xref: /freebsd/sys/kern/imgact_elf.c (revision 114350b9de8403b9e43bbad0d1534455a077d999)
1 /*-
2  * Copyright (c) 2000 David O'Brien
3  * Copyright (c) 1995-1996 Søren Schmidt
4  * Copyright (c) 1996 Peter Wemm
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_gzio.h"
37 
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
40 #include <sys/exec.h>
41 #include <sys/fcntl.h>
42 #include <sys/gzio.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mman.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
58 #include <sys/sbuf.h>
59 #include <sys/sf_buf.h>
60 #include <sys/smp.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
63 #include <sys/stat.h>
64 #include <sys/sx.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
71 #include <sys/user.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
80 
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
83 
84 #define ELF_NOTE_ROUNDSIZE	4
85 #define OLD_EI_BRAND	8
86 
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89     const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91     u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
93     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
94     size_t pagesize);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97     int32_t *osrel);
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100     Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
103 
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
105     "");
106 
107 #define	CORE_BUF_SIZE	(16 * 1024)
108 
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
113 
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116     &elf_legacy_coredump, 0, "");
117 
118 int __elfN(nxstack) =
119 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
120 	1;
121 #else
122 	0;
123 #endif
124 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
125     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
126     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
127 
128 #if __ELF_WORD_SIZE == 32
129 #if defined(__amd64__)
130 int i386_read_exec = 0;
131 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
132     "enable execution from readable segments");
133 #endif
134 #endif
135 
136 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
137 
138 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
139 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
140 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
141 
142 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
143 
144 Elf_Brandnote __elfN(freebsd_brandnote) = {
145 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
146 	.hdr.n_descsz	= sizeof(int32_t),
147 	.hdr.n_type	= 1,
148 	.vendor		= FREEBSD_ABI_VENDOR,
149 	.flags		= BN_TRANSLATE_OSREL,
150 	.trans_osrel	= __elfN(freebsd_trans_osrel)
151 };
152 
153 static boolean_t
154 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
155 {
156 	uintptr_t p;
157 
158 	p = (uintptr_t)(note + 1);
159 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
160 	*osrel = *(const int32_t *)(p);
161 
162 	return (TRUE);
163 }
164 
165 static const char GNU_ABI_VENDOR[] = "GNU";
166 static int GNU_KFREEBSD_ABI_DESC = 3;
167 
168 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
169 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
170 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
171 	.hdr.n_type	= 1,
172 	.vendor		= GNU_ABI_VENDOR,
173 	.flags		= BN_TRANSLATE_OSREL,
174 	.trans_osrel	= kfreebsd_trans_osrel
175 };
176 
177 static boolean_t
178 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
179 {
180 	const Elf32_Word *desc;
181 	uintptr_t p;
182 
183 	p = (uintptr_t)(note + 1);
184 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
185 
186 	desc = (const Elf32_Word *)p;
187 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
188 		return (FALSE);
189 
190 	/*
191 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
192 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
193 	 */
194 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
195 
196 	return (TRUE);
197 }
198 
199 int
200 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
201 {
202 	int i;
203 
204 	for (i = 0; i < MAX_BRANDS; i++) {
205 		if (elf_brand_list[i] == NULL) {
206 			elf_brand_list[i] = entry;
207 			break;
208 		}
209 	}
210 	if (i == MAX_BRANDS) {
211 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
212 			__func__, entry);
213 		return (-1);
214 	}
215 	return (0);
216 }
217 
218 int
219 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
220 {
221 	int i;
222 
223 	for (i = 0; i < MAX_BRANDS; i++) {
224 		if (elf_brand_list[i] == entry) {
225 			elf_brand_list[i] = NULL;
226 			break;
227 		}
228 	}
229 	if (i == MAX_BRANDS)
230 		return (-1);
231 	return (0);
232 }
233 
234 int
235 __elfN(brand_inuse)(Elf_Brandinfo *entry)
236 {
237 	struct proc *p;
238 	int rval = FALSE;
239 
240 	sx_slock(&allproc_lock);
241 	FOREACH_PROC_IN_SYSTEM(p) {
242 		if (p->p_sysent == entry->sysvec) {
243 			rval = TRUE;
244 			break;
245 		}
246 	}
247 	sx_sunlock(&allproc_lock);
248 
249 	return (rval);
250 }
251 
252 static Elf_Brandinfo *
253 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
254     int interp_name_len, int32_t *osrel)
255 {
256 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
257 	Elf_Brandinfo *bi;
258 	boolean_t ret;
259 	int i;
260 
261 	/*
262 	 * We support four types of branding -- (1) the ELF EI_OSABI field
263 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
264 	 * branding w/in the ELF header, (3) path of the `interp_path'
265 	 * field, and (4) the ".note.ABI-tag" ELF section.
266 	 */
267 
268 	/* Look for an ".note.ABI-tag" ELF section */
269 	for (i = 0; i < MAX_BRANDS; i++) {
270 		bi = elf_brand_list[i];
271 		if (bi == NULL)
272 			continue;
273 		if (hdr->e_machine == bi->machine && (bi->flags &
274 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
275 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
276 			if (ret)
277 				return (bi);
278 		}
279 	}
280 
281 	/* If the executable has a brand, search for it in the brand list. */
282 	for (i = 0; i < MAX_BRANDS; i++) {
283 		bi = elf_brand_list[i];
284 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
285 			continue;
286 		if (hdr->e_machine == bi->machine &&
287 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
288 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
289 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
290 			/* Looks good, but give brand a chance to veto */
291 			if (!bi->header_supported || bi->header_supported(imgp))
292 				return (bi);
293 		}
294 	}
295 
296 	/* No known brand, see if the header is recognized by any brand */
297 	for (i = 0; i < MAX_BRANDS; i++) {
298 		bi = elf_brand_list[i];
299 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
300 		    bi->header_supported == NULL)
301 			continue;
302 		if (hdr->e_machine == bi->machine) {
303 			ret = bi->header_supported(imgp);
304 			if (ret)
305 				return (bi);
306 		}
307 	}
308 
309 	/* Lacking a known brand, search for a recognized interpreter. */
310 	if (interp != NULL) {
311 		for (i = 0; i < MAX_BRANDS; i++) {
312 			bi = elf_brand_list[i];
313 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
314 				continue;
315 			if (hdr->e_machine == bi->machine &&
316 			    /* ELF image p_filesz includes terminating zero */
317 			    strlen(bi->interp_path) + 1 == interp_name_len &&
318 			    strncmp(interp, bi->interp_path, interp_name_len)
319 			    == 0)
320 				return (bi);
321 		}
322 	}
323 
324 	/* Lacking a recognized interpreter, try the default brand */
325 	for (i = 0; i < MAX_BRANDS; i++) {
326 		bi = elf_brand_list[i];
327 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
328 			continue;
329 		if (hdr->e_machine == bi->machine &&
330 		    __elfN(fallback_brand) == bi->brand)
331 			return (bi);
332 	}
333 	return (NULL);
334 }
335 
336 static int
337 __elfN(check_header)(const Elf_Ehdr *hdr)
338 {
339 	Elf_Brandinfo *bi;
340 	int i;
341 
342 	if (!IS_ELF(*hdr) ||
343 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
344 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
345 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
346 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
347 	    hdr->e_version != ELF_TARG_VER)
348 		return (ENOEXEC);
349 
350 	/*
351 	 * Make sure we have at least one brand for this machine.
352 	 */
353 
354 	for (i = 0; i < MAX_BRANDS; i++) {
355 		bi = elf_brand_list[i];
356 		if (bi != NULL && bi->machine == hdr->e_machine)
357 			break;
358 	}
359 	if (i == MAX_BRANDS)
360 		return (ENOEXEC);
361 
362 	return (0);
363 }
364 
365 static int
366 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
367     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
368 {
369 	struct sf_buf *sf;
370 	int error;
371 	vm_offset_t off;
372 
373 	/*
374 	 * Create the page if it doesn't exist yet. Ignore errors.
375 	 */
376 	vm_map_lock(map);
377 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
378 	    VM_PROT_ALL, VM_PROT_ALL, 0);
379 	vm_map_unlock(map);
380 
381 	/*
382 	 * Find the page from the underlying object.
383 	 */
384 	if (object) {
385 		sf = vm_imgact_map_page(object, offset);
386 		if (sf == NULL)
387 			return (KERN_FAILURE);
388 		off = offset - trunc_page(offset);
389 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
390 		    end - start);
391 		vm_imgact_unmap_page(sf);
392 		if (error) {
393 			return (KERN_FAILURE);
394 		}
395 	}
396 
397 	return (KERN_SUCCESS);
398 }
399 
400 static int
401 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
402     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
403 {
404 	struct sf_buf *sf;
405 	vm_offset_t off;
406 	vm_size_t sz;
407 	int error, rv;
408 
409 	if (start != trunc_page(start)) {
410 		rv = __elfN(map_partial)(map, object, offset, start,
411 		    round_page(start), prot);
412 		if (rv)
413 			return (rv);
414 		offset += round_page(start) - start;
415 		start = round_page(start);
416 	}
417 	if (end != round_page(end)) {
418 		rv = __elfN(map_partial)(map, object, offset +
419 		    trunc_page(end) - start, trunc_page(end), end, prot);
420 		if (rv)
421 			return (rv);
422 		end = trunc_page(end);
423 	}
424 	if (end > start) {
425 		if (offset & PAGE_MASK) {
426 			/*
427 			 * The mapping is not page aligned. This means we have
428 			 * to copy the data. Sigh.
429 			 */
430 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
431 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
432 			    0);
433 			if (rv)
434 				return (rv);
435 			if (object == NULL)
436 				return (KERN_SUCCESS);
437 			for (; start < end; start += sz) {
438 				sf = vm_imgact_map_page(object, offset);
439 				if (sf == NULL)
440 					return (KERN_FAILURE);
441 				off = offset - trunc_page(offset);
442 				sz = end - start;
443 				if (sz > PAGE_SIZE - off)
444 					sz = PAGE_SIZE - off;
445 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
446 				    (caddr_t)start, sz);
447 				vm_imgact_unmap_page(sf);
448 				if (error) {
449 					return (KERN_FAILURE);
450 				}
451 				offset += sz;
452 			}
453 			rv = KERN_SUCCESS;
454 		} else {
455 			vm_object_reference(object);
456 			vm_map_lock(map);
457 			rv = vm_map_insert(map, object, offset, start, end,
458 			    prot, VM_PROT_ALL, cow);
459 			vm_map_unlock(map);
460 			if (rv != KERN_SUCCESS)
461 				vm_object_deallocate(object);
462 		}
463 		return (rv);
464 	} else {
465 		return (KERN_SUCCESS);
466 	}
467 }
468 
469 static int
470 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
471     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
472     size_t pagesize)
473 {
474 	struct sf_buf *sf;
475 	size_t map_len;
476 	vm_map_t map;
477 	vm_object_t object;
478 	vm_offset_t map_addr;
479 	int error, rv, cow;
480 	size_t copy_len;
481 	vm_offset_t file_addr;
482 
483 	/*
484 	 * It's necessary to fail if the filsz + offset taken from the
485 	 * header is greater than the actual file pager object's size.
486 	 * If we were to allow this, then the vm_map_find() below would
487 	 * walk right off the end of the file object and into the ether.
488 	 *
489 	 * While I'm here, might as well check for something else that
490 	 * is invalid: filsz cannot be greater than memsz.
491 	 */
492 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
493 		uprintf("elf_load_section: truncated ELF file\n");
494 		return (ENOEXEC);
495 	}
496 
497 	object = imgp->object;
498 	map = &imgp->proc->p_vmspace->vm_map;
499 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
500 	file_addr = trunc_page_ps(offset, pagesize);
501 
502 	/*
503 	 * We have two choices.  We can either clear the data in the last page
504 	 * of an oversized mapping, or we can start the anon mapping a page
505 	 * early and copy the initialized data into that first page.  We
506 	 * choose the second..
507 	 */
508 	if (memsz > filsz)
509 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
510 	else
511 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
512 
513 	if (map_len != 0) {
514 		/* cow flags: don't dump readonly sections in core */
515 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
516 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
517 
518 		rv = __elfN(map_insert)(map,
519 				      object,
520 				      file_addr,	/* file offset */
521 				      map_addr,		/* virtual start */
522 				      map_addr + map_len,/* virtual end */
523 				      prot,
524 				      cow);
525 		if (rv != KERN_SUCCESS)
526 			return (EINVAL);
527 
528 		/* we can stop now if we've covered it all */
529 		if (memsz == filsz) {
530 			return (0);
531 		}
532 	}
533 
534 
535 	/*
536 	 * We have to get the remaining bit of the file into the first part
537 	 * of the oversized map segment.  This is normally because the .data
538 	 * segment in the file is extended to provide bss.  It's a neat idea
539 	 * to try and save a page, but it's a pain in the behind to implement.
540 	 */
541 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
542 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
543 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
544 	    map_addr;
545 
546 	/* This had damn well better be true! */
547 	if (map_len != 0) {
548 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
549 		    map_len, VM_PROT_ALL, 0);
550 		if (rv != KERN_SUCCESS) {
551 			return (EINVAL);
552 		}
553 	}
554 
555 	if (copy_len != 0) {
556 		vm_offset_t off;
557 
558 		sf = vm_imgact_map_page(object, offset + filsz);
559 		if (sf == NULL)
560 			return (EIO);
561 
562 		/* send the page fragment to user space */
563 		off = trunc_page_ps(offset + filsz, pagesize) -
564 		    trunc_page(offset + filsz);
565 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
566 		    (caddr_t)map_addr, copy_len);
567 		vm_imgact_unmap_page(sf);
568 		if (error) {
569 			return (error);
570 		}
571 	}
572 
573 	/*
574 	 * set it to the specified protection.
575 	 * XXX had better undo the damage from pasting over the cracks here!
576 	 */
577 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
578 	    map_len), prot, FALSE);
579 
580 	return (0);
581 }
582 
583 /*
584  * Load the file "file" into memory.  It may be either a shared object
585  * or an executable.
586  *
587  * The "addr" reference parameter is in/out.  On entry, it specifies
588  * the address where a shared object should be loaded.  If the file is
589  * an executable, this value is ignored.  On exit, "addr" specifies
590  * where the file was actually loaded.
591  *
592  * The "entry" reference parameter is out only.  On exit, it specifies
593  * the entry point for the loaded file.
594  */
595 static int
596 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
597 	u_long *entry, size_t pagesize)
598 {
599 	struct {
600 		struct nameidata nd;
601 		struct vattr attr;
602 		struct image_params image_params;
603 	} *tempdata;
604 	const Elf_Ehdr *hdr = NULL;
605 	const Elf_Phdr *phdr = NULL;
606 	struct nameidata *nd;
607 	struct vattr *attr;
608 	struct image_params *imgp;
609 	vm_prot_t prot;
610 	u_long rbase;
611 	u_long base_addr = 0;
612 	int error, i, numsegs;
613 
614 #ifdef CAPABILITY_MODE
615 	/*
616 	 * XXXJA: This check can go away once we are sufficiently confident
617 	 * that the checks in namei() are correct.
618 	 */
619 	if (IN_CAPABILITY_MODE(curthread))
620 		return (ECAPMODE);
621 #endif
622 
623 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
624 	nd = &tempdata->nd;
625 	attr = &tempdata->attr;
626 	imgp = &tempdata->image_params;
627 
628 	/*
629 	 * Initialize part of the common data
630 	 */
631 	imgp->proc = p;
632 	imgp->attr = attr;
633 	imgp->firstpage = NULL;
634 	imgp->image_header = NULL;
635 	imgp->object = NULL;
636 	imgp->execlabel = NULL;
637 
638 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
639 	if ((error = namei(nd)) != 0) {
640 		nd->ni_vp = NULL;
641 		goto fail;
642 	}
643 	NDFREE(nd, NDF_ONLY_PNBUF);
644 	imgp->vp = nd->ni_vp;
645 
646 	/*
647 	 * Check permissions, modes, uid, etc on the file, and "open" it.
648 	 */
649 	error = exec_check_permissions(imgp);
650 	if (error)
651 		goto fail;
652 
653 	error = exec_map_first_page(imgp);
654 	if (error)
655 		goto fail;
656 
657 	/*
658 	 * Also make certain that the interpreter stays the same, so set
659 	 * its VV_TEXT flag, too.
660 	 */
661 	VOP_SET_TEXT(nd->ni_vp);
662 
663 	imgp->object = nd->ni_vp->v_object;
664 
665 	hdr = (const Elf_Ehdr *)imgp->image_header;
666 	if ((error = __elfN(check_header)(hdr)) != 0)
667 		goto fail;
668 	if (hdr->e_type == ET_DYN)
669 		rbase = *addr;
670 	else if (hdr->e_type == ET_EXEC)
671 		rbase = 0;
672 	else {
673 		error = ENOEXEC;
674 		goto fail;
675 	}
676 
677 	/* Only support headers that fit within first page for now      */
678 	if ((hdr->e_phoff > PAGE_SIZE) ||
679 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
680 		error = ENOEXEC;
681 		goto fail;
682 	}
683 
684 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
685 	if (!aligned(phdr, Elf_Addr)) {
686 		error = ENOEXEC;
687 		goto fail;
688 	}
689 
690 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
691 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
692 			/* Loadable segment */
693 			prot = __elfN(trans_prot)(phdr[i].p_flags);
694 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
695 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
696 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
697 			if (error != 0)
698 				goto fail;
699 			/*
700 			 * Establish the base address if this is the
701 			 * first segment.
702 			 */
703 			if (numsegs == 0)
704   				base_addr = trunc_page(phdr[i].p_vaddr +
705 				    rbase);
706 			numsegs++;
707 		}
708 	}
709 	*addr = base_addr;
710 	*entry = (unsigned long)hdr->e_entry + rbase;
711 
712 fail:
713 	if (imgp->firstpage)
714 		exec_unmap_first_page(imgp);
715 
716 	if (nd->ni_vp)
717 		vput(nd->ni_vp);
718 
719 	free(tempdata, M_TEMP);
720 
721 	return (error);
722 }
723 
724 static int
725 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
726 {
727 	struct thread *td;
728 	const Elf_Ehdr *hdr;
729 	const Elf_Phdr *phdr;
730 	Elf_Auxargs *elf_auxargs;
731 	struct vmspace *vmspace;
732 	const char *err_str, *newinterp;
733 	char *interp, *interp_buf, *path;
734 	Elf_Brandinfo *brand_info;
735 	struct sysentvec *sv;
736 	vm_prot_t prot;
737 	u_long text_size, data_size, total_size, text_addr, data_addr;
738 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
739 	int32_t osrel;
740 	int error, i, n, interp_name_len, have_interp;
741 
742 	hdr = (const Elf_Ehdr *)imgp->image_header;
743 
744 	/*
745 	 * Do we have a valid ELF header ?
746 	 *
747 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
748 	 * if particular brand doesn't support it.
749 	 */
750 	if (__elfN(check_header)(hdr) != 0 ||
751 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
752 		return (-1);
753 
754 	/*
755 	 * From here on down, we return an errno, not -1, as we've
756 	 * detected an ELF file.
757 	 */
758 
759 	if ((hdr->e_phoff > PAGE_SIZE) ||
760 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
761 		/* Only support headers in first page for now */
762 		uprintf("Program headers not in the first page\n");
763 		return (ENOEXEC);
764 	}
765 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
766 	if (!aligned(phdr, Elf_Addr)) {
767 		uprintf("Unaligned program headers\n");
768 		return (ENOEXEC);
769 	}
770 
771 	n = error = 0;
772 	baddr = 0;
773 	osrel = 0;
774 	text_size = data_size = total_size = text_addr = data_addr = 0;
775 	entry = proghdr = 0;
776 	interp_name_len = 0;
777 	err_str = newinterp = NULL;
778 	interp = interp_buf = NULL;
779 	td = curthread;
780 
781 	for (i = 0; i < hdr->e_phnum; i++) {
782 		switch (phdr[i].p_type) {
783 		case PT_LOAD:
784 			if (n == 0)
785 				baddr = phdr[i].p_vaddr;
786 			n++;
787 			break;
788 		case PT_INTERP:
789 			/* Path to interpreter */
790 			if (phdr[i].p_filesz > MAXPATHLEN) {
791 				uprintf("Invalid PT_INTERP\n");
792 				error = ENOEXEC;
793 				goto ret;
794 			}
795 			interp_name_len = phdr[i].p_filesz;
796 			if (phdr[i].p_offset > PAGE_SIZE ||
797 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
798 				VOP_UNLOCK(imgp->vp, 0);
799 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
800 				    M_WAITOK);
801 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
802 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
803 				    interp_name_len, phdr[i].p_offset,
804 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
805 				    NOCRED, NULL, td);
806 				if (error != 0) {
807 					uprintf("i/o error PT_INTERP\n");
808 					goto ret;
809 				}
810 				interp_buf[interp_name_len] = '\0';
811 				interp = interp_buf;
812 			} else {
813 				interp = __DECONST(char *, imgp->image_header) +
814 				    phdr[i].p_offset;
815 			}
816 			break;
817 		case PT_GNU_STACK:
818 			if (__elfN(nxstack))
819 				imgp->stack_prot =
820 				    __elfN(trans_prot)(phdr[i].p_flags);
821 			imgp->stack_sz = phdr[i].p_memsz;
822 			break;
823 		}
824 	}
825 
826 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
827 	    &osrel);
828 	if (brand_info == NULL) {
829 		uprintf("ELF binary type \"%u\" not known.\n",
830 		    hdr->e_ident[EI_OSABI]);
831 		error = ENOEXEC;
832 		goto ret;
833 	}
834 	if (hdr->e_type == ET_DYN) {
835 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
836 			uprintf("Cannot execute shared object\n");
837 			error = ENOEXEC;
838 			goto ret;
839 		}
840 		/*
841 		 * Honour the base load address from the dso if it is
842 		 * non-zero for some reason.
843 		 */
844 		if (baddr == 0)
845 			et_dyn_addr = ET_DYN_LOAD_ADDR;
846 		else
847 			et_dyn_addr = 0;
848 	} else
849 		et_dyn_addr = 0;
850 	sv = brand_info->sysvec;
851 	if (interp != NULL && brand_info->interp_newpath != NULL)
852 		newinterp = brand_info->interp_newpath;
853 
854 	/*
855 	 * Avoid a possible deadlock if the current address space is destroyed
856 	 * and that address space maps the locked vnode.  In the common case,
857 	 * the locked vnode's v_usecount is decremented but remains greater
858 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
859 	 * However, in cases where the vnode lock is external, such as nullfs,
860 	 * v_usecount may become zero.
861 	 *
862 	 * The VV_TEXT flag prevents modifications to the executable while
863 	 * the vnode is unlocked.
864 	 */
865 	VOP_UNLOCK(imgp->vp, 0);
866 
867 	error = exec_new_vmspace(imgp, sv);
868 	imgp->proc->p_sysent = sv;
869 
870 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
871 	if (error != 0)
872 		goto ret;
873 
874 	for (i = 0; i < hdr->e_phnum; i++) {
875 		switch (phdr[i].p_type) {
876 		case PT_LOAD:	/* Loadable segment */
877 			if (phdr[i].p_memsz == 0)
878 				break;
879 			prot = __elfN(trans_prot)(phdr[i].p_flags);
880 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
881 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
882 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
883 			    sv->sv_pagesize);
884 			if (error != 0)
885 				goto ret;
886 
887 			/*
888 			 * If this segment contains the program headers,
889 			 * remember their virtual address for the AT_PHDR
890 			 * aux entry. Static binaries don't usually include
891 			 * a PT_PHDR entry.
892 			 */
893 			if (phdr[i].p_offset == 0 &&
894 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
895 				<= phdr[i].p_filesz)
896 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
897 				    et_dyn_addr;
898 
899 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
900 			seg_size = round_page(phdr[i].p_memsz +
901 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
902 
903 			/*
904 			 * Make the largest executable segment the official
905 			 * text segment and all others data.
906 			 *
907 			 * Note that obreak() assumes that data_addr +
908 			 * data_size == end of data load area, and the ELF
909 			 * file format expects segments to be sorted by
910 			 * address.  If multiple data segments exist, the
911 			 * last one will be used.
912 			 */
913 
914 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
915 				text_size = seg_size;
916 				text_addr = seg_addr;
917 			} else {
918 				data_size = seg_size;
919 				data_addr = seg_addr;
920 			}
921 			total_size += seg_size;
922 			break;
923 		case PT_PHDR: 	/* Program header table info */
924 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
925 			break;
926 		default:
927 			break;
928 		}
929 	}
930 
931 	if (data_addr == 0 && data_size == 0) {
932 		data_addr = text_addr;
933 		data_size = text_size;
934 	}
935 
936 	entry = (u_long)hdr->e_entry + et_dyn_addr;
937 
938 	/*
939 	 * Check limits.  It should be safe to check the
940 	 * limits after loading the segments since we do
941 	 * not actually fault in all the segments pages.
942 	 */
943 	PROC_LOCK(imgp->proc);
944 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
945 		err_str = "Data segment size exceeds process limit";
946 	else if (text_size > maxtsiz)
947 		err_str = "Text segment size exceeds system limit";
948 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
949 		err_str = "Total segment size exceeds process limit";
950 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
951 		err_str = "Data segment size exceeds resource limit";
952 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
953 		err_str = "Total segment size exceeds resource limit";
954 	if (err_str != NULL) {
955 		PROC_UNLOCK(imgp->proc);
956 		uprintf("%s\n", err_str);
957 		error = ENOMEM;
958 		goto ret;
959 	}
960 
961 	vmspace = imgp->proc->p_vmspace;
962 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
963 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
964 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
965 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
966 
967 	/*
968 	 * We load the dynamic linker where a userland call
969 	 * to mmap(0, ...) would put it.  The rationale behind this
970 	 * calculation is that it leaves room for the heap to grow to
971 	 * its maximum allowed size.
972 	 */
973 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
974 	    RLIMIT_DATA));
975 	PROC_UNLOCK(imgp->proc);
976 
977 	imgp->entry_addr = entry;
978 
979 	if (interp != NULL) {
980 		have_interp = FALSE;
981 		VOP_UNLOCK(imgp->vp, 0);
982 		if (brand_info->emul_path != NULL &&
983 		    brand_info->emul_path[0] != '\0') {
984 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
985 			snprintf(path, MAXPATHLEN, "%s%s",
986 			    brand_info->emul_path, interp);
987 			error = __elfN(load_file)(imgp->proc, path, &addr,
988 			    &imgp->entry_addr, sv->sv_pagesize);
989 			free(path, M_TEMP);
990 			if (error == 0)
991 				have_interp = TRUE;
992 		}
993 		if (!have_interp && newinterp != NULL) {
994 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
995 			    &imgp->entry_addr, sv->sv_pagesize);
996 			if (error == 0)
997 				have_interp = TRUE;
998 		}
999 		if (!have_interp) {
1000 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1001 			    &imgp->entry_addr, sv->sv_pagesize);
1002 		}
1003 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1004 		if (error != 0) {
1005 			uprintf("ELF interpreter %s not found\n", interp);
1006 			goto ret;
1007 		}
1008 	} else
1009 		addr = et_dyn_addr;
1010 
1011 	/*
1012 	 * Construct auxargs table (used by the fixup routine)
1013 	 */
1014 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1015 	elf_auxargs->execfd = -1;
1016 	elf_auxargs->phdr = proghdr;
1017 	elf_auxargs->phent = hdr->e_phentsize;
1018 	elf_auxargs->phnum = hdr->e_phnum;
1019 	elf_auxargs->pagesz = PAGE_SIZE;
1020 	elf_auxargs->base = addr;
1021 	elf_auxargs->flags = 0;
1022 	elf_auxargs->entry = entry;
1023 	elf_auxargs->hdr_eflags = hdr->e_flags;
1024 
1025 	imgp->auxargs = elf_auxargs;
1026 	imgp->interpreted = 0;
1027 	imgp->reloc_base = addr;
1028 	imgp->proc->p_osrel = osrel;
1029 
1030  ret:
1031 	free(interp_buf, M_TEMP);
1032 	return (error);
1033 }
1034 
1035 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1036 
1037 int
1038 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1039 {
1040 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1041 	Elf_Addr *base;
1042 	Elf_Addr *pos;
1043 
1044 	base = (Elf_Addr *)*stack_base;
1045 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1046 
1047 	if (args->execfd != -1)
1048 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1049 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1050 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1051 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1052 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1053 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1054 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1055 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1056 #ifdef AT_EHDRFLAGS
1057 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1058 #endif
1059 	if (imgp->execpathp != 0)
1060 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1061 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1062 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1063 	if (imgp->canary != 0) {
1064 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1065 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1066 	}
1067 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1068 	if (imgp->pagesizes != 0) {
1069 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1070 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1071 	}
1072 	if (imgp->sysent->sv_timekeep_base != 0) {
1073 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1074 		    imgp->sysent->sv_timekeep_base);
1075 	}
1076 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1077 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1078 	    imgp->sysent->sv_stackprot);
1079 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1080 
1081 	free(imgp->auxargs, M_TEMP);
1082 	imgp->auxargs = NULL;
1083 
1084 	base--;
1085 	suword(base, (long)imgp->args->argc);
1086 	*stack_base = (register_t *)base;
1087 	return (0);
1088 }
1089 
1090 /*
1091  * Code for generating ELF core dumps.
1092  */
1093 
1094 typedef void (*segment_callback)(vm_map_entry_t, void *);
1095 
1096 /* Closure for cb_put_phdr(). */
1097 struct phdr_closure {
1098 	Elf_Phdr *phdr;		/* Program header to fill in */
1099 	Elf_Off offset;		/* Offset of segment in core file */
1100 };
1101 
1102 /* Closure for cb_size_segment(). */
1103 struct sseg_closure {
1104 	int count;		/* Count of writable segments. */
1105 	size_t size;		/* Total size of all writable segments. */
1106 };
1107 
1108 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1109 
1110 struct note_info {
1111 	int		type;		/* Note type. */
1112 	outfunc_t 	outfunc; 	/* Output function. */
1113 	void		*outarg;	/* Argument for the output function. */
1114 	size_t		outsize;	/* Output size. */
1115 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1116 };
1117 
1118 TAILQ_HEAD(note_info_list, note_info);
1119 
1120 /* Coredump output parameters. */
1121 struct coredump_params {
1122 	off_t		offset;
1123 	struct ucred	*active_cred;
1124 	struct ucred	*file_cred;
1125 	struct thread	*td;
1126 	struct vnode	*vp;
1127 	struct gzio_stream *gzs;
1128 };
1129 
1130 static void cb_put_phdr(vm_map_entry_t, void *);
1131 static void cb_size_segment(vm_map_entry_t, void *);
1132 static int core_write(struct coredump_params *, void *, size_t, off_t,
1133     enum uio_seg);
1134 static void each_writable_segment(struct thread *, segment_callback, void *);
1135 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1136     struct note_info_list *, size_t);
1137 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1138     size_t *);
1139 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1140 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1141 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1142 static int sbuf_drain_core_output(void *, const char *, int);
1143 static int sbuf_drain_count(void *arg, const char *data, int len);
1144 
1145 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1146 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1147 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1148 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1149 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1150 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1151 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1152 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1153 static void note_procstat_files(void *, struct sbuf *, size_t *);
1154 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1155 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1156 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1157 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1158 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1159 
1160 #ifdef GZIO
1161 extern int compress_user_cores_gzlevel;
1162 
1163 /*
1164  * Write out a core segment to the compression stream.
1165  */
1166 static int
1167 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1168 {
1169 	u_int chunk_len;
1170 	int error;
1171 
1172 	while (len > 0) {
1173 		chunk_len = MIN(len, CORE_BUF_SIZE);
1174 		copyin(base, buf, chunk_len);
1175 		error = gzio_write(p->gzs, buf, chunk_len);
1176 		if (error != 0)
1177 			break;
1178 		base += chunk_len;
1179 		len -= chunk_len;
1180 	}
1181 	return (error);
1182 }
1183 
1184 static int
1185 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1186 {
1187 
1188 	return (core_write((struct coredump_params *)arg, base, len, offset,
1189 	    UIO_SYSSPACE));
1190 }
1191 #endif /* GZIO */
1192 
1193 static int
1194 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1195     enum uio_seg seg)
1196 {
1197 
1198 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1199 	    seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1200 	    p->active_cred, p->file_cred, NULL, p->td));
1201 }
1202 
1203 static int
1204 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1205     void *tmpbuf)
1206 {
1207 
1208 #ifdef GZIO
1209 	if (p->gzs != NULL)
1210 		return (compress_chunk(p, base, tmpbuf, len));
1211 #endif
1212 	return (core_write(p, base, len, offset, UIO_USERSPACE));
1213 }
1214 
1215 /*
1216  * Drain into a core file.
1217  */
1218 static int
1219 sbuf_drain_core_output(void *arg, const char *data, int len)
1220 {
1221 	struct coredump_params *p;
1222 	int error, locked;
1223 
1224 	p = (struct coredump_params *)arg;
1225 
1226 	/*
1227 	 * Some kern_proc out routines that print to this sbuf may
1228 	 * call us with the process lock held. Draining with the
1229 	 * non-sleepable lock held is unsafe. The lock is needed for
1230 	 * those routines when dumping a live process. In our case we
1231 	 * can safely release the lock before draining and acquire
1232 	 * again after.
1233 	 */
1234 	locked = PROC_LOCKED(p->td->td_proc);
1235 	if (locked)
1236 		PROC_UNLOCK(p->td->td_proc);
1237 #ifdef GZIO
1238 	if (p->gzs != NULL)
1239 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1240 	else
1241 #endif
1242 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1243 		    UIO_SYSSPACE);
1244 	if (locked)
1245 		PROC_LOCK(p->td->td_proc);
1246 	if (error != 0)
1247 		return (-error);
1248 	p->offset += len;
1249 	return (len);
1250 }
1251 
1252 /*
1253  * Drain into a counter.
1254  */
1255 static int
1256 sbuf_drain_count(void *arg, const char *data __unused, int len)
1257 {
1258 	size_t *sizep;
1259 
1260 	sizep = (size_t *)arg;
1261 	*sizep += len;
1262 	return (len);
1263 }
1264 
1265 int
1266 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1267 {
1268 	struct ucred *cred = td->td_ucred;
1269 	int error = 0;
1270 	struct sseg_closure seginfo;
1271 	struct note_info_list notelst;
1272 	struct coredump_params params;
1273 	struct note_info *ninfo;
1274 	void *hdr, *tmpbuf;
1275 	size_t hdrsize, notesz, coresize;
1276 #ifdef GZIO
1277 	boolean_t compress;
1278 
1279 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1280 #endif
1281 	hdr = NULL;
1282 	tmpbuf = NULL;
1283 	TAILQ_INIT(&notelst);
1284 
1285 	/* Size the program segments. */
1286 	seginfo.count = 0;
1287 	seginfo.size = 0;
1288 	each_writable_segment(td, cb_size_segment, &seginfo);
1289 
1290 	/*
1291 	 * Collect info about the core file header area.
1292 	 */
1293 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1294 	__elfN(prepare_notes)(td, &notelst, &notesz);
1295 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1296 
1297 	/* Set up core dump parameters. */
1298 	params.offset = 0;
1299 	params.active_cred = cred;
1300 	params.file_cred = NOCRED;
1301 	params.td = td;
1302 	params.vp = vp;
1303 	params.gzs = NULL;
1304 
1305 #ifdef RACCT
1306 	if (racct_enable) {
1307 		PROC_LOCK(td->td_proc);
1308 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1309 		PROC_UNLOCK(td->td_proc);
1310 		if (error != 0) {
1311 			error = EFAULT;
1312 			goto done;
1313 		}
1314 	}
1315 #endif
1316 	if (coresize >= limit) {
1317 		error = EFAULT;
1318 		goto done;
1319 	}
1320 
1321 #ifdef GZIO
1322 	/* Create a compression stream if necessary. */
1323 	if (compress) {
1324 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1325 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1326 		if (params.gzs == NULL) {
1327 			error = EFAULT;
1328 			goto done;
1329 		}
1330 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1331         }
1332 #endif
1333 
1334 	/*
1335 	 * Allocate memory for building the header, fill it up,
1336 	 * and write it out following the notes.
1337 	 */
1338 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1339 	if (hdr == NULL) {
1340 		error = EINVAL;
1341 		goto done;
1342 	}
1343 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1344 	    notesz);
1345 
1346 	/* Write the contents of all of the writable segments. */
1347 	if (error == 0) {
1348 		Elf_Phdr *php;
1349 		off_t offset;
1350 		int i;
1351 
1352 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1353 		offset = round_page(hdrsize + notesz);
1354 		for (i = 0; i < seginfo.count; i++) {
1355 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1356 			    php->p_filesz, offset, &params, tmpbuf);
1357 			if (error != 0)
1358 				break;
1359 			offset += php->p_filesz;
1360 			php++;
1361 		}
1362 #ifdef GZIO
1363 		if (error == 0 && compress)
1364 			error = gzio_flush(params.gzs);
1365 #endif
1366 	}
1367 	if (error) {
1368 		log(LOG_WARNING,
1369 		    "Failed to write core file for process %s (error %d)\n",
1370 		    curproc->p_comm, error);
1371 	}
1372 
1373 done:
1374 #ifdef GZIO
1375 	if (compress) {
1376 		free(tmpbuf, M_TEMP);
1377 		if (params.gzs != NULL)
1378 			gzio_fini(params.gzs);
1379 	}
1380 #endif
1381 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1382 		TAILQ_REMOVE(&notelst, ninfo, link);
1383 		free(ninfo, M_TEMP);
1384 	}
1385 	if (hdr != NULL)
1386 		free(hdr, M_TEMP);
1387 
1388 	return (error);
1389 }
1390 
1391 /*
1392  * A callback for each_writable_segment() to write out the segment's
1393  * program header entry.
1394  */
1395 static void
1396 cb_put_phdr(entry, closure)
1397 	vm_map_entry_t entry;
1398 	void *closure;
1399 {
1400 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1401 	Elf_Phdr *phdr = phc->phdr;
1402 
1403 	phc->offset = round_page(phc->offset);
1404 
1405 	phdr->p_type = PT_LOAD;
1406 	phdr->p_offset = phc->offset;
1407 	phdr->p_vaddr = entry->start;
1408 	phdr->p_paddr = 0;
1409 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1410 	phdr->p_align = PAGE_SIZE;
1411 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1412 
1413 	phc->offset += phdr->p_filesz;
1414 	phc->phdr++;
1415 }
1416 
1417 /*
1418  * A callback for each_writable_segment() to gather information about
1419  * the number of segments and their total size.
1420  */
1421 static void
1422 cb_size_segment(entry, closure)
1423 	vm_map_entry_t entry;
1424 	void *closure;
1425 {
1426 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1427 
1428 	ssc->count++;
1429 	ssc->size += entry->end - entry->start;
1430 }
1431 
1432 /*
1433  * For each writable segment in the process's memory map, call the given
1434  * function with a pointer to the map entry and some arbitrary
1435  * caller-supplied data.
1436  */
1437 static void
1438 each_writable_segment(td, func, closure)
1439 	struct thread *td;
1440 	segment_callback func;
1441 	void *closure;
1442 {
1443 	struct proc *p = td->td_proc;
1444 	vm_map_t map = &p->p_vmspace->vm_map;
1445 	vm_map_entry_t entry;
1446 	vm_object_t backing_object, object;
1447 	boolean_t ignore_entry;
1448 
1449 	vm_map_lock_read(map);
1450 	for (entry = map->header.next; entry != &map->header;
1451 	    entry = entry->next) {
1452 		/*
1453 		 * Don't dump inaccessible mappings, deal with legacy
1454 		 * coredump mode.
1455 		 *
1456 		 * Note that read-only segments related to the elf binary
1457 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1458 		 * need to arbitrarily ignore such segments.
1459 		 */
1460 		if (elf_legacy_coredump) {
1461 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1462 				continue;
1463 		} else {
1464 			if ((entry->protection & VM_PROT_ALL) == 0)
1465 				continue;
1466 		}
1467 
1468 		/*
1469 		 * Dont include memory segment in the coredump if
1470 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1471 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1472 		 * kernel map).
1473 		 */
1474 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1475 			continue;
1476 
1477 		if ((object = entry->object.vm_object) == NULL)
1478 			continue;
1479 
1480 		/* Ignore memory-mapped devices and such things. */
1481 		VM_OBJECT_RLOCK(object);
1482 		while ((backing_object = object->backing_object) != NULL) {
1483 			VM_OBJECT_RLOCK(backing_object);
1484 			VM_OBJECT_RUNLOCK(object);
1485 			object = backing_object;
1486 		}
1487 		ignore_entry = object->type != OBJT_DEFAULT &&
1488 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1489 		    object->type != OBJT_PHYS;
1490 		VM_OBJECT_RUNLOCK(object);
1491 		if (ignore_entry)
1492 			continue;
1493 
1494 		(*func)(entry, closure);
1495 	}
1496 	vm_map_unlock_read(map);
1497 }
1498 
1499 /*
1500  * Write the core file header to the file, including padding up to
1501  * the page boundary.
1502  */
1503 static int
1504 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1505     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1506 {
1507 	struct note_info *ninfo;
1508 	struct sbuf *sb;
1509 	int error;
1510 
1511 	/* Fill in the header. */
1512 	bzero(hdr, hdrsize);
1513 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1514 
1515 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1516 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1517 	sbuf_start_section(sb, NULL);
1518 	sbuf_bcat(sb, hdr, hdrsize);
1519 	TAILQ_FOREACH(ninfo, notelst, link)
1520 	    __elfN(putnote)(ninfo, sb);
1521 	/* Align up to a page boundary for the program segments. */
1522 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1523 	error = sbuf_finish(sb);
1524 	sbuf_delete(sb);
1525 
1526 	return (error);
1527 }
1528 
1529 static void
1530 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1531     size_t *sizep)
1532 {
1533 	struct proc *p;
1534 	struct thread *thr;
1535 	size_t size;
1536 
1537 	p = td->td_proc;
1538 	size = 0;
1539 
1540 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1541 
1542 	/*
1543 	 * To have the debugger select the right thread (LWP) as the initial
1544 	 * thread, we dump the state of the thread passed to us in td first.
1545 	 * This is the thread that causes the core dump and thus likely to
1546 	 * be the right thread one wants to have selected in the debugger.
1547 	 */
1548 	thr = td;
1549 	while (thr != NULL) {
1550 		size += register_note(list, NT_PRSTATUS,
1551 		    __elfN(note_prstatus), thr);
1552 		size += register_note(list, NT_FPREGSET,
1553 		    __elfN(note_fpregset), thr);
1554 		size += register_note(list, NT_THRMISC,
1555 		    __elfN(note_thrmisc), thr);
1556 		size += register_note(list, -1,
1557 		    __elfN(note_threadmd), thr);
1558 
1559 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1560 		    TAILQ_NEXT(thr, td_plist);
1561 		if (thr == td)
1562 			thr = TAILQ_NEXT(thr, td_plist);
1563 	}
1564 
1565 	size += register_note(list, NT_PROCSTAT_PROC,
1566 	    __elfN(note_procstat_proc), p);
1567 	size += register_note(list, NT_PROCSTAT_FILES,
1568 	    note_procstat_files, p);
1569 	size += register_note(list, NT_PROCSTAT_VMMAP,
1570 	    note_procstat_vmmap, p);
1571 	size += register_note(list, NT_PROCSTAT_GROUPS,
1572 	    note_procstat_groups, p);
1573 	size += register_note(list, NT_PROCSTAT_UMASK,
1574 	    note_procstat_umask, p);
1575 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1576 	    note_procstat_rlimit, p);
1577 	size += register_note(list, NT_PROCSTAT_OSREL,
1578 	    note_procstat_osrel, p);
1579 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1580 	    __elfN(note_procstat_psstrings), p);
1581 	size += register_note(list, NT_PROCSTAT_AUXV,
1582 	    __elfN(note_procstat_auxv), p);
1583 
1584 	*sizep = size;
1585 }
1586 
1587 static void
1588 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1589     size_t notesz)
1590 {
1591 	Elf_Ehdr *ehdr;
1592 	Elf_Phdr *phdr;
1593 	struct phdr_closure phc;
1594 
1595 	ehdr = (Elf_Ehdr *)hdr;
1596 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1597 
1598 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1599 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1600 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1601 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1602 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1603 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1604 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1605 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1606 	ehdr->e_ident[EI_ABIVERSION] = 0;
1607 	ehdr->e_ident[EI_PAD] = 0;
1608 	ehdr->e_type = ET_CORE;
1609 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1610 	ehdr->e_machine = ELF_ARCH32;
1611 #else
1612 	ehdr->e_machine = ELF_ARCH;
1613 #endif
1614 	ehdr->e_version = EV_CURRENT;
1615 	ehdr->e_entry = 0;
1616 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1617 	ehdr->e_flags = 0;
1618 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1619 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1620 	ehdr->e_phnum = numsegs + 1;
1621 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1622 	ehdr->e_shnum = 0;
1623 	ehdr->e_shstrndx = SHN_UNDEF;
1624 
1625 	/*
1626 	 * Fill in the program header entries.
1627 	 */
1628 
1629 	/* The note segement. */
1630 	phdr->p_type = PT_NOTE;
1631 	phdr->p_offset = hdrsize;
1632 	phdr->p_vaddr = 0;
1633 	phdr->p_paddr = 0;
1634 	phdr->p_filesz = notesz;
1635 	phdr->p_memsz = 0;
1636 	phdr->p_flags = PF_R;
1637 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1638 	phdr++;
1639 
1640 	/* All the writable segments from the program. */
1641 	phc.phdr = phdr;
1642 	phc.offset = round_page(hdrsize + notesz);
1643 	each_writable_segment(td, cb_put_phdr, &phc);
1644 }
1645 
1646 static size_t
1647 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1648 {
1649 	struct note_info *ninfo;
1650 	size_t size, notesize;
1651 
1652 	size = 0;
1653 	out(arg, NULL, &size);
1654 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1655 	ninfo->type = type;
1656 	ninfo->outfunc = out;
1657 	ninfo->outarg = arg;
1658 	ninfo->outsize = size;
1659 	TAILQ_INSERT_TAIL(list, ninfo, link);
1660 
1661 	if (type == -1)
1662 		return (size);
1663 
1664 	notesize = sizeof(Elf_Note) +		/* note header */
1665 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1666 						/* note name */
1667 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1668 
1669 	return (notesize);
1670 }
1671 
1672 static size_t
1673 append_note_data(const void *src, void *dst, size_t len)
1674 {
1675 	size_t padded_len;
1676 
1677 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1678 	if (dst != NULL) {
1679 		bcopy(src, dst, len);
1680 		bzero((char *)dst + len, padded_len - len);
1681 	}
1682 	return (padded_len);
1683 }
1684 
1685 size_t
1686 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1687 {
1688 	Elf_Note *note;
1689 	char *buf;
1690 	size_t notesize;
1691 
1692 	buf = dst;
1693 	if (buf != NULL) {
1694 		note = (Elf_Note *)buf;
1695 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1696 		note->n_descsz = size;
1697 		note->n_type = type;
1698 		buf += sizeof(*note);
1699 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1700 		    sizeof(FREEBSD_ABI_VENDOR));
1701 		append_note_data(src, buf, size);
1702 		if (descp != NULL)
1703 			*descp = buf;
1704 	}
1705 
1706 	notesize = sizeof(Elf_Note) +		/* note header */
1707 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1708 						/* note name */
1709 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1710 
1711 	return (notesize);
1712 }
1713 
1714 static void
1715 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1716 {
1717 	Elf_Note note;
1718 	ssize_t old_len, sect_len;
1719 	size_t new_len, descsz, i;
1720 
1721 	if (ninfo->type == -1) {
1722 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1723 		return;
1724 	}
1725 
1726 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1727 	note.n_descsz = ninfo->outsize;
1728 	note.n_type = ninfo->type;
1729 
1730 	sbuf_bcat(sb, &note, sizeof(note));
1731 	sbuf_start_section(sb, &old_len);
1732 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1733 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1734 	if (note.n_descsz == 0)
1735 		return;
1736 	sbuf_start_section(sb, &old_len);
1737 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1738 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1739 	if (sect_len < 0)
1740 		return;
1741 
1742 	new_len = (size_t)sect_len;
1743 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1744 	if (new_len < descsz) {
1745 		/*
1746 		 * It is expected that individual note emitters will correctly
1747 		 * predict their expected output size and fill up to that size
1748 		 * themselves, padding in a format-specific way if needed.
1749 		 * However, in case they don't, just do it here with zeros.
1750 		 */
1751 		for (i = 0; i < descsz - new_len; i++)
1752 			sbuf_putc(sb, 0);
1753 	} else if (new_len > descsz) {
1754 		/*
1755 		 * We can't always truncate sb -- we may have drained some
1756 		 * of it already.
1757 		 */
1758 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1759 		    "read it (%zu > %zu).  Since it is longer than "
1760 		    "expected, this coredump's notes are corrupt.  THIS "
1761 		    "IS A BUG in the note_procstat routine for type %u.\n",
1762 		    __func__, (unsigned)note.n_type, new_len, descsz,
1763 		    (unsigned)note.n_type));
1764 	}
1765 }
1766 
1767 /*
1768  * Miscellaneous note out functions.
1769  */
1770 
1771 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1772 #include <compat/freebsd32/freebsd32.h>
1773 
1774 typedef struct prstatus32 elf_prstatus_t;
1775 typedef struct prpsinfo32 elf_prpsinfo_t;
1776 typedef struct fpreg32 elf_prfpregset_t;
1777 typedef struct fpreg32 elf_fpregset_t;
1778 typedef struct reg32 elf_gregset_t;
1779 typedef struct thrmisc32 elf_thrmisc_t;
1780 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1781 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1782 typedef uint32_t elf_ps_strings_t;
1783 #else
1784 typedef prstatus_t elf_prstatus_t;
1785 typedef prpsinfo_t elf_prpsinfo_t;
1786 typedef prfpregset_t elf_prfpregset_t;
1787 typedef prfpregset_t elf_fpregset_t;
1788 typedef gregset_t elf_gregset_t;
1789 typedef thrmisc_t elf_thrmisc_t;
1790 #define ELF_KERN_PROC_MASK	0
1791 typedef struct kinfo_proc elf_kinfo_proc_t;
1792 typedef vm_offset_t elf_ps_strings_t;
1793 #endif
1794 
1795 static void
1796 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1797 {
1798 	struct proc *p;
1799 	elf_prpsinfo_t *psinfo;
1800 
1801 	p = (struct proc *)arg;
1802 	if (sb != NULL) {
1803 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1804 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1805 		psinfo->pr_version = PRPSINFO_VERSION;
1806 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1807 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1808 		/*
1809 		 * XXX - We don't fill in the command line arguments properly
1810 		 * yet.
1811 		 */
1812 		strlcpy(psinfo->pr_psargs, p->p_comm,
1813 		    sizeof(psinfo->pr_psargs));
1814 
1815 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1816 		free(psinfo, M_TEMP);
1817 	}
1818 	*sizep = sizeof(*psinfo);
1819 }
1820 
1821 static void
1822 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1823 {
1824 	struct thread *td;
1825 	elf_prstatus_t *status;
1826 
1827 	td = (struct thread *)arg;
1828 	if (sb != NULL) {
1829 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1830 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1831 		status->pr_version = PRSTATUS_VERSION;
1832 		status->pr_statussz = sizeof(elf_prstatus_t);
1833 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1834 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1835 		status->pr_osreldate = osreldate;
1836 		status->pr_cursig = td->td_proc->p_sig;
1837 		status->pr_pid = td->td_tid;
1838 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1839 		fill_regs32(td, &status->pr_reg);
1840 #else
1841 		fill_regs(td, &status->pr_reg);
1842 #endif
1843 		sbuf_bcat(sb, status, sizeof(*status));
1844 		free(status, M_TEMP);
1845 	}
1846 	*sizep = sizeof(*status);
1847 }
1848 
1849 static void
1850 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1851 {
1852 	struct thread *td;
1853 	elf_prfpregset_t *fpregset;
1854 
1855 	td = (struct thread *)arg;
1856 	if (sb != NULL) {
1857 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1858 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1859 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1860 		fill_fpregs32(td, fpregset);
1861 #else
1862 		fill_fpregs(td, fpregset);
1863 #endif
1864 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1865 		free(fpregset, M_TEMP);
1866 	}
1867 	*sizep = sizeof(*fpregset);
1868 }
1869 
1870 static void
1871 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1872 {
1873 	struct thread *td;
1874 	elf_thrmisc_t thrmisc;
1875 
1876 	td = (struct thread *)arg;
1877 	if (sb != NULL) {
1878 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1879 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1880 		strcpy(thrmisc.pr_tname, td->td_name);
1881 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1882 	}
1883 	*sizep = sizeof(thrmisc);
1884 }
1885 
1886 /*
1887  * Allow for MD specific notes, as well as any MD
1888  * specific preparations for writing MI notes.
1889  */
1890 static void
1891 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1892 {
1893 	struct thread *td;
1894 	void *buf;
1895 	size_t size;
1896 
1897 	td = (struct thread *)arg;
1898 	size = *sizep;
1899 	if (size != 0 && sb != NULL)
1900 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1901 	else
1902 		buf = NULL;
1903 	size = 0;
1904 	__elfN(dump_thread)(td, buf, &size);
1905 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1906 	if (size != 0 && sb != NULL)
1907 		sbuf_bcat(sb, buf, size);
1908 	free(buf, M_TEMP);
1909 	*sizep = size;
1910 }
1911 
1912 #ifdef KINFO_PROC_SIZE
1913 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1914 #endif
1915 
1916 static void
1917 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1918 {
1919 	struct proc *p;
1920 	size_t size;
1921 	int structsize;
1922 
1923 	p = (struct proc *)arg;
1924 	size = sizeof(structsize) + p->p_numthreads *
1925 	    sizeof(elf_kinfo_proc_t);
1926 
1927 	if (sb != NULL) {
1928 		KASSERT(*sizep == size, ("invalid size"));
1929 		structsize = sizeof(elf_kinfo_proc_t);
1930 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1931 		sx_slock(&proctree_lock);
1932 		PROC_LOCK(p);
1933 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1934 		sx_sunlock(&proctree_lock);
1935 	}
1936 	*sizep = size;
1937 }
1938 
1939 #ifdef KINFO_FILE_SIZE
1940 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1941 #endif
1942 
1943 static void
1944 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1945 {
1946 	struct proc *p;
1947 	size_t size, sect_sz, i;
1948 	ssize_t start_len, sect_len;
1949 	int structsize, filedesc_flags;
1950 
1951 	if (coredump_pack_fileinfo)
1952 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1953 	else
1954 		filedesc_flags = 0;
1955 
1956 	p = (struct proc *)arg;
1957 	structsize = sizeof(struct kinfo_file);
1958 	if (sb == NULL) {
1959 		size = 0;
1960 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1961 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1962 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1963 		PROC_LOCK(p);
1964 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1965 		sbuf_finish(sb);
1966 		sbuf_delete(sb);
1967 		*sizep = size;
1968 	} else {
1969 		sbuf_start_section(sb, &start_len);
1970 
1971 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1972 		PROC_LOCK(p);
1973 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1974 		    filedesc_flags);
1975 
1976 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
1977 		if (sect_len < 0)
1978 			return;
1979 		sect_sz = sect_len;
1980 
1981 		KASSERT(sect_sz <= *sizep,
1982 		    ("kern_proc_filedesc_out did not respect maxlen; "
1983 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
1984 		     sect_sz - sizeof(structsize)));
1985 
1986 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
1987 			sbuf_putc(sb, 0);
1988 	}
1989 }
1990 
1991 #ifdef KINFO_VMENTRY_SIZE
1992 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1993 #endif
1994 
1995 static void
1996 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1997 {
1998 	struct proc *p;
1999 	size_t size;
2000 	int structsize, vmmap_flags;
2001 
2002 	if (coredump_pack_vmmapinfo)
2003 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2004 	else
2005 		vmmap_flags = 0;
2006 
2007 	p = (struct proc *)arg;
2008 	structsize = sizeof(struct kinfo_vmentry);
2009 	if (sb == NULL) {
2010 		size = 0;
2011 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2012 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2013 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2014 		PROC_LOCK(p);
2015 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2016 		sbuf_finish(sb);
2017 		sbuf_delete(sb);
2018 		*sizep = size;
2019 	} else {
2020 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2021 		PROC_LOCK(p);
2022 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2023 		    vmmap_flags);
2024 	}
2025 }
2026 
2027 static void
2028 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2029 {
2030 	struct proc *p;
2031 	size_t size;
2032 	int structsize;
2033 
2034 	p = (struct proc *)arg;
2035 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2036 	if (sb != NULL) {
2037 		KASSERT(*sizep == size, ("invalid size"));
2038 		structsize = sizeof(gid_t);
2039 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2040 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2041 		    sizeof(gid_t));
2042 	}
2043 	*sizep = size;
2044 }
2045 
2046 static void
2047 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2048 {
2049 	struct proc *p;
2050 	size_t size;
2051 	int structsize;
2052 
2053 	p = (struct proc *)arg;
2054 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2055 	if (sb != NULL) {
2056 		KASSERT(*sizep == size, ("invalid size"));
2057 		structsize = sizeof(p->p_fd->fd_cmask);
2058 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2059 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2060 	}
2061 	*sizep = size;
2062 }
2063 
2064 static void
2065 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2066 {
2067 	struct proc *p;
2068 	struct rlimit rlim[RLIM_NLIMITS];
2069 	size_t size;
2070 	int structsize, i;
2071 
2072 	p = (struct proc *)arg;
2073 	size = sizeof(structsize) + sizeof(rlim);
2074 	if (sb != NULL) {
2075 		KASSERT(*sizep == size, ("invalid size"));
2076 		structsize = sizeof(rlim);
2077 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2078 		PROC_LOCK(p);
2079 		for (i = 0; i < RLIM_NLIMITS; i++)
2080 			lim_rlimit_proc(p, i, &rlim[i]);
2081 		PROC_UNLOCK(p);
2082 		sbuf_bcat(sb, rlim, sizeof(rlim));
2083 	}
2084 	*sizep = size;
2085 }
2086 
2087 static void
2088 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2089 {
2090 	struct proc *p;
2091 	size_t size;
2092 	int structsize;
2093 
2094 	p = (struct proc *)arg;
2095 	size = sizeof(structsize) + sizeof(p->p_osrel);
2096 	if (sb != NULL) {
2097 		KASSERT(*sizep == size, ("invalid size"));
2098 		structsize = sizeof(p->p_osrel);
2099 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2100 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2101 	}
2102 	*sizep = size;
2103 }
2104 
2105 static void
2106 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2107 {
2108 	struct proc *p;
2109 	elf_ps_strings_t ps_strings;
2110 	size_t size;
2111 	int structsize;
2112 
2113 	p = (struct proc *)arg;
2114 	size = sizeof(structsize) + sizeof(ps_strings);
2115 	if (sb != NULL) {
2116 		KASSERT(*sizep == size, ("invalid size"));
2117 		structsize = sizeof(ps_strings);
2118 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2119 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2120 #else
2121 		ps_strings = p->p_sysent->sv_psstrings;
2122 #endif
2123 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2124 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2125 	}
2126 	*sizep = size;
2127 }
2128 
2129 static void
2130 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2131 {
2132 	struct proc *p;
2133 	size_t size;
2134 	int structsize;
2135 
2136 	p = (struct proc *)arg;
2137 	if (sb == NULL) {
2138 		size = 0;
2139 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2140 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2141 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2142 		PHOLD(p);
2143 		proc_getauxv(curthread, p, sb);
2144 		PRELE(p);
2145 		sbuf_finish(sb);
2146 		sbuf_delete(sb);
2147 		*sizep = size;
2148 	} else {
2149 		structsize = sizeof(Elf_Auxinfo);
2150 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2151 		PHOLD(p);
2152 		proc_getauxv(curthread, p, sb);
2153 		PRELE(p);
2154 	}
2155 }
2156 
2157 static boolean_t
2158 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2159     int32_t *osrel, const Elf_Phdr *pnote)
2160 {
2161 	const Elf_Note *note, *note0, *note_end;
2162 	const char *note_name;
2163 	char *buf;
2164 	int i, error;
2165 	boolean_t res;
2166 
2167 	/* We need some limit, might as well use PAGE_SIZE. */
2168 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2169 		return (FALSE);
2170 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2171 	if (pnote->p_offset > PAGE_SIZE ||
2172 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2173 		VOP_UNLOCK(imgp->vp, 0);
2174 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2175 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2176 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2177 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2178 		    curthread->td_ucred, NOCRED, NULL, curthread);
2179 		if (error != 0) {
2180 			uprintf("i/o error PT_NOTE\n");
2181 			res = FALSE;
2182 			goto ret;
2183 		}
2184 		note = note0 = (const Elf_Note *)buf;
2185 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2186 	} else {
2187 		note = note0 = (const Elf_Note *)(imgp->image_header +
2188 		    pnote->p_offset);
2189 		note_end = (const Elf_Note *)(imgp->image_header +
2190 		    pnote->p_offset + pnote->p_filesz);
2191 		buf = NULL;
2192 	}
2193 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2194 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2195 		    (const char *)note < sizeof(Elf_Note)) {
2196 			res = FALSE;
2197 			goto ret;
2198 		}
2199 		if (note->n_namesz != checknote->hdr.n_namesz ||
2200 		    note->n_descsz != checknote->hdr.n_descsz ||
2201 		    note->n_type != checknote->hdr.n_type)
2202 			goto nextnote;
2203 		note_name = (const char *)(note + 1);
2204 		if (note_name + checknote->hdr.n_namesz >=
2205 		    (const char *)note_end || strncmp(checknote->vendor,
2206 		    note_name, checknote->hdr.n_namesz) != 0)
2207 			goto nextnote;
2208 
2209 		/*
2210 		 * Fetch the osreldate for binary
2211 		 * from the ELF OSABI-note if necessary.
2212 		 */
2213 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2214 		    checknote->trans_osrel != NULL) {
2215 			res = checknote->trans_osrel(note, osrel);
2216 			goto ret;
2217 		}
2218 		res = TRUE;
2219 		goto ret;
2220 nextnote:
2221 		note = (const Elf_Note *)((const char *)(note + 1) +
2222 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2223 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2224 	}
2225 	res = FALSE;
2226 ret:
2227 	free(buf, M_TEMP);
2228 	return (res);
2229 }
2230 
2231 /*
2232  * Try to find the appropriate ABI-note section for checknote,
2233  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2234  * first page of the image is searched, the same as for headers.
2235  */
2236 static boolean_t
2237 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2238     int32_t *osrel)
2239 {
2240 	const Elf_Phdr *phdr;
2241 	const Elf_Ehdr *hdr;
2242 	int i;
2243 
2244 	hdr = (const Elf_Ehdr *)imgp->image_header;
2245 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2246 
2247 	for (i = 0; i < hdr->e_phnum; i++) {
2248 		if (phdr[i].p_type == PT_NOTE &&
2249 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2250 			return (TRUE);
2251 	}
2252 	return (FALSE);
2253 
2254 }
2255 
2256 /*
2257  * Tell kern_execve.c about it, with a little help from the linker.
2258  */
2259 static struct execsw __elfN(execsw) = {
2260 	__CONCAT(exec_, __elfN(imgact)),
2261 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2262 };
2263 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2264 
2265 static vm_prot_t
2266 __elfN(trans_prot)(Elf_Word flags)
2267 {
2268 	vm_prot_t prot;
2269 
2270 	prot = 0;
2271 	if (flags & PF_X)
2272 		prot |= VM_PROT_EXECUTE;
2273 	if (flags & PF_W)
2274 		prot |= VM_PROT_WRITE;
2275 	if (flags & PF_R)
2276 		prot |= VM_PROT_READ;
2277 #if __ELF_WORD_SIZE == 32
2278 #if defined(__amd64__)
2279 	if (i386_read_exec && (flags & PF_R))
2280 		prot |= VM_PROT_EXECUTE;
2281 #endif
2282 #endif
2283 	return (prot);
2284 }
2285 
2286 static Elf_Word
2287 __elfN(untrans_prot)(vm_prot_t prot)
2288 {
2289 	Elf_Word flags;
2290 
2291 	flags = 0;
2292 	if (prot & VM_PROT_EXECUTE)
2293 		flags |= PF_X;
2294 	if (prot & VM_PROT_READ)
2295 		flags |= PF_R;
2296 	if (prot & VM_PROT_WRITE)
2297 		flags |= PF_W;
2298 	return (flags);
2299 }
2300