xref: /freebsd/sys/kern/imgact_elf.c (revision c6a33c8e88c5684876e670c8189d03ad25108d8a)
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 			interp_name_len = phdr[i].p_filesz;
803 			if (phdr[i].p_offset > PAGE_SIZE ||
804 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
805 				VOP_UNLOCK(imgp->vp, 0);
806 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
807 				    M_WAITOK);
808 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
809 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
810 				    interp_name_len, phdr[i].p_offset,
811 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
812 				    NOCRED, NULL, td);
813 				if (error != 0) {
814 					uprintf("i/o error PT_INTERP\n");
815 					goto ret;
816 				}
817 				interp_buf[interp_name_len] = '\0';
818 				interp = interp_buf;
819 			} else {
820 				interp = __DECONST(char *, imgp->image_header) +
821 				    phdr[i].p_offset;
822 			}
823 			break;
824 		case PT_GNU_STACK:
825 			if (__elfN(nxstack))
826 				imgp->stack_prot =
827 				    __elfN(trans_prot)(phdr[i].p_flags);
828 			imgp->stack_sz = phdr[i].p_memsz;
829 			break;
830 		}
831 	}
832 
833 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
834 	    &osrel);
835 	if (brand_info == NULL) {
836 		uprintf("ELF binary type \"%u\" not known.\n",
837 		    hdr->e_ident[EI_OSABI]);
838 		error = ENOEXEC;
839 		goto ret;
840 	}
841 	if (hdr->e_type == ET_DYN) {
842 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
843 			uprintf("Cannot execute shared object\n");
844 			error = ENOEXEC;
845 			goto ret;
846 		}
847 		/*
848 		 * Honour the base load address from the dso if it is
849 		 * non-zero for some reason.
850 		 */
851 		if (baddr == 0)
852 			et_dyn_addr = ET_DYN_LOAD_ADDR;
853 		else
854 			et_dyn_addr = 0;
855 	} else
856 		et_dyn_addr = 0;
857 	sv = brand_info->sysvec;
858 	if (interp != NULL && brand_info->interp_newpath != NULL)
859 		newinterp = brand_info->interp_newpath;
860 
861 	/*
862 	 * Avoid a possible deadlock if the current address space is destroyed
863 	 * and that address space maps the locked vnode.  In the common case,
864 	 * the locked vnode's v_usecount is decremented but remains greater
865 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
866 	 * However, in cases where the vnode lock is external, such as nullfs,
867 	 * v_usecount may become zero.
868 	 *
869 	 * The VV_TEXT flag prevents modifications to the executable while
870 	 * the vnode is unlocked.
871 	 */
872 	VOP_UNLOCK(imgp->vp, 0);
873 
874 	error = exec_new_vmspace(imgp, sv);
875 	imgp->proc->p_sysent = sv;
876 
877 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
878 	if (error != 0)
879 		goto ret;
880 
881 	for (i = 0; i < hdr->e_phnum; i++) {
882 		switch (phdr[i].p_type) {
883 		case PT_LOAD:	/* Loadable segment */
884 			if (phdr[i].p_memsz == 0)
885 				break;
886 			prot = __elfN(trans_prot)(phdr[i].p_flags);
887 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
888 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
889 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
890 			    sv->sv_pagesize);
891 			if (error != 0)
892 				goto ret;
893 
894 			/*
895 			 * If this segment contains the program headers,
896 			 * remember their virtual address for the AT_PHDR
897 			 * aux entry. Static binaries don't usually include
898 			 * a PT_PHDR entry.
899 			 */
900 			if (phdr[i].p_offset == 0 &&
901 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
902 				<= phdr[i].p_filesz)
903 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
904 				    et_dyn_addr;
905 
906 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
907 			seg_size = round_page(phdr[i].p_memsz +
908 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
909 
910 			/*
911 			 * Make the largest executable segment the official
912 			 * text segment and all others data.
913 			 *
914 			 * Note that obreak() assumes that data_addr +
915 			 * data_size == end of data load area, and the ELF
916 			 * file format expects segments to be sorted by
917 			 * address.  If multiple data segments exist, the
918 			 * last one will be used.
919 			 */
920 
921 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
922 				text_size = seg_size;
923 				text_addr = seg_addr;
924 			} else {
925 				data_size = seg_size;
926 				data_addr = seg_addr;
927 			}
928 			total_size += seg_size;
929 			break;
930 		case PT_PHDR: 	/* Program header table info */
931 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
932 			break;
933 		default:
934 			break;
935 		}
936 	}
937 
938 	if (data_addr == 0 && data_size == 0) {
939 		data_addr = text_addr;
940 		data_size = text_size;
941 	}
942 
943 	entry = (u_long)hdr->e_entry + et_dyn_addr;
944 
945 	/*
946 	 * Check limits.  It should be safe to check the
947 	 * limits after loading the segments since we do
948 	 * not actually fault in all the segments pages.
949 	 */
950 	PROC_LOCK(imgp->proc);
951 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
952 		err_str = "Data segment size exceeds process limit";
953 	else if (text_size > maxtsiz)
954 		err_str = "Text segment size exceeds system limit";
955 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
956 		err_str = "Total segment size exceeds process limit";
957 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
958 		err_str = "Data segment size exceeds resource limit";
959 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
960 		err_str = "Total segment size exceeds resource limit";
961 	if (err_str != NULL) {
962 		PROC_UNLOCK(imgp->proc);
963 		uprintf("%s\n", err_str);
964 		error = ENOMEM;
965 		goto ret;
966 	}
967 
968 	vmspace = imgp->proc->p_vmspace;
969 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
970 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
971 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
972 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
973 
974 	/*
975 	 * We load the dynamic linker where a userland call
976 	 * to mmap(0, ...) would put it.  The rationale behind this
977 	 * calculation is that it leaves room for the heap to grow to
978 	 * its maximum allowed size.
979 	 */
980 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
981 	    RLIMIT_DATA));
982 	PROC_UNLOCK(imgp->proc);
983 
984 	imgp->entry_addr = entry;
985 
986 	if (interp != NULL) {
987 		have_interp = FALSE;
988 		VOP_UNLOCK(imgp->vp, 0);
989 		if (brand_info->emul_path != NULL &&
990 		    brand_info->emul_path[0] != '\0') {
991 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
992 			snprintf(path, MAXPATHLEN, "%s%s",
993 			    brand_info->emul_path, interp);
994 			error = __elfN(load_file)(imgp->proc, path, &addr,
995 			    &imgp->entry_addr, sv->sv_pagesize);
996 			free(path, M_TEMP);
997 			if (error == 0)
998 				have_interp = TRUE;
999 		}
1000 		if (!have_interp && newinterp != NULL) {
1001 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1002 			    &imgp->entry_addr, sv->sv_pagesize);
1003 			if (error == 0)
1004 				have_interp = TRUE;
1005 		}
1006 		if (!have_interp) {
1007 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1008 			    &imgp->entry_addr, sv->sv_pagesize);
1009 		}
1010 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1011 		if (error != 0) {
1012 			uprintf("ELF interpreter %s not found\n", interp);
1013 			goto ret;
1014 		}
1015 	} else
1016 		addr = et_dyn_addr;
1017 
1018 	/*
1019 	 * Construct auxargs table (used by the fixup routine)
1020 	 */
1021 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1022 	elf_auxargs->execfd = -1;
1023 	elf_auxargs->phdr = proghdr;
1024 	elf_auxargs->phent = hdr->e_phentsize;
1025 	elf_auxargs->phnum = hdr->e_phnum;
1026 	elf_auxargs->pagesz = PAGE_SIZE;
1027 	elf_auxargs->base = addr;
1028 	elf_auxargs->flags = 0;
1029 	elf_auxargs->entry = entry;
1030 	elf_auxargs->hdr_eflags = hdr->e_flags;
1031 
1032 	imgp->auxargs = elf_auxargs;
1033 	imgp->interpreted = 0;
1034 	imgp->reloc_base = addr;
1035 	imgp->proc->p_osrel = osrel;
1036 
1037  ret:
1038 	free(interp_buf, M_TEMP);
1039 	return (error);
1040 }
1041 
1042 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1043 
1044 int
1045 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1046 {
1047 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1048 	Elf_Addr *base;
1049 	Elf_Addr *pos;
1050 
1051 	base = (Elf_Addr *)*stack_base;
1052 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1053 
1054 	if (args->execfd != -1)
1055 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1056 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1057 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1058 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1059 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1060 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1061 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1062 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1063 #ifdef AT_EHDRFLAGS
1064 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1065 #endif
1066 	if (imgp->execpathp != 0)
1067 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1068 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1069 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1070 	if (imgp->canary != 0) {
1071 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1072 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1073 	}
1074 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1075 	if (imgp->pagesizes != 0) {
1076 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1077 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1078 	}
1079 	if (imgp->sysent->sv_timekeep_base != 0) {
1080 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1081 		    imgp->sysent->sv_timekeep_base);
1082 	}
1083 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1084 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1085 	    imgp->sysent->sv_stackprot);
1086 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1087 
1088 	free(imgp->auxargs, M_TEMP);
1089 	imgp->auxargs = NULL;
1090 
1091 	base--;
1092 	suword(base, (long)imgp->args->argc);
1093 	*stack_base = (register_t *)base;
1094 	return (0);
1095 }
1096 
1097 /*
1098  * Code for generating ELF core dumps.
1099  */
1100 
1101 typedef void (*segment_callback)(vm_map_entry_t, void *);
1102 
1103 /* Closure for cb_put_phdr(). */
1104 struct phdr_closure {
1105 	Elf_Phdr *phdr;		/* Program header to fill in */
1106 	Elf_Off offset;		/* Offset of segment in core file */
1107 };
1108 
1109 /* Closure for cb_size_segment(). */
1110 struct sseg_closure {
1111 	int count;		/* Count of writable segments. */
1112 	size_t size;		/* Total size of all writable segments. */
1113 };
1114 
1115 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1116 
1117 struct note_info {
1118 	int		type;		/* Note type. */
1119 	outfunc_t 	outfunc; 	/* Output function. */
1120 	void		*outarg;	/* Argument for the output function. */
1121 	size_t		outsize;	/* Output size. */
1122 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1123 };
1124 
1125 TAILQ_HEAD(note_info_list, note_info);
1126 
1127 /* Coredump output parameters. */
1128 struct coredump_params {
1129 	off_t		offset;
1130 	struct ucred	*active_cred;
1131 	struct ucred	*file_cred;
1132 	struct thread	*td;
1133 	struct vnode	*vp;
1134 	struct gzio_stream *gzs;
1135 };
1136 
1137 static void cb_put_phdr(vm_map_entry_t, void *);
1138 static void cb_size_segment(vm_map_entry_t, void *);
1139 static int core_write(struct coredump_params *, void *, size_t, off_t,
1140     enum uio_seg);
1141 static void each_writable_segment(struct thread *, segment_callback, void *);
1142 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1143     struct note_info_list *, size_t);
1144 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1145     size_t *);
1146 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1147 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1148 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1149 static int sbuf_drain_core_output(void *, const char *, int);
1150 static int sbuf_drain_count(void *arg, const char *data, int len);
1151 
1152 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1153 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1154 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1155 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1156 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1157 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1158 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1159 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1160 static void note_procstat_files(void *, struct sbuf *, size_t *);
1161 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1162 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1163 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1164 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1165 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1166 
1167 #ifdef GZIO
1168 extern int compress_user_cores_gzlevel;
1169 
1170 /*
1171  * Write out a core segment to the compression stream.
1172  */
1173 static int
1174 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1175 {
1176 	u_int chunk_len;
1177 	int error;
1178 
1179 	while (len > 0) {
1180 		chunk_len = MIN(len, CORE_BUF_SIZE);
1181 		copyin(base, buf, chunk_len);
1182 		error = gzio_write(p->gzs, buf, chunk_len);
1183 		if (error != 0)
1184 			break;
1185 		base += chunk_len;
1186 		len -= chunk_len;
1187 	}
1188 	return (error);
1189 }
1190 
1191 static int
1192 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1193 {
1194 
1195 	return (core_write((struct coredump_params *)arg, base, len, offset,
1196 	    UIO_SYSSPACE));
1197 }
1198 #endif /* GZIO */
1199 
1200 static int
1201 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1202     enum uio_seg seg)
1203 {
1204 
1205 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1206 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1207 	    p->active_cred, p->file_cred, NULL, p->td));
1208 }
1209 
1210 static int
1211 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1212     void *tmpbuf)
1213 {
1214 
1215 #ifdef GZIO
1216 	if (p->gzs != NULL)
1217 		return (compress_chunk(p, base, tmpbuf, len));
1218 #endif
1219 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1220 }
1221 
1222 /*
1223  * Drain into a core file.
1224  */
1225 static int
1226 sbuf_drain_core_output(void *arg, const char *data, int len)
1227 {
1228 	struct coredump_params *p;
1229 	int error, locked;
1230 
1231 	p = (struct coredump_params *)arg;
1232 
1233 	/*
1234 	 * Some kern_proc out routines that print to this sbuf may
1235 	 * call us with the process lock held. Draining with the
1236 	 * non-sleepable lock held is unsafe. The lock is needed for
1237 	 * those routines when dumping a live process. In our case we
1238 	 * can safely release the lock before draining and acquire
1239 	 * again after.
1240 	 */
1241 	locked = PROC_LOCKED(p->td->td_proc);
1242 	if (locked)
1243 		PROC_UNLOCK(p->td->td_proc);
1244 #ifdef GZIO
1245 	if (p->gzs != NULL)
1246 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1247 	else
1248 #endif
1249 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1250 		    UIO_SYSSPACE);
1251 	if (locked)
1252 		PROC_LOCK(p->td->td_proc);
1253 	if (error != 0)
1254 		return (-error);
1255 	p->offset += len;
1256 	return (len);
1257 }
1258 
1259 /*
1260  * Drain into a counter.
1261  */
1262 static int
1263 sbuf_drain_count(void *arg, const char *data __unused, int len)
1264 {
1265 	size_t *sizep;
1266 
1267 	sizep = (size_t *)arg;
1268 	*sizep += len;
1269 	return (len);
1270 }
1271 
1272 int
1273 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1274 {
1275 	struct ucred *cred = td->td_ucred;
1276 	int error = 0;
1277 	struct sseg_closure seginfo;
1278 	struct note_info_list notelst;
1279 	struct coredump_params params;
1280 	struct note_info *ninfo;
1281 	void *hdr, *tmpbuf;
1282 	size_t hdrsize, notesz, coresize;
1283 #ifdef GZIO
1284 	boolean_t compress;
1285 
1286 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1287 #endif
1288 	hdr = NULL;
1289 	tmpbuf = NULL;
1290 	TAILQ_INIT(&notelst);
1291 
1292 	/* Size the program segments. */
1293 	seginfo.count = 0;
1294 	seginfo.size = 0;
1295 	each_writable_segment(td, cb_size_segment, &seginfo);
1296 
1297 	/*
1298 	 * Collect info about the core file header area.
1299 	 */
1300 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1301 	__elfN(prepare_notes)(td, &notelst, &notesz);
1302 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1303 
1304 	/* Set up core dump parameters. */
1305 	params.offset = 0;
1306 	params.active_cred = cred;
1307 	params.file_cred = NOCRED;
1308 	params.td = td;
1309 	params.vp = vp;
1310 	params.gzs = NULL;
1311 
1312 #ifdef RACCT
1313 	if (racct_enable) {
1314 		PROC_LOCK(td->td_proc);
1315 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1316 		PROC_UNLOCK(td->td_proc);
1317 		if (error != 0) {
1318 			error = EFAULT;
1319 			goto done;
1320 		}
1321 	}
1322 #endif
1323 	if (coresize >= limit) {
1324 		error = EFAULT;
1325 		goto done;
1326 	}
1327 
1328 #ifdef GZIO
1329 	/* Create a compression stream if necessary. */
1330 	if (compress) {
1331 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1332 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1333 		if (params.gzs == NULL) {
1334 			error = EFAULT;
1335 			goto done;
1336 		}
1337 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1338         }
1339 #endif
1340 
1341 	/*
1342 	 * Allocate memory for building the header, fill it up,
1343 	 * and write it out following the notes.
1344 	 */
1345 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1346 	if (hdr == NULL) {
1347 		error = EINVAL;
1348 		goto done;
1349 	}
1350 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1351 	    notesz);
1352 
1353 	/* Write the contents of all of the writable segments. */
1354 	if (error == 0) {
1355 		Elf_Phdr *php;
1356 		off_t offset;
1357 		int i;
1358 
1359 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1360 		offset = round_page(hdrsize + notesz);
1361 		for (i = 0; i < seginfo.count; i++) {
1362 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1363 			    php->p_filesz, offset, &params, tmpbuf);
1364 			if (error != 0)
1365 				break;
1366 			offset += php->p_filesz;
1367 			php++;
1368 		}
1369 #ifdef GZIO
1370 		if (error == 0 && compress)
1371 			error = gzio_flush(params.gzs);
1372 #endif
1373 	}
1374 	if (error) {
1375 		log(LOG_WARNING,
1376 		    "Failed to write core file for process %s (error %d)\n",
1377 		    curproc->p_comm, error);
1378 	}
1379 
1380 done:
1381 #ifdef GZIO
1382 	if (compress) {
1383 		free(tmpbuf, M_TEMP);
1384 		if (params.gzs != NULL)
1385 			gzio_fini(params.gzs);
1386 	}
1387 #endif
1388 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1389 		TAILQ_REMOVE(&notelst, ninfo, link);
1390 		free(ninfo, M_TEMP);
1391 	}
1392 	if (hdr != NULL)
1393 		free(hdr, M_TEMP);
1394 
1395 	return (error);
1396 }
1397 
1398 /*
1399  * A callback for each_writable_segment() to write out the segment's
1400  * program header entry.
1401  */
1402 static void
1403 cb_put_phdr(entry, closure)
1404 	vm_map_entry_t entry;
1405 	void *closure;
1406 {
1407 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1408 	Elf_Phdr *phdr = phc->phdr;
1409 
1410 	phc->offset = round_page(phc->offset);
1411 
1412 	phdr->p_type = PT_LOAD;
1413 	phdr->p_offset = phc->offset;
1414 	phdr->p_vaddr = entry->start;
1415 	phdr->p_paddr = 0;
1416 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1417 	phdr->p_align = PAGE_SIZE;
1418 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1419 
1420 	phc->offset += phdr->p_filesz;
1421 	phc->phdr++;
1422 }
1423 
1424 /*
1425  * A callback for each_writable_segment() to gather information about
1426  * the number of segments and their total size.
1427  */
1428 static void
1429 cb_size_segment(entry, closure)
1430 	vm_map_entry_t entry;
1431 	void *closure;
1432 {
1433 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1434 
1435 	ssc->count++;
1436 	ssc->size += entry->end - entry->start;
1437 }
1438 
1439 /*
1440  * For each writable segment in the process's memory map, call the given
1441  * function with a pointer to the map entry and some arbitrary
1442  * caller-supplied data.
1443  */
1444 static void
1445 each_writable_segment(td, func, closure)
1446 	struct thread *td;
1447 	segment_callback func;
1448 	void *closure;
1449 {
1450 	struct proc *p = td->td_proc;
1451 	vm_map_t map = &p->p_vmspace->vm_map;
1452 	vm_map_entry_t entry;
1453 	vm_object_t backing_object, object;
1454 	boolean_t ignore_entry;
1455 
1456 	vm_map_lock_read(map);
1457 	for (entry = map->header.next; entry != &map->header;
1458 	    entry = entry->next) {
1459 		/*
1460 		 * Don't dump inaccessible mappings, deal with legacy
1461 		 * coredump mode.
1462 		 *
1463 		 * Note that read-only segments related to the elf binary
1464 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1465 		 * need to arbitrarily ignore such segments.
1466 		 */
1467 		if (elf_legacy_coredump) {
1468 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1469 				continue;
1470 		} else {
1471 			if ((entry->protection & VM_PROT_ALL) == 0)
1472 				continue;
1473 		}
1474 
1475 		/*
1476 		 * Dont include memory segment in the coredump if
1477 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1478 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1479 		 * kernel map).
1480 		 */
1481 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1482 			continue;
1483 
1484 		if ((object = entry->object.vm_object) == NULL)
1485 			continue;
1486 
1487 		/* Ignore memory-mapped devices and such things. */
1488 		VM_OBJECT_RLOCK(object);
1489 		while ((backing_object = object->backing_object) != NULL) {
1490 			VM_OBJECT_RLOCK(backing_object);
1491 			VM_OBJECT_RUNLOCK(object);
1492 			object = backing_object;
1493 		}
1494 		ignore_entry = object->type != OBJT_DEFAULT &&
1495 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1496 		    object->type != OBJT_PHYS;
1497 		VM_OBJECT_RUNLOCK(object);
1498 		if (ignore_entry)
1499 			continue;
1500 
1501 		(*func)(entry, closure);
1502 	}
1503 	vm_map_unlock_read(map);
1504 }
1505 
1506 /*
1507  * Write the core file header to the file, including padding up to
1508  * the page boundary.
1509  */
1510 static int
1511 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1512     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1513 {
1514 	struct note_info *ninfo;
1515 	struct sbuf *sb;
1516 	int error;
1517 
1518 	/* Fill in the header. */
1519 	bzero(hdr, hdrsize);
1520 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1521 
1522 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1523 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1524 	sbuf_start_section(sb, NULL);
1525 	sbuf_bcat(sb, hdr, hdrsize);
1526 	TAILQ_FOREACH(ninfo, notelst, link)
1527 	    __elfN(putnote)(ninfo, sb);
1528 	/* Align up to a page boundary for the program segments. */
1529 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1530 	error = sbuf_finish(sb);
1531 	sbuf_delete(sb);
1532 
1533 	return (error);
1534 }
1535 
1536 static void
1537 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1538     size_t *sizep)
1539 {
1540 	struct proc *p;
1541 	struct thread *thr;
1542 	size_t size;
1543 
1544 	p = td->td_proc;
1545 	size = 0;
1546 
1547 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1548 
1549 	/*
1550 	 * To have the debugger select the right thread (LWP) as the initial
1551 	 * thread, we dump the state of the thread passed to us in td first.
1552 	 * This is the thread that causes the core dump and thus likely to
1553 	 * be the right thread one wants to have selected in the debugger.
1554 	 */
1555 	thr = td;
1556 	while (thr != NULL) {
1557 		size += register_note(list, NT_PRSTATUS,
1558 		    __elfN(note_prstatus), thr);
1559 		size += register_note(list, NT_FPREGSET,
1560 		    __elfN(note_fpregset), thr);
1561 		size += register_note(list, NT_THRMISC,
1562 		    __elfN(note_thrmisc), thr);
1563 		size += register_note(list, -1,
1564 		    __elfN(note_threadmd), thr);
1565 
1566 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1567 		    TAILQ_NEXT(thr, td_plist);
1568 		if (thr == td)
1569 			thr = TAILQ_NEXT(thr, td_plist);
1570 	}
1571 
1572 	size += register_note(list, NT_PROCSTAT_PROC,
1573 	    __elfN(note_procstat_proc), p);
1574 	size += register_note(list, NT_PROCSTAT_FILES,
1575 	    note_procstat_files, p);
1576 	size += register_note(list, NT_PROCSTAT_VMMAP,
1577 	    note_procstat_vmmap, p);
1578 	size += register_note(list, NT_PROCSTAT_GROUPS,
1579 	    note_procstat_groups, p);
1580 	size += register_note(list, NT_PROCSTAT_UMASK,
1581 	    note_procstat_umask, p);
1582 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1583 	    note_procstat_rlimit, p);
1584 	size += register_note(list, NT_PROCSTAT_OSREL,
1585 	    note_procstat_osrel, p);
1586 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1587 	    __elfN(note_procstat_psstrings), p);
1588 	size += register_note(list, NT_PROCSTAT_AUXV,
1589 	    __elfN(note_procstat_auxv), p);
1590 
1591 	*sizep = size;
1592 }
1593 
1594 static void
1595 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1596     size_t notesz)
1597 {
1598 	Elf_Ehdr *ehdr;
1599 	Elf_Phdr *phdr;
1600 	struct phdr_closure phc;
1601 
1602 	ehdr = (Elf_Ehdr *)hdr;
1603 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1604 
1605 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1606 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1607 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1608 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1609 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1610 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1611 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1612 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1613 	ehdr->e_ident[EI_ABIVERSION] = 0;
1614 	ehdr->e_ident[EI_PAD] = 0;
1615 	ehdr->e_type = ET_CORE;
1616 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1617 	ehdr->e_machine = ELF_ARCH32;
1618 #else
1619 	ehdr->e_machine = ELF_ARCH;
1620 #endif
1621 	ehdr->e_version = EV_CURRENT;
1622 	ehdr->e_entry = 0;
1623 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1624 	ehdr->e_flags = 0;
1625 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1626 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1627 	ehdr->e_phnum = numsegs + 1;
1628 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1629 	ehdr->e_shnum = 0;
1630 	ehdr->e_shstrndx = SHN_UNDEF;
1631 
1632 	/*
1633 	 * Fill in the program header entries.
1634 	 */
1635 
1636 	/* The note segement. */
1637 	phdr->p_type = PT_NOTE;
1638 	phdr->p_offset = hdrsize;
1639 	phdr->p_vaddr = 0;
1640 	phdr->p_paddr = 0;
1641 	phdr->p_filesz = notesz;
1642 	phdr->p_memsz = 0;
1643 	phdr->p_flags = PF_R;
1644 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1645 	phdr++;
1646 
1647 	/* All the writable segments from the program. */
1648 	phc.phdr = phdr;
1649 	phc.offset = round_page(hdrsize + notesz);
1650 	each_writable_segment(td, cb_put_phdr, &phc);
1651 }
1652 
1653 static size_t
1654 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1655 {
1656 	struct note_info *ninfo;
1657 	size_t size, notesize;
1658 
1659 	size = 0;
1660 	out(arg, NULL, &size);
1661 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1662 	ninfo->type = type;
1663 	ninfo->outfunc = out;
1664 	ninfo->outarg = arg;
1665 	ninfo->outsize = size;
1666 	TAILQ_INSERT_TAIL(list, ninfo, link);
1667 
1668 	if (type == -1)
1669 		return (size);
1670 
1671 	notesize = sizeof(Elf_Note) +		/* note header */
1672 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1673 						/* note name */
1674 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1675 
1676 	return (notesize);
1677 }
1678 
1679 static size_t
1680 append_note_data(const void *src, void *dst, size_t len)
1681 {
1682 	size_t padded_len;
1683 
1684 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1685 	if (dst != NULL) {
1686 		bcopy(src, dst, len);
1687 		bzero((char *)dst + len, padded_len - len);
1688 	}
1689 	return (padded_len);
1690 }
1691 
1692 size_t
1693 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1694 {
1695 	Elf_Note *note;
1696 	char *buf;
1697 	size_t notesize;
1698 
1699 	buf = dst;
1700 	if (buf != NULL) {
1701 		note = (Elf_Note *)buf;
1702 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1703 		note->n_descsz = size;
1704 		note->n_type = type;
1705 		buf += sizeof(*note);
1706 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1707 		    sizeof(FREEBSD_ABI_VENDOR));
1708 		append_note_data(src, buf, size);
1709 		if (descp != NULL)
1710 			*descp = buf;
1711 	}
1712 
1713 	notesize = sizeof(Elf_Note) +		/* note header */
1714 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1715 						/* note name */
1716 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1717 
1718 	return (notesize);
1719 }
1720 
1721 static void
1722 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1723 {
1724 	Elf_Note note;
1725 	ssize_t old_len, sect_len;
1726 	size_t new_len, descsz, i;
1727 
1728 	if (ninfo->type == -1) {
1729 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1730 		return;
1731 	}
1732 
1733 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1734 	note.n_descsz = ninfo->outsize;
1735 	note.n_type = ninfo->type;
1736 
1737 	sbuf_bcat(sb, &note, sizeof(note));
1738 	sbuf_start_section(sb, &old_len);
1739 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1740 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1741 	if (note.n_descsz == 0)
1742 		return;
1743 	sbuf_start_section(sb, &old_len);
1744 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1745 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1746 	if (sect_len < 0)
1747 		return;
1748 
1749 	new_len = (size_t)sect_len;
1750 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1751 	if (new_len < descsz) {
1752 		/*
1753 		 * It is expected that individual note emitters will correctly
1754 		 * predict their expected output size and fill up to that size
1755 		 * themselves, padding in a format-specific way if needed.
1756 		 * However, in case they don't, just do it here with zeros.
1757 		 */
1758 		for (i = 0; i < descsz - new_len; i++)
1759 			sbuf_putc(sb, 0);
1760 	} else if (new_len > descsz) {
1761 		/*
1762 		 * We can't always truncate sb -- we may have drained some
1763 		 * of it already.
1764 		 */
1765 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1766 		    "read it (%zu > %zu).  Since it is longer than "
1767 		    "expected, this coredump's notes are corrupt.  THIS "
1768 		    "IS A BUG in the note_procstat routine for type %u.\n",
1769 		    __func__, (unsigned)note.n_type, new_len, descsz,
1770 		    (unsigned)note.n_type));
1771 	}
1772 }
1773 
1774 /*
1775  * Miscellaneous note out functions.
1776  */
1777 
1778 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1779 #include <compat/freebsd32/freebsd32.h>
1780 
1781 typedef struct prstatus32 elf_prstatus_t;
1782 typedef struct prpsinfo32 elf_prpsinfo_t;
1783 typedef struct fpreg32 elf_prfpregset_t;
1784 typedef struct fpreg32 elf_fpregset_t;
1785 typedef struct reg32 elf_gregset_t;
1786 typedef struct thrmisc32 elf_thrmisc_t;
1787 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1788 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1789 typedef uint32_t elf_ps_strings_t;
1790 #else
1791 typedef prstatus_t elf_prstatus_t;
1792 typedef prpsinfo_t elf_prpsinfo_t;
1793 typedef prfpregset_t elf_prfpregset_t;
1794 typedef prfpregset_t elf_fpregset_t;
1795 typedef gregset_t elf_gregset_t;
1796 typedef thrmisc_t elf_thrmisc_t;
1797 #define ELF_KERN_PROC_MASK	0
1798 typedef struct kinfo_proc elf_kinfo_proc_t;
1799 typedef vm_offset_t elf_ps_strings_t;
1800 #endif
1801 
1802 static void
1803 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1804 {
1805 	struct proc *p;
1806 	elf_prpsinfo_t *psinfo;
1807 
1808 	p = (struct proc *)arg;
1809 	if (sb != NULL) {
1810 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1811 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1812 		psinfo->pr_version = PRPSINFO_VERSION;
1813 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1814 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1815 		/*
1816 		 * XXX - We don't fill in the command line arguments properly
1817 		 * yet.
1818 		 */
1819 		strlcpy(psinfo->pr_psargs, p->p_comm,
1820 		    sizeof(psinfo->pr_psargs));
1821 
1822 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1823 		free(psinfo, M_TEMP);
1824 	}
1825 	*sizep = sizeof(*psinfo);
1826 }
1827 
1828 static void
1829 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1830 {
1831 	struct thread *td;
1832 	elf_prstatus_t *status;
1833 
1834 	td = (struct thread *)arg;
1835 	if (sb != NULL) {
1836 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1837 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1838 		status->pr_version = PRSTATUS_VERSION;
1839 		status->pr_statussz = sizeof(elf_prstatus_t);
1840 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1841 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1842 		status->pr_osreldate = osreldate;
1843 		status->pr_cursig = td->td_proc->p_sig;
1844 		status->pr_pid = td->td_tid;
1845 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1846 		fill_regs32(td, &status->pr_reg);
1847 #else
1848 		fill_regs(td, &status->pr_reg);
1849 #endif
1850 		sbuf_bcat(sb, status, sizeof(*status));
1851 		free(status, M_TEMP);
1852 	}
1853 	*sizep = sizeof(*status);
1854 }
1855 
1856 static void
1857 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1858 {
1859 	struct thread *td;
1860 	elf_prfpregset_t *fpregset;
1861 
1862 	td = (struct thread *)arg;
1863 	if (sb != NULL) {
1864 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1865 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1866 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1867 		fill_fpregs32(td, fpregset);
1868 #else
1869 		fill_fpregs(td, fpregset);
1870 #endif
1871 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1872 		free(fpregset, M_TEMP);
1873 	}
1874 	*sizep = sizeof(*fpregset);
1875 }
1876 
1877 static void
1878 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1879 {
1880 	struct thread *td;
1881 	elf_thrmisc_t thrmisc;
1882 
1883 	td = (struct thread *)arg;
1884 	if (sb != NULL) {
1885 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1886 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1887 		strcpy(thrmisc.pr_tname, td->td_name);
1888 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1889 	}
1890 	*sizep = sizeof(thrmisc);
1891 }
1892 
1893 /*
1894  * Allow for MD specific notes, as well as any MD
1895  * specific preparations for writing MI notes.
1896  */
1897 static void
1898 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1899 {
1900 	struct thread *td;
1901 	void *buf;
1902 	size_t size;
1903 
1904 	td = (struct thread *)arg;
1905 	size = *sizep;
1906 	if (size != 0 && sb != NULL)
1907 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1908 	else
1909 		buf = NULL;
1910 	size = 0;
1911 	__elfN(dump_thread)(td, buf, &size);
1912 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1913 	if (size != 0 && sb != NULL)
1914 		sbuf_bcat(sb, buf, size);
1915 	free(buf, M_TEMP);
1916 	*sizep = size;
1917 }
1918 
1919 #ifdef KINFO_PROC_SIZE
1920 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1921 #endif
1922 
1923 static void
1924 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1925 {
1926 	struct proc *p;
1927 	size_t size;
1928 	int structsize;
1929 
1930 	p = (struct proc *)arg;
1931 	size = sizeof(structsize) + p->p_numthreads *
1932 	    sizeof(elf_kinfo_proc_t);
1933 
1934 	if (sb != NULL) {
1935 		KASSERT(*sizep == size, ("invalid size"));
1936 		structsize = sizeof(elf_kinfo_proc_t);
1937 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1938 		sx_slock(&proctree_lock);
1939 		PROC_LOCK(p);
1940 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1941 		sx_sunlock(&proctree_lock);
1942 	}
1943 	*sizep = size;
1944 }
1945 
1946 #ifdef KINFO_FILE_SIZE
1947 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1948 #endif
1949 
1950 static void
1951 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1952 {
1953 	struct proc *p;
1954 	size_t size, sect_sz, i;
1955 	ssize_t start_len, sect_len;
1956 	int structsize, filedesc_flags;
1957 
1958 	if (coredump_pack_fileinfo)
1959 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1960 	else
1961 		filedesc_flags = 0;
1962 
1963 	p = (struct proc *)arg;
1964 	structsize = sizeof(struct kinfo_file);
1965 	if (sb == NULL) {
1966 		size = 0;
1967 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1968 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1969 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1970 		PROC_LOCK(p);
1971 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1972 		sbuf_finish(sb);
1973 		sbuf_delete(sb);
1974 		*sizep = size;
1975 	} else {
1976 		sbuf_start_section(sb, &start_len);
1977 
1978 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1979 		PROC_LOCK(p);
1980 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1981 		    filedesc_flags);
1982 
1983 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
1984 		if (sect_len < 0)
1985 			return;
1986 		sect_sz = sect_len;
1987 
1988 		KASSERT(sect_sz <= *sizep,
1989 		    ("kern_proc_filedesc_out did not respect maxlen; "
1990 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
1991 		     sect_sz - sizeof(structsize)));
1992 
1993 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
1994 			sbuf_putc(sb, 0);
1995 	}
1996 }
1997 
1998 #ifdef KINFO_VMENTRY_SIZE
1999 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2000 #endif
2001 
2002 static void
2003 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2004 {
2005 	struct proc *p;
2006 	size_t size;
2007 	int structsize, vmmap_flags;
2008 
2009 	if (coredump_pack_vmmapinfo)
2010 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2011 	else
2012 		vmmap_flags = 0;
2013 
2014 	p = (struct proc *)arg;
2015 	structsize = sizeof(struct kinfo_vmentry);
2016 	if (sb == NULL) {
2017 		size = 0;
2018 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2019 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2020 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2021 		PROC_LOCK(p);
2022 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2023 		sbuf_finish(sb);
2024 		sbuf_delete(sb);
2025 		*sizep = size;
2026 	} else {
2027 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2028 		PROC_LOCK(p);
2029 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2030 		    vmmap_flags);
2031 	}
2032 }
2033 
2034 static void
2035 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2036 {
2037 	struct proc *p;
2038 	size_t size;
2039 	int structsize;
2040 
2041 	p = (struct proc *)arg;
2042 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2043 	if (sb != NULL) {
2044 		KASSERT(*sizep == size, ("invalid size"));
2045 		structsize = sizeof(gid_t);
2046 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2047 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2048 		    sizeof(gid_t));
2049 	}
2050 	*sizep = size;
2051 }
2052 
2053 static void
2054 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2055 {
2056 	struct proc *p;
2057 	size_t size;
2058 	int structsize;
2059 
2060 	p = (struct proc *)arg;
2061 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2062 	if (sb != NULL) {
2063 		KASSERT(*sizep == size, ("invalid size"));
2064 		structsize = sizeof(p->p_fd->fd_cmask);
2065 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2066 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2067 	}
2068 	*sizep = size;
2069 }
2070 
2071 static void
2072 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2073 {
2074 	struct proc *p;
2075 	struct rlimit rlim[RLIM_NLIMITS];
2076 	size_t size;
2077 	int structsize, i;
2078 
2079 	p = (struct proc *)arg;
2080 	size = sizeof(structsize) + sizeof(rlim);
2081 	if (sb != NULL) {
2082 		KASSERT(*sizep == size, ("invalid size"));
2083 		structsize = sizeof(rlim);
2084 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2085 		PROC_LOCK(p);
2086 		for (i = 0; i < RLIM_NLIMITS; i++)
2087 			lim_rlimit_proc(p, i, &rlim[i]);
2088 		PROC_UNLOCK(p);
2089 		sbuf_bcat(sb, rlim, sizeof(rlim));
2090 	}
2091 	*sizep = size;
2092 }
2093 
2094 static void
2095 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2096 {
2097 	struct proc *p;
2098 	size_t size;
2099 	int structsize;
2100 
2101 	p = (struct proc *)arg;
2102 	size = sizeof(structsize) + sizeof(p->p_osrel);
2103 	if (sb != NULL) {
2104 		KASSERT(*sizep == size, ("invalid size"));
2105 		structsize = sizeof(p->p_osrel);
2106 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2107 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2108 	}
2109 	*sizep = size;
2110 }
2111 
2112 static void
2113 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2114 {
2115 	struct proc *p;
2116 	elf_ps_strings_t ps_strings;
2117 	size_t size;
2118 	int structsize;
2119 
2120 	p = (struct proc *)arg;
2121 	size = sizeof(structsize) + sizeof(ps_strings);
2122 	if (sb != NULL) {
2123 		KASSERT(*sizep == size, ("invalid size"));
2124 		structsize = sizeof(ps_strings);
2125 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2126 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2127 #else
2128 		ps_strings = p->p_sysent->sv_psstrings;
2129 #endif
2130 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2131 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2132 	}
2133 	*sizep = size;
2134 }
2135 
2136 static void
2137 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2138 {
2139 	struct proc *p;
2140 	size_t size;
2141 	int structsize;
2142 
2143 	p = (struct proc *)arg;
2144 	if (sb == NULL) {
2145 		size = 0;
2146 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2147 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2148 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2149 		PHOLD(p);
2150 		proc_getauxv(curthread, p, sb);
2151 		PRELE(p);
2152 		sbuf_finish(sb);
2153 		sbuf_delete(sb);
2154 		*sizep = size;
2155 	} else {
2156 		structsize = sizeof(Elf_Auxinfo);
2157 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2158 		PHOLD(p);
2159 		proc_getauxv(curthread, p, sb);
2160 		PRELE(p);
2161 	}
2162 }
2163 
2164 static boolean_t
2165 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2166     int32_t *osrel, const Elf_Phdr *pnote)
2167 {
2168 	const Elf_Note *note, *note0, *note_end;
2169 	const char *note_name;
2170 	char *buf;
2171 	int i, error;
2172 	boolean_t res;
2173 
2174 	/* We need some limit, might as well use PAGE_SIZE. */
2175 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2176 		return (FALSE);
2177 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2178 	if (pnote->p_offset > PAGE_SIZE ||
2179 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2180 		VOP_UNLOCK(imgp->vp, 0);
2181 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2182 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2183 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2184 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2185 		    curthread->td_ucred, NOCRED, NULL, curthread);
2186 		if (error != 0) {
2187 			uprintf("i/o error PT_NOTE\n");
2188 			res = FALSE;
2189 			goto ret;
2190 		}
2191 		note = note0 = (const Elf_Note *)buf;
2192 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2193 	} else {
2194 		note = note0 = (const Elf_Note *)(imgp->image_header +
2195 		    pnote->p_offset);
2196 		note_end = (const Elf_Note *)(imgp->image_header +
2197 		    pnote->p_offset + pnote->p_filesz);
2198 		buf = NULL;
2199 	}
2200 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2201 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2202 		    (const char *)note < sizeof(Elf_Note)) {
2203 			res = FALSE;
2204 			goto ret;
2205 		}
2206 		if (note->n_namesz != checknote->hdr.n_namesz ||
2207 		    note->n_descsz != checknote->hdr.n_descsz ||
2208 		    note->n_type != checknote->hdr.n_type)
2209 			goto nextnote;
2210 		note_name = (const char *)(note + 1);
2211 		if (note_name + checknote->hdr.n_namesz >=
2212 		    (const char *)note_end || strncmp(checknote->vendor,
2213 		    note_name, checknote->hdr.n_namesz) != 0)
2214 			goto nextnote;
2215 
2216 		/*
2217 		 * Fetch the osreldate for binary
2218 		 * from the ELF OSABI-note if necessary.
2219 		 */
2220 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2221 		    checknote->trans_osrel != NULL) {
2222 			res = checknote->trans_osrel(note, osrel);
2223 			goto ret;
2224 		}
2225 		res = TRUE;
2226 		goto ret;
2227 nextnote:
2228 		note = (const Elf_Note *)((const char *)(note + 1) +
2229 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2230 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2231 	}
2232 	res = FALSE;
2233 ret:
2234 	free(buf, M_TEMP);
2235 	return (res);
2236 }
2237 
2238 /*
2239  * Try to find the appropriate ABI-note section for checknote,
2240  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2241  * first page of the image is searched, the same as for headers.
2242  */
2243 static boolean_t
2244 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2245     int32_t *osrel)
2246 {
2247 	const Elf_Phdr *phdr;
2248 	const Elf_Ehdr *hdr;
2249 	int i;
2250 
2251 	hdr = (const Elf_Ehdr *)imgp->image_header;
2252 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2253 
2254 	for (i = 0; i < hdr->e_phnum; i++) {
2255 		if (phdr[i].p_type == PT_NOTE &&
2256 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2257 			return (TRUE);
2258 	}
2259 	return (FALSE);
2260 
2261 }
2262 
2263 /*
2264  * Tell kern_execve.c about it, with a little help from the linker.
2265  */
2266 static struct execsw __elfN(execsw) = {
2267 	__CONCAT(exec_, __elfN(imgact)),
2268 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2269 };
2270 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2271 
2272 static vm_prot_t
2273 __elfN(trans_prot)(Elf_Word flags)
2274 {
2275 	vm_prot_t prot;
2276 
2277 	prot = 0;
2278 	if (flags & PF_X)
2279 		prot |= VM_PROT_EXECUTE;
2280 	if (flags & PF_W)
2281 		prot |= VM_PROT_WRITE;
2282 	if (flags & PF_R)
2283 		prot |= VM_PROT_READ;
2284 #if __ELF_WORD_SIZE == 32
2285 #if defined(__amd64__)
2286 	if (i386_read_exec && (flags & PF_R))
2287 		prot |= VM_PROT_EXECUTE;
2288 #endif
2289 #endif
2290 	return (prot);
2291 }
2292 
2293 static Elf_Word
2294 __elfN(untrans_prot)(vm_prot_t prot)
2295 {
2296 	Elf_Word flags;
2297 
2298 	flags = 0;
2299 	if (prot & VM_PROT_EXECUTE)
2300 		flags |= PF_X;
2301 	if (prot & VM_PROT_READ)
2302 		flags |= PF_R;
2303 	if (prot & VM_PROT_WRITE)
2304 		flags |= PF_W;
2305 	return (flags);
2306 }
2307