xref: /dflybsd-src/sys/kern/imgact_elf.c (revision ae788f37fe53d5d1ca1e12a184a662192caad3c5)
1 /*-
2  * Copyright (c) 1995-1996 Søren Schmidt
3  * Copyright (c) 1996 Peter Wemm
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer
11  *    in this position and unchanged.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. The name of the author may not be used to endorse or promote products
16  *    derived from this software withough specific prior written permission
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  *
29  * $FreeBSD: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $
30  * $DragonFly: src/sys/kern/imgact_elf.c,v 1.55 2008/08/17 17:21:36 nth Exp $
31  */
32 
33 #include <sys/param.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/file.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
41 #include <sys/mman.h>
42 #include <sys/systm.h>
43 #include <sys/proc.h>
44 #include <sys/nlookup.h>
45 #include <sys/pioctl.h>
46 #include <sys/procfs.h>
47 #include <sys/resourcevar.h>
48 #include <sys/signalvar.h>
49 #include <sys/stat.h>
50 #include <sys/syscall.h>
51 #include <sys/sysctl.h>
52 #include <sys/sysent.h>
53 #include <sys/vnode.h>
54 #include <sys/sfbuf.h>
55 
56 #include <vm/vm.h>
57 #include <vm/vm_kern.h>
58 #include <vm/vm_param.h>
59 #include <vm/pmap.h>
60 #include <sys/lock.h>
61 #include <vm/vm_map.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_extern.h>
64 
65 #include <machine/elf.h>
66 #include <machine/md_var.h>
67 #include <sys/mount.h>
68 #include <sys/ckpt.h>
69 #define OLD_EI_BRAND	8
70 
71 __ElfType(Brandinfo);
72 __ElfType(Auxargs);
73 
74 static int elf_check_header (const Elf_Ehdr *hdr);
75 static int elf_freebsd_fixup (register_t **stack_base,
76     struct image_params *imgp);
77 static int elf_load_file (struct proc *p, const char *file, u_long *addr,
78     u_long *entry);
79 static int elf_load_section (struct proc *p,
80     struct vmspace *vmspace, struct vnode *vp,
81     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
82     vm_prot_t prot);
83 static int exec_elf_imgact (struct image_params *imgp);
84 
85 static int elf_trace = 0;
86 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
87 static int elf_legacy_coredump = 0;
88 SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
89     &elf_legacy_coredump, 0, "");
90 
91 static int dragonfly_match_abi_note(const Elf_Note *);
92 static int freebsd_match_abi_note(const Elf_Note *);
93 
94 static struct sysentvec elf_freebsd_sysvec = {
95         SYS_MAXSYSCALL,
96         sysent,
97         -1,
98         0,
99         0,
100         0,
101         0,
102         0,
103         elf_freebsd_fixup,
104         sendsig,
105         sigcode,
106         &szsigcode,
107         0,
108 	"FreeBSD ELF",
109 	elf_coredump,
110 	NULL,
111 	MINSIGSTKSZ
112 };
113 
114 static Elf_Brandinfo freebsd_brand_info = {
115 						ELFOSABI_FREEBSD,
116 						"FreeBSD",
117 						freebsd_match_abi_note,
118 						"",
119 						"/usr/libexec/ld-elf.so.1",
120 						&elf_freebsd_sysvec
121 					  };
122 
123 static Elf_Brandinfo dragonfly_brand_info = {
124 						ELFOSABI_NONE,
125 						"DragonFly",
126 						dragonfly_match_abi_note,
127 						"",
128 						"/usr/libexec/ld-elf.so.2",
129 						&elf_freebsd_sysvec
130 					  };
131 
132 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
133 							&dragonfly_brand_info,
134 							&freebsd_brand_info,
135 							NULL, NULL, NULL,
136 							NULL, NULL, NULL
137 						    };
138 
139 static int
140 freebsd_match_abi_note(const Elf_Note *abi_note)
141 {
142 	const char *abi_name = (const char *)
143 	    ((const uint8_t *)abi_note + sizeof(*abi_note));
144 
145 	if (abi_note->n_namesz != sizeof("FreeBSD"))
146 		return(FALSE);
147 	if (memcmp(abi_name, "FreeBSD", sizeof("FreeBSD")))
148 		return(FALSE);
149 	return(TRUE);
150 }
151 
152 static int
153 dragonfly_match_abi_note(const Elf_Note *abi_note)
154 {
155 	const char *abi_name = (const char *)
156 	    ((const uint8_t *)abi_note + sizeof(*abi_note));
157 
158 	if (abi_note->n_namesz != sizeof("DragonFly"))
159 		return(FALSE);
160 	if (memcmp(abi_name, "DragonFly", sizeof("DragonFly")))
161 		return(FALSE);
162 	return(TRUE);
163 }
164 
165 int
166 elf_insert_brand_entry(Elf_Brandinfo *entry)
167 {
168 	int i;
169 
170 	for (i=1; i<MAX_BRANDS; i++) {
171 		if (elf_brand_list[i] == NULL) {
172 			elf_brand_list[i] = entry;
173 			break;
174 		}
175 	}
176 	if (i == MAX_BRANDS)
177 		return -1;
178 	return 0;
179 }
180 
181 int
182 elf_remove_brand_entry(Elf_Brandinfo *entry)
183 {
184 	int i;
185 
186 	for (i=1; i<MAX_BRANDS; i++) {
187 		if (elf_brand_list[i] == entry) {
188 			elf_brand_list[i] = NULL;
189 			break;
190 		}
191 	}
192 	if (i == MAX_BRANDS)
193 		return -1;
194 	return 0;
195 }
196 
197 /*
198  * Check if an elf brand is being used anywhere in the system.
199  *
200  * Used by the linux emulation module unloader.  This isn't safe from
201  * races.
202  */
203 struct elf_brand_inuse_info {
204 	int rval;
205 	Elf_Brandinfo *entry;
206 };
207 
208 static int elf_brand_inuse_callback(struct proc *p, void *data);
209 
210 int
211 elf_brand_inuse(Elf_Brandinfo *entry)
212 {
213 	struct elf_brand_inuse_info info;
214 
215 	info.rval = FALSE;
216 	info.entry = entry;
217 	allproc_scan(elf_brand_inuse_callback, entry);
218 	return (info.rval);
219 }
220 
221 static
222 int
223 elf_brand_inuse_callback(struct proc *p, void *data)
224 {
225 	struct elf_brand_inuse_info *info = data;
226 
227 	if (p->p_sysent == info->entry->sysvec) {
228 		info->rval = TRUE;
229 		return(-1);
230 	}
231 	return(0);
232 }
233 
234 static int
235 elf_check_header(const Elf_Ehdr *hdr)
236 {
237 	if (!IS_ELF(*hdr) ||
238 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
239 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
240 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
241 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
242 	    hdr->e_ehsize != sizeof(Elf_Ehdr) ||
243 	    hdr->e_version != ELF_TARG_VER)
244 		return ENOEXEC;
245 
246 	if (!ELF_MACHINE_OK(hdr->e_machine))
247 		return ENOEXEC;
248 
249 	return 0;
250 }
251 
252 static Elf_Brandinfo *
253 elf_check_abi_note(struct image_params *imgp, const Elf_Phdr *ph)
254 {
255 	Elf_Brandinfo *match = NULL;
256 	const Elf_Note *tmp_note;
257 	struct sf_buf *sfb;
258 	const char *page;
259 	char *data = NULL;
260 	Elf_Off off;
261 	size_t firstoff;
262 	size_t len;
263 	size_t firstlen;
264 
265 	len = ph->p_filesz;
266 	off = ph->p_offset;
267 
268 	firstoff = off & PAGE_MASK;
269 	firstlen = PAGE_SIZE - firstoff;
270 
271 	if (len < sizeof(Elf_Note) || len > PAGE_SIZE)
272 		return NULL; /* ENOEXEC? */
273 
274 	if (exec_map_page(imgp, off >> PAGE_SHIFT, &sfb, &page))
275 		return NULL;
276 
277 	/*
278 	 * Crosses page boundary?  Is that allowed?
279 	 */
280 	if (firstlen < len) {
281 		data = kmalloc(len, M_TEMP, M_WAITOK);
282 
283 		bcopy(page + firstoff, data, firstlen);
284 
285 		exec_unmap_page(sfb);
286 		if (exec_map_page(imgp, (off >> PAGE_SHIFT) + 1, &sfb, &page)) {
287 			kfree(data, M_TEMP);
288 			return NULL;
289 		}
290 		bcopy(page, data + firstlen, len - firstlen);
291 		tmp_note = (void *)data;
292 	} else {
293 		tmp_note = (const void *)(page + firstoff);
294 	}
295 
296 	while (len >= sizeof(Elf_Note)) {
297 		int i;
298 		size_t nlen = roundup(tmp_note->n_namesz, sizeof(Elf_Word)) +
299 			      roundup(tmp_note->n_descsz, sizeof(Elf_Word)) +
300 			      sizeof(Elf_Note);
301 
302 		if (nlen > len)
303 			break;
304 
305 		if (tmp_note->n_type != 1)
306 			goto next;
307 
308 		for (i = 0; i < MAX_BRANDS; i++) {
309 			Elf_Brandinfo *bi = elf_brand_list[i];
310 
311 			if (bi != NULL && bi->match_abi_note != NULL &&
312 			    bi->match_abi_note(tmp_note)) {
313 				match = bi;
314 				break;
315 			}
316 		}
317 
318 		if (match != NULL)
319 			break;
320 
321 next:
322 		len -= nlen;
323 		tmp_note += nlen;
324 	}
325 
326 	if (data != NULL)
327 		kfree(data, M_TEMP);
328 	exec_unmap_page(sfb);
329 
330 	return (match);
331 }
332 
333 static int
334 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp,
335 		 vm_offset_t offset, caddr_t vmaddr, size_t memsz,
336 		 size_t filsz, vm_prot_t prot)
337 {
338 	size_t map_len;
339 	vm_offset_t map_addr;
340 	int error, rv, cow;
341 	int count;
342 	size_t copy_len;
343 	vm_object_t object;
344 	vm_offset_t file_addr;
345 
346 	object = vp->v_object;
347 	error = 0;
348 
349 	/*
350 	 * It's necessary to fail if the filsz + offset taken from the
351 	 * header is greater than the actual file pager object's size.
352 	 * If we were to allow this, then the vm_map_find() below would
353 	 * walk right off the end of the file object and into the ether.
354 	 *
355 	 * While I'm here, might as well check for something else that
356 	 * is invalid: filsz cannot be greater than memsz.
357 	 */
358 	if ((off_t)filsz + offset > vp->v_filesize || filsz > memsz) {
359 		uprintf("elf_load_section: truncated ELF file\n");
360 		return (ENOEXEC);
361 	}
362 
363 	map_addr = trunc_page((vm_offset_t)vmaddr);
364 	file_addr = trunc_page(offset);
365 
366 	/*
367 	 * We have two choices.  We can either clear the data in the last page
368 	 * of an oversized mapping, or we can start the anon mapping a page
369 	 * early and copy the initialized data into that first page.  We
370 	 * choose the second..
371 	 */
372 	if (memsz > filsz)
373 		map_len = trunc_page(offset+filsz) - file_addr;
374 	else
375 		map_len = round_page(offset+filsz) - file_addr;
376 
377 	if (map_len != 0) {
378 		vm_object_reference(object);
379 
380 		/* cow flags: don't dump readonly sections in core */
381 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
382 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
383 
384 		count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
385 		vm_map_lock(&vmspace->vm_map);
386 		rv = vm_map_insert(&vmspace->vm_map, &count,
387 				      object,
388 				      file_addr,	/* file offset */
389 				      map_addr,		/* virtual start */
390 				      map_addr + map_len,/* virtual end */
391 				      VM_MAPTYPE_NORMAL,
392 				      prot, VM_PROT_ALL,
393 				      cow);
394 		vm_map_unlock(&vmspace->vm_map);
395 		vm_map_entry_release(count);
396 		if (rv != KERN_SUCCESS) {
397 			vm_object_deallocate(object);
398 			return EINVAL;
399 		}
400 
401 		/* we can stop now if we've covered it all */
402 		if (memsz == filsz) {
403 			return 0;
404 		}
405 	}
406 
407 
408 	/*
409 	 * We have to get the remaining bit of the file into the first part
410 	 * of the oversized map segment.  This is normally because the .data
411 	 * segment in the file is extended to provide bss.  It's a neat idea
412 	 * to try and save a page, but it's a pain in the behind to implement.
413 	 */
414 	copy_len = (offset + filsz) - trunc_page(offset + filsz);
415 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
416 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
417 
418 	/* This had damn well better be true! */
419         if (map_len != 0) {
420 		count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
421 		vm_map_lock(&vmspace->vm_map);
422 		rv = vm_map_insert(&vmspace->vm_map, &count,
423 					NULL, 0,
424 					map_addr, map_addr + map_len,
425 					VM_MAPTYPE_NORMAL,
426 					VM_PROT_ALL, VM_PROT_ALL,
427 					0);
428 		vm_map_unlock(&vmspace->vm_map);
429 		vm_map_entry_release(count);
430 		if (rv != KERN_SUCCESS) {
431 			return EINVAL;
432 		}
433 	}
434 
435 	if (copy_len != 0) {
436 		vm_page_t m;
437 		struct sf_buf *sf;
438 
439 		m = vm_fault_object_page(object, trunc_page(offset + filsz),
440 					 VM_PROT_READ, 0, &error);
441 		if (m) {
442 			sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
443 			error = copyout((caddr_t)sf_buf_kva(sf),
444 					(caddr_t)map_addr, copy_len);
445 			sf_buf_free(sf);
446 			vm_page_unhold(m);
447 		}
448 		if (error) {
449 			return (error);
450 		}
451 	}
452 
453 	/*
454 	 * set it to the specified protection
455 	 */
456 	vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len,  prot,
457 		       FALSE);
458 
459 	return error;
460 }
461 
462 /*
463  * Load the file "file" into memory.  It may be either a shared object
464  * or an executable.
465  *
466  * The "addr" reference parameter is in/out.  On entry, it specifies
467  * the address where a shared object should be loaded.  If the file is
468  * an executable, this value is ignored.  On exit, "addr" specifies
469  * where the file was actually loaded.
470  *
471  * The "entry" reference parameter is out only.  On exit, it specifies
472  * the entry point for the loaded file.
473  */
474 static int
475 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
476 {
477 	struct {
478 		struct nlookupdata nd;
479 		struct vattr attr;
480 		struct image_params image_params;
481 	} *tempdata;
482 	const Elf_Ehdr *hdr = NULL;
483 	const Elf_Phdr *phdr = NULL;
484 	struct nlookupdata *nd;
485 	struct vmspace *vmspace = p->p_vmspace;
486 	struct vattr *attr;
487 	struct image_params *imgp;
488 	vm_prot_t prot;
489 	u_long rbase;
490 	u_long base_addr = 0;
491 	int error, i, numsegs;
492 
493 	tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
494 	nd = &tempdata->nd;
495 	attr = &tempdata->attr;
496 	imgp = &tempdata->image_params;
497 
498 	/*
499 	 * Initialize part of the common data
500 	 */
501 	imgp->proc = p;
502 	imgp->attr = attr;
503 	imgp->firstpage = NULL;
504 	imgp->image_header = NULL;
505 	imgp->vp = NULL;
506 
507 	error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW);
508 	if (error == 0)
509 		error = nlookup(nd);
510 	if (error == 0)
511 		error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
512 	nlookup_done(nd);
513 	if (error)
514 		goto fail;
515 
516 	/*
517 	 * Check permissions, modes, uid, etc on the file, and "open" it.
518 	 */
519 	error = exec_check_permissions(imgp);
520 	if (error) {
521 		vn_unlock(imgp->vp);
522 		goto fail;
523 	}
524 
525 	error = exec_map_first_page(imgp);
526 	/*
527 	 * Also make certain that the interpreter stays the same, so set
528 	 * its VTEXT flag, too.
529 	 */
530 	if (error == 0)
531 		imgp->vp->v_flag |= VTEXT;
532 	vn_unlock(imgp->vp);
533 	if (error)
534                 goto fail;
535 
536 	hdr = (const Elf_Ehdr *)imgp->image_header;
537 	if ((error = elf_check_header(hdr)) != 0)
538 		goto fail;
539 	if (hdr->e_type == ET_DYN)
540 		rbase = *addr;
541 	else if (hdr->e_type == ET_EXEC)
542 		rbase = 0;
543 	else {
544 		error = ENOEXEC;
545 		goto fail;
546 	}
547 
548 	/* Only support headers that fit within first page for now
549 	 * (multiplication of two Elf_Half fields will not overflow) */
550 	if ((hdr->e_phoff > PAGE_SIZE) ||
551 	    (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
552 		error = ENOEXEC;
553 		goto fail;
554 	}
555 
556 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
557 
558 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
559 		if (phdr[i].p_type == PT_LOAD) {	/* Loadable segment */
560 			prot = 0;
561 			if (phdr[i].p_flags & PF_X)
562   				prot |= VM_PROT_EXECUTE;
563 			if (phdr[i].p_flags & PF_W)
564   				prot |= VM_PROT_WRITE;
565 			if (phdr[i].p_flags & PF_R)
566   				prot |= VM_PROT_READ;
567 
568 			error = elf_load_section(
569 				    p, vmspace, imgp->vp,
570 				    phdr[i].p_offset,
571 				    (caddr_t)phdr[i].p_vaddr +
572 				    rbase,
573 				    phdr[i].p_memsz,
574 				    phdr[i].p_filesz, prot);
575 			if (error != 0)
576 				goto fail;
577 			/*
578 			 * Establish the base address if this is the
579 			 * first segment.
580 			 */
581 			if (numsegs == 0)
582   				base_addr = trunc_page(phdr[i].p_vaddr + rbase);
583 			numsegs++;
584 		}
585 	}
586 	*addr = base_addr;
587 	*entry=(unsigned long)hdr->e_entry + rbase;
588 
589 fail:
590 	if (imgp->firstpage)
591 		exec_unmap_first_page(imgp);
592 	if (imgp->vp) {
593 		vrele(imgp->vp);
594 		imgp->vp = NULL;
595 	}
596 	kfree(tempdata, M_TEMP);
597 
598 	return error;
599 }
600 
601 /*
602  * non static, as it can be overridden by start_init()
603  */
604 int fallback_elf_brand = -1;
605 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
606 		&fallback_elf_brand, -1,
607 		"ELF brand of last resort");
608 
609 static int can_exec_dyn = 1;
610 SYSCTL_INT(_kern, OID_AUTO, elf_exec_dyn, CTLFLAG_RW,
611 		&can_exec_dyn, 1,
612 		"ELF: can exec shared libraries");
613 
614 static int
615 exec_elf_imgact(struct image_params *imgp)
616 {
617 	const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
618 	const Elf_Phdr *phdr;
619 	Elf_Auxargs *elf_auxargs = NULL;
620 	struct vmspace *vmspace;
621 	vm_prot_t prot;
622 	u_long text_size = 0, data_size = 0, total_size = 0;
623 	u_long text_addr = 0, data_addr = 0;
624 	u_long seg_size, seg_addr;
625 	u_long addr, entry = 0, proghdr = 0;
626 	int error, i;
627 	const char *interp = NULL;
628 	const Elf_Note *abi_note = NULL;
629 	Elf_Brandinfo *brand_info = NULL;
630 	char *path;
631 
632 	error = 0;
633 
634 	/*
635 	 * Do we have a valid ELF header ?
636 	 * We allow execution of ET_EXEC and, if kern.elf_exec_dyn is 1, ET_DYN.
637 	 */
638 	if (elf_check_header(hdr) != 0 ||
639 	    (hdr->e_type != ET_EXEC && (!can_exec_dyn || hdr->e_type != ET_DYN)))
640 		return -1;
641 
642 	/*
643 	 * From here on down, we return an errno, not -1, as we've
644 	 * detected an ELF file.
645 	 */
646 
647 	if ((hdr->e_phoff > PAGE_SIZE) ||
648 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
649 		/* Only support headers in first page for now */
650 		return ENOEXEC;
651 	}
652 	phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
653 
654 	/*
655 	 * From this point on, we may have resources that need to be freed.
656 	 */
657 
658 	exec_new_vmspace(imgp, NULL);
659 
660 	/*
661 	 * Yeah, I'm paranoid.  There is every reason in the world to get
662 	 * VTEXT now since from here on out, there are places we can have
663 	 * a context switch.  Better safe than sorry; I really don't want
664 	 * the file to change while it's being loaded.
665 	 */
666 	vsetflags(imgp->vp, VTEXT);
667 
668 	vmspace = imgp->proc->p_vmspace;
669 
670 	for (i = 0; i < hdr->e_phnum; i++) {
671 		switch(phdr[i].p_type) {
672 
673 		case PT_LOAD:	/* Loadable segment */
674 			prot = 0;
675 			if (phdr[i].p_flags & PF_X)
676   				prot |= VM_PROT_EXECUTE;
677 			if (phdr[i].p_flags & PF_W)
678   				prot |= VM_PROT_WRITE;
679 			if (phdr[i].p_flags & PF_R)
680   				prot |= VM_PROT_READ;
681 
682 			if ((error = elf_load_section(imgp->proc,
683 						     vmspace, imgp->vp,
684   						     phdr[i].p_offset,
685   						     (caddr_t)phdr[i].p_vaddr,
686   						     phdr[i].p_memsz,
687   						     phdr[i].p_filesz, prot)) != 0)
688   				goto fail;
689 
690 			/*
691 			 * If this segment contains the program headers,
692 			 * remember their virtual address for the AT_PHDR
693 			 * aux entry. Static binaries don't usually include
694 			 * a PT_PHDR entry.
695 			 */
696 			if (phdr[i].p_offset == 0 &&
697 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
698 				<= phdr[i].p_filesz)
699 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
700 
701 			seg_addr = trunc_page(phdr[i].p_vaddr);
702 			seg_size = round_page(phdr[i].p_memsz +
703 				phdr[i].p_vaddr - seg_addr);
704 
705 			/*
706 			 * Is this .text or .data?  We can't use
707 			 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
708 			 * alpha terribly and possibly does other bad
709 			 * things so we stick to the old way of figuring
710 			 * it out:  If the segment contains the program
711 			 * entry point, it's a text segment, otherwise it
712 			 * is a data segment.
713 			 *
714 			 * Note that obreak() assumes that data_addr +
715 			 * data_size == end of data load area, and the ELF
716 			 * file format expects segments to be sorted by
717 			 * address.  If multiple data segments exist, the
718 			 * last one will be used.
719 			 */
720 			if (hdr->e_entry >= phdr[i].p_vaddr &&
721 			    hdr->e_entry < (phdr[i].p_vaddr +
722 			    phdr[i].p_memsz)) {
723 				text_size = seg_size;
724 				text_addr = seg_addr;
725 				entry = (u_long)hdr->e_entry;
726 			} else {
727 				data_size = seg_size;
728 				data_addr = seg_addr;
729 			}
730 			total_size += seg_size;
731 
732 			/*
733 			 * Check limits.  It should be safe to check the
734 			 * limits after loading the segment since we do
735 			 * not actually fault in all the segment's pages.
736 			 */
737 			if (data_size >
738 			    imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
739 			    text_size > maxtsiz ||
740 			    total_size >
741 			    imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
742 				error = ENOMEM;
743 				goto fail;
744 			}
745 			break;
746 	  	case PT_INTERP:	/* Path to interpreter */
747 			if (phdr[i].p_filesz > MAXPATHLEN ||
748 			    phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
749 				error = ENOEXEC;
750 				goto fail;
751 			}
752 			interp = imgp->image_header + phdr[i].p_offset;
753 			break;
754 		case PT_NOTE:	/* Check for .note.ABI-tag */
755 			if (brand_info == NULL)
756 				brand_info = elf_check_abi_note(imgp, &phdr[i]);
757 			break;
758 		case PT_PHDR: 	/* Program header table info */
759 			proghdr = phdr[i].p_vaddr;
760 			break;
761 		default:
762 			break;
763 		}
764 	}
765 
766 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
767 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
768 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
769 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
770 
771 	addr = ELF_RTLD_ADDR(vmspace);
772 
773 	imgp->entry_addr = entry;
774 
775 	/* We support three types of branding -- (1) the ELF EI_OSABI field
776 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
777 	 * branding w/in the ELF header, and (3) path of the `interp_path'
778 	 * field.  We should also look for an ".note.ABI-tag" ELF section now
779 	 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
780 	 */
781 
782 	/* If the executable has a brand, search for it in the brand list. */
783 	if (brand_info == NULL && hdr->e_ident[EI_OSABI] != ELFOSABI_NONE) {
784 		for (i = 0;  i < MAX_BRANDS;  i++) {
785 			Elf_Brandinfo *bi = elf_brand_list[i];
786 
787 			if (bi != NULL &&
788 			    (hdr->e_ident[EI_OSABI] == bi->brand
789 			    || 0 ==
790 			    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
791 			    bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
792 				brand_info = bi;
793 				break;
794 			}
795 		}
796 	}
797 
798 	/* Search for a recognized ABI. */
799 	if (brand_info == NULL && abi_note != NULL) {
800 	}
801 
802 	/*
803 	 * ELFOSABI_NONE == ELFOSABI_SYSV, so a SYSV binary misses all
804 	 * checks so far, since it is neither branded nor does it have
805 	 * an ABI note.  If the EI_OSABI field is ELFOSABI_NONE, assume
806 	 * it is svr4 and look for an entry in the elf_brand_list with
807 	 * match_abi_note == NULL.
808 	 */
809 	if (brand_info == NULL && hdr->e_ident[EI_OSABI] == ELFOSABI_NONE) {
810 		for (i = 0; i < MAX_BRANDS; i++) {
811 			Elf_Brandinfo *bi = elf_brand_list[i];
812 
813 			if (bi != NULL && bi->match_abi_note == NULL &&
814 			    ELFOSABI_SYSV == bi->brand) {
815 				brand_info = bi;
816 				break;
817 			}
818 		}
819 	}
820 
821 	/* Lacking a recognized ABI, search for a recognized interpreter. */
822 	if (brand_info == NULL && interp != NULL) {
823 		for (i = 0;  i < MAX_BRANDS;  i++) {
824 			Elf_Brandinfo *bi = elf_brand_list[i];
825 
826 			if (bi != NULL &&
827 			    strcmp(interp, bi->interp_path) == 0) {
828 				brand_info = bi;
829 				break;
830 			}
831 		}
832 	}
833 
834 	/* Lacking a recognized interpreter, try the default brand */
835 	if (brand_info == NULL) {
836 		for (i = 0; i < MAX_BRANDS; i++) {
837 			Elf_Brandinfo *bi = elf_brand_list[i];
838 
839 			if (bi != NULL && fallback_elf_brand == bi->brand) {
840 				brand_info = bi;
841 				break;
842 			}
843 		}
844 	}
845 
846 	if (brand_info == NULL) {
847 		uprintf("ELF binary type \"%u\" not known.\n",
848 		    hdr->e_ident[EI_OSABI]);
849 		error = ENOEXEC;
850 		goto fail;
851 	}
852 
853 	imgp->proc->p_sysent = brand_info->sysvec;
854 	if (interp != NULL) {
855 		path = kmalloc(MAXPATHLEN, M_TEMP, M_WAITOK);
856 	        ksnprintf(path, MAXPATHLEN, "%s%s",
857 			 brand_info->emul_path, interp);
858 		if ((error = elf_load_file(imgp->proc, path, &addr,
859 					   &imgp->entry_addr)) != 0) {
860 		        if ((error = elf_load_file(imgp->proc, interp, &addr,
861 						   &imgp->entry_addr)) != 0) {
862 			        uprintf("ELF interpreter %s not found\n", path);
863 				kfree(path, M_TEMP);
864 				goto fail;
865 			}
866                 }
867 		kfree(path, M_TEMP);
868 	} else {
869 		addr = 0;
870 	}
871 
872 	/*
873 	 * Construct auxargs table (used by the fixup routine)
874 	 */
875 	elf_auxargs = kmalloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
876 	elf_auxargs->execfd = -1;
877 	elf_auxargs->phdr = proghdr;
878 	elf_auxargs->phent = hdr->e_phentsize;
879 	elf_auxargs->phnum = hdr->e_phnum;
880 	elf_auxargs->pagesz = PAGE_SIZE;
881 	elf_auxargs->base = addr;
882 	elf_auxargs->flags = 0;
883 	elf_auxargs->entry = entry;
884 	elf_auxargs->trace = elf_trace;
885 
886 	imgp->auxargs = elf_auxargs;
887 	imgp->interpreted = 0;
888 
889 fail:
890 	return error;
891 }
892 
893 static int
894 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
895 {
896 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
897 	register_t *pos;
898 
899 	pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
900 
901 	if (args->trace) {
902 		AUXARGS_ENTRY(pos, AT_DEBUG, 1);
903 	}
904 	if (args->execfd != -1) {
905 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
906 	}
907 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
908 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
909 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
910 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
911 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
912 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
913 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
914 	AUXARGS_ENTRY(pos, AT_NULL, 0);
915 
916 	kfree(imgp->auxargs, M_TEMP);
917 	imgp->auxargs = NULL;
918 
919 	(*stack_base)--;
920 	suword(*stack_base, (long) imgp->args->argc);
921 	return 0;
922 }
923 
924 /*
925  * Code for generating ELF core dumps.
926  */
927 
928 typedef int (*segment_callback) (vm_map_entry_t, void *);
929 
930 /* Closure for cb_put_phdr(). */
931 struct phdr_closure {
932 	Elf_Phdr *phdr;		/* Program header to fill in (incremented) */
933 	Elf_Phdr *phdr_max;	/* Pointer bound for error check */
934 	Elf_Off offset;		/* Offset of segment in core file */
935 };
936 
937 /* Closure for cb_size_segment(). */
938 struct sseg_closure {
939 	int count;		/* Count of writable segments. */
940 	size_t vsize;		/* Total size of all writable segments. */
941 };
942 
943 /* Closure for cb_put_fp(). */
944 struct fp_closure {
945 	struct vn_hdr *vnh;
946 	struct vn_hdr *vnh_max;
947 	int count;
948 	struct stat *sb;
949 };
950 
951 typedef struct elf_buf {
952 	char	*buf;
953 	size_t	off;
954 	size_t	off_max;
955 } *elf_buf_t;
956 
957 static void *target_reserve(elf_buf_t target, size_t bytes, int *error);
958 
959 static int cb_put_phdr (vm_map_entry_t, void *);
960 static int cb_size_segment (vm_map_entry_t, void *);
961 static int cb_fpcount_segment(vm_map_entry_t, void *);
962 static int cb_put_fp(vm_map_entry_t, void *);
963 
964 
965 static int each_segment (struct proc *, segment_callback, void *, int);
966 static int elf_corehdr (struct lwp *, int, struct file *, struct ucred *,
967 			int, elf_buf_t);
968 enum putmode { WRITE, DRYRUN };
969 static int elf_puthdr (struct lwp *, elf_buf_t, int sig, enum putmode,
970 			int, struct file *);
971 static int elf_putallnotes(struct lwp *, elf_buf_t, int, enum putmode);
972 static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t);
973 
974 static int elf_putsigs(struct lwp *, elf_buf_t);
975 static int elf_puttextvp(struct proc *, elf_buf_t);
976 static int elf_putfiles(struct proc *, elf_buf_t, struct file *);
977 
978 extern int osreldate;
979 
980 int
981 elf_coredump(struct lwp *lp, int sig, struct vnode *vp, off_t limit)
982 {
983 	struct file *fp;
984 	int error;
985 
986 	if ((error = falloc(NULL, &fp, NULL)) != 0)
987 		return (error);
988 	fsetcred(fp, lp->lwp_proc->p_ucred);
989 
990 	/*
991 	 * XXX fixme.
992 	 */
993 	fp->f_type = DTYPE_VNODE;
994 	fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW;
995 	fp->f_ops = &vnode_fileops;
996 	fp->f_data = vp;
997 	vn_unlock(vp);
998 
999 	error = generic_elf_coredump(lp, sig, fp, limit);
1000 
1001 	fp->f_type = 0;
1002 	fp->f_flag = 0;
1003 	fp->f_ops = &badfileops;
1004 	fp->f_data = NULL;
1005 	fdrop(fp);
1006 	return (error);
1007 }
1008 
1009 int
1010 generic_elf_coredump(struct lwp *lp, int sig, struct file *fp, off_t limit)
1011 {
1012 	struct proc *p = lp->lwp_proc;
1013 	struct ucred *cred = p->p_ucred;
1014 	int error = 0;
1015 	struct sseg_closure seginfo;
1016 	struct elf_buf target;
1017 
1018 	if (!fp)
1019 		kprintf("can't dump core - null fp\n");
1020 
1021 	/*
1022 	 * Size the program segments
1023 	 */
1024 	seginfo.count = 0;
1025 	seginfo.vsize = 0;
1026 	each_segment(p, cb_size_segment, &seginfo, 1);
1027 
1028 	/*
1029 	 * Calculate the size of the core file header area by making
1030 	 * a dry run of generating it.  Nothing is written, but the
1031 	 * size is calculated.
1032 	 */
1033 	bzero(&target, sizeof(target));
1034 	elf_puthdr(lp, &target, sig, DRYRUN, seginfo.count, fp);
1035 
1036 	if (target.off + seginfo.vsize >= limit)
1037 		return (EFAULT);
1038 
1039 	/*
1040 	 * Allocate memory for building the header, fill it up,
1041 	 * and write it out.
1042 	 */
1043 	target.off_max = target.off;
1044 	target.off = 0;
1045 	target.buf = kmalloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO);
1046 
1047 	error = elf_corehdr(lp, sig, fp, cred, seginfo.count, &target);
1048 
1049 	/* Write the contents of all of the writable segments. */
1050 	if (error == 0) {
1051 		Elf_Phdr *php;
1052 		int i;
1053 		ssize_t nbytes;
1054 
1055 		php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1;
1056 		for (i = 0; i < seginfo.count; i++) {
1057 			error = fp_write(fp, (caddr_t)php->p_vaddr,
1058 					php->p_filesz, &nbytes, UIO_USERSPACE);
1059 			if (error != 0)
1060 				break;
1061 			php++;
1062 		}
1063 	}
1064 	kfree(target.buf, M_TEMP);
1065 
1066 	return error;
1067 }
1068 
1069 /*
1070  * A callback for each_segment() to write out the segment's
1071  * program header entry.
1072  */
1073 static int
1074 cb_put_phdr(vm_map_entry_t entry, void *closure)
1075 {
1076 	struct phdr_closure *phc = closure;
1077 	Elf_Phdr *phdr = phc->phdr;
1078 
1079 	if (phc->phdr == phc->phdr_max)
1080 		return EINVAL;
1081 
1082 	phc->offset = round_page(phc->offset);
1083 
1084 	phdr->p_type = PT_LOAD;
1085 	phdr->p_offset = phc->offset;
1086 	phdr->p_vaddr = entry->start;
1087 	phdr->p_paddr = 0;
1088 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1089 	phdr->p_align = PAGE_SIZE;
1090 	phdr->p_flags = 0;
1091 	if (entry->protection & VM_PROT_READ)
1092 		phdr->p_flags |= PF_R;
1093 	if (entry->protection & VM_PROT_WRITE)
1094 		phdr->p_flags |= PF_W;
1095 	if (entry->protection & VM_PROT_EXECUTE)
1096 		phdr->p_flags |= PF_X;
1097 
1098 	phc->offset += phdr->p_filesz;
1099 	++phc->phdr;
1100 	return 0;
1101 }
1102 
1103 /*
1104  * A callback for each_writable_segment() to gather information about
1105  * the number of segments and their total size.
1106  */
1107 static int
1108 cb_size_segment(vm_map_entry_t entry, void *closure)
1109 {
1110 	struct sseg_closure *ssc = closure;
1111 
1112 	++ssc->count;
1113 	ssc->vsize += entry->end - entry->start;
1114 	return 0;
1115 }
1116 
1117 /*
1118  * A callback for each_segment() to gather information about
1119  * the number of text segments.
1120  */
1121 static int
1122 cb_fpcount_segment(vm_map_entry_t entry, void *closure)
1123 {
1124 	int *count = closure;
1125 	struct vnode *vp;
1126 
1127 	if (entry->object.vm_object->type == OBJT_VNODE) {
1128 		vp = (struct vnode *)entry->object.vm_object->handle;
1129 		if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1130 			return 0;
1131 		++*count;
1132 	}
1133 	return 0;
1134 }
1135 
1136 static int
1137 cb_put_fp(vm_map_entry_t entry, void *closure)
1138 {
1139 	struct fp_closure *fpc = closure;
1140 	struct vn_hdr *vnh = fpc->vnh;
1141 	Elf_Phdr *phdr = &vnh->vnh_phdr;
1142 	struct vnode *vp;
1143 	int error;
1144 
1145 	/*
1146 	 * If an entry represents a vnode then write out a file handle.
1147 	 *
1148 	 * If we are checkpointing a checkpoint-restored program we do
1149 	 * NOT record the filehandle for the old checkpoint vnode (which
1150 	 * is mapped all over the place).  Instead we rely on the fact
1151 	 * that a checkpoint-restored program does not mmap() the checkpt
1152 	 * vnode NOCORE, so its contents will be written out to the
1153 	 * new checkpoint file.  This is necessary because the 'old'
1154 	 * checkpoint file is typically destroyed when a new one is created
1155 	 * and thus cannot be used to restore the new checkpoint.
1156 	 *
1157 	 * Theoretically we could create a chain of checkpoint files and
1158 	 * operate the checkpointing operation kinda like an incremental
1159 	 * checkpoint, but a checkpoint restore would then likely wind up
1160 	 * referencing many prior checkpoint files and that is a bit over
1161 	 * the top for the purpose of the checkpoint API.
1162 	 */
1163 	if (entry->object.vm_object->type == OBJT_VNODE) {
1164 		vp = (struct vnode *)entry->object.vm_object->handle;
1165 		if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1166 			return 0;
1167 		if (vnh == fpc->vnh_max)
1168 			return EINVAL;
1169 
1170 		if (vp->v_mount)
1171 			vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1172 		error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid);
1173 		if (error) {
1174 			char *freepath, *fullpath;
1175 
1176 			if (vn_fullpath(curproc, vp, &fullpath, &freepath)) {
1177 				kprintf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp);
1178 			} else {
1179 				kprintf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath);
1180 				kfree(freepath, M_TEMP);
1181 			}
1182 			error = 0;
1183 		}
1184 
1185 		phdr->p_type = PT_LOAD;
1186 		phdr->p_offset = 0;        /* not written to core */
1187 		phdr->p_vaddr = entry->start;
1188 		phdr->p_paddr = 0;
1189 		phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1190 		phdr->p_align = PAGE_SIZE;
1191 		phdr->p_flags = 0;
1192 		if (entry->protection & VM_PROT_READ)
1193 			phdr->p_flags |= PF_R;
1194 		if (entry->protection & VM_PROT_WRITE)
1195 			phdr->p_flags |= PF_W;
1196 		if (entry->protection & VM_PROT_EXECUTE)
1197 			phdr->p_flags |= PF_X;
1198 		++fpc->vnh;
1199 		++fpc->count;
1200 	}
1201 	return 0;
1202 }
1203 
1204 /*
1205  * For each writable segment in the process's memory map, call the given
1206  * function with a pointer to the map entry and some arbitrary
1207  * caller-supplied data.
1208  */
1209 static int
1210 each_segment(struct proc *p, segment_callback func, void *closure, int writable)
1211 {
1212 	int error = 0;
1213 	vm_map_t map = &p->p_vmspace->vm_map;
1214 	vm_map_entry_t entry;
1215 
1216 	for (entry = map->header.next; error == 0 && entry != &map->header;
1217 	    entry = entry->next) {
1218 		vm_object_t obj;
1219 
1220 		/*
1221 		 * Don't dump inaccessible mappings, deal with legacy
1222 		 * coredump mode.
1223 		 *
1224 		 * Note that read-only segments related to the elf binary
1225 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1226 		 * need to arbitrarily ignore such segments.
1227 		 */
1228 		if (elf_legacy_coredump) {
1229 			if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW)
1230 				continue;
1231 		} else {
1232 			if (writable && (entry->protection & VM_PROT_ALL) == 0)
1233 				continue;
1234 		}
1235 
1236 		/*
1237 		 * Dont include memory segment in the coredump if
1238 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1239 		 * madvise(2).
1240 		 *
1241 		 * Currently we only dump normal VM object maps.  We do
1242 		 * not dump submaps or virtual page tables.
1243 		 */
1244 		if (writable && (entry->eflags & MAP_ENTRY_NOCOREDUMP))
1245 			continue;
1246 		if (entry->maptype != VM_MAPTYPE_NORMAL)
1247 			continue;
1248 		if ((obj = entry->object.vm_object) == NULL)
1249 			continue;
1250 
1251 		/* Find the deepest backing object. */
1252 		while (obj->backing_object != NULL)
1253 			obj = obj->backing_object;
1254 
1255 		/* Ignore memory-mapped devices and such things. */
1256 		if (obj->type != OBJT_DEFAULT &&
1257 		    obj->type != OBJT_SWAP &&
1258 		    obj->type != OBJT_VNODE)
1259 			continue;
1260 
1261 		error = (*func)(entry, closure);
1262 	}
1263 	return error;
1264 }
1265 
1266 static
1267 void *
1268 target_reserve(elf_buf_t target, size_t bytes, int *error)
1269 {
1270     void *res = NULL;
1271 
1272     if (target->buf) {
1273 	    if (target->off + bytes > target->off_max)
1274 		    *error = EINVAL;
1275 	    else
1276 		    res = target->buf + target->off;
1277     }
1278     target->off += bytes;
1279     return (res);
1280 }
1281 
1282 /*
1283  * Write the core file header to the file, including padding up to
1284  * the page boundary.
1285  */
1286 static int
1287 elf_corehdr(struct lwp *lp, int sig, struct file *fp, struct ucred *cred,
1288 	    int numsegs, elf_buf_t target)
1289 {
1290 	int error;
1291 	ssize_t nbytes;
1292 
1293 	/*
1294 	 * Fill in the header.  The fp is passed so we can detect and flag
1295 	 * a checkpoint file pointer within the core file itself, because
1296 	 * it may not be restored from the same file handle.
1297 	 */
1298 	error = elf_puthdr(lp, target, sig, WRITE, numsegs, fp);
1299 
1300 	/* Write it to the core file. */
1301 	if (error == 0) {
1302 		error = fp_write(fp, target->buf, target->off, &nbytes,
1303 				 UIO_SYSSPACE);
1304 	}
1305 	return error;
1306 }
1307 
1308 static int
1309 elf_puthdr(struct lwp *lp, elf_buf_t target, int sig, enum putmode mode,
1310     int numsegs, struct file *fp)
1311 {
1312 	struct proc *p = lp->lwp_proc;
1313 	int error = 0;
1314 	size_t phoff;
1315 	size_t noteoff;
1316 	size_t notesz;
1317 	Elf_Ehdr *ehdr;
1318 	Elf_Phdr *phdr;
1319 
1320 	ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error);
1321 
1322 	phoff = target->off;
1323 	phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error);
1324 
1325 	noteoff = target->off;
1326 	if (error == 0)
1327 		elf_putallnotes(lp, target, sig, mode);
1328 	notesz = target->off - noteoff;
1329 
1330 	/*
1331 	 * put extra cruft for dumping process state here
1332 	 *  - we really want it be before all the program
1333 	 *    mappings
1334 	 *  - we just need to update the offset accordingly
1335 	 *    and GDB will be none the wiser.
1336 	 */
1337 	if (error == 0)
1338 		error = elf_puttextvp(p, target);
1339 	if (error == 0)
1340 		error = elf_putsigs(lp, target);
1341 	if (error == 0)
1342 		error = elf_putfiles(p, target, fp);
1343 
1344 	/*
1345 	 * Align up to a page boundary for the program segments.  The
1346 	 * actual data will be written to the outptu file, not to elf_buf_t,
1347 	 * so we do not have to do any further bounds checking.
1348 	 */
1349 	target->off = round_page(target->off);
1350 	if (error == 0 && ehdr != NULL) {
1351 		/*
1352 		 * Fill in the ELF header.
1353 		 */
1354 		ehdr->e_ident[EI_MAG0] = ELFMAG0;
1355 		ehdr->e_ident[EI_MAG1] = ELFMAG1;
1356 		ehdr->e_ident[EI_MAG2] = ELFMAG2;
1357 		ehdr->e_ident[EI_MAG3] = ELFMAG3;
1358 		ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1359 		ehdr->e_ident[EI_DATA] = ELF_DATA;
1360 		ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1361 		ehdr->e_ident[EI_OSABI] = ELFOSABI_NONE;
1362 		ehdr->e_ident[EI_ABIVERSION] = 0;
1363 		ehdr->e_ident[EI_PAD] = 0;
1364 		ehdr->e_type = ET_CORE;
1365 		ehdr->e_machine = ELF_ARCH;
1366 		ehdr->e_version = EV_CURRENT;
1367 		ehdr->e_entry = 0;
1368 		ehdr->e_phoff = phoff;
1369 		ehdr->e_flags = 0;
1370 		ehdr->e_ehsize = sizeof(Elf_Ehdr);
1371 		ehdr->e_phentsize = sizeof(Elf_Phdr);
1372 		ehdr->e_phnum = numsegs + 1;
1373 		ehdr->e_shentsize = sizeof(Elf_Shdr);
1374 		ehdr->e_shnum = 0;
1375 		ehdr->e_shstrndx = SHN_UNDEF;
1376 	}
1377 	if (error == 0 && phdr != NULL) {
1378 		/*
1379 		 * Fill in the program header entries.
1380 		 */
1381 		struct phdr_closure phc;
1382 
1383 		/* The note segement. */
1384 		phdr->p_type = PT_NOTE;
1385 		phdr->p_offset = noteoff;
1386 		phdr->p_vaddr = 0;
1387 		phdr->p_paddr = 0;
1388 		phdr->p_filesz = notesz;
1389 		phdr->p_memsz = 0;
1390 		phdr->p_flags = 0;
1391 		phdr->p_align = 0;
1392 		++phdr;
1393 
1394 		/* All the writable segments from the program. */
1395 		phc.phdr = phdr;
1396 		phc.phdr_max = phdr + numsegs;
1397 		phc.offset = target->off;
1398 		each_segment(p, cb_put_phdr, &phc, 1);
1399 	}
1400 	return (error);
1401 }
1402 
1403 /*
1404  * Append core dump notes to target ELF buffer or simply update target size
1405  * if dryrun selected.
1406  */
1407 static int
1408 elf_putallnotes(struct lwp *corelp, elf_buf_t target, int sig,
1409     enum putmode mode)
1410 {
1411 	struct proc *p = corelp->lwp_proc;
1412 	int error;
1413 	struct {
1414 		prstatus_t status;
1415 		prfpregset_t fpregs;
1416 		prpsinfo_t psinfo;
1417 	} *tmpdata;
1418 	prstatus_t *status;
1419 	prfpregset_t *fpregs;
1420 	prpsinfo_t *psinfo;
1421 	struct lwp *lp;
1422 
1423 	/*
1424 	 * Allocate temporary storage for notes on heap to avoid stack overflow.
1425 	 */
1426 	if (mode != DRYRUN) {
1427 		tmpdata = kmalloc(sizeof(*tmpdata), M_TEMP, M_ZERO | M_WAITOK);
1428 		status = &tmpdata->status;
1429 		fpregs = &tmpdata->fpregs;
1430 		psinfo = &tmpdata->psinfo;
1431 	} else {
1432 		tmpdata = NULL;
1433 		status = NULL;
1434 		fpregs = NULL;
1435 		psinfo = NULL;
1436 	}
1437 
1438 	/*
1439 	 * Append LWP-agnostic note.
1440 	 */
1441 	if (mode != DRYRUN) {
1442 		psinfo->pr_version = PRPSINFO_VERSION;
1443 		psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1444 		strncpy(psinfo->pr_fname, p->p_comm,
1445 			sizeof(psinfo->pr_fname) - 1);
1446 		/*
1447 		 * XXX - We don't fill in the command line arguments
1448 		 * properly yet.
1449 		 */
1450 		strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
1451 	}
1452 	error =
1453 	    elf_putnote(target, "CORE", NT_PRPSINFO, psinfo, sizeof *psinfo);
1454 	if (error)
1455 		goto exit;
1456 
1457 	/*
1458 	 * Append first note for LWP that triggered core so that it is
1459 	 * the selected one when the debugger starts.
1460 	 */
1461 	if (mode != DRYRUN) {
1462 		status->pr_version = PRSTATUS_VERSION;
1463 		status->pr_statussz = sizeof(prstatus_t);
1464 		status->pr_gregsetsz = sizeof(gregset_t);
1465 		status->pr_fpregsetsz = sizeof(fpregset_t);
1466 		status->pr_osreldate = osreldate;
1467 		status->pr_cursig = sig;
1468 		/*
1469 		 * XXX GDB needs unique pr_pid for each LWP and does not
1470 		 * not support pr_pid==0 but lwp_tid can be 0, so hack unique
1471 		 * value.
1472 		 */
1473 		status->pr_pid = corelp->lwp_tid;
1474 		fill_regs(corelp, &status->pr_reg);
1475 		fill_fpregs(corelp, fpregs);
1476 	}
1477 	error =
1478 	    elf_putnote(target, "CORE", NT_PRSTATUS, status, sizeof *status);
1479 	if (error)
1480 		goto exit;
1481 	error =
1482 	    elf_putnote(target, "CORE", NT_FPREGSET, fpregs, sizeof *fpregs);
1483 	if (error)
1484 		goto exit;
1485 
1486 	/*
1487 	 * Then append notes for other LWPs.
1488 	 */
1489 	FOREACH_LWP_IN_PROC(lp, p) {
1490 		if (lp == corelp)
1491 			continue;
1492 		/* skip lwps being created */
1493 		if (lp->lwp_thread == NULL)
1494 			continue;
1495 		if (mode != DRYRUN) {
1496 			status->pr_pid = lp->lwp_tid;
1497 			fill_regs(lp, &status->pr_reg);
1498 			fill_fpregs(lp, fpregs);
1499 		}
1500 		error = elf_putnote(target, "CORE", NT_PRSTATUS,
1501 					status, sizeof *status);
1502 		if (error)
1503 			goto exit;
1504 		error = elf_putnote(target, "CORE", NT_FPREGSET,
1505 					fpregs, sizeof *fpregs);
1506 		if (error)
1507 			goto exit;
1508 	}
1509 
1510 exit:
1511 	if (tmpdata != NULL)
1512 		kfree(tmpdata, M_TEMP);
1513 	return (error);
1514 }
1515 
1516 /*
1517  * Generate a note sub-structure.
1518  *
1519  * NOTE: 4-byte alignment.
1520  */
1521 static int
1522 elf_putnote(elf_buf_t target, const char *name, int type,
1523 	    const void *desc, size_t descsz)
1524 {
1525 	int error = 0;
1526 	char *dst;
1527 	Elf_Note note;
1528 
1529 	note.n_namesz = strlen(name) + 1;
1530 	note.n_descsz = descsz;
1531 	note.n_type = type;
1532 	dst = target_reserve(target, sizeof(note), &error);
1533 	if (dst != NULL)
1534 		bcopy(&note, dst, sizeof note);
1535 	dst = target_reserve(target, note.n_namesz, &error);
1536 	if (dst != NULL)
1537 		bcopy(name, dst, note.n_namesz);
1538 	target->off = roundup2(target->off, sizeof(Elf_Word));
1539 	dst = target_reserve(target, note.n_descsz, &error);
1540 	if (dst != NULL)
1541 		bcopy(desc, dst, note.n_descsz);
1542 	target->off = roundup2(target->off, sizeof(Elf_Word));
1543 	return(error);
1544 }
1545 
1546 
1547 static int
1548 elf_putsigs(struct lwp *lp, elf_buf_t target)
1549 {
1550 	/* XXX lwp handle more than one lwp */
1551 	struct proc *p = lp->lwp_proc;
1552 	int error = 0;
1553 	struct ckpt_siginfo *csi;
1554 
1555 	csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error);
1556 	if (csi) {
1557 		csi->csi_ckptpisz = sizeof(struct ckpt_siginfo);
1558 		bcopy(p->p_sigacts, &csi->csi_sigacts, sizeof(*p->p_sigacts));
1559 		bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval));
1560 		bcopy(&lp->lwp_sigmask, &csi->csi_sigmask,
1561 			sizeof(sigset_t));
1562 		csi->csi_sigparent = p->p_sigparent;
1563 	}
1564 	return(error);
1565 }
1566 
1567 static int
1568 elf_putfiles(struct proc *p, elf_buf_t target, struct file *ckfp)
1569 {
1570 	int error = 0;
1571 	int i;
1572 	struct ckpt_filehdr *cfh = NULL;
1573 	struct ckpt_fileinfo *cfi;
1574 	struct file *fp;
1575 	struct vnode *vp;
1576 	/*
1577 	 * the duplicated loop is gross, but it was the only way
1578 	 * to eliminate uninitialized variable warnings
1579 	 */
1580 	cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error);
1581 	if (cfh) {
1582 		cfh->cfh_nfiles = 0;
1583 	}
1584 
1585 	/*
1586 	 * ignore STDIN/STDERR/STDOUT.
1587 	 */
1588 	for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) {
1589 		fp = holdfp(p->p_fd, i, -1);
1590 		if (fp == NULL)
1591 			continue;
1592 		/*
1593 		 * XXX Only checkpoint vnodes for now.
1594 		 */
1595 		if (fp->f_type != DTYPE_VNODE) {
1596 			fdrop(fp);
1597 			continue;
1598 		}
1599 		cfi = target_reserve(target, sizeof(struct ckpt_fileinfo),
1600 					&error);
1601 		if (cfi == NULL) {
1602 			fdrop(fp);
1603 			continue;
1604 		}
1605 		cfi->cfi_index = -1;
1606 		cfi->cfi_type = fp->f_type;
1607 		cfi->cfi_flags = fp->f_flag;
1608 		cfi->cfi_offset = fp->f_offset;
1609 		cfi->cfi_ckflags = 0;
1610 
1611 		if (fp == ckfp)
1612 			cfi->cfi_ckflags |= CKFIF_ISCKPTFD;
1613 		/* f_count and f_msgcount should not be saved/restored */
1614 		/* XXX save cred info */
1615 
1616 		switch(fp->f_type) {
1617 		case DTYPE_VNODE:
1618 			vp = (struct vnode *)fp->f_data;
1619 			/*
1620 			 * it looks like a bug in ptrace is marking
1621 			 * a non-vnode as a vnode - until we find the
1622 			 * root cause this will at least prevent
1623 			 * further panics from truss
1624 			 */
1625 			if (vp == NULL || vp->v_mount == NULL)
1626 				break;
1627 			cfh->cfh_nfiles++;
1628 			cfi->cfi_index = i;
1629 			cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1630 			error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid);
1631 			break;
1632 		default:
1633 			break;
1634 		}
1635 		fdrop(fp);
1636 	}
1637 	return(error);
1638 }
1639 
1640 static int
1641 elf_puttextvp(struct proc *p, elf_buf_t target)
1642 {
1643 	int error = 0;
1644 	int *vn_count;
1645 	struct fp_closure fpc;
1646 	struct ckpt_vminfo *vminfo;
1647 
1648 	vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error);
1649 	if (vminfo != NULL) {
1650 		vminfo->cvm_dsize = p->p_vmspace->vm_dsize;
1651 		vminfo->cvm_tsize = p->p_vmspace->vm_tsize;
1652 		vminfo->cvm_daddr = p->p_vmspace->vm_daddr;
1653 		vminfo->cvm_taddr = p->p_vmspace->vm_taddr;
1654 	}
1655 
1656 	fpc.count = 0;
1657 	vn_count = target_reserve(target, sizeof(int), &error);
1658 	if (target->buf != NULL) {
1659 		fpc.vnh = (struct vn_hdr *)(target->buf + target->off);
1660 		fpc.vnh_max = fpc.vnh +
1661 			(target->off_max - target->off) / sizeof(struct vn_hdr);
1662 		error = each_segment(p, cb_put_fp, &fpc, 0);
1663 		if (vn_count)
1664 			*vn_count = fpc.count;
1665 	} else {
1666 		error = each_segment(p, cb_fpcount_segment, &fpc.count, 0);
1667 	}
1668 	target->off += fpc.count * sizeof(struct vn_hdr);
1669 	return(error);
1670 }
1671 
1672 
1673 /*
1674  * Tell kern_execve.c about it, with a little help from the linker.
1675  */
1676 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1677 EXEC_SET_ORDERED(elf, elf_execsw, SI_ORDER_FIRST);
1678