xref: /freebsd/sys/kern/imgact_elf.c (revision 38f0b757fd84d17d0fc24739a7cda160c4516d81)
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_core.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/imgact.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mman.h>
49 #include <sys/namei.h>
50 #include <sys/pioctl.h>
51 #include <sys/proc.h>
52 #include <sys/procfs.h>
53 #include <sys/racct.h>
54 #include <sys/resourcevar.h>
55 #include <sys/rwlock.h>
56 #include <sys/sbuf.h>
57 #include <sys/sf_buf.h>
58 #include <sys/smp.h>
59 #include <sys/systm.h>
60 #include <sys/signalvar.h>
61 #include <sys/stat.h>
62 #include <sys/sx.h>
63 #include <sys/syscall.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/vnode.h>
67 #include <sys/syslog.h>
68 #include <sys/eventhandler.h>
69 #include <sys/user.h>
70 
71 #include <net/zlib.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 #ifdef COMPRESS_USER_CORES
108 static int compress_core(gzFile, char *, char *, unsigned int,
109     struct thread * td);
110 #endif
111 #define CORE_BUF_SIZE	(16 * 1024)
112 
113 int __elfN(fallback_brand) = -1;
114 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
115     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
116     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
118     &__elfN(fallback_brand));
119 
120 static int elf_legacy_coredump = 0;
121 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
122     &elf_legacy_coredump, 0, "");
123 
124 int __elfN(nxstack) =
125 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
126 	1;
127 #else
128 	0;
129 #endif
130 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
131     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
132     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
133 
134 #if __ELF_WORD_SIZE == 32
135 #if defined(__amd64__) || defined(__ia64__)
136 int i386_read_exec = 0;
137 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
138     "enable execution from readable segments");
139 #endif
140 #endif
141 
142 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
143 
144 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
145 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
146 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
147 
148 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
149 
150 Elf_Brandnote __elfN(freebsd_brandnote) = {
151 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
152 	.hdr.n_descsz	= sizeof(int32_t),
153 	.hdr.n_type	= 1,
154 	.vendor		= FREEBSD_ABI_VENDOR,
155 	.flags		= BN_TRANSLATE_OSREL,
156 	.trans_osrel	= __elfN(freebsd_trans_osrel)
157 };
158 
159 static boolean_t
160 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
161 {
162 	uintptr_t p;
163 
164 	p = (uintptr_t)(note + 1);
165 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
166 	*osrel = *(const int32_t *)(p);
167 
168 	return (TRUE);
169 }
170 
171 static const char GNU_ABI_VENDOR[] = "GNU";
172 static int GNU_KFREEBSD_ABI_DESC = 3;
173 
174 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
175 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
176 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
177 	.hdr.n_type	= 1,
178 	.vendor		= GNU_ABI_VENDOR,
179 	.flags		= BN_TRANSLATE_OSREL,
180 	.trans_osrel	= kfreebsd_trans_osrel
181 };
182 
183 static boolean_t
184 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
185 {
186 	const Elf32_Word *desc;
187 	uintptr_t p;
188 
189 	p = (uintptr_t)(note + 1);
190 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
191 
192 	desc = (const Elf32_Word *)p;
193 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
194 		return (FALSE);
195 
196 	/*
197 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
198 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
199 	 */
200 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
201 
202 	return (TRUE);
203 }
204 
205 int
206 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
207 {
208 	int i;
209 
210 	for (i = 0; i < MAX_BRANDS; i++) {
211 		if (elf_brand_list[i] == NULL) {
212 			elf_brand_list[i] = entry;
213 			break;
214 		}
215 	}
216 	if (i == MAX_BRANDS) {
217 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
218 			__func__, entry);
219 		return (-1);
220 	}
221 	return (0);
222 }
223 
224 int
225 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
226 {
227 	int i;
228 
229 	for (i = 0; i < MAX_BRANDS; i++) {
230 		if (elf_brand_list[i] == entry) {
231 			elf_brand_list[i] = NULL;
232 			break;
233 		}
234 	}
235 	if (i == MAX_BRANDS)
236 		return (-1);
237 	return (0);
238 }
239 
240 int
241 __elfN(brand_inuse)(Elf_Brandinfo *entry)
242 {
243 	struct proc *p;
244 	int rval = FALSE;
245 
246 	sx_slock(&allproc_lock);
247 	FOREACH_PROC_IN_SYSTEM(p) {
248 		if (p->p_sysent == entry->sysvec) {
249 			rval = TRUE;
250 			break;
251 		}
252 	}
253 	sx_sunlock(&allproc_lock);
254 
255 	return (rval);
256 }
257 
258 static Elf_Brandinfo *
259 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
260     int interp_name_len, int32_t *osrel)
261 {
262 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
263 	Elf_Brandinfo *bi;
264 	boolean_t ret;
265 	int i;
266 
267 	/*
268 	 * We support four types of branding -- (1) the ELF EI_OSABI field
269 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
270 	 * branding w/in the ELF header, (3) path of the `interp_path'
271 	 * field, and (4) the ".note.ABI-tag" ELF section.
272 	 */
273 
274 	/* Look for an ".note.ABI-tag" ELF section */
275 	for (i = 0; i < MAX_BRANDS; i++) {
276 		bi = elf_brand_list[i];
277 		if (bi == NULL)
278 			continue;
279 		if (hdr->e_machine == bi->machine && (bi->flags &
280 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
281 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
282 			if (ret)
283 				return (bi);
284 		}
285 	}
286 
287 	/* If the executable has a brand, search for it in the brand list. */
288 	for (i = 0; i < MAX_BRANDS; i++) {
289 		bi = elf_brand_list[i];
290 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
291 			continue;
292 		if (hdr->e_machine == bi->machine &&
293 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
294 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
295 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
296 			return (bi);
297 	}
298 
299 	/* Lacking a known brand, search for a recognized interpreter. */
300 	if (interp != NULL) {
301 		for (i = 0; i < MAX_BRANDS; i++) {
302 			bi = elf_brand_list[i];
303 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
304 				continue;
305 			if (hdr->e_machine == bi->machine &&
306 			    /* ELF image p_filesz includes terminating zero */
307 			    strlen(bi->interp_path) + 1 == interp_name_len &&
308 			    strncmp(interp, bi->interp_path, interp_name_len)
309 			    == 0)
310 				return (bi);
311 		}
312 	}
313 
314 	/* Lacking a recognized interpreter, try the default brand */
315 	for (i = 0; i < MAX_BRANDS; i++) {
316 		bi = elf_brand_list[i];
317 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
318 			continue;
319 		if (hdr->e_machine == bi->machine &&
320 		    __elfN(fallback_brand) == bi->brand)
321 			return (bi);
322 	}
323 	return (NULL);
324 }
325 
326 static int
327 __elfN(check_header)(const Elf_Ehdr *hdr)
328 {
329 	Elf_Brandinfo *bi;
330 	int i;
331 
332 	if (!IS_ELF(*hdr) ||
333 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
334 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
335 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
336 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
337 	    hdr->e_version != ELF_TARG_VER)
338 		return (ENOEXEC);
339 
340 	/*
341 	 * Make sure we have at least one brand for this machine.
342 	 */
343 
344 	for (i = 0; i < MAX_BRANDS; i++) {
345 		bi = elf_brand_list[i];
346 		if (bi != NULL && bi->machine == hdr->e_machine)
347 			break;
348 	}
349 	if (i == MAX_BRANDS)
350 		return (ENOEXEC);
351 
352 	return (0);
353 }
354 
355 static int
356 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
357     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
358 {
359 	struct sf_buf *sf;
360 	int error;
361 	vm_offset_t off;
362 
363 	/*
364 	 * Create the page if it doesn't exist yet. Ignore errors.
365 	 */
366 	vm_map_lock(map);
367 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
368 	    VM_PROT_ALL, VM_PROT_ALL, 0);
369 	vm_map_unlock(map);
370 
371 	/*
372 	 * Find the page from the underlying object.
373 	 */
374 	if (object) {
375 		sf = vm_imgact_map_page(object, offset);
376 		if (sf == NULL)
377 			return (KERN_FAILURE);
378 		off = offset - trunc_page(offset);
379 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
380 		    end - start);
381 		vm_imgact_unmap_page(sf);
382 		if (error) {
383 			return (KERN_FAILURE);
384 		}
385 	}
386 
387 	return (KERN_SUCCESS);
388 }
389 
390 static int
391 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
392     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
393 {
394 	struct sf_buf *sf;
395 	vm_offset_t off;
396 	vm_size_t sz;
397 	int error, rv;
398 
399 	if (start != trunc_page(start)) {
400 		rv = __elfN(map_partial)(map, object, offset, start,
401 		    round_page(start), prot);
402 		if (rv)
403 			return (rv);
404 		offset += round_page(start) - start;
405 		start = round_page(start);
406 	}
407 	if (end != round_page(end)) {
408 		rv = __elfN(map_partial)(map, object, offset +
409 		    trunc_page(end) - start, trunc_page(end), end, prot);
410 		if (rv)
411 			return (rv);
412 		end = trunc_page(end);
413 	}
414 	if (end > start) {
415 		if (offset & PAGE_MASK) {
416 			/*
417 			 * The mapping is not page aligned. This means we have
418 			 * to copy the data. Sigh.
419 			 */
420 			rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
421 			    VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
422 			    0);
423 			if (rv)
424 				return (rv);
425 			if (object == NULL)
426 				return (KERN_SUCCESS);
427 			for (; start < end; start += sz) {
428 				sf = vm_imgact_map_page(object, offset);
429 				if (sf == NULL)
430 					return (KERN_FAILURE);
431 				off = offset - trunc_page(offset);
432 				sz = end - start;
433 				if (sz > PAGE_SIZE - off)
434 					sz = PAGE_SIZE - off;
435 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
436 				    (caddr_t)start, sz);
437 				vm_imgact_unmap_page(sf);
438 				if (error) {
439 					return (KERN_FAILURE);
440 				}
441 				offset += sz;
442 			}
443 			rv = KERN_SUCCESS;
444 		} else {
445 			vm_object_reference(object);
446 			vm_map_lock(map);
447 			rv = vm_map_insert(map, object, offset, start, end,
448 			    prot, VM_PROT_ALL, cow);
449 			vm_map_unlock(map);
450 			if (rv != KERN_SUCCESS)
451 				vm_object_deallocate(object);
452 		}
453 		return (rv);
454 	} else {
455 		return (KERN_SUCCESS);
456 	}
457 }
458 
459 static int
460 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
461     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
462     size_t pagesize)
463 {
464 	struct sf_buf *sf;
465 	size_t map_len;
466 	vm_map_t map;
467 	vm_object_t object;
468 	vm_offset_t map_addr;
469 	int error, rv, cow;
470 	size_t copy_len;
471 	vm_offset_t file_addr;
472 
473 	/*
474 	 * It's necessary to fail if the filsz + offset taken from the
475 	 * header is greater than the actual file pager object's size.
476 	 * If we were to allow this, then the vm_map_find() below would
477 	 * walk right off the end of the file object and into the ether.
478 	 *
479 	 * While I'm here, might as well check for something else that
480 	 * is invalid: filsz cannot be greater than memsz.
481 	 */
482 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
483 		uprintf("elf_load_section: truncated ELF file\n");
484 		return (ENOEXEC);
485 	}
486 
487 	object = imgp->object;
488 	map = &imgp->proc->p_vmspace->vm_map;
489 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
490 	file_addr = trunc_page_ps(offset, pagesize);
491 
492 	/*
493 	 * We have two choices.  We can either clear the data in the last page
494 	 * of an oversized mapping, or we can start the anon mapping a page
495 	 * early and copy the initialized data into that first page.  We
496 	 * choose the second..
497 	 */
498 	if (memsz > filsz)
499 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
500 	else
501 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
502 
503 	if (map_len != 0) {
504 		/* cow flags: don't dump readonly sections in core */
505 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
506 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
507 
508 		rv = __elfN(map_insert)(map,
509 				      object,
510 				      file_addr,	/* file offset */
511 				      map_addr,		/* virtual start */
512 				      map_addr + map_len,/* virtual end */
513 				      prot,
514 				      cow);
515 		if (rv != KERN_SUCCESS)
516 			return (EINVAL);
517 
518 		/* we can stop now if we've covered it all */
519 		if (memsz == filsz) {
520 			return (0);
521 		}
522 	}
523 
524 
525 	/*
526 	 * We have to get the remaining bit of the file into the first part
527 	 * of the oversized map segment.  This is normally because the .data
528 	 * segment in the file is extended to provide bss.  It's a neat idea
529 	 * to try and save a page, but it's a pain in the behind to implement.
530 	 */
531 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
532 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
533 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
534 	    map_addr;
535 
536 	/* This had damn well better be true! */
537 	if (map_len != 0) {
538 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
539 		    map_len, VM_PROT_ALL, 0);
540 		if (rv != KERN_SUCCESS) {
541 			return (EINVAL);
542 		}
543 	}
544 
545 	if (copy_len != 0) {
546 		vm_offset_t off;
547 
548 		sf = vm_imgact_map_page(object, offset + filsz);
549 		if (sf == NULL)
550 			return (EIO);
551 
552 		/* send the page fragment to user space */
553 		off = trunc_page_ps(offset + filsz, pagesize) -
554 		    trunc_page(offset + filsz);
555 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
556 		    (caddr_t)map_addr, copy_len);
557 		vm_imgact_unmap_page(sf);
558 		if (error) {
559 			return (error);
560 		}
561 	}
562 
563 	/*
564 	 * set it to the specified protection.
565 	 * XXX had better undo the damage from pasting over the cracks here!
566 	 */
567 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
568 	    map_len), prot, FALSE);
569 
570 	return (0);
571 }
572 
573 /*
574  * Load the file "file" into memory.  It may be either a shared object
575  * or an executable.
576  *
577  * The "addr" reference parameter is in/out.  On entry, it specifies
578  * the address where a shared object should be loaded.  If the file is
579  * an executable, this value is ignored.  On exit, "addr" specifies
580  * where the file was actually loaded.
581  *
582  * The "entry" reference parameter is out only.  On exit, it specifies
583  * the entry point for the loaded file.
584  */
585 static int
586 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
587 	u_long *entry, size_t pagesize)
588 {
589 	struct {
590 		struct nameidata nd;
591 		struct vattr attr;
592 		struct image_params image_params;
593 	} *tempdata;
594 	const Elf_Ehdr *hdr = NULL;
595 	const Elf_Phdr *phdr = NULL;
596 	struct nameidata *nd;
597 	struct vattr *attr;
598 	struct image_params *imgp;
599 	vm_prot_t prot;
600 	u_long rbase;
601 	u_long base_addr = 0;
602 	int error, i, numsegs;
603 
604 #ifdef CAPABILITY_MODE
605 	/*
606 	 * XXXJA: This check can go away once we are sufficiently confident
607 	 * that the checks in namei() are correct.
608 	 */
609 	if (IN_CAPABILITY_MODE(curthread))
610 		return (ECAPMODE);
611 #endif
612 
613 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
614 	nd = &tempdata->nd;
615 	attr = &tempdata->attr;
616 	imgp = &tempdata->image_params;
617 
618 	/*
619 	 * Initialize part of the common data
620 	 */
621 	imgp->proc = p;
622 	imgp->attr = attr;
623 	imgp->firstpage = NULL;
624 	imgp->image_header = NULL;
625 	imgp->object = NULL;
626 	imgp->execlabel = NULL;
627 
628 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
629 	if ((error = namei(nd)) != 0) {
630 		nd->ni_vp = NULL;
631 		goto fail;
632 	}
633 	NDFREE(nd, NDF_ONLY_PNBUF);
634 	imgp->vp = nd->ni_vp;
635 
636 	/*
637 	 * Check permissions, modes, uid, etc on the file, and "open" it.
638 	 */
639 	error = exec_check_permissions(imgp);
640 	if (error)
641 		goto fail;
642 
643 	error = exec_map_first_page(imgp);
644 	if (error)
645 		goto fail;
646 
647 	/*
648 	 * Also make certain that the interpreter stays the same, so set
649 	 * its VV_TEXT flag, too.
650 	 */
651 	VOP_SET_TEXT(nd->ni_vp);
652 
653 	imgp->object = nd->ni_vp->v_object;
654 
655 	hdr = (const Elf_Ehdr *)imgp->image_header;
656 	if ((error = __elfN(check_header)(hdr)) != 0)
657 		goto fail;
658 	if (hdr->e_type == ET_DYN)
659 		rbase = *addr;
660 	else if (hdr->e_type == ET_EXEC)
661 		rbase = 0;
662 	else {
663 		error = ENOEXEC;
664 		goto fail;
665 	}
666 
667 	/* Only support headers that fit within first page for now      */
668 	if ((hdr->e_phoff > PAGE_SIZE) ||
669 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
670 		error = ENOEXEC;
671 		goto fail;
672 	}
673 
674 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
675 	if (!aligned(phdr, Elf_Addr)) {
676 		error = ENOEXEC;
677 		goto fail;
678 	}
679 
680 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
681 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
682 			/* Loadable segment */
683 			prot = __elfN(trans_prot)(phdr[i].p_flags);
684 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
685 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
686 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
687 			if (error != 0)
688 				goto fail;
689 			/*
690 			 * Establish the base address if this is the
691 			 * first segment.
692 			 */
693 			if (numsegs == 0)
694   				base_addr = trunc_page(phdr[i].p_vaddr +
695 				    rbase);
696 			numsegs++;
697 		}
698 	}
699 	*addr = base_addr;
700 	*entry = (unsigned long)hdr->e_entry + rbase;
701 
702 fail:
703 	if (imgp->firstpage)
704 		exec_unmap_first_page(imgp);
705 
706 	if (nd->ni_vp)
707 		vput(nd->ni_vp);
708 
709 	free(tempdata, M_TEMP);
710 
711 	return (error);
712 }
713 
714 static int
715 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
716 {
717 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
718 	const Elf_Phdr *phdr;
719 	Elf_Auxargs *elf_auxargs;
720 	struct vmspace *vmspace;
721 	vm_prot_t prot;
722 	u_long text_size = 0, data_size = 0, total_size = 0;
723 	u_long text_addr = 0, data_addr = 0;
724 	u_long seg_size, seg_addr;
725 	u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
726 	int32_t osrel = 0;
727 	int error = 0, i, n, interp_name_len = 0;
728 	const char *interp = NULL, *newinterp = NULL;
729 	Elf_Brandinfo *brand_info;
730 	char *path;
731 	struct sysentvec *sv;
732 
733 	/*
734 	 * Do we have a valid ELF header ?
735 	 *
736 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
737 	 * if particular brand doesn't support it.
738 	 */
739 	if (__elfN(check_header)(hdr) != 0 ||
740 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
741 		return (-1);
742 
743 	/*
744 	 * From here on down, we return an errno, not -1, as we've
745 	 * detected an ELF file.
746 	 */
747 
748 	if ((hdr->e_phoff > PAGE_SIZE) ||
749 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
750 		/* Only support headers in first page for now */
751 		return (ENOEXEC);
752 	}
753 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
754 	if (!aligned(phdr, Elf_Addr))
755 		return (ENOEXEC);
756 	n = 0;
757 	baddr = 0;
758 	for (i = 0; i < hdr->e_phnum; i++) {
759 		switch (phdr[i].p_type) {
760 		case PT_LOAD:
761 			if (n == 0)
762 				baddr = phdr[i].p_vaddr;
763 			n++;
764 			break;
765 		case PT_INTERP:
766 			/* Path to interpreter */
767 			if (phdr[i].p_filesz > MAXPATHLEN ||
768 			    phdr[i].p_offset > PAGE_SIZE ||
769 			    phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
770 				return (ENOEXEC);
771 			interp = imgp->image_header + phdr[i].p_offset;
772 			interp_name_len = phdr[i].p_filesz;
773 			break;
774 		case PT_GNU_STACK:
775 			if (__elfN(nxstack))
776 				imgp->stack_prot =
777 				    __elfN(trans_prot)(phdr[i].p_flags);
778 			break;
779 		}
780 	}
781 
782 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
783 	    &osrel);
784 	if (brand_info == NULL) {
785 		uprintf("ELF binary type \"%u\" not known.\n",
786 		    hdr->e_ident[EI_OSABI]);
787 		return (ENOEXEC);
788 	}
789 	if (hdr->e_type == ET_DYN) {
790 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
791 			return (ENOEXEC);
792 		/*
793 		 * Honour the base load address from the dso if it is
794 		 * non-zero for some reason.
795 		 */
796 		if (baddr == 0)
797 			et_dyn_addr = ET_DYN_LOAD_ADDR;
798 		else
799 			et_dyn_addr = 0;
800 	} else
801 		et_dyn_addr = 0;
802 	sv = brand_info->sysvec;
803 	if (interp != NULL && brand_info->interp_newpath != NULL)
804 		newinterp = brand_info->interp_newpath;
805 
806 	/*
807 	 * Avoid a possible deadlock if the current address space is destroyed
808 	 * and that address space maps the locked vnode.  In the common case,
809 	 * the locked vnode's v_usecount is decremented but remains greater
810 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
811 	 * However, in cases where the vnode lock is external, such as nullfs,
812 	 * v_usecount may become zero.
813 	 *
814 	 * The VV_TEXT flag prevents modifications to the executable while
815 	 * the vnode is unlocked.
816 	 */
817 	VOP_UNLOCK(imgp->vp, 0);
818 
819 	error = exec_new_vmspace(imgp, sv);
820 	imgp->proc->p_sysent = sv;
821 
822 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
823 	if (error)
824 		return (error);
825 
826 	for (i = 0; i < hdr->e_phnum; i++) {
827 		switch (phdr[i].p_type) {
828 		case PT_LOAD:	/* Loadable segment */
829 			if (phdr[i].p_memsz == 0)
830 				break;
831 			prot = __elfN(trans_prot)(phdr[i].p_flags);
832 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
833 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
834 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
835 			    sv->sv_pagesize);
836 			if (error != 0)
837 				return (error);
838 
839 			/*
840 			 * If this segment contains the program headers,
841 			 * remember their virtual address for the AT_PHDR
842 			 * aux entry. Static binaries don't usually include
843 			 * a PT_PHDR entry.
844 			 */
845 			if (phdr[i].p_offset == 0 &&
846 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
847 				<= phdr[i].p_filesz)
848 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
849 				    et_dyn_addr;
850 
851 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
852 			seg_size = round_page(phdr[i].p_memsz +
853 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
854 
855 			/*
856 			 * Make the largest executable segment the official
857 			 * text segment and all others data.
858 			 *
859 			 * Note that obreak() assumes that data_addr +
860 			 * data_size == end of data load area, and the ELF
861 			 * file format expects segments to be sorted by
862 			 * address.  If multiple data segments exist, the
863 			 * last one will be used.
864 			 */
865 
866 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
867 				text_size = seg_size;
868 				text_addr = seg_addr;
869 			} else {
870 				data_size = seg_size;
871 				data_addr = seg_addr;
872 			}
873 			total_size += seg_size;
874 			break;
875 		case PT_PHDR: 	/* Program header table info */
876 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
877 			break;
878 		default:
879 			break;
880 		}
881 	}
882 
883 	if (data_addr == 0 && data_size == 0) {
884 		data_addr = text_addr;
885 		data_size = text_size;
886 	}
887 
888 	entry = (u_long)hdr->e_entry + et_dyn_addr;
889 
890 	/*
891 	 * Check limits.  It should be safe to check the
892 	 * limits after loading the segments since we do
893 	 * not actually fault in all the segments pages.
894 	 */
895 	PROC_LOCK(imgp->proc);
896 	if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
897 	    text_size > maxtsiz ||
898 	    total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
899 	    racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
900 	    racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
901 		PROC_UNLOCK(imgp->proc);
902 		return (ENOMEM);
903 	}
904 
905 	vmspace = imgp->proc->p_vmspace;
906 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
907 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
908 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
909 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
910 
911 	/*
912 	 * We load the dynamic linker where a userland call
913 	 * to mmap(0, ...) would put it.  The rationale behind this
914 	 * calculation is that it leaves room for the heap to grow to
915 	 * its maximum allowed size.
916 	 */
917 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
918 	    RLIMIT_DATA));
919 	PROC_UNLOCK(imgp->proc);
920 
921 	imgp->entry_addr = entry;
922 
923 	if (interp != NULL) {
924 		int have_interp = FALSE;
925 		VOP_UNLOCK(imgp->vp, 0);
926 		if (brand_info->emul_path != NULL &&
927 		    brand_info->emul_path[0] != '\0') {
928 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
929 			snprintf(path, MAXPATHLEN, "%s%s",
930 			    brand_info->emul_path, interp);
931 			error = __elfN(load_file)(imgp->proc, path, &addr,
932 			    &imgp->entry_addr, sv->sv_pagesize);
933 			free(path, M_TEMP);
934 			if (error == 0)
935 				have_interp = TRUE;
936 		}
937 		if (!have_interp && newinterp != NULL) {
938 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
939 			    &imgp->entry_addr, sv->sv_pagesize);
940 			if (error == 0)
941 				have_interp = TRUE;
942 		}
943 		if (!have_interp) {
944 			error = __elfN(load_file)(imgp->proc, interp, &addr,
945 			    &imgp->entry_addr, sv->sv_pagesize);
946 		}
947 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
948 		if (error != 0) {
949 			uprintf("ELF interpreter %s not found\n", interp);
950 			return (error);
951 		}
952 	} else
953 		addr = et_dyn_addr;
954 
955 	/*
956 	 * Construct auxargs table (used by the fixup routine)
957 	 */
958 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
959 	elf_auxargs->execfd = -1;
960 	elf_auxargs->phdr = proghdr;
961 	elf_auxargs->phent = hdr->e_phentsize;
962 	elf_auxargs->phnum = hdr->e_phnum;
963 	elf_auxargs->pagesz = PAGE_SIZE;
964 	elf_auxargs->base = addr;
965 	elf_auxargs->flags = 0;
966 	elf_auxargs->entry = entry;
967 
968 	imgp->auxargs = elf_auxargs;
969 	imgp->interpreted = 0;
970 	imgp->reloc_base = addr;
971 	imgp->proc->p_osrel = osrel;
972 
973 	return (error);
974 }
975 
976 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
977 
978 int
979 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
980 {
981 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
982 	Elf_Addr *base;
983 	Elf_Addr *pos;
984 
985 	base = (Elf_Addr *)*stack_base;
986 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
987 
988 	if (args->execfd != -1)
989 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
990 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
991 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
992 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
993 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
994 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
995 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
996 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
997 	if (imgp->execpathp != 0)
998 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
999 	AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate);
1000 	if (imgp->canary != 0) {
1001 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1002 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1003 	}
1004 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1005 	if (imgp->pagesizes != 0) {
1006 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1007 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1008 	}
1009 	if (imgp->sysent->sv_timekeep_base != 0) {
1010 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1011 		    imgp->sysent->sv_timekeep_base);
1012 	}
1013 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1014 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1015 	    imgp->sysent->sv_stackprot);
1016 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1017 
1018 	free(imgp->auxargs, M_TEMP);
1019 	imgp->auxargs = NULL;
1020 
1021 	base--;
1022 	suword(base, (long)imgp->args->argc);
1023 	*stack_base = (register_t *)base;
1024 	return (0);
1025 }
1026 
1027 /*
1028  * Code for generating ELF core dumps.
1029  */
1030 
1031 typedef void (*segment_callback)(vm_map_entry_t, void *);
1032 
1033 /* Closure for cb_put_phdr(). */
1034 struct phdr_closure {
1035 	Elf_Phdr *phdr;		/* Program header to fill in */
1036 	Elf_Off offset;		/* Offset of segment in core file */
1037 };
1038 
1039 /* Closure for cb_size_segment(). */
1040 struct sseg_closure {
1041 	int count;		/* Count of writable segments. */
1042 	size_t size;		/* Total size of all writable segments. */
1043 };
1044 
1045 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1046 
1047 struct note_info {
1048 	int		type;		/* Note type. */
1049 	outfunc_t 	outfunc; 	/* Output function. */
1050 	void		*outarg;	/* Argument for the output function. */
1051 	size_t		outsize;	/* Output size. */
1052 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1053 };
1054 
1055 TAILQ_HEAD(note_info_list, note_info);
1056 
1057 static void cb_put_phdr(vm_map_entry_t, void *);
1058 static void cb_size_segment(vm_map_entry_t, void *);
1059 static void each_writable_segment(struct thread *, segment_callback, void *);
1060 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1061     int, void *, size_t, struct note_info_list *, size_t, gzFile);
1062 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1063     size_t *);
1064 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1065 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1066 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1067 static int sbuf_drain_core_output(void *, const char *, int);
1068 static int sbuf_drain_count(void *arg, const char *data, int len);
1069 
1070 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1071 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1072 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1073 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1074 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1075 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1076 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1077 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1078 static void note_procstat_files(void *, struct sbuf *, size_t *);
1079 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1080 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1081 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1082 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1083 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1084 
1085 #ifdef COMPRESS_USER_CORES
1086 extern int compress_user_cores;
1087 extern int compress_user_cores_gzlevel;
1088 #endif
1089 
1090 static int
1091 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1092     struct ucred *active_cred, struct ucred *file_cred,
1093     struct thread *td, char *core_buf, gzFile gzfile) {
1094 
1095 	int error;
1096 	if (gzfile) {
1097 #ifdef COMPRESS_USER_CORES
1098 		error = compress_core(gzfile, base, core_buf, len, td);
1099 #else
1100 		panic("shouldn't be here");
1101 #endif
1102 	} else {
1103 		error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1104 		    UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1105 		    NULL, td);
1106 	}
1107 	return (error);
1108 }
1109 
1110 /* Coredump output parameters for sbuf drain routine. */
1111 struct sbuf_drain_core_params {
1112 	off_t		offset;
1113 	struct ucred	*active_cred;
1114 	struct ucred	*file_cred;
1115 	struct thread	*td;
1116 	struct vnode	*vp;
1117 #ifdef COMPRESS_USER_CORES
1118 	gzFile		gzfile;
1119 #endif
1120 };
1121 
1122 /*
1123  * Drain into a core file.
1124  */
1125 static int
1126 sbuf_drain_core_output(void *arg, const char *data, int len)
1127 {
1128 	struct sbuf_drain_core_params *p;
1129 	int error, locked;
1130 
1131 	p = (struct sbuf_drain_core_params *)arg;
1132 
1133 	/*
1134 	 * Some kern_proc out routines that print to this sbuf may
1135 	 * call us with the process lock held. Draining with the
1136 	 * non-sleepable lock held is unsafe. The lock is needed for
1137 	 * those routines when dumping a live process. In our case we
1138 	 * can safely release the lock before draining and acquire
1139 	 * again after.
1140 	 */
1141 	locked = PROC_LOCKED(p->td->td_proc);
1142 	if (locked)
1143 		PROC_UNLOCK(p->td->td_proc);
1144 #ifdef COMPRESS_USER_CORES
1145 	if (p->gzfile != Z_NULL)
1146 		error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1147 		    len, p->td);
1148 	else
1149 #endif
1150 		error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1151 		    __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1152 		    IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1153 		    p->td);
1154 	if (locked)
1155 		PROC_LOCK(p->td->td_proc);
1156 	if (error != 0)
1157 		return (-error);
1158 	p->offset += len;
1159 	return (len);
1160 }
1161 
1162 /*
1163  * Drain into a counter.
1164  */
1165 static int
1166 sbuf_drain_count(void *arg, const char *data __unused, int len)
1167 {
1168 	size_t *sizep;
1169 
1170 	sizep = (size_t *)arg;
1171 	*sizep += len;
1172 	return (len);
1173 }
1174 
1175 int
1176 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1177 {
1178 	struct ucred *cred = td->td_ucred;
1179 	int error = 0;
1180 	struct sseg_closure seginfo;
1181 	struct note_info_list notelst;
1182 	struct note_info *ninfo;
1183 	void *hdr;
1184 	size_t hdrsize, notesz, coresize;
1185 
1186 	gzFile gzfile = Z_NULL;
1187 	char *core_buf = NULL;
1188 #ifdef COMPRESS_USER_CORES
1189 	char gzopen_flags[8];
1190 	char *p;
1191 	int doing_compress = flags & IMGACT_CORE_COMPRESS;
1192 #endif
1193 
1194 	hdr = NULL;
1195 	TAILQ_INIT(&notelst);
1196 
1197 #ifdef COMPRESS_USER_CORES
1198         if (doing_compress) {
1199                 p = gzopen_flags;
1200                 *p++ = 'w';
1201                 if (compress_user_cores_gzlevel >= 0 &&
1202                     compress_user_cores_gzlevel <= 9)
1203                         *p++ = '0' + compress_user_cores_gzlevel;
1204                 *p = 0;
1205                 gzfile = gz_open("", gzopen_flags, vp);
1206                 if (gzfile == Z_NULL) {
1207                         error = EFAULT;
1208                         goto done;
1209                 }
1210                 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1211                 if (!core_buf) {
1212                         error = ENOMEM;
1213                         goto done;
1214                 }
1215         }
1216 #endif
1217 
1218 	/* Size the program segments. */
1219 	seginfo.count = 0;
1220 	seginfo.size = 0;
1221 	each_writable_segment(td, cb_size_segment, &seginfo);
1222 
1223 	/*
1224 	 * Collect info about the core file header area.
1225 	 */
1226 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1227 	__elfN(prepare_notes)(td, &notelst, &notesz);
1228 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1229 
1230 #ifdef RACCT
1231 	PROC_LOCK(td->td_proc);
1232 	error = racct_add(td->td_proc, RACCT_CORE, coresize);
1233 	PROC_UNLOCK(td->td_proc);
1234 	if (error != 0) {
1235 		error = EFAULT;
1236 		goto done;
1237 	}
1238 #endif
1239 	if (coresize >= limit) {
1240 		error = EFAULT;
1241 		goto done;
1242 	}
1243 
1244 	/*
1245 	 * Allocate memory for building the header, fill it up,
1246 	 * and write it out following the notes.
1247 	 */
1248 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1249 	if (hdr == NULL) {
1250 		error = EINVAL;
1251 		goto done;
1252 	}
1253 	error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1254 	    &notelst, notesz, gzfile);
1255 
1256 	/* Write the contents of all of the writable segments. */
1257 	if (error == 0) {
1258 		Elf_Phdr *php;
1259 		off_t offset;
1260 		int i;
1261 
1262 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1263 		offset = round_page(hdrsize + notesz);
1264 		for (i = 0; i < seginfo.count; i++) {
1265 			error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1266 			    php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1267 			if (error != 0)
1268 				break;
1269 			offset += php->p_filesz;
1270 			php++;
1271 		}
1272 	}
1273 	if (error) {
1274 		log(LOG_WARNING,
1275 		    "Failed to write core file for process %s (error %d)\n",
1276 		    curproc->p_comm, error);
1277 	}
1278 
1279 done:
1280 #ifdef COMPRESS_USER_CORES
1281 	if (core_buf)
1282 		free(core_buf, M_TEMP);
1283 	if (gzfile)
1284 		gzclose(gzfile);
1285 #endif
1286 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1287 		TAILQ_REMOVE(&notelst, ninfo, link);
1288 		free(ninfo, M_TEMP);
1289 	}
1290 	if (hdr != NULL)
1291 		free(hdr, M_TEMP);
1292 
1293 	return (error);
1294 }
1295 
1296 /*
1297  * A callback for each_writable_segment() to write out the segment's
1298  * program header entry.
1299  */
1300 static void
1301 cb_put_phdr(entry, closure)
1302 	vm_map_entry_t entry;
1303 	void *closure;
1304 {
1305 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1306 	Elf_Phdr *phdr = phc->phdr;
1307 
1308 	phc->offset = round_page(phc->offset);
1309 
1310 	phdr->p_type = PT_LOAD;
1311 	phdr->p_offset = phc->offset;
1312 	phdr->p_vaddr = entry->start;
1313 	phdr->p_paddr = 0;
1314 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1315 	phdr->p_align = PAGE_SIZE;
1316 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1317 
1318 	phc->offset += phdr->p_filesz;
1319 	phc->phdr++;
1320 }
1321 
1322 /*
1323  * A callback for each_writable_segment() to gather information about
1324  * the number of segments and their total size.
1325  */
1326 static void
1327 cb_size_segment(entry, closure)
1328 	vm_map_entry_t entry;
1329 	void *closure;
1330 {
1331 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1332 
1333 	ssc->count++;
1334 	ssc->size += entry->end - entry->start;
1335 }
1336 
1337 /*
1338  * For each writable segment in the process's memory map, call the given
1339  * function with a pointer to the map entry and some arbitrary
1340  * caller-supplied data.
1341  */
1342 static void
1343 each_writable_segment(td, func, closure)
1344 	struct thread *td;
1345 	segment_callback func;
1346 	void *closure;
1347 {
1348 	struct proc *p = td->td_proc;
1349 	vm_map_t map = &p->p_vmspace->vm_map;
1350 	vm_map_entry_t entry;
1351 	vm_object_t backing_object, object;
1352 	boolean_t ignore_entry;
1353 
1354 	vm_map_lock_read(map);
1355 	for (entry = map->header.next; entry != &map->header;
1356 	    entry = entry->next) {
1357 		/*
1358 		 * Don't dump inaccessible mappings, deal with legacy
1359 		 * coredump mode.
1360 		 *
1361 		 * Note that read-only segments related to the elf binary
1362 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1363 		 * need to arbitrarily ignore such segments.
1364 		 */
1365 		if (elf_legacy_coredump) {
1366 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1367 				continue;
1368 		} else {
1369 			if ((entry->protection & VM_PROT_ALL) == 0)
1370 				continue;
1371 		}
1372 
1373 		/*
1374 		 * Dont include memory segment in the coredump if
1375 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1376 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1377 		 * kernel map).
1378 		 */
1379 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1380 			continue;
1381 
1382 		if ((object = entry->object.vm_object) == NULL)
1383 			continue;
1384 
1385 		/* Ignore memory-mapped devices and such things. */
1386 		VM_OBJECT_RLOCK(object);
1387 		while ((backing_object = object->backing_object) != NULL) {
1388 			VM_OBJECT_RLOCK(backing_object);
1389 			VM_OBJECT_RUNLOCK(object);
1390 			object = backing_object;
1391 		}
1392 		ignore_entry = object->type != OBJT_DEFAULT &&
1393 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1394 		VM_OBJECT_RUNLOCK(object);
1395 		if (ignore_entry)
1396 			continue;
1397 
1398 		(*func)(entry, closure);
1399 	}
1400 	vm_map_unlock_read(map);
1401 }
1402 
1403 /*
1404  * Write the core file header to the file, including padding up to
1405  * the page boundary.
1406  */
1407 static int
1408 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1409     int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1410     size_t notesz, gzFile gzfile)
1411 {
1412 	struct sbuf_drain_core_params params;
1413 	struct note_info *ninfo;
1414 	struct sbuf *sb;
1415 	int error;
1416 
1417 	/* Fill in the header. */
1418 	bzero(hdr, hdrsize);
1419 	__elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1420 
1421 	params.offset = 0;
1422 	params.active_cred = cred;
1423 	params.file_cred = NOCRED;
1424 	params.td = td;
1425 	params.vp = vp;
1426 #ifdef COMPRESS_USER_CORES
1427 	params.gzfile = gzfile;
1428 #endif
1429 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1430 	sbuf_set_drain(sb, sbuf_drain_core_output, &params);
1431 	sbuf_start_section(sb, NULL);
1432 	sbuf_bcat(sb, hdr, hdrsize);
1433 	TAILQ_FOREACH(ninfo, notelst, link)
1434 	    __elfN(putnote)(ninfo, sb);
1435 	/* Align up to a page boundary for the program segments. */
1436 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1437 	error = sbuf_finish(sb);
1438 	sbuf_delete(sb);
1439 
1440 	return (error);
1441 }
1442 
1443 static void
1444 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1445     size_t *sizep)
1446 {
1447 	struct proc *p;
1448 	struct thread *thr;
1449 	size_t size;
1450 
1451 	p = td->td_proc;
1452 	size = 0;
1453 
1454 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1455 
1456 	/*
1457 	 * To have the debugger select the right thread (LWP) as the initial
1458 	 * thread, we dump the state of the thread passed to us in td first.
1459 	 * This is the thread that causes the core dump and thus likely to
1460 	 * be the right thread one wants to have selected in the debugger.
1461 	 */
1462 	thr = td;
1463 	while (thr != NULL) {
1464 		size += register_note(list, NT_PRSTATUS,
1465 		    __elfN(note_prstatus), thr);
1466 		size += register_note(list, NT_FPREGSET,
1467 		    __elfN(note_fpregset), thr);
1468 		size += register_note(list, NT_THRMISC,
1469 		    __elfN(note_thrmisc), thr);
1470 		size += register_note(list, -1,
1471 		    __elfN(note_threadmd), thr);
1472 
1473 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1474 		    TAILQ_NEXT(thr, td_plist);
1475 		if (thr == td)
1476 			thr = TAILQ_NEXT(thr, td_plist);
1477 	}
1478 
1479 	size += register_note(list, NT_PROCSTAT_PROC,
1480 	    __elfN(note_procstat_proc), p);
1481 	size += register_note(list, NT_PROCSTAT_FILES,
1482 	    note_procstat_files, p);
1483 	size += register_note(list, NT_PROCSTAT_VMMAP,
1484 	    note_procstat_vmmap, p);
1485 	size += register_note(list, NT_PROCSTAT_GROUPS,
1486 	    note_procstat_groups, p);
1487 	size += register_note(list, NT_PROCSTAT_UMASK,
1488 	    note_procstat_umask, p);
1489 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1490 	    note_procstat_rlimit, p);
1491 	size += register_note(list, NT_PROCSTAT_OSREL,
1492 	    note_procstat_osrel, p);
1493 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1494 	    __elfN(note_procstat_psstrings), p);
1495 	size += register_note(list, NT_PROCSTAT_AUXV,
1496 	    __elfN(note_procstat_auxv), p);
1497 
1498 	*sizep = size;
1499 }
1500 
1501 static void
1502 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1503     size_t notesz)
1504 {
1505 	Elf_Ehdr *ehdr;
1506 	Elf_Phdr *phdr;
1507 	struct phdr_closure phc;
1508 
1509 	ehdr = (Elf_Ehdr *)hdr;
1510 	phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1511 
1512 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1513 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1514 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1515 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1516 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1517 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1518 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1519 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1520 	ehdr->e_ident[EI_ABIVERSION] = 0;
1521 	ehdr->e_ident[EI_PAD] = 0;
1522 	ehdr->e_type = ET_CORE;
1523 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1524 	ehdr->e_machine = ELF_ARCH32;
1525 #else
1526 	ehdr->e_machine = ELF_ARCH;
1527 #endif
1528 	ehdr->e_version = EV_CURRENT;
1529 	ehdr->e_entry = 0;
1530 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1531 	ehdr->e_flags = 0;
1532 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1533 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1534 	ehdr->e_phnum = numsegs + 1;
1535 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1536 	ehdr->e_shnum = 0;
1537 	ehdr->e_shstrndx = SHN_UNDEF;
1538 
1539 	/*
1540 	 * Fill in the program header entries.
1541 	 */
1542 
1543 	/* The note segement. */
1544 	phdr->p_type = PT_NOTE;
1545 	phdr->p_offset = hdrsize;
1546 	phdr->p_vaddr = 0;
1547 	phdr->p_paddr = 0;
1548 	phdr->p_filesz = notesz;
1549 	phdr->p_memsz = 0;
1550 	phdr->p_flags = PF_R;
1551 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1552 	phdr++;
1553 
1554 	/* All the writable segments from the program. */
1555 	phc.phdr = phdr;
1556 	phc.offset = round_page(hdrsize + notesz);
1557 	each_writable_segment(td, cb_put_phdr, &phc);
1558 }
1559 
1560 static size_t
1561 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1562 {
1563 	struct note_info *ninfo;
1564 	size_t size, notesize;
1565 
1566 	size = 0;
1567 	out(arg, NULL, &size);
1568 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1569 	ninfo->type = type;
1570 	ninfo->outfunc = out;
1571 	ninfo->outarg = arg;
1572 	ninfo->outsize = size;
1573 	TAILQ_INSERT_TAIL(list, ninfo, link);
1574 
1575 	if (type == -1)
1576 		return (size);
1577 
1578 	notesize = sizeof(Elf_Note) +		/* note header */
1579 	    roundup2(8, ELF_NOTE_ROUNDSIZE) +	/* note name ("FreeBSD") */
1580 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1581 
1582 	return (notesize);
1583 }
1584 
1585 static void
1586 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1587 {
1588 	Elf_Note note;
1589 	ssize_t old_len;
1590 
1591 	if (ninfo->type == -1) {
1592 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1593 		return;
1594 	}
1595 
1596 	note.n_namesz = 8; /* strlen("FreeBSD") + 1 */
1597 	note.n_descsz = ninfo->outsize;
1598 	note.n_type = ninfo->type;
1599 
1600 	sbuf_bcat(sb, &note, sizeof(note));
1601 	sbuf_start_section(sb, &old_len);
1602 	sbuf_bcat(sb, "FreeBSD", note.n_namesz);
1603 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1604 	if (note.n_descsz == 0)
1605 		return;
1606 	sbuf_start_section(sb, &old_len);
1607 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1608 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1609 }
1610 
1611 /*
1612  * Miscellaneous note out functions.
1613  */
1614 
1615 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1616 #include <compat/freebsd32/freebsd32.h>
1617 
1618 typedef struct prstatus32 elf_prstatus_t;
1619 typedef struct prpsinfo32 elf_prpsinfo_t;
1620 typedef struct fpreg32 elf_prfpregset_t;
1621 typedef struct fpreg32 elf_fpregset_t;
1622 typedef struct reg32 elf_gregset_t;
1623 typedef struct thrmisc32 elf_thrmisc_t;
1624 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1625 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1626 typedef uint32_t elf_ps_strings_t;
1627 #else
1628 typedef prstatus_t elf_prstatus_t;
1629 typedef prpsinfo_t elf_prpsinfo_t;
1630 typedef prfpregset_t elf_prfpregset_t;
1631 typedef prfpregset_t elf_fpregset_t;
1632 typedef gregset_t elf_gregset_t;
1633 typedef thrmisc_t elf_thrmisc_t;
1634 #define ELF_KERN_PROC_MASK	0
1635 typedef struct kinfo_proc elf_kinfo_proc_t;
1636 typedef vm_offset_t elf_ps_strings_t;
1637 #endif
1638 
1639 static void
1640 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1641 {
1642 	struct proc *p;
1643 	elf_prpsinfo_t *psinfo;
1644 
1645 	p = (struct proc *)arg;
1646 	if (sb != NULL) {
1647 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1648 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1649 		psinfo->pr_version = PRPSINFO_VERSION;
1650 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1651 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1652 		/*
1653 		 * XXX - We don't fill in the command line arguments properly
1654 		 * yet.
1655 		 */
1656 		strlcpy(psinfo->pr_psargs, p->p_comm,
1657 		    sizeof(psinfo->pr_psargs));
1658 
1659 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1660 		free(psinfo, M_TEMP);
1661 	}
1662 	*sizep = sizeof(*psinfo);
1663 }
1664 
1665 static void
1666 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1667 {
1668 	struct thread *td;
1669 	elf_prstatus_t *status;
1670 
1671 	td = (struct thread *)arg;
1672 	if (sb != NULL) {
1673 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1674 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1675 		status->pr_version = PRSTATUS_VERSION;
1676 		status->pr_statussz = sizeof(elf_prstatus_t);
1677 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1678 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1679 		status->pr_osreldate = osreldate;
1680 		status->pr_cursig = td->td_proc->p_sig;
1681 		status->pr_pid = td->td_tid;
1682 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1683 		fill_regs32(td, &status->pr_reg);
1684 #else
1685 		fill_regs(td, &status->pr_reg);
1686 #endif
1687 		sbuf_bcat(sb, status, sizeof(*status));
1688 		free(status, M_TEMP);
1689 	}
1690 	*sizep = sizeof(*status);
1691 }
1692 
1693 static void
1694 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1695 {
1696 	struct thread *td;
1697 	elf_prfpregset_t *fpregset;
1698 
1699 	td = (struct thread *)arg;
1700 	if (sb != NULL) {
1701 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1702 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1703 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1704 		fill_fpregs32(td, fpregset);
1705 #else
1706 		fill_fpregs(td, fpregset);
1707 #endif
1708 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1709 		free(fpregset, M_TEMP);
1710 	}
1711 	*sizep = sizeof(*fpregset);
1712 }
1713 
1714 static void
1715 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1716 {
1717 	struct thread *td;
1718 	elf_thrmisc_t thrmisc;
1719 
1720 	td = (struct thread *)arg;
1721 	if (sb != NULL) {
1722 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1723 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1724 		strcpy(thrmisc.pr_tname, td->td_name);
1725 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1726 	}
1727 	*sizep = sizeof(thrmisc);
1728 }
1729 
1730 /*
1731  * Allow for MD specific notes, as well as any MD
1732  * specific preparations for writing MI notes.
1733  */
1734 static void
1735 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1736 {
1737 	struct thread *td;
1738 	void *buf;
1739 	size_t size;
1740 
1741 	td = (struct thread *)arg;
1742 	size = *sizep;
1743 	if (size != 0 && sb != NULL)
1744 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1745 	else
1746 		buf = NULL;
1747 	size = 0;
1748 	__elfN(dump_thread)(td, buf, &size);
1749 	KASSERT(*sizep == size, ("invalid size"));
1750 	if (size != 0 && sb != NULL)
1751 		sbuf_bcat(sb, buf, size);
1752 	free(buf, M_TEMP);
1753 	*sizep = size;
1754 }
1755 
1756 #ifdef KINFO_PROC_SIZE
1757 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1758 #endif
1759 
1760 static void
1761 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1762 {
1763 	struct proc *p;
1764 	size_t size;
1765 	int structsize;
1766 
1767 	p = (struct proc *)arg;
1768 	size = sizeof(structsize) + p->p_numthreads *
1769 	    sizeof(elf_kinfo_proc_t);
1770 
1771 	if (sb != NULL) {
1772 		KASSERT(*sizep == size, ("invalid size"));
1773 		structsize = sizeof(elf_kinfo_proc_t);
1774 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1775 		PROC_LOCK(p);
1776 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1777 	}
1778 	*sizep = size;
1779 }
1780 
1781 #ifdef KINFO_FILE_SIZE
1782 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1783 #endif
1784 
1785 static void
1786 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1787 {
1788 	struct proc *p;
1789 	size_t size;
1790 	int structsize;
1791 
1792 	p = (struct proc *)arg;
1793 	if (sb == NULL) {
1794 		size = 0;
1795 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1796 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1797 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1798 		PROC_LOCK(p);
1799 		kern_proc_filedesc_out(p, sb, -1);
1800 		sbuf_finish(sb);
1801 		sbuf_delete(sb);
1802 		*sizep = size;
1803 	} else {
1804 		structsize = sizeof(struct kinfo_file);
1805 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1806 		PROC_LOCK(p);
1807 		kern_proc_filedesc_out(p, sb, -1);
1808 	}
1809 }
1810 
1811 #ifdef KINFO_VMENTRY_SIZE
1812 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1813 #endif
1814 
1815 static void
1816 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1817 {
1818 	struct proc *p;
1819 	size_t size;
1820 	int structsize;
1821 
1822 	p = (struct proc *)arg;
1823 	if (sb == NULL) {
1824 		size = 0;
1825 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1826 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1827 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1828 		PROC_LOCK(p);
1829 		kern_proc_vmmap_out(p, sb);
1830 		sbuf_finish(sb);
1831 		sbuf_delete(sb);
1832 		*sizep = size;
1833 	} else {
1834 		structsize = sizeof(struct kinfo_vmentry);
1835 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1836 		PROC_LOCK(p);
1837 		kern_proc_vmmap_out(p, sb);
1838 	}
1839 }
1840 
1841 static void
1842 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1843 {
1844 	struct proc *p;
1845 	size_t size;
1846 	int structsize;
1847 
1848 	p = (struct proc *)arg;
1849 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1850 	if (sb != NULL) {
1851 		KASSERT(*sizep == size, ("invalid size"));
1852 		structsize = sizeof(gid_t);
1853 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1854 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1855 		    sizeof(gid_t));
1856 	}
1857 	*sizep = size;
1858 }
1859 
1860 static void
1861 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1862 {
1863 	struct proc *p;
1864 	size_t size;
1865 	int structsize;
1866 
1867 	p = (struct proc *)arg;
1868 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1869 	if (sb != NULL) {
1870 		KASSERT(*sizep == size, ("invalid size"));
1871 		structsize = sizeof(p->p_fd->fd_cmask);
1872 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1873 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1874 	}
1875 	*sizep = size;
1876 }
1877 
1878 static void
1879 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1880 {
1881 	struct proc *p;
1882 	struct rlimit rlim[RLIM_NLIMITS];
1883 	size_t size;
1884 	int structsize, i;
1885 
1886 	p = (struct proc *)arg;
1887 	size = sizeof(structsize) + sizeof(rlim);
1888 	if (sb != NULL) {
1889 		KASSERT(*sizep == size, ("invalid size"));
1890 		structsize = sizeof(rlim);
1891 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1892 		PROC_LOCK(p);
1893 		for (i = 0; i < RLIM_NLIMITS; i++)
1894 			lim_rlimit(p, i, &rlim[i]);
1895 		PROC_UNLOCK(p);
1896 		sbuf_bcat(sb, rlim, sizeof(rlim));
1897 	}
1898 	*sizep = size;
1899 }
1900 
1901 static void
1902 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1903 {
1904 	struct proc *p;
1905 	size_t size;
1906 	int structsize;
1907 
1908 	p = (struct proc *)arg;
1909 	size = sizeof(structsize) + sizeof(p->p_osrel);
1910 	if (sb != NULL) {
1911 		KASSERT(*sizep == size, ("invalid size"));
1912 		structsize = sizeof(p->p_osrel);
1913 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1914 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1915 	}
1916 	*sizep = size;
1917 }
1918 
1919 static void
1920 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1921 {
1922 	struct proc *p;
1923 	elf_ps_strings_t ps_strings;
1924 	size_t size;
1925 	int structsize;
1926 
1927 	p = (struct proc *)arg;
1928 	size = sizeof(structsize) + sizeof(ps_strings);
1929 	if (sb != NULL) {
1930 		KASSERT(*sizep == size, ("invalid size"));
1931 		structsize = sizeof(ps_strings);
1932 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1933 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1934 #else
1935 		ps_strings = p->p_sysent->sv_psstrings;
1936 #endif
1937 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1938 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
1939 	}
1940 	*sizep = size;
1941 }
1942 
1943 static void
1944 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
1945 {
1946 	struct proc *p;
1947 	size_t size;
1948 	int structsize;
1949 
1950 	p = (struct proc *)arg;
1951 	if (sb == NULL) {
1952 		size = 0;
1953 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1954 		sbuf_set_drain(sb, sbuf_drain_count, &size);
1955 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1956 		PHOLD(p);
1957 		proc_getauxv(curthread, p, sb);
1958 		PRELE(p);
1959 		sbuf_finish(sb);
1960 		sbuf_delete(sb);
1961 		*sizep = size;
1962 	} else {
1963 		structsize = sizeof(Elf_Auxinfo);
1964 		sbuf_bcat(sb, &structsize, sizeof(structsize));
1965 		PHOLD(p);
1966 		proc_getauxv(curthread, p, sb);
1967 		PRELE(p);
1968 	}
1969 }
1970 
1971 static boolean_t
1972 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
1973     int32_t *osrel, const Elf_Phdr *pnote)
1974 {
1975 	const Elf_Note *note, *note0, *note_end;
1976 	const char *note_name;
1977 	int i;
1978 
1979 	if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
1980 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
1981 		return (FALSE);
1982 
1983 	note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1984 	note_end = (const Elf_Note *)(imgp->image_header +
1985 	    pnote->p_offset + pnote->p_filesz);
1986 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1987 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
1988 		    (const char *)note < sizeof(Elf_Note))
1989 			return (FALSE);
1990 		if (note->n_namesz != checknote->hdr.n_namesz ||
1991 		    note->n_descsz != checknote->hdr.n_descsz ||
1992 		    note->n_type != checknote->hdr.n_type)
1993 			goto nextnote;
1994 		note_name = (const char *)(note + 1);
1995 		if (note_name + checknote->hdr.n_namesz >=
1996 		    (const char *)note_end || strncmp(checknote->vendor,
1997 		    note_name, checknote->hdr.n_namesz) != 0)
1998 			goto nextnote;
1999 
2000 		/*
2001 		 * Fetch the osreldate for binary
2002 		 * from the ELF OSABI-note if necessary.
2003 		 */
2004 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2005 		    checknote->trans_osrel != NULL)
2006 			return (checknote->trans_osrel(note, osrel));
2007 		return (TRUE);
2008 
2009 nextnote:
2010 		note = (const Elf_Note *)((const char *)(note + 1) +
2011 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2012 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2013 	}
2014 
2015 	return (FALSE);
2016 }
2017 
2018 /*
2019  * Try to find the appropriate ABI-note section for checknote,
2020  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2021  * first page of the image is searched, the same as for headers.
2022  */
2023 static boolean_t
2024 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2025     int32_t *osrel)
2026 {
2027 	const Elf_Phdr *phdr;
2028 	const Elf_Ehdr *hdr;
2029 	int i;
2030 
2031 	hdr = (const Elf_Ehdr *)imgp->image_header;
2032 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2033 
2034 	for (i = 0; i < hdr->e_phnum; i++) {
2035 		if (phdr[i].p_type == PT_NOTE &&
2036 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2037 			return (TRUE);
2038 	}
2039 	return (FALSE);
2040 
2041 }
2042 
2043 /*
2044  * Tell kern_execve.c about it, with a little help from the linker.
2045  */
2046 static struct execsw __elfN(execsw) = {
2047 	__CONCAT(exec_, __elfN(imgact)),
2048 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2049 };
2050 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2051 
2052 #ifdef COMPRESS_USER_CORES
2053 /*
2054  * Compress and write out a core segment for a user process.
2055  *
2056  * 'inbuf' is the starting address of a VM segment in the process' address
2057  * space that is to be compressed and written out to the core file.  'dest_buf'
2058  * is a buffer in the kernel's address space.  The segment is copied from
2059  * 'inbuf' to 'dest_buf' first before being processed by the compression
2060  * routine gzwrite().  This copying is necessary because the content of the VM
2061  * segment may change between the compression pass and the crc-computation pass
2062  * in gzwrite().  This is because realtime threads may preempt the UNIX kernel.
2063  *
2064  * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2065  */
2066 static int
2067 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2068     struct thread *td)
2069 {
2070 	int len_compressed;
2071 	int error = 0;
2072 	unsigned int chunk_len;
2073 
2074 	while (len) {
2075 		if (inbuf != NULL) {
2076 			chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2077 			copyin(inbuf, dest_buf, chunk_len);
2078 			inbuf += chunk_len;
2079 		} else {
2080 			chunk_len = len;
2081 		}
2082 		len_compressed = gzwrite(file, dest_buf, chunk_len);
2083 
2084 		EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2085 
2086 		if ((unsigned int)len_compressed != chunk_len) {
2087 			log(LOG_WARNING,
2088 			    "compress_core: length mismatch (0x%x returned, "
2089 			    "0x%x expected)\n", len_compressed, chunk_len);
2090 			EVENTHANDLER_INVOKE(app_coredump_error, td,
2091 			    "compress_core: length mismatch %x -> %x",
2092 			    chunk_len, len_compressed);
2093 			error = EFAULT;
2094 			break;
2095 		}
2096 		len -= chunk_len;
2097 		maybe_yield();
2098 	}
2099 
2100 	return (error);
2101 }
2102 #endif /* COMPRESS_USER_CORES */
2103 
2104 static vm_prot_t
2105 __elfN(trans_prot)(Elf_Word flags)
2106 {
2107 	vm_prot_t prot;
2108 
2109 	prot = 0;
2110 	if (flags & PF_X)
2111 		prot |= VM_PROT_EXECUTE;
2112 	if (flags & PF_W)
2113 		prot |= VM_PROT_WRITE;
2114 	if (flags & PF_R)
2115 		prot |= VM_PROT_READ;
2116 #if __ELF_WORD_SIZE == 32
2117 #if defined(__amd64__) || defined(__ia64__)
2118 	if (i386_read_exec && (flags & PF_R))
2119 		prot |= VM_PROT_EXECUTE;
2120 #endif
2121 #endif
2122 	return (prot);
2123 }
2124 
2125 static Elf_Word
2126 __elfN(untrans_prot)(vm_prot_t prot)
2127 {
2128 	Elf_Word flags;
2129 
2130 	flags = 0;
2131 	if (prot & VM_PROT_EXECUTE)
2132 		flags |= PF_X;
2133 	if (prot & VM_PROT_READ)
2134 		flags |= PF_R;
2135 	if (prot & VM_PROT_WRITE)
2136 		flags |= PF_W;
2137 	return (flags);
2138 }
2139