xref: /freebsd/sys/kern/imgact_elf.c (revision 9fe48b8076ae9c6dc486d713c468ff03ec056eda)
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 
84 #define ELF_NOTE_ROUNDSIZE	4
85 #define OLD_EI_BRAND	8
86 
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89     const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91     u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
93     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
94     size_t pagesize);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97     int32_t *osrel);
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100     Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
103 
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
105     "");
106 
107 #define	CORE_BUF_SIZE	(16 * 1024)
108 
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
113 
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116     &elf_legacy_coredump, 0, "");
117 
118 int __elfN(nxstack) =
119 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
120 	1;
121 #else
122 	0;
123 #endif
124 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
125     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
126     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
127 
128 #if __ELF_WORD_SIZE == 32
129 #if defined(__amd64__)
130 int i386_read_exec = 0;
131 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
132     "enable execution from readable segments");
133 #endif
134 #endif
135 
136 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
137 
138 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
139 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
140 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
141 
142 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
143 
144 Elf_Brandnote __elfN(freebsd_brandnote) = {
145 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
146 	.hdr.n_descsz	= sizeof(int32_t),
147 	.hdr.n_type	= 1,
148 	.vendor		= FREEBSD_ABI_VENDOR,
149 	.flags		= BN_TRANSLATE_OSREL,
150 	.trans_osrel	= __elfN(freebsd_trans_osrel)
151 };
152 
153 static boolean_t
154 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
155 {
156 	uintptr_t p;
157 
158 	p = (uintptr_t)(note + 1);
159 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
160 	*osrel = *(const int32_t *)(p);
161 
162 	return (TRUE);
163 }
164 
165 static const char GNU_ABI_VENDOR[] = "GNU";
166 static int GNU_KFREEBSD_ABI_DESC = 3;
167 
168 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
169 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
170 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
171 	.hdr.n_type	= 1,
172 	.vendor		= GNU_ABI_VENDOR,
173 	.flags		= BN_TRANSLATE_OSREL,
174 	.trans_osrel	= kfreebsd_trans_osrel
175 };
176 
177 static boolean_t
178 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
179 {
180 	const Elf32_Word *desc;
181 	uintptr_t p;
182 
183 	p = (uintptr_t)(note + 1);
184 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
185 
186 	desc = (const Elf32_Word *)p;
187 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
188 		return (FALSE);
189 
190 	/*
191 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
192 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
193 	 */
194 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
195 
196 	return (TRUE);
197 }
198 
199 int
200 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
201 {
202 	int i;
203 
204 	for (i = 0; i < MAX_BRANDS; i++) {
205 		if (elf_brand_list[i] == NULL) {
206 			elf_brand_list[i] = entry;
207 			break;
208 		}
209 	}
210 	if (i == MAX_BRANDS) {
211 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
212 			__func__, entry);
213 		return (-1);
214 	}
215 	return (0);
216 }
217 
218 int
219 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
220 {
221 	int i;
222 
223 	for (i = 0; i < MAX_BRANDS; i++) {
224 		if (elf_brand_list[i] == entry) {
225 			elf_brand_list[i] = NULL;
226 			break;
227 		}
228 	}
229 	if (i == MAX_BRANDS)
230 		return (-1);
231 	return (0);
232 }
233 
234 int
235 __elfN(brand_inuse)(Elf_Brandinfo *entry)
236 {
237 	struct proc *p;
238 	int rval = FALSE;
239 
240 	sx_slock(&allproc_lock);
241 	FOREACH_PROC_IN_SYSTEM(p) {
242 		if (p->p_sysent == entry->sysvec) {
243 			rval = TRUE;
244 			break;
245 		}
246 	}
247 	sx_sunlock(&allproc_lock);
248 
249 	return (rval);
250 }
251 
252 static Elf_Brandinfo *
253 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
254     int interp_name_len, int32_t *osrel)
255 {
256 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
257 	Elf_Brandinfo *bi;
258 	boolean_t ret;
259 	int i;
260 
261 	/*
262 	 * We support four types of branding -- (1) the ELF EI_OSABI field
263 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
264 	 * branding w/in the ELF header, (3) path of the `interp_path'
265 	 * field, and (4) the ".note.ABI-tag" ELF section.
266 	 */
267 
268 	/* Look for an ".note.ABI-tag" ELF section */
269 	for (i = 0; i < MAX_BRANDS; i++) {
270 		bi = elf_brand_list[i];
271 		if (bi == NULL)
272 			continue;
273 		if (hdr->e_machine == bi->machine && (bi->flags &
274 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
275 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
276 			if (ret)
277 				return (bi);
278 		}
279 	}
280 
281 	/* If the executable has a brand, search for it in the brand list. */
282 	for (i = 0; i < MAX_BRANDS; i++) {
283 		bi = elf_brand_list[i];
284 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
285 			continue;
286 		if (hdr->e_machine == bi->machine &&
287 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
288 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
289 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
290 			return (bi);
291 	}
292 
293 	/* No known brand, see if the header is recognized by any brand */
294 	for (i = 0; i < MAX_BRANDS; i++) {
295 		bi = elf_brand_list[i];
296 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
297 		    bi->header_supported == NULL)
298 			continue;
299 		if (hdr->e_machine == bi->machine) {
300 			ret = bi->header_supported(imgp);
301 			if (ret)
302 				return (bi);
303 		}
304 	}
305 
306 	/* Lacking a known brand, search for a recognized interpreter. */
307 	if (interp != NULL) {
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 			    /* ELF image p_filesz includes terminating zero */
314 			    strlen(bi->interp_path) + 1 == interp_name_len &&
315 			    strncmp(interp, bi->interp_path, interp_name_len)
316 			    == 0)
317 				return (bi);
318 		}
319 	}
320 
321 	/* Lacking a recognized interpreter, try the default brand */
322 	for (i = 0; i < MAX_BRANDS; i++) {
323 		bi = elf_brand_list[i];
324 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
325 			continue;
326 		if (hdr->e_machine == bi->machine &&
327 		    __elfN(fallback_brand) == bi->brand)
328 			return (bi);
329 	}
330 	return (NULL);
331 }
332 
333 static int
334 __elfN(check_header)(const Elf_Ehdr *hdr)
335 {
336 	Elf_Brandinfo *bi;
337 	int i;
338 
339 	if (!IS_ELF(*hdr) ||
340 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
341 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
342 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
343 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
344 	    hdr->e_version != ELF_TARG_VER)
345 		return (ENOEXEC);
346 
347 	/*
348 	 * Make sure we have at least one brand for this machine.
349 	 */
350 
351 	for (i = 0; i < MAX_BRANDS; i++) {
352 		bi = elf_brand_list[i];
353 		if (bi != NULL && bi->machine == hdr->e_machine)
354 			break;
355 	}
356 	if (i == MAX_BRANDS)
357 		return (ENOEXEC);
358 
359 	return (0);
360 }
361 
362 static int
363 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
364     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
365 {
366 	struct sf_buf *sf;
367 	int error;
368 	vm_offset_t off;
369 
370 	/*
371 	 * Create the page if it doesn't exist yet. Ignore errors.
372 	 */
373 	vm_map_lock(map);
374 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
375 	    VM_PROT_ALL, VM_PROT_ALL, 0);
376 	vm_map_unlock(map);
377 
378 	/*
379 	 * Find the page from the underlying object.
380 	 */
381 	if (object) {
382 		sf = vm_imgact_map_page(object, offset);
383 		if (sf == NULL)
384 			return (KERN_FAILURE);
385 		off = offset - trunc_page(offset);
386 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
387 		    end - start);
388 		vm_imgact_unmap_page(sf);
389 		if (error) {
390 			return (KERN_FAILURE);
391 		}
392 	}
393 
394 	return (KERN_SUCCESS);
395 }
396 
397 static int
398 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
399     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
400 {
401 	struct sf_buf *sf;
402 	vm_offset_t off;
403 	vm_size_t sz;
404 	int error, rv;
405 
406 	if (start != trunc_page(start)) {
407 		rv = __elfN(map_partial)(map, object, offset, start,
408 		    round_page(start), prot);
409 		if (rv)
410 			return (rv);
411 		offset += round_page(start) - start;
412 		start = round_page(start);
413 	}
414 	if (end != round_page(end)) {
415 		rv = __elfN(map_partial)(map, object, offset +
416 		    trunc_page(end) - start, trunc_page(end), end, prot);
417 		if (rv)
418 			return (rv);
419 		end = trunc_page(end);
420 	}
421 	if (end > start) {
422 		if (offset & PAGE_MASK) {
423 			/*
424 			 * The mapping is not page aligned. This means we have
425 			 * to copy the data. Sigh.
426 			 */
427 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
428 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
429 			    0);
430 			if (rv)
431 				return (rv);
432 			if (object == NULL)
433 				return (KERN_SUCCESS);
434 			for (; start < end; start += sz) {
435 				sf = vm_imgact_map_page(object, offset);
436 				if (sf == NULL)
437 					return (KERN_FAILURE);
438 				off = offset - trunc_page(offset);
439 				sz = end - start;
440 				if (sz > PAGE_SIZE - off)
441 					sz = PAGE_SIZE - off;
442 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
443 				    (caddr_t)start, sz);
444 				vm_imgact_unmap_page(sf);
445 				if (error) {
446 					return (KERN_FAILURE);
447 				}
448 				offset += sz;
449 			}
450 			rv = KERN_SUCCESS;
451 		} else {
452 			vm_object_reference(object);
453 			vm_map_lock(map);
454 			rv = vm_map_insert(map, object, offset, start, end,
455 			    prot, VM_PROT_ALL, cow);
456 			vm_map_unlock(map);
457 			if (rv != KERN_SUCCESS)
458 				vm_object_deallocate(object);
459 		}
460 		return (rv);
461 	} else {
462 		return (KERN_SUCCESS);
463 	}
464 }
465 
466 static int
467 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
468     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
469     size_t pagesize)
470 {
471 	struct sf_buf *sf;
472 	size_t map_len;
473 	vm_map_t map;
474 	vm_object_t object;
475 	vm_offset_t map_addr;
476 	int error, rv, cow;
477 	size_t copy_len;
478 	vm_offset_t file_addr;
479 
480 	/*
481 	 * It's necessary to fail if the filsz + offset taken from the
482 	 * header is greater than the actual file pager object's size.
483 	 * If we were to allow this, then the vm_map_find() below would
484 	 * walk right off the end of the file object and into the ether.
485 	 *
486 	 * While I'm here, might as well check for something else that
487 	 * is invalid: filsz cannot be greater than memsz.
488 	 */
489 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
490 		uprintf("elf_load_section: truncated ELF file\n");
491 		return (ENOEXEC);
492 	}
493 
494 	object = imgp->object;
495 	map = &imgp->proc->p_vmspace->vm_map;
496 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
497 	file_addr = trunc_page_ps(offset, pagesize);
498 
499 	/*
500 	 * We have two choices.  We can either clear the data in the last page
501 	 * of an oversized mapping, or we can start the anon mapping a page
502 	 * early and copy the initialized data into that first page.  We
503 	 * choose the second..
504 	 */
505 	if (memsz > filsz)
506 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
507 	else
508 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
509 
510 	if (map_len != 0) {
511 		/* cow flags: don't dump readonly sections in core */
512 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
513 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
514 
515 		rv = __elfN(map_insert)(map,
516 				      object,
517 				      file_addr,	/* file offset */
518 				      map_addr,		/* virtual start */
519 				      map_addr + map_len,/* virtual end */
520 				      prot,
521 				      cow);
522 		if (rv != KERN_SUCCESS)
523 			return (EINVAL);
524 
525 		/* we can stop now if we've covered it all */
526 		if (memsz == filsz) {
527 			return (0);
528 		}
529 	}
530 
531 
532 	/*
533 	 * We have to get the remaining bit of the file into the first part
534 	 * of the oversized map segment.  This is normally because the .data
535 	 * segment in the file is extended to provide bss.  It's a neat idea
536 	 * to try and save a page, but it's a pain in the behind to implement.
537 	 */
538 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
539 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
540 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
541 	    map_addr;
542 
543 	/* This had damn well better be true! */
544 	if (map_len != 0) {
545 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
546 		    map_len, VM_PROT_ALL, 0);
547 		if (rv != KERN_SUCCESS) {
548 			return (EINVAL);
549 		}
550 	}
551 
552 	if (copy_len != 0) {
553 		vm_offset_t off;
554 
555 		sf = vm_imgact_map_page(object, offset + filsz);
556 		if (sf == NULL)
557 			return (EIO);
558 
559 		/* send the page fragment to user space */
560 		off = trunc_page_ps(offset + filsz, pagesize) -
561 		    trunc_page(offset + filsz);
562 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
563 		    (caddr_t)map_addr, copy_len);
564 		vm_imgact_unmap_page(sf);
565 		if (error) {
566 			return (error);
567 		}
568 	}
569 
570 	/*
571 	 * set it to the specified protection.
572 	 * XXX had better undo the damage from pasting over the cracks here!
573 	 */
574 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
575 	    map_len), prot, FALSE);
576 
577 	return (0);
578 }
579 
580 /*
581  * Load the file "file" into memory.  It may be either a shared object
582  * or an executable.
583  *
584  * The "addr" reference parameter is in/out.  On entry, it specifies
585  * the address where a shared object should be loaded.  If the file is
586  * an executable, this value is ignored.  On exit, "addr" specifies
587  * where the file was actually loaded.
588  *
589  * The "entry" reference parameter is out only.  On exit, it specifies
590  * the entry point for the loaded file.
591  */
592 static int
593 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
594 	u_long *entry, size_t pagesize)
595 {
596 	struct {
597 		struct nameidata nd;
598 		struct vattr attr;
599 		struct image_params image_params;
600 	} *tempdata;
601 	const Elf_Ehdr *hdr = NULL;
602 	const Elf_Phdr *phdr = NULL;
603 	struct nameidata *nd;
604 	struct vattr *attr;
605 	struct image_params *imgp;
606 	vm_prot_t prot;
607 	u_long rbase;
608 	u_long base_addr = 0;
609 	int error, i, numsegs;
610 
611 #ifdef CAPABILITY_MODE
612 	/*
613 	 * XXXJA: This check can go away once we are sufficiently confident
614 	 * that the checks in namei() are correct.
615 	 */
616 	if (IN_CAPABILITY_MODE(curthread))
617 		return (ECAPMODE);
618 #endif
619 
620 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
621 	nd = &tempdata->nd;
622 	attr = &tempdata->attr;
623 	imgp = &tempdata->image_params;
624 
625 	/*
626 	 * Initialize part of the common data
627 	 */
628 	imgp->proc = p;
629 	imgp->attr = attr;
630 	imgp->firstpage = NULL;
631 	imgp->image_header = NULL;
632 	imgp->object = NULL;
633 	imgp->execlabel = NULL;
634 
635 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
636 	if ((error = namei(nd)) != 0) {
637 		nd->ni_vp = NULL;
638 		goto fail;
639 	}
640 	NDFREE(nd, NDF_ONLY_PNBUF);
641 	imgp->vp = nd->ni_vp;
642 
643 	/*
644 	 * Check permissions, modes, uid, etc on the file, and "open" it.
645 	 */
646 	error = exec_check_permissions(imgp);
647 	if (error)
648 		goto fail;
649 
650 	error = exec_map_first_page(imgp);
651 	if (error)
652 		goto fail;
653 
654 	/*
655 	 * Also make certain that the interpreter stays the same, so set
656 	 * its VV_TEXT flag, too.
657 	 */
658 	VOP_SET_TEXT(nd->ni_vp);
659 
660 	imgp->object = nd->ni_vp->v_object;
661 
662 	hdr = (const Elf_Ehdr *)imgp->image_header;
663 	if ((error = __elfN(check_header)(hdr)) != 0)
664 		goto fail;
665 	if (hdr->e_type == ET_DYN)
666 		rbase = *addr;
667 	else if (hdr->e_type == ET_EXEC)
668 		rbase = 0;
669 	else {
670 		error = ENOEXEC;
671 		goto fail;
672 	}
673 
674 	/* Only support headers that fit within first page for now      */
675 	if ((hdr->e_phoff > PAGE_SIZE) ||
676 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
677 		error = ENOEXEC;
678 		goto fail;
679 	}
680 
681 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
682 	if (!aligned(phdr, Elf_Addr)) {
683 		error = ENOEXEC;
684 		goto fail;
685 	}
686 
687 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
688 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
689 			/* Loadable segment */
690 			prot = __elfN(trans_prot)(phdr[i].p_flags);
691 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
692 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
693 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
694 			if (error != 0)
695 				goto fail;
696 			/*
697 			 * Establish the base address if this is the
698 			 * first segment.
699 			 */
700 			if (numsegs == 0)
701   				base_addr = trunc_page(phdr[i].p_vaddr +
702 				    rbase);
703 			numsegs++;
704 		}
705 	}
706 	*addr = base_addr;
707 	*entry = (unsigned long)hdr->e_entry + rbase;
708 
709 fail:
710 	if (imgp->firstpage)
711 		exec_unmap_first_page(imgp);
712 
713 	if (nd->ni_vp)
714 		vput(nd->ni_vp);
715 
716 	free(tempdata, M_TEMP);
717 
718 	return (error);
719 }
720 
721 static int
722 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
723 {
724 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
725 	const Elf_Phdr *phdr;
726 	Elf_Auxargs *elf_auxargs;
727 	struct vmspace *vmspace;
728 	vm_prot_t prot;
729 	u_long text_size = 0, data_size = 0, total_size = 0;
730 	u_long text_addr = 0, data_addr = 0;
731 	u_long seg_size, seg_addr;
732 	u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
733 	int32_t osrel = 0;
734 	int error = 0, i, n, interp_name_len = 0;
735 	const char *interp = NULL, *newinterp = NULL;
736 	Elf_Brandinfo *brand_info;
737 	char *path;
738 	struct sysentvec *sv;
739 
740 	/*
741 	 * Do we have a valid ELF header ?
742 	 *
743 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
744 	 * if particular brand doesn't support it.
745 	 */
746 	if (__elfN(check_header)(hdr) != 0 ||
747 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
748 		return (-1);
749 
750 	/*
751 	 * From here on down, we return an errno, not -1, as we've
752 	 * detected an ELF file.
753 	 */
754 
755 	if ((hdr->e_phoff > PAGE_SIZE) ||
756 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
757 		/* Only support headers in first page for now */
758 		return (ENOEXEC);
759 	}
760 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
761 	if (!aligned(phdr, Elf_Addr))
762 		return (ENOEXEC);
763 	n = 0;
764 	baddr = 0;
765 	for (i = 0; i < hdr->e_phnum; i++) {
766 		switch (phdr[i].p_type) {
767 		case PT_LOAD:
768 			if (n == 0)
769 				baddr = phdr[i].p_vaddr;
770 			n++;
771 			break;
772 		case PT_INTERP:
773 			/* Path to interpreter */
774 			if (phdr[i].p_filesz > MAXPATHLEN ||
775 			    phdr[i].p_offset > PAGE_SIZE ||
776 			    phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
777 				return (ENOEXEC);
778 			interp = imgp->image_header + phdr[i].p_offset;
779 			interp_name_len = phdr[i].p_filesz;
780 			break;
781 		case PT_GNU_STACK:
782 			if (__elfN(nxstack))
783 				imgp->stack_prot =
784 				    __elfN(trans_prot)(phdr[i].p_flags);
785 			imgp->stack_sz = phdr[i].p_memsz;
786 			break;
787 		}
788 	}
789 
790 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
791 	    &osrel);
792 	if (brand_info == NULL) {
793 		uprintf("ELF binary type \"%u\" not known.\n",
794 		    hdr->e_ident[EI_OSABI]);
795 		return (ENOEXEC);
796 	}
797 	if (hdr->e_type == ET_DYN) {
798 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
799 			return (ENOEXEC);
800 		/*
801 		 * Honour the base load address from the dso if it is
802 		 * non-zero for some reason.
803 		 */
804 		if (baddr == 0)
805 			et_dyn_addr = ET_DYN_LOAD_ADDR;
806 		else
807 			et_dyn_addr = 0;
808 	} else
809 		et_dyn_addr = 0;
810 	sv = brand_info->sysvec;
811 	if (interp != NULL && brand_info->interp_newpath != NULL)
812 		newinterp = brand_info->interp_newpath;
813 
814 	/*
815 	 * Avoid a possible deadlock if the current address space is destroyed
816 	 * and that address space maps the locked vnode.  In the common case,
817 	 * the locked vnode's v_usecount is decremented but remains greater
818 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
819 	 * However, in cases where the vnode lock is external, such as nullfs,
820 	 * v_usecount may become zero.
821 	 *
822 	 * The VV_TEXT flag prevents modifications to the executable while
823 	 * the vnode is unlocked.
824 	 */
825 	VOP_UNLOCK(imgp->vp, 0);
826 
827 	error = exec_new_vmspace(imgp, sv);
828 	imgp->proc->p_sysent = sv;
829 
830 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
831 	if (error)
832 		return (error);
833 
834 	for (i = 0; i < hdr->e_phnum; i++) {
835 		switch (phdr[i].p_type) {
836 		case PT_LOAD:	/* Loadable segment */
837 			if (phdr[i].p_memsz == 0)
838 				break;
839 			prot = __elfN(trans_prot)(phdr[i].p_flags);
840 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
841 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
842 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
843 			    sv->sv_pagesize);
844 			if (error != 0)
845 				return (error);
846 
847 			/*
848 			 * If this segment contains the program headers,
849 			 * remember their virtual address for the AT_PHDR
850 			 * aux entry. Static binaries don't usually include
851 			 * a PT_PHDR entry.
852 			 */
853 			if (phdr[i].p_offset == 0 &&
854 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
855 				<= phdr[i].p_filesz)
856 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
857 				    et_dyn_addr;
858 
859 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
860 			seg_size = round_page(phdr[i].p_memsz +
861 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
862 
863 			/*
864 			 * Make the largest executable segment the official
865 			 * text segment and all others data.
866 			 *
867 			 * Note that obreak() assumes that data_addr +
868 			 * data_size == end of data load area, and the ELF
869 			 * file format expects segments to be sorted by
870 			 * address.  If multiple data segments exist, the
871 			 * last one will be used.
872 			 */
873 
874 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
875 				text_size = seg_size;
876 				text_addr = seg_addr;
877 			} else {
878 				data_size = seg_size;
879 				data_addr = seg_addr;
880 			}
881 			total_size += seg_size;
882 			break;
883 		case PT_PHDR: 	/* Program header table info */
884 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
885 			break;
886 		default:
887 			break;
888 		}
889 	}
890 
891 	if (data_addr == 0 && data_size == 0) {
892 		data_addr = text_addr;
893 		data_size = text_size;
894 	}
895 
896 	entry = (u_long)hdr->e_entry + et_dyn_addr;
897 
898 	/*
899 	 * Check limits.  It should be safe to check the
900 	 * limits after loading the segments since we do
901 	 * not actually fault in all the segments pages.
902 	 */
903 	PROC_LOCK(imgp->proc);
904 	if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
905 	    text_size > maxtsiz ||
906 	    total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
907 	    racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
908 	    racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
909 		PROC_UNLOCK(imgp->proc);
910 		return (ENOMEM);
911 	}
912 
913 	vmspace = imgp->proc->p_vmspace;
914 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
915 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
916 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
917 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
918 
919 	/*
920 	 * We load the dynamic linker where a userland call
921 	 * to mmap(0, ...) would put it.  The rationale behind this
922 	 * calculation is that it leaves room for the heap to grow to
923 	 * its maximum allowed size.
924 	 */
925 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
926 	    RLIMIT_DATA));
927 	PROC_UNLOCK(imgp->proc);
928 
929 	imgp->entry_addr = entry;
930 
931 	if (interp != NULL) {
932 		int have_interp = FALSE;
933 		VOP_UNLOCK(imgp->vp, 0);
934 		if (brand_info->emul_path != NULL &&
935 		    brand_info->emul_path[0] != '\0') {
936 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
937 			snprintf(path, MAXPATHLEN, "%s%s",
938 			    brand_info->emul_path, interp);
939 			error = __elfN(load_file)(imgp->proc, path, &addr,
940 			    &imgp->entry_addr, sv->sv_pagesize);
941 			free(path, M_TEMP);
942 			if (error == 0)
943 				have_interp = TRUE;
944 		}
945 		if (!have_interp && newinterp != NULL) {
946 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
947 			    &imgp->entry_addr, sv->sv_pagesize);
948 			if (error == 0)
949 				have_interp = TRUE;
950 		}
951 		if (!have_interp) {
952 			error = __elfN(load_file)(imgp->proc, interp, &addr,
953 			    &imgp->entry_addr, sv->sv_pagesize);
954 		}
955 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
956 		if (error != 0) {
957 			uprintf("ELF interpreter %s not found\n", interp);
958 			return (error);
959 		}
960 	} else
961 		addr = et_dyn_addr;
962 
963 	/*
964 	 * Construct auxargs table (used by the fixup routine)
965 	 */
966 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
967 	elf_auxargs->execfd = -1;
968 	elf_auxargs->phdr = proghdr;
969 	elf_auxargs->phent = hdr->e_phentsize;
970 	elf_auxargs->phnum = hdr->e_phnum;
971 	elf_auxargs->pagesz = PAGE_SIZE;
972 	elf_auxargs->base = addr;
973 	elf_auxargs->flags = 0;
974 	elf_auxargs->entry = entry;
975 
976 	imgp->auxargs = elf_auxargs;
977 	imgp->interpreted = 0;
978 	imgp->reloc_base = addr;
979 	imgp->proc->p_osrel = osrel;
980 
981 	return (error);
982 }
983 
984 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
985 
986 int
987 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
988 {
989 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
990 	Elf_Addr *base;
991 	Elf_Addr *pos;
992 
993 	base = (Elf_Addr *)*stack_base;
994 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
995 
996 	if (args->execfd != -1)
997 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
998 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
999 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1000 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1001 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1002 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1003 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1004 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1005 	if (imgp->execpathp != 0)
1006 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1007 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1008 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1009 	if (imgp->canary != 0) {
1010 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1011 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1012 	}
1013 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1014 	if (imgp->pagesizes != 0) {
1015 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1016 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1017 	}
1018 	if (imgp->sysent->sv_timekeep_base != 0) {
1019 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1020 		    imgp->sysent->sv_timekeep_base);
1021 	}
1022 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1023 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1024 	    imgp->sysent->sv_stackprot);
1025 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1026 
1027 	free(imgp->auxargs, M_TEMP);
1028 	imgp->auxargs = NULL;
1029 
1030 	base--;
1031 	suword(base, (long)imgp->args->argc);
1032 	*stack_base = (register_t *)base;
1033 	return (0);
1034 }
1035 
1036 /*
1037  * Code for generating ELF core dumps.
1038  */
1039 
1040 typedef void (*segment_callback)(vm_map_entry_t, void *);
1041 
1042 /* Closure for cb_put_phdr(). */
1043 struct phdr_closure {
1044 	Elf_Phdr *phdr;		/* Program header to fill in */
1045 	Elf_Off offset;		/* Offset of segment in core file */
1046 };
1047 
1048 /* Closure for cb_size_segment(). */
1049 struct sseg_closure {
1050 	int count;		/* Count of writable segments. */
1051 	size_t size;		/* Total size of all writable segments. */
1052 };
1053 
1054 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1055 
1056 struct note_info {
1057 	int		type;		/* Note type. */
1058 	outfunc_t 	outfunc; 	/* Output function. */
1059 	void		*outarg;	/* Argument for the output function. */
1060 	size_t		outsize;	/* Output size. */
1061 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1062 };
1063 
1064 TAILQ_HEAD(note_info_list, note_info);
1065 
1066 /* Coredump output parameters. */
1067 struct coredump_params {
1068 	off_t		offset;
1069 	struct ucred	*active_cred;
1070 	struct ucred	*file_cred;
1071 	struct thread	*td;
1072 	struct vnode	*vp;
1073 	struct gzio_stream *gzs;
1074 };
1075 
1076 static void cb_put_phdr(vm_map_entry_t, void *);
1077 static void cb_size_segment(vm_map_entry_t, void *);
1078 static int core_write(struct coredump_params *, void *, size_t, off_t,
1079     enum uio_seg);
1080 static void each_writable_segment(struct thread *, segment_callback, void *);
1081 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1082     struct note_info_list *, size_t);
1083 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1084     size_t *);
1085 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1086 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1087 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1088 static int sbuf_drain_core_output(void *, const char *, int);
1089 static int sbuf_drain_count(void *arg, const char *data, int len);
1090 
1091 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1092 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1093 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1094 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1095 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1096 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1097 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1098 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1099 static void note_procstat_files(void *, struct sbuf *, size_t *);
1100 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1101 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1102 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1103 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1104 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1105 
1106 #ifdef GZIO
1107 extern int compress_user_cores_gzlevel;
1108 
1109 /*
1110  * Write out a core segment to the compression stream.
1111  */
1112 static int
1113 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1114 {
1115 	u_int chunk_len;
1116 	int error;
1117 
1118 	while (len > 0) {
1119 		chunk_len = MIN(len, CORE_BUF_SIZE);
1120 		copyin(base, buf, chunk_len);
1121 		error = gzio_write(p->gzs, buf, chunk_len);
1122 		if (error != 0)
1123 			break;
1124 		base += chunk_len;
1125 		len -= chunk_len;
1126 	}
1127 	return (error);
1128 }
1129 
1130 static int
1131 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1132 {
1133 
1134 	return (core_write((struct coredump_params *)arg, base, len, offset,
1135 	    UIO_SYSSPACE));
1136 }
1137 #endif /* GZIO */
1138 
1139 static int
1140 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1141     enum uio_seg seg)
1142 {
1143 
1144 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1145 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1146 	    p->active_cred, p->file_cred, NULL, p->td));
1147 }
1148 
1149 static int
1150 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1151     void *tmpbuf)
1152 {
1153 
1154 #ifdef GZIO
1155 	if (p->gzs != NULL)
1156 		return (compress_chunk(p, base, tmpbuf, len));
1157 #endif
1158 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1159 }
1160 
1161 /*
1162  * Drain into a core file.
1163  */
1164 static int
1165 sbuf_drain_core_output(void *arg, const char *data, int len)
1166 {
1167 	struct coredump_params *p;
1168 	int error, locked;
1169 
1170 	p = (struct coredump_params *)arg;
1171 
1172 	/*
1173 	 * Some kern_proc out routines that print to this sbuf may
1174 	 * call us with the process lock held. Draining with the
1175 	 * non-sleepable lock held is unsafe. The lock is needed for
1176 	 * those routines when dumping a live process. In our case we
1177 	 * can safely release the lock before draining and acquire
1178 	 * again after.
1179 	 */
1180 	locked = PROC_LOCKED(p->td->td_proc);
1181 	if (locked)
1182 		PROC_UNLOCK(p->td->td_proc);
1183 #ifdef GZIO
1184 	if (p->gzs != NULL)
1185 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1186 	else
1187 #endif
1188 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1189 		    UIO_SYSSPACE);
1190 	if (locked)
1191 		PROC_LOCK(p->td->td_proc);
1192 	if (error != 0)
1193 		return (-error);
1194 	p->offset += len;
1195 	return (len);
1196 }
1197 
1198 /*
1199  * Drain into a counter.
1200  */
1201 static int
1202 sbuf_drain_count(void *arg, const char *data __unused, int len)
1203 {
1204 	size_t *sizep;
1205 
1206 	sizep = (size_t *)arg;
1207 	*sizep += len;
1208 	return (len);
1209 }
1210 
1211 int
1212 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1213 {
1214 	struct ucred *cred = td->td_ucred;
1215 	int error = 0;
1216 	struct sseg_closure seginfo;
1217 	struct note_info_list notelst;
1218 	struct coredump_params params;
1219 	struct note_info *ninfo;
1220 	void *hdr, *tmpbuf;
1221 	size_t hdrsize, notesz, coresize;
1222 	boolean_t compress;
1223 
1224 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1225 	hdr = NULL;
1226 	TAILQ_INIT(&notelst);
1227 
1228 	/* Size the program segments. */
1229 	seginfo.count = 0;
1230 	seginfo.size = 0;
1231 	each_writable_segment(td, cb_size_segment, &seginfo);
1232 
1233 	/*
1234 	 * Collect info about the core file header area.
1235 	 */
1236 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1237 	__elfN(prepare_notes)(td, &notelst, &notesz);
1238 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1239 
1240 #ifdef RACCT
1241 	if (racct_enable) {
1242 		PROC_LOCK(td->td_proc);
1243 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1244 		PROC_UNLOCK(td->td_proc);
1245 		if (error != 0) {
1246 			error = EFAULT;
1247 			goto done;
1248 		}
1249 	}
1250 #endif
1251 	if (coresize >= limit) {
1252 		error = EFAULT;
1253 		goto done;
1254 	}
1255 
1256 	/* Set up core dump parameters. */
1257 	params.offset = 0;
1258 	params.active_cred = cred;
1259 	params.file_cred = NOCRED;
1260 	params.td = td;
1261 	params.vp = vp;
1262 	params.gzs = NULL;
1263 
1264 	tmpbuf = NULL;
1265 #ifdef GZIO
1266 	/* Create a compression stream if necessary. */
1267 	if (compress) {
1268 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1269 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1270 		if (params.gzs == NULL) {
1271 			error = EFAULT;
1272 			goto done;
1273 		}
1274 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1275         }
1276 #endif
1277 
1278 	/*
1279 	 * Allocate memory for building the header, fill it up,
1280 	 * and write it out following the notes.
1281 	 */
1282 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1283 	if (hdr == NULL) {
1284 		error = EINVAL;
1285 		goto done;
1286 	}
1287 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1288 	    notesz);
1289 
1290 	/* Write the contents of all of the writable segments. */
1291 	if (error == 0) {
1292 		Elf_Phdr *php;
1293 		off_t offset;
1294 		int i;
1295 
1296 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1297 		offset = round_page(hdrsize + notesz);
1298 		for (i = 0; i < seginfo.count; i++) {
1299 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1300 			    php->p_filesz, offset, &params, tmpbuf);
1301 			if (error != 0)
1302 				break;
1303 			offset += php->p_filesz;
1304 			php++;
1305 		}
1306 #ifdef GZIO
1307 		if (error == 0 && compress)
1308 			error = gzio_flush(params.gzs);
1309 #endif
1310 	}
1311 	if (error) {
1312 		log(LOG_WARNING,
1313 		    "Failed to write core file for process %s (error %d)\n",
1314 		    curproc->p_comm, error);
1315 	}
1316 
1317 done:
1318 #ifdef GZIO
1319 	if (compress) {
1320 		free(tmpbuf, M_TEMP);
1321 		gzio_fini(params.gzs);
1322 	}
1323 #endif
1324 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1325 		TAILQ_REMOVE(&notelst, ninfo, link);
1326 		free(ninfo, M_TEMP);
1327 	}
1328 	if (hdr != NULL)
1329 		free(hdr, M_TEMP);
1330 
1331 	return (error);
1332 }
1333 
1334 /*
1335  * A callback for each_writable_segment() to write out the segment's
1336  * program header entry.
1337  */
1338 static void
1339 cb_put_phdr(entry, closure)
1340 	vm_map_entry_t entry;
1341 	void *closure;
1342 {
1343 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1344 	Elf_Phdr *phdr = phc->phdr;
1345 
1346 	phc->offset = round_page(phc->offset);
1347 
1348 	phdr->p_type = PT_LOAD;
1349 	phdr->p_offset = phc->offset;
1350 	phdr->p_vaddr = entry->start;
1351 	phdr->p_paddr = 0;
1352 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1353 	phdr->p_align = PAGE_SIZE;
1354 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1355 
1356 	phc->offset += phdr->p_filesz;
1357 	phc->phdr++;
1358 }
1359 
1360 /*
1361  * A callback for each_writable_segment() to gather information about
1362  * the number of segments and their total size.
1363  */
1364 static void
1365 cb_size_segment(entry, closure)
1366 	vm_map_entry_t entry;
1367 	void *closure;
1368 {
1369 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1370 
1371 	ssc->count++;
1372 	ssc->size += entry->end - entry->start;
1373 }
1374 
1375 /*
1376  * For each writable segment in the process's memory map, call the given
1377  * function with a pointer to the map entry and some arbitrary
1378  * caller-supplied data.
1379  */
1380 static void
1381 each_writable_segment(td, func, closure)
1382 	struct thread *td;
1383 	segment_callback func;
1384 	void *closure;
1385 {
1386 	struct proc *p = td->td_proc;
1387 	vm_map_t map = &p->p_vmspace->vm_map;
1388 	vm_map_entry_t entry;
1389 	vm_object_t backing_object, object;
1390 	boolean_t ignore_entry;
1391 
1392 	vm_map_lock_read(map);
1393 	for (entry = map->header.next; entry != &map->header;
1394 	    entry = entry->next) {
1395 		/*
1396 		 * Don't dump inaccessible mappings, deal with legacy
1397 		 * coredump mode.
1398 		 *
1399 		 * Note that read-only segments related to the elf binary
1400 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1401 		 * need to arbitrarily ignore such segments.
1402 		 */
1403 		if (elf_legacy_coredump) {
1404 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1405 				continue;
1406 		} else {
1407 			if ((entry->protection & VM_PROT_ALL) == 0)
1408 				continue;
1409 		}
1410 
1411 		/*
1412 		 * Dont include memory segment in the coredump if
1413 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1414 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1415 		 * kernel map).
1416 		 */
1417 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1418 			continue;
1419 
1420 		if ((object = entry->object.vm_object) == NULL)
1421 			continue;
1422 
1423 		/* Ignore memory-mapped devices and such things. */
1424 		VM_OBJECT_RLOCK(object);
1425 		while ((backing_object = object->backing_object) != NULL) {
1426 			VM_OBJECT_RLOCK(backing_object);
1427 			VM_OBJECT_RUNLOCK(object);
1428 			object = backing_object;
1429 		}
1430 		ignore_entry = object->type != OBJT_DEFAULT &&
1431 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1432 		    object->type != OBJT_PHYS;
1433 		VM_OBJECT_RUNLOCK(object);
1434 		if (ignore_entry)
1435 			continue;
1436 
1437 		(*func)(entry, closure);
1438 	}
1439 	vm_map_unlock_read(map);
1440 }
1441 
1442 /*
1443  * Write the core file header to the file, including padding up to
1444  * the page boundary.
1445  */
1446 static int
1447 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1448     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1449 {
1450 	struct note_info *ninfo;
1451 	struct sbuf *sb;
1452 	int error;
1453 
1454 	/* Fill in the header. */
1455 	bzero(hdr, hdrsize);
1456 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1457 
1458 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1459 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1460 	sbuf_start_section(sb, NULL);
1461 	sbuf_bcat(sb, hdr, hdrsize);
1462 	TAILQ_FOREACH(ninfo, notelst, link)
1463 	    __elfN(putnote)(ninfo, sb);
1464 	/* Align up to a page boundary for the program segments. */
1465 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1466 	error = sbuf_finish(sb);
1467 	sbuf_delete(sb);
1468 
1469 	return (error);
1470 }
1471 
1472 static void
1473 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1474     size_t *sizep)
1475 {
1476 	struct proc *p;
1477 	struct thread *thr;
1478 	size_t size;
1479 
1480 	p = td->td_proc;
1481 	size = 0;
1482 
1483 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1484 
1485 	/*
1486 	 * To have the debugger select the right thread (LWP) as the initial
1487 	 * thread, we dump the state of the thread passed to us in td first.
1488 	 * This is the thread that causes the core dump and thus likely to
1489 	 * be the right thread one wants to have selected in the debugger.
1490 	 */
1491 	thr = td;
1492 	while (thr != NULL) {
1493 		size += register_note(list, NT_PRSTATUS,
1494 		    __elfN(note_prstatus), thr);
1495 		size += register_note(list, NT_FPREGSET,
1496 		    __elfN(note_fpregset), thr);
1497 		size += register_note(list, NT_THRMISC,
1498 		    __elfN(note_thrmisc), thr);
1499 		size += register_note(list, -1,
1500 		    __elfN(note_threadmd), thr);
1501 
1502 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1503 		    TAILQ_NEXT(thr, td_plist);
1504 		if (thr == td)
1505 			thr = TAILQ_NEXT(thr, td_plist);
1506 	}
1507 
1508 	size += register_note(list, NT_PROCSTAT_PROC,
1509 	    __elfN(note_procstat_proc), p);
1510 	size += register_note(list, NT_PROCSTAT_FILES,
1511 	    note_procstat_files, p);
1512 	size += register_note(list, NT_PROCSTAT_VMMAP,
1513 	    note_procstat_vmmap, p);
1514 	size += register_note(list, NT_PROCSTAT_GROUPS,
1515 	    note_procstat_groups, p);
1516 	size += register_note(list, NT_PROCSTAT_UMASK,
1517 	    note_procstat_umask, p);
1518 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1519 	    note_procstat_rlimit, p);
1520 	size += register_note(list, NT_PROCSTAT_OSREL,
1521 	    note_procstat_osrel, p);
1522 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1523 	    __elfN(note_procstat_psstrings), p);
1524 	size += register_note(list, NT_PROCSTAT_AUXV,
1525 	    __elfN(note_procstat_auxv), p);
1526 
1527 	*sizep = size;
1528 }
1529 
1530 static void
1531 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1532     size_t notesz)
1533 {
1534 	Elf_Ehdr *ehdr;
1535 	Elf_Phdr *phdr;
1536 	struct phdr_closure phc;
1537 
1538 	ehdr = (Elf_Ehdr *)hdr;
1539 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1540 
1541 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1542 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1543 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1544 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1545 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1546 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1547 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1548 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1549 	ehdr->e_ident[EI_ABIVERSION] = 0;
1550 	ehdr->e_ident[EI_PAD] = 0;
1551 	ehdr->e_type = ET_CORE;
1552 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1553 	ehdr->e_machine = ELF_ARCH32;
1554 #else
1555 	ehdr->e_machine = ELF_ARCH;
1556 #endif
1557 	ehdr->e_version = EV_CURRENT;
1558 	ehdr->e_entry = 0;
1559 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1560 	ehdr->e_flags = 0;
1561 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1562 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1563 	ehdr->e_phnum = numsegs + 1;
1564 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1565 	ehdr->e_shnum = 0;
1566 	ehdr->e_shstrndx = SHN_UNDEF;
1567 
1568 	/*
1569 	 * Fill in the program header entries.
1570 	 */
1571 
1572 	/* The note segement. */
1573 	phdr->p_type = PT_NOTE;
1574 	phdr->p_offset = hdrsize;
1575 	phdr->p_vaddr = 0;
1576 	phdr->p_paddr = 0;
1577 	phdr->p_filesz = notesz;
1578 	phdr->p_memsz = 0;
1579 	phdr->p_flags = PF_R;
1580 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1581 	phdr++;
1582 
1583 	/* All the writable segments from the program. */
1584 	phc.phdr = phdr;
1585 	phc.offset = round_page(hdrsize + notesz);
1586 	each_writable_segment(td, cb_put_phdr, &phc);
1587 }
1588 
1589 static size_t
1590 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1591 {
1592 	struct note_info *ninfo;
1593 	size_t size, notesize;
1594 
1595 	size = 0;
1596 	out(arg, NULL, &size);
1597 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1598 	ninfo->type = type;
1599 	ninfo->outfunc = out;
1600 	ninfo->outarg = arg;
1601 	ninfo->outsize = size;
1602 	TAILQ_INSERT_TAIL(list, ninfo, link);
1603 
1604 	if (type == -1)
1605 		return (size);
1606 
1607 	notesize = sizeof(Elf_Note) +		/* note header */
1608 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1609 						/* note name */
1610 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1611 
1612 	return (notesize);
1613 }
1614 
1615 static size_t
1616 append_note_data(const void *src, void *dst, size_t len)
1617 {
1618 	size_t padded_len;
1619 
1620 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1621 	if (dst != NULL) {
1622 		bcopy(src, dst, len);
1623 		bzero((char *)dst + len, padded_len - len);
1624 	}
1625 	return (padded_len);
1626 }
1627 
1628 size_t
1629 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1630 {
1631 	Elf_Note *note;
1632 	char *buf;
1633 	size_t notesize;
1634 
1635 	buf = dst;
1636 	if (buf != NULL) {
1637 		note = (Elf_Note *)buf;
1638 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1639 		note->n_descsz = size;
1640 		note->n_type = type;
1641 		buf += sizeof(*note);
1642 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1643 		    sizeof(FREEBSD_ABI_VENDOR));
1644 		append_note_data(src, buf, size);
1645 		if (descp != NULL)
1646 			*descp = buf;
1647 	}
1648 
1649 	notesize = sizeof(Elf_Note) +		/* note header */
1650 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1651 						/* note name */
1652 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1653 
1654 	return (notesize);
1655 }
1656 
1657 static void
1658 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1659 {
1660 	Elf_Note note;
1661 	ssize_t old_len;
1662 
1663 	if (ninfo->type == -1) {
1664 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1665 		return;
1666 	}
1667 
1668 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1669 	note.n_descsz = ninfo->outsize;
1670 	note.n_type = ninfo->type;
1671 
1672 	sbuf_bcat(sb, &note, sizeof(note));
1673 	sbuf_start_section(sb, &old_len);
1674 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1675 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1676 	if (note.n_descsz == 0)
1677 		return;
1678 	sbuf_start_section(sb, &old_len);
1679 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1680 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1681 }
1682 
1683 /*
1684  * Miscellaneous note out functions.
1685  */
1686 
1687 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1688 #include <compat/freebsd32/freebsd32.h>
1689 
1690 typedef struct prstatus32 elf_prstatus_t;
1691 typedef struct prpsinfo32 elf_prpsinfo_t;
1692 typedef struct fpreg32 elf_prfpregset_t;
1693 typedef struct fpreg32 elf_fpregset_t;
1694 typedef struct reg32 elf_gregset_t;
1695 typedef struct thrmisc32 elf_thrmisc_t;
1696 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1697 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1698 typedef uint32_t elf_ps_strings_t;
1699 #else
1700 typedef prstatus_t elf_prstatus_t;
1701 typedef prpsinfo_t elf_prpsinfo_t;
1702 typedef prfpregset_t elf_prfpregset_t;
1703 typedef prfpregset_t elf_fpregset_t;
1704 typedef gregset_t elf_gregset_t;
1705 typedef thrmisc_t elf_thrmisc_t;
1706 #define ELF_KERN_PROC_MASK	0
1707 typedef struct kinfo_proc elf_kinfo_proc_t;
1708 typedef vm_offset_t elf_ps_strings_t;
1709 #endif
1710 
1711 static void
1712 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1713 {
1714 	struct proc *p;
1715 	elf_prpsinfo_t *psinfo;
1716 
1717 	p = (struct proc *)arg;
1718 	if (sb != NULL) {
1719 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1720 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1721 		psinfo->pr_version = PRPSINFO_VERSION;
1722 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1723 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1724 		/*
1725 		 * XXX - We don't fill in the command line arguments properly
1726 		 * yet.
1727 		 */
1728 		strlcpy(psinfo->pr_psargs, p->p_comm,
1729 		    sizeof(psinfo->pr_psargs));
1730 
1731 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1732 		free(psinfo, M_TEMP);
1733 	}
1734 	*sizep = sizeof(*psinfo);
1735 }
1736 
1737 static void
1738 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1739 {
1740 	struct thread *td;
1741 	elf_prstatus_t *status;
1742 
1743 	td = (struct thread *)arg;
1744 	if (sb != NULL) {
1745 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1746 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1747 		status->pr_version = PRSTATUS_VERSION;
1748 		status->pr_statussz = sizeof(elf_prstatus_t);
1749 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1750 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1751 		status->pr_osreldate = osreldate;
1752 		status->pr_cursig = td->td_proc->p_sig;
1753 		status->pr_pid = td->td_tid;
1754 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1755 		fill_regs32(td, &status->pr_reg);
1756 #else
1757 		fill_regs(td, &status->pr_reg);
1758 #endif
1759 		sbuf_bcat(sb, status, sizeof(*status));
1760 		free(status, M_TEMP);
1761 	}
1762 	*sizep = sizeof(*status);
1763 }
1764 
1765 static void
1766 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1767 {
1768 	struct thread *td;
1769 	elf_prfpregset_t *fpregset;
1770 
1771 	td = (struct thread *)arg;
1772 	if (sb != NULL) {
1773 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1774 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1775 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1776 		fill_fpregs32(td, fpregset);
1777 #else
1778 		fill_fpregs(td, fpregset);
1779 #endif
1780 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1781 		free(fpregset, M_TEMP);
1782 	}
1783 	*sizep = sizeof(*fpregset);
1784 }
1785 
1786 static void
1787 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1788 {
1789 	struct thread *td;
1790 	elf_thrmisc_t thrmisc;
1791 
1792 	td = (struct thread *)arg;
1793 	if (sb != NULL) {
1794 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1795 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1796 		strcpy(thrmisc.pr_tname, td->td_name);
1797 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1798 	}
1799 	*sizep = sizeof(thrmisc);
1800 }
1801 
1802 /*
1803  * Allow for MD specific notes, as well as any MD
1804  * specific preparations for writing MI notes.
1805  */
1806 static void
1807 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1808 {
1809 	struct thread *td;
1810 	void *buf;
1811 	size_t size;
1812 
1813 	td = (struct thread *)arg;
1814 	size = *sizep;
1815 	if (size != 0 && sb != NULL)
1816 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1817 	else
1818 		buf = NULL;
1819 	size = 0;
1820 	__elfN(dump_thread)(td, buf, &size);
1821 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1822 	if (size != 0 && sb != NULL)
1823 		sbuf_bcat(sb, buf, size);
1824 	free(buf, M_TEMP);
1825 	*sizep = size;
1826 }
1827 
1828 #ifdef KINFO_PROC_SIZE
1829 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1830 #endif
1831 
1832 static void
1833 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1834 {
1835 	struct proc *p;
1836 	size_t size;
1837 	int structsize;
1838 
1839 	p = (struct proc *)arg;
1840 	size = sizeof(structsize) + p->p_numthreads *
1841 	    sizeof(elf_kinfo_proc_t);
1842 
1843 	if (sb != NULL) {
1844 		KASSERT(*sizep == size, ("invalid size"));
1845 		structsize = sizeof(elf_kinfo_proc_t);
1846 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1847 		sx_slock(&proctree_lock);
1848 		PROC_LOCK(p);
1849 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1850 		sx_sunlock(&proctree_lock);
1851 	}
1852 	*sizep = size;
1853 }
1854 
1855 #ifdef KINFO_FILE_SIZE
1856 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1857 #endif
1858 
1859 static void
1860 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1861 {
1862 	struct proc *p;
1863 	size_t size;
1864 	int structsize;
1865 
1866 	p = (struct proc *)arg;
1867 	if (sb == NULL) {
1868 		size = 0;
1869 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1870 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1871 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1872 		PROC_LOCK(p);
1873 		kern_proc_filedesc_out(p, sb, -1);
1874 		sbuf_finish(sb);
1875 		sbuf_delete(sb);
1876 		*sizep = size;
1877 	} else {
1878 		structsize = sizeof(struct kinfo_file);
1879 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1880 		PROC_LOCK(p);
1881 		kern_proc_filedesc_out(p, sb, -1);
1882 	}
1883 }
1884 
1885 #ifdef KINFO_VMENTRY_SIZE
1886 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1887 #endif
1888 
1889 static void
1890 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1891 {
1892 	struct proc *p;
1893 	size_t size;
1894 	int structsize;
1895 
1896 	p = (struct proc *)arg;
1897 	if (sb == NULL) {
1898 		size = 0;
1899 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1900 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1901 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1902 		PROC_LOCK(p);
1903 		kern_proc_vmmap_out(p, sb);
1904 		sbuf_finish(sb);
1905 		sbuf_delete(sb);
1906 		*sizep = size;
1907 	} else {
1908 		structsize = sizeof(struct kinfo_vmentry);
1909 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1910 		PROC_LOCK(p);
1911 		kern_proc_vmmap_out(p, sb);
1912 	}
1913 }
1914 
1915 static void
1916 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1917 {
1918 	struct proc *p;
1919 	size_t size;
1920 	int structsize;
1921 
1922 	p = (struct proc *)arg;
1923 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1924 	if (sb != NULL) {
1925 		KASSERT(*sizep == size, ("invalid size"));
1926 		structsize = sizeof(gid_t);
1927 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1928 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1929 		    sizeof(gid_t));
1930 	}
1931 	*sizep = size;
1932 }
1933 
1934 static void
1935 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1936 {
1937 	struct proc *p;
1938 	size_t size;
1939 	int structsize;
1940 
1941 	p = (struct proc *)arg;
1942 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1943 	if (sb != NULL) {
1944 		KASSERT(*sizep == size, ("invalid size"));
1945 		structsize = sizeof(p->p_fd->fd_cmask);
1946 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1947 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1948 	}
1949 	*sizep = size;
1950 }
1951 
1952 static void
1953 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1954 {
1955 	struct proc *p;
1956 	struct rlimit rlim[RLIM_NLIMITS];
1957 	size_t size;
1958 	int structsize, i;
1959 
1960 	p = (struct proc *)arg;
1961 	size = sizeof(structsize) + sizeof(rlim);
1962 	if (sb != NULL) {
1963 		KASSERT(*sizep == size, ("invalid size"));
1964 		structsize = sizeof(rlim);
1965 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1966 		PROC_LOCK(p);
1967 		for (i = 0; i < RLIM_NLIMITS; i++)
1968 			lim_rlimit(p, i, &rlim[i]);
1969 		PROC_UNLOCK(p);
1970 		sbuf_bcat(sb, rlim, sizeof(rlim));
1971 	}
1972 	*sizep = size;
1973 }
1974 
1975 static void
1976 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1977 {
1978 	struct proc *p;
1979 	size_t size;
1980 	int structsize;
1981 
1982 	p = (struct proc *)arg;
1983 	size = sizeof(structsize) + sizeof(p->p_osrel);
1984 	if (sb != NULL) {
1985 		KASSERT(*sizep == size, ("invalid size"));
1986 		structsize = sizeof(p->p_osrel);
1987 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1988 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1989 	}
1990 	*sizep = size;
1991 }
1992 
1993 static void
1994 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1995 {
1996 	struct proc *p;
1997 	elf_ps_strings_t ps_strings;
1998 	size_t size;
1999 	int structsize;
2000 
2001 	p = (struct proc *)arg;
2002 	size = sizeof(structsize) + sizeof(ps_strings);
2003 	if (sb != NULL) {
2004 		KASSERT(*sizep == size, ("invalid size"));
2005 		structsize = sizeof(ps_strings);
2006 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2007 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2008 #else
2009 		ps_strings = p->p_sysent->sv_psstrings;
2010 #endif
2011 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2012 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2013 	}
2014 	*sizep = size;
2015 }
2016 
2017 static void
2018 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2019 {
2020 	struct proc *p;
2021 	size_t size;
2022 	int structsize;
2023 
2024 	p = (struct proc *)arg;
2025 	if (sb == NULL) {
2026 		size = 0;
2027 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2028 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2029 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2030 		PHOLD(p);
2031 		proc_getauxv(curthread, p, sb);
2032 		PRELE(p);
2033 		sbuf_finish(sb);
2034 		sbuf_delete(sb);
2035 		*sizep = size;
2036 	} else {
2037 		structsize = sizeof(Elf_Auxinfo);
2038 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2039 		PHOLD(p);
2040 		proc_getauxv(curthread, p, sb);
2041 		PRELE(p);
2042 	}
2043 }
2044 
2045 static boolean_t
2046 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2047     int32_t *osrel, const Elf_Phdr *pnote)
2048 {
2049 	const Elf_Note *note, *note0, *note_end;
2050 	const char *note_name;
2051 	int i;
2052 
2053 	if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
2054 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
2055 		return (FALSE);
2056 
2057 	note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
2058 	note_end = (const Elf_Note *)(imgp->image_header +
2059 	    pnote->p_offset + pnote->p_filesz);
2060 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2061 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2062 		    (const char *)note < sizeof(Elf_Note))
2063 			return (FALSE);
2064 		if (note->n_namesz != checknote->hdr.n_namesz ||
2065 		    note->n_descsz != checknote->hdr.n_descsz ||
2066 		    note->n_type != checknote->hdr.n_type)
2067 			goto nextnote;
2068 		note_name = (const char *)(note + 1);
2069 		if (note_name + checknote->hdr.n_namesz >=
2070 		    (const char *)note_end || strncmp(checknote->vendor,
2071 		    note_name, checknote->hdr.n_namesz) != 0)
2072 			goto nextnote;
2073 
2074 		/*
2075 		 * Fetch the osreldate for binary
2076 		 * from the ELF OSABI-note if necessary.
2077 		 */
2078 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2079 		    checknote->trans_osrel != NULL)
2080 			return (checknote->trans_osrel(note, osrel));
2081 		return (TRUE);
2082 
2083 nextnote:
2084 		note = (const Elf_Note *)((const char *)(note + 1) +
2085 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2086 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2087 	}
2088 
2089 	return (FALSE);
2090 }
2091 
2092 /*
2093  * Try to find the appropriate ABI-note section for checknote,
2094  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2095  * first page of the image is searched, the same as for headers.
2096  */
2097 static boolean_t
2098 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2099     int32_t *osrel)
2100 {
2101 	const Elf_Phdr *phdr;
2102 	const Elf_Ehdr *hdr;
2103 	int i;
2104 
2105 	hdr = (const Elf_Ehdr *)imgp->image_header;
2106 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2107 
2108 	for (i = 0; i < hdr->e_phnum; i++) {
2109 		if (phdr[i].p_type == PT_NOTE &&
2110 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2111 			return (TRUE);
2112 	}
2113 	return (FALSE);
2114 
2115 }
2116 
2117 /*
2118  * Tell kern_execve.c about it, with a little help from the linker.
2119  */
2120 static struct execsw __elfN(execsw) = {
2121 	__CONCAT(exec_, __elfN(imgact)),
2122 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2123 };
2124 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2125 
2126 static vm_prot_t
2127 __elfN(trans_prot)(Elf_Word flags)
2128 {
2129 	vm_prot_t prot;
2130 
2131 	prot = 0;
2132 	if (flags & PF_X)
2133 		prot |= VM_PROT_EXECUTE;
2134 	if (flags & PF_W)
2135 		prot |= VM_PROT_WRITE;
2136 	if (flags & PF_R)
2137 		prot |= VM_PROT_READ;
2138 #if __ELF_WORD_SIZE == 32
2139 #if defined(__amd64__)
2140 	if (i386_read_exec && (flags & PF_R))
2141 		prot |= VM_PROT_EXECUTE;
2142 #endif
2143 #endif
2144 	return (prot);
2145 }
2146 
2147 static Elf_Word
2148 __elfN(untrans_prot)(vm_prot_t prot)
2149 {
2150 	Elf_Word flags;
2151 
2152 	flags = 0;
2153 	if (prot & VM_PROT_EXECUTE)
2154 		flags |= PF_X;
2155 	if (prot & VM_PROT_READ)
2156 		flags |= PF_R;
2157 	if (prot & VM_PROT_WRITE)
2158 		flags |= PF_W;
2159 	return (flags);
2160 }
2161