xref: /freebsd/sys/kern/imgact_elf.c (revision 43faedc1339a9624c7acedb7f3e5624e64da5b99)
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 #ifdef __arm__
84 #include <machine/acle-compat.h>
85 #endif
86 
87 #define ELF_NOTE_ROUNDSIZE	4
88 #define OLD_EI_BRAND	8
89 
90 static int __elfN(check_header)(const Elf_Ehdr *hdr);
91 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
92     const char *interp, int interp_name_len, int32_t *osrel);
93 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94     u_long *entry, size_t pagesize);
95 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
96     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
97     size_t pagesize);
98 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
99 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
100     int32_t *osrel);
101 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
102 static boolean_t __elfN(check_note)(struct image_params *imgp,
103     Elf_Brandnote *checknote, int32_t *osrel);
104 static vm_prot_t __elfN(trans_prot)(Elf_Word);
105 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
106 
107 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108     "");
109 
110 #define	CORE_BUF_SIZE	(16 * 1024)
111 
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119     &elf_legacy_coredump, 0, "");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
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	= NT_FREEBSD_ABI_TAG,
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 			/* Give brand a chance to veto check_note's guess */
281 			if (ret && bi->header_supported)
282 				ret = bi->header_supported(imgp);
283 			if (ret)
284 				return (bi);
285 		}
286 	}
287 
288 	/* If the executable has a brand, search for it in the brand list. */
289 	for (i = 0; i < MAX_BRANDS; i++) {
290 		bi = elf_brand_list[i];
291 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
292 			continue;
293 		if (hdr->e_machine == bi->machine &&
294 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
295 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
296 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
297 			/* Looks good, but give brand a chance to veto */
298 			if (!bi->header_supported || bi->header_supported(imgp))
299 				return (bi);
300 		}
301 	}
302 
303 	/* No known brand, see if the header is recognized by any brand */
304 	for (i = 0; i < MAX_BRANDS; i++) {
305 		bi = elf_brand_list[i];
306 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
307 		    bi->header_supported == NULL)
308 			continue;
309 		if (hdr->e_machine == bi->machine) {
310 			ret = bi->header_supported(imgp);
311 			if (ret)
312 				return (bi);
313 		}
314 	}
315 
316 	/* Lacking a known brand, search for a recognized interpreter. */
317 	if (interp != NULL) {
318 		for (i = 0; i < MAX_BRANDS; i++) {
319 			bi = elf_brand_list[i];
320 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
321 				continue;
322 			if (hdr->e_machine == bi->machine &&
323 			    /* ELF image p_filesz includes terminating zero */
324 			    strlen(bi->interp_path) + 1 == interp_name_len &&
325 			    strncmp(interp, bi->interp_path, interp_name_len)
326 			    == 0)
327 				return (bi);
328 		}
329 	}
330 
331 	/* Lacking a recognized interpreter, try the default brand */
332 	for (i = 0; i < MAX_BRANDS; i++) {
333 		bi = elf_brand_list[i];
334 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
335 			continue;
336 		if (hdr->e_machine == bi->machine &&
337 		    __elfN(fallback_brand) == bi->brand)
338 			return (bi);
339 	}
340 	return (NULL);
341 }
342 
343 static int
344 __elfN(check_header)(const Elf_Ehdr *hdr)
345 {
346 	Elf_Brandinfo *bi;
347 	int i;
348 
349 	if (!IS_ELF(*hdr) ||
350 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
351 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
352 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
353 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
354 	    hdr->e_version != ELF_TARG_VER)
355 		return (ENOEXEC);
356 
357 	/*
358 	 * Make sure we have at least one brand for this machine.
359 	 */
360 
361 	for (i = 0; i < MAX_BRANDS; i++) {
362 		bi = elf_brand_list[i];
363 		if (bi != NULL && bi->machine == hdr->e_machine)
364 			break;
365 	}
366 	if (i == MAX_BRANDS)
367 		return (ENOEXEC);
368 
369 	return (0);
370 }
371 
372 static int
373 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
374     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
375 {
376 	struct sf_buf *sf;
377 	int error;
378 	vm_offset_t off;
379 
380 	/*
381 	 * Create the page if it doesn't exist yet. Ignore errors.
382 	 */
383 	vm_map_lock(map);
384 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
385 	    VM_PROT_ALL, VM_PROT_ALL, 0);
386 	vm_map_unlock(map);
387 
388 	/*
389 	 * Find the page from the underlying object.
390 	 */
391 	if (object) {
392 		sf = vm_imgact_map_page(object, offset);
393 		if (sf == NULL)
394 			return (KERN_FAILURE);
395 		off = offset - trunc_page(offset);
396 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
397 		    end - start);
398 		vm_imgact_unmap_page(sf);
399 		if (error) {
400 			return (KERN_FAILURE);
401 		}
402 	}
403 
404 	return (KERN_SUCCESS);
405 }
406 
407 static int
408 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
409     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
410 {
411 	struct sf_buf *sf;
412 	vm_offset_t off;
413 	vm_size_t sz;
414 	int error, rv;
415 
416 	if (start != trunc_page(start)) {
417 		rv = __elfN(map_partial)(map, object, offset, start,
418 		    round_page(start), prot);
419 		if (rv)
420 			return (rv);
421 		offset += round_page(start) - start;
422 		start = round_page(start);
423 	}
424 	if (end != round_page(end)) {
425 		rv = __elfN(map_partial)(map, object, offset +
426 		    trunc_page(end) - start, trunc_page(end), end, prot);
427 		if (rv)
428 			return (rv);
429 		end = trunc_page(end);
430 	}
431 	if (end > start) {
432 		if (offset & PAGE_MASK) {
433 			/*
434 			 * The mapping is not page aligned. This means we have
435 			 * to copy the data. Sigh.
436 			 */
437 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
438 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
439 			    0);
440 			if (rv)
441 				return (rv);
442 			if (object == NULL)
443 				return (KERN_SUCCESS);
444 			for (; start < end; start += sz) {
445 				sf = vm_imgact_map_page(object, offset);
446 				if (sf == NULL)
447 					return (KERN_FAILURE);
448 				off = offset - trunc_page(offset);
449 				sz = end - start;
450 				if (sz > PAGE_SIZE - off)
451 					sz = PAGE_SIZE - off;
452 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
453 				    (caddr_t)start, sz);
454 				vm_imgact_unmap_page(sf);
455 				if (error) {
456 					return (KERN_FAILURE);
457 				}
458 				offset += sz;
459 			}
460 			rv = KERN_SUCCESS;
461 		} else {
462 			vm_object_reference(object);
463 			vm_map_lock(map);
464 			rv = vm_map_insert(map, object, offset, start, end,
465 			    prot, VM_PROT_ALL, cow);
466 			vm_map_unlock(map);
467 			if (rv != KERN_SUCCESS)
468 				vm_object_deallocate(object);
469 		}
470 		return (rv);
471 	} else {
472 		return (KERN_SUCCESS);
473 	}
474 }
475 
476 static int
477 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
478     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
479     size_t pagesize)
480 {
481 	struct sf_buf *sf;
482 	size_t map_len;
483 	vm_map_t map;
484 	vm_object_t object;
485 	vm_offset_t map_addr;
486 	int error, rv, cow;
487 	size_t copy_len;
488 	vm_offset_t file_addr;
489 
490 	/*
491 	 * It's necessary to fail if the filsz + offset taken from the
492 	 * header is greater than the actual file pager object's size.
493 	 * If we were to allow this, then the vm_map_find() below would
494 	 * walk right off the end of the file object and into the ether.
495 	 *
496 	 * While I'm here, might as well check for something else that
497 	 * is invalid: filsz cannot be greater than memsz.
498 	 */
499 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
500 		uprintf("elf_load_section: truncated ELF file\n");
501 		return (ENOEXEC);
502 	}
503 
504 	object = imgp->object;
505 	map = &imgp->proc->p_vmspace->vm_map;
506 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
507 	file_addr = trunc_page_ps(offset, pagesize);
508 
509 	/*
510 	 * We have two choices.  We can either clear the data in the last page
511 	 * of an oversized mapping, or we can start the anon mapping a page
512 	 * early and copy the initialized data into that first page.  We
513 	 * choose the second..
514 	 */
515 	if (memsz > filsz)
516 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
517 	else
518 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
519 
520 	if (map_len != 0) {
521 		/* cow flags: don't dump readonly sections in core */
522 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
523 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
524 
525 		rv = __elfN(map_insert)(map,
526 				      object,
527 				      file_addr,	/* file offset */
528 				      map_addr,		/* virtual start */
529 				      map_addr + map_len,/* virtual end */
530 				      prot,
531 				      cow);
532 		if (rv != KERN_SUCCESS)
533 			return (EINVAL);
534 
535 		/* we can stop now if we've covered it all */
536 		if (memsz == filsz) {
537 			return (0);
538 		}
539 	}
540 
541 
542 	/*
543 	 * We have to get the remaining bit of the file into the first part
544 	 * of the oversized map segment.  This is normally because the .data
545 	 * segment in the file is extended to provide bss.  It's a neat idea
546 	 * to try and save a page, but it's a pain in the behind to implement.
547 	 */
548 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
549 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
550 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
551 	    map_addr;
552 
553 	/* This had damn well better be true! */
554 	if (map_len != 0) {
555 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
556 		    map_len, VM_PROT_ALL, 0);
557 		if (rv != KERN_SUCCESS) {
558 			return (EINVAL);
559 		}
560 	}
561 
562 	if (copy_len != 0) {
563 		vm_offset_t off;
564 
565 		sf = vm_imgact_map_page(object, offset + filsz);
566 		if (sf == NULL)
567 			return (EIO);
568 
569 		/* send the page fragment to user space */
570 		off = trunc_page_ps(offset + filsz, pagesize) -
571 		    trunc_page(offset + filsz);
572 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
573 		    (caddr_t)map_addr, copy_len);
574 		vm_imgact_unmap_page(sf);
575 		if (error) {
576 			return (error);
577 		}
578 	}
579 
580 	/*
581 	 * set it to the specified protection.
582 	 * XXX had better undo the damage from pasting over the cracks here!
583 	 */
584 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
585 	    map_len), prot, FALSE);
586 
587 	return (0);
588 }
589 
590 /*
591  * Load the file "file" into memory.  It may be either a shared object
592  * or an executable.
593  *
594  * The "addr" reference parameter is in/out.  On entry, it specifies
595  * the address where a shared object should be loaded.  If the file is
596  * an executable, this value is ignored.  On exit, "addr" specifies
597  * where the file was actually loaded.
598  *
599  * The "entry" reference parameter is out only.  On exit, it specifies
600  * the entry point for the loaded file.
601  */
602 static int
603 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
604 	u_long *entry, size_t pagesize)
605 {
606 	struct {
607 		struct nameidata nd;
608 		struct vattr attr;
609 		struct image_params image_params;
610 	} *tempdata;
611 	const Elf_Ehdr *hdr = NULL;
612 	const Elf_Phdr *phdr = NULL;
613 	struct nameidata *nd;
614 	struct vattr *attr;
615 	struct image_params *imgp;
616 	vm_prot_t prot;
617 	u_long rbase;
618 	u_long base_addr = 0;
619 	int error, i, numsegs;
620 
621 #ifdef CAPABILITY_MODE
622 	/*
623 	 * XXXJA: This check can go away once we are sufficiently confident
624 	 * that the checks in namei() are correct.
625 	 */
626 	if (IN_CAPABILITY_MODE(curthread))
627 		return (ECAPMODE);
628 #endif
629 
630 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
631 	nd = &tempdata->nd;
632 	attr = &tempdata->attr;
633 	imgp = &tempdata->image_params;
634 
635 	/*
636 	 * Initialize part of the common data
637 	 */
638 	imgp->proc = p;
639 	imgp->attr = attr;
640 	imgp->firstpage = NULL;
641 	imgp->image_header = NULL;
642 	imgp->object = NULL;
643 	imgp->execlabel = NULL;
644 
645 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
646 	if ((error = namei(nd)) != 0) {
647 		nd->ni_vp = NULL;
648 		goto fail;
649 	}
650 	NDFREE(nd, NDF_ONLY_PNBUF);
651 	imgp->vp = nd->ni_vp;
652 
653 	/*
654 	 * Check permissions, modes, uid, etc on the file, and "open" it.
655 	 */
656 	error = exec_check_permissions(imgp);
657 	if (error)
658 		goto fail;
659 
660 	error = exec_map_first_page(imgp);
661 	if (error)
662 		goto fail;
663 
664 	/*
665 	 * Also make certain that the interpreter stays the same, so set
666 	 * its VV_TEXT flag, too.
667 	 */
668 	VOP_SET_TEXT(nd->ni_vp);
669 
670 	imgp->object = nd->ni_vp->v_object;
671 
672 	hdr = (const Elf_Ehdr *)imgp->image_header;
673 	if ((error = __elfN(check_header)(hdr)) != 0)
674 		goto fail;
675 	if (hdr->e_type == ET_DYN)
676 		rbase = *addr;
677 	else if (hdr->e_type == ET_EXEC)
678 		rbase = 0;
679 	else {
680 		error = ENOEXEC;
681 		goto fail;
682 	}
683 
684 	/* Only support headers that fit within first page for now      */
685 	if ((hdr->e_phoff > PAGE_SIZE) ||
686 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
687 		error = ENOEXEC;
688 		goto fail;
689 	}
690 
691 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
692 	if (!aligned(phdr, Elf_Addr)) {
693 		error = ENOEXEC;
694 		goto fail;
695 	}
696 
697 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
698 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
699 			/* Loadable segment */
700 			prot = __elfN(trans_prot)(phdr[i].p_flags);
701 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
702 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
703 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
704 			if (error != 0)
705 				goto fail;
706 			/*
707 			 * Establish the base address if this is the
708 			 * first segment.
709 			 */
710 			if (numsegs == 0)
711   				base_addr = trunc_page(phdr[i].p_vaddr +
712 				    rbase);
713 			numsegs++;
714 		}
715 	}
716 	*addr = base_addr;
717 	*entry = (unsigned long)hdr->e_entry + rbase;
718 
719 fail:
720 	if (imgp->firstpage)
721 		exec_unmap_first_page(imgp);
722 
723 	if (nd->ni_vp)
724 		vput(nd->ni_vp);
725 
726 	free(tempdata, M_TEMP);
727 
728 	return (error);
729 }
730 
731 static int
732 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
733 {
734 	struct thread *td;
735 	const Elf_Ehdr *hdr;
736 	const Elf_Phdr *phdr;
737 	Elf_Auxargs *elf_auxargs;
738 	struct vmspace *vmspace;
739 	const char *err_str, *newinterp;
740 	char *interp, *interp_buf, *path;
741 	Elf_Brandinfo *brand_info;
742 	struct sysentvec *sv;
743 	vm_prot_t prot;
744 	u_long text_size, data_size, total_size, text_addr, data_addr;
745 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
746 	int32_t osrel;
747 	int error, i, n, interp_name_len, have_interp;
748 
749 	hdr = (const Elf_Ehdr *)imgp->image_header;
750 
751 	/*
752 	 * Do we have a valid ELF header ?
753 	 *
754 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
755 	 * if particular brand doesn't support it.
756 	 */
757 	if (__elfN(check_header)(hdr) != 0 ||
758 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
759 		return (-1);
760 
761 	/*
762 	 * From here on down, we return an errno, not -1, as we've
763 	 * detected an ELF file.
764 	 */
765 
766 	if ((hdr->e_phoff > PAGE_SIZE) ||
767 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
768 		/* Only support headers in first page for now */
769 		uprintf("Program headers not in the first page\n");
770 		return (ENOEXEC);
771 	}
772 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
773 	if (!aligned(phdr, Elf_Addr)) {
774 		uprintf("Unaligned program headers\n");
775 		return (ENOEXEC);
776 	}
777 
778 	n = error = 0;
779 	baddr = 0;
780 	osrel = 0;
781 	text_size = data_size = total_size = text_addr = data_addr = 0;
782 	entry = proghdr = 0;
783 	interp_name_len = 0;
784 	err_str = newinterp = NULL;
785 	interp = interp_buf = NULL;
786 	td = curthread;
787 
788 	for (i = 0; i < hdr->e_phnum; i++) {
789 		switch (phdr[i].p_type) {
790 		case PT_LOAD:
791 			if (n == 0)
792 				baddr = phdr[i].p_vaddr;
793 			n++;
794 			break;
795 		case PT_INTERP:
796 			/* Path to interpreter */
797 			if (phdr[i].p_filesz > MAXPATHLEN) {
798 				uprintf("Invalid PT_INTERP\n");
799 				error = ENOEXEC;
800 				goto ret;
801 			}
802 			if (interp != NULL) {
803 				uprintf("Multiple PT_INTERP headers\n");
804 				error = ENOEXEC;
805 				goto ret;
806 			}
807 			interp_name_len = phdr[i].p_filesz;
808 			if (phdr[i].p_offset > PAGE_SIZE ||
809 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
810 				VOP_UNLOCK(imgp->vp, 0);
811 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
812 				    M_WAITOK);
813 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
814 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
815 				    interp_name_len, phdr[i].p_offset,
816 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
817 				    NOCRED, NULL, td);
818 				if (error != 0) {
819 					uprintf("i/o error PT_INTERP\n");
820 					goto ret;
821 				}
822 				interp_buf[interp_name_len] = '\0';
823 				interp = interp_buf;
824 			} else {
825 				interp = __DECONST(char *, imgp->image_header) +
826 				    phdr[i].p_offset;
827 			}
828 			break;
829 		case PT_GNU_STACK:
830 			if (__elfN(nxstack))
831 				imgp->stack_prot =
832 				    __elfN(trans_prot)(phdr[i].p_flags);
833 			imgp->stack_sz = phdr[i].p_memsz;
834 			break;
835 		}
836 	}
837 
838 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
839 	    &osrel);
840 	if (brand_info == NULL) {
841 		uprintf("ELF binary type \"%u\" not known.\n",
842 		    hdr->e_ident[EI_OSABI]);
843 		error = ENOEXEC;
844 		goto ret;
845 	}
846 	if (hdr->e_type == ET_DYN) {
847 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
848 			uprintf("Cannot execute shared object\n");
849 			error = ENOEXEC;
850 			goto ret;
851 		}
852 		/*
853 		 * Honour the base load address from the dso if it is
854 		 * non-zero for some reason.
855 		 */
856 		if (baddr == 0)
857 			et_dyn_addr = ET_DYN_LOAD_ADDR;
858 		else
859 			et_dyn_addr = 0;
860 	} else
861 		et_dyn_addr = 0;
862 	sv = brand_info->sysvec;
863 	if (interp != NULL && brand_info->interp_newpath != NULL)
864 		newinterp = brand_info->interp_newpath;
865 
866 	/*
867 	 * Avoid a possible deadlock if the current address space is destroyed
868 	 * and that address space maps the locked vnode.  In the common case,
869 	 * the locked vnode's v_usecount is decremented but remains greater
870 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
871 	 * However, in cases where the vnode lock is external, such as nullfs,
872 	 * v_usecount may become zero.
873 	 *
874 	 * The VV_TEXT flag prevents modifications to the executable while
875 	 * the vnode is unlocked.
876 	 */
877 	VOP_UNLOCK(imgp->vp, 0);
878 
879 	error = exec_new_vmspace(imgp, sv);
880 	imgp->proc->p_sysent = sv;
881 
882 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
883 	if (error != 0)
884 		goto ret;
885 
886 	for (i = 0; i < hdr->e_phnum; i++) {
887 		switch (phdr[i].p_type) {
888 		case PT_LOAD:	/* Loadable segment */
889 			if (phdr[i].p_memsz == 0)
890 				break;
891 			prot = __elfN(trans_prot)(phdr[i].p_flags);
892 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
893 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
894 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
895 			    sv->sv_pagesize);
896 			if (error != 0)
897 				goto ret;
898 
899 			/*
900 			 * If this segment contains the program headers,
901 			 * remember their virtual address for the AT_PHDR
902 			 * aux entry. Static binaries don't usually include
903 			 * a PT_PHDR entry.
904 			 */
905 			if (phdr[i].p_offset == 0 &&
906 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
907 				<= phdr[i].p_filesz)
908 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
909 				    et_dyn_addr;
910 
911 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
912 			seg_size = round_page(phdr[i].p_memsz +
913 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
914 
915 			/*
916 			 * Make the largest executable segment the official
917 			 * text segment and all others data.
918 			 *
919 			 * Note that obreak() assumes that data_addr +
920 			 * data_size == end of data load area, and the ELF
921 			 * file format expects segments to be sorted by
922 			 * address.  If multiple data segments exist, the
923 			 * last one will be used.
924 			 */
925 
926 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
927 				text_size = seg_size;
928 				text_addr = seg_addr;
929 			} else {
930 				data_size = seg_size;
931 				data_addr = seg_addr;
932 			}
933 			total_size += seg_size;
934 			break;
935 		case PT_PHDR: 	/* Program header table info */
936 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
937 			break;
938 		default:
939 			break;
940 		}
941 	}
942 
943 	if (data_addr == 0 && data_size == 0) {
944 		data_addr = text_addr;
945 		data_size = text_size;
946 	}
947 
948 	entry = (u_long)hdr->e_entry + et_dyn_addr;
949 
950 	/*
951 	 * Check limits.  It should be safe to check the
952 	 * limits after loading the segments since we do
953 	 * not actually fault in all the segments pages.
954 	 */
955 	PROC_LOCK(imgp->proc);
956 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
957 		err_str = "Data segment size exceeds process limit";
958 	else if (text_size > maxtsiz)
959 		err_str = "Text segment size exceeds system limit";
960 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
961 		err_str = "Total segment size exceeds process limit";
962 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
963 		err_str = "Data segment size exceeds resource limit";
964 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
965 		err_str = "Total segment size exceeds resource limit";
966 	if (err_str != NULL) {
967 		PROC_UNLOCK(imgp->proc);
968 		uprintf("%s\n", err_str);
969 		error = ENOMEM;
970 		goto ret;
971 	}
972 
973 	vmspace = imgp->proc->p_vmspace;
974 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
975 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
976 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
977 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
978 
979 	/*
980 	 * We load the dynamic linker where a userland call
981 	 * to mmap(0, ...) would put it.  The rationale behind this
982 	 * calculation is that it leaves room for the heap to grow to
983 	 * its maximum allowed size.
984 	 */
985 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
986 	    RLIMIT_DATA));
987 	PROC_UNLOCK(imgp->proc);
988 
989 	imgp->entry_addr = entry;
990 
991 	if (interp != NULL) {
992 		have_interp = FALSE;
993 		VOP_UNLOCK(imgp->vp, 0);
994 		if (brand_info->emul_path != NULL &&
995 		    brand_info->emul_path[0] != '\0') {
996 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
997 			snprintf(path, MAXPATHLEN, "%s%s",
998 			    brand_info->emul_path, interp);
999 			error = __elfN(load_file)(imgp->proc, path, &addr,
1000 			    &imgp->entry_addr, sv->sv_pagesize);
1001 			free(path, M_TEMP);
1002 			if (error == 0)
1003 				have_interp = TRUE;
1004 		}
1005 		if (!have_interp && newinterp != NULL &&
1006 		    (brand_info->interp_path == NULL ||
1007 		    strcmp(interp, brand_info->interp_path) == 0)) {
1008 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1009 			    &imgp->entry_addr, sv->sv_pagesize);
1010 			if (error == 0)
1011 				have_interp = TRUE;
1012 		}
1013 		if (!have_interp) {
1014 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1015 			    &imgp->entry_addr, sv->sv_pagesize);
1016 		}
1017 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1018 		if (error != 0) {
1019 			uprintf("ELF interpreter %s not found, error %d\n",
1020 			    interp, error);
1021 			goto ret;
1022 		}
1023 	} else
1024 		addr = et_dyn_addr;
1025 
1026 	/*
1027 	 * Construct auxargs table (used by the fixup routine)
1028 	 */
1029 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1030 	elf_auxargs->execfd = -1;
1031 	elf_auxargs->phdr = proghdr;
1032 	elf_auxargs->phent = hdr->e_phentsize;
1033 	elf_auxargs->phnum = hdr->e_phnum;
1034 	elf_auxargs->pagesz = PAGE_SIZE;
1035 	elf_auxargs->base = addr;
1036 	elf_auxargs->flags = 0;
1037 	elf_auxargs->entry = entry;
1038 	elf_auxargs->hdr_eflags = hdr->e_flags;
1039 
1040 	imgp->auxargs = elf_auxargs;
1041 	imgp->interpreted = 0;
1042 	imgp->reloc_base = addr;
1043 	imgp->proc->p_osrel = osrel;
1044 
1045  ret:
1046 	free(interp_buf, M_TEMP);
1047 	return (error);
1048 }
1049 
1050 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1051 
1052 int
1053 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1054 {
1055 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1056 	Elf_Addr *base;
1057 	Elf_Addr *pos;
1058 
1059 	base = (Elf_Addr *)*stack_base;
1060 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1061 
1062 	if (args->execfd != -1)
1063 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1064 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1065 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1066 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1067 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1068 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1069 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1070 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1071 #ifdef AT_EHDRFLAGS
1072 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1073 #endif
1074 	if (imgp->execpathp != 0)
1075 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1076 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1077 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1078 	if (imgp->canary != 0) {
1079 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1080 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1081 	}
1082 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1083 	if (imgp->pagesizes != 0) {
1084 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1085 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1086 	}
1087 	if (imgp->sysent->sv_timekeep_base != 0) {
1088 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1089 		    imgp->sysent->sv_timekeep_base);
1090 	}
1091 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1092 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1093 	    imgp->sysent->sv_stackprot);
1094 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1095 
1096 	free(imgp->auxargs, M_TEMP);
1097 	imgp->auxargs = NULL;
1098 
1099 	base--;
1100 	suword(base, (long)imgp->args->argc);
1101 	*stack_base = (register_t *)base;
1102 	return (0);
1103 }
1104 
1105 /*
1106  * Code for generating ELF core dumps.
1107  */
1108 
1109 typedef void (*segment_callback)(vm_map_entry_t, void *);
1110 
1111 /* Closure for cb_put_phdr(). */
1112 struct phdr_closure {
1113 	Elf_Phdr *phdr;		/* Program header to fill in */
1114 	Elf_Off offset;		/* Offset of segment in core file */
1115 };
1116 
1117 /* Closure for cb_size_segment(). */
1118 struct sseg_closure {
1119 	int count;		/* Count of writable segments. */
1120 	size_t size;		/* Total size of all writable segments. */
1121 };
1122 
1123 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1124 
1125 struct note_info {
1126 	int		type;		/* Note type. */
1127 	outfunc_t 	outfunc; 	/* Output function. */
1128 	void		*outarg;	/* Argument for the output function. */
1129 	size_t		outsize;	/* Output size. */
1130 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1131 };
1132 
1133 TAILQ_HEAD(note_info_list, note_info);
1134 
1135 /* Coredump output parameters. */
1136 struct coredump_params {
1137 	off_t		offset;
1138 	struct ucred	*active_cred;
1139 	struct ucred	*file_cred;
1140 	struct thread	*td;
1141 	struct vnode	*vp;
1142 	struct gzio_stream *gzs;
1143 };
1144 
1145 static void cb_put_phdr(vm_map_entry_t, void *);
1146 static void cb_size_segment(vm_map_entry_t, void *);
1147 static int core_write(struct coredump_params *, void *, size_t, off_t,
1148     enum uio_seg);
1149 static void each_writable_segment(struct thread *, segment_callback, void *);
1150 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1151     struct note_info_list *, size_t);
1152 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1153     size_t *);
1154 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1155 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1156 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1157 static int sbuf_drain_core_output(void *, const char *, int);
1158 static int sbuf_drain_count(void *arg, const char *data, int len);
1159 
1160 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1161 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1162 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1163 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1164 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1165 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1166 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1167 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1168 static void note_procstat_files(void *, struct sbuf *, size_t *);
1169 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1170 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1171 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1172 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1173 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1174 
1175 #ifdef GZIO
1176 extern int compress_user_cores_gzlevel;
1177 
1178 /*
1179  * Write out a core segment to the compression stream.
1180  */
1181 static int
1182 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1183 {
1184 	u_int chunk_len;
1185 	int error;
1186 
1187 	while (len > 0) {
1188 		chunk_len = MIN(len, CORE_BUF_SIZE);
1189 		copyin(base, buf, chunk_len);
1190 		error = gzio_write(p->gzs, buf, chunk_len);
1191 		if (error != 0)
1192 			break;
1193 		base += chunk_len;
1194 		len -= chunk_len;
1195 	}
1196 	return (error);
1197 }
1198 
1199 static int
1200 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1201 {
1202 
1203 	return (core_write((struct coredump_params *)arg, base, len, offset,
1204 	    UIO_SYSSPACE));
1205 }
1206 #endif /* GZIO */
1207 
1208 static int
1209 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1210     enum uio_seg seg)
1211 {
1212 
1213 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1214 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1215 	    p->active_cred, p->file_cred, NULL, p->td));
1216 }
1217 
1218 static int
1219 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1220     void *tmpbuf)
1221 {
1222 
1223 #ifdef GZIO
1224 	if (p->gzs != NULL)
1225 		return (compress_chunk(p, base, tmpbuf, len));
1226 #endif
1227 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1228 }
1229 
1230 /*
1231  * Drain into a core file.
1232  */
1233 static int
1234 sbuf_drain_core_output(void *arg, const char *data, int len)
1235 {
1236 	struct coredump_params *p;
1237 	int error, locked;
1238 
1239 	p = (struct coredump_params *)arg;
1240 
1241 	/*
1242 	 * Some kern_proc out routines that print to this sbuf may
1243 	 * call us with the process lock held. Draining with the
1244 	 * non-sleepable lock held is unsafe. The lock is needed for
1245 	 * those routines when dumping a live process. In our case we
1246 	 * can safely release the lock before draining and acquire
1247 	 * again after.
1248 	 */
1249 	locked = PROC_LOCKED(p->td->td_proc);
1250 	if (locked)
1251 		PROC_UNLOCK(p->td->td_proc);
1252 #ifdef GZIO
1253 	if (p->gzs != NULL)
1254 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1255 	else
1256 #endif
1257 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1258 		    UIO_SYSSPACE);
1259 	if (locked)
1260 		PROC_LOCK(p->td->td_proc);
1261 	if (error != 0)
1262 		return (-error);
1263 	p->offset += len;
1264 	return (len);
1265 }
1266 
1267 /*
1268  * Drain into a counter.
1269  */
1270 static int
1271 sbuf_drain_count(void *arg, const char *data __unused, int len)
1272 {
1273 	size_t *sizep;
1274 
1275 	sizep = (size_t *)arg;
1276 	*sizep += len;
1277 	return (len);
1278 }
1279 
1280 int
1281 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1282 {
1283 	struct ucred *cred = td->td_ucred;
1284 	int error = 0;
1285 	struct sseg_closure seginfo;
1286 	struct note_info_list notelst;
1287 	struct coredump_params params;
1288 	struct note_info *ninfo;
1289 	void *hdr, *tmpbuf;
1290 	size_t hdrsize, notesz, coresize;
1291 #ifdef GZIO
1292 	boolean_t compress;
1293 
1294 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1295 #endif
1296 	hdr = NULL;
1297 	tmpbuf = NULL;
1298 	TAILQ_INIT(&notelst);
1299 
1300 	/* Size the program segments. */
1301 	seginfo.count = 0;
1302 	seginfo.size = 0;
1303 	each_writable_segment(td, cb_size_segment, &seginfo);
1304 
1305 	/*
1306 	 * Collect info about the core file header area.
1307 	 */
1308 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1309 	__elfN(prepare_notes)(td, &notelst, &notesz);
1310 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1311 
1312 	/* Set up core dump parameters. */
1313 	params.offset = 0;
1314 	params.active_cred = cred;
1315 	params.file_cred = NOCRED;
1316 	params.td = td;
1317 	params.vp = vp;
1318 	params.gzs = NULL;
1319 
1320 #ifdef RACCT
1321 	if (racct_enable) {
1322 		PROC_LOCK(td->td_proc);
1323 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1324 		PROC_UNLOCK(td->td_proc);
1325 		if (error != 0) {
1326 			error = EFAULT;
1327 			goto done;
1328 		}
1329 	}
1330 #endif
1331 	if (coresize >= limit) {
1332 		error = EFAULT;
1333 		goto done;
1334 	}
1335 
1336 #ifdef GZIO
1337 	/* Create a compression stream if necessary. */
1338 	if (compress) {
1339 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1340 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1341 		if (params.gzs == NULL) {
1342 			error = EFAULT;
1343 			goto done;
1344 		}
1345 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1346         }
1347 #endif
1348 
1349 	/*
1350 	 * Allocate memory for building the header, fill it up,
1351 	 * and write it out following the notes.
1352 	 */
1353 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1354 	if (hdr == NULL) {
1355 		error = EINVAL;
1356 		goto done;
1357 	}
1358 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1359 	    notesz);
1360 
1361 	/* Write the contents of all of the writable segments. */
1362 	if (error == 0) {
1363 		Elf_Phdr *php;
1364 		off_t offset;
1365 		int i;
1366 
1367 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1368 		offset = round_page(hdrsize + notesz);
1369 		for (i = 0; i < seginfo.count; i++) {
1370 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1371 			    php->p_filesz, offset, &params, tmpbuf);
1372 			if (error != 0)
1373 				break;
1374 			offset += php->p_filesz;
1375 			php++;
1376 		}
1377 #ifdef GZIO
1378 		if (error == 0 && compress)
1379 			error = gzio_flush(params.gzs);
1380 #endif
1381 	}
1382 	if (error) {
1383 		log(LOG_WARNING,
1384 		    "Failed to write core file for process %s (error %d)\n",
1385 		    curproc->p_comm, error);
1386 	}
1387 
1388 done:
1389 #ifdef GZIO
1390 	if (compress) {
1391 		free(tmpbuf, M_TEMP);
1392 		if (params.gzs != NULL)
1393 			gzio_fini(params.gzs);
1394 	}
1395 #endif
1396 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1397 		TAILQ_REMOVE(&notelst, ninfo, link);
1398 		free(ninfo, M_TEMP);
1399 	}
1400 	if (hdr != NULL)
1401 		free(hdr, M_TEMP);
1402 
1403 	return (error);
1404 }
1405 
1406 /*
1407  * A callback for each_writable_segment() to write out the segment's
1408  * program header entry.
1409  */
1410 static void
1411 cb_put_phdr(entry, closure)
1412 	vm_map_entry_t entry;
1413 	void *closure;
1414 {
1415 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1416 	Elf_Phdr *phdr = phc->phdr;
1417 
1418 	phc->offset = round_page(phc->offset);
1419 
1420 	phdr->p_type = PT_LOAD;
1421 	phdr->p_offset = phc->offset;
1422 	phdr->p_vaddr = entry->start;
1423 	phdr->p_paddr = 0;
1424 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1425 	phdr->p_align = PAGE_SIZE;
1426 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1427 
1428 	phc->offset += phdr->p_filesz;
1429 	phc->phdr++;
1430 }
1431 
1432 /*
1433  * A callback for each_writable_segment() to gather information about
1434  * the number of segments and their total size.
1435  */
1436 static void
1437 cb_size_segment(entry, closure)
1438 	vm_map_entry_t entry;
1439 	void *closure;
1440 {
1441 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1442 
1443 	ssc->count++;
1444 	ssc->size += entry->end - entry->start;
1445 }
1446 
1447 /*
1448  * For each writable segment in the process's memory map, call the given
1449  * function with a pointer to the map entry and some arbitrary
1450  * caller-supplied data.
1451  */
1452 static void
1453 each_writable_segment(td, func, closure)
1454 	struct thread *td;
1455 	segment_callback func;
1456 	void *closure;
1457 {
1458 	struct proc *p = td->td_proc;
1459 	vm_map_t map = &p->p_vmspace->vm_map;
1460 	vm_map_entry_t entry;
1461 	vm_object_t backing_object, object;
1462 	boolean_t ignore_entry;
1463 
1464 	vm_map_lock_read(map);
1465 	for (entry = map->header.next; entry != &map->header;
1466 	    entry = entry->next) {
1467 		/*
1468 		 * Don't dump inaccessible mappings, deal with legacy
1469 		 * coredump mode.
1470 		 *
1471 		 * Note that read-only segments related to the elf binary
1472 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1473 		 * need to arbitrarily ignore such segments.
1474 		 */
1475 		if (elf_legacy_coredump) {
1476 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1477 				continue;
1478 		} else {
1479 			if ((entry->protection & VM_PROT_ALL) == 0)
1480 				continue;
1481 		}
1482 
1483 		/*
1484 		 * Dont include memory segment in the coredump if
1485 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1486 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1487 		 * kernel map).
1488 		 */
1489 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1490 			continue;
1491 
1492 		if ((object = entry->object.vm_object) == NULL)
1493 			continue;
1494 
1495 		/* Ignore memory-mapped devices and such things. */
1496 		VM_OBJECT_RLOCK(object);
1497 		while ((backing_object = object->backing_object) != NULL) {
1498 			VM_OBJECT_RLOCK(backing_object);
1499 			VM_OBJECT_RUNLOCK(object);
1500 			object = backing_object;
1501 		}
1502 		ignore_entry = object->type != OBJT_DEFAULT &&
1503 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1504 		    object->type != OBJT_PHYS;
1505 		VM_OBJECT_RUNLOCK(object);
1506 		if (ignore_entry)
1507 			continue;
1508 
1509 		(*func)(entry, closure);
1510 	}
1511 	vm_map_unlock_read(map);
1512 }
1513 
1514 /*
1515  * Write the core file header to the file, including padding up to
1516  * the page boundary.
1517  */
1518 static int
1519 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1520     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1521 {
1522 	struct note_info *ninfo;
1523 	struct sbuf *sb;
1524 	int error;
1525 
1526 	/* Fill in the header. */
1527 	bzero(hdr, hdrsize);
1528 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1529 
1530 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1531 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1532 	sbuf_start_section(sb, NULL);
1533 	sbuf_bcat(sb, hdr, hdrsize);
1534 	TAILQ_FOREACH(ninfo, notelst, link)
1535 	    __elfN(putnote)(ninfo, sb);
1536 	/* Align up to a page boundary for the program segments. */
1537 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1538 	error = sbuf_finish(sb);
1539 	sbuf_delete(sb);
1540 
1541 	return (error);
1542 }
1543 
1544 static void
1545 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1546     size_t *sizep)
1547 {
1548 	struct proc *p;
1549 	struct thread *thr;
1550 	size_t size;
1551 
1552 	p = td->td_proc;
1553 	size = 0;
1554 
1555 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1556 
1557 	/*
1558 	 * To have the debugger select the right thread (LWP) as the initial
1559 	 * thread, we dump the state of the thread passed to us in td first.
1560 	 * This is the thread that causes the core dump and thus likely to
1561 	 * be the right thread one wants to have selected in the debugger.
1562 	 */
1563 	thr = td;
1564 	while (thr != NULL) {
1565 		size += register_note(list, NT_PRSTATUS,
1566 		    __elfN(note_prstatus), thr);
1567 		size += register_note(list, NT_FPREGSET,
1568 		    __elfN(note_fpregset), thr);
1569 		size += register_note(list, NT_THRMISC,
1570 		    __elfN(note_thrmisc), thr);
1571 		size += register_note(list, -1,
1572 		    __elfN(note_threadmd), thr);
1573 
1574 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1575 		    TAILQ_NEXT(thr, td_plist);
1576 		if (thr == td)
1577 			thr = TAILQ_NEXT(thr, td_plist);
1578 	}
1579 
1580 	size += register_note(list, NT_PROCSTAT_PROC,
1581 	    __elfN(note_procstat_proc), p);
1582 	size += register_note(list, NT_PROCSTAT_FILES,
1583 	    note_procstat_files, p);
1584 	size += register_note(list, NT_PROCSTAT_VMMAP,
1585 	    note_procstat_vmmap, p);
1586 	size += register_note(list, NT_PROCSTAT_GROUPS,
1587 	    note_procstat_groups, p);
1588 	size += register_note(list, NT_PROCSTAT_UMASK,
1589 	    note_procstat_umask, p);
1590 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1591 	    note_procstat_rlimit, p);
1592 	size += register_note(list, NT_PROCSTAT_OSREL,
1593 	    note_procstat_osrel, p);
1594 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1595 	    __elfN(note_procstat_psstrings), p);
1596 	size += register_note(list, NT_PROCSTAT_AUXV,
1597 	    __elfN(note_procstat_auxv), p);
1598 
1599 	*sizep = size;
1600 }
1601 
1602 static void
1603 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1604     size_t notesz)
1605 {
1606 	Elf_Ehdr *ehdr;
1607 	Elf_Phdr *phdr;
1608 	struct phdr_closure phc;
1609 
1610 	ehdr = (Elf_Ehdr *)hdr;
1611 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1612 
1613 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1614 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1615 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1616 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1617 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1618 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1619 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1620 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1621 	ehdr->e_ident[EI_ABIVERSION] = 0;
1622 	ehdr->e_ident[EI_PAD] = 0;
1623 	ehdr->e_type = ET_CORE;
1624 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1625 	ehdr->e_machine = ELF_ARCH32;
1626 #else
1627 	ehdr->e_machine = ELF_ARCH;
1628 #endif
1629 	ehdr->e_version = EV_CURRENT;
1630 	ehdr->e_entry = 0;
1631 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1632 	ehdr->e_flags = 0;
1633 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1634 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1635 	ehdr->e_phnum = numsegs + 1;
1636 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1637 	ehdr->e_shnum = 0;
1638 	ehdr->e_shstrndx = SHN_UNDEF;
1639 
1640 	/*
1641 	 * Fill in the program header entries.
1642 	 */
1643 
1644 	/* The note segement. */
1645 	phdr->p_type = PT_NOTE;
1646 	phdr->p_offset = hdrsize;
1647 	phdr->p_vaddr = 0;
1648 	phdr->p_paddr = 0;
1649 	phdr->p_filesz = notesz;
1650 	phdr->p_memsz = 0;
1651 	phdr->p_flags = PF_R;
1652 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1653 	phdr++;
1654 
1655 	/* All the writable segments from the program. */
1656 	phc.phdr = phdr;
1657 	phc.offset = round_page(hdrsize + notesz);
1658 	each_writable_segment(td, cb_put_phdr, &phc);
1659 }
1660 
1661 static size_t
1662 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1663 {
1664 	struct note_info *ninfo;
1665 	size_t size, notesize;
1666 
1667 	size = 0;
1668 	out(arg, NULL, &size);
1669 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1670 	ninfo->type = type;
1671 	ninfo->outfunc = out;
1672 	ninfo->outarg = arg;
1673 	ninfo->outsize = size;
1674 	TAILQ_INSERT_TAIL(list, ninfo, link);
1675 
1676 	if (type == -1)
1677 		return (size);
1678 
1679 	notesize = sizeof(Elf_Note) +		/* note header */
1680 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1681 						/* note name */
1682 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1683 
1684 	return (notesize);
1685 }
1686 
1687 static size_t
1688 append_note_data(const void *src, void *dst, size_t len)
1689 {
1690 	size_t padded_len;
1691 
1692 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1693 	if (dst != NULL) {
1694 		bcopy(src, dst, len);
1695 		bzero((char *)dst + len, padded_len - len);
1696 	}
1697 	return (padded_len);
1698 }
1699 
1700 size_t
1701 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1702 {
1703 	Elf_Note *note;
1704 	char *buf;
1705 	size_t notesize;
1706 
1707 	buf = dst;
1708 	if (buf != NULL) {
1709 		note = (Elf_Note *)buf;
1710 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1711 		note->n_descsz = size;
1712 		note->n_type = type;
1713 		buf += sizeof(*note);
1714 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1715 		    sizeof(FREEBSD_ABI_VENDOR));
1716 		append_note_data(src, buf, size);
1717 		if (descp != NULL)
1718 			*descp = buf;
1719 	}
1720 
1721 	notesize = sizeof(Elf_Note) +		/* note header */
1722 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1723 						/* note name */
1724 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1725 
1726 	return (notesize);
1727 }
1728 
1729 static void
1730 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1731 {
1732 	Elf_Note note;
1733 	ssize_t old_len, sect_len;
1734 	size_t new_len, descsz, i;
1735 
1736 	if (ninfo->type == -1) {
1737 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1738 		return;
1739 	}
1740 
1741 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1742 	note.n_descsz = ninfo->outsize;
1743 	note.n_type = ninfo->type;
1744 
1745 	sbuf_bcat(sb, &note, sizeof(note));
1746 	sbuf_start_section(sb, &old_len);
1747 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1748 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1749 	if (note.n_descsz == 0)
1750 		return;
1751 	sbuf_start_section(sb, &old_len);
1752 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1753 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1754 	if (sect_len < 0)
1755 		return;
1756 
1757 	new_len = (size_t)sect_len;
1758 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1759 	if (new_len < descsz) {
1760 		/*
1761 		 * It is expected that individual note emitters will correctly
1762 		 * predict their expected output size and fill up to that size
1763 		 * themselves, padding in a format-specific way if needed.
1764 		 * However, in case they don't, just do it here with zeros.
1765 		 */
1766 		for (i = 0; i < descsz - new_len; i++)
1767 			sbuf_putc(sb, 0);
1768 	} else if (new_len > descsz) {
1769 		/*
1770 		 * We can't always truncate sb -- we may have drained some
1771 		 * of it already.
1772 		 */
1773 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1774 		    "read it (%zu > %zu).  Since it is longer than "
1775 		    "expected, this coredump's notes are corrupt.  THIS "
1776 		    "IS A BUG in the note_procstat routine for type %u.\n",
1777 		    __func__, (unsigned)note.n_type, new_len, descsz,
1778 		    (unsigned)note.n_type));
1779 	}
1780 }
1781 
1782 /*
1783  * Miscellaneous note out functions.
1784  */
1785 
1786 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1787 #include <compat/freebsd32/freebsd32.h>
1788 
1789 typedef struct prstatus32 elf_prstatus_t;
1790 typedef struct prpsinfo32 elf_prpsinfo_t;
1791 typedef struct fpreg32 elf_prfpregset_t;
1792 typedef struct fpreg32 elf_fpregset_t;
1793 typedef struct reg32 elf_gregset_t;
1794 typedef struct thrmisc32 elf_thrmisc_t;
1795 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1796 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1797 typedef uint32_t elf_ps_strings_t;
1798 #else
1799 typedef prstatus_t elf_prstatus_t;
1800 typedef prpsinfo_t elf_prpsinfo_t;
1801 typedef prfpregset_t elf_prfpregset_t;
1802 typedef prfpregset_t elf_fpregset_t;
1803 typedef gregset_t elf_gregset_t;
1804 typedef thrmisc_t elf_thrmisc_t;
1805 #define ELF_KERN_PROC_MASK	0
1806 typedef struct kinfo_proc elf_kinfo_proc_t;
1807 typedef vm_offset_t elf_ps_strings_t;
1808 #endif
1809 
1810 static void
1811 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1812 {
1813 	struct proc *p;
1814 	elf_prpsinfo_t *psinfo;
1815 
1816 	p = (struct proc *)arg;
1817 	if (sb != NULL) {
1818 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1819 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1820 		psinfo->pr_version = PRPSINFO_VERSION;
1821 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1822 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1823 		/*
1824 		 * XXX - We don't fill in the command line arguments properly
1825 		 * yet.
1826 		 */
1827 		strlcpy(psinfo->pr_psargs, p->p_comm,
1828 		    sizeof(psinfo->pr_psargs));
1829 
1830 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1831 		free(psinfo, M_TEMP);
1832 	}
1833 	*sizep = sizeof(*psinfo);
1834 }
1835 
1836 static void
1837 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1838 {
1839 	struct thread *td;
1840 	elf_prstatus_t *status;
1841 
1842 	td = (struct thread *)arg;
1843 	if (sb != NULL) {
1844 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1845 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1846 		status->pr_version = PRSTATUS_VERSION;
1847 		status->pr_statussz = sizeof(elf_prstatus_t);
1848 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1849 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1850 		status->pr_osreldate = osreldate;
1851 		status->pr_cursig = td->td_proc->p_sig;
1852 		status->pr_pid = td->td_tid;
1853 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1854 		fill_regs32(td, &status->pr_reg);
1855 #else
1856 		fill_regs(td, &status->pr_reg);
1857 #endif
1858 		sbuf_bcat(sb, status, sizeof(*status));
1859 		free(status, M_TEMP);
1860 	}
1861 	*sizep = sizeof(*status);
1862 }
1863 
1864 static void
1865 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1866 {
1867 	struct thread *td;
1868 	elf_prfpregset_t *fpregset;
1869 
1870 	td = (struct thread *)arg;
1871 	if (sb != NULL) {
1872 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1873 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1874 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1875 		fill_fpregs32(td, fpregset);
1876 #else
1877 		fill_fpregs(td, fpregset);
1878 #endif
1879 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1880 		free(fpregset, M_TEMP);
1881 	}
1882 	*sizep = sizeof(*fpregset);
1883 }
1884 
1885 static void
1886 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1887 {
1888 	struct thread *td;
1889 	elf_thrmisc_t thrmisc;
1890 
1891 	td = (struct thread *)arg;
1892 	if (sb != NULL) {
1893 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1894 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1895 		strcpy(thrmisc.pr_tname, td->td_name);
1896 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1897 	}
1898 	*sizep = sizeof(thrmisc);
1899 }
1900 
1901 /*
1902  * Allow for MD specific notes, as well as any MD
1903  * specific preparations for writing MI notes.
1904  */
1905 static void
1906 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1907 {
1908 	struct thread *td;
1909 	void *buf;
1910 	size_t size;
1911 
1912 	td = (struct thread *)arg;
1913 	size = *sizep;
1914 	if (size != 0 && sb != NULL)
1915 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1916 	else
1917 		buf = NULL;
1918 	size = 0;
1919 	__elfN(dump_thread)(td, buf, &size);
1920 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1921 	if (size != 0 && sb != NULL)
1922 		sbuf_bcat(sb, buf, size);
1923 	free(buf, M_TEMP);
1924 	*sizep = size;
1925 }
1926 
1927 #ifdef KINFO_PROC_SIZE
1928 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1929 #endif
1930 
1931 static void
1932 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1933 {
1934 	struct proc *p;
1935 	size_t size;
1936 	int structsize;
1937 
1938 	p = (struct proc *)arg;
1939 	size = sizeof(structsize) + p->p_numthreads *
1940 	    sizeof(elf_kinfo_proc_t);
1941 
1942 	if (sb != NULL) {
1943 		KASSERT(*sizep == size, ("invalid size"));
1944 		structsize = sizeof(elf_kinfo_proc_t);
1945 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1946 		sx_slock(&proctree_lock);
1947 		PROC_LOCK(p);
1948 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1949 		sx_sunlock(&proctree_lock);
1950 	}
1951 	*sizep = size;
1952 }
1953 
1954 #ifdef KINFO_FILE_SIZE
1955 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1956 #endif
1957 
1958 static void
1959 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1960 {
1961 	struct proc *p;
1962 	size_t size, sect_sz, i;
1963 	ssize_t start_len, sect_len;
1964 	int structsize, filedesc_flags;
1965 
1966 	if (coredump_pack_fileinfo)
1967 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1968 	else
1969 		filedesc_flags = 0;
1970 
1971 	p = (struct proc *)arg;
1972 	structsize = sizeof(struct kinfo_file);
1973 	if (sb == NULL) {
1974 		size = 0;
1975 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1976 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1977 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1978 		PROC_LOCK(p);
1979 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1980 		sbuf_finish(sb);
1981 		sbuf_delete(sb);
1982 		*sizep = size;
1983 	} else {
1984 		sbuf_start_section(sb, &start_len);
1985 
1986 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1987 		PROC_LOCK(p);
1988 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1989 		    filedesc_flags);
1990 
1991 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
1992 		if (sect_len < 0)
1993 			return;
1994 		sect_sz = sect_len;
1995 
1996 		KASSERT(sect_sz <= *sizep,
1997 		    ("kern_proc_filedesc_out did not respect maxlen; "
1998 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
1999 		     sect_sz - sizeof(structsize)));
2000 
2001 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2002 			sbuf_putc(sb, 0);
2003 	}
2004 }
2005 
2006 #ifdef KINFO_VMENTRY_SIZE
2007 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2008 #endif
2009 
2010 static void
2011 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2012 {
2013 	struct proc *p;
2014 	size_t size;
2015 	int structsize, vmmap_flags;
2016 
2017 	if (coredump_pack_vmmapinfo)
2018 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2019 	else
2020 		vmmap_flags = 0;
2021 
2022 	p = (struct proc *)arg;
2023 	structsize = sizeof(struct kinfo_vmentry);
2024 	if (sb == NULL) {
2025 		size = 0;
2026 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2027 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2028 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2029 		PROC_LOCK(p);
2030 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2031 		sbuf_finish(sb);
2032 		sbuf_delete(sb);
2033 		*sizep = size;
2034 	} else {
2035 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2036 		PROC_LOCK(p);
2037 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2038 		    vmmap_flags);
2039 	}
2040 }
2041 
2042 static void
2043 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2044 {
2045 	struct proc *p;
2046 	size_t size;
2047 	int structsize;
2048 
2049 	p = (struct proc *)arg;
2050 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2051 	if (sb != NULL) {
2052 		KASSERT(*sizep == size, ("invalid size"));
2053 		structsize = sizeof(gid_t);
2054 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2055 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2056 		    sizeof(gid_t));
2057 	}
2058 	*sizep = size;
2059 }
2060 
2061 static void
2062 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2063 {
2064 	struct proc *p;
2065 	size_t size;
2066 	int structsize;
2067 
2068 	p = (struct proc *)arg;
2069 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2070 	if (sb != NULL) {
2071 		KASSERT(*sizep == size, ("invalid size"));
2072 		structsize = sizeof(p->p_fd->fd_cmask);
2073 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2074 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2075 	}
2076 	*sizep = size;
2077 }
2078 
2079 static void
2080 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2081 {
2082 	struct proc *p;
2083 	struct rlimit rlim[RLIM_NLIMITS];
2084 	size_t size;
2085 	int structsize, i;
2086 
2087 	p = (struct proc *)arg;
2088 	size = sizeof(structsize) + sizeof(rlim);
2089 	if (sb != NULL) {
2090 		KASSERT(*sizep == size, ("invalid size"));
2091 		structsize = sizeof(rlim);
2092 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2093 		PROC_LOCK(p);
2094 		for (i = 0; i < RLIM_NLIMITS; i++)
2095 			lim_rlimit_proc(p, i, &rlim[i]);
2096 		PROC_UNLOCK(p);
2097 		sbuf_bcat(sb, rlim, sizeof(rlim));
2098 	}
2099 	*sizep = size;
2100 }
2101 
2102 static void
2103 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2104 {
2105 	struct proc *p;
2106 	size_t size;
2107 	int structsize;
2108 
2109 	p = (struct proc *)arg;
2110 	size = sizeof(structsize) + sizeof(p->p_osrel);
2111 	if (sb != NULL) {
2112 		KASSERT(*sizep == size, ("invalid size"));
2113 		structsize = sizeof(p->p_osrel);
2114 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2115 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2116 	}
2117 	*sizep = size;
2118 }
2119 
2120 static void
2121 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2122 {
2123 	struct proc *p;
2124 	elf_ps_strings_t ps_strings;
2125 	size_t size;
2126 	int structsize;
2127 
2128 	p = (struct proc *)arg;
2129 	size = sizeof(structsize) + sizeof(ps_strings);
2130 	if (sb != NULL) {
2131 		KASSERT(*sizep == size, ("invalid size"));
2132 		structsize = sizeof(ps_strings);
2133 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2134 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2135 #else
2136 		ps_strings = p->p_sysent->sv_psstrings;
2137 #endif
2138 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2139 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2140 	}
2141 	*sizep = size;
2142 }
2143 
2144 static void
2145 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2146 {
2147 	struct proc *p;
2148 	size_t size;
2149 	int structsize;
2150 
2151 	p = (struct proc *)arg;
2152 	if (sb == NULL) {
2153 		size = 0;
2154 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2155 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2156 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2157 		PHOLD(p);
2158 		proc_getauxv(curthread, p, sb);
2159 		PRELE(p);
2160 		sbuf_finish(sb);
2161 		sbuf_delete(sb);
2162 		*sizep = size;
2163 	} else {
2164 		structsize = sizeof(Elf_Auxinfo);
2165 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2166 		PHOLD(p);
2167 		proc_getauxv(curthread, p, sb);
2168 		PRELE(p);
2169 	}
2170 }
2171 
2172 static boolean_t
2173 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2174     int32_t *osrel, const Elf_Phdr *pnote)
2175 {
2176 	const Elf_Note *note, *note0, *note_end;
2177 	const char *note_name;
2178 	char *buf;
2179 	int i, error;
2180 	boolean_t res;
2181 
2182 	/* We need some limit, might as well use PAGE_SIZE. */
2183 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2184 		return (FALSE);
2185 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2186 	if (pnote->p_offset > PAGE_SIZE ||
2187 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2188 		VOP_UNLOCK(imgp->vp, 0);
2189 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2190 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2191 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2192 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2193 		    curthread->td_ucred, NOCRED, NULL, curthread);
2194 		if (error != 0) {
2195 			uprintf("i/o error PT_NOTE\n");
2196 			res = FALSE;
2197 			goto ret;
2198 		}
2199 		note = note0 = (const Elf_Note *)buf;
2200 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2201 	} else {
2202 		note = note0 = (const Elf_Note *)(imgp->image_header +
2203 		    pnote->p_offset);
2204 		note_end = (const Elf_Note *)(imgp->image_header +
2205 		    pnote->p_offset + pnote->p_filesz);
2206 		buf = NULL;
2207 	}
2208 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2209 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2210 		    (const char *)note < sizeof(Elf_Note)) {
2211 			res = FALSE;
2212 			goto ret;
2213 		}
2214 		if (note->n_namesz != checknote->hdr.n_namesz ||
2215 		    note->n_descsz != checknote->hdr.n_descsz ||
2216 		    note->n_type != checknote->hdr.n_type)
2217 			goto nextnote;
2218 		note_name = (const char *)(note + 1);
2219 		if (note_name + checknote->hdr.n_namesz >=
2220 		    (const char *)note_end || strncmp(checknote->vendor,
2221 		    note_name, checknote->hdr.n_namesz) != 0)
2222 			goto nextnote;
2223 
2224 		/*
2225 		 * Fetch the osreldate for binary
2226 		 * from the ELF OSABI-note if necessary.
2227 		 */
2228 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2229 		    checknote->trans_osrel != NULL) {
2230 			res = checknote->trans_osrel(note, osrel);
2231 			goto ret;
2232 		}
2233 		res = TRUE;
2234 		goto ret;
2235 nextnote:
2236 		note = (const Elf_Note *)((const char *)(note + 1) +
2237 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2238 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2239 	}
2240 	res = FALSE;
2241 ret:
2242 	free(buf, M_TEMP);
2243 	return (res);
2244 }
2245 
2246 /*
2247  * Try to find the appropriate ABI-note section for checknote,
2248  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2249  * first page of the image is searched, the same as for headers.
2250  */
2251 static boolean_t
2252 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2253     int32_t *osrel)
2254 {
2255 	const Elf_Phdr *phdr;
2256 	const Elf_Ehdr *hdr;
2257 	int i;
2258 
2259 	hdr = (const Elf_Ehdr *)imgp->image_header;
2260 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2261 
2262 	for (i = 0; i < hdr->e_phnum; i++) {
2263 		if (phdr[i].p_type == PT_NOTE &&
2264 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2265 			return (TRUE);
2266 	}
2267 	return (FALSE);
2268 
2269 }
2270 
2271 /*
2272  * Tell kern_execve.c about it, with a little help from the linker.
2273  */
2274 static struct execsw __elfN(execsw) = {
2275 	__CONCAT(exec_, __elfN(imgact)),
2276 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2277 };
2278 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2279 
2280 static vm_prot_t
2281 __elfN(trans_prot)(Elf_Word flags)
2282 {
2283 	vm_prot_t prot;
2284 
2285 	prot = 0;
2286 	if (flags & PF_X)
2287 		prot |= VM_PROT_EXECUTE;
2288 	if (flags & PF_W)
2289 		prot |= VM_PROT_WRITE;
2290 	if (flags & PF_R)
2291 		prot |= VM_PROT_READ;
2292 #if __ELF_WORD_SIZE == 32
2293 #if defined(__amd64__)
2294 	if (i386_read_exec && (flags & PF_R))
2295 		prot |= VM_PROT_EXECUTE;
2296 #endif
2297 #endif
2298 	return (prot);
2299 }
2300 
2301 static Elf_Word
2302 __elfN(untrans_prot)(vm_prot_t prot)
2303 {
2304 	Elf_Word flags;
2305 
2306 	flags = 0;
2307 	if (prot & VM_PROT_EXECUTE)
2308 		flags |= PF_X;
2309 	if (prot & VM_PROT_READ)
2310 		flags |= PF_R;
2311 	if (prot & VM_PROT_WRITE)
2312 		flags |= PF_W;
2313 	return (flags);
2314 }
2315