xref: /freebsd/sys/kern/imgact_elf.c (revision c77547d2f9e419914c3dc50d8d0fc038c1c8ea05)
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     "include all and only RW pages in core dumps");
118 
119 int __elfN(nxstack) =
120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
121     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
122 	1;
123 #else
124 	0;
125 #endif
126 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
127     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
128     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
129 
130 #if __ELF_WORD_SIZE == 32
131 #if defined(__amd64__)
132 int i386_read_exec = 0;
133 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
134     "enable execution from readable segments");
135 #endif
136 #endif
137 
138 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
139 
140 #define	trunc_page_ps(va, ps)	rounddown2(va, ps)
141 #define	round_page_ps(va, ps)	roundup2(va, ps)
142 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
143 
144 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
145 
146 Elf_Brandnote __elfN(freebsd_brandnote) = {
147 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
148 	.hdr.n_descsz	= sizeof(int32_t),
149 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
150 	.vendor		= FREEBSD_ABI_VENDOR,
151 	.flags		= BN_TRANSLATE_OSREL,
152 	.trans_osrel	= __elfN(freebsd_trans_osrel)
153 };
154 
155 static boolean_t
156 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
157 {
158 	uintptr_t p;
159 
160 	p = (uintptr_t)(note + 1);
161 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
162 	*osrel = *(const int32_t *)(p);
163 
164 	return (TRUE);
165 }
166 
167 static const char GNU_ABI_VENDOR[] = "GNU";
168 static int GNU_KFREEBSD_ABI_DESC = 3;
169 
170 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
171 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
172 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
173 	.hdr.n_type	= 1,
174 	.vendor		= GNU_ABI_VENDOR,
175 	.flags		= BN_TRANSLATE_OSREL,
176 	.trans_osrel	= kfreebsd_trans_osrel
177 };
178 
179 static boolean_t
180 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
181 {
182 	const Elf32_Word *desc;
183 	uintptr_t p;
184 
185 	p = (uintptr_t)(note + 1);
186 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
187 
188 	desc = (const Elf32_Word *)p;
189 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
190 		return (FALSE);
191 
192 	/*
193 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
194 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
195 	 */
196 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
197 
198 	return (TRUE);
199 }
200 
201 int
202 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
203 {
204 	int i;
205 
206 	for (i = 0; i < MAX_BRANDS; i++) {
207 		if (elf_brand_list[i] == NULL) {
208 			elf_brand_list[i] = entry;
209 			break;
210 		}
211 	}
212 	if (i == MAX_BRANDS) {
213 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
214 			__func__, entry);
215 		return (-1);
216 	}
217 	return (0);
218 }
219 
220 int
221 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
222 {
223 	int i;
224 
225 	for (i = 0; i < MAX_BRANDS; i++) {
226 		if (elf_brand_list[i] == entry) {
227 			elf_brand_list[i] = NULL;
228 			break;
229 		}
230 	}
231 	if (i == MAX_BRANDS)
232 		return (-1);
233 	return (0);
234 }
235 
236 int
237 __elfN(brand_inuse)(Elf_Brandinfo *entry)
238 {
239 	struct proc *p;
240 	int rval = FALSE;
241 
242 	sx_slock(&allproc_lock);
243 	FOREACH_PROC_IN_SYSTEM(p) {
244 		if (p->p_sysent == entry->sysvec) {
245 			rval = TRUE;
246 			break;
247 		}
248 	}
249 	sx_sunlock(&allproc_lock);
250 
251 	return (rval);
252 }
253 
254 static Elf_Brandinfo *
255 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
256     int interp_name_len, int32_t *osrel)
257 {
258 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
259 	Elf_Brandinfo *bi, *bi_m;
260 	boolean_t ret;
261 	int i;
262 
263 	/*
264 	 * We support four types of branding -- (1) the ELF EI_OSABI field
265 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
266 	 * branding w/in the ELF header, (3) path of the `interp_path'
267 	 * field, and (4) the ".note.ABI-tag" ELF section.
268 	 */
269 
270 	/* Look for an ".note.ABI-tag" ELF section */
271 	bi_m = NULL;
272 	for (i = 0; i < MAX_BRANDS; i++) {
273 		bi = elf_brand_list[i];
274 		if (bi == NULL)
275 			continue;
276 		if (hdr->e_machine == bi->machine && (bi->flags &
277 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
278 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
279 			/* Give brand a chance to veto check_note's guess */
280 			if (ret && bi->header_supported)
281 				ret = bi->header_supported(imgp);
282 			/*
283 			 * If note checker claimed the binary, but the
284 			 * interpreter path in the image does not
285 			 * match default one for the brand, try to
286 			 * search for other brands with the same
287 			 * interpreter.  Either there is better brand
288 			 * with the right interpreter, or, failing
289 			 * this, we return first brand which accepted
290 			 * our note and, optionally, header.
291 			 */
292 			if (ret && bi_m == NULL && (strlen(bi->interp_path) +
293 			    1 != interp_name_len || strncmp(interp,
294 			    bi->interp_path, interp_name_len) != 0)) {
295 				bi_m = bi;
296 				ret = 0;
297 			}
298 			if (ret)
299 				return (bi);
300 		}
301 	}
302 	if (bi_m != NULL)
303 		return (bi_m);
304 
305 	/* If the executable has a brand, search for it in the brand list. */
306 	for (i = 0; i < MAX_BRANDS; i++) {
307 		bi = elf_brand_list[i];
308 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
309 			continue;
310 		if (hdr->e_machine == bi->machine &&
311 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
312 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
313 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
314 			/* Looks good, but give brand a chance to veto */
315 			if (!bi->header_supported || bi->header_supported(imgp))
316 				return (bi);
317 		}
318 	}
319 
320 	/* No known brand, see if the header is recognized by any brand */
321 	for (i = 0; i < MAX_BRANDS; i++) {
322 		bi = elf_brand_list[i];
323 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
324 		    bi->header_supported == NULL)
325 			continue;
326 		if (hdr->e_machine == bi->machine) {
327 			ret = bi->header_supported(imgp);
328 			if (ret)
329 				return (bi);
330 		}
331 	}
332 
333 	/* Lacking a known brand, search for a recognized interpreter. */
334 	if (interp != NULL) {
335 		for (i = 0; i < MAX_BRANDS; i++) {
336 			bi = elf_brand_list[i];
337 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
338 				continue;
339 			if (hdr->e_machine == bi->machine &&
340 			    /* ELF image p_filesz includes terminating zero */
341 			    strlen(bi->interp_path) + 1 == interp_name_len &&
342 			    strncmp(interp, bi->interp_path, interp_name_len)
343 			    == 0)
344 				return (bi);
345 		}
346 	}
347 
348 	/* Lacking a recognized interpreter, try the default brand */
349 	for (i = 0; i < MAX_BRANDS; i++) {
350 		bi = elf_brand_list[i];
351 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
352 			continue;
353 		if (hdr->e_machine == bi->machine &&
354 		    __elfN(fallback_brand) == bi->brand)
355 			return (bi);
356 	}
357 	return (NULL);
358 }
359 
360 static int
361 __elfN(check_header)(const Elf_Ehdr *hdr)
362 {
363 	Elf_Brandinfo *bi;
364 	int i;
365 
366 	if (!IS_ELF(*hdr) ||
367 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
368 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
369 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
370 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
371 	    hdr->e_version != ELF_TARG_VER)
372 		return (ENOEXEC);
373 
374 	/*
375 	 * Make sure we have at least one brand for this machine.
376 	 */
377 
378 	for (i = 0; i < MAX_BRANDS; i++) {
379 		bi = elf_brand_list[i];
380 		if (bi != NULL && bi->machine == hdr->e_machine)
381 			break;
382 	}
383 	if (i == MAX_BRANDS)
384 		return (ENOEXEC);
385 
386 	return (0);
387 }
388 
389 static int
390 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
391     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
392 {
393 	struct sf_buf *sf;
394 	int error;
395 	vm_offset_t off;
396 
397 	/*
398 	 * Create the page if it doesn't exist yet. Ignore errors.
399 	 */
400 	vm_map_lock(map);
401 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
402 	    VM_PROT_ALL, VM_PROT_ALL, 0);
403 	vm_map_unlock(map);
404 
405 	/*
406 	 * Find the page from the underlying object.
407 	 */
408 	if (object) {
409 		sf = vm_imgact_map_page(object, offset);
410 		if (sf == NULL)
411 			return (KERN_FAILURE);
412 		off = offset - trunc_page(offset);
413 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
414 		    end - start);
415 		vm_imgact_unmap_page(sf);
416 		if (error) {
417 			return (KERN_FAILURE);
418 		}
419 	}
420 
421 	return (KERN_SUCCESS);
422 }
423 
424 static int
425 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
426     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
427 {
428 	struct sf_buf *sf;
429 	vm_offset_t off;
430 	vm_size_t sz;
431 	int error, rv;
432 
433 	if (start != trunc_page(start)) {
434 		rv = __elfN(map_partial)(map, object, offset, start,
435 		    round_page(start), prot);
436 		if (rv)
437 			return (rv);
438 		offset += round_page(start) - start;
439 		start = round_page(start);
440 	}
441 	if (end != round_page(end)) {
442 		rv = __elfN(map_partial)(map, object, offset +
443 		    trunc_page(end) - start, trunc_page(end), end, prot);
444 		if (rv)
445 			return (rv);
446 		end = trunc_page(end);
447 	}
448 	if (end > start) {
449 		if (offset & PAGE_MASK) {
450 			/*
451 			 * The mapping is not page aligned. This means we have
452 			 * to copy the data. Sigh.
453 			 */
454 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
455 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
456 			    0);
457 			if (rv)
458 				return (rv);
459 			if (object == NULL)
460 				return (KERN_SUCCESS);
461 			for (; start < end; start += sz) {
462 				sf = vm_imgact_map_page(object, offset);
463 				if (sf == NULL)
464 					return (KERN_FAILURE);
465 				off = offset - trunc_page(offset);
466 				sz = end - start;
467 				if (sz > PAGE_SIZE - off)
468 					sz = PAGE_SIZE - off;
469 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
470 				    (caddr_t)start, sz);
471 				vm_imgact_unmap_page(sf);
472 				if (error) {
473 					return (KERN_FAILURE);
474 				}
475 				offset += sz;
476 			}
477 			rv = KERN_SUCCESS;
478 		} else {
479 			vm_object_reference(object);
480 			vm_map_lock(map);
481 			rv = vm_map_insert(map, object, offset, start, end,
482 			    prot, VM_PROT_ALL, cow);
483 			vm_map_unlock(map);
484 			if (rv != KERN_SUCCESS)
485 				vm_object_deallocate(object);
486 		}
487 		return (rv);
488 	} else {
489 		return (KERN_SUCCESS);
490 	}
491 }
492 
493 static int
494 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
495     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
496     size_t pagesize)
497 {
498 	struct sf_buf *sf;
499 	size_t map_len;
500 	vm_map_t map;
501 	vm_object_t object;
502 	vm_offset_t map_addr;
503 	int error, rv, cow;
504 	size_t copy_len;
505 	vm_offset_t file_addr;
506 
507 	/*
508 	 * It's necessary to fail if the filsz + offset taken from the
509 	 * header is greater than the actual file pager object's size.
510 	 * If we were to allow this, then the vm_map_find() below would
511 	 * walk right off the end of the file object and into the ether.
512 	 *
513 	 * While I'm here, might as well check for something else that
514 	 * is invalid: filsz cannot be greater than memsz.
515 	 */
516 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
517 		uprintf("elf_load_section: truncated ELF file\n");
518 		return (ENOEXEC);
519 	}
520 
521 	object = imgp->object;
522 	map = &imgp->proc->p_vmspace->vm_map;
523 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
524 	file_addr = trunc_page_ps(offset, pagesize);
525 
526 	/*
527 	 * We have two choices.  We can either clear the data in the last page
528 	 * of an oversized mapping, or we can start the anon mapping a page
529 	 * early and copy the initialized data into that first page.  We
530 	 * choose the second..
531 	 */
532 	if (memsz > filsz)
533 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
534 	else
535 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
536 
537 	if (map_len != 0) {
538 		/* cow flags: don't dump readonly sections in core */
539 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
540 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
541 
542 		rv = __elfN(map_insert)(map,
543 				      object,
544 				      file_addr,	/* file offset */
545 				      map_addr,		/* virtual start */
546 				      map_addr + map_len,/* virtual end */
547 				      prot,
548 				      cow);
549 		if (rv != KERN_SUCCESS)
550 			return (EINVAL);
551 
552 		/* we can stop now if we've covered it all */
553 		if (memsz == filsz) {
554 			return (0);
555 		}
556 	}
557 
558 
559 	/*
560 	 * We have to get the remaining bit of the file into the first part
561 	 * of the oversized map segment.  This is normally because the .data
562 	 * segment in the file is extended to provide bss.  It's a neat idea
563 	 * to try and save a page, but it's a pain in the behind to implement.
564 	 */
565 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
566 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
567 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
568 	    map_addr;
569 
570 	/* This had damn well better be true! */
571 	if (map_len != 0) {
572 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
573 		    map_len, VM_PROT_ALL, 0);
574 		if (rv != KERN_SUCCESS) {
575 			return (EINVAL);
576 		}
577 	}
578 
579 	if (copy_len != 0) {
580 		vm_offset_t off;
581 
582 		sf = vm_imgact_map_page(object, offset + filsz);
583 		if (sf == NULL)
584 			return (EIO);
585 
586 		/* send the page fragment to user space */
587 		off = trunc_page_ps(offset + filsz, pagesize) -
588 		    trunc_page(offset + filsz);
589 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
590 		    (caddr_t)map_addr, copy_len);
591 		vm_imgact_unmap_page(sf);
592 		if (error) {
593 			return (error);
594 		}
595 	}
596 
597 	/*
598 	 * set it to the specified protection.
599 	 * XXX had better undo the damage from pasting over the cracks here!
600 	 */
601 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
602 	    map_len), prot, FALSE);
603 
604 	return (0);
605 }
606 
607 /*
608  * Load the file "file" into memory.  It may be either a shared object
609  * or an executable.
610  *
611  * The "addr" reference parameter is in/out.  On entry, it specifies
612  * the address where a shared object should be loaded.  If the file is
613  * an executable, this value is ignored.  On exit, "addr" specifies
614  * where the file was actually loaded.
615  *
616  * The "entry" reference parameter is out only.  On exit, it specifies
617  * the entry point for the loaded file.
618  */
619 static int
620 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
621 	u_long *entry, size_t pagesize)
622 {
623 	struct {
624 		struct nameidata nd;
625 		struct vattr attr;
626 		struct image_params image_params;
627 	} *tempdata;
628 	const Elf_Ehdr *hdr = NULL;
629 	const Elf_Phdr *phdr = NULL;
630 	struct nameidata *nd;
631 	struct vattr *attr;
632 	struct image_params *imgp;
633 	vm_prot_t prot;
634 	u_long rbase;
635 	u_long base_addr = 0;
636 	int error, i, numsegs;
637 
638 #ifdef CAPABILITY_MODE
639 	/*
640 	 * XXXJA: This check can go away once we are sufficiently confident
641 	 * that the checks in namei() are correct.
642 	 */
643 	if (IN_CAPABILITY_MODE(curthread))
644 		return (ECAPMODE);
645 #endif
646 
647 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
648 	nd = &tempdata->nd;
649 	attr = &tempdata->attr;
650 	imgp = &tempdata->image_params;
651 
652 	/*
653 	 * Initialize part of the common data
654 	 */
655 	imgp->proc = p;
656 	imgp->attr = attr;
657 	imgp->firstpage = NULL;
658 	imgp->image_header = NULL;
659 	imgp->object = NULL;
660 	imgp->execlabel = NULL;
661 
662 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
663 	if ((error = namei(nd)) != 0) {
664 		nd->ni_vp = NULL;
665 		goto fail;
666 	}
667 	NDFREE(nd, NDF_ONLY_PNBUF);
668 	imgp->vp = nd->ni_vp;
669 
670 	/*
671 	 * Check permissions, modes, uid, etc on the file, and "open" it.
672 	 */
673 	error = exec_check_permissions(imgp);
674 	if (error)
675 		goto fail;
676 
677 	error = exec_map_first_page(imgp);
678 	if (error)
679 		goto fail;
680 
681 	/*
682 	 * Also make certain that the interpreter stays the same, so set
683 	 * its VV_TEXT flag, too.
684 	 */
685 	VOP_SET_TEXT(nd->ni_vp);
686 
687 	imgp->object = nd->ni_vp->v_object;
688 
689 	hdr = (const Elf_Ehdr *)imgp->image_header;
690 	if ((error = __elfN(check_header)(hdr)) != 0)
691 		goto fail;
692 	if (hdr->e_type == ET_DYN)
693 		rbase = *addr;
694 	else if (hdr->e_type == ET_EXEC)
695 		rbase = 0;
696 	else {
697 		error = ENOEXEC;
698 		goto fail;
699 	}
700 
701 	/* Only support headers that fit within first page for now      */
702 	if ((hdr->e_phoff > PAGE_SIZE) ||
703 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
704 		error = ENOEXEC;
705 		goto fail;
706 	}
707 
708 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
709 	if (!aligned(phdr, Elf_Addr)) {
710 		error = ENOEXEC;
711 		goto fail;
712 	}
713 
714 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
715 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
716 			/* Loadable segment */
717 			prot = __elfN(trans_prot)(phdr[i].p_flags);
718 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
719 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
720 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
721 			if (error != 0)
722 				goto fail;
723 			/*
724 			 * Establish the base address if this is the
725 			 * first segment.
726 			 */
727 			if (numsegs == 0)
728   				base_addr = trunc_page(phdr[i].p_vaddr +
729 				    rbase);
730 			numsegs++;
731 		}
732 	}
733 	*addr = base_addr;
734 	*entry = (unsigned long)hdr->e_entry + rbase;
735 
736 fail:
737 	if (imgp->firstpage)
738 		exec_unmap_first_page(imgp);
739 
740 	if (nd->ni_vp)
741 		vput(nd->ni_vp);
742 
743 	free(tempdata, M_TEMP);
744 
745 	return (error);
746 }
747 
748 static int
749 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
750 {
751 	struct thread *td;
752 	const Elf_Ehdr *hdr;
753 	const Elf_Phdr *phdr;
754 	Elf_Auxargs *elf_auxargs;
755 	struct vmspace *vmspace;
756 	const char *err_str, *newinterp;
757 	char *interp, *interp_buf, *path;
758 	Elf_Brandinfo *brand_info;
759 	struct sysentvec *sv;
760 	vm_prot_t prot;
761 	u_long text_size, data_size, total_size, text_addr, data_addr;
762 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
763 	int32_t osrel;
764 	int error, i, n, interp_name_len, have_interp;
765 
766 	hdr = (const Elf_Ehdr *)imgp->image_header;
767 
768 	/*
769 	 * Do we have a valid ELF header ?
770 	 *
771 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
772 	 * if particular brand doesn't support it.
773 	 */
774 	if (__elfN(check_header)(hdr) != 0 ||
775 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
776 		return (-1);
777 
778 	/*
779 	 * From here on down, we return an errno, not -1, as we've
780 	 * detected an ELF file.
781 	 */
782 
783 	if ((hdr->e_phoff > PAGE_SIZE) ||
784 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
785 		/* Only support headers in first page for now */
786 		uprintf("Program headers not in the first page\n");
787 		return (ENOEXEC);
788 	}
789 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
790 	if (!aligned(phdr, Elf_Addr)) {
791 		uprintf("Unaligned program headers\n");
792 		return (ENOEXEC);
793 	}
794 
795 	n = error = 0;
796 	baddr = 0;
797 	osrel = 0;
798 	text_size = data_size = total_size = text_addr = data_addr = 0;
799 	entry = proghdr = 0;
800 	interp_name_len = 0;
801 	err_str = newinterp = NULL;
802 	interp = interp_buf = NULL;
803 	td = curthread;
804 
805 	for (i = 0; i < hdr->e_phnum; i++) {
806 		switch (phdr[i].p_type) {
807 		case PT_LOAD:
808 			if (n == 0)
809 				baddr = phdr[i].p_vaddr;
810 			n++;
811 			break;
812 		case PT_INTERP:
813 			/* Path to interpreter */
814 			if (phdr[i].p_filesz > MAXPATHLEN) {
815 				uprintf("Invalid PT_INTERP\n");
816 				error = ENOEXEC;
817 				goto ret;
818 			}
819 			if (interp != NULL) {
820 				uprintf("Multiple PT_INTERP headers\n");
821 				error = ENOEXEC;
822 				goto ret;
823 			}
824 			interp_name_len = phdr[i].p_filesz;
825 			if (phdr[i].p_offset > PAGE_SIZE ||
826 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
827 				VOP_UNLOCK(imgp->vp, 0);
828 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
829 				    M_WAITOK);
830 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
831 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
832 				    interp_name_len, phdr[i].p_offset,
833 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
834 				    NOCRED, NULL, td);
835 				if (error != 0) {
836 					uprintf("i/o error PT_INTERP\n");
837 					goto ret;
838 				}
839 				interp_buf[interp_name_len] = '\0';
840 				interp = interp_buf;
841 			} else {
842 				interp = __DECONST(char *, imgp->image_header) +
843 				    phdr[i].p_offset;
844 			}
845 			break;
846 		case PT_GNU_STACK:
847 			if (__elfN(nxstack))
848 				imgp->stack_prot =
849 				    __elfN(trans_prot)(phdr[i].p_flags);
850 			imgp->stack_sz = phdr[i].p_memsz;
851 			break;
852 		}
853 	}
854 
855 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
856 	    &osrel);
857 	if (brand_info == NULL) {
858 		uprintf("ELF binary type \"%u\" not known.\n",
859 		    hdr->e_ident[EI_OSABI]);
860 		error = ENOEXEC;
861 		goto ret;
862 	}
863 	if (hdr->e_type == ET_DYN) {
864 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
865 			uprintf("Cannot execute shared object\n");
866 			error = ENOEXEC;
867 			goto ret;
868 		}
869 		/*
870 		 * Honour the base load address from the dso if it is
871 		 * non-zero for some reason.
872 		 */
873 		if (baddr == 0)
874 			et_dyn_addr = ET_DYN_LOAD_ADDR;
875 		else
876 			et_dyn_addr = 0;
877 	} else
878 		et_dyn_addr = 0;
879 	sv = brand_info->sysvec;
880 	if (interp != NULL && brand_info->interp_newpath != NULL)
881 		newinterp = brand_info->interp_newpath;
882 
883 	/*
884 	 * Avoid a possible deadlock if the current address space is destroyed
885 	 * and that address space maps the locked vnode.  In the common case,
886 	 * the locked vnode's v_usecount is decremented but remains greater
887 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
888 	 * However, in cases where the vnode lock is external, such as nullfs,
889 	 * v_usecount may become zero.
890 	 *
891 	 * The VV_TEXT flag prevents modifications to the executable while
892 	 * the vnode is unlocked.
893 	 */
894 	VOP_UNLOCK(imgp->vp, 0);
895 
896 	error = exec_new_vmspace(imgp, sv);
897 	imgp->proc->p_sysent = sv;
898 
899 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
900 	if (error != 0)
901 		goto ret;
902 
903 	for (i = 0; i < hdr->e_phnum; i++) {
904 		switch (phdr[i].p_type) {
905 		case PT_LOAD:	/* Loadable segment */
906 			if (phdr[i].p_memsz == 0)
907 				break;
908 			prot = __elfN(trans_prot)(phdr[i].p_flags);
909 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
910 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
911 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
912 			    sv->sv_pagesize);
913 			if (error != 0)
914 				goto ret;
915 
916 			/*
917 			 * If this segment contains the program headers,
918 			 * remember their virtual address for the AT_PHDR
919 			 * aux entry. Static binaries don't usually include
920 			 * a PT_PHDR entry.
921 			 */
922 			if (phdr[i].p_offset == 0 &&
923 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
924 				<= phdr[i].p_filesz)
925 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
926 				    et_dyn_addr;
927 
928 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
929 			seg_size = round_page(phdr[i].p_memsz +
930 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
931 
932 			/*
933 			 * Make the largest executable segment the official
934 			 * text segment and all others data.
935 			 *
936 			 * Note that obreak() assumes that data_addr +
937 			 * data_size == end of data load area, and the ELF
938 			 * file format expects segments to be sorted by
939 			 * address.  If multiple data segments exist, the
940 			 * last one will be used.
941 			 */
942 
943 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
944 				text_size = seg_size;
945 				text_addr = seg_addr;
946 			} else {
947 				data_size = seg_size;
948 				data_addr = seg_addr;
949 			}
950 			total_size += seg_size;
951 			break;
952 		case PT_PHDR: 	/* Program header table info */
953 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
954 			break;
955 		default:
956 			break;
957 		}
958 	}
959 
960 	if (data_addr == 0 && data_size == 0) {
961 		data_addr = text_addr;
962 		data_size = text_size;
963 	}
964 
965 	entry = (u_long)hdr->e_entry + et_dyn_addr;
966 
967 	/*
968 	 * Check limits.  It should be safe to check the
969 	 * limits after loading the segments since we do
970 	 * not actually fault in all the segments pages.
971 	 */
972 	PROC_LOCK(imgp->proc);
973 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
974 		err_str = "Data segment size exceeds process limit";
975 	else if (text_size > maxtsiz)
976 		err_str = "Text segment size exceeds system limit";
977 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
978 		err_str = "Total segment size exceeds process limit";
979 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
980 		err_str = "Data segment size exceeds resource limit";
981 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
982 		err_str = "Total segment size exceeds resource limit";
983 	if (err_str != NULL) {
984 		PROC_UNLOCK(imgp->proc);
985 		uprintf("%s\n", err_str);
986 		error = ENOMEM;
987 		goto ret;
988 	}
989 
990 	vmspace = imgp->proc->p_vmspace;
991 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
992 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
993 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
994 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
995 
996 	/*
997 	 * We load the dynamic linker where a userland call
998 	 * to mmap(0, ...) would put it.  The rationale behind this
999 	 * calculation is that it leaves room for the heap to grow to
1000 	 * its maximum allowed size.
1001 	 */
1002 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1003 	    RLIMIT_DATA));
1004 	PROC_UNLOCK(imgp->proc);
1005 
1006 	imgp->entry_addr = entry;
1007 
1008 	if (interp != NULL) {
1009 		have_interp = FALSE;
1010 		VOP_UNLOCK(imgp->vp, 0);
1011 		if (brand_info->emul_path != NULL &&
1012 		    brand_info->emul_path[0] != '\0') {
1013 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1014 			snprintf(path, MAXPATHLEN, "%s%s",
1015 			    brand_info->emul_path, interp);
1016 			error = __elfN(load_file)(imgp->proc, path, &addr,
1017 			    &imgp->entry_addr, sv->sv_pagesize);
1018 			free(path, M_TEMP);
1019 			if (error == 0)
1020 				have_interp = TRUE;
1021 		}
1022 		if (!have_interp && newinterp != NULL &&
1023 		    (brand_info->interp_path == NULL ||
1024 		    strcmp(interp, brand_info->interp_path) == 0)) {
1025 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1026 			    &imgp->entry_addr, sv->sv_pagesize);
1027 			if (error == 0)
1028 				have_interp = TRUE;
1029 		}
1030 		if (!have_interp) {
1031 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1032 			    &imgp->entry_addr, sv->sv_pagesize);
1033 		}
1034 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1035 		if (error != 0) {
1036 			uprintf("ELF interpreter %s not found, error %d\n",
1037 			    interp, error);
1038 			goto ret;
1039 		}
1040 	} else
1041 		addr = et_dyn_addr;
1042 
1043 	/*
1044 	 * Construct auxargs table (used by the fixup routine)
1045 	 */
1046 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1047 	elf_auxargs->execfd = -1;
1048 	elf_auxargs->phdr = proghdr;
1049 	elf_auxargs->phent = hdr->e_phentsize;
1050 	elf_auxargs->phnum = hdr->e_phnum;
1051 	elf_auxargs->pagesz = PAGE_SIZE;
1052 	elf_auxargs->base = addr;
1053 	elf_auxargs->flags = 0;
1054 	elf_auxargs->entry = entry;
1055 	elf_auxargs->hdr_eflags = hdr->e_flags;
1056 
1057 	imgp->auxargs = elf_auxargs;
1058 	imgp->interpreted = 0;
1059 	imgp->reloc_base = addr;
1060 	imgp->proc->p_osrel = osrel;
1061 
1062  ret:
1063 	free(interp_buf, M_TEMP);
1064 	return (error);
1065 }
1066 
1067 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1068 
1069 int
1070 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1071 {
1072 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1073 	Elf_Addr *base;
1074 	Elf_Addr *pos;
1075 
1076 	base = (Elf_Addr *)*stack_base;
1077 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1078 
1079 	if (args->execfd != -1)
1080 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1081 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1082 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1083 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1084 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1085 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1086 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1087 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1088 #ifdef AT_EHDRFLAGS
1089 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1090 #endif
1091 	if (imgp->execpathp != 0)
1092 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1093 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1094 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1095 	if (imgp->canary != 0) {
1096 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1097 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1098 	}
1099 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1100 	if (imgp->pagesizes != 0) {
1101 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1102 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1103 	}
1104 	if (imgp->sysent->sv_timekeep_base != 0) {
1105 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1106 		    imgp->sysent->sv_timekeep_base);
1107 	}
1108 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1109 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1110 	    imgp->sysent->sv_stackprot);
1111 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1112 
1113 	free(imgp->auxargs, M_TEMP);
1114 	imgp->auxargs = NULL;
1115 
1116 	base--;
1117 	suword(base, (long)imgp->args->argc);
1118 	*stack_base = (register_t *)base;
1119 	return (0);
1120 }
1121 
1122 /*
1123  * Code for generating ELF core dumps.
1124  */
1125 
1126 typedef void (*segment_callback)(vm_map_entry_t, void *);
1127 
1128 /* Closure for cb_put_phdr(). */
1129 struct phdr_closure {
1130 	Elf_Phdr *phdr;		/* Program header to fill in */
1131 	Elf_Off offset;		/* Offset of segment in core file */
1132 };
1133 
1134 /* Closure for cb_size_segment(). */
1135 struct sseg_closure {
1136 	int count;		/* Count of writable segments. */
1137 	size_t size;		/* Total size of all writable segments. */
1138 };
1139 
1140 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1141 
1142 struct note_info {
1143 	int		type;		/* Note type. */
1144 	outfunc_t 	outfunc; 	/* Output function. */
1145 	void		*outarg;	/* Argument for the output function. */
1146 	size_t		outsize;	/* Output size. */
1147 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1148 };
1149 
1150 TAILQ_HEAD(note_info_list, note_info);
1151 
1152 /* Coredump output parameters. */
1153 struct coredump_params {
1154 	off_t		offset;
1155 	struct ucred	*active_cred;
1156 	struct ucred	*file_cred;
1157 	struct thread	*td;
1158 	struct vnode	*vp;
1159 	struct gzio_stream *gzs;
1160 };
1161 
1162 static void cb_put_phdr(vm_map_entry_t, void *);
1163 static void cb_size_segment(vm_map_entry_t, void *);
1164 static int core_write(struct coredump_params *, void *, size_t, off_t,
1165     enum uio_seg);
1166 static void each_writable_segment(struct thread *, segment_callback, void *);
1167 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1168     struct note_info_list *, size_t);
1169 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1170     size_t *);
1171 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1172 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1173 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1174 static int sbuf_drain_core_output(void *, const char *, int);
1175 static int sbuf_drain_count(void *arg, const char *data, int len);
1176 
1177 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1178 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1179 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1180 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1181 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1182 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1183 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1184 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1185 static void note_procstat_files(void *, struct sbuf *, size_t *);
1186 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1187 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1188 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1189 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1190 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1191 
1192 #ifdef GZIO
1193 extern int compress_user_cores_gzlevel;
1194 
1195 /*
1196  * Write out a core segment to the compression stream.
1197  */
1198 static int
1199 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1200 {
1201 	u_int chunk_len;
1202 	int error;
1203 
1204 	while (len > 0) {
1205 		chunk_len = MIN(len, CORE_BUF_SIZE);
1206 		copyin(base, buf, chunk_len);
1207 		error = gzio_write(p->gzs, buf, chunk_len);
1208 		if (error != 0)
1209 			break;
1210 		base += chunk_len;
1211 		len -= chunk_len;
1212 	}
1213 	return (error);
1214 }
1215 
1216 static int
1217 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1218 {
1219 
1220 	return (core_write((struct coredump_params *)arg, base, len, offset,
1221 	    UIO_SYSSPACE));
1222 }
1223 #endif /* GZIO */
1224 
1225 static int
1226 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1227     enum uio_seg seg)
1228 {
1229 
1230 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1231 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1232 	    p->active_cred, p->file_cred, NULL, p->td));
1233 }
1234 
1235 static int
1236 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1237     void *tmpbuf)
1238 {
1239 
1240 #ifdef GZIO
1241 	if (p->gzs != NULL)
1242 		return (compress_chunk(p, base, tmpbuf, len));
1243 #endif
1244 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1245 }
1246 
1247 /*
1248  * Drain into a core file.
1249  */
1250 static int
1251 sbuf_drain_core_output(void *arg, const char *data, int len)
1252 {
1253 	struct coredump_params *p;
1254 	int error, locked;
1255 
1256 	p = (struct coredump_params *)arg;
1257 
1258 	/*
1259 	 * Some kern_proc out routines that print to this sbuf may
1260 	 * call us with the process lock held. Draining with the
1261 	 * non-sleepable lock held is unsafe. The lock is needed for
1262 	 * those routines when dumping a live process. In our case we
1263 	 * can safely release the lock before draining and acquire
1264 	 * again after.
1265 	 */
1266 	locked = PROC_LOCKED(p->td->td_proc);
1267 	if (locked)
1268 		PROC_UNLOCK(p->td->td_proc);
1269 #ifdef GZIO
1270 	if (p->gzs != NULL)
1271 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1272 	else
1273 #endif
1274 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1275 		    UIO_SYSSPACE);
1276 	if (locked)
1277 		PROC_LOCK(p->td->td_proc);
1278 	if (error != 0)
1279 		return (-error);
1280 	p->offset += len;
1281 	return (len);
1282 }
1283 
1284 /*
1285  * Drain into a counter.
1286  */
1287 static int
1288 sbuf_drain_count(void *arg, const char *data __unused, int len)
1289 {
1290 	size_t *sizep;
1291 
1292 	sizep = (size_t *)arg;
1293 	*sizep += len;
1294 	return (len);
1295 }
1296 
1297 int
1298 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1299 {
1300 	struct ucred *cred = td->td_ucred;
1301 	int error = 0;
1302 	struct sseg_closure seginfo;
1303 	struct note_info_list notelst;
1304 	struct coredump_params params;
1305 	struct note_info *ninfo;
1306 	void *hdr, *tmpbuf;
1307 	size_t hdrsize, notesz, coresize;
1308 #ifdef GZIO
1309 	boolean_t compress;
1310 
1311 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1312 #endif
1313 	hdr = NULL;
1314 	tmpbuf = NULL;
1315 	TAILQ_INIT(&notelst);
1316 
1317 	/* Size the program segments. */
1318 	seginfo.count = 0;
1319 	seginfo.size = 0;
1320 	each_writable_segment(td, cb_size_segment, &seginfo);
1321 
1322 	/*
1323 	 * Collect info about the core file header area.
1324 	 */
1325 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1326 	__elfN(prepare_notes)(td, &notelst, &notesz);
1327 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1328 
1329 	/* Set up core dump parameters. */
1330 	params.offset = 0;
1331 	params.active_cred = cred;
1332 	params.file_cred = NOCRED;
1333 	params.td = td;
1334 	params.vp = vp;
1335 	params.gzs = NULL;
1336 
1337 #ifdef RACCT
1338 	if (racct_enable) {
1339 		PROC_LOCK(td->td_proc);
1340 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1341 		PROC_UNLOCK(td->td_proc);
1342 		if (error != 0) {
1343 			error = EFAULT;
1344 			goto done;
1345 		}
1346 	}
1347 #endif
1348 	if (coresize >= limit) {
1349 		error = EFAULT;
1350 		goto done;
1351 	}
1352 
1353 #ifdef GZIO
1354 	/* Create a compression stream if necessary. */
1355 	if (compress) {
1356 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1357 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1358 		if (params.gzs == NULL) {
1359 			error = EFAULT;
1360 			goto done;
1361 		}
1362 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1363         }
1364 #endif
1365 
1366 	/*
1367 	 * Allocate memory for building the header, fill it up,
1368 	 * and write it out following the notes.
1369 	 */
1370 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1371 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1372 	    notesz);
1373 
1374 	/* Write the contents of all of the writable segments. */
1375 	if (error == 0) {
1376 		Elf_Phdr *php;
1377 		off_t offset;
1378 		int i;
1379 
1380 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1381 		offset = round_page(hdrsize + notesz);
1382 		for (i = 0; i < seginfo.count; i++) {
1383 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1384 			    php->p_filesz, offset, &params, tmpbuf);
1385 			if (error != 0)
1386 				break;
1387 			offset += php->p_filesz;
1388 			php++;
1389 		}
1390 #ifdef GZIO
1391 		if (error == 0 && compress)
1392 			error = gzio_flush(params.gzs);
1393 #endif
1394 	}
1395 	if (error) {
1396 		log(LOG_WARNING,
1397 		    "Failed to write core file for process %s (error %d)\n",
1398 		    curproc->p_comm, error);
1399 	}
1400 
1401 done:
1402 #ifdef GZIO
1403 	if (compress) {
1404 		free(tmpbuf, M_TEMP);
1405 		if (params.gzs != NULL)
1406 			gzio_fini(params.gzs);
1407 	}
1408 #endif
1409 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1410 		TAILQ_REMOVE(&notelst, ninfo, link);
1411 		free(ninfo, M_TEMP);
1412 	}
1413 	if (hdr != NULL)
1414 		free(hdr, M_TEMP);
1415 
1416 	return (error);
1417 }
1418 
1419 /*
1420  * A callback for each_writable_segment() to write out the segment's
1421  * program header entry.
1422  */
1423 static void
1424 cb_put_phdr(entry, closure)
1425 	vm_map_entry_t entry;
1426 	void *closure;
1427 {
1428 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1429 	Elf_Phdr *phdr = phc->phdr;
1430 
1431 	phc->offset = round_page(phc->offset);
1432 
1433 	phdr->p_type = PT_LOAD;
1434 	phdr->p_offset = phc->offset;
1435 	phdr->p_vaddr = entry->start;
1436 	phdr->p_paddr = 0;
1437 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1438 	phdr->p_align = PAGE_SIZE;
1439 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1440 
1441 	phc->offset += phdr->p_filesz;
1442 	phc->phdr++;
1443 }
1444 
1445 /*
1446  * A callback for each_writable_segment() to gather information about
1447  * the number of segments and their total size.
1448  */
1449 static void
1450 cb_size_segment(entry, closure)
1451 	vm_map_entry_t entry;
1452 	void *closure;
1453 {
1454 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1455 
1456 	ssc->count++;
1457 	ssc->size += entry->end - entry->start;
1458 }
1459 
1460 /*
1461  * For each writable segment in the process's memory map, call the given
1462  * function with a pointer to the map entry and some arbitrary
1463  * caller-supplied data.
1464  */
1465 static void
1466 each_writable_segment(td, func, closure)
1467 	struct thread *td;
1468 	segment_callback func;
1469 	void *closure;
1470 {
1471 	struct proc *p = td->td_proc;
1472 	vm_map_t map = &p->p_vmspace->vm_map;
1473 	vm_map_entry_t entry;
1474 	vm_object_t backing_object, object;
1475 	boolean_t ignore_entry;
1476 
1477 	vm_map_lock_read(map);
1478 	for (entry = map->header.next; entry != &map->header;
1479 	    entry = entry->next) {
1480 		/*
1481 		 * Don't dump inaccessible mappings, deal with legacy
1482 		 * coredump mode.
1483 		 *
1484 		 * Note that read-only segments related to the elf binary
1485 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1486 		 * need to arbitrarily ignore such segments.
1487 		 */
1488 		if (elf_legacy_coredump) {
1489 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1490 				continue;
1491 		} else {
1492 			if ((entry->protection & VM_PROT_ALL) == 0)
1493 				continue;
1494 		}
1495 
1496 		/*
1497 		 * Dont include memory segment in the coredump if
1498 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1499 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1500 		 * kernel map).
1501 		 */
1502 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1503 			continue;
1504 
1505 		if ((object = entry->object.vm_object) == NULL)
1506 			continue;
1507 
1508 		/* Ignore memory-mapped devices and such things. */
1509 		VM_OBJECT_RLOCK(object);
1510 		while ((backing_object = object->backing_object) != NULL) {
1511 			VM_OBJECT_RLOCK(backing_object);
1512 			VM_OBJECT_RUNLOCK(object);
1513 			object = backing_object;
1514 		}
1515 		ignore_entry = object->type != OBJT_DEFAULT &&
1516 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1517 		    object->type != OBJT_PHYS;
1518 		VM_OBJECT_RUNLOCK(object);
1519 		if (ignore_entry)
1520 			continue;
1521 
1522 		(*func)(entry, closure);
1523 	}
1524 	vm_map_unlock_read(map);
1525 }
1526 
1527 /*
1528  * Write the core file header to the file, including padding up to
1529  * the page boundary.
1530  */
1531 static int
1532 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1533     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1534 {
1535 	struct note_info *ninfo;
1536 	struct sbuf *sb;
1537 	int error;
1538 
1539 	/* Fill in the header. */
1540 	bzero(hdr, hdrsize);
1541 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1542 
1543 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1544 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1545 	sbuf_start_section(sb, NULL);
1546 	sbuf_bcat(sb, hdr, hdrsize);
1547 	TAILQ_FOREACH(ninfo, notelst, link)
1548 	    __elfN(putnote)(ninfo, sb);
1549 	/* Align up to a page boundary for the program segments. */
1550 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1551 	error = sbuf_finish(sb);
1552 	sbuf_delete(sb);
1553 
1554 	return (error);
1555 }
1556 
1557 static void
1558 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1559     size_t *sizep)
1560 {
1561 	struct proc *p;
1562 	struct thread *thr;
1563 	size_t size;
1564 
1565 	p = td->td_proc;
1566 	size = 0;
1567 
1568 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1569 
1570 	/*
1571 	 * To have the debugger select the right thread (LWP) as the initial
1572 	 * thread, we dump the state of the thread passed to us in td first.
1573 	 * This is the thread that causes the core dump and thus likely to
1574 	 * be the right thread one wants to have selected in the debugger.
1575 	 */
1576 	thr = td;
1577 	while (thr != NULL) {
1578 		size += register_note(list, NT_PRSTATUS,
1579 		    __elfN(note_prstatus), thr);
1580 		size += register_note(list, NT_FPREGSET,
1581 		    __elfN(note_fpregset), thr);
1582 		size += register_note(list, NT_THRMISC,
1583 		    __elfN(note_thrmisc), thr);
1584 		size += register_note(list, -1,
1585 		    __elfN(note_threadmd), thr);
1586 
1587 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1588 		    TAILQ_NEXT(thr, td_plist);
1589 		if (thr == td)
1590 			thr = TAILQ_NEXT(thr, td_plist);
1591 	}
1592 
1593 	size += register_note(list, NT_PROCSTAT_PROC,
1594 	    __elfN(note_procstat_proc), p);
1595 	size += register_note(list, NT_PROCSTAT_FILES,
1596 	    note_procstat_files, p);
1597 	size += register_note(list, NT_PROCSTAT_VMMAP,
1598 	    note_procstat_vmmap, p);
1599 	size += register_note(list, NT_PROCSTAT_GROUPS,
1600 	    note_procstat_groups, p);
1601 	size += register_note(list, NT_PROCSTAT_UMASK,
1602 	    note_procstat_umask, p);
1603 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1604 	    note_procstat_rlimit, p);
1605 	size += register_note(list, NT_PROCSTAT_OSREL,
1606 	    note_procstat_osrel, p);
1607 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1608 	    __elfN(note_procstat_psstrings), p);
1609 	size += register_note(list, NT_PROCSTAT_AUXV,
1610 	    __elfN(note_procstat_auxv), p);
1611 
1612 	*sizep = size;
1613 }
1614 
1615 static void
1616 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1617     size_t notesz)
1618 {
1619 	Elf_Ehdr *ehdr;
1620 	Elf_Phdr *phdr;
1621 	struct phdr_closure phc;
1622 
1623 	ehdr = (Elf_Ehdr *)hdr;
1624 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1625 
1626 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1627 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1628 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1629 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1630 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1631 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1632 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1633 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1634 	ehdr->e_ident[EI_ABIVERSION] = 0;
1635 	ehdr->e_ident[EI_PAD] = 0;
1636 	ehdr->e_type = ET_CORE;
1637 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1638 	ehdr->e_machine = ELF_ARCH32;
1639 #else
1640 	ehdr->e_machine = ELF_ARCH;
1641 #endif
1642 	ehdr->e_version = EV_CURRENT;
1643 	ehdr->e_entry = 0;
1644 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1645 	ehdr->e_flags = 0;
1646 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1647 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1648 	ehdr->e_phnum = numsegs + 1;
1649 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1650 	ehdr->e_shnum = 0;
1651 	ehdr->e_shstrndx = SHN_UNDEF;
1652 
1653 	/*
1654 	 * Fill in the program header entries.
1655 	 */
1656 
1657 	/* The note segement. */
1658 	phdr->p_type = PT_NOTE;
1659 	phdr->p_offset = hdrsize;
1660 	phdr->p_vaddr = 0;
1661 	phdr->p_paddr = 0;
1662 	phdr->p_filesz = notesz;
1663 	phdr->p_memsz = 0;
1664 	phdr->p_flags = PF_R;
1665 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1666 	phdr++;
1667 
1668 	/* All the writable segments from the program. */
1669 	phc.phdr = phdr;
1670 	phc.offset = round_page(hdrsize + notesz);
1671 	each_writable_segment(td, cb_put_phdr, &phc);
1672 }
1673 
1674 static size_t
1675 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1676 {
1677 	struct note_info *ninfo;
1678 	size_t size, notesize;
1679 
1680 	size = 0;
1681 	out(arg, NULL, &size);
1682 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1683 	ninfo->type = type;
1684 	ninfo->outfunc = out;
1685 	ninfo->outarg = arg;
1686 	ninfo->outsize = size;
1687 	TAILQ_INSERT_TAIL(list, ninfo, link);
1688 
1689 	if (type == -1)
1690 		return (size);
1691 
1692 	notesize = sizeof(Elf_Note) +		/* note header */
1693 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1694 						/* note name */
1695 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1696 
1697 	return (notesize);
1698 }
1699 
1700 static size_t
1701 append_note_data(const void *src, void *dst, size_t len)
1702 {
1703 	size_t padded_len;
1704 
1705 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1706 	if (dst != NULL) {
1707 		bcopy(src, dst, len);
1708 		bzero((char *)dst + len, padded_len - len);
1709 	}
1710 	return (padded_len);
1711 }
1712 
1713 size_t
1714 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1715 {
1716 	Elf_Note *note;
1717 	char *buf;
1718 	size_t notesize;
1719 
1720 	buf = dst;
1721 	if (buf != NULL) {
1722 		note = (Elf_Note *)buf;
1723 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1724 		note->n_descsz = size;
1725 		note->n_type = type;
1726 		buf += sizeof(*note);
1727 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1728 		    sizeof(FREEBSD_ABI_VENDOR));
1729 		append_note_data(src, buf, size);
1730 		if (descp != NULL)
1731 			*descp = buf;
1732 	}
1733 
1734 	notesize = sizeof(Elf_Note) +		/* note header */
1735 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1736 						/* note name */
1737 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1738 
1739 	return (notesize);
1740 }
1741 
1742 static void
1743 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1744 {
1745 	Elf_Note note;
1746 	ssize_t old_len, sect_len;
1747 	size_t new_len, descsz, i;
1748 
1749 	if (ninfo->type == -1) {
1750 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1751 		return;
1752 	}
1753 
1754 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1755 	note.n_descsz = ninfo->outsize;
1756 	note.n_type = ninfo->type;
1757 
1758 	sbuf_bcat(sb, &note, sizeof(note));
1759 	sbuf_start_section(sb, &old_len);
1760 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1761 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1762 	if (note.n_descsz == 0)
1763 		return;
1764 	sbuf_start_section(sb, &old_len);
1765 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1766 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1767 	if (sect_len < 0)
1768 		return;
1769 
1770 	new_len = (size_t)sect_len;
1771 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1772 	if (new_len < descsz) {
1773 		/*
1774 		 * It is expected that individual note emitters will correctly
1775 		 * predict their expected output size and fill up to that size
1776 		 * themselves, padding in a format-specific way if needed.
1777 		 * However, in case they don't, just do it here with zeros.
1778 		 */
1779 		for (i = 0; i < descsz - new_len; i++)
1780 			sbuf_putc(sb, 0);
1781 	} else if (new_len > descsz) {
1782 		/*
1783 		 * We can't always truncate sb -- we may have drained some
1784 		 * of it already.
1785 		 */
1786 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1787 		    "read it (%zu > %zu).  Since it is longer than "
1788 		    "expected, this coredump's notes are corrupt.  THIS "
1789 		    "IS A BUG in the note_procstat routine for type %u.\n",
1790 		    __func__, (unsigned)note.n_type, new_len, descsz,
1791 		    (unsigned)note.n_type));
1792 	}
1793 }
1794 
1795 /*
1796  * Miscellaneous note out functions.
1797  */
1798 
1799 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1800 #include <compat/freebsd32/freebsd32.h>
1801 
1802 typedef struct prstatus32 elf_prstatus_t;
1803 typedef struct prpsinfo32 elf_prpsinfo_t;
1804 typedef struct fpreg32 elf_prfpregset_t;
1805 typedef struct fpreg32 elf_fpregset_t;
1806 typedef struct reg32 elf_gregset_t;
1807 typedef struct thrmisc32 elf_thrmisc_t;
1808 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1809 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1810 typedef uint32_t elf_ps_strings_t;
1811 #else
1812 typedef prstatus_t elf_prstatus_t;
1813 typedef prpsinfo_t elf_prpsinfo_t;
1814 typedef prfpregset_t elf_prfpregset_t;
1815 typedef prfpregset_t elf_fpregset_t;
1816 typedef gregset_t elf_gregset_t;
1817 typedef thrmisc_t elf_thrmisc_t;
1818 #define ELF_KERN_PROC_MASK	0
1819 typedef struct kinfo_proc elf_kinfo_proc_t;
1820 typedef vm_offset_t elf_ps_strings_t;
1821 #endif
1822 
1823 static void
1824 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1825 {
1826 	struct sbuf sbarg;
1827 	size_t len;
1828 	char *cp, *end;
1829 	struct proc *p;
1830 	elf_prpsinfo_t *psinfo;
1831 	int error;
1832 
1833 	p = (struct proc *)arg;
1834 	if (sb != NULL) {
1835 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1836 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1837 		psinfo->pr_version = PRPSINFO_VERSION;
1838 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1839 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1840 		PROC_LOCK(p);
1841 		if (p->p_args != NULL) {
1842 			len = sizeof(psinfo->pr_psargs) - 1;
1843 			if (len > p->p_args->ar_length)
1844 				len = p->p_args->ar_length;
1845 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1846 			PROC_UNLOCK(p);
1847 			error = 0;
1848 		} else {
1849 			_PHOLD(p);
1850 			PROC_UNLOCK(p);
1851 			sbuf_new(&sbarg, psinfo->pr_psargs,
1852 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1853 			error = proc_getargv(curthread, p, &sbarg);
1854 			PRELE(p);
1855 			if (sbuf_finish(&sbarg) == 0)
1856 				len = sbuf_len(&sbarg) - 1;
1857 			else
1858 				len = sizeof(psinfo->pr_psargs) - 1;
1859 			sbuf_delete(&sbarg);
1860 		}
1861 		if (error || len == 0)
1862 			strlcpy(psinfo->pr_psargs, p->p_comm,
1863 			    sizeof(psinfo->pr_psargs));
1864 		else {
1865 			KASSERT(len < sizeof(psinfo->pr_psargs),
1866 			    ("len is too long: %zu vs %zu", len,
1867 			    sizeof(psinfo->pr_psargs)));
1868 			cp = psinfo->pr_psargs;
1869 			end = cp + len - 1;
1870 			for (;;) {
1871 				cp = memchr(cp, '\0', end - cp);
1872 				if (cp == NULL)
1873 					break;
1874 				*cp = ' ';
1875 			}
1876 		}
1877 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1878 		free(psinfo, M_TEMP);
1879 	}
1880 	*sizep = sizeof(*psinfo);
1881 }
1882 
1883 static void
1884 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1885 {
1886 	struct thread *td;
1887 	elf_prstatus_t *status;
1888 
1889 	td = (struct thread *)arg;
1890 	if (sb != NULL) {
1891 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1892 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1893 		status->pr_version = PRSTATUS_VERSION;
1894 		status->pr_statussz = sizeof(elf_prstatus_t);
1895 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1896 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1897 		status->pr_osreldate = osreldate;
1898 		status->pr_cursig = td->td_proc->p_sig;
1899 		status->pr_pid = td->td_tid;
1900 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1901 		fill_regs32(td, &status->pr_reg);
1902 #else
1903 		fill_regs(td, &status->pr_reg);
1904 #endif
1905 		sbuf_bcat(sb, status, sizeof(*status));
1906 		free(status, M_TEMP);
1907 	}
1908 	*sizep = sizeof(*status);
1909 }
1910 
1911 static void
1912 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1913 {
1914 	struct thread *td;
1915 	elf_prfpregset_t *fpregset;
1916 
1917 	td = (struct thread *)arg;
1918 	if (sb != NULL) {
1919 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1920 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1921 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1922 		fill_fpregs32(td, fpregset);
1923 #else
1924 		fill_fpregs(td, fpregset);
1925 #endif
1926 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1927 		free(fpregset, M_TEMP);
1928 	}
1929 	*sizep = sizeof(*fpregset);
1930 }
1931 
1932 static void
1933 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1934 {
1935 	struct thread *td;
1936 	elf_thrmisc_t thrmisc;
1937 
1938 	td = (struct thread *)arg;
1939 	if (sb != NULL) {
1940 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1941 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1942 		strcpy(thrmisc.pr_tname, td->td_name);
1943 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1944 	}
1945 	*sizep = sizeof(thrmisc);
1946 }
1947 
1948 /*
1949  * Allow for MD specific notes, as well as any MD
1950  * specific preparations for writing MI notes.
1951  */
1952 static void
1953 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1954 {
1955 	struct thread *td;
1956 	void *buf;
1957 	size_t size;
1958 
1959 	td = (struct thread *)arg;
1960 	size = *sizep;
1961 	if (size != 0 && sb != NULL)
1962 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1963 	else
1964 		buf = NULL;
1965 	size = 0;
1966 	__elfN(dump_thread)(td, buf, &size);
1967 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1968 	if (size != 0 && sb != NULL)
1969 		sbuf_bcat(sb, buf, size);
1970 	free(buf, M_TEMP);
1971 	*sizep = size;
1972 }
1973 
1974 #ifdef KINFO_PROC_SIZE
1975 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1976 #endif
1977 
1978 static void
1979 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1980 {
1981 	struct proc *p;
1982 	size_t size;
1983 	int structsize;
1984 
1985 	p = (struct proc *)arg;
1986 	size = sizeof(structsize) + p->p_numthreads *
1987 	    sizeof(elf_kinfo_proc_t);
1988 
1989 	if (sb != NULL) {
1990 		KASSERT(*sizep == size, ("invalid size"));
1991 		structsize = sizeof(elf_kinfo_proc_t);
1992 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1993 		sx_slock(&proctree_lock);
1994 		PROC_LOCK(p);
1995 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1996 		sx_sunlock(&proctree_lock);
1997 	}
1998 	*sizep = size;
1999 }
2000 
2001 #ifdef KINFO_FILE_SIZE
2002 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2003 #endif
2004 
2005 static void
2006 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2007 {
2008 	struct proc *p;
2009 	size_t size, sect_sz, i;
2010 	ssize_t start_len, sect_len;
2011 	int structsize, filedesc_flags;
2012 
2013 	if (coredump_pack_fileinfo)
2014 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2015 	else
2016 		filedesc_flags = 0;
2017 
2018 	p = (struct proc *)arg;
2019 	structsize = sizeof(struct kinfo_file);
2020 	if (sb == NULL) {
2021 		size = 0;
2022 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2023 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2024 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2025 		PROC_LOCK(p);
2026 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2027 		sbuf_finish(sb);
2028 		sbuf_delete(sb);
2029 		*sizep = size;
2030 	} else {
2031 		sbuf_start_section(sb, &start_len);
2032 
2033 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2034 		PROC_LOCK(p);
2035 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2036 		    filedesc_flags);
2037 
2038 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2039 		if (sect_len < 0)
2040 			return;
2041 		sect_sz = sect_len;
2042 
2043 		KASSERT(sect_sz <= *sizep,
2044 		    ("kern_proc_filedesc_out did not respect maxlen; "
2045 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2046 		     sect_sz - sizeof(structsize)));
2047 
2048 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2049 			sbuf_putc(sb, 0);
2050 	}
2051 }
2052 
2053 #ifdef KINFO_VMENTRY_SIZE
2054 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2055 #endif
2056 
2057 static void
2058 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2059 {
2060 	struct proc *p;
2061 	size_t size;
2062 	int structsize, vmmap_flags;
2063 
2064 	if (coredump_pack_vmmapinfo)
2065 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2066 	else
2067 		vmmap_flags = 0;
2068 
2069 	p = (struct proc *)arg;
2070 	structsize = sizeof(struct kinfo_vmentry);
2071 	if (sb == NULL) {
2072 		size = 0;
2073 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2074 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2075 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2076 		PROC_LOCK(p);
2077 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2078 		sbuf_finish(sb);
2079 		sbuf_delete(sb);
2080 		*sizep = size;
2081 	} else {
2082 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2083 		PROC_LOCK(p);
2084 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2085 		    vmmap_flags);
2086 	}
2087 }
2088 
2089 static void
2090 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2091 {
2092 	struct proc *p;
2093 	size_t size;
2094 	int structsize;
2095 
2096 	p = (struct proc *)arg;
2097 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2098 	if (sb != NULL) {
2099 		KASSERT(*sizep == size, ("invalid size"));
2100 		structsize = sizeof(gid_t);
2101 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2102 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2103 		    sizeof(gid_t));
2104 	}
2105 	*sizep = size;
2106 }
2107 
2108 static void
2109 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2110 {
2111 	struct proc *p;
2112 	size_t size;
2113 	int structsize;
2114 
2115 	p = (struct proc *)arg;
2116 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2117 	if (sb != NULL) {
2118 		KASSERT(*sizep == size, ("invalid size"));
2119 		structsize = sizeof(p->p_fd->fd_cmask);
2120 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2121 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2122 	}
2123 	*sizep = size;
2124 }
2125 
2126 static void
2127 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2128 {
2129 	struct proc *p;
2130 	struct rlimit rlim[RLIM_NLIMITS];
2131 	size_t size;
2132 	int structsize, i;
2133 
2134 	p = (struct proc *)arg;
2135 	size = sizeof(structsize) + sizeof(rlim);
2136 	if (sb != NULL) {
2137 		KASSERT(*sizep == size, ("invalid size"));
2138 		structsize = sizeof(rlim);
2139 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2140 		PROC_LOCK(p);
2141 		for (i = 0; i < RLIM_NLIMITS; i++)
2142 			lim_rlimit_proc(p, i, &rlim[i]);
2143 		PROC_UNLOCK(p);
2144 		sbuf_bcat(sb, rlim, sizeof(rlim));
2145 	}
2146 	*sizep = size;
2147 }
2148 
2149 static void
2150 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2151 {
2152 	struct proc *p;
2153 	size_t size;
2154 	int structsize;
2155 
2156 	p = (struct proc *)arg;
2157 	size = sizeof(structsize) + sizeof(p->p_osrel);
2158 	if (sb != NULL) {
2159 		KASSERT(*sizep == size, ("invalid size"));
2160 		structsize = sizeof(p->p_osrel);
2161 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2162 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2163 	}
2164 	*sizep = size;
2165 }
2166 
2167 static void
2168 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2169 {
2170 	struct proc *p;
2171 	elf_ps_strings_t ps_strings;
2172 	size_t size;
2173 	int structsize;
2174 
2175 	p = (struct proc *)arg;
2176 	size = sizeof(structsize) + sizeof(ps_strings);
2177 	if (sb != NULL) {
2178 		KASSERT(*sizep == size, ("invalid size"));
2179 		structsize = sizeof(ps_strings);
2180 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2181 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2182 #else
2183 		ps_strings = p->p_sysent->sv_psstrings;
2184 #endif
2185 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2186 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2187 	}
2188 	*sizep = size;
2189 }
2190 
2191 static void
2192 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2193 {
2194 	struct proc *p;
2195 	size_t size;
2196 	int structsize;
2197 
2198 	p = (struct proc *)arg;
2199 	if (sb == NULL) {
2200 		size = 0;
2201 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2202 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2203 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2204 		PHOLD(p);
2205 		proc_getauxv(curthread, p, sb);
2206 		PRELE(p);
2207 		sbuf_finish(sb);
2208 		sbuf_delete(sb);
2209 		*sizep = size;
2210 	} else {
2211 		structsize = sizeof(Elf_Auxinfo);
2212 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2213 		PHOLD(p);
2214 		proc_getauxv(curthread, p, sb);
2215 		PRELE(p);
2216 	}
2217 }
2218 
2219 static boolean_t
2220 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2221     int32_t *osrel, const Elf_Phdr *pnote)
2222 {
2223 	const Elf_Note *note, *note0, *note_end;
2224 	const char *note_name;
2225 	char *buf;
2226 	int i, error;
2227 	boolean_t res;
2228 
2229 	/* We need some limit, might as well use PAGE_SIZE. */
2230 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2231 		return (FALSE);
2232 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2233 	if (pnote->p_offset > PAGE_SIZE ||
2234 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2235 		VOP_UNLOCK(imgp->vp, 0);
2236 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2237 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2238 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2239 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2240 		    curthread->td_ucred, NOCRED, NULL, curthread);
2241 		if (error != 0) {
2242 			uprintf("i/o error PT_NOTE\n");
2243 			res = FALSE;
2244 			goto ret;
2245 		}
2246 		note = note0 = (const Elf_Note *)buf;
2247 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2248 	} else {
2249 		note = note0 = (const Elf_Note *)(imgp->image_header +
2250 		    pnote->p_offset);
2251 		note_end = (const Elf_Note *)(imgp->image_header +
2252 		    pnote->p_offset + pnote->p_filesz);
2253 		buf = NULL;
2254 	}
2255 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2256 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2257 		    (const char *)note < sizeof(Elf_Note)) {
2258 			res = FALSE;
2259 			goto ret;
2260 		}
2261 		if (note->n_namesz != checknote->hdr.n_namesz ||
2262 		    note->n_descsz != checknote->hdr.n_descsz ||
2263 		    note->n_type != checknote->hdr.n_type)
2264 			goto nextnote;
2265 		note_name = (const char *)(note + 1);
2266 		if (note_name + checknote->hdr.n_namesz >=
2267 		    (const char *)note_end || strncmp(checknote->vendor,
2268 		    note_name, checknote->hdr.n_namesz) != 0)
2269 			goto nextnote;
2270 
2271 		/*
2272 		 * Fetch the osreldate for binary
2273 		 * from the ELF OSABI-note if necessary.
2274 		 */
2275 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2276 		    checknote->trans_osrel != NULL) {
2277 			res = checknote->trans_osrel(note, osrel);
2278 			goto ret;
2279 		}
2280 		res = TRUE;
2281 		goto ret;
2282 nextnote:
2283 		note = (const Elf_Note *)((const char *)(note + 1) +
2284 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2285 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2286 	}
2287 	res = FALSE;
2288 ret:
2289 	free(buf, M_TEMP);
2290 	return (res);
2291 }
2292 
2293 /*
2294  * Try to find the appropriate ABI-note section for checknote,
2295  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2296  * first page of the image is searched, the same as for headers.
2297  */
2298 static boolean_t
2299 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2300     int32_t *osrel)
2301 {
2302 	const Elf_Phdr *phdr;
2303 	const Elf_Ehdr *hdr;
2304 	int i;
2305 
2306 	hdr = (const Elf_Ehdr *)imgp->image_header;
2307 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2308 
2309 	for (i = 0; i < hdr->e_phnum; i++) {
2310 		if (phdr[i].p_type == PT_NOTE &&
2311 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2312 			return (TRUE);
2313 	}
2314 	return (FALSE);
2315 
2316 }
2317 
2318 /*
2319  * Tell kern_execve.c about it, with a little help from the linker.
2320  */
2321 static struct execsw __elfN(execsw) = {
2322 	__CONCAT(exec_, __elfN(imgact)),
2323 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2324 };
2325 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2326 
2327 static vm_prot_t
2328 __elfN(trans_prot)(Elf_Word flags)
2329 {
2330 	vm_prot_t prot;
2331 
2332 	prot = 0;
2333 	if (flags & PF_X)
2334 		prot |= VM_PROT_EXECUTE;
2335 	if (flags & PF_W)
2336 		prot |= VM_PROT_WRITE;
2337 	if (flags & PF_R)
2338 		prot |= VM_PROT_READ;
2339 #if __ELF_WORD_SIZE == 32
2340 #if defined(__amd64__)
2341 	if (i386_read_exec && (flags & PF_R))
2342 		prot |= VM_PROT_EXECUTE;
2343 #endif
2344 #endif
2345 	return (prot);
2346 }
2347 
2348 static Elf_Word
2349 __elfN(untrans_prot)(vm_prot_t prot)
2350 {
2351 	Elf_Word flags;
2352 
2353 	flags = 0;
2354 	if (prot & VM_PROT_EXECUTE)
2355 		flags |= PF_X;
2356 	if (prot & VM_PROT_READ)
2357 		flags |= PF_R;
2358 	if (prot & VM_PROT_WRITE)
2359 		flags |= PF_W;
2360 	return (flags);
2361 }
2362