xref: /freebsd/sys/kern/imgact_elf.c (revision 43d55325406e96e37be661de52289c2592c82ec9)
1 /*-
2  * Copyright (c) 2000 David O'Brien
3  * Copyright (c) 1995-1996 Søren Schmidt
4  * Copyright (c) 1996 Peter Wemm
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_core.h"
37 
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
40 #include <sys/exec.h>
41 #include <sys/fcntl.h>
42 #include <sys/imgact.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mman.h>
49 #include <sys/namei.h>
50 #include <sys/pioctl.h>
51 #include <sys/proc.h>
52 #include <sys/procfs.h>
53 #include <sys/racct.h>
54 #include <sys/resourcevar.h>
55 #include <sys/rwlock.h>
56 #include <sys/sbuf.h>
57 #include <sys/sf_buf.h>
58 #include <sys/smp.h>
59 #include <sys/systm.h>
60 #include <sys/signalvar.h>
61 #include <sys/stat.h>
62 #include <sys/sx.h>
63 #include <sys/syscall.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/vnode.h>
67 #include <sys/syslog.h>
68 #include <sys/eventhandler.h>
69 #include <sys/user.h>
70 
71 #include <net/zlib.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 #ifdef COMPRESS_USER_CORES
108 static int compress_core(gzFile, char *, char *, unsigned int,
109     struct thread * td);
110 #endif
111 #define CORE_BUF_SIZE	(16 * 1024)
112 
113 int __elfN(fallback_brand) = -1;
114 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
115     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
116     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 
118 static int elf_legacy_coredump = 0;
119 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
120     &elf_legacy_coredump, 0, "");
121 
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if __ELF_WORD_SIZE == 32
133 #if defined(__amd64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136     "enable execution from readable segments");
137 #endif
138 #endif
139 
140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
141 
142 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
143 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
144 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
145 
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
147 
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
150 	.hdr.n_descsz	= sizeof(int32_t),
151 	.hdr.n_type	= 1,
152 	.vendor		= FREEBSD_ABI_VENDOR,
153 	.flags		= BN_TRANSLATE_OSREL,
154 	.trans_osrel	= __elfN(freebsd_trans_osrel)
155 };
156 
157 static boolean_t
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
159 {
160 	uintptr_t p;
161 
162 	p = (uintptr_t)(note + 1);
163 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 	*osrel = *(const int32_t *)(p);
165 
166 	return (TRUE);
167 }
168 
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
171 
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
174 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
175 	.hdr.n_type	= 1,
176 	.vendor		= GNU_ABI_VENDOR,
177 	.flags		= BN_TRANSLATE_OSREL,
178 	.trans_osrel	= kfreebsd_trans_osrel
179 };
180 
181 static boolean_t
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
183 {
184 	const Elf32_Word *desc;
185 	uintptr_t p;
186 
187 	p = (uintptr_t)(note + 1);
188 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
189 
190 	desc = (const Elf32_Word *)p;
191 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
192 		return (FALSE);
193 
194 	/*
195 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
197 	 */
198 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
199 
200 	return (TRUE);
201 }
202 
203 int
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
205 {
206 	int i;
207 
208 	for (i = 0; i < MAX_BRANDS; i++) {
209 		if (elf_brand_list[i] == NULL) {
210 			elf_brand_list[i] = entry;
211 			break;
212 		}
213 	}
214 	if (i == MAX_BRANDS) {
215 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
216 			__func__, entry);
217 		return (-1);
218 	}
219 	return (0);
220 }
221 
222 int
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
224 {
225 	int i;
226 
227 	for (i = 0; i < MAX_BRANDS; i++) {
228 		if (elf_brand_list[i] == entry) {
229 			elf_brand_list[i] = NULL;
230 			break;
231 		}
232 	}
233 	if (i == MAX_BRANDS)
234 		return (-1);
235 	return (0);
236 }
237 
238 int
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
240 {
241 	struct proc *p;
242 	int rval = FALSE;
243 
244 	sx_slock(&allproc_lock);
245 	FOREACH_PROC_IN_SYSTEM(p) {
246 		if (p->p_sysent == entry->sysvec) {
247 			rval = TRUE;
248 			break;
249 		}
250 	}
251 	sx_sunlock(&allproc_lock);
252 
253 	return (rval);
254 }
255 
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258     int interp_name_len, int32_t *osrel)
259 {
260 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
261 	Elf_Brandinfo *bi;
262 	boolean_t ret;
263 	int i;
264 
265 	/*
266 	 * We support four types of branding -- (1) the ELF EI_OSABI field
267 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 	 * branding w/in the ELF header, (3) path of the `interp_path'
269 	 * field, and (4) the ".note.ABI-tag" ELF section.
270 	 */
271 
272 	/* Look for an ".note.ABI-tag" ELF section */
273 	for (i = 0; i < MAX_BRANDS; i++) {
274 		bi = elf_brand_list[i];
275 		if (bi == NULL)
276 			continue;
277 		if (hdr->e_machine == bi->machine && (bi->flags &
278 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
279 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
280 			if (ret)
281 				return (bi);
282 		}
283 	}
284 
285 	/* If the executable has a brand, search for it in the brand list. */
286 	for (i = 0; i < MAX_BRANDS; i++) {
287 		bi = elf_brand_list[i];
288 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
289 			continue;
290 		if (hdr->e_machine == bi->machine &&
291 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
292 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
293 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
294 			return (bi);
295 	}
296 
297 	/* No known brand, see if the header is recognized by any brand */
298 	for (i = 0; i < MAX_BRANDS; i++) {
299 		bi = elf_brand_list[i];
300 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
301 		    bi->header_supported == NULL)
302 			continue;
303 		if (hdr->e_machine == bi->machine) {
304 			ret = bi->header_supported(imgp);
305 			if (ret)
306 				return (bi);
307 		}
308 	}
309 
310 	/* Lacking a known brand, search for a recognized interpreter. */
311 	if (interp != NULL) {
312 		for (i = 0; i < MAX_BRANDS; i++) {
313 			bi = elf_brand_list[i];
314 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
315 				continue;
316 			if (hdr->e_machine == bi->machine &&
317 			    /* ELF image p_filesz includes terminating zero */
318 			    strlen(bi->interp_path) + 1 == interp_name_len &&
319 			    strncmp(interp, bi->interp_path, interp_name_len)
320 			    == 0)
321 				return (bi);
322 		}
323 	}
324 
325 	/* Lacking a recognized interpreter, try the default brand */
326 	for (i = 0; i < MAX_BRANDS; i++) {
327 		bi = elf_brand_list[i];
328 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
329 			continue;
330 		if (hdr->e_machine == bi->machine &&
331 		    __elfN(fallback_brand) == bi->brand)
332 			return (bi);
333 	}
334 	return (NULL);
335 }
336 
337 static int
338 __elfN(check_header)(const Elf_Ehdr *hdr)
339 {
340 	Elf_Brandinfo *bi;
341 	int i;
342 
343 	if (!IS_ELF(*hdr) ||
344 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
345 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
346 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
347 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
348 	    hdr->e_version != ELF_TARG_VER)
349 		return (ENOEXEC);
350 
351 	/*
352 	 * Make sure we have at least one brand for this machine.
353 	 */
354 
355 	for (i = 0; i < MAX_BRANDS; i++) {
356 		bi = elf_brand_list[i];
357 		if (bi != NULL && bi->machine == hdr->e_machine)
358 			break;
359 	}
360 	if (i == MAX_BRANDS)
361 		return (ENOEXEC);
362 
363 	return (0);
364 }
365 
366 static int
367 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
368     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
369 {
370 	struct sf_buf *sf;
371 	int error;
372 	vm_offset_t off;
373 
374 	/*
375 	 * Create the page if it doesn't exist yet. Ignore errors.
376 	 */
377 	vm_map_lock(map);
378 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
379 	    VM_PROT_ALL, VM_PROT_ALL, 0);
380 	vm_map_unlock(map);
381 
382 	/*
383 	 * Find the page from the underlying object.
384 	 */
385 	if (object) {
386 		sf = vm_imgact_map_page(object, offset);
387 		if (sf == NULL)
388 			return (KERN_FAILURE);
389 		off = offset - trunc_page(offset);
390 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
391 		    end - start);
392 		vm_imgact_unmap_page(sf);
393 		if (error) {
394 			return (KERN_FAILURE);
395 		}
396 	}
397 
398 	return (KERN_SUCCESS);
399 }
400 
401 static int
402 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
403     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
404 {
405 	struct sf_buf *sf;
406 	vm_offset_t off;
407 	vm_size_t sz;
408 	int error, rv;
409 
410 	if (start != trunc_page(start)) {
411 		rv = __elfN(map_partial)(map, object, offset, start,
412 		    round_page(start), prot);
413 		if (rv)
414 			return (rv);
415 		offset += round_page(start) - start;
416 		start = round_page(start);
417 	}
418 	if (end != round_page(end)) {
419 		rv = __elfN(map_partial)(map, object, offset +
420 		    trunc_page(end) - start, trunc_page(end), end, prot);
421 		if (rv)
422 			return (rv);
423 		end = trunc_page(end);
424 	}
425 	if (end > start) {
426 		if (offset & PAGE_MASK) {
427 			/*
428 			 * The mapping is not page aligned. This means we have
429 			 * to copy the data. Sigh.
430 			 */
431 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
432 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
433 			    0);
434 			if (rv)
435 				return (rv);
436 			if (object == NULL)
437 				return (KERN_SUCCESS);
438 			for (; start < end; start += sz) {
439 				sf = vm_imgact_map_page(object, offset);
440 				if (sf == NULL)
441 					return (KERN_FAILURE);
442 				off = offset - trunc_page(offset);
443 				sz = end - start;
444 				if (sz > PAGE_SIZE - off)
445 					sz = PAGE_SIZE - off;
446 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
447 				    (caddr_t)start, sz);
448 				vm_imgact_unmap_page(sf);
449 				if (error) {
450 					return (KERN_FAILURE);
451 				}
452 				offset += sz;
453 			}
454 			rv = KERN_SUCCESS;
455 		} else {
456 			vm_object_reference(object);
457 			vm_map_lock(map);
458 			rv = vm_map_insert(map, object, offset, start, end,
459 			    prot, VM_PROT_ALL, cow);
460 			vm_map_unlock(map);
461 			if (rv != KERN_SUCCESS)
462 				vm_object_deallocate(object);
463 		}
464 		return (rv);
465 	} else {
466 		return (KERN_SUCCESS);
467 	}
468 }
469 
470 static int
471 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
472     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
473     size_t pagesize)
474 {
475 	struct sf_buf *sf;
476 	size_t map_len;
477 	vm_map_t map;
478 	vm_object_t object;
479 	vm_offset_t map_addr;
480 	int error, rv, cow;
481 	size_t copy_len;
482 	vm_offset_t file_addr;
483 
484 	/*
485 	 * It's necessary to fail if the filsz + offset taken from the
486 	 * header is greater than the actual file pager object's size.
487 	 * If we were to allow this, then the vm_map_find() below would
488 	 * walk right off the end of the file object and into the ether.
489 	 *
490 	 * While I'm here, might as well check for something else that
491 	 * is invalid: filsz cannot be greater than memsz.
492 	 */
493 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
494 		uprintf("elf_load_section: truncated ELF file\n");
495 		return (ENOEXEC);
496 	}
497 
498 	object = imgp->object;
499 	map = &imgp->proc->p_vmspace->vm_map;
500 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
501 	file_addr = trunc_page_ps(offset, pagesize);
502 
503 	/*
504 	 * We have two choices.  We can either clear the data in the last page
505 	 * of an oversized mapping, or we can start the anon mapping a page
506 	 * early and copy the initialized data into that first page.  We
507 	 * choose the second..
508 	 */
509 	if (memsz > filsz)
510 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
511 	else
512 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
513 
514 	if (map_len != 0) {
515 		/* cow flags: don't dump readonly sections in core */
516 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
517 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
518 
519 		rv = __elfN(map_insert)(map,
520 				      object,
521 				      file_addr,	/* file offset */
522 				      map_addr,		/* virtual start */
523 				      map_addr + map_len,/* virtual end */
524 				      prot,
525 				      cow);
526 		if (rv != KERN_SUCCESS)
527 			return (EINVAL);
528 
529 		/* we can stop now if we've covered it all */
530 		if (memsz == filsz) {
531 			return (0);
532 		}
533 	}
534 
535 
536 	/*
537 	 * We have to get the remaining bit of the file into the first part
538 	 * of the oversized map segment.  This is normally because the .data
539 	 * segment in the file is extended to provide bss.  It's a neat idea
540 	 * to try and save a page, but it's a pain in the behind to implement.
541 	 */
542 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
543 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
544 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
545 	    map_addr;
546 
547 	/* This had damn well better be true! */
548 	if (map_len != 0) {
549 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
550 		    map_len, VM_PROT_ALL, 0);
551 		if (rv != KERN_SUCCESS) {
552 			return (EINVAL);
553 		}
554 	}
555 
556 	if (copy_len != 0) {
557 		vm_offset_t off;
558 
559 		sf = vm_imgact_map_page(object, offset + filsz);
560 		if (sf == NULL)
561 			return (EIO);
562 
563 		/* send the page fragment to user space */
564 		off = trunc_page_ps(offset + filsz, pagesize) -
565 		    trunc_page(offset + filsz);
566 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
567 		    (caddr_t)map_addr, copy_len);
568 		vm_imgact_unmap_page(sf);
569 		if (error) {
570 			return (error);
571 		}
572 	}
573 
574 	/*
575 	 * set it to the specified protection.
576 	 * XXX had better undo the damage from pasting over the cracks here!
577 	 */
578 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
579 	    map_len), prot, FALSE);
580 
581 	return (0);
582 }
583 
584 /*
585  * Load the file "file" into memory.  It may be either a shared object
586  * or an executable.
587  *
588  * The "addr" reference parameter is in/out.  On entry, it specifies
589  * the address where a shared object should be loaded.  If the file is
590  * an executable, this value is ignored.  On exit, "addr" specifies
591  * where the file was actually loaded.
592  *
593  * The "entry" reference parameter is out only.  On exit, it specifies
594  * the entry point for the loaded file.
595  */
596 static int
597 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
598 	u_long *entry, size_t pagesize)
599 {
600 	struct {
601 		struct nameidata nd;
602 		struct vattr attr;
603 		struct image_params image_params;
604 	} *tempdata;
605 	const Elf_Ehdr *hdr = NULL;
606 	const Elf_Phdr *phdr = NULL;
607 	struct nameidata *nd;
608 	struct vattr *attr;
609 	struct image_params *imgp;
610 	vm_prot_t prot;
611 	u_long rbase;
612 	u_long base_addr = 0;
613 	int error, i, numsegs;
614 
615 #ifdef CAPABILITY_MODE
616 	/*
617 	 * XXXJA: This check can go away once we are sufficiently confident
618 	 * that the checks in namei() are correct.
619 	 */
620 	if (IN_CAPABILITY_MODE(curthread))
621 		return (ECAPMODE);
622 #endif
623 
624 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
625 	nd = &tempdata->nd;
626 	attr = &tempdata->attr;
627 	imgp = &tempdata->image_params;
628 
629 	/*
630 	 * Initialize part of the common data
631 	 */
632 	imgp->proc = p;
633 	imgp->attr = attr;
634 	imgp->firstpage = NULL;
635 	imgp->image_header = NULL;
636 	imgp->object = NULL;
637 	imgp->execlabel = NULL;
638 
639 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
640 	if ((error = namei(nd)) != 0) {
641 		nd->ni_vp = NULL;
642 		goto fail;
643 	}
644 	NDFREE(nd, NDF_ONLY_PNBUF);
645 	imgp->vp = nd->ni_vp;
646 
647 	/*
648 	 * Check permissions, modes, uid, etc on the file, and "open" it.
649 	 */
650 	error = exec_check_permissions(imgp);
651 	if (error)
652 		goto fail;
653 
654 	error = exec_map_first_page(imgp);
655 	if (error)
656 		goto fail;
657 
658 	/*
659 	 * Also make certain that the interpreter stays the same, so set
660 	 * its VV_TEXT flag, too.
661 	 */
662 	VOP_SET_TEXT(nd->ni_vp);
663 
664 	imgp->object = nd->ni_vp->v_object;
665 
666 	hdr = (const Elf_Ehdr *)imgp->image_header;
667 	if ((error = __elfN(check_header)(hdr)) != 0)
668 		goto fail;
669 	if (hdr->e_type == ET_DYN)
670 		rbase = *addr;
671 	else if (hdr->e_type == ET_EXEC)
672 		rbase = 0;
673 	else {
674 		error = ENOEXEC;
675 		goto fail;
676 	}
677 
678 	/* Only support headers that fit within first page for now      */
679 	if ((hdr->e_phoff > PAGE_SIZE) ||
680 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
681 		error = ENOEXEC;
682 		goto fail;
683 	}
684 
685 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
686 	if (!aligned(phdr, Elf_Addr)) {
687 		error = ENOEXEC;
688 		goto fail;
689 	}
690 
691 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
692 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
693 			/* Loadable segment */
694 			prot = __elfN(trans_prot)(phdr[i].p_flags);
695 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
696 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
697 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
698 			if (error != 0)
699 				goto fail;
700 			/*
701 			 * Establish the base address if this is the
702 			 * first segment.
703 			 */
704 			if (numsegs == 0)
705   				base_addr = trunc_page(phdr[i].p_vaddr +
706 				    rbase);
707 			numsegs++;
708 		}
709 	}
710 	*addr = base_addr;
711 	*entry = (unsigned long)hdr->e_entry + rbase;
712 
713 fail:
714 	if (imgp->firstpage)
715 		exec_unmap_first_page(imgp);
716 
717 	if (nd->ni_vp)
718 		vput(nd->ni_vp);
719 
720 	free(tempdata, M_TEMP);
721 
722 	return (error);
723 }
724 
725 static int
726 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
727 {
728 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
729 	const Elf_Phdr *phdr;
730 	Elf_Auxargs *elf_auxargs;
731 	struct vmspace *vmspace;
732 	vm_prot_t prot;
733 	u_long text_size = 0, data_size = 0, total_size = 0;
734 	u_long text_addr = 0, data_addr = 0;
735 	u_long seg_size, seg_addr;
736 	u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
737 	int32_t osrel = 0;
738 	int error = 0, i, n, interp_name_len = 0;
739 	const char *interp = NULL, *newinterp = NULL;
740 	Elf_Brandinfo *brand_info;
741 	char *path;
742 	struct sysentvec *sv;
743 
744 	/*
745 	 * Do we have a valid ELF header ?
746 	 *
747 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
748 	 * if particular brand doesn't support it.
749 	 */
750 	if (__elfN(check_header)(hdr) != 0 ||
751 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
752 		return (-1);
753 
754 	/*
755 	 * From here on down, we return an errno, not -1, as we've
756 	 * detected an ELF file.
757 	 */
758 
759 	if ((hdr->e_phoff > PAGE_SIZE) ||
760 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
761 		/* Only support headers in first page for now */
762 		return (ENOEXEC);
763 	}
764 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
765 	if (!aligned(phdr, Elf_Addr))
766 		return (ENOEXEC);
767 	n = 0;
768 	baddr = 0;
769 	for (i = 0; i < hdr->e_phnum; i++) {
770 		switch (phdr[i].p_type) {
771 		case PT_LOAD:
772 			if (n == 0)
773 				baddr = phdr[i].p_vaddr;
774 			n++;
775 			break;
776 		case PT_INTERP:
777 			/* Path to interpreter */
778 			if (phdr[i].p_filesz > MAXPATHLEN ||
779 			    phdr[i].p_offset > PAGE_SIZE ||
780 			    phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
781 				return (ENOEXEC);
782 			interp = imgp->image_header + phdr[i].p_offset;
783 			interp_name_len = phdr[i].p_filesz;
784 			break;
785 		case PT_GNU_STACK:
786 			if (__elfN(nxstack))
787 				imgp->stack_prot =
788 				    __elfN(trans_prot)(phdr[i].p_flags);
789 			break;
790 		}
791 	}
792 
793 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
794 	    &osrel);
795 	if (brand_info == NULL) {
796 		uprintf("ELF binary type \"%u\" not known.\n",
797 		    hdr->e_ident[EI_OSABI]);
798 		return (ENOEXEC);
799 	}
800 	if (hdr->e_type == ET_DYN) {
801 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
802 			return (ENOEXEC);
803 		/*
804 		 * Honour the base load address from the dso if it is
805 		 * non-zero for some reason.
806 		 */
807 		if (baddr == 0)
808 			et_dyn_addr = ET_DYN_LOAD_ADDR;
809 		else
810 			et_dyn_addr = 0;
811 	} else
812 		et_dyn_addr = 0;
813 	sv = brand_info->sysvec;
814 	if (interp != NULL && brand_info->interp_newpath != NULL)
815 		newinterp = brand_info->interp_newpath;
816 
817 	/*
818 	 * Avoid a possible deadlock if the current address space is destroyed
819 	 * and that address space maps the locked vnode.  In the common case,
820 	 * the locked vnode's v_usecount is decremented but remains greater
821 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
822 	 * However, in cases where the vnode lock is external, such as nullfs,
823 	 * v_usecount may become zero.
824 	 *
825 	 * The VV_TEXT flag prevents modifications to the executable while
826 	 * the vnode is unlocked.
827 	 */
828 	VOP_UNLOCK(imgp->vp, 0);
829 
830 	error = exec_new_vmspace(imgp, sv);
831 	imgp->proc->p_sysent = sv;
832 
833 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
834 	if (error)
835 		return (error);
836 
837 	for (i = 0; i < hdr->e_phnum; i++) {
838 		switch (phdr[i].p_type) {
839 		case PT_LOAD:	/* Loadable segment */
840 			if (phdr[i].p_memsz == 0)
841 				break;
842 			prot = __elfN(trans_prot)(phdr[i].p_flags);
843 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
844 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
845 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
846 			    sv->sv_pagesize);
847 			if (error != 0)
848 				return (error);
849 
850 			/*
851 			 * If this segment contains the program headers,
852 			 * remember their virtual address for the AT_PHDR
853 			 * aux entry. Static binaries don't usually include
854 			 * a PT_PHDR entry.
855 			 */
856 			if (phdr[i].p_offset == 0 &&
857 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
858 				<= phdr[i].p_filesz)
859 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
860 				    et_dyn_addr;
861 
862 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
863 			seg_size = round_page(phdr[i].p_memsz +
864 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
865 
866 			/*
867 			 * Make the largest executable segment the official
868 			 * text segment and all others data.
869 			 *
870 			 * Note that obreak() assumes that data_addr +
871 			 * data_size == end of data load area, and the ELF
872 			 * file format expects segments to be sorted by
873 			 * address.  If multiple data segments exist, the
874 			 * last one will be used.
875 			 */
876 
877 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
878 				text_size = seg_size;
879 				text_addr = seg_addr;
880 			} else {
881 				data_size = seg_size;
882 				data_addr = seg_addr;
883 			}
884 			total_size += seg_size;
885 			break;
886 		case PT_PHDR: 	/* Program header table info */
887 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
888 			break;
889 		default:
890 			break;
891 		}
892 	}
893 
894 	if (data_addr == 0 && data_size == 0) {
895 		data_addr = text_addr;
896 		data_size = text_size;
897 	}
898 
899 	entry = (u_long)hdr->e_entry + et_dyn_addr;
900 
901 	/*
902 	 * Check limits.  It should be safe to check the
903 	 * limits after loading the segments since we do
904 	 * not actually fault in all the segments pages.
905 	 */
906 	PROC_LOCK(imgp->proc);
907 	if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
908 	    text_size > maxtsiz ||
909 	    total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
910 	    racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
911 	    racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
912 		PROC_UNLOCK(imgp->proc);
913 		return (ENOMEM);
914 	}
915 
916 	vmspace = imgp->proc->p_vmspace;
917 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
918 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
919 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
920 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
921 
922 	/*
923 	 * We load the dynamic linker where a userland call
924 	 * to mmap(0, ...) would put it.  The rationale behind this
925 	 * calculation is that it leaves room for the heap to grow to
926 	 * its maximum allowed size.
927 	 */
928 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
929 	    RLIMIT_DATA));
930 	PROC_UNLOCK(imgp->proc);
931 
932 	imgp->entry_addr = entry;
933 
934 	if (interp != NULL) {
935 		int have_interp = FALSE;
936 		VOP_UNLOCK(imgp->vp, 0);
937 		if (brand_info->emul_path != NULL &&
938 		    brand_info->emul_path[0] != '\0') {
939 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
940 			snprintf(path, MAXPATHLEN, "%s%s",
941 			    brand_info->emul_path, interp);
942 			error = __elfN(load_file)(imgp->proc, path, &addr,
943 			    &imgp->entry_addr, sv->sv_pagesize);
944 			free(path, M_TEMP);
945 			if (error == 0)
946 				have_interp = TRUE;
947 		}
948 		if (!have_interp && newinterp != NULL) {
949 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
950 			    &imgp->entry_addr, sv->sv_pagesize);
951 			if (error == 0)
952 				have_interp = TRUE;
953 		}
954 		if (!have_interp) {
955 			error = __elfN(load_file)(imgp->proc, interp, &addr,
956 			    &imgp->entry_addr, sv->sv_pagesize);
957 		}
958 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
959 		if (error != 0) {
960 			uprintf("ELF interpreter %s not found\n", interp);
961 			return (error);
962 		}
963 	} else
964 		addr = et_dyn_addr;
965 
966 	/*
967 	 * Construct auxargs table (used by the fixup routine)
968 	 */
969 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
970 	elf_auxargs->execfd = -1;
971 	elf_auxargs->phdr = proghdr;
972 	elf_auxargs->phent = hdr->e_phentsize;
973 	elf_auxargs->phnum = hdr->e_phnum;
974 	elf_auxargs->pagesz = PAGE_SIZE;
975 	elf_auxargs->base = addr;
976 	elf_auxargs->flags = 0;
977 	elf_auxargs->entry = entry;
978 
979 	imgp->auxargs = elf_auxargs;
980 	imgp->interpreted = 0;
981 	imgp->reloc_base = addr;
982 	imgp->proc->p_osrel = osrel;
983 
984 	return (error);
985 }
986 
987 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
988 
989 int
990 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
991 {
992 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
993 	Elf_Addr *base;
994 	Elf_Addr *pos;
995 
996 	base = (Elf_Addr *)*stack_base;
997 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
998 
999 	if (args->execfd != -1)
1000 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1001 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1002 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1003 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1004 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1005 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1006 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1007 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1008 	if (imgp->execpathp != 0)
1009 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1010 	AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate);
1011 	if (imgp->canary != 0) {
1012 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1013 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1014 	}
1015 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1016 	if (imgp->pagesizes != 0) {
1017 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1018 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1019 	}
1020 	if (imgp->sysent->sv_timekeep_base != 0) {
1021 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1022 		    imgp->sysent->sv_timekeep_base);
1023 	}
1024 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1025 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1026 	    imgp->sysent->sv_stackprot);
1027 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1028 
1029 	free(imgp->auxargs, M_TEMP);
1030 	imgp->auxargs = NULL;
1031 
1032 	base--;
1033 	suword(base, (long)imgp->args->argc);
1034 	*stack_base = (register_t *)base;
1035 	return (0);
1036 }
1037 
1038 /*
1039  * Code for generating ELF core dumps.
1040  */
1041 
1042 typedef void (*segment_callback)(vm_map_entry_t, void *);
1043 
1044 /* Closure for cb_put_phdr(). */
1045 struct phdr_closure {
1046 	Elf_Phdr *phdr;		/* Program header to fill in */
1047 	Elf_Off offset;		/* Offset of segment in core file */
1048 };
1049 
1050 /* Closure for cb_size_segment(). */
1051 struct sseg_closure {
1052 	int count;		/* Count of writable segments. */
1053 	size_t size;		/* Total size of all writable segments. */
1054 };
1055 
1056 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1057 
1058 struct note_info {
1059 	int		type;		/* Note type. */
1060 	outfunc_t 	outfunc; 	/* Output function. */
1061 	void		*outarg;	/* Argument for the output function. */
1062 	size_t		outsize;	/* Output size. */
1063 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1064 };
1065 
1066 TAILQ_HEAD(note_info_list, note_info);
1067 
1068 static void cb_put_phdr(vm_map_entry_t, void *);
1069 static void cb_size_segment(vm_map_entry_t, void *);
1070 static void each_writable_segment(struct thread *, segment_callback, void *);
1071 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1072     int, void *, size_t, struct note_info_list *, size_t, gzFile);
1073 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1074     size_t *);
1075 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1076 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1077 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1078 static int sbuf_drain_core_output(void *, const char *, int);
1079 static int sbuf_drain_count(void *arg, const char *data, int len);
1080 
1081 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1082 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1083 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1084 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1085 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1086 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1087 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1088 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1089 static void note_procstat_files(void *, struct sbuf *, size_t *);
1090 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1091 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1092 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1093 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1094 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1095 
1096 #ifdef COMPRESS_USER_CORES
1097 extern int compress_user_cores;
1098 extern int compress_user_cores_gzlevel;
1099 #endif
1100 
1101 static int
1102 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1103     struct ucred *active_cred, struct ucred *file_cred,
1104     struct thread *td, char *core_buf, gzFile gzfile) {
1105 
1106 	int error;
1107 	if (gzfile) {
1108 #ifdef COMPRESS_USER_CORES
1109 		error = compress_core(gzfile, base, core_buf, len, td);
1110 #else
1111 		panic("shouldn't be here");
1112 #endif
1113 	} else {
1114 		error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1115 		    UIO_USERSPACE, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1116 		    active_cred, file_cred, NULL, td);
1117 	}
1118 	return (error);
1119 }
1120 
1121 /* Coredump output parameters for sbuf drain routine. */
1122 struct sbuf_drain_core_params {
1123 	off_t		offset;
1124 	struct ucred	*active_cred;
1125 	struct ucred	*file_cred;
1126 	struct thread	*td;
1127 	struct vnode	*vp;
1128 #ifdef COMPRESS_USER_CORES
1129 	gzFile		gzfile;
1130 #endif
1131 };
1132 
1133 /*
1134  * Drain into a core file.
1135  */
1136 static int
1137 sbuf_drain_core_output(void *arg, const char *data, int len)
1138 {
1139 	struct sbuf_drain_core_params *p;
1140 	int error, locked;
1141 
1142 	p = (struct sbuf_drain_core_params *)arg;
1143 
1144 	/*
1145 	 * Some kern_proc out routines that print to this sbuf may
1146 	 * call us with the process lock held. Draining with the
1147 	 * non-sleepable lock held is unsafe. The lock is needed for
1148 	 * those routines when dumping a live process. In our case we
1149 	 * can safely release the lock before draining and acquire
1150 	 * again after.
1151 	 */
1152 	locked = PROC_LOCKED(p->td->td_proc);
1153 	if (locked)
1154 		PROC_UNLOCK(p->td->td_proc);
1155 #ifdef COMPRESS_USER_CORES
1156 	if (p->gzfile != Z_NULL)
1157 		error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1158 		    len, p->td);
1159 	else
1160 #endif
1161 		error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1162 		    __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1163 		    IO_UNIT | IO_DIRECT | IO_RANGELOCKED, p->active_cred,
1164 		    p->file_cred, NULL, p->td);
1165 	if (locked)
1166 		PROC_LOCK(p->td->td_proc);
1167 	if (error != 0)
1168 		return (-error);
1169 	p->offset += len;
1170 	return (len);
1171 }
1172 
1173 /*
1174  * Drain into a counter.
1175  */
1176 static int
1177 sbuf_drain_count(void *arg, const char *data __unused, int len)
1178 {
1179 	size_t *sizep;
1180 
1181 	sizep = (size_t *)arg;
1182 	*sizep += len;
1183 	return (len);
1184 }
1185 
1186 int
1187 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1188 {
1189 	struct ucred *cred = td->td_ucred;
1190 	int error = 0;
1191 	struct sseg_closure seginfo;
1192 	struct note_info_list notelst;
1193 	struct note_info *ninfo;
1194 	void *hdr;
1195 	size_t hdrsize, notesz, coresize;
1196 
1197 	gzFile gzfile = Z_NULL;
1198 	char *core_buf = NULL;
1199 #ifdef COMPRESS_USER_CORES
1200 	char gzopen_flags[8];
1201 	char *p;
1202 	int doing_compress = flags & IMGACT_CORE_COMPRESS;
1203 #endif
1204 
1205 	hdr = NULL;
1206 	TAILQ_INIT(&notelst);
1207 
1208 #ifdef COMPRESS_USER_CORES
1209         if (doing_compress) {
1210                 p = gzopen_flags;
1211                 *p++ = 'w';
1212                 if (compress_user_cores_gzlevel >= 0 &&
1213                     compress_user_cores_gzlevel <= 9)
1214                         *p++ = '0' + compress_user_cores_gzlevel;
1215                 *p = 0;
1216                 gzfile = gz_open("", gzopen_flags, vp);
1217                 if (gzfile == Z_NULL) {
1218                         error = EFAULT;
1219                         goto done;
1220                 }
1221                 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1222                 if (!core_buf) {
1223                         error = ENOMEM;
1224                         goto done;
1225                 }
1226         }
1227 #endif
1228 
1229 	/* Size the program segments. */
1230 	seginfo.count = 0;
1231 	seginfo.size = 0;
1232 	each_writable_segment(td, cb_size_segment, &seginfo);
1233 
1234 	/*
1235 	 * Collect info about the core file header area.
1236 	 */
1237 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1238 	__elfN(prepare_notes)(td, &notelst, &notesz);
1239 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1240 
1241 #ifdef RACCT
1242 	PROC_LOCK(td->td_proc);
1243 	error = racct_add(td->td_proc, RACCT_CORE, coresize);
1244 	PROC_UNLOCK(td->td_proc);
1245 	if (error != 0) {
1246 		error = EFAULT;
1247 		goto done;
1248 	}
1249 #endif
1250 	if (coresize >= limit) {
1251 		error = EFAULT;
1252 		goto done;
1253 	}
1254 
1255 	/*
1256 	 * Allocate memory for building the header, fill it up,
1257 	 * and write it out following the notes.
1258 	 */
1259 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1260 	if (hdr == NULL) {
1261 		error = EINVAL;
1262 		goto done;
1263 	}
1264 	error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1265 	    &notelst, notesz, gzfile);
1266 
1267 	/* Write the contents of all of the writable segments. */
1268 	if (error == 0) {
1269 		Elf_Phdr *php;
1270 		off_t offset;
1271 		int i;
1272 
1273 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1274 		offset = round_page(hdrsize + notesz);
1275 		for (i = 0; i < seginfo.count; i++) {
1276 			error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1277 			    php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1278 			if (error != 0)
1279 				break;
1280 			offset += php->p_filesz;
1281 			php++;
1282 		}
1283 	}
1284 	if (error) {
1285 		log(LOG_WARNING,
1286 		    "Failed to write core file for process %s (error %d)\n",
1287 		    curproc->p_comm, error);
1288 	}
1289 
1290 done:
1291 #ifdef COMPRESS_USER_CORES
1292 	if (core_buf)
1293 		free(core_buf, M_TEMP);
1294 	if (gzfile)
1295 		gzclose(gzfile);
1296 #endif
1297 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1298 		TAILQ_REMOVE(&notelst, ninfo, link);
1299 		free(ninfo, M_TEMP);
1300 	}
1301 	if (hdr != NULL)
1302 		free(hdr, M_TEMP);
1303 
1304 	return (error);
1305 }
1306 
1307 /*
1308  * A callback for each_writable_segment() to write out the segment's
1309  * program header entry.
1310  */
1311 static void
1312 cb_put_phdr(entry, closure)
1313 	vm_map_entry_t entry;
1314 	void *closure;
1315 {
1316 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1317 	Elf_Phdr *phdr = phc->phdr;
1318 
1319 	phc->offset = round_page(phc->offset);
1320 
1321 	phdr->p_type = PT_LOAD;
1322 	phdr->p_offset = phc->offset;
1323 	phdr->p_vaddr = entry->start;
1324 	phdr->p_paddr = 0;
1325 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1326 	phdr->p_align = PAGE_SIZE;
1327 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1328 
1329 	phc->offset += phdr->p_filesz;
1330 	phc->phdr++;
1331 }
1332 
1333 /*
1334  * A callback for each_writable_segment() to gather information about
1335  * the number of segments and their total size.
1336  */
1337 static void
1338 cb_size_segment(entry, closure)
1339 	vm_map_entry_t entry;
1340 	void *closure;
1341 {
1342 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1343 
1344 	ssc->count++;
1345 	ssc->size += entry->end - entry->start;
1346 }
1347 
1348 /*
1349  * For each writable segment in the process's memory map, call the given
1350  * function with a pointer to the map entry and some arbitrary
1351  * caller-supplied data.
1352  */
1353 static void
1354 each_writable_segment(td, func, closure)
1355 	struct thread *td;
1356 	segment_callback func;
1357 	void *closure;
1358 {
1359 	struct proc *p = td->td_proc;
1360 	vm_map_t map = &p->p_vmspace->vm_map;
1361 	vm_map_entry_t entry;
1362 	vm_object_t backing_object, object;
1363 	boolean_t ignore_entry;
1364 
1365 	vm_map_lock_read(map);
1366 	for (entry = map->header.next; entry != &map->header;
1367 	    entry = entry->next) {
1368 		/*
1369 		 * Don't dump inaccessible mappings, deal with legacy
1370 		 * coredump mode.
1371 		 *
1372 		 * Note that read-only segments related to the elf binary
1373 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1374 		 * need to arbitrarily ignore such segments.
1375 		 */
1376 		if (elf_legacy_coredump) {
1377 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1378 				continue;
1379 		} else {
1380 			if ((entry->protection & VM_PROT_ALL) == 0)
1381 				continue;
1382 		}
1383 
1384 		/*
1385 		 * Dont include memory segment in the coredump if
1386 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1387 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1388 		 * kernel map).
1389 		 */
1390 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1391 			continue;
1392 
1393 		if ((object = entry->object.vm_object) == NULL)
1394 			continue;
1395 
1396 		/* Ignore memory-mapped devices and such things. */
1397 		VM_OBJECT_RLOCK(object);
1398 		while ((backing_object = object->backing_object) != NULL) {
1399 			VM_OBJECT_RLOCK(backing_object);
1400 			VM_OBJECT_RUNLOCK(object);
1401 			object = backing_object;
1402 		}
1403 		ignore_entry = object->type != OBJT_DEFAULT &&
1404 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1405 		VM_OBJECT_RUNLOCK(object);
1406 		if (ignore_entry)
1407 			continue;
1408 
1409 		(*func)(entry, closure);
1410 	}
1411 	vm_map_unlock_read(map);
1412 }
1413 
1414 /*
1415  * Write the core file header to the file, including padding up to
1416  * the page boundary.
1417  */
1418 static int
1419 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1420     int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1421     size_t notesz, gzFile gzfile)
1422 {
1423 	struct sbuf_drain_core_params params;
1424 	struct note_info *ninfo;
1425 	struct sbuf *sb;
1426 	int error;
1427 
1428 	/* Fill in the header. */
1429 	bzero(hdr, hdrsize);
1430 	__elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1431 
1432 	params.offset = 0;
1433 	params.active_cred = cred;
1434 	params.file_cred = NOCRED;
1435 	params.td = td;
1436 	params.vp = vp;
1437 #ifdef COMPRESS_USER_CORES
1438 	params.gzfile = gzfile;
1439 #endif
1440 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1441 	sbuf_set_drain(sb, sbuf_drain_core_output, &params);
1442 	sbuf_start_section(sb, NULL);
1443 	sbuf_bcat(sb, hdr, hdrsize);
1444 	TAILQ_FOREACH(ninfo, notelst, link)
1445 	    __elfN(putnote)(ninfo, sb);
1446 	/* Align up to a page boundary for the program segments. */
1447 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1448 	error = sbuf_finish(sb);
1449 	sbuf_delete(sb);
1450 
1451 	return (error);
1452 }
1453 
1454 static void
1455 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1456     size_t *sizep)
1457 {
1458 	struct proc *p;
1459 	struct thread *thr;
1460 	size_t size;
1461 
1462 	p = td->td_proc;
1463 	size = 0;
1464 
1465 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1466 
1467 	/*
1468 	 * To have the debugger select the right thread (LWP) as the initial
1469 	 * thread, we dump the state of the thread passed to us in td first.
1470 	 * This is the thread that causes the core dump and thus likely to
1471 	 * be the right thread one wants to have selected in the debugger.
1472 	 */
1473 	thr = td;
1474 	while (thr != NULL) {
1475 		size += register_note(list, NT_PRSTATUS,
1476 		    __elfN(note_prstatus), thr);
1477 		size += register_note(list, NT_FPREGSET,
1478 		    __elfN(note_fpregset), thr);
1479 		size += register_note(list, NT_THRMISC,
1480 		    __elfN(note_thrmisc), thr);
1481 		size += register_note(list, -1,
1482 		    __elfN(note_threadmd), thr);
1483 
1484 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1485 		    TAILQ_NEXT(thr, td_plist);
1486 		if (thr == td)
1487 			thr = TAILQ_NEXT(thr, td_plist);
1488 	}
1489 
1490 	size += register_note(list, NT_PROCSTAT_PROC,
1491 	    __elfN(note_procstat_proc), p);
1492 	size += register_note(list, NT_PROCSTAT_FILES,
1493 	    note_procstat_files, p);
1494 	size += register_note(list, NT_PROCSTAT_VMMAP,
1495 	    note_procstat_vmmap, p);
1496 	size += register_note(list, NT_PROCSTAT_GROUPS,
1497 	    note_procstat_groups, p);
1498 	size += register_note(list, NT_PROCSTAT_UMASK,
1499 	    note_procstat_umask, p);
1500 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1501 	    note_procstat_rlimit, p);
1502 	size += register_note(list, NT_PROCSTAT_OSREL,
1503 	    note_procstat_osrel, p);
1504 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1505 	    __elfN(note_procstat_psstrings), p);
1506 	size += register_note(list, NT_PROCSTAT_AUXV,
1507 	    __elfN(note_procstat_auxv), p);
1508 
1509 	*sizep = size;
1510 }
1511 
1512 static void
1513 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1514     size_t notesz)
1515 {
1516 	Elf_Ehdr *ehdr;
1517 	Elf_Phdr *phdr;
1518 	struct phdr_closure phc;
1519 
1520 	ehdr = (Elf_Ehdr *)hdr;
1521 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1522 
1523 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1524 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1525 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1526 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1527 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1528 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1529 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1530 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1531 	ehdr->e_ident[EI_ABIVERSION] = 0;
1532 	ehdr->e_ident[EI_PAD] = 0;
1533 	ehdr->e_type = ET_CORE;
1534 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1535 	ehdr->e_machine = ELF_ARCH32;
1536 #else
1537 	ehdr->e_machine = ELF_ARCH;
1538 #endif
1539 	ehdr->e_version = EV_CURRENT;
1540 	ehdr->e_entry = 0;
1541 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1542 	ehdr->e_flags = 0;
1543 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1544 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1545 	ehdr->e_phnum = numsegs + 1;
1546 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1547 	ehdr->e_shnum = 0;
1548 	ehdr->e_shstrndx = SHN_UNDEF;
1549 
1550 	/*
1551 	 * Fill in the program header entries.
1552 	 */
1553 
1554 	/* The note segement. */
1555 	phdr->p_type = PT_NOTE;
1556 	phdr->p_offset = hdrsize;
1557 	phdr->p_vaddr = 0;
1558 	phdr->p_paddr = 0;
1559 	phdr->p_filesz = notesz;
1560 	phdr->p_memsz = 0;
1561 	phdr->p_flags = PF_R;
1562 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1563 	phdr++;
1564 
1565 	/* All the writable segments from the program. */
1566 	phc.phdr = phdr;
1567 	phc.offset = round_page(hdrsize + notesz);
1568 	each_writable_segment(td, cb_put_phdr, &phc);
1569 }
1570 
1571 static size_t
1572 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1573 {
1574 	struct note_info *ninfo;
1575 	size_t size, notesize;
1576 
1577 	size = 0;
1578 	out(arg, NULL, &size);
1579 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1580 	ninfo->type = type;
1581 	ninfo->outfunc = out;
1582 	ninfo->outarg = arg;
1583 	ninfo->outsize = size;
1584 	TAILQ_INSERT_TAIL(list, ninfo, link);
1585 
1586 	if (type == -1)
1587 		return (size);
1588 
1589 	notesize = sizeof(Elf_Note) +		/* note header */
1590 	    roundup2(8, ELF_NOTE_ROUNDSIZE) +	/* note name ("FreeBSD") */
1591 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1592 
1593 	return (notesize);
1594 }
1595 
1596 static void
1597 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1598 {
1599 	Elf_Note note;
1600 	ssize_t old_len;
1601 
1602 	if (ninfo->type == -1) {
1603 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1604 		return;
1605 	}
1606 
1607 	note.n_namesz = 8; /* strlen("FreeBSD") + 1 */
1608 	note.n_descsz = ninfo->outsize;
1609 	note.n_type = ninfo->type;
1610 
1611 	sbuf_bcat(sb, &note, sizeof(note));
1612 	sbuf_start_section(sb, &old_len);
1613 	sbuf_bcat(sb, "FreeBSD", note.n_namesz);
1614 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1615 	if (note.n_descsz == 0)
1616 		return;
1617 	sbuf_start_section(sb, &old_len);
1618 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1619 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1620 }
1621 
1622 /*
1623  * Miscellaneous note out functions.
1624  */
1625 
1626 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1627 #include <compat/freebsd32/freebsd32.h>
1628 
1629 typedef struct prstatus32 elf_prstatus_t;
1630 typedef struct prpsinfo32 elf_prpsinfo_t;
1631 typedef struct fpreg32 elf_prfpregset_t;
1632 typedef struct fpreg32 elf_fpregset_t;
1633 typedef struct reg32 elf_gregset_t;
1634 typedef struct thrmisc32 elf_thrmisc_t;
1635 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1636 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1637 typedef uint32_t elf_ps_strings_t;
1638 #else
1639 typedef prstatus_t elf_prstatus_t;
1640 typedef prpsinfo_t elf_prpsinfo_t;
1641 typedef prfpregset_t elf_prfpregset_t;
1642 typedef prfpregset_t elf_fpregset_t;
1643 typedef gregset_t elf_gregset_t;
1644 typedef thrmisc_t elf_thrmisc_t;
1645 #define ELF_KERN_PROC_MASK	0
1646 typedef struct kinfo_proc elf_kinfo_proc_t;
1647 typedef vm_offset_t elf_ps_strings_t;
1648 #endif
1649 
1650 static void
1651 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1652 {
1653 	struct proc *p;
1654 	elf_prpsinfo_t *psinfo;
1655 
1656 	p = (struct proc *)arg;
1657 	if (sb != NULL) {
1658 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1659 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1660 		psinfo->pr_version = PRPSINFO_VERSION;
1661 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1662 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1663 		/*
1664 		 * XXX - We don't fill in the command line arguments properly
1665 		 * yet.
1666 		 */
1667 		strlcpy(psinfo->pr_psargs, p->p_comm,
1668 		    sizeof(psinfo->pr_psargs));
1669 
1670 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1671 		free(psinfo, M_TEMP);
1672 	}
1673 	*sizep = sizeof(*psinfo);
1674 }
1675 
1676 static void
1677 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1678 {
1679 	struct thread *td;
1680 	elf_prstatus_t *status;
1681 
1682 	td = (struct thread *)arg;
1683 	if (sb != NULL) {
1684 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1685 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1686 		status->pr_version = PRSTATUS_VERSION;
1687 		status->pr_statussz = sizeof(elf_prstatus_t);
1688 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1689 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1690 		status->pr_osreldate = osreldate;
1691 		status->pr_cursig = td->td_proc->p_sig;
1692 		status->pr_pid = td->td_tid;
1693 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1694 		fill_regs32(td, &status->pr_reg);
1695 #else
1696 		fill_regs(td, &status->pr_reg);
1697 #endif
1698 		sbuf_bcat(sb, status, sizeof(*status));
1699 		free(status, M_TEMP);
1700 	}
1701 	*sizep = sizeof(*status);
1702 }
1703 
1704 static void
1705 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1706 {
1707 	struct thread *td;
1708 	elf_prfpregset_t *fpregset;
1709 
1710 	td = (struct thread *)arg;
1711 	if (sb != NULL) {
1712 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1713 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1714 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1715 		fill_fpregs32(td, fpregset);
1716 #else
1717 		fill_fpregs(td, fpregset);
1718 #endif
1719 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1720 		free(fpregset, M_TEMP);
1721 	}
1722 	*sizep = sizeof(*fpregset);
1723 }
1724 
1725 static void
1726 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1727 {
1728 	struct thread *td;
1729 	elf_thrmisc_t thrmisc;
1730 
1731 	td = (struct thread *)arg;
1732 	if (sb != NULL) {
1733 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1734 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1735 		strcpy(thrmisc.pr_tname, td->td_name);
1736 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1737 	}
1738 	*sizep = sizeof(thrmisc);
1739 }
1740 
1741 /*
1742  * Allow for MD specific notes, as well as any MD
1743  * specific preparations for writing MI notes.
1744  */
1745 static void
1746 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1747 {
1748 	struct thread *td;
1749 	void *buf;
1750 	size_t size;
1751 
1752 	td = (struct thread *)arg;
1753 	size = *sizep;
1754 	if (size != 0 && sb != NULL)
1755 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1756 	else
1757 		buf = NULL;
1758 	size = 0;
1759 	__elfN(dump_thread)(td, buf, &size);
1760 	KASSERT(*sizep == size, ("invalid size"));
1761 	if (size != 0 && sb != NULL)
1762 		sbuf_bcat(sb, buf, size);
1763 	free(buf, M_TEMP);
1764 	*sizep = size;
1765 }
1766 
1767 #ifdef KINFO_PROC_SIZE
1768 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1769 #endif
1770 
1771 static void
1772 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1773 {
1774 	struct proc *p;
1775 	size_t size;
1776 	int structsize;
1777 
1778 	p = (struct proc *)arg;
1779 	size = sizeof(structsize) + p->p_numthreads *
1780 	    sizeof(elf_kinfo_proc_t);
1781 
1782 	if (sb != NULL) {
1783 		KASSERT(*sizep == size, ("invalid size"));
1784 		structsize = sizeof(elf_kinfo_proc_t);
1785 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1786 		sx_slock(&proctree_lock);
1787 		PROC_LOCK(p);
1788 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1789 		sx_sunlock(&proctree_lock);
1790 	}
1791 	*sizep = size;
1792 }
1793 
1794 #ifdef KINFO_FILE_SIZE
1795 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1796 #endif
1797 
1798 static void
1799 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1800 {
1801 	struct proc *p;
1802 	size_t size;
1803 	int structsize;
1804 
1805 	p = (struct proc *)arg;
1806 	if (sb == NULL) {
1807 		size = 0;
1808 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1809 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1810 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1811 		PROC_LOCK(p);
1812 		kern_proc_filedesc_out(p, sb, -1);
1813 		sbuf_finish(sb);
1814 		sbuf_delete(sb);
1815 		*sizep = size;
1816 	} else {
1817 		structsize = sizeof(struct kinfo_file);
1818 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1819 		PROC_LOCK(p);
1820 		kern_proc_filedesc_out(p, sb, -1);
1821 	}
1822 }
1823 
1824 #ifdef KINFO_VMENTRY_SIZE
1825 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1826 #endif
1827 
1828 static void
1829 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1830 {
1831 	struct proc *p;
1832 	size_t size;
1833 	int structsize;
1834 
1835 	p = (struct proc *)arg;
1836 	if (sb == NULL) {
1837 		size = 0;
1838 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1839 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1840 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1841 		PROC_LOCK(p);
1842 		kern_proc_vmmap_out(p, sb);
1843 		sbuf_finish(sb);
1844 		sbuf_delete(sb);
1845 		*sizep = size;
1846 	} else {
1847 		structsize = sizeof(struct kinfo_vmentry);
1848 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1849 		PROC_LOCK(p);
1850 		kern_proc_vmmap_out(p, sb);
1851 	}
1852 }
1853 
1854 static void
1855 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1856 {
1857 	struct proc *p;
1858 	size_t size;
1859 	int structsize;
1860 
1861 	p = (struct proc *)arg;
1862 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1863 	if (sb != NULL) {
1864 		KASSERT(*sizep == size, ("invalid size"));
1865 		structsize = sizeof(gid_t);
1866 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1867 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1868 		    sizeof(gid_t));
1869 	}
1870 	*sizep = size;
1871 }
1872 
1873 static void
1874 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1875 {
1876 	struct proc *p;
1877 	size_t size;
1878 	int structsize;
1879 
1880 	p = (struct proc *)arg;
1881 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1882 	if (sb != NULL) {
1883 		KASSERT(*sizep == size, ("invalid size"));
1884 		structsize = sizeof(p->p_fd->fd_cmask);
1885 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1886 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1887 	}
1888 	*sizep = size;
1889 }
1890 
1891 static void
1892 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1893 {
1894 	struct proc *p;
1895 	struct rlimit rlim[RLIM_NLIMITS];
1896 	size_t size;
1897 	int structsize, i;
1898 
1899 	p = (struct proc *)arg;
1900 	size = sizeof(structsize) + sizeof(rlim);
1901 	if (sb != NULL) {
1902 		KASSERT(*sizep == size, ("invalid size"));
1903 		structsize = sizeof(rlim);
1904 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1905 		PROC_LOCK(p);
1906 		for (i = 0; i < RLIM_NLIMITS; i++)
1907 			lim_rlimit(p, i, &rlim[i]);
1908 		PROC_UNLOCK(p);
1909 		sbuf_bcat(sb, rlim, sizeof(rlim));
1910 	}
1911 	*sizep = size;
1912 }
1913 
1914 static void
1915 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1916 {
1917 	struct proc *p;
1918 	size_t size;
1919 	int structsize;
1920 
1921 	p = (struct proc *)arg;
1922 	size = sizeof(structsize) + sizeof(p->p_osrel);
1923 	if (sb != NULL) {
1924 		KASSERT(*sizep == size, ("invalid size"));
1925 		structsize = sizeof(p->p_osrel);
1926 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1927 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1928 	}
1929 	*sizep = size;
1930 }
1931 
1932 static void
1933 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1934 {
1935 	struct proc *p;
1936 	elf_ps_strings_t ps_strings;
1937 	size_t size;
1938 	int structsize;
1939 
1940 	p = (struct proc *)arg;
1941 	size = sizeof(structsize) + sizeof(ps_strings);
1942 	if (sb != NULL) {
1943 		KASSERT(*sizep == size, ("invalid size"));
1944 		structsize = sizeof(ps_strings);
1945 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1946 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1947 #else
1948 		ps_strings = p->p_sysent->sv_psstrings;
1949 #endif
1950 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1951 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
1952 	}
1953 	*sizep = size;
1954 }
1955 
1956 static void
1957 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
1958 {
1959 	struct proc *p;
1960 	size_t size;
1961 	int structsize;
1962 
1963 	p = (struct proc *)arg;
1964 	if (sb == NULL) {
1965 		size = 0;
1966 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1967 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1968 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1969 		PHOLD(p);
1970 		proc_getauxv(curthread, p, sb);
1971 		PRELE(p);
1972 		sbuf_finish(sb);
1973 		sbuf_delete(sb);
1974 		*sizep = size;
1975 	} else {
1976 		structsize = sizeof(Elf_Auxinfo);
1977 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1978 		PHOLD(p);
1979 		proc_getauxv(curthread, p, sb);
1980 		PRELE(p);
1981 	}
1982 }
1983 
1984 static boolean_t
1985 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
1986     int32_t *osrel, const Elf_Phdr *pnote)
1987 {
1988 	const Elf_Note *note, *note0, *note_end;
1989 	const char *note_name;
1990 	int i;
1991 
1992 	if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
1993 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
1994 		return (FALSE);
1995 
1996 	note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1997 	note_end = (const Elf_Note *)(imgp->image_header +
1998 	    pnote->p_offset + pnote->p_filesz);
1999 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2000 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2001 		    (const char *)note < sizeof(Elf_Note))
2002 			return (FALSE);
2003 		if (note->n_namesz != checknote->hdr.n_namesz ||
2004 		    note->n_descsz != checknote->hdr.n_descsz ||
2005 		    note->n_type != checknote->hdr.n_type)
2006 			goto nextnote;
2007 		note_name = (const char *)(note + 1);
2008 		if (note_name + checknote->hdr.n_namesz >=
2009 		    (const char *)note_end || strncmp(checknote->vendor,
2010 		    note_name, checknote->hdr.n_namesz) != 0)
2011 			goto nextnote;
2012 
2013 		/*
2014 		 * Fetch the osreldate for binary
2015 		 * from the ELF OSABI-note if necessary.
2016 		 */
2017 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2018 		    checknote->trans_osrel != NULL)
2019 			return (checknote->trans_osrel(note, osrel));
2020 		return (TRUE);
2021 
2022 nextnote:
2023 		note = (const Elf_Note *)((const char *)(note + 1) +
2024 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2025 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2026 	}
2027 
2028 	return (FALSE);
2029 }
2030 
2031 /*
2032  * Try to find the appropriate ABI-note section for checknote,
2033  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2034  * first page of the image is searched, the same as for headers.
2035  */
2036 static boolean_t
2037 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2038     int32_t *osrel)
2039 {
2040 	const Elf_Phdr *phdr;
2041 	const Elf_Ehdr *hdr;
2042 	int i;
2043 
2044 	hdr = (const Elf_Ehdr *)imgp->image_header;
2045 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2046 
2047 	for (i = 0; i < hdr->e_phnum; i++) {
2048 		if (phdr[i].p_type == PT_NOTE &&
2049 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2050 			return (TRUE);
2051 	}
2052 	return (FALSE);
2053 
2054 }
2055 
2056 /*
2057  * Tell kern_execve.c about it, with a little help from the linker.
2058  */
2059 static struct execsw __elfN(execsw) = {
2060 	__CONCAT(exec_, __elfN(imgact)),
2061 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2062 };
2063 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2064 
2065 #ifdef COMPRESS_USER_CORES
2066 /*
2067  * Compress and write out a core segment for a user process.
2068  *
2069  * 'inbuf' is the starting address of a VM segment in the process' address
2070  * space that is to be compressed and written out to the core file.  'dest_buf'
2071  * is a buffer in the kernel's address space.  The segment is copied from
2072  * 'inbuf' to 'dest_buf' first before being processed by the compression
2073  * routine gzwrite().  This copying is necessary because the content of the VM
2074  * segment may change between the compression pass and the crc-computation pass
2075  * in gzwrite().  This is because realtime threads may preempt the UNIX kernel.
2076  *
2077  * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2078  */
2079 static int
2080 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2081     struct thread *td)
2082 {
2083 	int len_compressed;
2084 	int error = 0;
2085 	unsigned int chunk_len;
2086 
2087 	while (len) {
2088 		if (inbuf != NULL) {
2089 			chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2090 			copyin(inbuf, dest_buf, chunk_len);
2091 			inbuf += chunk_len;
2092 		} else {
2093 			chunk_len = len;
2094 		}
2095 		len_compressed = gzwrite(file, dest_buf, chunk_len);
2096 
2097 		EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2098 
2099 		if ((unsigned int)len_compressed != chunk_len) {
2100 			log(LOG_WARNING,
2101 			    "compress_core: length mismatch (0x%x returned, "
2102 			    "0x%x expected)\n", len_compressed, chunk_len);
2103 			EVENTHANDLER_INVOKE(app_coredump_error, td,
2104 			    "compress_core: length mismatch %x -> %x",
2105 			    chunk_len, len_compressed);
2106 			error = EFAULT;
2107 			break;
2108 		}
2109 		len -= chunk_len;
2110 		maybe_yield();
2111 	}
2112 
2113 	return (error);
2114 }
2115 #endif /* COMPRESS_USER_CORES */
2116 
2117 static vm_prot_t
2118 __elfN(trans_prot)(Elf_Word flags)
2119 {
2120 	vm_prot_t prot;
2121 
2122 	prot = 0;
2123 	if (flags & PF_X)
2124 		prot |= VM_PROT_EXECUTE;
2125 	if (flags & PF_W)
2126 		prot |= VM_PROT_WRITE;
2127 	if (flags & PF_R)
2128 		prot |= VM_PROT_READ;
2129 #if __ELF_WORD_SIZE == 32
2130 #if defined(__amd64__)
2131 	if (i386_read_exec && (flags & PF_R))
2132 		prot |= VM_PROT_EXECUTE;
2133 #endif
2134 #endif
2135 	return (prot);
2136 }
2137 
2138 static Elf_Word
2139 __elfN(untrans_prot)(vm_prot_t prot)
2140 {
2141 	Elf_Word flags;
2142 
2143 	flags = 0;
2144 	if (prot & VM_PROT_EXECUTE)
2145 		flags |= PF_X;
2146 	if (prot & VM_PROT_READ)
2147 		flags |= PF_R;
2148 	if (prot & VM_PROT_WRITE)
2149 		flags |= PF_W;
2150 	return (flags);
2151 }
2152