xref: /freebsd/sys/kern/imgact_elf.c (revision 895f86f15fbf6540071feb9328c3c50ed1f027b8)
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 			/* Give brand a chance to veto check_note's guess */
277 			if (ret && bi->header_supported)
278 				ret = bi->header_supported(imgp);
279 			if (ret)
280 				return (bi);
281 		}
282 	}
283 
284 	/* If the executable has a brand, search for it in the brand list. */
285 	for (i = 0; i < MAX_BRANDS; i++) {
286 		bi = elf_brand_list[i];
287 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
288 			continue;
289 		if (hdr->e_machine == bi->machine &&
290 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
291 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
292 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
293 			/* Looks good, but give brand a chance to veto */
294 			if (!bi->header_supported || bi->header_supported(imgp))
295 				return (bi);
296 		}
297 	}
298 
299 	/* No known brand, see if the header is recognized by any brand */
300 	for (i = 0; i < MAX_BRANDS; i++) {
301 		bi = elf_brand_list[i];
302 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
303 		    bi->header_supported == NULL)
304 			continue;
305 		if (hdr->e_machine == bi->machine) {
306 			ret = bi->header_supported(imgp);
307 			if (ret)
308 				return (bi);
309 		}
310 	}
311 
312 	/* Lacking a known brand, search for a recognized interpreter. */
313 	if (interp != NULL) {
314 		for (i = 0; i < MAX_BRANDS; i++) {
315 			bi = elf_brand_list[i];
316 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
317 				continue;
318 			if (hdr->e_machine == bi->machine &&
319 			    /* ELF image p_filesz includes terminating zero */
320 			    strlen(bi->interp_path) + 1 == interp_name_len &&
321 			    strncmp(interp, bi->interp_path, interp_name_len)
322 			    == 0)
323 				return (bi);
324 		}
325 	}
326 
327 	/* Lacking a recognized interpreter, try the default brand */
328 	for (i = 0; i < MAX_BRANDS; i++) {
329 		bi = elf_brand_list[i];
330 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
331 			continue;
332 		if (hdr->e_machine == bi->machine &&
333 		    __elfN(fallback_brand) == bi->brand)
334 			return (bi);
335 	}
336 	return (NULL);
337 }
338 
339 static int
340 __elfN(check_header)(const Elf_Ehdr *hdr)
341 {
342 	Elf_Brandinfo *bi;
343 	int i;
344 
345 	if (!IS_ELF(*hdr) ||
346 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
347 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
348 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
349 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
350 	    hdr->e_version != ELF_TARG_VER)
351 		return (ENOEXEC);
352 
353 	/*
354 	 * Make sure we have at least one brand for this machine.
355 	 */
356 
357 	for (i = 0; i < MAX_BRANDS; i++) {
358 		bi = elf_brand_list[i];
359 		if (bi != NULL && bi->machine == hdr->e_machine)
360 			break;
361 	}
362 	if (i == MAX_BRANDS)
363 		return (ENOEXEC);
364 
365 	return (0);
366 }
367 
368 static int
369 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
370     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
371 {
372 	struct sf_buf *sf;
373 	int error;
374 	vm_offset_t off;
375 
376 	/*
377 	 * Create the page if it doesn't exist yet. Ignore errors.
378 	 */
379 	vm_map_lock(map);
380 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
381 	    VM_PROT_ALL, VM_PROT_ALL, 0);
382 	vm_map_unlock(map);
383 
384 	/*
385 	 * Find the page from the underlying object.
386 	 */
387 	if (object) {
388 		sf = vm_imgact_map_page(object, offset);
389 		if (sf == NULL)
390 			return (KERN_FAILURE);
391 		off = offset - trunc_page(offset);
392 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
393 		    end - start);
394 		vm_imgact_unmap_page(sf);
395 		if (error) {
396 			return (KERN_FAILURE);
397 		}
398 	}
399 
400 	return (KERN_SUCCESS);
401 }
402 
403 static int
404 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
405     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
406 {
407 	struct sf_buf *sf;
408 	vm_offset_t off;
409 	vm_size_t sz;
410 	int error, rv;
411 
412 	if (start != trunc_page(start)) {
413 		rv = __elfN(map_partial)(map, object, offset, start,
414 		    round_page(start), prot);
415 		if (rv)
416 			return (rv);
417 		offset += round_page(start) - start;
418 		start = round_page(start);
419 	}
420 	if (end != round_page(end)) {
421 		rv = __elfN(map_partial)(map, object, offset +
422 		    trunc_page(end) - start, trunc_page(end), end, prot);
423 		if (rv)
424 			return (rv);
425 		end = trunc_page(end);
426 	}
427 	if (end > start) {
428 		if (offset & PAGE_MASK) {
429 			/*
430 			 * The mapping is not page aligned. This means we have
431 			 * to copy the data. Sigh.
432 			 */
433 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
434 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
435 			    0);
436 			if (rv)
437 				return (rv);
438 			if (object == NULL)
439 				return (KERN_SUCCESS);
440 			for (; start < end; start += sz) {
441 				sf = vm_imgact_map_page(object, offset);
442 				if (sf == NULL)
443 					return (KERN_FAILURE);
444 				off = offset - trunc_page(offset);
445 				sz = end - start;
446 				if (sz > PAGE_SIZE - off)
447 					sz = PAGE_SIZE - off;
448 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
449 				    (caddr_t)start, sz);
450 				vm_imgact_unmap_page(sf);
451 				if (error) {
452 					return (KERN_FAILURE);
453 				}
454 				offset += sz;
455 			}
456 			rv = KERN_SUCCESS;
457 		} else {
458 			vm_object_reference(object);
459 			vm_map_lock(map);
460 			rv = vm_map_insert(map, object, offset, start, end,
461 			    prot, VM_PROT_ALL, cow);
462 			vm_map_unlock(map);
463 			if (rv != KERN_SUCCESS)
464 				vm_object_deallocate(object);
465 		}
466 		return (rv);
467 	} else {
468 		return (KERN_SUCCESS);
469 	}
470 }
471 
472 static int
473 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
474     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
475     size_t pagesize)
476 {
477 	struct sf_buf *sf;
478 	size_t map_len;
479 	vm_map_t map;
480 	vm_object_t object;
481 	vm_offset_t map_addr;
482 	int error, rv, cow;
483 	size_t copy_len;
484 	vm_offset_t file_addr;
485 
486 	/*
487 	 * It's necessary to fail if the filsz + offset taken from the
488 	 * header is greater than the actual file pager object's size.
489 	 * If we were to allow this, then the vm_map_find() below would
490 	 * walk right off the end of the file object and into the ether.
491 	 *
492 	 * While I'm here, might as well check for something else that
493 	 * is invalid: filsz cannot be greater than memsz.
494 	 */
495 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
496 		uprintf("elf_load_section: truncated ELF file\n");
497 		return (ENOEXEC);
498 	}
499 
500 	object = imgp->object;
501 	map = &imgp->proc->p_vmspace->vm_map;
502 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
503 	file_addr = trunc_page_ps(offset, pagesize);
504 
505 	/*
506 	 * We have two choices.  We can either clear the data in the last page
507 	 * of an oversized mapping, or we can start the anon mapping a page
508 	 * early and copy the initialized data into that first page.  We
509 	 * choose the second..
510 	 */
511 	if (memsz > filsz)
512 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
513 	else
514 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
515 
516 	if (map_len != 0) {
517 		/* cow flags: don't dump readonly sections in core */
518 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
519 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
520 
521 		rv = __elfN(map_insert)(map,
522 				      object,
523 				      file_addr,	/* file offset */
524 				      map_addr,		/* virtual start */
525 				      map_addr + map_len,/* virtual end */
526 				      prot,
527 				      cow);
528 		if (rv != KERN_SUCCESS)
529 			return (EINVAL);
530 
531 		/* we can stop now if we've covered it all */
532 		if (memsz == filsz) {
533 			return (0);
534 		}
535 	}
536 
537 
538 	/*
539 	 * We have to get the remaining bit of the file into the first part
540 	 * of the oversized map segment.  This is normally because the .data
541 	 * segment in the file is extended to provide bss.  It's a neat idea
542 	 * to try and save a page, but it's a pain in the behind to implement.
543 	 */
544 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
545 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
546 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
547 	    map_addr;
548 
549 	/* This had damn well better be true! */
550 	if (map_len != 0) {
551 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
552 		    map_len, VM_PROT_ALL, 0);
553 		if (rv != KERN_SUCCESS) {
554 			return (EINVAL);
555 		}
556 	}
557 
558 	if (copy_len != 0) {
559 		vm_offset_t off;
560 
561 		sf = vm_imgact_map_page(object, offset + filsz);
562 		if (sf == NULL)
563 			return (EIO);
564 
565 		/* send the page fragment to user space */
566 		off = trunc_page_ps(offset + filsz, pagesize) -
567 		    trunc_page(offset + filsz);
568 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
569 		    (caddr_t)map_addr, copy_len);
570 		vm_imgact_unmap_page(sf);
571 		if (error) {
572 			return (error);
573 		}
574 	}
575 
576 	/*
577 	 * set it to the specified protection.
578 	 * XXX had better undo the damage from pasting over the cracks here!
579 	 */
580 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
581 	    map_len), prot, FALSE);
582 
583 	return (0);
584 }
585 
586 /*
587  * Load the file "file" into memory.  It may be either a shared object
588  * or an executable.
589  *
590  * The "addr" reference parameter is in/out.  On entry, it specifies
591  * the address where a shared object should be loaded.  If the file is
592  * an executable, this value is ignored.  On exit, "addr" specifies
593  * where the file was actually loaded.
594  *
595  * The "entry" reference parameter is out only.  On exit, it specifies
596  * the entry point for the loaded file.
597  */
598 static int
599 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
600 	u_long *entry, size_t pagesize)
601 {
602 	struct {
603 		struct nameidata nd;
604 		struct vattr attr;
605 		struct image_params image_params;
606 	} *tempdata;
607 	const Elf_Ehdr *hdr = NULL;
608 	const Elf_Phdr *phdr = NULL;
609 	struct nameidata *nd;
610 	struct vattr *attr;
611 	struct image_params *imgp;
612 	vm_prot_t prot;
613 	u_long rbase;
614 	u_long base_addr = 0;
615 	int error, i, numsegs;
616 
617 #ifdef CAPABILITY_MODE
618 	/*
619 	 * XXXJA: This check can go away once we are sufficiently confident
620 	 * that the checks in namei() are correct.
621 	 */
622 	if (IN_CAPABILITY_MODE(curthread))
623 		return (ECAPMODE);
624 #endif
625 
626 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
627 	nd = &tempdata->nd;
628 	attr = &tempdata->attr;
629 	imgp = &tempdata->image_params;
630 
631 	/*
632 	 * Initialize part of the common data
633 	 */
634 	imgp->proc = p;
635 	imgp->attr = attr;
636 	imgp->firstpage = NULL;
637 	imgp->image_header = NULL;
638 	imgp->object = NULL;
639 	imgp->execlabel = NULL;
640 
641 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
642 	if ((error = namei(nd)) != 0) {
643 		nd->ni_vp = NULL;
644 		goto fail;
645 	}
646 	NDFREE(nd, NDF_ONLY_PNBUF);
647 	imgp->vp = nd->ni_vp;
648 
649 	/*
650 	 * Check permissions, modes, uid, etc on the file, and "open" it.
651 	 */
652 	error = exec_check_permissions(imgp);
653 	if (error)
654 		goto fail;
655 
656 	error = exec_map_first_page(imgp);
657 	if (error)
658 		goto fail;
659 
660 	/*
661 	 * Also make certain that the interpreter stays the same, so set
662 	 * its VV_TEXT flag, too.
663 	 */
664 	VOP_SET_TEXT(nd->ni_vp);
665 
666 	imgp->object = nd->ni_vp->v_object;
667 
668 	hdr = (const Elf_Ehdr *)imgp->image_header;
669 	if ((error = __elfN(check_header)(hdr)) != 0)
670 		goto fail;
671 	if (hdr->e_type == ET_DYN)
672 		rbase = *addr;
673 	else if (hdr->e_type == ET_EXEC)
674 		rbase = 0;
675 	else {
676 		error = ENOEXEC;
677 		goto fail;
678 	}
679 
680 	/* Only support headers that fit within first page for now      */
681 	if ((hdr->e_phoff > PAGE_SIZE) ||
682 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
683 		error = ENOEXEC;
684 		goto fail;
685 	}
686 
687 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
688 	if (!aligned(phdr, Elf_Addr)) {
689 		error = ENOEXEC;
690 		goto fail;
691 	}
692 
693 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
694 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
695 			/* Loadable segment */
696 			prot = __elfN(trans_prot)(phdr[i].p_flags);
697 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
698 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
699 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
700 			if (error != 0)
701 				goto fail;
702 			/*
703 			 * Establish the base address if this is the
704 			 * first segment.
705 			 */
706 			if (numsegs == 0)
707   				base_addr = trunc_page(phdr[i].p_vaddr +
708 				    rbase);
709 			numsegs++;
710 		}
711 	}
712 	*addr = base_addr;
713 	*entry = (unsigned long)hdr->e_entry + rbase;
714 
715 fail:
716 	if (imgp->firstpage)
717 		exec_unmap_first_page(imgp);
718 
719 	if (nd->ni_vp)
720 		vput(nd->ni_vp);
721 
722 	free(tempdata, M_TEMP);
723 
724 	return (error);
725 }
726 
727 static int
728 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
729 {
730 	struct thread *td;
731 	const Elf_Ehdr *hdr;
732 	const Elf_Phdr *phdr;
733 	Elf_Auxargs *elf_auxargs;
734 	struct vmspace *vmspace;
735 	const char *err_str, *newinterp;
736 	char *interp, *interp_buf, *path;
737 	Elf_Brandinfo *brand_info;
738 	struct sysentvec *sv;
739 	vm_prot_t prot;
740 	u_long text_size, data_size, total_size, text_addr, data_addr;
741 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
742 	int32_t osrel;
743 	int error, i, n, interp_name_len, have_interp;
744 
745 	hdr = (const Elf_Ehdr *)imgp->image_header;
746 
747 	/*
748 	 * Do we have a valid ELF header ?
749 	 *
750 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
751 	 * if particular brand doesn't support it.
752 	 */
753 	if (__elfN(check_header)(hdr) != 0 ||
754 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
755 		return (-1);
756 
757 	/*
758 	 * From here on down, we return an errno, not -1, as we've
759 	 * detected an ELF file.
760 	 */
761 
762 	if ((hdr->e_phoff > PAGE_SIZE) ||
763 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
764 		/* Only support headers in first page for now */
765 		uprintf("Program headers not in the first page\n");
766 		return (ENOEXEC);
767 	}
768 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
769 	if (!aligned(phdr, Elf_Addr)) {
770 		uprintf("Unaligned program headers\n");
771 		return (ENOEXEC);
772 	}
773 
774 	n = error = 0;
775 	baddr = 0;
776 	osrel = 0;
777 	text_size = data_size = total_size = text_addr = data_addr = 0;
778 	entry = proghdr = 0;
779 	interp_name_len = 0;
780 	err_str = newinterp = NULL;
781 	interp = interp_buf = NULL;
782 	td = curthread;
783 
784 	for (i = 0; i < hdr->e_phnum; i++) {
785 		switch (phdr[i].p_type) {
786 		case PT_LOAD:
787 			if (n == 0)
788 				baddr = phdr[i].p_vaddr;
789 			n++;
790 			break;
791 		case PT_INTERP:
792 			/* Path to interpreter */
793 			if (phdr[i].p_filesz > MAXPATHLEN) {
794 				uprintf("Invalid PT_INTERP\n");
795 				error = ENOEXEC;
796 				goto ret;
797 			}
798 			interp_name_len = phdr[i].p_filesz;
799 			if (phdr[i].p_offset > PAGE_SIZE ||
800 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
801 				VOP_UNLOCK(imgp->vp, 0);
802 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
803 				    M_WAITOK);
804 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
805 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
806 				    interp_name_len, phdr[i].p_offset,
807 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
808 				    NOCRED, NULL, td);
809 				if (error != 0) {
810 					uprintf("i/o error PT_INTERP\n");
811 					goto ret;
812 				}
813 				interp_buf[interp_name_len] = '\0';
814 				interp = interp_buf;
815 			} else {
816 				interp = __DECONST(char *, imgp->image_header) +
817 				    phdr[i].p_offset;
818 			}
819 			break;
820 		case PT_GNU_STACK:
821 			if (__elfN(nxstack))
822 				imgp->stack_prot =
823 				    __elfN(trans_prot)(phdr[i].p_flags);
824 			imgp->stack_sz = phdr[i].p_memsz;
825 			break;
826 		}
827 	}
828 
829 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
830 	    &osrel);
831 	if (brand_info == NULL) {
832 		uprintf("ELF binary type \"%u\" not known.\n",
833 		    hdr->e_ident[EI_OSABI]);
834 		error = ENOEXEC;
835 		goto ret;
836 	}
837 	if (hdr->e_type == ET_DYN) {
838 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
839 			uprintf("Cannot execute shared object\n");
840 			error = ENOEXEC;
841 			goto ret;
842 		}
843 		/*
844 		 * Honour the base load address from the dso if it is
845 		 * non-zero for some reason.
846 		 */
847 		if (baddr == 0)
848 			et_dyn_addr = ET_DYN_LOAD_ADDR;
849 		else
850 			et_dyn_addr = 0;
851 	} else
852 		et_dyn_addr = 0;
853 	sv = brand_info->sysvec;
854 	if (interp != NULL && brand_info->interp_newpath != NULL)
855 		newinterp = brand_info->interp_newpath;
856 
857 	/*
858 	 * Avoid a possible deadlock if the current address space is destroyed
859 	 * and that address space maps the locked vnode.  In the common case,
860 	 * the locked vnode's v_usecount is decremented but remains greater
861 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
862 	 * However, in cases where the vnode lock is external, such as nullfs,
863 	 * v_usecount may become zero.
864 	 *
865 	 * The VV_TEXT flag prevents modifications to the executable while
866 	 * the vnode is unlocked.
867 	 */
868 	VOP_UNLOCK(imgp->vp, 0);
869 
870 	error = exec_new_vmspace(imgp, sv);
871 	imgp->proc->p_sysent = sv;
872 
873 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
874 	if (error != 0)
875 		goto ret;
876 
877 	for (i = 0; i < hdr->e_phnum; i++) {
878 		switch (phdr[i].p_type) {
879 		case PT_LOAD:	/* Loadable segment */
880 			if (phdr[i].p_memsz == 0)
881 				break;
882 			prot = __elfN(trans_prot)(phdr[i].p_flags);
883 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
884 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
885 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
886 			    sv->sv_pagesize);
887 			if (error != 0)
888 				goto ret;
889 
890 			/*
891 			 * If this segment contains the program headers,
892 			 * remember their virtual address for the AT_PHDR
893 			 * aux entry. Static binaries don't usually include
894 			 * a PT_PHDR entry.
895 			 */
896 			if (phdr[i].p_offset == 0 &&
897 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
898 				<= phdr[i].p_filesz)
899 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
900 				    et_dyn_addr;
901 
902 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
903 			seg_size = round_page(phdr[i].p_memsz +
904 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
905 
906 			/*
907 			 * Make the largest executable segment the official
908 			 * text segment and all others data.
909 			 *
910 			 * Note that obreak() assumes that data_addr +
911 			 * data_size == end of data load area, and the ELF
912 			 * file format expects segments to be sorted by
913 			 * address.  If multiple data segments exist, the
914 			 * last one will be used.
915 			 */
916 
917 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
918 				text_size = seg_size;
919 				text_addr = seg_addr;
920 			} else {
921 				data_size = seg_size;
922 				data_addr = seg_addr;
923 			}
924 			total_size += seg_size;
925 			break;
926 		case PT_PHDR: 	/* Program header table info */
927 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
928 			break;
929 		default:
930 			break;
931 		}
932 	}
933 
934 	if (data_addr == 0 && data_size == 0) {
935 		data_addr = text_addr;
936 		data_size = text_size;
937 	}
938 
939 	entry = (u_long)hdr->e_entry + et_dyn_addr;
940 
941 	/*
942 	 * Check limits.  It should be safe to check the
943 	 * limits after loading the segments since we do
944 	 * not actually fault in all the segments pages.
945 	 */
946 	PROC_LOCK(imgp->proc);
947 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
948 		err_str = "Data segment size exceeds process limit";
949 	else if (text_size > maxtsiz)
950 		err_str = "Text segment size exceeds system limit";
951 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
952 		err_str = "Total segment size exceeds process limit";
953 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
954 		err_str = "Data segment size exceeds resource limit";
955 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
956 		err_str = "Total segment size exceeds resource limit";
957 	if (err_str != NULL) {
958 		PROC_UNLOCK(imgp->proc);
959 		uprintf("%s\n", err_str);
960 		error = ENOMEM;
961 		goto ret;
962 	}
963 
964 	vmspace = imgp->proc->p_vmspace;
965 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
966 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
967 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
968 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
969 
970 	/*
971 	 * We load the dynamic linker where a userland call
972 	 * to mmap(0, ...) would put it.  The rationale behind this
973 	 * calculation is that it leaves room for the heap to grow to
974 	 * its maximum allowed size.
975 	 */
976 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
977 	    RLIMIT_DATA));
978 	PROC_UNLOCK(imgp->proc);
979 
980 	imgp->entry_addr = entry;
981 
982 	if (interp != NULL) {
983 		have_interp = FALSE;
984 		VOP_UNLOCK(imgp->vp, 0);
985 		if (brand_info->emul_path != NULL &&
986 		    brand_info->emul_path[0] != '\0') {
987 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
988 			snprintf(path, MAXPATHLEN, "%s%s",
989 			    brand_info->emul_path, interp);
990 			error = __elfN(load_file)(imgp->proc, path, &addr,
991 			    &imgp->entry_addr, sv->sv_pagesize);
992 			free(path, M_TEMP);
993 			if (error == 0)
994 				have_interp = TRUE;
995 		}
996 		if (!have_interp && newinterp != NULL) {
997 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
998 			    &imgp->entry_addr, sv->sv_pagesize);
999 			if (error == 0)
1000 				have_interp = TRUE;
1001 		}
1002 		if (!have_interp) {
1003 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1004 			    &imgp->entry_addr, sv->sv_pagesize);
1005 		}
1006 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1007 		if (error != 0) {
1008 			uprintf("ELF interpreter %s not found\n", interp);
1009 			goto ret;
1010 		}
1011 	} else
1012 		addr = et_dyn_addr;
1013 
1014 	/*
1015 	 * Construct auxargs table (used by the fixup routine)
1016 	 */
1017 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1018 	elf_auxargs->execfd = -1;
1019 	elf_auxargs->phdr = proghdr;
1020 	elf_auxargs->phent = hdr->e_phentsize;
1021 	elf_auxargs->phnum = hdr->e_phnum;
1022 	elf_auxargs->pagesz = PAGE_SIZE;
1023 	elf_auxargs->base = addr;
1024 	elf_auxargs->flags = 0;
1025 	elf_auxargs->entry = entry;
1026 	elf_auxargs->hdr_eflags = hdr->e_flags;
1027 
1028 	imgp->auxargs = elf_auxargs;
1029 	imgp->interpreted = 0;
1030 	imgp->reloc_base = addr;
1031 	imgp->proc->p_osrel = osrel;
1032 
1033  ret:
1034 	free(interp_buf, M_TEMP);
1035 	return (error);
1036 }
1037 
1038 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1039 
1040 int
1041 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1042 {
1043 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1044 	Elf_Addr *base;
1045 	Elf_Addr *pos;
1046 
1047 	base = (Elf_Addr *)*stack_base;
1048 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1049 
1050 	if (args->execfd != -1)
1051 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1052 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1053 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1054 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1055 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1056 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1057 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1058 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1059 #ifdef AT_EHDRFLAGS
1060 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1061 #endif
1062 	if (imgp->execpathp != 0)
1063 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1064 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1065 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1066 	if (imgp->canary != 0) {
1067 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1068 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1069 	}
1070 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1071 	if (imgp->pagesizes != 0) {
1072 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1073 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1074 	}
1075 	if (imgp->sysent->sv_timekeep_base != 0) {
1076 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1077 		    imgp->sysent->sv_timekeep_base);
1078 	}
1079 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1080 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1081 	    imgp->sysent->sv_stackprot);
1082 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1083 
1084 	free(imgp->auxargs, M_TEMP);
1085 	imgp->auxargs = NULL;
1086 
1087 	base--;
1088 	suword(base, (long)imgp->args->argc);
1089 	*stack_base = (register_t *)base;
1090 	return (0);
1091 }
1092 
1093 /*
1094  * Code for generating ELF core dumps.
1095  */
1096 
1097 typedef void (*segment_callback)(vm_map_entry_t, void *);
1098 
1099 /* Closure for cb_put_phdr(). */
1100 struct phdr_closure {
1101 	Elf_Phdr *phdr;		/* Program header to fill in */
1102 	Elf_Off offset;		/* Offset of segment in core file */
1103 };
1104 
1105 /* Closure for cb_size_segment(). */
1106 struct sseg_closure {
1107 	int count;		/* Count of writable segments. */
1108 	size_t size;		/* Total size of all writable segments. */
1109 };
1110 
1111 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1112 
1113 struct note_info {
1114 	int		type;		/* Note type. */
1115 	outfunc_t 	outfunc; 	/* Output function. */
1116 	void		*outarg;	/* Argument for the output function. */
1117 	size_t		outsize;	/* Output size. */
1118 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1119 };
1120 
1121 TAILQ_HEAD(note_info_list, note_info);
1122 
1123 /* Coredump output parameters. */
1124 struct coredump_params {
1125 	off_t		offset;
1126 	struct ucred	*active_cred;
1127 	struct ucred	*file_cred;
1128 	struct thread	*td;
1129 	struct vnode	*vp;
1130 	struct gzio_stream *gzs;
1131 };
1132 
1133 static void cb_put_phdr(vm_map_entry_t, void *);
1134 static void cb_size_segment(vm_map_entry_t, void *);
1135 static int core_write(struct coredump_params *, void *, size_t, off_t,
1136     enum uio_seg);
1137 static void each_writable_segment(struct thread *, segment_callback, void *);
1138 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1139     struct note_info_list *, size_t);
1140 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1141     size_t *);
1142 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1143 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1144 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1145 static int sbuf_drain_core_output(void *, const char *, int);
1146 static int sbuf_drain_count(void *arg, const char *data, int len);
1147 
1148 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1149 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1150 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1151 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1152 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1153 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1154 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1155 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1156 static void note_procstat_files(void *, struct sbuf *, size_t *);
1157 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1158 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1159 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1160 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1161 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1162 
1163 #ifdef GZIO
1164 extern int compress_user_cores_gzlevel;
1165 
1166 /*
1167  * Write out a core segment to the compression stream.
1168  */
1169 static int
1170 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1171 {
1172 	u_int chunk_len;
1173 	int error;
1174 
1175 	while (len > 0) {
1176 		chunk_len = MIN(len, CORE_BUF_SIZE);
1177 		copyin(base, buf, chunk_len);
1178 		error = gzio_write(p->gzs, buf, chunk_len);
1179 		if (error != 0)
1180 			break;
1181 		base += chunk_len;
1182 		len -= chunk_len;
1183 	}
1184 	return (error);
1185 }
1186 
1187 static int
1188 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1189 {
1190 
1191 	return (core_write((struct coredump_params *)arg, base, len, offset,
1192 	    UIO_SYSSPACE));
1193 }
1194 #endif /* GZIO */
1195 
1196 static int
1197 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1198     enum uio_seg seg)
1199 {
1200 
1201 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1202 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1203 	    p->active_cred, p->file_cred, NULL, p->td));
1204 }
1205 
1206 static int
1207 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1208     void *tmpbuf)
1209 {
1210 
1211 #ifdef GZIO
1212 	if (p->gzs != NULL)
1213 		return (compress_chunk(p, base, tmpbuf, len));
1214 #endif
1215 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1216 }
1217 
1218 /*
1219  * Drain into a core file.
1220  */
1221 static int
1222 sbuf_drain_core_output(void *arg, const char *data, int len)
1223 {
1224 	struct coredump_params *p;
1225 	int error, locked;
1226 
1227 	p = (struct coredump_params *)arg;
1228 
1229 	/*
1230 	 * Some kern_proc out routines that print to this sbuf may
1231 	 * call us with the process lock held. Draining with the
1232 	 * non-sleepable lock held is unsafe. The lock is needed for
1233 	 * those routines when dumping a live process. In our case we
1234 	 * can safely release the lock before draining and acquire
1235 	 * again after.
1236 	 */
1237 	locked = PROC_LOCKED(p->td->td_proc);
1238 	if (locked)
1239 		PROC_UNLOCK(p->td->td_proc);
1240 #ifdef GZIO
1241 	if (p->gzs != NULL)
1242 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1243 	else
1244 #endif
1245 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1246 		    UIO_SYSSPACE);
1247 	if (locked)
1248 		PROC_LOCK(p->td->td_proc);
1249 	if (error != 0)
1250 		return (-error);
1251 	p->offset += len;
1252 	return (len);
1253 }
1254 
1255 /*
1256  * Drain into a counter.
1257  */
1258 static int
1259 sbuf_drain_count(void *arg, const char *data __unused, int len)
1260 {
1261 	size_t *sizep;
1262 
1263 	sizep = (size_t *)arg;
1264 	*sizep += len;
1265 	return (len);
1266 }
1267 
1268 int
1269 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1270 {
1271 	struct ucred *cred = td->td_ucred;
1272 	int error = 0;
1273 	struct sseg_closure seginfo;
1274 	struct note_info_list notelst;
1275 	struct coredump_params params;
1276 	struct note_info *ninfo;
1277 	void *hdr, *tmpbuf;
1278 	size_t hdrsize, notesz, coresize;
1279 #ifdef GZIO
1280 	boolean_t compress;
1281 
1282 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1283 #endif
1284 	hdr = NULL;
1285 	tmpbuf = NULL;
1286 	TAILQ_INIT(&notelst);
1287 
1288 	/* Size the program segments. */
1289 	seginfo.count = 0;
1290 	seginfo.size = 0;
1291 	each_writable_segment(td, cb_size_segment, &seginfo);
1292 
1293 	/*
1294 	 * Collect info about the core file header area.
1295 	 */
1296 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1297 	__elfN(prepare_notes)(td, &notelst, &notesz);
1298 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1299 
1300 	/* Set up core dump parameters. */
1301 	params.offset = 0;
1302 	params.active_cred = cred;
1303 	params.file_cred = NOCRED;
1304 	params.td = td;
1305 	params.vp = vp;
1306 	params.gzs = NULL;
1307 
1308 #ifdef RACCT
1309 	if (racct_enable) {
1310 		PROC_LOCK(td->td_proc);
1311 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1312 		PROC_UNLOCK(td->td_proc);
1313 		if (error != 0) {
1314 			error = EFAULT;
1315 			goto done;
1316 		}
1317 	}
1318 #endif
1319 	if (coresize >= limit) {
1320 		error = EFAULT;
1321 		goto done;
1322 	}
1323 
1324 #ifdef GZIO
1325 	/* Create a compression stream if necessary. */
1326 	if (compress) {
1327 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1328 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1329 		if (params.gzs == NULL) {
1330 			error = EFAULT;
1331 			goto done;
1332 		}
1333 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1334         }
1335 #endif
1336 
1337 	/*
1338 	 * Allocate memory for building the header, fill it up,
1339 	 * and write it out following the notes.
1340 	 */
1341 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1342 	if (hdr == NULL) {
1343 		error = EINVAL;
1344 		goto done;
1345 	}
1346 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1347 	    notesz);
1348 
1349 	/* Write the contents of all of the writable segments. */
1350 	if (error == 0) {
1351 		Elf_Phdr *php;
1352 		off_t offset;
1353 		int i;
1354 
1355 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1356 		offset = round_page(hdrsize + notesz);
1357 		for (i = 0; i < seginfo.count; i++) {
1358 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1359 			    php->p_filesz, offset, &params, tmpbuf);
1360 			if (error != 0)
1361 				break;
1362 			offset += php->p_filesz;
1363 			php++;
1364 		}
1365 #ifdef GZIO
1366 		if (error == 0 && compress)
1367 			error = gzio_flush(params.gzs);
1368 #endif
1369 	}
1370 	if (error) {
1371 		log(LOG_WARNING,
1372 		    "Failed to write core file for process %s (error %d)\n",
1373 		    curproc->p_comm, error);
1374 	}
1375 
1376 done:
1377 #ifdef GZIO
1378 	if (compress) {
1379 		free(tmpbuf, M_TEMP);
1380 		if (params.gzs != NULL)
1381 			gzio_fini(params.gzs);
1382 	}
1383 #endif
1384 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1385 		TAILQ_REMOVE(&notelst, ninfo, link);
1386 		free(ninfo, M_TEMP);
1387 	}
1388 	if (hdr != NULL)
1389 		free(hdr, M_TEMP);
1390 
1391 	return (error);
1392 }
1393 
1394 /*
1395  * A callback for each_writable_segment() to write out the segment's
1396  * program header entry.
1397  */
1398 static void
1399 cb_put_phdr(entry, closure)
1400 	vm_map_entry_t entry;
1401 	void *closure;
1402 {
1403 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1404 	Elf_Phdr *phdr = phc->phdr;
1405 
1406 	phc->offset = round_page(phc->offset);
1407 
1408 	phdr->p_type = PT_LOAD;
1409 	phdr->p_offset = phc->offset;
1410 	phdr->p_vaddr = entry->start;
1411 	phdr->p_paddr = 0;
1412 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1413 	phdr->p_align = PAGE_SIZE;
1414 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1415 
1416 	phc->offset += phdr->p_filesz;
1417 	phc->phdr++;
1418 }
1419 
1420 /*
1421  * A callback for each_writable_segment() to gather information about
1422  * the number of segments and their total size.
1423  */
1424 static void
1425 cb_size_segment(entry, closure)
1426 	vm_map_entry_t entry;
1427 	void *closure;
1428 {
1429 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1430 
1431 	ssc->count++;
1432 	ssc->size += entry->end - entry->start;
1433 }
1434 
1435 /*
1436  * For each writable segment in the process's memory map, call the given
1437  * function with a pointer to the map entry and some arbitrary
1438  * caller-supplied data.
1439  */
1440 static void
1441 each_writable_segment(td, func, closure)
1442 	struct thread *td;
1443 	segment_callback func;
1444 	void *closure;
1445 {
1446 	struct proc *p = td->td_proc;
1447 	vm_map_t map = &p->p_vmspace->vm_map;
1448 	vm_map_entry_t entry;
1449 	vm_object_t backing_object, object;
1450 	boolean_t ignore_entry;
1451 
1452 	vm_map_lock_read(map);
1453 	for (entry = map->header.next; entry != &map->header;
1454 	    entry = entry->next) {
1455 		/*
1456 		 * Don't dump inaccessible mappings, deal with legacy
1457 		 * coredump mode.
1458 		 *
1459 		 * Note that read-only segments related to the elf binary
1460 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1461 		 * need to arbitrarily ignore such segments.
1462 		 */
1463 		if (elf_legacy_coredump) {
1464 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1465 				continue;
1466 		} else {
1467 			if ((entry->protection & VM_PROT_ALL) == 0)
1468 				continue;
1469 		}
1470 
1471 		/*
1472 		 * Dont include memory segment in the coredump if
1473 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1474 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1475 		 * kernel map).
1476 		 */
1477 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1478 			continue;
1479 
1480 		if ((object = entry->object.vm_object) == NULL)
1481 			continue;
1482 
1483 		/* Ignore memory-mapped devices and such things. */
1484 		VM_OBJECT_RLOCK(object);
1485 		while ((backing_object = object->backing_object) != NULL) {
1486 			VM_OBJECT_RLOCK(backing_object);
1487 			VM_OBJECT_RUNLOCK(object);
1488 			object = backing_object;
1489 		}
1490 		ignore_entry = object->type != OBJT_DEFAULT &&
1491 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1492 		    object->type != OBJT_PHYS;
1493 		VM_OBJECT_RUNLOCK(object);
1494 		if (ignore_entry)
1495 			continue;
1496 
1497 		(*func)(entry, closure);
1498 	}
1499 	vm_map_unlock_read(map);
1500 }
1501 
1502 /*
1503  * Write the core file header to the file, including padding up to
1504  * the page boundary.
1505  */
1506 static int
1507 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1508     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1509 {
1510 	struct note_info *ninfo;
1511 	struct sbuf *sb;
1512 	int error;
1513 
1514 	/* Fill in the header. */
1515 	bzero(hdr, hdrsize);
1516 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1517 
1518 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1519 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1520 	sbuf_start_section(sb, NULL);
1521 	sbuf_bcat(sb, hdr, hdrsize);
1522 	TAILQ_FOREACH(ninfo, notelst, link)
1523 	    __elfN(putnote)(ninfo, sb);
1524 	/* Align up to a page boundary for the program segments. */
1525 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1526 	error = sbuf_finish(sb);
1527 	sbuf_delete(sb);
1528 
1529 	return (error);
1530 }
1531 
1532 static void
1533 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1534     size_t *sizep)
1535 {
1536 	struct proc *p;
1537 	struct thread *thr;
1538 	size_t size;
1539 
1540 	p = td->td_proc;
1541 	size = 0;
1542 
1543 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1544 
1545 	/*
1546 	 * To have the debugger select the right thread (LWP) as the initial
1547 	 * thread, we dump the state of the thread passed to us in td first.
1548 	 * This is the thread that causes the core dump and thus likely to
1549 	 * be the right thread one wants to have selected in the debugger.
1550 	 */
1551 	thr = td;
1552 	while (thr != NULL) {
1553 		size += register_note(list, NT_PRSTATUS,
1554 		    __elfN(note_prstatus), thr);
1555 		size += register_note(list, NT_FPREGSET,
1556 		    __elfN(note_fpregset), thr);
1557 		size += register_note(list, NT_THRMISC,
1558 		    __elfN(note_thrmisc), thr);
1559 		size += register_note(list, -1,
1560 		    __elfN(note_threadmd), thr);
1561 
1562 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1563 		    TAILQ_NEXT(thr, td_plist);
1564 		if (thr == td)
1565 			thr = TAILQ_NEXT(thr, td_plist);
1566 	}
1567 
1568 	size += register_note(list, NT_PROCSTAT_PROC,
1569 	    __elfN(note_procstat_proc), p);
1570 	size += register_note(list, NT_PROCSTAT_FILES,
1571 	    note_procstat_files, p);
1572 	size += register_note(list, NT_PROCSTAT_VMMAP,
1573 	    note_procstat_vmmap, p);
1574 	size += register_note(list, NT_PROCSTAT_GROUPS,
1575 	    note_procstat_groups, p);
1576 	size += register_note(list, NT_PROCSTAT_UMASK,
1577 	    note_procstat_umask, p);
1578 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1579 	    note_procstat_rlimit, p);
1580 	size += register_note(list, NT_PROCSTAT_OSREL,
1581 	    note_procstat_osrel, p);
1582 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1583 	    __elfN(note_procstat_psstrings), p);
1584 	size += register_note(list, NT_PROCSTAT_AUXV,
1585 	    __elfN(note_procstat_auxv), p);
1586 
1587 	*sizep = size;
1588 }
1589 
1590 static void
1591 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1592     size_t notesz)
1593 {
1594 	Elf_Ehdr *ehdr;
1595 	Elf_Phdr *phdr;
1596 	struct phdr_closure phc;
1597 
1598 	ehdr = (Elf_Ehdr *)hdr;
1599 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1600 
1601 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1602 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1603 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1604 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1605 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1606 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1607 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1608 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1609 	ehdr->e_ident[EI_ABIVERSION] = 0;
1610 	ehdr->e_ident[EI_PAD] = 0;
1611 	ehdr->e_type = ET_CORE;
1612 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1613 	ehdr->e_machine = ELF_ARCH32;
1614 #else
1615 	ehdr->e_machine = ELF_ARCH;
1616 #endif
1617 	ehdr->e_version = EV_CURRENT;
1618 	ehdr->e_entry = 0;
1619 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1620 	ehdr->e_flags = 0;
1621 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1622 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1623 	ehdr->e_phnum = numsegs + 1;
1624 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1625 	ehdr->e_shnum = 0;
1626 	ehdr->e_shstrndx = SHN_UNDEF;
1627 
1628 	/*
1629 	 * Fill in the program header entries.
1630 	 */
1631 
1632 	/* The note segement. */
1633 	phdr->p_type = PT_NOTE;
1634 	phdr->p_offset = hdrsize;
1635 	phdr->p_vaddr = 0;
1636 	phdr->p_paddr = 0;
1637 	phdr->p_filesz = notesz;
1638 	phdr->p_memsz = 0;
1639 	phdr->p_flags = PF_R;
1640 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1641 	phdr++;
1642 
1643 	/* All the writable segments from the program. */
1644 	phc.phdr = phdr;
1645 	phc.offset = round_page(hdrsize + notesz);
1646 	each_writable_segment(td, cb_put_phdr, &phc);
1647 }
1648 
1649 static size_t
1650 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1651 {
1652 	struct note_info *ninfo;
1653 	size_t size, notesize;
1654 
1655 	size = 0;
1656 	out(arg, NULL, &size);
1657 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1658 	ninfo->type = type;
1659 	ninfo->outfunc = out;
1660 	ninfo->outarg = arg;
1661 	ninfo->outsize = size;
1662 	TAILQ_INSERT_TAIL(list, ninfo, link);
1663 
1664 	if (type == -1)
1665 		return (size);
1666 
1667 	notesize = sizeof(Elf_Note) +		/* note header */
1668 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1669 						/* note name */
1670 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1671 
1672 	return (notesize);
1673 }
1674 
1675 static size_t
1676 append_note_data(const void *src, void *dst, size_t len)
1677 {
1678 	size_t padded_len;
1679 
1680 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1681 	if (dst != NULL) {
1682 		bcopy(src, dst, len);
1683 		bzero((char *)dst + len, padded_len - len);
1684 	}
1685 	return (padded_len);
1686 }
1687 
1688 size_t
1689 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1690 {
1691 	Elf_Note *note;
1692 	char *buf;
1693 	size_t notesize;
1694 
1695 	buf = dst;
1696 	if (buf != NULL) {
1697 		note = (Elf_Note *)buf;
1698 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1699 		note->n_descsz = size;
1700 		note->n_type = type;
1701 		buf += sizeof(*note);
1702 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1703 		    sizeof(FREEBSD_ABI_VENDOR));
1704 		append_note_data(src, buf, size);
1705 		if (descp != NULL)
1706 			*descp = buf;
1707 	}
1708 
1709 	notesize = sizeof(Elf_Note) +		/* note header */
1710 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1711 						/* note name */
1712 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1713 
1714 	return (notesize);
1715 }
1716 
1717 static void
1718 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1719 {
1720 	Elf_Note note;
1721 	ssize_t old_len, sect_len;
1722 	size_t new_len, descsz, i;
1723 
1724 	if (ninfo->type == -1) {
1725 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1726 		return;
1727 	}
1728 
1729 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1730 	note.n_descsz = ninfo->outsize;
1731 	note.n_type = ninfo->type;
1732 
1733 	sbuf_bcat(sb, &note, sizeof(note));
1734 	sbuf_start_section(sb, &old_len);
1735 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1736 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1737 	if (note.n_descsz == 0)
1738 		return;
1739 	sbuf_start_section(sb, &old_len);
1740 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1741 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1742 	if (sect_len < 0)
1743 		return;
1744 
1745 	new_len = (size_t)sect_len;
1746 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1747 	if (new_len < descsz) {
1748 		/*
1749 		 * It is expected that individual note emitters will correctly
1750 		 * predict their expected output size and fill up to that size
1751 		 * themselves, padding in a format-specific way if needed.
1752 		 * However, in case they don't, just do it here with zeros.
1753 		 */
1754 		for (i = 0; i < descsz - new_len; i++)
1755 			sbuf_putc(sb, 0);
1756 	} else if (new_len > descsz) {
1757 		/*
1758 		 * We can't always truncate sb -- we may have drained some
1759 		 * of it already.
1760 		 */
1761 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1762 		    "read it (%zu > %zu).  Since it is longer than "
1763 		    "expected, this coredump's notes are corrupt.  THIS "
1764 		    "IS A BUG in the note_procstat routine for type %u.\n",
1765 		    __func__, (unsigned)note.n_type, new_len, descsz,
1766 		    (unsigned)note.n_type));
1767 	}
1768 }
1769 
1770 /*
1771  * Miscellaneous note out functions.
1772  */
1773 
1774 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1775 #include <compat/freebsd32/freebsd32.h>
1776 
1777 typedef struct prstatus32 elf_prstatus_t;
1778 typedef struct prpsinfo32 elf_prpsinfo_t;
1779 typedef struct fpreg32 elf_prfpregset_t;
1780 typedef struct fpreg32 elf_fpregset_t;
1781 typedef struct reg32 elf_gregset_t;
1782 typedef struct thrmisc32 elf_thrmisc_t;
1783 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1784 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1785 typedef uint32_t elf_ps_strings_t;
1786 #else
1787 typedef prstatus_t elf_prstatus_t;
1788 typedef prpsinfo_t elf_prpsinfo_t;
1789 typedef prfpregset_t elf_prfpregset_t;
1790 typedef prfpregset_t elf_fpregset_t;
1791 typedef gregset_t elf_gregset_t;
1792 typedef thrmisc_t elf_thrmisc_t;
1793 #define ELF_KERN_PROC_MASK	0
1794 typedef struct kinfo_proc elf_kinfo_proc_t;
1795 typedef vm_offset_t elf_ps_strings_t;
1796 #endif
1797 
1798 static void
1799 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1800 {
1801 	struct proc *p;
1802 	elf_prpsinfo_t *psinfo;
1803 
1804 	p = (struct proc *)arg;
1805 	if (sb != NULL) {
1806 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1807 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1808 		psinfo->pr_version = PRPSINFO_VERSION;
1809 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1810 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1811 		/*
1812 		 * XXX - We don't fill in the command line arguments properly
1813 		 * yet.
1814 		 */
1815 		strlcpy(psinfo->pr_psargs, p->p_comm,
1816 		    sizeof(psinfo->pr_psargs));
1817 
1818 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1819 		free(psinfo, M_TEMP);
1820 	}
1821 	*sizep = sizeof(*psinfo);
1822 }
1823 
1824 static void
1825 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1826 {
1827 	struct thread *td;
1828 	elf_prstatus_t *status;
1829 
1830 	td = (struct thread *)arg;
1831 	if (sb != NULL) {
1832 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1833 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1834 		status->pr_version = PRSTATUS_VERSION;
1835 		status->pr_statussz = sizeof(elf_prstatus_t);
1836 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1837 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1838 		status->pr_osreldate = osreldate;
1839 		status->pr_cursig = td->td_proc->p_sig;
1840 		status->pr_pid = td->td_tid;
1841 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1842 		fill_regs32(td, &status->pr_reg);
1843 #else
1844 		fill_regs(td, &status->pr_reg);
1845 #endif
1846 		sbuf_bcat(sb, status, sizeof(*status));
1847 		free(status, M_TEMP);
1848 	}
1849 	*sizep = sizeof(*status);
1850 }
1851 
1852 static void
1853 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1854 {
1855 	struct thread *td;
1856 	elf_prfpregset_t *fpregset;
1857 
1858 	td = (struct thread *)arg;
1859 	if (sb != NULL) {
1860 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1861 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1862 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1863 		fill_fpregs32(td, fpregset);
1864 #else
1865 		fill_fpregs(td, fpregset);
1866 #endif
1867 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1868 		free(fpregset, M_TEMP);
1869 	}
1870 	*sizep = sizeof(*fpregset);
1871 }
1872 
1873 static void
1874 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1875 {
1876 	struct thread *td;
1877 	elf_thrmisc_t thrmisc;
1878 
1879 	td = (struct thread *)arg;
1880 	if (sb != NULL) {
1881 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1882 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1883 		strcpy(thrmisc.pr_tname, td->td_name);
1884 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1885 	}
1886 	*sizep = sizeof(thrmisc);
1887 }
1888 
1889 /*
1890  * Allow for MD specific notes, as well as any MD
1891  * specific preparations for writing MI notes.
1892  */
1893 static void
1894 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1895 {
1896 	struct thread *td;
1897 	void *buf;
1898 	size_t size;
1899 
1900 	td = (struct thread *)arg;
1901 	size = *sizep;
1902 	if (size != 0 && sb != NULL)
1903 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1904 	else
1905 		buf = NULL;
1906 	size = 0;
1907 	__elfN(dump_thread)(td, buf, &size);
1908 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1909 	if (size != 0 && sb != NULL)
1910 		sbuf_bcat(sb, buf, size);
1911 	free(buf, M_TEMP);
1912 	*sizep = size;
1913 }
1914 
1915 #ifdef KINFO_PROC_SIZE
1916 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1917 #endif
1918 
1919 static void
1920 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1921 {
1922 	struct proc *p;
1923 	size_t size;
1924 	int structsize;
1925 
1926 	p = (struct proc *)arg;
1927 	size = sizeof(structsize) + p->p_numthreads *
1928 	    sizeof(elf_kinfo_proc_t);
1929 
1930 	if (sb != NULL) {
1931 		KASSERT(*sizep == size, ("invalid size"));
1932 		structsize = sizeof(elf_kinfo_proc_t);
1933 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1934 		sx_slock(&proctree_lock);
1935 		PROC_LOCK(p);
1936 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1937 		sx_sunlock(&proctree_lock);
1938 	}
1939 	*sizep = size;
1940 }
1941 
1942 #ifdef KINFO_FILE_SIZE
1943 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1944 #endif
1945 
1946 static void
1947 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1948 {
1949 	struct proc *p;
1950 	size_t size, sect_sz, i;
1951 	ssize_t start_len, sect_len;
1952 	int structsize, filedesc_flags;
1953 
1954 	if (coredump_pack_fileinfo)
1955 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1956 	else
1957 		filedesc_flags = 0;
1958 
1959 	p = (struct proc *)arg;
1960 	structsize = sizeof(struct kinfo_file);
1961 	if (sb == NULL) {
1962 		size = 0;
1963 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1964 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1965 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1966 		PROC_LOCK(p);
1967 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1968 		sbuf_finish(sb);
1969 		sbuf_delete(sb);
1970 		*sizep = size;
1971 	} else {
1972 		sbuf_start_section(sb, &start_len);
1973 
1974 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1975 		PROC_LOCK(p);
1976 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1977 		    filedesc_flags);
1978 
1979 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
1980 		if (sect_len < 0)
1981 			return;
1982 		sect_sz = sect_len;
1983 
1984 		KASSERT(sect_sz <= *sizep,
1985 		    ("kern_proc_filedesc_out did not respect maxlen; "
1986 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
1987 		     sect_sz - sizeof(structsize)));
1988 
1989 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
1990 			sbuf_putc(sb, 0);
1991 	}
1992 }
1993 
1994 #ifdef KINFO_VMENTRY_SIZE
1995 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1996 #endif
1997 
1998 static void
1999 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2000 {
2001 	struct proc *p;
2002 	size_t size;
2003 	int structsize, vmmap_flags;
2004 
2005 	if (coredump_pack_vmmapinfo)
2006 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2007 	else
2008 		vmmap_flags = 0;
2009 
2010 	p = (struct proc *)arg;
2011 	structsize = sizeof(struct kinfo_vmentry);
2012 	if (sb == NULL) {
2013 		size = 0;
2014 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2015 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2016 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2017 		PROC_LOCK(p);
2018 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2019 		sbuf_finish(sb);
2020 		sbuf_delete(sb);
2021 		*sizep = size;
2022 	} else {
2023 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2024 		PROC_LOCK(p);
2025 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2026 		    vmmap_flags);
2027 	}
2028 }
2029 
2030 static void
2031 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2032 {
2033 	struct proc *p;
2034 	size_t size;
2035 	int structsize;
2036 
2037 	p = (struct proc *)arg;
2038 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2039 	if (sb != NULL) {
2040 		KASSERT(*sizep == size, ("invalid size"));
2041 		structsize = sizeof(gid_t);
2042 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2043 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2044 		    sizeof(gid_t));
2045 	}
2046 	*sizep = size;
2047 }
2048 
2049 static void
2050 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2051 {
2052 	struct proc *p;
2053 	size_t size;
2054 	int structsize;
2055 
2056 	p = (struct proc *)arg;
2057 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2058 	if (sb != NULL) {
2059 		KASSERT(*sizep == size, ("invalid size"));
2060 		structsize = sizeof(p->p_fd->fd_cmask);
2061 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2062 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2063 	}
2064 	*sizep = size;
2065 }
2066 
2067 static void
2068 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2069 {
2070 	struct proc *p;
2071 	struct rlimit rlim[RLIM_NLIMITS];
2072 	size_t size;
2073 	int structsize, i;
2074 
2075 	p = (struct proc *)arg;
2076 	size = sizeof(structsize) + sizeof(rlim);
2077 	if (sb != NULL) {
2078 		KASSERT(*sizep == size, ("invalid size"));
2079 		structsize = sizeof(rlim);
2080 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2081 		PROC_LOCK(p);
2082 		for (i = 0; i < RLIM_NLIMITS; i++)
2083 			lim_rlimit_proc(p, i, &rlim[i]);
2084 		PROC_UNLOCK(p);
2085 		sbuf_bcat(sb, rlim, sizeof(rlim));
2086 	}
2087 	*sizep = size;
2088 }
2089 
2090 static void
2091 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2092 {
2093 	struct proc *p;
2094 	size_t size;
2095 	int structsize;
2096 
2097 	p = (struct proc *)arg;
2098 	size = sizeof(structsize) + sizeof(p->p_osrel);
2099 	if (sb != NULL) {
2100 		KASSERT(*sizep == size, ("invalid size"));
2101 		structsize = sizeof(p->p_osrel);
2102 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2103 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2104 	}
2105 	*sizep = size;
2106 }
2107 
2108 static void
2109 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2110 {
2111 	struct proc *p;
2112 	elf_ps_strings_t ps_strings;
2113 	size_t size;
2114 	int structsize;
2115 
2116 	p = (struct proc *)arg;
2117 	size = sizeof(structsize) + sizeof(ps_strings);
2118 	if (sb != NULL) {
2119 		KASSERT(*sizep == size, ("invalid size"));
2120 		structsize = sizeof(ps_strings);
2121 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2122 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2123 #else
2124 		ps_strings = p->p_sysent->sv_psstrings;
2125 #endif
2126 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2127 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2128 	}
2129 	*sizep = size;
2130 }
2131 
2132 static void
2133 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2134 {
2135 	struct proc *p;
2136 	size_t size;
2137 	int structsize;
2138 
2139 	p = (struct proc *)arg;
2140 	if (sb == NULL) {
2141 		size = 0;
2142 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2143 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2144 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2145 		PHOLD(p);
2146 		proc_getauxv(curthread, p, sb);
2147 		PRELE(p);
2148 		sbuf_finish(sb);
2149 		sbuf_delete(sb);
2150 		*sizep = size;
2151 	} else {
2152 		structsize = sizeof(Elf_Auxinfo);
2153 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2154 		PHOLD(p);
2155 		proc_getauxv(curthread, p, sb);
2156 		PRELE(p);
2157 	}
2158 }
2159 
2160 static boolean_t
2161 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2162     int32_t *osrel, const Elf_Phdr *pnote)
2163 {
2164 	const Elf_Note *note, *note0, *note_end;
2165 	const char *note_name;
2166 	char *buf;
2167 	int i, error;
2168 	boolean_t res;
2169 
2170 	/* We need some limit, might as well use PAGE_SIZE. */
2171 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2172 		return (FALSE);
2173 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2174 	if (pnote->p_offset > PAGE_SIZE ||
2175 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2176 		VOP_UNLOCK(imgp->vp, 0);
2177 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2178 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2179 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2180 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2181 		    curthread->td_ucred, NOCRED, NULL, curthread);
2182 		if (error != 0) {
2183 			uprintf("i/o error PT_NOTE\n");
2184 			res = FALSE;
2185 			goto ret;
2186 		}
2187 		note = note0 = (const Elf_Note *)buf;
2188 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2189 	} else {
2190 		note = note0 = (const Elf_Note *)(imgp->image_header +
2191 		    pnote->p_offset);
2192 		note_end = (const Elf_Note *)(imgp->image_header +
2193 		    pnote->p_offset + pnote->p_filesz);
2194 		buf = NULL;
2195 	}
2196 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2197 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2198 		    (const char *)note < sizeof(Elf_Note)) {
2199 			res = FALSE;
2200 			goto ret;
2201 		}
2202 		if (note->n_namesz != checknote->hdr.n_namesz ||
2203 		    note->n_descsz != checknote->hdr.n_descsz ||
2204 		    note->n_type != checknote->hdr.n_type)
2205 			goto nextnote;
2206 		note_name = (const char *)(note + 1);
2207 		if (note_name + checknote->hdr.n_namesz >=
2208 		    (const char *)note_end || strncmp(checknote->vendor,
2209 		    note_name, checknote->hdr.n_namesz) != 0)
2210 			goto nextnote;
2211 
2212 		/*
2213 		 * Fetch the osreldate for binary
2214 		 * from the ELF OSABI-note if necessary.
2215 		 */
2216 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2217 		    checknote->trans_osrel != NULL) {
2218 			res = checknote->trans_osrel(note, osrel);
2219 			goto ret;
2220 		}
2221 		res = TRUE;
2222 		goto ret;
2223 nextnote:
2224 		note = (const Elf_Note *)((const char *)(note + 1) +
2225 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2226 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2227 	}
2228 	res = FALSE;
2229 ret:
2230 	free(buf, M_TEMP);
2231 	return (res);
2232 }
2233 
2234 /*
2235  * Try to find the appropriate ABI-note section for checknote,
2236  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2237  * first page of the image is searched, the same as for headers.
2238  */
2239 static boolean_t
2240 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2241     int32_t *osrel)
2242 {
2243 	const Elf_Phdr *phdr;
2244 	const Elf_Ehdr *hdr;
2245 	int i;
2246 
2247 	hdr = (const Elf_Ehdr *)imgp->image_header;
2248 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2249 
2250 	for (i = 0; i < hdr->e_phnum; i++) {
2251 		if (phdr[i].p_type == PT_NOTE &&
2252 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2253 			return (TRUE);
2254 	}
2255 	return (FALSE);
2256 
2257 }
2258 
2259 /*
2260  * Tell kern_execve.c about it, with a little help from the linker.
2261  */
2262 static struct execsw __elfN(execsw) = {
2263 	__CONCAT(exec_, __elfN(imgact)),
2264 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2265 };
2266 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2267 
2268 static vm_prot_t
2269 __elfN(trans_prot)(Elf_Word flags)
2270 {
2271 	vm_prot_t prot;
2272 
2273 	prot = 0;
2274 	if (flags & PF_X)
2275 		prot |= VM_PROT_EXECUTE;
2276 	if (flags & PF_W)
2277 		prot |= VM_PROT_WRITE;
2278 	if (flags & PF_R)
2279 		prot |= VM_PROT_READ;
2280 #if __ELF_WORD_SIZE == 32
2281 #if defined(__amd64__)
2282 	if (i386_read_exec && (flags & PF_R))
2283 		prot |= VM_PROT_EXECUTE;
2284 #endif
2285 #endif
2286 	return (prot);
2287 }
2288 
2289 static Elf_Word
2290 __elfN(untrans_prot)(vm_prot_t prot)
2291 {
2292 	Elf_Word flags;
2293 
2294 	flags = 0;
2295 	if (prot & VM_PROT_EXECUTE)
2296 		flags |= PF_X;
2297 	if (prot & VM_PROT_READ)
2298 		flags |= PF_R;
2299 	if (prot & VM_PROT_WRITE)
2300 		flags |= PF_W;
2301 	return (flags);
2302 }
2303