strings;
2140 	size_t size;
2141 	int structsize;
2142 
2143 	p = (struct proc *)arg;
2144 	size = sizeof(structsize) + sizeof(ps_strings);
2145 	if (sb != NULL) {
2146 		KASSERT(*sizep == size, ("invalid size"));
2147 		structsize = sizeof(ps_strings);
2148 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2149 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2150 #else
2151 		ps_strings = p->p_sysent->sv_psstrings;
2152 #endif
2153 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2154 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2155 	}
2156 	*sizep = size;
2157 }
2158 
2159 static void
2160 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2161 {
2162 	struct proc *p;
2163 	size_t size;
2164 	int structsize;
2165 
2166 	p = (struct proc *)arg;
2167 	if (sb == NULL) {
2168 		size = 0;
2169 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2170 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2171 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2172 		PHOLD(p);
2173 		proc_getauxv(curthread, p, sb);
2174 		PRELE(p);
2175 		sbuf_finish(sb);
2176 		sbuf_delete(sb);
2177 		*sizep = size;
2178 	} else {
2179 		structsize = sizeof(Elf_Auxinfo);
2180 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2181 		PHOLD(p);
2182 		proc_getauxv(curthread, p, sb);
2183 		PRELE(p);
2184 	}
2185 }
2186 
2187 static boolean_t
2188 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2189     int32_t *osrel, const Elf_Phdr *pnote)
2190 {
2191 	const Elf_Note *note, *note0, *note_end;
2192 	const char *note_name;
2193 	char *buf;
2194 	int i, error;
2195 	boolean_t res;
2196 
2197 	/* We need some limit, might as well use PAGE_SIZE. */
2198 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2199 		return (FALSE);
2200 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2201 	if (pnote->p_offset > PAGE_SIZE ||
2202 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2203 		VOP_UNLOCK(imgp->vp, 0);
2204 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2205 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2206 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2207 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2208 		    curthread->td_ucred, NOCRED, NULL, curthread);
2209 		if (error != 0) {
2210 			uprintf("i/o error PT_NOTE\n");
2211 			res = FALSE;
2212 			goto ret;
2213 		}
2214 		note = note0 = (const Elf_Note *)buf;
2215 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2216 	} else {
2217 		note = note0 = (const Elf_Note *)(imgp->image_header +
2218 		    pnote->p_offset);
2219 		note_end = (const Elf_Note *)(imgp->image_header +
2220 		    pnote->p_offset + pnote->p_filesz);
2221 		buf = NULL;
2222 	}
2223 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2224 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2225 		    (const char *)note < sizeof(Elf_Note)) {
2226 			res = FALSE;
2227 			goto ret;
2228 		}
2229 		if (note->n_namesz != checknote->hdr.n_namesz ||
2230 		    note->n_descsz != checknote->hdr.n_descsz ||
2231 		    note->n_type != checknote->hdr.n_type)
2232 			goto nextnote;
2233 		note_name = (const char *)(note + 1);
2234 		if (note_name + checknote->hdr.n_namesz >=
2235 		    (const char *)note_end || strncmp(checknote->vendor,
2236 		    note_name, checknote->hdr.n_namesz) != 0)
2237 			goto nextnote;
2238 
2239 		/*
2240 		 * Fetch the osreldate for binary
2241 		 * from the ELF OSABI-note if necessary.
2242 		 */
2243 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2244 		    checknote->trans_osrel != NULL) {
2245 			res = checknote->trans_osrel(note, osrel);
2246 			goto ret;
2247 		}
2248 		res = TRUE;
2249 		goto ret;
2250 nextnote:
2251 		note = (const Elf_Note *)((const char *)(note + 1) +
2252 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2253 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2254 	}
2255 	res = FALSE;
2256 ret:
2257 	free(buf, M_TEMP);
2258 	return (res);
2259 }
2260 
2261 /*
2262  * Try to find the appropriate ABI-note section for checknote,
2263  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2264  * first page of the image is searched, the same as for headers.
2265  */
2266 static boolean_t
2267 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2268     int32_t *osrel)
2269 {
2270 	const Elf_Phdr *phdr;
2271 	const Elf_Ehdr *hdr;
2272 	int i;
2273 
2274 	hdr = (const Elf_Ehdr *)imgp->image_header;
2275 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2276 
2277 	for (i = 0; i < hdr->e_phnum; i++) {
2278 		if (phdr[i].p_type == PT_NOTE &&
2279 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2280 			return (TRUE);
2281 	}
2282 	return (FALSE);
2283 
2284 }
2285 
2286 /*
2287  * Tell kern_execve.c about it, with a little help from the linker.
2288  */
2289 static struct execsw __elfN(execsw) = {
2290 	__CONCAT(exec_, __elfN(imgact)),
2291 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2292 };
2293 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2294 
2295 static vm_prot_t
2296 __elfN(trans_prot)(Elf_Word flags)
2297 {
2298 	vm_prot_t prot;
2299 
2300 	prot = 0;
2301 	if (flags & PF_X)
2302 		prot |= VM_PROT_EXECUTE;
2303 	if (flags & PF_W)
2304 		prot |= VM_PROT_WRITE;
2305 	if (flags & PF_R)
2306 		prot |= VM_PROT_READ;
2307 #if __ELF_WORD_SIZE == 32
2308 #if defined(__amd64__)
2309 	if (i386_read_exec && (flags & PF_R))
2310 		prot |= VM_PROT_EXECUTE;
2311 #endif
2312 #endif
2313 	return (prot);
2314 }
2315 
2316 static Elf_Word
2317 __elfN(untrans_prot)(vm_prot_t prot)
2318 {
2319 	Elf_Word flags;
2320 
2321 	flags = 0;
2322 	if (prot & VM_PROT_EXECUTE)
2323 		flags |= PF_X;
2324 	if (prot & VM_PROT_READ)
2325 		flags |= PF_R;
2326 	if (prot & VM_PROT_WRITE)
2327 		flags |= PF_W;
2328 	return (flags);
2329 }
2330