xref: /freebsd/sys/kern/imgact_elf.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
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 *err_str = NULL, *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 		uprintf("Program headers not in the first page\n");
759 		return (ENOEXEC);
760 	}
761 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
762 	if (!aligned(phdr, Elf_Addr)) {
763 		uprintf("Unaligned program headers\n");
764 		return (ENOEXEC);
765 	}
766 	n = 0;
767 	baddr = 0;
768 	for (i = 0; i < hdr->e_phnum; i++) {
769 		switch (phdr[i].p_type) {
770 		case PT_LOAD:
771 			if (n == 0)
772 				baddr = phdr[i].p_vaddr;
773 			n++;
774 			break;
775 		case PT_INTERP:
776 			/* Path to interpreter */
777 			if (phdr[i].p_filesz > MAXPATHLEN ||
778 			    phdr[i].p_offset > PAGE_SIZE ||
779 			    phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset) {
780 				uprintf("Invalid PT_INTERP\n");
781 				return (ENOEXEC);
782 			}
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 			imgp->stack_sz = phdr[i].p_memsz;
791 			break;
792 		}
793 	}
794 
795 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
796 	    &osrel);
797 	if (brand_info == NULL) {
798 		uprintf("ELF binary type \"%u\" not known.\n",
799 		    hdr->e_ident[EI_OSABI]);
800 		return (ENOEXEC);
801 	}
802 	if (hdr->e_type == ET_DYN) {
803 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
804 			uprintf("Cannot execute shared object\n");
805 			return (ENOEXEC);
806 		}
807 		/*
808 		 * Honour the base load address from the dso if it is
809 		 * non-zero for some reason.
810 		 */
811 		if (baddr == 0)
812 			et_dyn_addr = ET_DYN_LOAD_ADDR;
813 		else
814 			et_dyn_addr = 0;
815 	} else
816 		et_dyn_addr = 0;
817 	sv = brand_info->sysvec;
818 	if (interp != NULL && brand_info->interp_newpath != NULL)
819 		newinterp = brand_info->interp_newpath;
820 
821 	/*
822 	 * Avoid a possible deadlock if the current address space is destroyed
823 	 * and that address space maps the locked vnode.  In the common case,
824 	 * the locked vnode's v_usecount is decremented but remains greater
825 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
826 	 * However, in cases where the vnode lock is external, such as nullfs,
827 	 * v_usecount may become zero.
828 	 *
829 	 * The VV_TEXT flag prevents modifications to the executable while
830 	 * the vnode is unlocked.
831 	 */
832 	VOP_UNLOCK(imgp->vp, 0);
833 
834 	error = exec_new_vmspace(imgp, sv);
835 	imgp->proc->p_sysent = sv;
836 
837 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
838 	if (error)
839 		return (error);
840 
841 	for (i = 0; i < hdr->e_phnum; i++) {
842 		switch (phdr[i].p_type) {
843 		case PT_LOAD:	/* Loadable segment */
844 			if (phdr[i].p_memsz == 0)
845 				break;
846 			prot = __elfN(trans_prot)(phdr[i].p_flags);
847 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
848 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
849 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
850 			    sv->sv_pagesize);
851 			if (error != 0)
852 				return (error);
853 
854 			/*
855 			 * If this segment contains the program headers,
856 			 * remember their virtual address for the AT_PHDR
857 			 * aux entry. Static binaries don't usually include
858 			 * a PT_PHDR entry.
859 			 */
860 			if (phdr[i].p_offset == 0 &&
861 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
862 				<= phdr[i].p_filesz)
863 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
864 				    et_dyn_addr;
865 
866 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
867 			seg_size = round_page(phdr[i].p_memsz +
868 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
869 
870 			/*
871 			 * Make the largest executable segment the official
872 			 * text segment and all others data.
873 			 *
874 			 * Note that obreak() assumes that data_addr +
875 			 * data_size == end of data load area, and the ELF
876 			 * file format expects segments to be sorted by
877 			 * address.  If multiple data segments exist, the
878 			 * last one will be used.
879 			 */
880 
881 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
882 				text_size = seg_size;
883 				text_addr = seg_addr;
884 			} else {
885 				data_size = seg_size;
886 				data_addr = seg_addr;
887 			}
888 			total_size += seg_size;
889 			break;
890 		case PT_PHDR: 	/* Program header table info */
891 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
892 			break;
893 		default:
894 			break;
895 		}
896 	}
897 
898 	if (data_addr == 0 && data_size == 0) {
899 		data_addr = text_addr;
900 		data_size = text_size;
901 	}
902 
903 	entry = (u_long)hdr->e_entry + et_dyn_addr;
904 
905 	/*
906 	 * Check limits.  It should be safe to check the
907 	 * limits after loading the segments since we do
908 	 * not actually fault in all the segments pages.
909 	 */
910 	PROC_LOCK(imgp->proc);
911 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
912 		err_str = "Data segment size exceeds process limit";
913 	else if (text_size > maxtsiz)
914 		err_str = "Text segment size exceeds system limit";
915 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
916 		err_str = "Total segment size exceeds process limit";
917 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
918 		err_str = "Data segment size exceeds resource limit";
919 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
920 		err_str = "Total segment size exceeds resource limit";
921 	if (err_str != NULL) {
922 		PROC_UNLOCK(imgp->proc);
923 		uprintf("%s\n", err_str);
924 		return (ENOMEM);
925 	}
926 
927 	vmspace = imgp->proc->p_vmspace;
928 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
929 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
930 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
931 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
932 
933 	/*
934 	 * We load the dynamic linker where a userland call
935 	 * to mmap(0, ...) would put it.  The rationale behind this
936 	 * calculation is that it leaves room for the heap to grow to
937 	 * its maximum allowed size.
938 	 */
939 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(curthread,
940 	    RLIMIT_DATA));
941 	PROC_UNLOCK(imgp->proc);
942 
943 	imgp->entry_addr = entry;
944 
945 	if (interp != NULL) {
946 		int have_interp = FALSE;
947 		VOP_UNLOCK(imgp->vp, 0);
948 		if (brand_info->emul_path != NULL &&
949 		    brand_info->emul_path[0] != '\0') {
950 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
951 			snprintf(path, MAXPATHLEN, "%s%s",
952 			    brand_info->emul_path, interp);
953 			error = __elfN(load_file)(imgp->proc, path, &addr,
954 			    &imgp->entry_addr, sv->sv_pagesize);
955 			free(path, M_TEMP);
956 			if (error == 0)
957 				have_interp = TRUE;
958 		}
959 		if (!have_interp && newinterp != NULL) {
960 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
961 			    &imgp->entry_addr, sv->sv_pagesize);
962 			if (error == 0)
963 				have_interp = TRUE;
964 		}
965 		if (!have_interp) {
966 			error = __elfN(load_file)(imgp->proc, interp, &addr,
967 			    &imgp->entry_addr, sv->sv_pagesize);
968 		}
969 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
970 		if (error != 0) {
971 			uprintf("ELF interpreter %s not found\n", interp);
972 			return (error);
973 		}
974 	} else
975 		addr = et_dyn_addr;
976 
977 	/*
978 	 * Construct auxargs table (used by the fixup routine)
979 	 */
980 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
981 	elf_auxargs->execfd = -1;
982 	elf_auxargs->phdr = proghdr;
983 	elf_auxargs->phent = hdr->e_phentsize;
984 	elf_auxargs->phnum = hdr->e_phnum;
985 	elf_auxargs->pagesz = PAGE_SIZE;
986 	elf_auxargs->base = addr;
987 	elf_auxargs->flags = 0;
988 	elf_auxargs->entry = entry;
989 	elf_auxargs->hdr_eflags = hdr->e_flags;
990 
991 	imgp->auxargs = elf_auxargs;
992 	imgp->interpreted = 0;
993 	imgp->reloc_base = addr;
994 	imgp->proc->p_osrel = osrel;
995 
996 	return (error);
997 }
998 
999 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1000 
1001 int
1002 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1003 {
1004 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1005 	Elf_Addr *base;
1006 	Elf_Addr *pos;
1007 
1008 	base = (Elf_Addr *)*stack_base;
1009 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1010 
1011 	if (args->execfd != -1)
1012 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1013 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1014 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1015 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1016 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1017 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1018 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1019 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1020 #ifdef AT_EHDRFLAGS
1021 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1022 #endif
1023 	if (imgp->execpathp != 0)
1024 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1025 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1026 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1027 	if (imgp->canary != 0) {
1028 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1029 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1030 	}
1031 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1032 	if (imgp->pagesizes != 0) {
1033 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1034 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1035 	}
1036 	if (imgp->sysent->sv_timekeep_base != 0) {
1037 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1038 		    imgp->sysent->sv_timekeep_base);
1039 	}
1040 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1041 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1042 	    imgp->sysent->sv_stackprot);
1043 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1044 
1045 	free(imgp->auxargs, M_TEMP);
1046 	imgp->auxargs = NULL;
1047 
1048 	base--;
1049 	suword(base, (long)imgp->args->argc);
1050 	*stack_base = (register_t *)base;
1051 	return (0);
1052 }
1053 
1054 /*
1055  * Code for generating ELF core dumps.
1056  */
1057 
1058 typedef void (*segment_callback)(vm_map_entry_t, void *);
1059 
1060 /* Closure for cb_put_phdr(). */
1061 struct phdr_closure {
1062 	Elf_Phdr *phdr;		/* Program header to fill in */
1063 	Elf_Off offset;		/* Offset of segment in core file */
1064 };
1065 
1066 /* Closure for cb_size_segment(). */
1067 struct sseg_closure {
1068 	int count;		/* Count of writable segments. */
1069 	size_t size;		/* Total size of all writable segments. */
1070 };
1071 
1072 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1073 
1074 struct note_info {
1075 	int		type;		/* Note type. */
1076 	outfunc_t 	outfunc; 	/* Output function. */
1077 	void		*outarg;	/* Argument for the output function. */
1078 	size_t		outsize;	/* Output size. */
1079 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1080 };
1081 
1082 TAILQ_HEAD(note_info_list, note_info);
1083 
1084 /* Coredump output parameters. */
1085 struct coredump_params {
1086 	off_t		offset;
1087 	struct ucred	*active_cred;
1088 	struct ucred	*file_cred;
1089 	struct thread	*td;
1090 	struct vnode	*vp;
1091 	struct gzio_stream *gzs;
1092 };
1093 
1094 static void cb_put_phdr(vm_map_entry_t, void *);
1095 static void cb_size_segment(vm_map_entry_t, void *);
1096 static int core_write(struct coredump_params *, void *, size_t, off_t,
1097     enum uio_seg);
1098 static void each_writable_segment(struct thread *, segment_callback, void *);
1099 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1100     struct note_info_list *, size_t);
1101 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1102     size_t *);
1103 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1104 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1105 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1106 static int sbuf_drain_core_output(void *, const char *, int);
1107 static int sbuf_drain_count(void *arg, const char *data, int len);
1108 
1109 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1110 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1111 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1112 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1113 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1114 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1115 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1116 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1117 static void note_procstat_files(void *, struct sbuf *, size_t *);
1118 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1119 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1120 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1121 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1122 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1123 
1124 #ifdef GZIO
1125 extern int compress_user_cores_gzlevel;
1126 
1127 /*
1128  * Write out a core segment to the compression stream.
1129  */
1130 static int
1131 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1132 {
1133 	u_int chunk_len;
1134 	int error;
1135 
1136 	while (len > 0) {
1137 		chunk_len = MIN(len, CORE_BUF_SIZE);
1138 		copyin(base, buf, chunk_len);
1139 		error = gzio_write(p->gzs, buf, chunk_len);
1140 		if (error != 0)
1141 			break;
1142 		base += chunk_len;
1143 		len -= chunk_len;
1144 	}
1145 	return (error);
1146 }
1147 
1148 static int
1149 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1150 {
1151 
1152 	return (core_write((struct coredump_params *)arg, base, len, offset,
1153 	    UIO_SYSSPACE));
1154 }
1155 #endif /* GZIO */
1156 
1157 static int
1158 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1159     enum uio_seg seg)
1160 {
1161 
1162 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1163 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1164 	    p->active_cred, p->file_cred, NULL, p->td));
1165 }
1166 
1167 static int
1168 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1169     void *tmpbuf)
1170 {
1171 
1172 #ifdef GZIO
1173 	if (p->gzs != NULL)
1174 		return (compress_chunk(p, base, tmpbuf, len));
1175 #endif
1176 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1177 }
1178 
1179 /*
1180  * Drain into a core file.
1181  */
1182 static int
1183 sbuf_drain_core_output(void *arg, const char *data, int len)
1184 {
1185 	struct coredump_params *p;
1186 	int error, locked;
1187 
1188 	p = (struct coredump_params *)arg;
1189 
1190 	/*
1191 	 * Some kern_proc out routines that print to this sbuf may
1192 	 * call us with the process lock held. Draining with the
1193 	 * non-sleepable lock held is unsafe. The lock is needed for
1194 	 * those routines when dumping a live process. In our case we
1195 	 * can safely release the lock before draining and acquire
1196 	 * again after.
1197 	 */
1198 	locked = PROC_LOCKED(p->td->td_proc);
1199 	if (locked)
1200 		PROC_UNLOCK(p->td->td_proc);
1201 #ifdef GZIO
1202 	if (p->gzs != NULL)
1203 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1204 	else
1205 #endif
1206 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1207 		    UIO_SYSSPACE);
1208 	if (locked)
1209 		PROC_LOCK(p->td->td_proc);
1210 	if (error != 0)
1211 		return (-error);
1212 	p->offset += len;
1213 	return (len);
1214 }
1215 
1216 /*
1217  * Drain into a counter.
1218  */
1219 static int
1220 sbuf_drain_count(void *arg, const char *data __unused, int len)
1221 {
1222 	size_t *sizep;
1223 
1224 	sizep = (size_t *)arg;
1225 	*sizep += len;
1226 	return (len);
1227 }
1228 
1229 int
1230 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1231 {
1232 	struct ucred *cred = td->td_ucred;
1233 	int error = 0;
1234 	struct sseg_closure seginfo;
1235 	struct note_info_list notelst;
1236 	struct coredump_params params;
1237 	struct note_info *ninfo;
1238 	void *hdr, *tmpbuf;
1239 	size_t hdrsize, notesz, coresize;
1240 	boolean_t compress;
1241 
1242 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1243 	hdr = NULL;
1244 	tmpbuf = NULL;
1245 	TAILQ_INIT(&notelst);
1246 
1247 	/* Size the program segments. */
1248 	seginfo.count = 0;
1249 	seginfo.size = 0;
1250 	each_writable_segment(td, cb_size_segment, &seginfo);
1251 
1252 	/*
1253 	 * Collect info about the core file header area.
1254 	 */
1255 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1256 	__elfN(prepare_notes)(td, &notelst, &notesz);
1257 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1258 
1259 	/* Set up core dump parameters. */
1260 	params.offset = 0;
1261 	params.active_cred = cred;
1262 	params.file_cred = NOCRED;
1263 	params.td = td;
1264 	params.vp = vp;
1265 	params.gzs = NULL;
1266 
1267 #ifdef RACCT
1268 	if (racct_enable) {
1269 		PROC_LOCK(td->td_proc);
1270 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1271 		PROC_UNLOCK(td->td_proc);
1272 		if (error != 0) {
1273 			error = EFAULT;
1274 			goto done;
1275 		}
1276 	}
1277 #endif
1278 	if (coresize >= limit) {
1279 		error = EFAULT;
1280 		goto done;
1281 	}
1282 
1283 #ifdef GZIO
1284 	/* Create a compression stream if necessary. */
1285 	if (compress) {
1286 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1287 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1288 		if (params.gzs == NULL) {
1289 			error = EFAULT;
1290 			goto done;
1291 		}
1292 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1293         }
1294 #endif
1295 
1296 	/*
1297 	 * Allocate memory for building the header, fill it up,
1298 	 * and write it out following the notes.
1299 	 */
1300 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1301 	if (hdr == NULL) {
1302 		error = EINVAL;
1303 		goto done;
1304 	}
1305 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1306 	    notesz);
1307 
1308 	/* Write the contents of all of the writable segments. */
1309 	if (error == 0) {
1310 		Elf_Phdr *php;
1311 		off_t offset;
1312 		int i;
1313 
1314 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1315 		offset = round_page(hdrsize + notesz);
1316 		for (i = 0; i < seginfo.count; i++) {
1317 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1318 			    php->p_filesz, offset, &params, tmpbuf);
1319 			if (error != 0)
1320 				break;
1321 			offset += php->p_filesz;
1322 			php++;
1323 		}
1324 #ifdef GZIO
1325 		if (error == 0 && compress)
1326 			error = gzio_flush(params.gzs);
1327 #endif
1328 	}
1329 	if (error) {
1330 		log(LOG_WARNING,
1331 		    "Failed to write core file for process %s (error %d)\n",
1332 		    curproc->p_comm, error);
1333 	}
1334 
1335 done:
1336 #ifdef GZIO
1337 	if (compress) {
1338 		free(tmpbuf, M_TEMP);
1339 		if (params.gzs != NULL)
1340 			gzio_fini(params.gzs);
1341 	}
1342 #endif
1343 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1344 		TAILQ_REMOVE(&notelst, ninfo, link);
1345 		free(ninfo, M_TEMP);
1346 	}
1347 	if (hdr != NULL)
1348 		free(hdr, M_TEMP);
1349 
1350 	return (error);
1351 }
1352 
1353 /*
1354  * A callback for each_writable_segment() to write out the segment's
1355  * program header entry.
1356  */
1357 static void
1358 cb_put_phdr(entry, closure)
1359 	vm_map_entry_t entry;
1360 	void *closure;
1361 {
1362 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1363 	Elf_Phdr *phdr = phc->phdr;
1364 
1365 	phc->offset = round_page(phc->offset);
1366 
1367 	phdr->p_type = PT_LOAD;
1368 	phdr->p_offset = phc->offset;
1369 	phdr->p_vaddr = entry->start;
1370 	phdr->p_paddr = 0;
1371 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1372 	phdr->p_align = PAGE_SIZE;
1373 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1374 
1375 	phc->offset += phdr->p_filesz;
1376 	phc->phdr++;
1377 }
1378 
1379 /*
1380  * A callback for each_writable_segment() to gather information about
1381  * the number of segments and their total size.
1382  */
1383 static void
1384 cb_size_segment(entry, closure)
1385 	vm_map_entry_t entry;
1386 	void *closure;
1387 {
1388 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1389 
1390 	ssc->count++;
1391 	ssc->size += entry->end - entry->start;
1392 }
1393 
1394 /*
1395  * For each writable segment in the process's memory map, call the given
1396  * function with a pointer to the map entry and some arbitrary
1397  * caller-supplied data.
1398  */
1399 static void
1400 each_writable_segment(td, func, closure)
1401 	struct thread *td;
1402 	segment_callback func;
1403 	void *closure;
1404 {
1405 	struct proc *p = td->td_proc;
1406 	vm_map_t map = &p->p_vmspace->vm_map;
1407 	vm_map_entry_t entry;
1408 	vm_object_t backing_object, object;
1409 	boolean_t ignore_entry;
1410 
1411 	vm_map_lock_read(map);
1412 	for (entry = map->header.next; entry != &map->header;
1413 	    entry = entry->next) {
1414 		/*
1415 		 * Don't dump inaccessible mappings, deal with legacy
1416 		 * coredump mode.
1417 		 *
1418 		 * Note that read-only segments related to the elf binary
1419 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1420 		 * need to arbitrarily ignore such segments.
1421 		 */
1422 		if (elf_legacy_coredump) {
1423 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1424 				continue;
1425 		} else {
1426 			if ((entry->protection & VM_PROT_ALL) == 0)
1427 				continue;
1428 		}
1429 
1430 		/*
1431 		 * Dont include memory segment in the coredump if
1432 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1433 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1434 		 * kernel map).
1435 		 */
1436 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1437 			continue;
1438 
1439 		if ((object = entry->object.vm_object) == NULL)
1440 			continue;
1441 
1442 		/* Ignore memory-mapped devices and such things. */
1443 		VM_OBJECT_RLOCK(object);
1444 		while ((backing_object = object->backing_object) != NULL) {
1445 			VM_OBJECT_RLOCK(backing_object);
1446 			VM_OBJECT_RUNLOCK(object);
1447 			object = backing_object;
1448 		}
1449 		ignore_entry = object->type != OBJT_DEFAULT &&
1450 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1451 		    object->type != OBJT_PHYS;
1452 		VM_OBJECT_RUNLOCK(object);
1453 		if (ignore_entry)
1454 			continue;
1455 
1456 		(*func)(entry, closure);
1457 	}
1458 	vm_map_unlock_read(map);
1459 }
1460 
1461 /*
1462  * Write the core file header to the file, including padding up to
1463  * the page boundary.
1464  */
1465 static int
1466 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1467     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1468 {
1469 	struct note_info *ninfo;
1470 	struct sbuf *sb;
1471 	int error;
1472 
1473 	/* Fill in the header. */
1474 	bzero(hdr, hdrsize);
1475 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1476 
1477 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1478 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1479 	sbuf_start_section(sb, NULL);
1480 	sbuf_bcat(sb, hdr, hdrsize);
1481 	TAILQ_FOREACH(ninfo, notelst, link)
1482 	    __elfN(putnote)(ninfo, sb);
1483 	/* Align up to a page boundary for the program segments. */
1484 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1485 	error = sbuf_finish(sb);
1486 	sbuf_delete(sb);
1487 
1488 	return (error);
1489 }
1490 
1491 static void
1492 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1493     size_t *sizep)
1494 {
1495 	struct proc *p;
1496 	struct thread *thr;
1497 	size_t size;
1498 
1499 	p = td->td_proc;
1500 	size = 0;
1501 
1502 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1503 
1504 	/*
1505 	 * To have the debugger select the right thread (LWP) as the initial
1506 	 * thread, we dump the state of the thread passed to us in td first.
1507 	 * This is the thread that causes the core dump and thus likely to
1508 	 * be the right thread one wants to have selected in the debugger.
1509 	 */
1510 	thr = td;
1511 	while (thr != NULL) {
1512 		size += register_note(list, NT_PRSTATUS,
1513 		    __elfN(note_prstatus), thr);
1514 		size += register_note(list, NT_FPREGSET,
1515 		    __elfN(note_fpregset), thr);
1516 		size += register_note(list, NT_THRMISC,
1517 		    __elfN(note_thrmisc), thr);
1518 		size += register_note(list, -1,
1519 		    __elfN(note_threadmd), thr);
1520 
1521 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1522 		    TAILQ_NEXT(thr, td_plist);
1523 		if (thr == td)
1524 			thr = TAILQ_NEXT(thr, td_plist);
1525 	}
1526 
1527 	size += register_note(list, NT_PROCSTAT_PROC,
1528 	    __elfN(note_procstat_proc), p);
1529 	size += register_note(list, NT_PROCSTAT_FILES,
1530 	    note_procstat_files, p);
1531 	size += register_note(list, NT_PROCSTAT_VMMAP,
1532 	    note_procstat_vmmap, p);
1533 	size += register_note(list, NT_PROCSTAT_GROUPS,
1534 	    note_procstat_groups, p);
1535 	size += register_note(list, NT_PROCSTAT_UMASK,
1536 	    note_procstat_umask, p);
1537 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1538 	    note_procstat_rlimit, p);
1539 	size += register_note(list, NT_PROCSTAT_OSREL,
1540 	    note_procstat_osrel, p);
1541 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1542 	    __elfN(note_procstat_psstrings), p);
1543 	size += register_note(list, NT_PROCSTAT_AUXV,
1544 	    __elfN(note_procstat_auxv), p);
1545 
1546 	*sizep = size;
1547 }
1548 
1549 static void
1550 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1551     size_t notesz)
1552 {
1553 	Elf_Ehdr *ehdr;
1554 	Elf_Phdr *phdr;
1555 	struct phdr_closure phc;
1556 
1557 	ehdr = (Elf_Ehdr *)hdr;
1558 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1559 
1560 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1561 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1562 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1563 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1564 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1565 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1566 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1567 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1568 	ehdr->e_ident[EI_ABIVERSION] = 0;
1569 	ehdr->e_ident[EI_PAD] = 0;
1570 	ehdr->e_type = ET_CORE;
1571 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1572 	ehdr->e_machine = ELF_ARCH32;
1573 #else
1574 	ehdr->e_machine = ELF_ARCH;
1575 #endif
1576 	ehdr->e_version = EV_CURRENT;
1577 	ehdr->e_entry = 0;
1578 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1579 	ehdr->e_flags = 0;
1580 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1581 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1582 	ehdr->e_phnum = numsegs + 1;
1583 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1584 	ehdr->e_shnum = 0;
1585 	ehdr->e_shstrndx = SHN_UNDEF;
1586 
1587 	/*
1588 	 * Fill in the program header entries.
1589 	 */
1590 
1591 	/* The note segement. */
1592 	phdr->p_type = PT_NOTE;
1593 	phdr->p_offset = hdrsize;
1594 	phdr->p_vaddr = 0;
1595 	phdr->p_paddr = 0;
1596 	phdr->p_filesz = notesz;
1597 	phdr->p_memsz = 0;
1598 	phdr->p_flags = PF_R;
1599 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1600 	phdr++;
1601 
1602 	/* All the writable segments from the program. */
1603 	phc.phdr = phdr;
1604 	phc.offset = round_page(hdrsize + notesz);
1605 	each_writable_segment(td, cb_put_phdr, &phc);
1606 }
1607 
1608 static size_t
1609 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1610 {
1611 	struct note_info *ninfo;
1612 	size_t size, notesize;
1613 
1614 	size = 0;
1615 	out(arg, NULL, &size);
1616 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1617 	ninfo->type = type;
1618 	ninfo->outfunc = out;
1619 	ninfo->outarg = arg;
1620 	ninfo->outsize = size;
1621 	TAILQ_INSERT_TAIL(list, ninfo, link);
1622 
1623 	if (type == -1)
1624 		return (size);
1625 
1626 	notesize = sizeof(Elf_Note) +		/* note header */
1627 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1628 						/* note name */
1629 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1630 
1631 	return (notesize);
1632 }
1633 
1634 static size_t
1635 append_note_data(const void *src, void *dst, size_t len)
1636 {
1637 	size_t padded_len;
1638 
1639 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1640 	if (dst != NULL) {
1641 		bcopy(src, dst, len);
1642 		bzero((char *)dst + len, padded_len - len);
1643 	}
1644 	return (padded_len);
1645 }
1646 
1647 size_t
1648 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1649 {
1650 	Elf_Note *note;
1651 	char *buf;
1652 	size_t notesize;
1653 
1654 	buf = dst;
1655 	if (buf != NULL) {
1656 		note = (Elf_Note *)buf;
1657 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1658 		note->n_descsz = size;
1659 		note->n_type = type;
1660 		buf += sizeof(*note);
1661 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1662 		    sizeof(FREEBSD_ABI_VENDOR));
1663 		append_note_data(src, buf, size);
1664 		if (descp != NULL)
1665 			*descp = buf;
1666 	}
1667 
1668 	notesize = sizeof(Elf_Note) +		/* note header */
1669 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1670 						/* note name */
1671 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1672 
1673 	return (notesize);
1674 }
1675 
1676 static void
1677 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1678 {
1679 	Elf_Note note;
1680 	ssize_t old_len;
1681 
1682 	if (ninfo->type == -1) {
1683 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1684 		return;
1685 	}
1686 
1687 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1688 	note.n_descsz = ninfo->outsize;
1689 	note.n_type = ninfo->type;
1690 
1691 	sbuf_bcat(sb, &note, sizeof(note));
1692 	sbuf_start_section(sb, &old_len);
1693 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1694 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1695 	if (note.n_descsz == 0)
1696 		return;
1697 	sbuf_start_section(sb, &old_len);
1698 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1699 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1700 }
1701 
1702 /*
1703  * Miscellaneous note out functions.
1704  */
1705 
1706 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1707 #include <compat/freebsd32/freebsd32.h>
1708 
1709 typedef struct prstatus32 elf_prstatus_t;
1710 typedef struct prpsinfo32 elf_prpsinfo_t;
1711 typedef struct fpreg32 elf_prfpregset_t;
1712 typedef struct fpreg32 elf_fpregset_t;
1713 typedef struct reg32 elf_gregset_t;
1714 typedef struct thrmisc32 elf_thrmisc_t;
1715 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1716 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1717 typedef uint32_t elf_ps_strings_t;
1718 #else
1719 typedef prstatus_t elf_prstatus_t;
1720 typedef prpsinfo_t elf_prpsinfo_t;
1721 typedef prfpregset_t elf_prfpregset_t;
1722 typedef prfpregset_t elf_fpregset_t;
1723 typedef gregset_t elf_gregset_t;
1724 typedef thrmisc_t elf_thrmisc_t;
1725 #define ELF_KERN_PROC_MASK	0
1726 typedef struct kinfo_proc elf_kinfo_proc_t;
1727 typedef vm_offset_t elf_ps_strings_t;
1728 #endif
1729 
1730 static void
1731 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1732 {
1733 	struct proc *p;
1734 	elf_prpsinfo_t *psinfo;
1735 
1736 	p = (struct proc *)arg;
1737 	if (sb != NULL) {
1738 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1739 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1740 		psinfo->pr_version = PRPSINFO_VERSION;
1741 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1742 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1743 		/*
1744 		 * XXX - We don't fill in the command line arguments properly
1745 		 * yet.
1746 		 */
1747 		strlcpy(psinfo->pr_psargs, p->p_comm,
1748 		    sizeof(psinfo->pr_psargs));
1749 
1750 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1751 		free(psinfo, M_TEMP);
1752 	}
1753 	*sizep = sizeof(*psinfo);
1754 }
1755 
1756 static void
1757 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1758 {
1759 	struct thread *td;
1760 	elf_prstatus_t *status;
1761 
1762 	td = (struct thread *)arg;
1763 	if (sb != NULL) {
1764 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1765 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1766 		status->pr_version = PRSTATUS_VERSION;
1767 		status->pr_statussz = sizeof(elf_prstatus_t);
1768 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1769 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1770 		status->pr_osreldate = osreldate;
1771 		status->pr_cursig = td->td_proc->p_sig;
1772 		status->pr_pid = td->td_tid;
1773 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1774 		fill_regs32(td, &status->pr_reg);
1775 #else
1776 		fill_regs(td, &status->pr_reg);
1777 #endif
1778 		sbuf_bcat(sb, status, sizeof(*status));
1779 		free(status, M_TEMP);
1780 	}
1781 	*sizep = sizeof(*status);
1782 }
1783 
1784 static void
1785 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1786 {
1787 	struct thread *td;
1788 	elf_prfpregset_t *fpregset;
1789 
1790 	td = (struct thread *)arg;
1791 	if (sb != NULL) {
1792 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1793 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1794 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1795 		fill_fpregs32(td, fpregset);
1796 #else
1797 		fill_fpregs(td, fpregset);
1798 #endif
1799 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1800 		free(fpregset, M_TEMP);
1801 	}
1802 	*sizep = sizeof(*fpregset);
1803 }
1804 
1805 static void
1806 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1807 {
1808 	struct thread *td;
1809 	elf_thrmisc_t thrmisc;
1810 
1811 	td = (struct thread *)arg;
1812 	if (sb != NULL) {
1813 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1814 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1815 		strcpy(thrmisc.pr_tname, td->td_name);
1816 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1817 	}
1818 	*sizep = sizeof(thrmisc);
1819 }
1820 
1821 /*
1822  * Allow for MD specific notes, as well as any MD
1823  * specific preparations for writing MI notes.
1824  */
1825 static void
1826 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1827 {
1828 	struct thread *td;
1829 	void *buf;
1830 	size_t size;
1831 
1832 	td = (struct thread *)arg;
1833 	size = *sizep;
1834 	if (size != 0 && sb != NULL)
1835 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1836 	else
1837 		buf = NULL;
1838 	size = 0;
1839 	__elfN(dump_thread)(td, buf, &size);
1840 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1841 	if (size != 0 && sb != NULL)
1842 		sbuf_bcat(sb, buf, size);
1843 	free(buf, M_TEMP);
1844 	*sizep = size;
1845 }
1846 
1847 #ifdef KINFO_PROC_SIZE
1848 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1849 #endif
1850 
1851 static void
1852 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1853 {
1854 	struct proc *p;
1855 	size_t size;
1856 	int structsize;
1857 
1858 	p = (struct proc *)arg;
1859 	size = sizeof(structsize) + p->p_numthreads *
1860 	    sizeof(elf_kinfo_proc_t);
1861 
1862 	if (sb != NULL) {
1863 		KASSERT(*sizep == size, ("invalid size"));
1864 		structsize = sizeof(elf_kinfo_proc_t);
1865 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1866 		sx_slock(&proctree_lock);
1867 		PROC_LOCK(p);
1868 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1869 		sx_sunlock(&proctree_lock);
1870 	}
1871 	*sizep = size;
1872 }
1873 
1874 #ifdef KINFO_FILE_SIZE
1875 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1876 #endif
1877 
1878 static void
1879 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1880 {
1881 	struct proc *p;
1882 	size_t size;
1883 	int structsize;
1884 
1885 	p = (struct proc *)arg;
1886 	if (sb == NULL) {
1887 		size = 0;
1888 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1889 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1890 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1891 		PROC_LOCK(p);
1892 		kern_proc_filedesc_out(p, sb, -1);
1893 		sbuf_finish(sb);
1894 		sbuf_delete(sb);
1895 		*sizep = size;
1896 	} else {
1897 		structsize = sizeof(struct kinfo_file);
1898 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1899 		PROC_LOCK(p);
1900 		kern_proc_filedesc_out(p, sb, -1);
1901 	}
1902 }
1903 
1904 #ifdef KINFO_VMENTRY_SIZE
1905 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1906 #endif
1907 
1908 static void
1909 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1910 {
1911 	struct proc *p;
1912 	size_t size;
1913 	int structsize;
1914 
1915 	p = (struct proc *)arg;
1916 	if (sb == NULL) {
1917 		size = 0;
1918 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1919 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1920 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1921 		PROC_LOCK(p);
1922 		kern_proc_vmmap_out(p, sb);
1923 		sbuf_finish(sb);
1924 		sbuf_delete(sb);
1925 		*sizep = size;
1926 	} else {
1927 		structsize = sizeof(struct kinfo_vmentry);
1928 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1929 		PROC_LOCK(p);
1930 		kern_proc_vmmap_out(p, sb);
1931 	}
1932 }
1933 
1934 static void
1935 note_procstat_groups(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) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1943 	if (sb != NULL) {
1944 		KASSERT(*sizep == size, ("invalid size"));
1945 		structsize = sizeof(gid_t);
1946 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1947 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1948 		    sizeof(gid_t));
1949 	}
1950 	*sizep = size;
1951 }
1952 
1953 static void
1954 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1955 {
1956 	struct proc *p;
1957 	size_t size;
1958 	int structsize;
1959 
1960 	p = (struct proc *)arg;
1961 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1962 	if (sb != NULL) {
1963 		KASSERT(*sizep == size, ("invalid size"));
1964 		structsize = sizeof(p->p_fd->fd_cmask);
1965 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1966 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1967 	}
1968 	*sizep = size;
1969 }
1970 
1971 static void
1972 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1973 {
1974 	struct proc *p;
1975 	struct rlimit rlim[RLIM_NLIMITS];
1976 	size_t size;
1977 	int structsize, i;
1978 
1979 	p = (struct proc *)arg;
1980 	size = sizeof(structsize) + sizeof(rlim);
1981 	if (sb != NULL) {
1982 		KASSERT(*sizep == size, ("invalid size"));
1983 		structsize = sizeof(rlim);
1984 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1985 		PROC_LOCK(p);
1986 		for (i = 0; i < RLIM_NLIMITS; i++)
1987 			lim_rlimit_proc(p, i, &rlim[i]);
1988 		PROC_UNLOCK(p);
1989 		sbuf_bcat(sb, rlim, sizeof(rlim));
1990 	}
1991 	*sizep = size;
1992 }
1993 
1994 static void
1995 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1996 {
1997 	struct proc *p;
1998 	size_t size;
1999 	int structsize;
2000 
2001 	p = (struct proc *)arg;
2002 	size = sizeof(structsize) + sizeof(p->p_osrel);
2003 	if (sb != NULL) {
2004 		KASSERT(*sizep == size, ("invalid size"));
2005 		structsize = sizeof(p->p_osrel);
2006 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2007 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2008 	}
2009 	*sizep = size;
2010 }
2011 
2012 static void
2013 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2014 {
2015 	struct proc *p;
2016 	elf_ps_strings_t ps_strings;
2017 	size_t size;
2018 	int structsize;
2019 
2020 	p = (struct proc *)arg;
2021 	size = sizeof(structsize) + sizeof(ps_strings);
2022 	if (sb != NULL) {
2023 		KASSERT(*sizep == size, ("invalid size"));
2024 		structsize = sizeof(ps_strings);
2025 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2026 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2027 #else
2028 		ps_strings = p->p_sysent->sv_psstrings;
2029 #endif
2030 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2031 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2032 	}
2033 	*sizep = size;
2034 }
2035 
2036 static void
2037 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2038 {
2039 	struct proc *p;
2040 	size_t size;
2041 	int structsize;
2042 
2043 	p = (struct proc *)arg;
2044 	if (sb == NULL) {
2045 		size = 0;
2046 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2047 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2048 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2049 		PHOLD(p);
2050 		proc_getauxv(curthread, p, sb);
2051 		PRELE(p);
2052 		sbuf_finish(sb);
2053 		sbuf_delete(sb);
2054 		*sizep = size;
2055 	} else {
2056 		structsize = sizeof(Elf_Auxinfo);
2057 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2058 		PHOLD(p);
2059 		proc_getauxv(curthread, p, sb);
2060 		PRELE(p);
2061 	}
2062 }
2063 
2064 static boolean_t
2065 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2066     int32_t *osrel, const Elf_Phdr *pnote)
2067 {
2068 	const Elf_Note *note, *note0, *note_end;
2069 	const char *note_name;
2070 	int i;
2071 
2072 	if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
2073 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
2074 		return (FALSE);
2075 
2076 	note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
2077 	note_end = (const Elf_Note *)(imgp->image_header +
2078 	    pnote->p_offset + pnote->p_filesz);
2079 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2080 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2081 		    (const char *)note < sizeof(Elf_Note))
2082 			return (FALSE);
2083 		if (note->n_namesz != checknote->hdr.n_namesz ||
2084 		    note->n_descsz != checknote->hdr.n_descsz ||
2085 		    note->n_type != checknote->hdr.n_type)
2086 			goto nextnote;
2087 		note_name = (const char *)(note + 1);
2088 		if (note_name + checknote->hdr.n_namesz >=
2089 		    (const char *)note_end || strncmp(checknote->vendor,
2090 		    note_name, checknote->hdr.n_namesz) != 0)
2091 			goto nextnote;
2092 
2093 		/*
2094 		 * Fetch the osreldate for binary
2095 		 * from the ELF OSABI-note if necessary.
2096 		 */
2097 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2098 		    checknote->trans_osrel != NULL)
2099 			return (checknote->trans_osrel(note, osrel));
2100 		return (TRUE);
2101 
2102 nextnote:
2103 		note = (const Elf_Note *)((const char *)(note + 1) +
2104 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2105 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2106 	}
2107 
2108 	return (FALSE);
2109 }
2110 
2111 /*
2112  * Try to find the appropriate ABI-note section for checknote,
2113  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2114  * first page of the image is searched, the same as for headers.
2115  */
2116 static boolean_t
2117 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2118     int32_t *osrel)
2119 {
2120 	const Elf_Phdr *phdr;
2121 	const Elf_Ehdr *hdr;
2122 	int i;
2123 
2124 	hdr = (const Elf_Ehdr *)imgp->image_header;
2125 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2126 
2127 	for (i = 0; i < hdr->e_phnum; i++) {
2128 		if (phdr[i].p_type == PT_NOTE &&
2129 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2130 			return (TRUE);
2131 	}
2132 	return (FALSE);
2133 
2134 }
2135 
2136 /*
2137  * Tell kern_execve.c about it, with a little help from the linker.
2138  */
2139 static struct execsw __elfN(execsw) = {
2140 	__CONCAT(exec_, __elfN(imgact)),
2141 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2142 };
2143 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2144 
2145 static vm_prot_t
2146 __elfN(trans_prot)(Elf_Word flags)
2147 {
2148 	vm_prot_t prot;
2149 
2150 	prot = 0;
2151 	if (flags & PF_X)
2152 		prot |= VM_PROT_EXECUTE;
2153 	if (flags & PF_W)
2154 		prot |= VM_PROT_WRITE;
2155 	if (flags & PF_R)
2156 		prot |= VM_PROT_READ;
2157 #if __ELF_WORD_SIZE == 32
2158 #if defined(__amd64__)
2159 	if (i386_read_exec && (flags & PF_R))
2160 		prot |= VM_PROT_EXECUTE;
2161 #endif
2162 #endif
2163 	return (prot);
2164 }
2165 
2166 static Elf_Word
2167 __elfN(untrans_prot)(vm_prot_t prot)
2168 {
2169 	Elf_Word flags;
2170 
2171 	flags = 0;
2172 	if (prot & VM_PROT_EXECUTE)
2173 		flags |= PF_X;
2174 	if (prot & VM_PROT_READ)
2175 		flags |= PF_R;
2176 	if (prot & VM_PROT_WRITE)
2177 		flags |= PF_W;
2178 	return (flags);
2179 }
2180