xref: /freebsd/sys/kern/imgact_elf.c (revision 01b792f1f535c12a1a14000cf3360ef6c36cee2d)
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_core.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/imgact.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mman.h>
50 #include <sys/namei.h>
51 #include <sys/pioctl.h>
52 #include <sys/proc.h>
53 #include <sys/procfs.h>
54 #include <sys/racct.h>
55 #include <sys/resourcevar.h>
56 #include <sys/rwlock.h>
57 #include <sys/sbuf.h>
58 #include <sys/sf_buf.h>
59 #include <sys/smp.h>
60 #include <sys/systm.h>
61 #include <sys/signalvar.h>
62 #include <sys/stat.h>
63 #include <sys/sx.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/syslog.h>
69 #include <sys/eventhandler.h>
70 #include <sys/user.h>
71 
72 #include <net/zlib.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
81 
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
84 
85 #define ELF_NOTE_ROUNDSIZE	4
86 #define OLD_EI_BRAND	8
87 
88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
90     const char *interp, int interp_name_len, int32_t *osrel);
91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
92     u_long *entry, size_t pagesize);
93 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
94     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
95     size_t pagesize);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
106     "");
107 
108 #ifdef COMPRESS_USER_CORES
109 static int compress_core(gzFile, char *, char *, unsigned int,
110     struct thread * td);
111 #endif
112 #define CORE_BUF_SIZE	(16 * 1024)
113 
114 int __elfN(fallback_brand) = -1;
115 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
116     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
117     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
118 
119 static int elf_legacy_coredump = 0;
120 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
121     &elf_legacy_coredump, 0, "");
122 
123 int __elfN(nxstack) =
124 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
125 	1;
126 #else
127 	0;
128 #endif
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 
133 #if __ELF_WORD_SIZE == 32
134 #if defined(__amd64__)
135 int i386_read_exec = 0;
136 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
137     "enable execution from readable segments");
138 #endif
139 #endif
140 
141 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
142 
143 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
144 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
145 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
146 
147 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
148 
149 Elf_Brandnote __elfN(freebsd_brandnote) = {
150 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
151 	.hdr.n_descsz	= sizeof(int32_t),
152 	.hdr.n_type	= 1,
153 	.vendor		= FREEBSD_ABI_VENDOR,
154 	.flags		= BN_TRANSLATE_OSREL,
155 	.trans_osrel	= __elfN(freebsd_trans_osrel)
156 };
157 
158 static boolean_t
159 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
160 {
161 	uintptr_t p;
162 
163 	p = (uintptr_t)(note + 1);
164 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
165 	*osrel = *(const int32_t *)(p);
166 
167 	return (TRUE);
168 }
169 
170 static const char GNU_ABI_VENDOR[] = "GNU";
171 static int GNU_KFREEBSD_ABI_DESC = 3;
172 
173 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
174 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
175 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
176 	.hdr.n_type	= 1,
177 	.vendor		= GNU_ABI_VENDOR,
178 	.flags		= BN_TRANSLATE_OSREL,
179 	.trans_osrel	= kfreebsd_trans_osrel
180 };
181 
182 static boolean_t
183 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
184 {
185 	const Elf32_Word *desc;
186 	uintptr_t p;
187 
188 	p = (uintptr_t)(note + 1);
189 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
190 
191 	desc = (const Elf32_Word *)p;
192 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
193 		return (FALSE);
194 
195 	/*
196 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
197 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
198 	 */
199 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
200 
201 	return (TRUE);
202 }
203 
204 int
205 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
206 {
207 	int i;
208 
209 	for (i = 0; i < MAX_BRANDS; i++) {
210 		if (elf_brand_list[i] == NULL) {
211 			elf_brand_list[i] = entry;
212 			break;
213 		}
214 	}
215 	if (i == MAX_BRANDS) {
216 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
217 			__func__, entry);
218 		return (-1);
219 	}
220 	return (0);
221 }
222 
223 int
224 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
225 {
226 	int i;
227 
228 	for (i = 0; i < MAX_BRANDS; i++) {
229 		if (elf_brand_list[i] == entry) {
230 			elf_brand_list[i] = NULL;
231 			break;
232 		}
233 	}
234 	if (i == MAX_BRANDS)
235 		return (-1);
236 	return (0);
237 }
238 
239 int
240 __elfN(brand_inuse)(Elf_Brandinfo *entry)
241 {
242 	struct proc *p;
243 	int rval = FALSE;
244 
245 	sx_slock(&allproc_lock);
246 	FOREACH_PROC_IN_SYSTEM(p) {
247 		if (p->p_sysent == entry->sysvec) {
248 			rval = TRUE;
249 			break;
250 		}
251 	}
252 	sx_sunlock(&allproc_lock);
253 
254 	return (rval);
255 }
256 
257 static Elf_Brandinfo *
258 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
259     int interp_name_len, int32_t *osrel)
260 {
261 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
262 	Elf_Brandinfo *bi;
263 	boolean_t ret;
264 	int i;
265 
266 	/*
267 	 * We support four types of branding -- (1) the ELF EI_OSABI field
268 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
269 	 * branding w/in the ELF header, (3) path of the `interp_path'
270 	 * field, and (4) the ".note.ABI-tag" ELF section.
271 	 */
272 
273 	/* Look for an ".note.ABI-tag" ELF section */
274 	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 			if (ret)
282 				return (bi);
283 		}
284 	}
285 
286 	/* If the executable has a brand, search for it in the brand list. */
287 	for (i = 0; i < MAX_BRANDS; i++) {
288 		bi = elf_brand_list[i];
289 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
290 			continue;
291 		if (hdr->e_machine == bi->machine &&
292 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
293 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
294 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
295 			return (bi);
296 	}
297 
298 	/* No known brand, see if the header is recognized by any brand */
299 	for (i = 0; i < MAX_BRANDS; i++) {
300 		bi = elf_brand_list[i];
301 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
302 		    bi->header_supported == NULL)
303 			continue;
304 		if (hdr->e_machine == bi->machine) {
305 			ret = bi->header_supported(imgp);
306 			if (ret)
307 				return (bi);
308 		}
309 	}
310 
311 	/* Lacking a known brand, search for a recognized interpreter. */
312 	if (interp != NULL) {
313 		for (i = 0; i < MAX_BRANDS; i++) {
314 			bi = elf_brand_list[i];
315 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
316 				continue;
317 			if (hdr->e_machine == bi->machine &&
318 			    /* ELF image p_filesz includes terminating zero */
319 			    strlen(bi->interp_path) + 1 == interp_name_len &&
320 			    strncmp(interp, bi->interp_path, interp_name_len)
321 			    == 0)
322 				return (bi);
323 		}
324 	}
325 
326 	/* Lacking a recognized interpreter, try the default brand */
327 	for (i = 0; i < MAX_BRANDS; i++) {
328 		bi = elf_brand_list[i];
329 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
330 			continue;
331 		if (hdr->e_machine == bi->machine &&
332 		    __elfN(fallback_brand) == bi->brand)
333 			return (bi);
334 	}
335 	return (NULL);
336 }
337 
338 static int
339 __elfN(check_header)(const Elf_Ehdr *hdr)
340 {
341 	Elf_Brandinfo *bi;
342 	int i;
343 
344 	if (!IS_ELF(*hdr) ||
345 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
346 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
347 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
348 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
349 	    hdr->e_version != ELF_TARG_VER)
350 		return (ENOEXEC);
351 
352 	/*
353 	 * Make sure we have at least one brand for this machine.
354 	 */
355 
356 	for (i = 0; i < MAX_BRANDS; i++) {
357 		bi = elf_brand_list[i];
358 		if (bi != NULL && bi->machine == hdr->e_machine)
359 			break;
360 	}
361 	if (i == MAX_BRANDS)
362 		return (ENOEXEC);
363 
364 	return (0);
365 }
366 
367 static int
368 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
369     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
370 {
371 	struct sf_buf *sf;
372 	int error;
373 	vm_offset_t off;
374 
375 	/*
376 	 * Create the page if it doesn't exist yet. Ignore errors.
377 	 */
378 	vm_map_lock(map);
379 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
380 	    VM_PROT_ALL, VM_PROT_ALL, 0);
381 	vm_map_unlock(map);
382 
383 	/*
384 	 * Find the page from the underlying object.
385 	 */
386 	if (object) {
387 		sf = vm_imgact_map_page(object, offset);
388 		if (sf == NULL)
389 			return (KERN_FAILURE);
390 		off = offset - trunc_page(offset);
391 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
392 		    end - start);
393 		vm_imgact_unmap_page(sf);
394 		if (error) {
395 			return (KERN_FAILURE);
396 		}
397 	}
398 
399 	return (KERN_SUCCESS);
400 }
401 
402 static int
403 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
404     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
405 {
406 	struct sf_buf *sf;
407 	vm_offset_t off;
408 	vm_size_t sz;
409 	int error, rv;
410 
411 	if (start != trunc_page(start)) {
412 		rv = __elfN(map_partial)(map, object, offset, start,
413 		    round_page(start), prot);
414 		if (rv)
415 			return (rv);
416 		offset += round_page(start) - start;
417 		start = round_page(start);
418 	}
419 	if (end != round_page(end)) {
420 		rv = __elfN(map_partial)(map, object, offset +
421 		    trunc_page(end) - start, trunc_page(end), end, prot);
422 		if (rv)
423 			return (rv);
424 		end = trunc_page(end);
425 	}
426 	if (end > start) {
427 		if (offset & PAGE_MASK) {
428 			/*
429 			 * The mapping is not page aligned. This means we have
430 			 * to copy the data. Sigh.
431 			 */
432 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
433 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
434 			    0);
435 			if (rv)
436 				return (rv);
437 			if (object == NULL)
438 				return (KERN_SUCCESS);
439 			for (; start < end; start += sz) {
440 				sf = vm_imgact_map_page(object, offset);
441 				if (sf == NULL)
442 					return (KERN_FAILURE);
443 				off = offset - trunc_page(offset);
444 				sz = end - start;
445 				if (sz > PAGE_SIZE - off)
446 					sz = PAGE_SIZE - off;
447 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
448 				    (caddr_t)start, sz);
449 				vm_imgact_unmap_page(sf);
450 				if (error) {
451 					return (KERN_FAILURE);
452 				}
453 				offset += sz;
454 			}
455 			rv = KERN_SUCCESS;
456 		} else {
457 			vm_object_reference(object);
458 			vm_map_lock(map);
459 			rv = vm_map_insert(map, object, offset, start, end,
460 			    prot, VM_PROT_ALL, cow);
461 			vm_map_unlock(map);
462 			if (rv != KERN_SUCCESS)
463 				vm_object_deallocate(object);
464 		}
465 		return (rv);
466 	} else {
467 		return (KERN_SUCCESS);
468 	}
469 }
470 
471 static int
472 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
473     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
474     size_t pagesize)
475 {
476 	struct sf_buf *sf;
477 	size_t map_len;
478 	vm_map_t map;
479 	vm_object_t object;
480 	vm_offset_t map_addr;
481 	int error, rv, cow;
482 	size_t copy_len;
483 	vm_offset_t file_addr;
484 
485 	/*
486 	 * It's necessary to fail if the filsz + offset taken from the
487 	 * header is greater than the actual file pager object's size.
488 	 * If we were to allow this, then the vm_map_find() below would
489 	 * walk right off the end of the file object and into the ether.
490 	 *
491 	 * While I'm here, might as well check for something else that
492 	 * is invalid: filsz cannot be greater than memsz.
493 	 */
494 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
495 		uprintf("elf_load_section: truncated ELF file\n");
496 		return (ENOEXEC);
497 	}
498 
499 	object = imgp->object;
500 	map = &imgp->proc->p_vmspace->vm_map;
501 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
502 	file_addr = trunc_page_ps(offset, pagesize);
503 
504 	/*
505 	 * We have two choices.  We can either clear the data in the last page
506 	 * of an oversized mapping, or we can start the anon mapping a page
507 	 * early and copy the initialized data into that first page.  We
508 	 * choose the second..
509 	 */
510 	if (memsz > filsz)
511 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
512 	else
513 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
514 
515 	if (map_len != 0) {
516 		/* cow flags: don't dump readonly sections in core */
517 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
518 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
519 
520 		rv = __elfN(map_insert)(map,
521 				      object,
522 				      file_addr,	/* file offset */
523 				      map_addr,		/* virtual start */
524 				      map_addr + map_len,/* virtual end */
525 				      prot,
526 				      cow);
527 		if (rv != KERN_SUCCESS)
528 			return (EINVAL);
529 
530 		/* we can stop now if we've covered it all */
531 		if (memsz == filsz) {
532 			return (0);
533 		}
534 	}
535 
536 
537 	/*
538 	 * We have to get the remaining bit of the file into the first part
539 	 * of the oversized map segment.  This is normally because the .data
540 	 * segment in the file is extended to provide bss.  It's a neat idea
541 	 * to try and save a page, but it's a pain in the behind to implement.
542 	 */
543 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
544 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
545 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
546 	    map_addr;
547 
548 	/* This had damn well better be true! */
549 	if (map_len != 0) {
550 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
551 		    map_len, VM_PROT_ALL, 0);
552 		if (rv != KERN_SUCCESS) {
553 			return (EINVAL);
554 		}
555 	}
556 
557 	if (copy_len != 0) {
558 		vm_offset_t off;
559 
560 		sf = vm_imgact_map_page(object, offset + filsz);
561 		if (sf == NULL)
562 			return (EIO);
563 
564 		/* send the page fragment to user space */
565 		off = trunc_page_ps(offset + filsz, pagesize) -
566 		    trunc_page(offset + filsz);
567 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
568 		    (caddr_t)map_addr, copy_len);
569 		vm_imgact_unmap_page(sf);
570 		if (error) {
571 			return (error);
572 		}
573 	}
574 
575 	/*
576 	 * set it to the specified protection.
577 	 * XXX had better undo the damage from pasting over the cracks here!
578 	 */
579 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
580 	    map_len), prot, FALSE);
581 
582 	return (0);
583 }
584 
585 /*
586  * Load the file "file" into memory.  It may be either a shared object
587  * or an executable.
588  *
589  * The "addr" reference parameter is in/out.  On entry, it specifies
590  * the address where a shared object should be loaded.  If the file is
591  * an executable, this value is ignored.  On exit, "addr" specifies
592  * where the file was actually loaded.
593  *
594  * The "entry" reference parameter is out only.  On exit, it specifies
595  * the entry point for the loaded file.
596  */
597 static int
598 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
599 	u_long *entry, size_t pagesize)
600 {
601 	struct {
602 		struct nameidata nd;
603 		struct vattr attr;
604 		struct image_params image_params;
605 	} *tempdata;
606 	const Elf_Ehdr *hdr = NULL;
607 	const Elf_Phdr *phdr = NULL;
608 	struct nameidata *nd;
609 	struct vattr *attr;
610 	struct image_params *imgp;
611 	vm_prot_t prot;
612 	u_long rbase;
613 	u_long base_addr = 0;
614 	int error, i, numsegs;
615 
616 #ifdef CAPABILITY_MODE
617 	/*
618 	 * XXXJA: This check can go away once we are sufficiently confident
619 	 * that the checks in namei() are correct.
620 	 */
621 	if (IN_CAPABILITY_MODE(curthread))
622 		return (ECAPMODE);
623 #endif
624 
625 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
626 	nd = &tempdata->nd;
627 	attr = &tempdata->attr;
628 	imgp = &tempdata->image_params;
629 
630 	/*
631 	 * Initialize part of the common data
632 	 */
633 	imgp->proc = p;
634 	imgp->attr = attr;
635 	imgp->firstpage = NULL;
636 	imgp->image_header = NULL;
637 	imgp->object = NULL;
638 	imgp->execlabel = NULL;
639 
640 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
641 	if ((error = namei(nd)) != 0) {
642 		nd->ni_vp = NULL;
643 		goto fail;
644 	}
645 	NDFREE(nd, NDF_ONLY_PNBUF);
646 	imgp->vp = nd->ni_vp;
647 
648 	/*
649 	 * Check permissions, modes, uid, etc on the file, and "open" it.
650 	 */
651 	error = exec_check_permissions(imgp);
652 	if (error)
653 		goto fail;
654 
655 	error = exec_map_first_page(imgp);
656 	if (error)
657 		goto fail;
658 
659 	/*
660 	 * Also make certain that the interpreter stays the same, so set
661 	 * its VV_TEXT flag, too.
662 	 */
663 	VOP_SET_TEXT(nd->ni_vp);
664 
665 	imgp->object = nd->ni_vp->v_object;
666 
667 	hdr = (const Elf_Ehdr *)imgp->image_header;
668 	if ((error = __elfN(check_header)(hdr)) != 0)
669 		goto fail;
670 	if (hdr->e_type == ET_DYN)
671 		rbase = *addr;
672 	else if (hdr->e_type == ET_EXEC)
673 		rbase = 0;
674 	else {
675 		error = ENOEXEC;
676 		goto fail;
677 	}
678 
679 	/* Only support headers that fit within first page for now      */
680 	if ((hdr->e_phoff > PAGE_SIZE) ||
681 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
682 		error = ENOEXEC;
683 		goto fail;
684 	}
685 
686 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
687 	if (!aligned(phdr, Elf_Addr)) {
688 		error = ENOEXEC;
689 		goto fail;
690 	}
691 
692 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
693 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
694 			/* Loadable segment */
695 			prot = __elfN(trans_prot)(phdr[i].p_flags);
696 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
697 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
698 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
699 			if (error != 0)
700 				goto fail;
701 			/*
702 			 * Establish the base address if this is the
703 			 * first segment.
704 			 */
705 			if (numsegs == 0)
706   				base_addr = trunc_page(phdr[i].p_vaddr +
707 				    rbase);
708 			numsegs++;
709 		}
710 	}
711 	*addr = base_addr;
712 	*entry = (unsigned long)hdr->e_entry + rbase;
713 
714 fail:
715 	if (imgp->firstpage)
716 		exec_unmap_first_page(imgp);
717 
718 	if (nd->ni_vp)
719 		vput(nd->ni_vp);
720 
721 	free(tempdata, M_TEMP);
722 
723 	return (error);
724 }
725 
726 static int
727 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
728 {
729 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
730 	const Elf_Phdr *phdr;
731 	Elf_Auxargs *elf_auxargs;
732 	struct vmspace *vmspace;
733 	vm_prot_t prot;
734 	u_long text_size = 0, data_size = 0, total_size = 0;
735 	u_long text_addr = 0, data_addr = 0;
736 	u_long seg_size, seg_addr;
737 	u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
738 	int32_t osrel = 0;
739 	int error = 0, i, n, interp_name_len = 0;
740 	const char *interp = NULL, *newinterp = NULL;
741 	Elf_Brandinfo *brand_info;
742 	char *path;
743 	struct sysentvec *sv;
744 
745 	/*
746 	 * Do we have a valid ELF header ?
747 	 *
748 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
749 	 * if particular brand doesn't support it.
750 	 */
751 	if (__elfN(check_header)(hdr) != 0 ||
752 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
753 		return (-1);
754 
755 	/*
756 	 * From here on down, we return an errno, not -1, as we've
757 	 * detected an ELF file.
758 	 */
759 
760 	if ((hdr->e_phoff > PAGE_SIZE) ||
761 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
762 		/* Only support headers in first page for now */
763 		return (ENOEXEC);
764 	}
765 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
766 	if (!aligned(phdr, Elf_Addr))
767 		return (ENOEXEC);
768 	n = 0;
769 	baddr = 0;
770 	for (i = 0; i < hdr->e_phnum; i++) {
771 		switch (phdr[i].p_type) {
772 		case PT_LOAD:
773 			if (n == 0)
774 				baddr = phdr[i].p_vaddr;
775 			n++;
776 			break;
777 		case PT_INTERP:
778 			/* Path to interpreter */
779 			if (phdr[i].p_filesz > MAXPATHLEN ||
780 			    phdr[i].p_offset > PAGE_SIZE ||
781 			    phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
782 				return (ENOEXEC);
783 			interp = imgp->image_header + phdr[i].p_offset;
784 			interp_name_len = phdr[i].p_filesz;
785 			break;
786 		case PT_GNU_STACK:
787 			if (__elfN(nxstack))
788 				imgp->stack_prot =
789 				    __elfN(trans_prot)(phdr[i].p_flags);
790 			break;
791 		}
792 	}
793 
794 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
795 	    &osrel);
796 	if (brand_info == NULL) {
797 		uprintf("ELF binary type \"%u\" not known.\n",
798 		    hdr->e_ident[EI_OSABI]);
799 		return (ENOEXEC);
800 	}
801 	if (hdr->e_type == ET_DYN) {
802 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
803 			return (ENOEXEC);
804 		/*
805 		 * Honour the base load address from the dso if it is
806 		 * non-zero for some reason.
807 		 */
808 		if (baddr == 0)
809 			et_dyn_addr = ET_DYN_LOAD_ADDR;
810 		else
811 			et_dyn_addr = 0;
812 	} else
813 		et_dyn_addr = 0;
814 	sv = brand_info->sysvec;
815 	if (interp != NULL && brand_info->interp_newpath != NULL)
816 		newinterp = brand_info->interp_newpath;
817 
818 	/*
819 	 * Avoid a possible deadlock if the current address space is destroyed
820 	 * and that address space maps the locked vnode.  In the common case,
821 	 * the locked vnode's v_usecount is decremented but remains greater
822 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
823 	 * However, in cases where the vnode lock is external, such as nullfs,
824 	 * v_usecount may become zero.
825 	 *
826 	 * The VV_TEXT flag prevents modifications to the executable while
827 	 * the vnode is unlocked.
828 	 */
829 	VOP_UNLOCK(imgp->vp, 0);
830 
831 	error = exec_new_vmspace(imgp, sv);
832 	imgp->proc->p_sysent = sv;
833 
834 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
835 	if (error)
836 		return (error);
837 
838 	for (i = 0; i < hdr->e_phnum; i++) {
839 		switch (phdr[i].p_type) {
840 		case PT_LOAD:	/* Loadable segment */
841 			if (phdr[i].p_memsz == 0)
842 				break;
843 			prot = __elfN(trans_prot)(phdr[i].p_flags);
844 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
845 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
846 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
847 			    sv->sv_pagesize);
848 			if (error != 0)
849 				return (error);
850 
851 			/*
852 			 * If this segment contains the program headers,
853 			 * remember their virtual address for the AT_PHDR
854 			 * aux entry. Static binaries don't usually include
855 			 * a PT_PHDR entry.
856 			 */
857 			if (phdr[i].p_offset == 0 &&
858 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
859 				<= phdr[i].p_filesz)
860 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
861 				    et_dyn_addr;
862 
863 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
864 			seg_size = round_page(phdr[i].p_memsz +
865 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
866 
867 			/*
868 			 * Make the largest executable segment the official
869 			 * text segment and all others data.
870 			 *
871 			 * Note that obreak() assumes that data_addr +
872 			 * data_size == end of data load area, and the ELF
873 			 * file format expects segments to be sorted by
874 			 * address.  If multiple data segments exist, the
875 			 * last one will be used.
876 			 */
877 
878 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
879 				text_size = seg_size;
880 				text_addr = seg_addr;
881 			} else {
882 				data_size = seg_size;
883 				data_addr = seg_addr;
884 			}
885 			total_size += seg_size;
886 			break;
887 		case PT_PHDR: 	/* Program header table info */
888 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
889 			break;
890 		default:
891 			break;
892 		}
893 	}
894 
895 	if (data_addr == 0 && data_size == 0) {
896 		data_addr = text_addr;
897 		data_size = text_size;
898 	}
899 
900 	entry = (u_long)hdr->e_entry + et_dyn_addr;
901 
902 	/*
903 	 * Check limits.  It should be safe to check the
904 	 * limits after loading the segments since we do
905 	 * not actually fault in all the segments pages.
906 	 */
907 	PROC_LOCK(imgp->proc);
908 	if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
909 	    text_size > maxtsiz ||
910 	    total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
911 	    racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
912 	    racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
913 		PROC_UNLOCK(imgp->proc);
914 		return (ENOMEM);
915 	}
916 
917 	vmspace = imgp->proc->p_vmspace;
918 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
919 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
920 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
921 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
922 
923 	/*
924 	 * We load the dynamic linker where a userland call
925 	 * to mmap(0, ...) would put it.  The rationale behind this
926 	 * calculation is that it leaves room for the heap to grow to
927 	 * its maximum allowed size.
928 	 */
929 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
930 	    RLIMIT_DATA));
931 	PROC_UNLOCK(imgp->proc);
932 
933 	imgp->entry_addr = entry;
934 
935 	if (interp != NULL) {
936 		int have_interp = FALSE;
937 		VOP_UNLOCK(imgp->vp, 0);
938 		if (brand_info->emul_path != NULL &&
939 		    brand_info->emul_path[0] != '\0') {
940 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
941 			snprintf(path, MAXPATHLEN, "%s%s",
942 			    brand_info->emul_path, interp);
943 			error = __elfN(load_file)(imgp->proc, path, &addr,
944 			    &imgp->entry_addr, sv->sv_pagesize);
945 			free(path, M_TEMP);
946 			if (error == 0)
947 				have_interp = TRUE;
948 		}
949 		if (!have_interp && newinterp != NULL) {
950 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
951 			    &imgp->entry_addr, sv->sv_pagesize);
952 			if (error == 0)
953 				have_interp = TRUE;
954 		}
955 		if (!have_interp) {
956 			error = __elfN(load_file)(imgp->proc, interp, &addr,
957 			    &imgp->entry_addr, sv->sv_pagesize);
958 		}
959 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
960 		if (error != 0) {
961 			uprintf("ELF interpreter %s not found\n", interp);
962 			return (error);
963 		}
964 	} else
965 		addr = et_dyn_addr;
966 
967 	/*
968 	 * Construct auxargs table (used by the fixup routine)
969 	 */
970 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
971 	elf_auxargs->execfd = -1;
972 	elf_auxargs->phdr = proghdr;
973 	elf_auxargs->phent = hdr->e_phentsize;
974 	elf_auxargs->phnum = hdr->e_phnum;
975 	elf_auxargs->pagesz = PAGE_SIZE;
976 	elf_auxargs->base = addr;
977 	elf_auxargs->flags = 0;
978 	elf_auxargs->entry = entry;
979 
980 	imgp->auxargs = elf_auxargs;
981 	imgp->interpreted = 0;
982 	imgp->reloc_base = addr;
983 	imgp->proc->p_osrel = osrel;
984 
985 	return (error);
986 }
987 
988 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
989 
990 int
991 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
992 {
993 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
994 	Elf_Addr *base;
995 	Elf_Addr *pos;
996 
997 	base = (Elf_Addr *)*stack_base;
998 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
999 
1000 	if (args->execfd != -1)
1001 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1002 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1003 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1004 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1005 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1006 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1007 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1008 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1009 	if (imgp->execpathp != 0)
1010 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1011 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1012 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1013 	if (imgp->canary != 0) {
1014 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1015 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1016 	}
1017 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1018 	if (imgp->pagesizes != 0) {
1019 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1020 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1021 	}
1022 	if (imgp->sysent->sv_timekeep_base != 0) {
1023 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1024 		    imgp->sysent->sv_timekeep_base);
1025 	}
1026 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1027 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1028 	    imgp->sysent->sv_stackprot);
1029 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1030 
1031 	free(imgp->auxargs, M_TEMP);
1032 	imgp->auxargs = NULL;
1033 
1034 	base--;
1035 	suword(base, (long)imgp->args->argc);
1036 	*stack_base = (register_t *)base;
1037 	return (0);
1038 }
1039 
1040 /*
1041  * Code for generating ELF core dumps.
1042  */
1043 
1044 typedef void (*segment_callback)(vm_map_entry_t, void *);
1045 
1046 /* Closure for cb_put_phdr(). */
1047 struct phdr_closure {
1048 	Elf_Phdr *phdr;		/* Program header to fill in */
1049 	Elf_Off offset;		/* Offset of segment in core file */
1050 };
1051 
1052 /* Closure for cb_size_segment(). */
1053 struct sseg_closure {
1054 	int count;		/* Count of writable segments. */
1055 	size_t size;		/* Total size of all writable segments. */
1056 };
1057 
1058 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1059 
1060 struct note_info {
1061 	int		type;		/* Note type. */
1062 	outfunc_t 	outfunc; 	/* Output function. */
1063 	void		*outarg;	/* Argument for the output function. */
1064 	size_t		outsize;	/* Output size. */
1065 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1066 };
1067 
1068 TAILQ_HEAD(note_info_list, note_info);
1069 
1070 static void cb_put_phdr(vm_map_entry_t, void *);
1071 static void cb_size_segment(vm_map_entry_t, void *);
1072 static void each_writable_segment(struct thread *, segment_callback, void *);
1073 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1074     int, void *, size_t, struct note_info_list *, size_t, gzFile);
1075 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1076     size_t *);
1077 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1078 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1079 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1080 static int sbuf_drain_core_output(void *, const char *, int);
1081 static int sbuf_drain_count(void *arg, const char *data, int len);
1082 
1083 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1084 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1085 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1086 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1087 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1088 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1089 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1090 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1091 static void note_procstat_files(void *, struct sbuf *, size_t *);
1092 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1093 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1094 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1095 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1096 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1097 
1098 #ifdef COMPRESS_USER_CORES
1099 extern int compress_user_cores;
1100 extern int compress_user_cores_gzlevel;
1101 #endif
1102 
1103 static int
1104 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1105     struct ucred *active_cred, struct ucred *file_cred,
1106     struct thread *td, char *core_buf, gzFile gzfile) {
1107 
1108 	int error;
1109 	if (gzfile) {
1110 #ifdef COMPRESS_USER_CORES
1111 		error = compress_core(gzfile, base, core_buf, len, td);
1112 #else
1113 		panic("shouldn't be here");
1114 #endif
1115 	} else {
1116 		error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1117 		    UIO_USERSPACE, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1118 		    active_cred, file_cred, NULL, td);
1119 	}
1120 	return (error);
1121 }
1122 
1123 /* Coredump output parameters for sbuf drain routine. */
1124 struct sbuf_drain_core_params {
1125 	off_t		offset;
1126 	struct ucred	*active_cred;
1127 	struct ucred	*file_cred;
1128 	struct thread	*td;
1129 	struct vnode	*vp;
1130 #ifdef COMPRESS_USER_CORES
1131 	gzFile		gzfile;
1132 #endif
1133 };
1134 
1135 /*
1136  * Drain into a core file.
1137  */
1138 static int
1139 sbuf_drain_core_output(void *arg, const char *data, int len)
1140 {
1141 	struct sbuf_drain_core_params *p;
1142 	int error, locked;
1143 
1144 	p = (struct sbuf_drain_core_params *)arg;
1145 
1146 	/*
1147 	 * Some kern_proc out routines that print to this sbuf may
1148 	 * call us with the process lock held. Draining with the
1149 	 * non-sleepable lock held is unsafe. The lock is needed for
1150 	 * those routines when dumping a live process. In our case we
1151 	 * can safely release the lock before draining and acquire
1152 	 * again after.
1153 	 */
1154 	locked = PROC_LOCKED(p->td->td_proc);
1155 	if (locked)
1156 		PROC_UNLOCK(p->td->td_proc);
1157 #ifdef COMPRESS_USER_CORES
1158 	if (p->gzfile != Z_NULL)
1159 		error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1160 		    len, p->td);
1161 	else
1162 #endif
1163 		error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1164 		    __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1165 		    IO_UNIT | IO_DIRECT | IO_RANGELOCKED, p->active_cred,
1166 		    p->file_cred, NULL, p->td);
1167 	if (locked)
1168 		PROC_LOCK(p->td->td_proc);
1169 	if (error != 0)
1170 		return (-error);
1171 	p->offset += len;
1172 	return (len);
1173 }
1174 
1175 /*
1176  * Drain into a counter.
1177  */
1178 static int
1179 sbuf_drain_count(void *arg, const char *data __unused, int len)
1180 {
1181 	size_t *sizep;
1182 
1183 	sizep = (size_t *)arg;
1184 	*sizep += len;
1185 	return (len);
1186 }
1187 
1188 int
1189 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1190 {
1191 	struct ucred *cred = td->td_ucred;
1192 	int error = 0;
1193 	struct sseg_closure seginfo;
1194 	struct note_info_list notelst;
1195 	struct note_info *ninfo;
1196 	void *hdr;
1197 	size_t hdrsize, notesz, coresize;
1198 
1199 	gzFile gzfile = Z_NULL;
1200 	char *core_buf = NULL;
1201 #ifdef COMPRESS_USER_CORES
1202 	char gzopen_flags[8];
1203 	char *p;
1204 	int doing_compress = flags & IMGACT_CORE_COMPRESS;
1205 #endif
1206 
1207 	hdr = NULL;
1208 	TAILQ_INIT(&notelst);
1209 
1210 #ifdef COMPRESS_USER_CORES
1211         if (doing_compress) {
1212                 p = gzopen_flags;
1213                 *p++ = 'w';
1214                 if (compress_user_cores_gzlevel >= 0 &&
1215                     compress_user_cores_gzlevel <= 9)
1216                         *p++ = '0' + compress_user_cores_gzlevel;
1217                 *p = 0;
1218                 gzfile = gz_open("", gzopen_flags, vp);
1219                 if (gzfile == Z_NULL) {
1220                         error = EFAULT;
1221                         goto done;
1222                 }
1223                 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1224                 if (!core_buf) {
1225                         error = ENOMEM;
1226                         goto done;
1227                 }
1228         }
1229 #endif
1230 
1231 	/* Size the program segments. */
1232 	seginfo.count = 0;
1233 	seginfo.size = 0;
1234 	each_writable_segment(td, cb_size_segment, &seginfo);
1235 
1236 	/*
1237 	 * Collect info about the core file header area.
1238 	 */
1239 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1240 	__elfN(prepare_notes)(td, &notelst, &notesz);
1241 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1242 
1243 #ifdef RACCT
1244 	PROC_LOCK(td->td_proc);
1245 	error = racct_add(td->td_proc, RACCT_CORE, coresize);
1246 	PROC_UNLOCK(td->td_proc);
1247 	if (error != 0) {
1248 		error = EFAULT;
1249 		goto done;
1250 	}
1251 #endif
1252 	if (coresize >= limit) {
1253 		error = EFAULT;
1254 		goto done;
1255 	}
1256 
1257 	/*
1258 	 * Allocate memory for building the header, fill it up,
1259 	 * and write it out following the notes.
1260 	 */
1261 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1262 	if (hdr == NULL) {
1263 		error = EINVAL;
1264 		goto done;
1265 	}
1266 	error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1267 	    &notelst, notesz, gzfile);
1268 
1269 	/* Write the contents of all of the writable segments. */
1270 	if (error == 0) {
1271 		Elf_Phdr *php;
1272 		off_t offset;
1273 		int i;
1274 
1275 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1276 		offset = round_page(hdrsize + notesz);
1277 		for (i = 0; i < seginfo.count; i++) {
1278 			error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1279 			    php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1280 			if (error != 0)
1281 				break;
1282 			offset += php->p_filesz;
1283 			php++;
1284 		}
1285 	}
1286 	if (error) {
1287 		log(LOG_WARNING,
1288 		    "Failed to write core file for process %s (error %d)\n",
1289 		    curproc->p_comm, error);
1290 	}
1291 
1292 done:
1293 #ifdef COMPRESS_USER_CORES
1294 	if (core_buf)
1295 		free(core_buf, M_TEMP);
1296 	if (gzfile)
1297 		gzclose(gzfile);
1298 #endif
1299 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1300 		TAILQ_REMOVE(&notelst, ninfo, link);
1301 		free(ninfo, M_TEMP);
1302 	}
1303 	if (hdr != NULL)
1304 		free(hdr, M_TEMP);
1305 
1306 	return (error);
1307 }
1308 
1309 /*
1310  * A callback for each_writable_segment() to write out the segment's
1311  * program header entry.
1312  */
1313 static void
1314 cb_put_phdr(entry, closure)
1315 	vm_map_entry_t entry;
1316 	void *closure;
1317 {
1318 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1319 	Elf_Phdr *phdr = phc->phdr;
1320 
1321 	phc->offset = round_page(phc->offset);
1322 
1323 	phdr->p_type = PT_LOAD;
1324 	phdr->p_offset = phc->offset;
1325 	phdr->p_vaddr = entry->start;
1326 	phdr->p_paddr = 0;
1327 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1328 	phdr->p_align = PAGE_SIZE;
1329 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1330 
1331 	phc->offset += phdr->p_filesz;
1332 	phc->phdr++;
1333 }
1334 
1335 /*
1336  * A callback for each_writable_segment() to gather information about
1337  * the number of segments and their total size.
1338  */
1339 static void
1340 cb_size_segment(entry, closure)
1341 	vm_map_entry_t entry;
1342 	void *closure;
1343 {
1344 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1345 
1346 	ssc->count++;
1347 	ssc->size += entry->end - entry->start;
1348 }
1349 
1350 /*
1351  * For each writable segment in the process's memory map, call the given
1352  * function with a pointer to the map entry and some arbitrary
1353  * caller-supplied data.
1354  */
1355 static void
1356 each_writable_segment(td, func, closure)
1357 	struct thread *td;
1358 	segment_callback func;
1359 	void *closure;
1360 {
1361 	struct proc *p = td->td_proc;
1362 	vm_map_t map = &p->p_vmspace->vm_map;
1363 	vm_map_entry_t entry;
1364 	vm_object_t backing_object, object;
1365 	boolean_t ignore_entry;
1366 
1367 	vm_map_lock_read(map);
1368 	for (entry = map->header.next; entry != &map->header;
1369 	    entry = entry->next) {
1370 		/*
1371 		 * Don't dump inaccessible mappings, deal with legacy
1372 		 * coredump mode.
1373 		 *
1374 		 * Note that read-only segments related to the elf binary
1375 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1376 		 * need to arbitrarily ignore such segments.
1377 		 */
1378 		if (elf_legacy_coredump) {
1379 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1380 				continue;
1381 		} else {
1382 			if ((entry->protection & VM_PROT_ALL) == 0)
1383 				continue;
1384 		}
1385 
1386 		/*
1387 		 * Dont include memory segment in the coredump if
1388 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1389 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1390 		 * kernel map).
1391 		 */
1392 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1393 			continue;
1394 
1395 		if ((object = entry->object.vm_object) == NULL)
1396 			continue;
1397 
1398 		/* Ignore memory-mapped devices and such things. */
1399 		VM_OBJECT_RLOCK(object);
1400 		while ((backing_object = object->backing_object) != NULL) {
1401 			VM_OBJECT_RLOCK(backing_object);
1402 			VM_OBJECT_RUNLOCK(object);
1403 			object = backing_object;
1404 		}
1405 		ignore_entry = object->type != OBJT_DEFAULT &&
1406 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1407 		    object->type != OBJT_PHYS;
1408 		VM_OBJECT_RUNLOCK(object);
1409 		if (ignore_entry)
1410 			continue;
1411 
1412 		(*func)(entry, closure);
1413 	}
1414 	vm_map_unlock_read(map);
1415 }
1416 
1417 /*
1418  * Write the core file header to the file, including padding up to
1419  * the page boundary.
1420  */
1421 static int
1422 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1423     int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1424     size_t notesz, gzFile gzfile)
1425 {
1426 	struct sbuf_drain_core_params params;
1427 	struct note_info *ninfo;
1428 	struct sbuf *sb;
1429 	int error;
1430 
1431 	/* Fill in the header. */
1432 	bzero(hdr, hdrsize);
1433 	__elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1434 
1435 	params.offset = 0;
1436 	params.active_cred = cred;
1437 	params.file_cred = NOCRED;
1438 	params.td = td;
1439 	params.vp = vp;
1440 #ifdef COMPRESS_USER_CORES
1441 	params.gzfile = gzfile;
1442 #endif
1443 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1444 	sbuf_set_drain(sb, sbuf_drain_core_output, &params);
1445 	sbuf_start_section(sb, NULL);
1446 	sbuf_bcat(sb, hdr, hdrsize);
1447 	TAILQ_FOREACH(ninfo, notelst, link)
1448 	    __elfN(putnote)(ninfo, sb);
1449 	/* Align up to a page boundary for the program segments. */
1450 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1451 	error = sbuf_finish(sb);
1452 	sbuf_delete(sb);
1453 
1454 	return (error);
1455 }
1456 
1457 static void
1458 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1459     size_t *sizep)
1460 {
1461 	struct proc *p;
1462 	struct thread *thr;
1463 	size_t size;
1464 
1465 	p = td->td_proc;
1466 	size = 0;
1467 
1468 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1469 
1470 	/*
1471 	 * To have the debugger select the right thread (LWP) as the initial
1472 	 * thread, we dump the state of the thread passed to us in td first.
1473 	 * This is the thread that causes the core dump and thus likely to
1474 	 * be the right thread one wants to have selected in the debugger.
1475 	 */
1476 	thr = td;
1477 	while (thr != NULL) {
1478 		size += register_note(list, NT_PRSTATUS,
1479 		    __elfN(note_prstatus), thr);
1480 		size += register_note(list, NT_FPREGSET,
1481 		    __elfN(note_fpregset), thr);
1482 		size += register_note(list, NT_THRMISC,
1483 		    __elfN(note_thrmisc), thr);
1484 		size += register_note(list, -1,
1485 		    __elfN(note_threadmd), thr);
1486 
1487 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1488 		    TAILQ_NEXT(thr, td_plist);
1489 		if (thr == td)
1490 			thr = TAILQ_NEXT(thr, td_plist);
1491 	}
1492 
1493 	size += register_note(list, NT_PROCSTAT_PROC,
1494 	    __elfN(note_procstat_proc), p);
1495 	size += register_note(list, NT_PROCSTAT_FILES,
1496 	    note_procstat_files, p);
1497 	size += register_note(list, NT_PROCSTAT_VMMAP,
1498 	    note_procstat_vmmap, p);
1499 	size += register_note(list, NT_PROCSTAT_GROUPS,
1500 	    note_procstat_groups, p);
1501 	size += register_note(list, NT_PROCSTAT_UMASK,
1502 	    note_procstat_umask, p);
1503 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1504 	    note_procstat_rlimit, p);
1505 	size += register_note(list, NT_PROCSTAT_OSREL,
1506 	    note_procstat_osrel, p);
1507 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1508 	    __elfN(note_procstat_psstrings), p);
1509 	size += register_note(list, NT_PROCSTAT_AUXV,
1510 	    __elfN(note_procstat_auxv), p);
1511 
1512 	*sizep = size;
1513 }
1514 
1515 static void
1516 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1517     size_t notesz)
1518 {
1519 	Elf_Ehdr *ehdr;
1520 	Elf_Phdr *phdr;
1521 	struct phdr_closure phc;
1522 
1523 	ehdr = (Elf_Ehdr *)hdr;
1524 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1525 
1526 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1527 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1528 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1529 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1530 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1531 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1532 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1533 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1534 	ehdr->e_ident[EI_ABIVERSION] = 0;
1535 	ehdr->e_ident[EI_PAD] = 0;
1536 	ehdr->e_type = ET_CORE;
1537 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1538 	ehdr->e_machine = ELF_ARCH32;
1539 #else
1540 	ehdr->e_machine = ELF_ARCH;
1541 #endif
1542 	ehdr->e_version = EV_CURRENT;
1543 	ehdr->e_entry = 0;
1544 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1545 	ehdr->e_flags = 0;
1546 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1547 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1548 	ehdr->e_phnum = numsegs + 1;
1549 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1550 	ehdr->e_shnum = 0;
1551 	ehdr->e_shstrndx = SHN_UNDEF;
1552 
1553 	/*
1554 	 * Fill in the program header entries.
1555 	 */
1556 
1557 	/* The note segement. */
1558 	phdr->p_type = PT_NOTE;
1559 	phdr->p_offset = hdrsize;
1560 	phdr->p_vaddr = 0;
1561 	phdr->p_paddr = 0;
1562 	phdr->p_filesz = notesz;
1563 	phdr->p_memsz = 0;
1564 	phdr->p_flags = PF_R;
1565 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1566 	phdr++;
1567 
1568 	/* All the writable segments from the program. */
1569 	phc.phdr = phdr;
1570 	phc.offset = round_page(hdrsize + notesz);
1571 	each_writable_segment(td, cb_put_phdr, &phc);
1572 }
1573 
1574 static size_t
1575 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1576 {
1577 	struct note_info *ninfo;
1578 	size_t size, notesize;
1579 
1580 	size = 0;
1581 	out(arg, NULL, &size);
1582 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1583 	ninfo->type = type;
1584 	ninfo->outfunc = out;
1585 	ninfo->outarg = arg;
1586 	ninfo->outsize = size;
1587 	TAILQ_INSERT_TAIL(list, ninfo, link);
1588 
1589 	if (type == -1)
1590 		return (size);
1591 
1592 	notesize = sizeof(Elf_Note) +		/* note header */
1593 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1594 						/* note name */
1595 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1596 
1597 	return (notesize);
1598 }
1599 
1600 static size_t
1601 append_note_data(const void *src, void *dst, size_t len)
1602 {
1603 	size_t padded_len;
1604 
1605 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1606 	if (dst != NULL) {
1607 		bcopy(src, dst, len);
1608 		bzero((char *)dst + len, padded_len - len);
1609 	}
1610 	return (padded_len);
1611 }
1612 
1613 size_t
1614 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1615 {
1616 	Elf_Note *note;
1617 	char *buf;
1618 	size_t notesize;
1619 
1620 	buf = dst;
1621 	if (buf != NULL) {
1622 		note = (Elf_Note *)buf;
1623 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1624 		note->n_descsz = size;
1625 		note->n_type = type;
1626 		buf += sizeof(*note);
1627 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1628 		    sizeof(FREEBSD_ABI_VENDOR));
1629 		append_note_data(src, buf, size);
1630 		if (descp != NULL)
1631 			*descp = buf;
1632 	}
1633 
1634 	notesize = sizeof(Elf_Note) +		/* note header */
1635 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1636 						/* note name */
1637 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1638 
1639 	return (notesize);
1640 }
1641 
1642 static void
1643 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1644 {
1645 	Elf_Note note;
1646 	ssize_t old_len;
1647 
1648 	if (ninfo->type == -1) {
1649 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1650 		return;
1651 	}
1652 
1653 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1654 	note.n_descsz = ninfo->outsize;
1655 	note.n_type = ninfo->type;
1656 
1657 	sbuf_bcat(sb, &note, sizeof(note));
1658 	sbuf_start_section(sb, &old_len);
1659 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1660 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1661 	if (note.n_descsz == 0)
1662 		return;
1663 	sbuf_start_section(sb, &old_len);
1664 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1665 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1666 }
1667 
1668 /*
1669  * Miscellaneous note out functions.
1670  */
1671 
1672 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1673 #include <compat/freebsd32/freebsd32.h>
1674 
1675 typedef struct prstatus32 elf_prstatus_t;
1676 typedef struct prpsinfo32 elf_prpsinfo_t;
1677 typedef struct fpreg32 elf_prfpregset_t;
1678 typedef struct fpreg32 elf_fpregset_t;
1679 typedef struct reg32 elf_gregset_t;
1680 typedef struct thrmisc32 elf_thrmisc_t;
1681 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1682 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1683 typedef uint32_t elf_ps_strings_t;
1684 #else
1685 typedef prstatus_t elf_prstatus_t;
1686 typedef prpsinfo_t elf_prpsinfo_t;
1687 typedef prfpregset_t elf_prfpregset_t;
1688 typedef prfpregset_t elf_fpregset_t;
1689 typedef gregset_t elf_gregset_t;
1690 typedef thrmisc_t elf_thrmisc_t;
1691 #define ELF_KERN_PROC_MASK	0
1692 typedef struct kinfo_proc elf_kinfo_proc_t;
1693 typedef vm_offset_t elf_ps_strings_t;
1694 #endif
1695 
1696 static void
1697 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1698 {
1699 	struct proc *p;
1700 	elf_prpsinfo_t *psinfo;
1701 
1702 	p = (struct proc *)arg;
1703 	if (sb != NULL) {
1704 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1705 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1706 		psinfo->pr_version = PRPSINFO_VERSION;
1707 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1708 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1709 		/*
1710 		 * XXX - We don't fill in the command line arguments properly
1711 		 * yet.
1712 		 */
1713 		strlcpy(psinfo->pr_psargs, p->p_comm,
1714 		    sizeof(psinfo->pr_psargs));
1715 
1716 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1717 		free(psinfo, M_TEMP);
1718 	}
1719 	*sizep = sizeof(*psinfo);
1720 }
1721 
1722 static void
1723 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1724 {
1725 	struct thread *td;
1726 	elf_prstatus_t *status;
1727 
1728 	td = (struct thread *)arg;
1729 	if (sb != NULL) {
1730 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1731 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1732 		status->pr_version = PRSTATUS_VERSION;
1733 		status->pr_statussz = sizeof(elf_prstatus_t);
1734 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1735 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1736 		status->pr_osreldate = osreldate;
1737 		status->pr_cursig = td->td_proc->p_sig;
1738 		status->pr_pid = td->td_tid;
1739 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1740 		fill_regs32(td, &status->pr_reg);
1741 #else
1742 		fill_regs(td, &status->pr_reg);
1743 #endif
1744 		sbuf_bcat(sb, status, sizeof(*status));
1745 		free(status, M_TEMP);
1746 	}
1747 	*sizep = sizeof(*status);
1748 }
1749 
1750 static void
1751 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1752 {
1753 	struct thread *td;
1754 	elf_prfpregset_t *fpregset;
1755 
1756 	td = (struct thread *)arg;
1757 	if (sb != NULL) {
1758 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1759 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1760 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1761 		fill_fpregs32(td, fpregset);
1762 #else
1763 		fill_fpregs(td, fpregset);
1764 #endif
1765 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1766 		free(fpregset, M_TEMP);
1767 	}
1768 	*sizep = sizeof(*fpregset);
1769 }
1770 
1771 static void
1772 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1773 {
1774 	struct thread *td;
1775 	elf_thrmisc_t thrmisc;
1776 
1777 	td = (struct thread *)arg;
1778 	if (sb != NULL) {
1779 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1780 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1781 		strcpy(thrmisc.pr_tname, td->td_name);
1782 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1783 	}
1784 	*sizep = sizeof(thrmisc);
1785 }
1786 
1787 /*
1788  * Allow for MD specific notes, as well as any MD
1789  * specific preparations for writing MI notes.
1790  */
1791 static void
1792 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1793 {
1794 	struct thread *td;
1795 	void *buf;
1796 	size_t size;
1797 
1798 	td = (struct thread *)arg;
1799 	size = *sizep;
1800 	if (size != 0 && sb != NULL)
1801 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1802 	else
1803 		buf = NULL;
1804 	size = 0;
1805 	__elfN(dump_thread)(td, buf, &size);
1806 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1807 	if (size != 0 && sb != NULL)
1808 		sbuf_bcat(sb, buf, size);
1809 	free(buf, M_TEMP);
1810 	*sizep = size;
1811 }
1812 
1813 #ifdef KINFO_PROC_SIZE
1814 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1815 #endif
1816 
1817 static void
1818 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1819 {
1820 	struct proc *p;
1821 	size_t size;
1822 	int structsize;
1823 
1824 	p = (struct proc *)arg;
1825 	size = sizeof(structsize) + p->p_numthreads *
1826 	    sizeof(elf_kinfo_proc_t);
1827 
1828 	if (sb != NULL) {
1829 		KASSERT(*sizep == size, ("invalid size"));
1830 		structsize = sizeof(elf_kinfo_proc_t);
1831 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1832 		sx_slock(&proctree_lock);
1833 		PROC_LOCK(p);
1834 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1835 		sx_sunlock(&proctree_lock);
1836 	}
1837 	*sizep = size;
1838 }
1839 
1840 #ifdef KINFO_FILE_SIZE
1841 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1842 #endif
1843 
1844 static void
1845 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1846 {
1847 	struct proc *p;
1848 	size_t size;
1849 	int structsize;
1850 
1851 	p = (struct proc *)arg;
1852 	if (sb == NULL) {
1853 		size = 0;
1854 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1855 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1856 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1857 		PROC_LOCK(p);
1858 		kern_proc_filedesc_out(p, sb, -1);
1859 		sbuf_finish(sb);
1860 		sbuf_delete(sb);
1861 		*sizep = size;
1862 	} else {
1863 		structsize = sizeof(struct kinfo_file);
1864 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1865 		PROC_LOCK(p);
1866 		kern_proc_filedesc_out(p, sb, -1);
1867 	}
1868 }
1869 
1870 #ifdef KINFO_VMENTRY_SIZE
1871 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1872 #endif
1873 
1874 static void
1875 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1876 {
1877 	struct proc *p;
1878 	size_t size;
1879 	int structsize;
1880 
1881 	p = (struct proc *)arg;
1882 	if (sb == NULL) {
1883 		size = 0;
1884 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1885 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1886 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1887 		PROC_LOCK(p);
1888 		kern_proc_vmmap_out(p, sb);
1889 		sbuf_finish(sb);
1890 		sbuf_delete(sb);
1891 		*sizep = size;
1892 	} else {
1893 		structsize = sizeof(struct kinfo_vmentry);
1894 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1895 		PROC_LOCK(p);
1896 		kern_proc_vmmap_out(p, sb);
1897 	}
1898 }
1899 
1900 static void
1901 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1902 {
1903 	struct proc *p;
1904 	size_t size;
1905 	int structsize;
1906 
1907 	p = (struct proc *)arg;
1908 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1909 	if (sb != NULL) {
1910 		KASSERT(*sizep == size, ("invalid size"));
1911 		structsize = sizeof(gid_t);
1912 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1913 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1914 		    sizeof(gid_t));
1915 	}
1916 	*sizep = size;
1917 }
1918 
1919 static void
1920 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1921 {
1922 	struct proc *p;
1923 	size_t size;
1924 	int structsize;
1925 
1926 	p = (struct proc *)arg;
1927 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1928 	if (sb != NULL) {
1929 		KASSERT(*sizep == size, ("invalid size"));
1930 		structsize = sizeof(p->p_fd->fd_cmask);
1931 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1932 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1933 	}
1934 	*sizep = size;
1935 }
1936 
1937 static void
1938 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1939 {
1940 	struct proc *p;
1941 	struct rlimit rlim[RLIM_NLIMITS];
1942 	size_t size;
1943 	int structsize, i;
1944 
1945 	p = (struct proc *)arg;
1946 	size = sizeof(structsize) + sizeof(rlim);
1947 	if (sb != NULL) {
1948 		KASSERT(*sizep == size, ("invalid size"));
1949 		structsize = sizeof(rlim);
1950 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1951 		PROC_LOCK(p);
1952 		for (i = 0; i < RLIM_NLIMITS; i++)
1953 			lim_rlimit(p, i, &rlim[i]);
1954 		PROC_UNLOCK(p);
1955 		sbuf_bcat(sb, rlim, sizeof(rlim));
1956 	}
1957 	*sizep = size;
1958 }
1959 
1960 static void
1961 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1962 {
1963 	struct proc *p;
1964 	size_t size;
1965 	int structsize;
1966 
1967 	p = (struct proc *)arg;
1968 	size = sizeof(structsize) + sizeof(p->p_osrel);
1969 	if (sb != NULL) {
1970 		KASSERT(*sizep == size, ("invalid size"));
1971 		structsize = sizeof(p->p_osrel);
1972 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1973 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1974 	}
1975 	*sizep = size;
1976 }
1977 
1978 static void
1979 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1980 {
1981 	struct proc *p;
1982 	elf_ps_strings_t ps_strings;
1983 	size_t size;
1984 	int structsize;
1985 
1986 	p = (struct proc *)arg;
1987 	size = sizeof(structsize) + sizeof(ps_strings);
1988 	if (sb != NULL) {
1989 		KASSERT(*sizep == size, ("invalid size"));
1990 		structsize = sizeof(ps_strings);
1991 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1992 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1993 #else
1994 		ps_strings = p->p_sysent->sv_psstrings;
1995 #endif
1996 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1997 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
1998 	}
1999 	*sizep = size;
2000 }
2001 
2002 static void
2003 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2004 {
2005 	struct proc *p;
2006 	size_t size;
2007 	int structsize;
2008 
2009 	p = (struct proc *)arg;
2010 	if (sb == NULL) {
2011 		size = 0;
2012 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2013 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2014 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2015 		PHOLD(p);
2016 		proc_getauxv(curthread, p, sb);
2017 		PRELE(p);
2018 		sbuf_finish(sb);
2019 		sbuf_delete(sb);
2020 		*sizep = size;
2021 	} else {
2022 		structsize = sizeof(Elf_Auxinfo);
2023 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2024 		PHOLD(p);
2025 		proc_getauxv(curthread, p, sb);
2026 		PRELE(p);
2027 	}
2028 }
2029 
2030 static boolean_t
2031 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2032     int32_t *osrel, const Elf_Phdr *pnote)
2033 {
2034 	const Elf_Note *note, *note0, *note_end;
2035 	const char *note_name;
2036 	int i;
2037 
2038 	if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
2039 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
2040 		return (FALSE);
2041 
2042 	note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
2043 	note_end = (const Elf_Note *)(imgp->image_header +
2044 	    pnote->p_offset + pnote->p_filesz);
2045 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2046 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2047 		    (const char *)note < sizeof(Elf_Note))
2048 			return (FALSE);
2049 		if (note->n_namesz != checknote->hdr.n_namesz ||
2050 		    note->n_descsz != checknote->hdr.n_descsz ||
2051 		    note->n_type != checknote->hdr.n_type)
2052 			goto nextnote;
2053 		note_name = (const char *)(note + 1);
2054 		if (note_name + checknote->hdr.n_namesz >=
2055 		    (const char *)note_end || strncmp(checknote->vendor,
2056 		    note_name, checknote->hdr.n_namesz) != 0)
2057 			goto nextnote;
2058 
2059 		/*
2060 		 * Fetch the osreldate for binary
2061 		 * from the ELF OSABI-note if necessary.
2062 		 */
2063 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2064 		    checknote->trans_osrel != NULL)
2065 			return (checknote->trans_osrel(note, osrel));
2066 		return (TRUE);
2067 
2068 nextnote:
2069 		note = (const Elf_Note *)((const char *)(note + 1) +
2070 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2071 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2072 	}
2073 
2074 	return (FALSE);
2075 }
2076 
2077 /*
2078  * Try to find the appropriate ABI-note section for checknote,
2079  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2080  * first page of the image is searched, the same as for headers.
2081  */
2082 static boolean_t
2083 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2084     int32_t *osrel)
2085 {
2086 	const Elf_Phdr *phdr;
2087 	const Elf_Ehdr *hdr;
2088 	int i;
2089 
2090 	hdr = (const Elf_Ehdr *)imgp->image_header;
2091 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2092 
2093 	for (i = 0; i < hdr->e_phnum; i++) {
2094 		if (phdr[i].p_type == PT_NOTE &&
2095 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2096 			return (TRUE);
2097 	}
2098 	return (FALSE);
2099 
2100 }
2101 
2102 /*
2103  * Tell kern_execve.c about it, with a little help from the linker.
2104  */
2105 static struct execsw __elfN(execsw) = {
2106 	__CONCAT(exec_, __elfN(imgact)),
2107 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2108 };
2109 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2110 
2111 #ifdef COMPRESS_USER_CORES
2112 /*
2113  * Compress and write out a core segment for a user process.
2114  *
2115  * 'inbuf' is the starting address of a VM segment in the process' address
2116  * space that is to be compressed and written out to the core file.  'dest_buf'
2117  * is a buffer in the kernel's address space.  The segment is copied from
2118  * 'inbuf' to 'dest_buf' first before being processed by the compression
2119  * routine gzwrite().  This copying is necessary because the content of the VM
2120  * segment may change between the compression pass and the crc-computation pass
2121  * in gzwrite().  This is because realtime threads may preempt the UNIX kernel.
2122  *
2123  * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2124  */
2125 static int
2126 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2127     struct thread *td)
2128 {
2129 	int len_compressed;
2130 	int error = 0;
2131 	unsigned int chunk_len;
2132 
2133 	while (len) {
2134 		if (inbuf != NULL) {
2135 			chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2136 			copyin(inbuf, dest_buf, chunk_len);
2137 			inbuf += chunk_len;
2138 		} else {
2139 			chunk_len = len;
2140 		}
2141 		len_compressed = gzwrite(file, dest_buf, chunk_len);
2142 
2143 		EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2144 
2145 		if ((unsigned int)len_compressed != chunk_len) {
2146 			log(LOG_WARNING,
2147 			    "compress_core: length mismatch (0x%x returned, "
2148 			    "0x%x expected)\n", len_compressed, chunk_len);
2149 			EVENTHANDLER_INVOKE(app_coredump_error, td,
2150 			    "compress_core: length mismatch %x -> %x",
2151 			    chunk_len, len_compressed);
2152 			error = EFAULT;
2153 			break;
2154 		}
2155 		len -= chunk_len;
2156 		maybe_yield();
2157 	}
2158 
2159 	return (error);
2160 }
2161 #endif /* COMPRESS_USER_CORES */
2162 
2163 static vm_prot_t
2164 __elfN(trans_prot)(Elf_Word flags)
2165 {
2166 	vm_prot_t prot;
2167 
2168 	prot = 0;
2169 	if (flags & PF_X)
2170 		prot |= VM_PROT_EXECUTE;
2171 	if (flags & PF_W)
2172 		prot |= VM_PROT_WRITE;
2173 	if (flags & PF_R)
2174 		prot |= VM_PROT_READ;
2175 #if __ELF_WORD_SIZE == 32
2176 #if defined(__amd64__)
2177 	if (i386_read_exec && (flags & PF_R))
2178 		prot |= VM_PROT_EXECUTE;
2179 #endif
2180 #endif
2181 	return (prot);
2182 }
2183 
2184 static Elf_Word
2185 __elfN(untrans_prot)(vm_prot_t prot)
2186 {
2187 	Elf_Word flags;
2188 
2189 	flags = 0;
2190 	if (prot & VM_PROT_EXECUTE)
2191 		flags |= PF_X;
2192 	if (prot & VM_PROT_READ)
2193 		flags |= PF_R;
2194 	if (prot & VM_PROT_WRITE)
2195 		flags |= PF_W;
2196 	return (flags);
2197 }
2198