xref: /freebsd/sys/kern/imgact_elf.c (revision 9034852c84a13f0e3b5527e1c886ca94b2863b2b)
1 /*-
2  * Copyright (c) 2000 David O'Brien
3  * Copyright (c) 1995-1996 Søren Schmidt
4  * Copyright (c) 1996 Peter Wemm
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_gzio.h"
37 
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
40 #include <sys/exec.h>
41 #include <sys/fcntl.h>
42 #include <sys/gzio.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mman.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
58 #include <sys/sbuf.h>
59 #include <sys/sf_buf.h>
60 #include <sys/smp.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
63 #include <sys/stat.h>
64 #include <sys/sx.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
71 #include <sys/user.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
80 
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
83 
84 #define ELF_NOTE_ROUNDSIZE	4
85 #define OLD_EI_BRAND	8
86 
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89     const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91     u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
93     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
94     size_t pagesize);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97     int32_t *osrel);
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100     Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
103 
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
105     "");
106 
107 #define	CORE_BUF_SIZE	(16 * 1024)
108 
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
113 
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116     &elf_legacy_coredump, 0, "");
117 
118 int __elfN(nxstack) =
119 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
120 	1;
121 #else
122 	0;
123 #endif
124 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
125     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
126     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
127 
128 #if __ELF_WORD_SIZE == 32
129 #if defined(__amd64__)
130 int i386_read_exec = 0;
131 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
132     "enable execution from readable segments");
133 #endif
134 #endif
135 
136 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
137 
138 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
139 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
140 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
141 
142 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
143 
144 Elf_Brandnote __elfN(freebsd_brandnote) = {
145 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
146 	.hdr.n_descsz	= sizeof(int32_t),
147 	.hdr.n_type	= 1,
148 	.vendor		= FREEBSD_ABI_VENDOR,
149 	.flags		= BN_TRANSLATE_OSREL,
150 	.trans_osrel	= __elfN(freebsd_trans_osrel)
151 };
152 
153 static boolean_t
154 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
155 {
156 	uintptr_t p;
157 
158 	p = (uintptr_t)(note + 1);
159 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
160 	*osrel = *(const int32_t *)(p);
161 
162 	return (TRUE);
163 }
164 
165 static const char GNU_ABI_VENDOR[] = "GNU";
166 static int GNU_KFREEBSD_ABI_DESC = 3;
167 
168 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
169 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
170 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
171 	.hdr.n_type	= 1,
172 	.vendor		= GNU_ABI_VENDOR,
173 	.flags		= BN_TRANSLATE_OSREL,
174 	.trans_osrel	= kfreebsd_trans_osrel
175 };
176 
177 static boolean_t
178 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
179 {
180 	const Elf32_Word *desc;
181 	uintptr_t p;
182 
183 	p = (uintptr_t)(note + 1);
184 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
185 
186 	desc = (const Elf32_Word *)p;
187 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
188 		return (FALSE);
189 
190 	/*
191 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
192 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
193 	 */
194 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
195 
196 	return (TRUE);
197 }
198 
199 int
200 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
201 {
202 	int i;
203 
204 	for (i = 0; i < MAX_BRANDS; i++) {
205 		if (elf_brand_list[i] == NULL) {
206 			elf_brand_list[i] = entry;
207 			break;
208 		}
209 	}
210 	if (i == MAX_BRANDS) {
211 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
212 			__func__, entry);
213 		return (-1);
214 	}
215 	return (0);
216 }
217 
218 int
219 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
220 {
221 	int i;
222 
223 	for (i = 0; i < MAX_BRANDS; i++) {
224 		if (elf_brand_list[i] == entry) {
225 			elf_brand_list[i] = NULL;
226 			break;
227 		}
228 	}
229 	if (i == MAX_BRANDS)
230 		return (-1);
231 	return (0);
232 }
233 
234 int
235 __elfN(brand_inuse)(Elf_Brandinfo *entry)
236 {
237 	struct proc *p;
238 	int rval = FALSE;
239 
240 	sx_slock(&allproc_lock);
241 	FOREACH_PROC_IN_SYSTEM(p) {
242 		if (p->p_sysent == entry->sysvec) {
243 			rval = TRUE;
244 			break;
245 		}
246 	}
247 	sx_sunlock(&allproc_lock);
248 
249 	return (rval);
250 }
251 
252 static Elf_Brandinfo *
253 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
254     int interp_name_len, int32_t *osrel)
255 {
256 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
257 	Elf_Brandinfo *bi;
258 	boolean_t ret;
259 	int i;
260 
261 	/*
262 	 * We support four types of branding -- (1) the ELF EI_OSABI field
263 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
264 	 * branding w/in the ELF header, (3) path of the `interp_path'
265 	 * field, and (4) the ".note.ABI-tag" ELF section.
266 	 */
267 
268 	/* Look for an ".note.ABI-tag" ELF section */
269 	for (i = 0; i < MAX_BRANDS; i++) {
270 		bi = elf_brand_list[i];
271 		if (bi == NULL)
272 			continue;
273 		if (hdr->e_machine == bi->machine && (bi->flags &
274 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
275 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
276 			if (ret)
277 				return (bi);
278 		}
279 	}
280 
281 	/* If the executable has a brand, search for it in the brand list. */
282 	for (i = 0; i < MAX_BRANDS; i++) {
283 		bi = elf_brand_list[i];
284 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
285 			continue;
286 		if (hdr->e_machine == bi->machine &&
287 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
288 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
289 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
290 			return (bi);
291 	}
292 
293 	/* No known brand, see if the header is recognized by any brand */
294 	for (i = 0; i < MAX_BRANDS; i++) {
295 		bi = elf_brand_list[i];
296 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
297 		    bi->header_supported == NULL)
298 			continue;
299 		if (hdr->e_machine == bi->machine) {
300 			ret = bi->header_supported(imgp);
301 			if (ret)
302 				return (bi);
303 		}
304 	}
305 
306 	/* Lacking a known brand, search for a recognized interpreter. */
307 	if (interp != NULL) {
308 		for (i = 0; i < MAX_BRANDS; i++) {
309 			bi = elf_brand_list[i];
310 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
311 				continue;
312 			if (hdr->e_machine == bi->machine &&
313 			    /* ELF image p_filesz includes terminating zero */
314 			    strlen(bi->interp_path) + 1 == interp_name_len &&
315 			    strncmp(interp, bi->interp_path, interp_name_len)
316 			    == 0)
317 				return (bi);
318 		}
319 	}
320 
321 	/* Lacking a recognized interpreter, try the default brand */
322 	for (i = 0; i < MAX_BRANDS; i++) {
323 		bi = elf_brand_list[i];
324 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
325 			continue;
326 		if (hdr->e_machine == bi->machine &&
327 		    __elfN(fallback_brand) == bi->brand)
328 			return (bi);
329 	}
330 	return (NULL);
331 }
332 
333 static int
334 __elfN(check_header)(const Elf_Ehdr *hdr)
335 {
336 	Elf_Brandinfo *bi;
337 	int i;
338 
339 	if (!IS_ELF(*hdr) ||
340 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
341 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
342 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
343 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
344 	    hdr->e_version != ELF_TARG_VER)
345 		return (ENOEXEC);
346 
347 	/*
348 	 * Make sure we have at least one brand for this machine.
349 	 */
350 
351 	for (i = 0; i < MAX_BRANDS; i++) {
352 		bi = elf_brand_list[i];
353 		if (bi != NULL && bi->machine == hdr->e_machine)
354 			break;
355 	}
356 	if (i == MAX_BRANDS)
357 		return (ENOEXEC);
358 
359 	return (0);
360 }
361 
362 static int
363 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
364     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
365 {
366 	struct sf_buf *sf;
367 	int error;
368 	vm_offset_t off;
369 
370 	/*
371 	 * Create the page if it doesn't exist yet. Ignore errors.
372 	 */
373 	vm_map_lock(map);
374 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
375 	    VM_PROT_ALL, VM_PROT_ALL, 0);
376 	vm_map_unlock(map);
377 
378 	/*
379 	 * Find the page from the underlying object.
380 	 */
381 	if (object) {
382 		sf = vm_imgact_map_page(object, offset);
383 		if (sf == NULL)
384 			return (KERN_FAILURE);
385 		off = offset - trunc_page(offset);
386 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
387 		    end - start);
388 		vm_imgact_unmap_page(sf);
389 		if (error) {
390 			return (KERN_FAILURE);
391 		}
392 	}
393 
394 	return (KERN_SUCCESS);
395 }
396 
397 static int
398 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
399     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
400 {
401 	struct sf_buf *sf;
402 	vm_offset_t off;
403 	vm_size_t sz;
404 	int error, rv;
405 
406 	if (start != trunc_page(start)) {
407 		rv = __elfN(map_partial)(map, object, offset, start,
408 		    round_page(start), prot);
409 		if (rv)
410 			return (rv);
411 		offset += round_page(start) - start;
412 		start = round_page(start);
413 	}
414 	if (end != round_page(end)) {
415 		rv = __elfN(map_partial)(map, object, offset +
416 		    trunc_page(end) - start, trunc_page(end), end, prot);
417 		if (rv)
418 			return (rv);
419 		end = trunc_page(end);
420 	}
421 	if (end > start) {
422 		if (offset & PAGE_MASK) {
423 			/*
424 			 * The mapping is not page aligned. This means we have
425 			 * to copy the data. Sigh.
426 			 */
427 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
428 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
429 			    0);
430 			if (rv)
431 				return (rv);
432 			if (object == NULL)
433 				return (KERN_SUCCESS);
434 			for (; start < end; start += sz) {
435 				sf = vm_imgact_map_page(object, offset);
436 				if (sf == NULL)
437 					return (KERN_FAILURE);
438 				off = offset - trunc_page(offset);
439 				sz = end - start;
440 				if (sz > PAGE_SIZE - off)
441 					sz = PAGE_SIZE - off;
442 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
443 				    (caddr_t)start, sz);
444 				vm_imgact_unmap_page(sf);
445 				if (error) {
446 					return (KERN_FAILURE);
447 				}
448 				offset += sz;
449 			}
450 			rv = KERN_SUCCESS;
451 		} else {
452 			vm_object_reference(object);
453 			vm_map_lock(map);
454 			rv = vm_map_insert(map, object, offset, start, end,
455 			    prot, VM_PROT_ALL, cow);
456 			vm_map_unlock(map);
457 			if (rv != KERN_SUCCESS)
458 				vm_object_deallocate(object);
459 		}
460 		return (rv);
461 	} else {
462 		return (KERN_SUCCESS);
463 	}
464 }
465 
466 static int
467 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
468     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
469     size_t pagesize)
470 {
471 	struct sf_buf *sf;
472 	size_t map_len;
473 	vm_map_t map;
474 	vm_object_t object;
475 	vm_offset_t map_addr;
476 	int error, rv, cow;
477 	size_t copy_len;
478 	vm_offset_t file_addr;
479 
480 	/*
481 	 * It's necessary to fail if the filsz + offset taken from the
482 	 * header is greater than the actual file pager object's size.
483 	 * If we were to allow this, then the vm_map_find() below would
484 	 * walk right off the end of the file object and into the ether.
485 	 *
486 	 * While I'm here, might as well check for something else that
487 	 * is invalid: filsz cannot be greater than memsz.
488 	 */
489 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
490 		uprintf("elf_load_section: truncated ELF file\n");
491 		return (ENOEXEC);
492 	}
493 
494 	object = imgp->object;
495 	map = &imgp->proc->p_vmspace->vm_map;
496 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
497 	file_addr = trunc_page_ps(offset, pagesize);
498 
499 	/*
500 	 * We have two choices.  We can either clear the data in the last page
501 	 * of an oversized mapping, or we can start the anon mapping a page
502 	 * early and copy the initialized data into that first page.  We
503 	 * choose the second..
504 	 */
505 	if (memsz > filsz)
506 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
507 	else
508 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
509 
510 	if (map_len != 0) {
511 		/* cow flags: don't dump readonly sections in core */
512 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
513 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
514 
515 		rv = __elfN(map_insert)(map,
516 				      object,
517 				      file_addr,	/* file offset */
518 				      map_addr,		/* virtual start */
519 				      map_addr + map_len,/* virtual end */
520 				      prot,
521 				      cow);
522 		if (rv != KERN_SUCCESS)
523 			return (EINVAL);
524 
525 		/* we can stop now if we've covered it all */
526 		if (memsz == filsz) {
527 			return (0);
528 		}
529 	}
530 
531 
532 	/*
533 	 * We have to get the remaining bit of the file into the first part
534 	 * of the oversized map segment.  This is normally because the .data
535 	 * segment in the file is extended to provide bss.  It's a neat idea
536 	 * to try and save a page, but it's a pain in the behind to implement.
537 	 */
538 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
539 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
540 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
541 	    map_addr;
542 
543 	/* This had damn well better be true! */
544 	if (map_len != 0) {
545 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
546 		    map_len, VM_PROT_ALL, 0);
547 		if (rv != KERN_SUCCESS) {
548 			return (EINVAL);
549 		}
550 	}
551 
552 	if (copy_len != 0) {
553 		vm_offset_t off;
554 
555 		sf = vm_imgact_map_page(object, offset + filsz);
556 		if (sf == NULL)
557 			return (EIO);
558 
559 		/* send the page fragment to user space */
560 		off = trunc_page_ps(offset + filsz, pagesize) -
561 		    trunc_page(offset + filsz);
562 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
563 		    (caddr_t)map_addr, copy_len);
564 		vm_imgact_unmap_page(sf);
565 		if (error) {
566 			return (error);
567 		}
568 	}
569 
570 	/*
571 	 * set it to the specified protection.
572 	 * XXX had better undo the damage from pasting over the cracks here!
573 	 */
574 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
575 	    map_len), prot, FALSE);
576 
577 	return (0);
578 }
579 
580 /*
581  * Load the file "file" into memory.  It may be either a shared object
582  * or an executable.
583  *
584  * The "addr" reference parameter is in/out.  On entry, it specifies
585  * the address where a shared object should be loaded.  If the file is
586  * an executable, this value is ignored.  On exit, "addr" specifies
587  * where the file was actually loaded.
588  *
589  * The "entry" reference parameter is out only.  On exit, it specifies
590  * the entry point for the loaded file.
591  */
592 static int
593 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
594 	u_long *entry, size_t pagesize)
595 {
596 	struct {
597 		struct nameidata nd;
598 		struct vattr attr;
599 		struct image_params image_params;
600 	} *tempdata;
601 	const Elf_Ehdr *hdr = NULL;
602 	const Elf_Phdr *phdr = NULL;
603 	struct nameidata *nd;
604 	struct vattr *attr;
605 	struct image_params *imgp;
606 	vm_prot_t prot;
607 	u_long rbase;
608 	u_long base_addr = 0;
609 	int error, i, numsegs;
610 
611 #ifdef CAPABILITY_MODE
612 	/*
613 	 * XXXJA: This check can go away once we are sufficiently confident
614 	 * that the checks in namei() are correct.
615 	 */
616 	if (IN_CAPABILITY_MODE(curthread))
617 		return (ECAPMODE);
618 #endif
619 
620 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
621 	nd = &tempdata->nd;
622 	attr = &tempdata->attr;
623 	imgp = &tempdata->image_params;
624 
625 	/*
626 	 * Initialize part of the common data
627 	 */
628 	imgp->proc = p;
629 	imgp->attr = attr;
630 	imgp->firstpage = NULL;
631 	imgp->image_header = NULL;
632 	imgp->object = NULL;
633 	imgp->execlabel = NULL;
634 
635 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
636 	if ((error = namei(nd)) != 0) {
637 		nd->ni_vp = NULL;
638 		goto fail;
639 	}
640 	NDFREE(nd, NDF_ONLY_PNBUF);
641 	imgp->vp = nd->ni_vp;
642 
643 	/*
644 	 * Check permissions, modes, uid, etc on the file, and "open" it.
645 	 */
646 	error = exec_check_permissions(imgp);
647 	if (error)
648 		goto fail;
649 
650 	error = exec_map_first_page(imgp);
651 	if (error)
652 		goto fail;
653 
654 	/*
655 	 * Also make certain that the interpreter stays the same, so set
656 	 * its VV_TEXT flag, too.
657 	 */
658 	VOP_SET_TEXT(nd->ni_vp);
659 
660 	imgp->object = nd->ni_vp->v_object;
661 
662 	hdr = (const Elf_Ehdr *)imgp->image_header;
663 	if ((error = __elfN(check_header)(hdr)) != 0)
664 		goto fail;
665 	if (hdr->e_type == ET_DYN)
666 		rbase = *addr;
667 	else if (hdr->e_type == ET_EXEC)
668 		rbase = 0;
669 	else {
670 		error = ENOEXEC;
671 		goto fail;
672 	}
673 
674 	/* Only support headers that fit within first page for now      */
675 	if ((hdr->e_phoff > PAGE_SIZE) ||
676 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
677 		error = ENOEXEC;
678 		goto fail;
679 	}
680 
681 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
682 	if (!aligned(phdr, Elf_Addr)) {
683 		error = ENOEXEC;
684 		goto fail;
685 	}
686 
687 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
688 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
689 			/* Loadable segment */
690 			prot = __elfN(trans_prot)(phdr[i].p_flags);
691 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
692 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
693 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
694 			if (error != 0)
695 				goto fail;
696 			/*
697 			 * Establish the base address if this is the
698 			 * first segment.
699 			 */
700 			if (numsegs == 0)
701   				base_addr = trunc_page(phdr[i].p_vaddr +
702 				    rbase);
703 			numsegs++;
704 		}
705 	}
706 	*addr = base_addr;
707 	*entry = (unsigned long)hdr->e_entry + rbase;
708 
709 fail:
710 	if (imgp->firstpage)
711 		exec_unmap_first_page(imgp);
712 
713 	if (nd->ni_vp)
714 		vput(nd->ni_vp);
715 
716 	free(tempdata, M_TEMP);
717 
718 	return (error);
719 }
720 
721 static int
722 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
723 {
724 	struct thread *td;
725 	const Elf_Ehdr *hdr;
726 	const Elf_Phdr *phdr;
727 	Elf_Auxargs *elf_auxargs;
728 	struct vmspace *vmspace;
729 	const char *err_str, *newinterp;
730 	char *interp, *interp_buf, *path;
731 	Elf_Brandinfo *brand_info;
732 	struct sysentvec *sv;
733 	vm_prot_t prot;
734 	u_long text_size, data_size, total_size, text_addr, data_addr;
735 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
736 	int32_t osrel;
737 	int error, i, n, interp_name_len, have_interp;
738 
739 	hdr = (const Elf_Ehdr *)imgp->image_header;
740 
741 	/*
742 	 * Do we have a valid ELF header ?
743 	 *
744 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
745 	 * if particular brand doesn't support it.
746 	 */
747 	if (__elfN(check_header)(hdr) != 0 ||
748 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
749 		return (-1);
750 
751 	/*
752 	 * From here on down, we return an errno, not -1, as we've
753 	 * detected an ELF file.
754 	 */
755 
756 	if ((hdr->e_phoff > PAGE_SIZE) ||
757 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
758 		/* Only support headers in first page for now */
759 		uprintf("Program headers not in the first page\n");
760 		return (ENOEXEC);
761 	}
762 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
763 	if (!aligned(phdr, Elf_Addr)) {
764 		uprintf("Unaligned program headers\n");
765 		return (ENOEXEC);
766 	}
767 
768 	n = error = 0;
769 	baddr = 0;
770 	osrel = 0;
771 	text_size = data_size = total_size = text_addr = data_addr = 0;
772 	entry = proghdr = 0;
773 	interp_name_len = 0;
774 	err_str = newinterp = NULL;
775 	interp = interp_buf = NULL;
776 	td = curthread;
777 
778 	for (i = 0; i < hdr->e_phnum; i++) {
779 		switch (phdr[i].p_type) {
780 		case PT_LOAD:
781 			if (n == 0)
782 				baddr = phdr[i].p_vaddr;
783 			n++;
784 			break;
785 		case PT_INTERP:
786 			/* Path to interpreter */
787 			if (phdr[i].p_filesz > MAXPATHLEN) {
788 				uprintf("Invalid PT_INTERP\n");
789 				error = ENOEXEC;
790 				goto ret;
791 			}
792 			interp_name_len = phdr[i].p_filesz;
793 			if (phdr[i].p_offset > PAGE_SIZE ||
794 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
795 				VOP_UNLOCK(imgp->vp, 0);
796 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
797 				    M_WAITOK);
798 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
799 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
800 				    interp_name_len, phdr[i].p_offset,
801 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
802 				    NOCRED, NULL, td);
803 				if (error != 0) {
804 					uprintf("i/o error PT_INTERP\n");
805 					goto ret;
806 				}
807 				interp_buf[interp_name_len] = '\0';
808 				interp = interp_buf;
809 			} else {
810 				interp = __DECONST(char *, imgp->image_header) +
811 				    phdr[i].p_offset;
812 			}
813 			break;
814 		case PT_GNU_STACK:
815 			if (__elfN(nxstack))
816 				imgp->stack_prot =
817 				    __elfN(trans_prot)(phdr[i].p_flags);
818 			imgp->stack_sz = phdr[i].p_memsz;
819 			break;
820 		}
821 	}
822 
823 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
824 	    &osrel);
825 	if (brand_info == NULL) {
826 		uprintf("ELF binary type \"%u\" not known.\n",
827 		    hdr->e_ident[EI_OSABI]);
828 		error = ENOEXEC;
829 		goto ret;
830 	}
831 	if (hdr->e_type == ET_DYN) {
832 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
833 			uprintf("Cannot execute shared object\n");
834 			error = ENOEXEC;
835 			goto ret;
836 		}
837 		/*
838 		 * Honour the base load address from the dso if it is
839 		 * non-zero for some reason.
840 		 */
841 		if (baddr == 0)
842 			et_dyn_addr = ET_DYN_LOAD_ADDR;
843 		else
844 			et_dyn_addr = 0;
845 	} else
846 		et_dyn_addr = 0;
847 	sv = brand_info->sysvec;
848 	if (interp != NULL && brand_info->interp_newpath != NULL)
849 		newinterp = brand_info->interp_newpath;
850 
851 	/*
852 	 * Avoid a possible deadlock if the current address space is destroyed
853 	 * and that address space maps the locked vnode.  In the common case,
854 	 * the locked vnode's v_usecount is decremented but remains greater
855 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
856 	 * However, in cases where the vnode lock is external, such as nullfs,
857 	 * v_usecount may become zero.
858 	 *
859 	 * The VV_TEXT flag prevents modifications to the executable while
860 	 * the vnode is unlocked.
861 	 */
862 	VOP_UNLOCK(imgp->vp, 0);
863 
864 	error = exec_new_vmspace(imgp, sv);
865 	imgp->proc->p_sysent = sv;
866 
867 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
868 	if (error != 0)
869 		goto ret;
870 
871 	for (i = 0; i < hdr->e_phnum; i++) {
872 		switch (phdr[i].p_type) {
873 		case PT_LOAD:	/* Loadable segment */
874 			if (phdr[i].p_memsz == 0)
875 				break;
876 			prot = __elfN(trans_prot)(phdr[i].p_flags);
877 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
878 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
879 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
880 			    sv->sv_pagesize);
881 			if (error != 0)
882 				goto ret;
883 
884 			/*
885 			 * If this segment contains the program headers,
886 			 * remember their virtual address for the AT_PHDR
887 			 * aux entry. Static binaries don't usually include
888 			 * a PT_PHDR entry.
889 			 */
890 			if (phdr[i].p_offset == 0 &&
891 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
892 				<= phdr[i].p_filesz)
893 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
894 				    et_dyn_addr;
895 
896 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
897 			seg_size = round_page(phdr[i].p_memsz +
898 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
899 
900 			/*
901 			 * Make the largest executable segment the official
902 			 * text segment and all others data.
903 			 *
904 			 * Note that obreak() assumes that data_addr +
905 			 * data_size == end of data load area, and the ELF
906 			 * file format expects segments to be sorted by
907 			 * address.  If multiple data segments exist, the
908 			 * last one will be used.
909 			 */
910 
911 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
912 				text_size = seg_size;
913 				text_addr = seg_addr;
914 			} else {
915 				data_size = seg_size;
916 				data_addr = seg_addr;
917 			}
918 			total_size += seg_size;
919 			break;
920 		case PT_PHDR: 	/* Program header table info */
921 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
922 			break;
923 		default:
924 			break;
925 		}
926 	}
927 
928 	if (data_addr == 0 && data_size == 0) {
929 		data_addr = text_addr;
930 		data_size = text_size;
931 	}
932 
933 	entry = (u_long)hdr->e_entry + et_dyn_addr;
934 
935 	/*
936 	 * Check limits.  It should be safe to check the
937 	 * limits after loading the segments since we do
938 	 * not actually fault in all the segments pages.
939 	 */
940 	PROC_LOCK(imgp->proc);
941 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
942 		err_str = "Data segment size exceeds process limit";
943 	else if (text_size > maxtsiz)
944 		err_str = "Text segment size exceeds system limit";
945 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
946 		err_str = "Total segment size exceeds process limit";
947 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
948 		err_str = "Data segment size exceeds resource limit";
949 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
950 		err_str = "Total segment size exceeds resource limit";
951 	if (err_str != NULL) {
952 		PROC_UNLOCK(imgp->proc);
953 		uprintf("%s\n", err_str);
954 		error = ENOMEM;
955 		goto ret;
956 	}
957 
958 	vmspace = imgp->proc->p_vmspace;
959 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
960 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
961 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
962 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
963 
964 	/*
965 	 * We load the dynamic linker where a userland call
966 	 * to mmap(0, ...) would put it.  The rationale behind this
967 	 * calculation is that it leaves room for the heap to grow to
968 	 * its maximum allowed size.
969 	 */
970 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
971 	    RLIMIT_DATA));
972 	PROC_UNLOCK(imgp->proc);
973 
974 	imgp->entry_addr = entry;
975 
976 	if (interp != NULL) {
977 		have_interp = FALSE;
978 		VOP_UNLOCK(imgp->vp, 0);
979 		if (brand_info->emul_path != NULL &&
980 		    brand_info->emul_path[0] != '\0') {
981 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
982 			snprintf(path, MAXPATHLEN, "%s%s",
983 			    brand_info->emul_path, interp);
984 			error = __elfN(load_file)(imgp->proc, path, &addr,
985 			    &imgp->entry_addr, sv->sv_pagesize);
986 			free(path, M_TEMP);
987 			if (error == 0)
988 				have_interp = TRUE;
989 		}
990 		if (!have_interp && newinterp != NULL) {
991 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
992 			    &imgp->entry_addr, sv->sv_pagesize);
993 			if (error == 0)
994 				have_interp = TRUE;
995 		}
996 		if (!have_interp) {
997 			error = __elfN(load_file)(imgp->proc, interp, &addr,
998 			    &imgp->entry_addr, sv->sv_pagesize);
999 		}
1000 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1001 		if (error != 0) {
1002 			uprintf("ELF interpreter %s not found\n", interp);
1003 			goto ret;
1004 		}
1005 	} else
1006 		addr = et_dyn_addr;
1007 
1008 	/*
1009 	 * Construct auxargs table (used by the fixup routine)
1010 	 */
1011 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1012 	elf_auxargs->execfd = -1;
1013 	elf_auxargs->phdr = proghdr;
1014 	elf_auxargs->phent = hdr->e_phentsize;
1015 	elf_auxargs->phnum = hdr->e_phnum;
1016 	elf_auxargs->pagesz = PAGE_SIZE;
1017 	elf_auxargs->base = addr;
1018 	elf_auxargs->flags = 0;
1019 	elf_auxargs->entry = entry;
1020 	elf_auxargs->hdr_eflags = hdr->e_flags;
1021 
1022 	imgp->auxargs = elf_auxargs;
1023 	imgp->interpreted = 0;
1024 	imgp->reloc_base = addr;
1025 	imgp->proc->p_osrel = osrel;
1026 
1027  ret:
1028 	free(interp_buf, M_TEMP);
1029 	return (error);
1030 }
1031 
1032 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1033 
1034 int
1035 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1036 {
1037 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1038 	Elf_Addr *base;
1039 	Elf_Addr *pos;
1040 
1041 	base = (Elf_Addr *)*stack_base;
1042 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1043 
1044 	if (args->execfd != -1)
1045 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1046 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1047 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1048 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1049 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1050 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1051 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1052 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1053 #ifdef AT_EHDRFLAGS
1054 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1055 #endif
1056 	if (imgp->execpathp != 0)
1057 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1058 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1059 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1060 	if (imgp->canary != 0) {
1061 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1062 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1063 	}
1064 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1065 	if (imgp->pagesizes != 0) {
1066 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1067 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1068 	}
1069 	if (imgp->sysent->sv_timekeep_base != 0) {
1070 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1071 		    imgp->sysent->sv_timekeep_base);
1072 	}
1073 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1074 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1075 	    imgp->sysent->sv_stackprot);
1076 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1077 
1078 	free(imgp->auxargs, M_TEMP);
1079 	imgp->auxargs = NULL;
1080 
1081 	base--;
1082 	suword(base, (long)imgp->args->argc);
1083 	*stack_base = (register_t *)base;
1084 	return (0);
1085 }
1086 
1087 /*
1088  * Code for generating ELF core dumps.
1089  */
1090 
1091 typedef void (*segment_callback)(vm_map_entry_t, void *);
1092 
1093 /* Closure for cb_put_phdr(). */
1094 struct phdr_closure {
1095 	Elf_Phdr *phdr;		/* Program header to fill in */
1096 	Elf_Off offset;		/* Offset of segment in core file */
1097 };
1098 
1099 /* Closure for cb_size_segment(). */
1100 struct sseg_closure {
1101 	int count;		/* Count of writable segments. */
1102 	size_t size;		/* Total size of all writable segments. */
1103 };
1104 
1105 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1106 
1107 struct note_info {
1108 	int		type;		/* Note type. */
1109 	outfunc_t 	outfunc; 	/* Output function. */
1110 	void		*outarg;	/* Argument for the output function. */
1111 	size_t		outsize;	/* Output size. */
1112 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1113 };
1114 
1115 TAILQ_HEAD(note_info_list, note_info);
1116 
1117 /* Coredump output parameters. */
1118 struct coredump_params {
1119 	off_t		offset;
1120 	struct ucred	*active_cred;
1121 	struct ucred	*file_cred;
1122 	struct thread	*td;
1123 	struct vnode	*vp;
1124 	struct gzio_stream *gzs;
1125 };
1126 
1127 static void cb_put_phdr(vm_map_entry_t, void *);
1128 static void cb_size_segment(vm_map_entry_t, void *);
1129 static int core_write(struct coredump_params *, void *, size_t, off_t,
1130     enum uio_seg);
1131 static void each_writable_segment(struct thread *, segment_callback, void *);
1132 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1133     struct note_info_list *, size_t);
1134 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1135     size_t *);
1136 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1137 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1138 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1139 static int sbuf_drain_core_output(void *, const char *, int);
1140 static int sbuf_drain_count(void *arg, const char *data, int len);
1141 
1142 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1143 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1144 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1145 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1146 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1147 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1148 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1149 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1150 static void note_procstat_files(void *, struct sbuf *, size_t *);
1151 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1152 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1153 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1154 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1155 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1156 
1157 #ifdef GZIO
1158 extern int compress_user_cores_gzlevel;
1159 
1160 /*
1161  * Write out a core segment to the compression stream.
1162  */
1163 static int
1164 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1165 {
1166 	u_int chunk_len;
1167 	int error;
1168 
1169 	while (len > 0) {
1170 		chunk_len = MIN(len, CORE_BUF_SIZE);
1171 		copyin(base, buf, chunk_len);
1172 		error = gzio_write(p->gzs, buf, chunk_len);
1173 		if (error != 0)
1174 			break;
1175 		base += chunk_len;
1176 		len -= chunk_len;
1177 	}
1178 	return (error);
1179 }
1180 
1181 static int
1182 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1183 {
1184 
1185 	return (core_write((struct coredump_params *)arg, base, len, offset,
1186 	    UIO_SYSSPACE));
1187 }
1188 #endif /* GZIO */
1189 
1190 static int
1191 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1192     enum uio_seg seg)
1193 {
1194 
1195 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1196 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1197 	    p->active_cred, p->file_cred, NULL, p->td));
1198 }
1199 
1200 static int
1201 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1202     void *tmpbuf)
1203 {
1204 
1205 #ifdef GZIO
1206 	if (p->gzs != NULL)
1207 		return (compress_chunk(p, base, tmpbuf, len));
1208 #endif
1209 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1210 }
1211 
1212 /*
1213  * Drain into a core file.
1214  */
1215 static int
1216 sbuf_drain_core_output(void *arg, const char *data, int len)
1217 {
1218 	struct coredump_params *p;
1219 	int error, locked;
1220 
1221 	p = (struct coredump_params *)arg;
1222 
1223 	/*
1224 	 * Some kern_proc out routines that print to this sbuf may
1225 	 * call us with the process lock held. Draining with the
1226 	 * non-sleepable lock held is unsafe. The lock is needed for
1227 	 * those routines when dumping a live process. In our case we
1228 	 * can safely release the lock before draining and acquire
1229 	 * again after.
1230 	 */
1231 	locked = PROC_LOCKED(p->td->td_proc);
1232 	if (locked)
1233 		PROC_UNLOCK(p->td->td_proc);
1234 #ifdef GZIO
1235 	if (p->gzs != NULL)
1236 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1237 	else
1238 #endif
1239 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1240 		    UIO_SYSSPACE);
1241 	if (locked)
1242 		PROC_LOCK(p->td->td_proc);
1243 	if (error != 0)
1244 		return (-error);
1245 	p->offset += len;
1246 	return (len);
1247 }
1248 
1249 /*
1250  * Drain into a counter.
1251  */
1252 static int
1253 sbuf_drain_count(void *arg, const char *data __unused, int len)
1254 {
1255 	size_t *sizep;
1256 
1257 	sizep = (size_t *)arg;
1258 	*sizep += len;
1259 	return (len);
1260 }
1261 
1262 int
1263 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1264 {
1265 	struct ucred *cred = td->td_ucred;
1266 	int error = 0;
1267 	struct sseg_closure seginfo;
1268 	struct note_info_list notelst;
1269 	struct coredump_params params;
1270 	struct note_info *ninfo;
1271 	void *hdr, *tmpbuf;
1272 	size_t hdrsize, notesz, coresize;
1273 	boolean_t compress;
1274 
1275 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1276 	hdr = NULL;
1277 	tmpbuf = NULL;
1278 	TAILQ_INIT(&notelst);
1279 
1280 	/* Size the program segments. */
1281 	seginfo.count = 0;
1282 	seginfo.size = 0;
1283 	each_writable_segment(td, cb_size_segment, &seginfo);
1284 
1285 	/*
1286 	 * Collect info about the core file header area.
1287 	 */
1288 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1289 	__elfN(prepare_notes)(td, &notelst, &notesz);
1290 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1291 
1292 	/* Set up core dump parameters. */
1293 	params.offset = 0;
1294 	params.active_cred = cred;
1295 	params.file_cred = NOCRED;
1296 	params.td = td;
1297 	params.vp = vp;
1298 	params.gzs = NULL;
1299 
1300 #ifdef RACCT
1301 	if (racct_enable) {
1302 		PROC_LOCK(td->td_proc);
1303 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1304 		PROC_UNLOCK(td->td_proc);
1305 		if (error != 0) {
1306 			error = EFAULT;
1307 			goto done;
1308 		}
1309 	}
1310 #endif
1311 	if (coresize >= limit) {
1312 		error = EFAULT;
1313 		goto done;
1314 	}
1315 
1316 #ifdef GZIO
1317 	/* Create a compression stream if necessary. */
1318 	if (compress) {
1319 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1320 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1321 		if (params.gzs == NULL) {
1322 			error = EFAULT;
1323 			goto done;
1324 		}
1325 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1326         }
1327 #endif
1328 
1329 	/*
1330 	 * Allocate memory for building the header, fill it up,
1331 	 * and write it out following the notes.
1332 	 */
1333 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1334 	if (hdr == NULL) {
1335 		error = EINVAL;
1336 		goto done;
1337 	}
1338 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1339 	    notesz);
1340 
1341 	/* Write the contents of all of the writable segments. */
1342 	if (error == 0) {
1343 		Elf_Phdr *php;
1344 		off_t offset;
1345 		int i;
1346 
1347 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1348 		offset = round_page(hdrsize + notesz);
1349 		for (i = 0; i < seginfo.count; i++) {
1350 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1351 			    php->p_filesz, offset, &params, tmpbuf);
1352 			if (error != 0)
1353 				break;
1354 			offset += php->p_filesz;
1355 			php++;
1356 		}
1357 #ifdef GZIO
1358 		if (error == 0 && compress)
1359 			error = gzio_flush(params.gzs);
1360 #endif
1361 	}
1362 	if (error) {
1363 		log(LOG_WARNING,
1364 		    "Failed to write core file for process %s (error %d)\n",
1365 		    curproc->p_comm, error);
1366 	}
1367 
1368 done:
1369 #ifdef GZIO
1370 	if (compress) {
1371 		free(tmpbuf, M_TEMP);
1372 		if (params.gzs != NULL)
1373 			gzio_fini(params.gzs);
1374 	}
1375 #endif
1376 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1377 		TAILQ_REMOVE(&notelst, ninfo, link);
1378 		free(ninfo, M_TEMP);
1379 	}
1380 	if (hdr != NULL)
1381 		free(hdr, M_TEMP);
1382 
1383 	return (error);
1384 }
1385 
1386 /*
1387  * A callback for each_writable_segment() to write out the segment's
1388  * program header entry.
1389  */
1390 static void
1391 cb_put_phdr(entry, closure)
1392 	vm_map_entry_t entry;
1393 	void *closure;
1394 {
1395 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1396 	Elf_Phdr *phdr = phc->phdr;
1397 
1398 	phc->offset = round_page(phc->offset);
1399 
1400 	phdr->p_type = PT_LOAD;
1401 	phdr->p_offset = phc->offset;
1402 	phdr->p_vaddr = entry->start;
1403 	phdr->p_paddr = 0;
1404 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1405 	phdr->p_align = PAGE_SIZE;
1406 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1407 
1408 	phc->offset += phdr->p_filesz;
1409 	phc->phdr++;
1410 }
1411 
1412 /*
1413  * A callback for each_writable_segment() to gather information about
1414  * the number of segments and their total size.
1415  */
1416 static void
1417 cb_size_segment(entry, closure)
1418 	vm_map_entry_t entry;
1419 	void *closure;
1420 {
1421 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1422 
1423 	ssc->count++;
1424 	ssc->size += entry->end - entry->start;
1425 }
1426 
1427 /*
1428  * For each writable segment in the process's memory map, call the given
1429  * function with a pointer to the map entry and some arbitrary
1430  * caller-supplied data.
1431  */
1432 static void
1433 each_writable_segment(td, func, closure)
1434 	struct thread *td;
1435 	segment_callback func;
1436 	void *closure;
1437 {
1438 	struct proc *p = td->td_proc;
1439 	vm_map_t map = &p->p_vmspace->vm_map;
1440 	vm_map_entry_t entry;
1441 	vm_object_t backing_object, object;
1442 	boolean_t ignore_entry;
1443 
1444 	vm_map_lock_read(map);
1445 	for (entry = map->header.next; entry != &map->header;
1446 	    entry = entry->next) {
1447 		/*
1448 		 * Don't dump inaccessible mappings, deal with legacy
1449 		 * coredump mode.
1450 		 *
1451 		 * Note that read-only segments related to the elf binary
1452 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1453 		 * need to arbitrarily ignore such segments.
1454 		 */
1455 		if (elf_legacy_coredump) {
1456 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1457 				continue;
1458 		} else {
1459 			if ((entry->protection & VM_PROT_ALL) == 0)
1460 				continue;
1461 		}
1462 
1463 		/*
1464 		 * Dont include memory segment in the coredump if
1465 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1466 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1467 		 * kernel map).
1468 		 */
1469 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1470 			continue;
1471 
1472 		if ((object = entry->object.vm_object) == NULL)
1473 			continue;
1474 
1475 		/* Ignore memory-mapped devices and such things. */
1476 		VM_OBJECT_RLOCK(object);
1477 		while ((backing_object = object->backing_object) != NULL) {
1478 			VM_OBJECT_RLOCK(backing_object);
1479 			VM_OBJECT_RUNLOCK(object);
1480 			object = backing_object;
1481 		}
1482 		ignore_entry = object->type != OBJT_DEFAULT &&
1483 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1484 		    object->type != OBJT_PHYS;
1485 		VM_OBJECT_RUNLOCK(object);
1486 		if (ignore_entry)
1487 			continue;
1488 
1489 		(*func)(entry, closure);
1490 	}
1491 	vm_map_unlock_read(map);
1492 }
1493 
1494 /*
1495  * Write the core file header to the file, including padding up to
1496  * the page boundary.
1497  */
1498 static int
1499 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1500     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1501 {
1502 	struct note_info *ninfo;
1503 	struct sbuf *sb;
1504 	int error;
1505 
1506 	/* Fill in the header. */
1507 	bzero(hdr, hdrsize);
1508 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1509 
1510 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1511 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1512 	sbuf_start_section(sb, NULL);
1513 	sbuf_bcat(sb, hdr, hdrsize);
1514 	TAILQ_FOREACH(ninfo, notelst, link)
1515 	    __elfN(putnote)(ninfo, sb);
1516 	/* Align up to a page boundary for the program segments. */
1517 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1518 	error = sbuf_finish(sb);
1519 	sbuf_delete(sb);
1520 
1521 	return (error);
1522 }
1523 
1524 static void
1525 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1526     size_t *sizep)
1527 {
1528 	struct proc *p;
1529 	struct thread *thr;
1530 	size_t size;
1531 
1532 	p = td->td_proc;
1533 	size = 0;
1534 
1535 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1536 
1537 	/*
1538 	 * To have the debugger select the right thread (LWP) as the initial
1539 	 * thread, we dump the state of the thread passed to us in td first.
1540 	 * This is the thread that causes the core dump and thus likely to
1541 	 * be the right thread one wants to have selected in the debugger.
1542 	 */
1543 	thr = td;
1544 	while (thr != NULL) {
1545 		size += register_note(list, NT_PRSTATUS,
1546 		    __elfN(note_prstatus), thr);
1547 		size += register_note(list, NT_FPREGSET,
1548 		    __elfN(note_fpregset), thr);
1549 		size += register_note(list, NT_THRMISC,
1550 		    __elfN(note_thrmisc), thr);
1551 		size += register_note(list, -1,
1552 		    __elfN(note_threadmd), thr);
1553 
1554 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1555 		    TAILQ_NEXT(thr, td_plist);
1556 		if (thr == td)
1557 			thr = TAILQ_NEXT(thr, td_plist);
1558 	}
1559 
1560 	size += register_note(list, NT_PROCSTAT_PROC,
1561 	    __elfN(note_procstat_proc), p);
1562 	size += register_note(list, NT_PROCSTAT_FILES,
1563 	    note_procstat_files, p);
1564 	size += register_note(list, NT_PROCSTAT_VMMAP,
1565 	    note_procstat_vmmap, p);
1566 	size += register_note(list, NT_PROCSTAT_GROUPS,
1567 	    note_procstat_groups, p);
1568 	size += register_note(list, NT_PROCSTAT_UMASK,
1569 	    note_procstat_umask, p);
1570 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1571 	    note_procstat_rlimit, p);
1572 	size += register_note(list, NT_PROCSTAT_OSREL,
1573 	    note_procstat_osrel, p);
1574 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1575 	    __elfN(note_procstat_psstrings), p);
1576 	size += register_note(list, NT_PROCSTAT_AUXV,
1577 	    __elfN(note_procstat_auxv), p);
1578 
1579 	*sizep = size;
1580 }
1581 
1582 static void
1583 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1584     size_t notesz)
1585 {
1586 	Elf_Ehdr *ehdr;
1587 	Elf_Phdr *phdr;
1588 	struct phdr_closure phc;
1589 
1590 	ehdr = (Elf_Ehdr *)hdr;
1591 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1592 
1593 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1594 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1595 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1596 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1597 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1598 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1599 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1600 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1601 	ehdr->e_ident[EI_ABIVERSION] = 0;
1602 	ehdr->e_ident[EI_PAD] = 0;
1603 	ehdr->e_type = ET_CORE;
1604 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1605 	ehdr->e_machine = ELF_ARCH32;
1606 #else
1607 	ehdr->e_machine = ELF_ARCH;
1608 #endif
1609 	ehdr->e_version = EV_CURRENT;
1610 	ehdr->e_entry = 0;
1611 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1612 	ehdr->e_flags = 0;
1613 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1614 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1615 	ehdr->e_phnum = numsegs + 1;
1616 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1617 	ehdr->e_shnum = 0;
1618 	ehdr->e_shstrndx = SHN_UNDEF;
1619 
1620 	/*
1621 	 * Fill in the program header entries.
1622 	 */
1623 
1624 	/* The note segement. */
1625 	phdr->p_type = PT_NOTE;
1626 	phdr->p_offset = hdrsize;
1627 	phdr->p_vaddr = 0;
1628 	phdr->p_paddr = 0;
1629 	phdr->p_filesz = notesz;
1630 	phdr->p_memsz = 0;
1631 	phdr->p_flags = PF_R;
1632 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1633 	phdr++;
1634 
1635 	/* All the writable segments from the program. */
1636 	phc.phdr = phdr;
1637 	phc.offset = round_page(hdrsize + notesz);
1638 	each_writable_segment(td, cb_put_phdr, &phc);
1639 }
1640 
1641 static size_t
1642 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1643 {
1644 	struct note_info *ninfo;
1645 	size_t size, notesize;
1646 
1647 	size = 0;
1648 	out(arg, NULL, &size);
1649 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1650 	ninfo->type = type;
1651 	ninfo->outfunc = out;
1652 	ninfo->outarg = arg;
1653 	ninfo->outsize = size;
1654 	TAILQ_INSERT_TAIL(list, ninfo, link);
1655 
1656 	if (type == -1)
1657 		return (size);
1658 
1659 	notesize = sizeof(Elf_Note) +		/* note header */
1660 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1661 						/* note name */
1662 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1663 
1664 	return (notesize);
1665 }
1666 
1667 static size_t
1668 append_note_data(const void *src, void *dst, size_t len)
1669 {
1670 	size_t padded_len;
1671 
1672 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1673 	if (dst != NULL) {
1674 		bcopy(src, dst, len);
1675 		bzero((char *)dst + len, padded_len - len);
1676 	}
1677 	return (padded_len);
1678 }
1679 
1680 size_t
1681 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1682 {
1683 	Elf_Note *note;
1684 	char *buf;
1685 	size_t notesize;
1686 
1687 	buf = dst;
1688 	if (buf != NULL) {
1689 		note = (Elf_Note *)buf;
1690 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1691 		note->n_descsz = size;
1692 		note->n_type = type;
1693 		buf += sizeof(*note);
1694 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1695 		    sizeof(FREEBSD_ABI_VENDOR));
1696 		append_note_data(src, buf, size);
1697 		if (descp != NULL)
1698 			*descp = buf;
1699 	}
1700 
1701 	notesize = sizeof(Elf_Note) +		/* note header */
1702 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1703 						/* note name */
1704 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1705 
1706 	return (notesize);
1707 }
1708 
1709 static void
1710 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1711 {
1712 	Elf_Note note;
1713 	ssize_t old_len, sect_len;
1714 	size_t new_len, descsz, i;
1715 
1716 	if (ninfo->type == -1) {
1717 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1718 		return;
1719 	}
1720 
1721 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1722 	note.n_descsz = ninfo->outsize;
1723 	note.n_type = ninfo->type;
1724 
1725 	sbuf_bcat(sb, &note, sizeof(note));
1726 	sbuf_start_section(sb, &old_len);
1727 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1728 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1729 	if (note.n_descsz == 0)
1730 		return;
1731 	sbuf_start_section(sb, &old_len);
1732 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1733 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1734 	if (sect_len < 0)
1735 		return;
1736 
1737 	new_len = (size_t)sect_len;
1738 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1739 	if (new_len < descsz) {
1740 		/*
1741 		 * It is expected that individual note emitters will correctly
1742 		 * predict their expected output size and fill up to that size
1743 		 * themselves, padding in a format-specific way if needed.
1744 		 * However, in case they don't, just do it here with zeros.
1745 		 */
1746 		for (i = 0; i < descsz - new_len; i++)
1747 			sbuf_putc(sb, 0);
1748 	} else if (new_len > descsz) {
1749 		/*
1750 		 * We can't always truncate sb -- we may have drained some
1751 		 * of it already.
1752 		 */
1753 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1754 		    "read it (%zu > %zu).  Since it is longer than "
1755 		    "expected, this coredump's notes are corrupt.  THIS "
1756 		    "IS A BUG in the note_procstat routine for type %u.\n",
1757 		    __func__, (unsigned)note.n_type, new_len, descsz,
1758 		    (unsigned)note.n_type));
1759 	}
1760 }
1761 
1762 /*
1763  * Miscellaneous note out functions.
1764  */
1765 
1766 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1767 #include <compat/freebsd32/freebsd32.h>
1768 
1769 typedef struct prstatus32 elf_prstatus_t;
1770 typedef struct prpsinfo32 elf_prpsinfo_t;
1771 typedef struct fpreg32 elf_prfpregset_t;
1772 typedef struct fpreg32 elf_fpregset_t;
1773 typedef struct reg32 elf_gregset_t;
1774 typedef struct thrmisc32 elf_thrmisc_t;
1775 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1776 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1777 typedef uint32_t elf_ps_strings_t;
1778 #else
1779 typedef prstatus_t elf_prstatus_t;
1780 typedef prpsinfo_t elf_prpsinfo_t;
1781 typedef prfpregset_t elf_prfpregset_t;
1782 typedef prfpregset_t elf_fpregset_t;
1783 typedef gregset_t elf_gregset_t;
1784 typedef thrmisc_t elf_thrmisc_t;
1785 #define ELF_KERN_PROC_MASK	0
1786 typedef struct kinfo_proc elf_kinfo_proc_t;
1787 typedef vm_offset_t elf_ps_strings_t;
1788 #endif
1789 
1790 static void
1791 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1792 {
1793 	struct proc *p;
1794 	elf_prpsinfo_t *psinfo;
1795 
1796 	p = (struct proc *)arg;
1797 	if (sb != NULL) {
1798 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1799 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1800 		psinfo->pr_version = PRPSINFO_VERSION;
1801 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1802 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1803 		/*
1804 		 * XXX - We don't fill in the command line arguments properly
1805 		 * yet.
1806 		 */
1807 		strlcpy(psinfo->pr_psargs, p->p_comm,
1808 		    sizeof(psinfo->pr_psargs));
1809 
1810 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1811 		free(psinfo, M_TEMP);
1812 	}
1813 	*sizep = sizeof(*psinfo);
1814 }
1815 
1816 static void
1817 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1818 {
1819 	struct thread *td;
1820 	elf_prstatus_t *status;
1821 
1822 	td = (struct thread *)arg;
1823 	if (sb != NULL) {
1824 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1825 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1826 		status->pr_version = PRSTATUS_VERSION;
1827 		status->pr_statussz = sizeof(elf_prstatus_t);
1828 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1829 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1830 		status->pr_osreldate = osreldate;
1831 		status->pr_cursig = td->td_proc->p_sig;
1832 		status->pr_pid = td->td_tid;
1833 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1834 		fill_regs32(td, &status->pr_reg);
1835 #else
1836 		fill_regs(td, &status->pr_reg);
1837 #endif
1838 		sbuf_bcat(sb, status, sizeof(*status));
1839 		free(status, M_TEMP);
1840 	}
1841 	*sizep = sizeof(*status);
1842 }
1843 
1844 static void
1845 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1846 {
1847 	struct thread *td;
1848 	elf_prfpregset_t *fpregset;
1849 
1850 	td = (struct thread *)arg;
1851 	if (sb != NULL) {
1852 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1853 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1854 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1855 		fill_fpregs32(td, fpregset);
1856 #else
1857 		fill_fpregs(td, fpregset);
1858 #endif
1859 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1860 		free(fpregset, M_TEMP);
1861 	}
1862 	*sizep = sizeof(*fpregset);
1863 }
1864 
1865 static void
1866 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1867 {
1868 	struct thread *td;
1869 	elf_thrmisc_t thrmisc;
1870 
1871 	td = (struct thread *)arg;
1872 	if (sb != NULL) {
1873 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1874 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1875 		strcpy(thrmisc.pr_tname, td->td_name);
1876 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1877 	}
1878 	*sizep = sizeof(thrmisc);
1879 }
1880 
1881 /*
1882  * Allow for MD specific notes, as well as any MD
1883  * specific preparations for writing MI notes.
1884  */
1885 static void
1886 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1887 {
1888 	struct thread *td;
1889 	void *buf;
1890 	size_t size;
1891 
1892 	td = (struct thread *)arg;
1893 	size = *sizep;
1894 	if (size != 0 && sb != NULL)
1895 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1896 	else
1897 		buf = NULL;
1898 	size = 0;
1899 	__elfN(dump_thread)(td, buf, &size);
1900 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1901 	if (size != 0 && sb != NULL)
1902 		sbuf_bcat(sb, buf, size);
1903 	free(buf, M_TEMP);
1904 	*sizep = size;
1905 }
1906 
1907 #ifdef KINFO_PROC_SIZE
1908 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1909 #endif
1910 
1911 static void
1912 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1913 {
1914 	struct proc *p;
1915 	size_t size;
1916 	int structsize;
1917 
1918 	p = (struct proc *)arg;
1919 	size = sizeof(structsize) + p->p_numthreads *
1920 	    sizeof(elf_kinfo_proc_t);
1921 
1922 	if (sb != NULL) {
1923 		KASSERT(*sizep == size, ("invalid size"));
1924 		structsize = sizeof(elf_kinfo_proc_t);
1925 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1926 		sx_slock(&proctree_lock);
1927 		PROC_LOCK(p);
1928 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1929 		sx_sunlock(&proctree_lock);
1930 	}
1931 	*sizep = size;
1932 }
1933 
1934 #ifdef KINFO_FILE_SIZE
1935 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1936 #endif
1937 
1938 static void
1939 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1940 {
1941 	struct proc *p;
1942 	size_t size, sect_sz, i;
1943 	ssize_t start_len, sect_len;
1944 	int structsize, filedesc_flags;
1945 
1946 	if (coredump_pack_fileinfo)
1947 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1948 	else
1949 		filedesc_flags = 0;
1950 
1951 	p = (struct proc *)arg;
1952 	structsize = sizeof(struct kinfo_file);
1953 	if (sb == NULL) {
1954 		size = 0;
1955 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1956 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1957 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1958 		PROC_LOCK(p);
1959 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1960 		sbuf_finish(sb);
1961 		sbuf_delete(sb);
1962 		*sizep = size;
1963 	} else {
1964 		sbuf_start_section(sb, &start_len);
1965 
1966 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1967 		PROC_LOCK(p);
1968 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1969 		    filedesc_flags);
1970 
1971 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
1972 		if (sect_len < 0)
1973 			return;
1974 		sect_sz = sect_len;
1975 
1976 		KASSERT(sect_sz <= *sizep,
1977 		    ("kern_proc_filedesc_out did not respect maxlen; "
1978 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
1979 		     sect_sz - sizeof(structsize)));
1980 
1981 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
1982 			sbuf_putc(sb, 0);
1983 	}
1984 }
1985 
1986 #ifdef KINFO_VMENTRY_SIZE
1987 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1988 #endif
1989 
1990 static void
1991 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1992 {
1993 	struct proc *p;
1994 	size_t size;
1995 	int structsize, vmmap_flags;
1996 
1997 	if (coredump_pack_vmmapinfo)
1998 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
1999 	else
2000 		vmmap_flags = 0;
2001 
2002 	p = (struct proc *)arg;
2003 	structsize = sizeof(struct kinfo_vmentry);
2004 	if (sb == NULL) {
2005 		size = 0;
2006 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2007 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2008 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2009 		PROC_LOCK(p);
2010 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2011 		sbuf_finish(sb);
2012 		sbuf_delete(sb);
2013 		*sizep = size;
2014 	} else {
2015 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2016 		PROC_LOCK(p);
2017 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2018 		    vmmap_flags);
2019 	}
2020 }
2021 
2022 static void
2023 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2024 {
2025 	struct proc *p;
2026 	size_t size;
2027 	int structsize;
2028 
2029 	p = (struct proc *)arg;
2030 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2031 	if (sb != NULL) {
2032 		KASSERT(*sizep == size, ("invalid size"));
2033 		structsize = sizeof(gid_t);
2034 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2035 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2036 		    sizeof(gid_t));
2037 	}
2038 	*sizep = size;
2039 }
2040 
2041 static void
2042 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2043 {
2044 	struct proc *p;
2045 	size_t size;
2046 	int structsize;
2047 
2048 	p = (struct proc *)arg;
2049 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2050 	if (sb != NULL) {
2051 		KASSERT(*sizep == size, ("invalid size"));
2052 		structsize = sizeof(p->p_fd->fd_cmask);
2053 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2054 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2055 	}
2056 	*sizep = size;
2057 }
2058 
2059 static void
2060 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2061 {
2062 	struct proc *p;
2063 	struct rlimit rlim[RLIM_NLIMITS];
2064 	size_t size;
2065 	int structsize, i;
2066 
2067 	p = (struct proc *)arg;
2068 	size = sizeof(structsize) + sizeof(rlim);
2069 	if (sb != NULL) {
2070 		KASSERT(*sizep == size, ("invalid size"));
2071 		structsize = sizeof(rlim);
2072 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2073 		PROC_LOCK(p);
2074 		for (i = 0; i < RLIM_NLIMITS; i++)
2075 			lim_rlimit_proc(p, i, &rlim[i]);
2076 		PROC_UNLOCK(p);
2077 		sbuf_bcat(sb, rlim, sizeof(rlim));
2078 	}
2079 	*sizep = size;
2080 }
2081 
2082 static void
2083 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2084 {
2085 	struct proc *p;
2086 	size_t size;
2087 	int structsize;
2088 
2089 	p = (struct proc *)arg;
2090 	size = sizeof(structsize) + sizeof(p->p_osrel);
2091 	if (sb != NULL) {
2092 		KASSERT(*sizep == size, ("invalid size"));
2093 		structsize = sizeof(p->p_osrel);
2094 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2095 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2096 	}
2097 	*sizep = size;
2098 }
2099 
2100 static void
2101 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2102 {
2103 	struct proc *p;
2104 	elf_ps_strings_t ps_strings;
2105 	size_t size;
2106 	int structsize;
2107 
2108 	p = (struct proc *)arg;
2109 	size = sizeof(structsize) + sizeof(ps_strings);
2110 	if (sb != NULL) {
2111 		KASSERT(*sizep == size, ("invalid size"));
2112 		structsize = sizeof(ps_strings);
2113 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2114 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2115 #else
2116 		ps_strings = p->p_sysent->sv_psstrings;
2117 #endif
2118 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2119 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2120 	}
2121 	*sizep = size;
2122 }
2123 
2124 static void
2125 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2126 {
2127 	struct proc *p;
2128 	size_t size;
2129 	int structsize;
2130 
2131 	p = (struct proc *)arg;
2132 	if (sb == NULL) {
2133 		size = 0;
2134 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2135 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2136 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2137 		PHOLD(p);
2138 		proc_getauxv(curthread, p, sb);
2139 		PRELE(p);
2140 		sbuf_finish(sb);
2141 		sbuf_delete(sb);
2142 		*sizep = size;
2143 	} else {
2144 		structsize = sizeof(Elf_Auxinfo);
2145 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2146 		PHOLD(p);
2147 		proc_getauxv(curthread, p, sb);
2148 		PRELE(p);
2149 	}
2150 }
2151 
2152 static boolean_t
2153 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2154     int32_t *osrel, const Elf_Phdr *pnote)
2155 {
2156 	const Elf_Note *note, *note0, *note_end;
2157 	const char *note_name;
2158 	char *buf;
2159 	int i, error;
2160 	boolean_t res;
2161 
2162 	/* We need some limit, might as well use PAGE_SIZE. */
2163 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2164 		return (FALSE);
2165 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2166 	if (pnote->p_offset > PAGE_SIZE ||
2167 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2168 		VOP_UNLOCK(imgp->vp, 0);
2169 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2170 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2171 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2172 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2173 		    curthread->td_ucred, NOCRED, NULL, curthread);
2174 		if (error != 0) {
2175 			uprintf("i/o error PT_NOTE\n");
2176 			res = FALSE;
2177 			goto ret;
2178 		}
2179 		note = note0 = (const Elf_Note *)buf;
2180 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2181 	} else {
2182 		note = note0 = (const Elf_Note *)(imgp->image_header +
2183 		    pnote->p_offset);
2184 		note_end = (const Elf_Note *)(imgp->image_header +
2185 		    pnote->p_offset + pnote->p_filesz);
2186 		buf = NULL;
2187 	}
2188 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2189 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2190 		    (const char *)note < sizeof(Elf_Note)) {
2191 			res = FALSE;
2192 			goto ret;
2193 		}
2194 		if (note->n_namesz != checknote->hdr.n_namesz ||
2195 		    note->n_descsz != checknote->hdr.n_descsz ||
2196 		    note->n_type != checknote->hdr.n_type)
2197 			goto nextnote;
2198 		note_name = (const char *)(note + 1);
2199 		if (note_name + checknote->hdr.n_namesz >=
2200 		    (const char *)note_end || strncmp(checknote->vendor,
2201 		    note_name, checknote->hdr.n_namesz) != 0)
2202 			goto nextnote;
2203 
2204 		/*
2205 		 * Fetch the osreldate for binary
2206 		 * from the ELF OSABI-note if necessary.
2207 		 */
2208 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2209 		    checknote->trans_osrel != NULL) {
2210 			res = checknote->trans_osrel(note, osrel);
2211 			goto ret;
2212 		}
2213 		res = TRUE;
2214 		goto ret;
2215 nextnote:
2216 		note = (const Elf_Note *)((const char *)(note + 1) +
2217 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2218 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2219 	}
2220 	res = FALSE;
2221 ret:
2222 	free(buf, M_TEMP);
2223 	return (res);
2224 }
2225 
2226 /*
2227  * Try to find the appropriate ABI-note section for checknote,
2228  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2229  * first page of the image is searched, the same as for headers.
2230  */
2231 static boolean_t
2232 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2233     int32_t *osrel)
2234 {
2235 	const Elf_Phdr *phdr;
2236 	const Elf_Ehdr *hdr;
2237 	int i;
2238 
2239 	hdr = (const Elf_Ehdr *)imgp->image_header;
2240 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2241 
2242 	for (i = 0; i < hdr->e_phnum; i++) {
2243 		if (phdr[i].p_type == PT_NOTE &&
2244 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2245 			return (TRUE);
2246 	}
2247 	return (FALSE);
2248 
2249 }
2250 
2251 /*
2252  * Tell kern_execve.c about it, with a little help from the linker.
2253  */
2254 static struct execsw __elfN(execsw) = {
2255 	__CONCAT(exec_, __elfN(imgact)),
2256 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2257 };
2258 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2259 
2260 static vm_prot_t
2261 __elfN(trans_prot)(Elf_Word flags)
2262 {
2263 	vm_prot_t prot;
2264 
2265 	prot = 0;
2266 	if (flags & PF_X)
2267 		prot |= VM_PROT_EXECUTE;
2268 	if (flags & PF_W)
2269 		prot |= VM_PROT_WRITE;
2270 	if (flags & PF_R)
2271 		prot |= VM_PROT_READ;
2272 #if __ELF_WORD_SIZE == 32
2273 #if defined(__amd64__)
2274 	if (i386_read_exec && (flags & PF_R))
2275 		prot |= VM_PROT_EXECUTE;
2276 #endif
2277 #endif
2278 	return (prot);
2279 }
2280 
2281 static Elf_Word
2282 __elfN(untrans_prot)(vm_prot_t prot)
2283 {
2284 	Elf_Word flags;
2285 
2286 	flags = 0;
2287 	if (prot & VM_PROT_EXECUTE)
2288 		flags |= PF_X;
2289 	if (prot & VM_PROT_READ)
2290 		flags |= PF_R;
2291 	if (prot & VM_PROT_WRITE)
2292 		flags |= PF_W;
2293 	return (flags);
2294 }
2295