Make sure the ELF header size is not too large. This fixes a potential over
[dragonfly.git] / sys / kern / imgact_elf.c
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 1.18 2004/03/29 17:17:09 drhodus 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/namei.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
55 #include <vm/vm.h>
56 #include <vm/vm_kern.h>
57 #include <vm/vm_param.h>
58 #include <vm/pmap.h>
59 #include <sys/lock.h>
60 #include <vm/vm_map.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_extern.h>
63
64 #include <machine/elf.h>
65 #include <machine/md_var.h>
66 #include <sys/mount.h>
67 #include <sys/ckpt.h>
68 #define OLD_EI_BRAND    8
69
70 __ElfType(Brandinfo);
71 __ElfType(Auxargs);
72
73 static int elf_check_header (const Elf_Ehdr *hdr);
74 static int elf_freebsd_fixup (register_t **stack_base,
75     struct image_params *imgp);
76 static int elf_load_file (struct proc *p, const char *file, u_long *addr,
77     u_long *entry);
78 static int elf_load_section (struct proc *p,
79     struct vmspace *vmspace, struct vnode *vp,
80     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
81     vm_prot_t prot);
82 static int exec_elf_imgact (struct image_params *imgp);
83
84 static int elf_trace = 0;
85 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
86 static int elf_legacy_coredump = 0;
87 SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
88     &elf_legacy_coredump, 0, "");
89
90 static struct sysentvec elf_freebsd_sysvec = {
91         SYS_MAXSYSCALL,
92         sysent,
93         0,
94         0,
95         0,
96         0,
97         0,
98         0,
99         elf_freebsd_fixup,
100         sendsig,
101         sigcode,
102         &szsigcode,
103         0,
104         "FreeBSD ELF",
105         elf_coredump,
106         NULL,
107         MINSIGSTKSZ
108 };
109
110 static Elf_Brandinfo freebsd_brand_info = {
111                                                 ELFOSABI_FREEBSD,
112                                                 "FreeBSD",
113                                                 "",
114                                                 "/usr/libexec/ld-elf.so.1",
115                                                 &elf_freebsd_sysvec
116                                           };
117 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
118                                                         &freebsd_brand_info,
119                                                         NULL, NULL, NULL,
120                                                         NULL, NULL, NULL, NULL
121                                                     };
122
123 int
124 elf_insert_brand_entry(Elf_Brandinfo *entry)
125 {
126         int i;
127
128         for (i=1; i<MAX_BRANDS; i++) {
129                 if (elf_brand_list[i] == NULL) {
130                         elf_brand_list[i] = entry;
131                         break;
132                 }
133         }
134         if (i == MAX_BRANDS)
135                 return -1;
136         return 0;
137 }
138
139 int
140 elf_remove_brand_entry(Elf_Brandinfo *entry)
141 {
142         int i;
143
144         for (i=1; i<MAX_BRANDS; i++) {
145                 if (elf_brand_list[i] == entry) {
146                         elf_brand_list[i] = NULL;
147                         break;
148                 }
149         }
150         if (i == MAX_BRANDS)
151                 return -1;
152         return 0;
153 }
154
155 int
156 elf_brand_inuse(Elf_Brandinfo *entry)
157 {
158         struct proc *p;
159         int rval = FALSE;
160
161         FOREACH_PROC_IN_SYSTEM(p) {
162                 if (p->p_sysent == entry->sysvec) {
163                         rval = TRUE;
164                         break;
165                 }
166         }
167
168         return (rval);
169 }
170
171 static int
172 elf_check_header(const Elf_Ehdr *hdr)
173 {
174         if (!IS_ELF(*hdr) ||
175             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
176             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
177             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
178             hdr->e_phentsize != sizeof(Elf_Phdr) ||
179             hdr->e_ehsize != sizeof(Elf_Ehdr) ||
180             hdr->e_version != ELF_TARG_VER)
181                 return ENOEXEC;
182
183         if (!ELF_MACHINE_OK(hdr->e_machine))
184                 return ENOEXEC;
185
186         return 0;
187 }
188
189 static int
190 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, 
191                  vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 
192                  vm_prot_t prot)
193 {
194         size_t map_len;
195         vm_offset_t map_addr;
196         int error, rv, cow;
197         int count;
198         size_t copy_len;
199         vm_object_t object;
200         vm_offset_t file_addr;
201         vm_offset_t data_buf = 0;
202
203         VOP_GETVOBJECT(vp, &object);
204         error = 0;
205
206         /*
207          * It's necessary to fail if the filsz + offset taken from the
208          * header is greater than the actual file pager object's size.
209          * If we were to allow this, then the vm_map_find() below would
210          * walk right off the end of the file object and into the ether.
211          *
212          * While I'm here, might as well check for something else that
213          * is invalid: filsz cannot be greater than memsz.
214          */
215         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
216             filsz > memsz) {
217                 uprintf("elf_load_section: truncated ELF file\n");
218                 return (ENOEXEC);
219         }
220
221         map_addr = trunc_page((vm_offset_t)vmaddr);
222         file_addr = trunc_page(offset);
223
224         /*
225          * We have two choices.  We can either clear the data in the last page
226          * of an oversized mapping, or we can start the anon mapping a page
227          * early and copy the initialized data into that first page.  We
228          * choose the second..
229          */
230         if (memsz > filsz)
231                 map_len = trunc_page(offset+filsz) - file_addr;
232         else
233                 map_len = round_page(offset+filsz) - file_addr;
234
235         if (map_len != 0) {
236                 vm_object_reference(object);
237
238                 /* cow flags: don't dump readonly sections in core */
239                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
240                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
241
242                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
243                 vm_map_lock(&vmspace->vm_map);
244                 rv = vm_map_insert(&vmspace->vm_map, &count,
245                                       object,
246                                       file_addr,        /* file offset */
247                                       map_addr,         /* virtual start */
248                                       map_addr + map_len,/* virtual end */
249                                       prot,
250                                       VM_PROT_ALL,
251                                       cow);
252                 vm_map_unlock(&vmspace->vm_map);
253                 vm_map_entry_release(count);
254                 if (rv != KERN_SUCCESS) {
255                         vm_object_deallocate(object);
256                         return EINVAL;
257                 }
258
259                 /* we can stop now if we've covered it all */
260                 if (memsz == filsz) {
261                         return 0;
262                 }
263         }
264
265
266         /*
267          * We have to get the remaining bit of the file into the first part
268          * of the oversized map segment.  This is normally because the .data
269          * segment in the file is extended to provide bss.  It's a neat idea
270          * to try and save a page, but it's a pain in the behind to implement.
271          */
272         copy_len = (offset + filsz) - trunc_page(offset + filsz);
273         map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
274         map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
275
276         /* This had damn well better be true! */
277         if (map_len != 0) {
278                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
279                 vm_map_lock(&vmspace->vm_map);
280                 rv = vm_map_insert(&vmspace->vm_map, &count,
281                                         NULL, 0,
282                                         map_addr, map_addr + map_len,
283                                         VM_PROT_ALL, VM_PROT_ALL, 0);
284                 vm_map_unlock(&vmspace->vm_map);
285                 vm_map_entry_release(count);
286                 if (rv != KERN_SUCCESS) {
287                         return EINVAL; 
288                 }
289         }
290
291         if (copy_len != 0) {
292                 vm_object_reference(object);
293                 rv = vm_map_find(exec_map,
294                                  object, 
295                                  trunc_page(offset + filsz),
296                                  &data_buf,
297                                  PAGE_SIZE,
298                                  TRUE,
299                                  VM_PROT_READ,
300                                  VM_PROT_ALL,
301                                  MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
302                 if (rv != KERN_SUCCESS) {
303                         vm_object_deallocate(object);
304                         return EINVAL;
305                 }
306
307                 /* send the page fragment to user space */
308                 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
309                 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
310                 if (error) {
311                         return (error);
312                 }
313         }
314
315         /*
316          * set it to the specified protection
317          */
318         vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len,  prot,
319                        FALSE);
320
321         return error;
322 }
323
324 /*
325  * Load the file "file" into memory.  It may be either a shared object
326  * or an executable.
327  *
328  * The "addr" reference parameter is in/out.  On entry, it specifies
329  * the address where a shared object should be loaded.  If the file is
330  * an executable, this value is ignored.  On exit, "addr" specifies
331  * where the file was actually loaded.
332  *
333  * The "entry" reference parameter is out only.  On exit, it specifies
334  * the entry point for the loaded file.
335  */
336 static int
337 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
338 {
339         struct {
340                 struct nameidata nd;
341                 struct vattr attr;
342                 struct image_params image_params;
343         } *tempdata;
344         const Elf_Ehdr *hdr = NULL;
345         const Elf_Phdr *phdr = NULL;
346         struct nameidata *nd;
347         struct vmspace *vmspace = p->p_vmspace;
348         struct vattr *attr;
349         struct image_params *imgp;
350         vm_prot_t prot;
351         u_long rbase;
352         u_long base_addr = 0;
353         int error, i, numsegs;
354         struct thread *td = p->p_thread;
355
356         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
357         nd = &tempdata->nd;
358         attr = &tempdata->attr;
359         imgp = &tempdata->image_params;
360
361         /*
362          * Initialize part of the common data
363          */
364         imgp->proc = p;
365         imgp->attr = attr;
366         imgp->firstpage = NULL;
367         imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
368
369         if (imgp->image_header == NULL) {
370                 nd->ni_vp = NULL;
371                 error = ENOMEM;
372                 goto fail;
373         }
374
375         NDINIT(nd, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW,
376             UIO_SYSSPACE, file, td);
377                          
378         if ((error = namei(nd)) != 0) {
379                 nd->ni_vp = NULL;
380                 goto fail;
381         }
382         NDFREE(nd, NDF_ONLY_PNBUF);
383         imgp->vp = nd->ni_vp;
384
385         /*
386          * Check permissions, modes, uid, etc on the file, and "open" it.
387          */
388         error = exec_check_permissions(imgp);
389         if (error) {
390                 VOP_UNLOCK(nd->ni_vp, NULL, 0, td);
391                 goto fail;
392         }
393
394         error = exec_map_first_page(imgp);
395         /*
396          * Also make certain that the interpreter stays the same, so set
397          * its VTEXT flag, too.
398          */
399         if (error == 0)
400                 nd->ni_vp->v_flag |= VTEXT;
401         VOP_UNLOCK(nd->ni_vp, NULL, 0, td);
402         if (error)
403                 goto fail;
404
405         hdr = (const Elf_Ehdr *)imgp->image_header;
406         if ((error = elf_check_header(hdr)) != 0)
407                 goto fail;
408         if (hdr->e_type == ET_DYN)
409                 rbase = *addr;
410         else if (hdr->e_type == ET_EXEC)
411                 rbase = 0;
412         else {
413                 error = ENOEXEC;
414                 goto fail;
415         }
416
417         /* Only support headers that fit within first page for now
418          * (multiplication of two Elf_Half fields will not overflow) */
419         if ((hdr->e_phoff > PAGE_SIZE) ||
420             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
421                 error = ENOEXEC;
422                 goto fail;
423         }
424
425         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
426
427         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
428                 if (phdr[i].p_type == PT_LOAD) {        /* Loadable segment */
429                         prot = 0;
430                         if (phdr[i].p_flags & PF_X)
431                                 prot |= VM_PROT_EXECUTE;
432                         if (phdr[i].p_flags & PF_W)
433                                 prot |= VM_PROT_WRITE;
434                         if (phdr[i].p_flags & PF_R)
435                                 prot |= VM_PROT_READ;
436
437                         error = elf_load_section(
438                                     p, vmspace, nd->ni_vp,
439                                     phdr[i].p_offset,
440                                     (caddr_t)phdr[i].p_vaddr +
441                                     rbase,
442                                     phdr[i].p_memsz,
443                                     phdr[i].p_filesz, prot);
444                         if (error != 0)
445                                 goto fail;
446                         /*
447                          * Establish the base address if this is the
448                          * first segment.
449                          */
450                         if (numsegs == 0)
451                                 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
452                         numsegs++;
453                 }
454         }
455         *addr = base_addr;
456         *entry=(unsigned long)hdr->e_entry + rbase;
457
458 fail:
459         if (imgp->firstpage)
460                 exec_unmap_first_page(imgp);
461         if (imgp->image_header)
462                 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
463                         PAGE_SIZE);
464         if (nd->ni_vp)
465                 vrele(nd->ni_vp);
466
467         free(tempdata, M_TEMP);
468
469         return error;
470 }
471
472 /*
473  * non static, as it can be overridden by start_init()
474  */
475 int fallback_elf_brand = -1;
476 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
477                 &fallback_elf_brand, -1,
478                 "ELF brand of last resort");
479
480 static int
481 exec_elf_imgact(struct image_params *imgp)
482 {
483         const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
484         const Elf_Phdr *phdr;
485         Elf_Auxargs *elf_auxargs = NULL;
486         struct vmspace *vmspace;
487         vm_prot_t prot;
488         u_long text_size = 0, data_size = 0, total_size = 0;
489         u_long text_addr = 0, data_addr = 0;
490         u_long seg_size, seg_addr;
491         u_long addr, entry = 0, proghdr = 0;
492         int error, i;
493         const char *interp = NULL;
494         Elf_Brandinfo *brand_info;
495         char *path;
496         lwkt_tokref ilock;
497
498         error = 0;
499
500         /*
501          * Do we have a valid ELF header ?
502          */
503         if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
504                 return -1;
505
506         /*
507          * From here on down, we return an errno, not -1, as we've
508          * detected an ELF file.
509          */
510
511         if ((hdr->e_phoff > PAGE_SIZE) ||
512             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
513                 /* Only support headers in first page for now */
514                 return ENOEXEC;
515         }
516         phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
517         
518         /*
519          * From this point on, we may have resources that need to be freed.
520          */
521
522         exec_new_vmspace(imgp, NULL);
523
524         /*
525          * Yeah, I'm paranoid.  There is every reason in the world to get
526          * VTEXT now since from here on out, there are places we can have
527          * a context switch.  Better safe than sorry; I really don't want
528          * the file to change while it's being loaded.
529          */
530         lwkt_gettoken(&ilock, imgp->vp->v_interlock);
531         imgp->vp->v_flag |= VTEXT;
532         lwkt_reltoken(&ilock);
533
534         vmspace = imgp->proc->p_vmspace;
535
536         for (i = 0; i < hdr->e_phnum; i++) {
537                 switch(phdr[i].p_type) {
538
539                 case PT_LOAD:   /* Loadable segment */
540                         prot = 0;
541                         if (phdr[i].p_flags & PF_X)
542                                 prot |= VM_PROT_EXECUTE;
543                         if (phdr[i].p_flags & PF_W)
544                                 prot |= VM_PROT_WRITE;
545                         if (phdr[i].p_flags & PF_R)
546                                 prot |= VM_PROT_READ;
547
548                         if ((error = elf_load_section(imgp->proc,
549                                                      vmspace, imgp->vp,
550                                                      phdr[i].p_offset,
551                                                      (caddr_t)phdr[i].p_vaddr,
552                                                      phdr[i].p_memsz,
553                                                      phdr[i].p_filesz, prot)) != 0)
554                                 goto fail;
555
556                         seg_addr = trunc_page(phdr[i].p_vaddr);
557                         seg_size = round_page(phdr[i].p_memsz +
558                                 phdr[i].p_vaddr - seg_addr);
559
560                         /*
561                          * Is this .text or .data?  We can't use
562                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
563                          * alpha terribly and possibly does other bad
564                          * things so we stick to the old way of figuring
565                          * it out:  If the segment contains the program
566                          * entry point, it's a text segment, otherwise it
567                          * is a data segment.
568                          *
569                          * Note that obreak() assumes that data_addr + 
570                          * data_size == end of data load area, and the ELF
571                          * file format expects segments to be sorted by
572                          * address.  If multiple data segments exist, the
573                          * last one will be used.
574                          */
575                         if (hdr->e_entry >= phdr[i].p_vaddr &&
576                             hdr->e_entry < (phdr[i].p_vaddr +
577                             phdr[i].p_memsz)) {
578                                 text_size = seg_size;
579                                 text_addr = seg_addr;
580                                 entry = (u_long)hdr->e_entry;
581                         } else {
582                                 data_size = seg_size;
583                                 data_addr = seg_addr;
584                         }
585                         total_size += seg_size;
586
587                         /*
588                          * Check limits.  It should be safe to check the
589                          * limits after loading the segment since we do
590                          * not actually fault in all the segment's pages.
591                          */
592                         if (data_size >
593                             imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
594                             text_size > maxtsiz ||
595                             total_size >
596                             imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
597                                 error = ENOMEM;
598                                 goto fail;
599                         }
600                         break;
601                 case PT_INTERP: /* Path to interpreter */
602                         if (phdr[i].p_filesz > MAXPATHLEN ||
603                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
604                                 error = ENOEXEC;
605                                 goto fail;
606                         }
607                         interp = imgp->image_header + phdr[i].p_offset;
608                         break;
609                 case PT_PHDR:   /* Program header table info */
610                         proghdr = phdr[i].p_vaddr;
611                         break;
612                 default:
613                         break;
614                 }
615         }
616
617         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
618         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
619         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
620         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
621
622         addr = ELF_RTLD_ADDR(vmspace);
623
624         imgp->entry_addr = entry;
625
626         brand_info = NULL;
627
628         /* We support three types of branding -- (1) the ELF EI_OSABI field
629          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
630          * branding w/in the ELF header, and (3) path of the `interp_path'
631          * field.  We should also look for an ".note.ABI-tag" ELF section now
632          * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
633          */
634
635         /* If the executable has a brand, search for it in the brand list. */
636         if (brand_info == NULL) {
637                 for (i = 0;  i < MAX_BRANDS;  i++) {
638                         Elf_Brandinfo *bi = elf_brand_list[i];
639
640                         if (bi != NULL && 
641                             (hdr->e_ident[EI_OSABI] == bi->brand
642                             || 0 == 
643                             strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 
644                             bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
645                                 brand_info = bi;
646                                 break;
647                         }
648                 }
649         }
650
651         /* Lacking a known brand, search for a recognized interpreter. */
652         if (brand_info == NULL && interp != NULL) {
653                 for (i = 0;  i < MAX_BRANDS;  i++) {
654                         Elf_Brandinfo *bi = elf_brand_list[i];
655
656                         if (bi != NULL &&
657                             strcmp(interp, bi->interp_path) == 0) {
658                                 brand_info = bi;
659                                 break;
660                         }
661                 }
662         }
663
664         /* Lacking a recognized interpreter, try the default brand */
665         if (brand_info == NULL) {
666                 for (i = 0; i < MAX_BRANDS; i++) {
667                         Elf_Brandinfo *bi = elf_brand_list[i];
668
669                         if (bi != NULL && fallback_elf_brand == bi->brand) {
670                                 brand_info = bi;
671                                 break;
672                         }
673                 }
674         }
675
676         if (brand_info == NULL) {
677                 uprintf("ELF binary type \"%u\" not known.\n",
678                     hdr->e_ident[EI_OSABI]);
679                 error = ENOEXEC;
680                 goto fail;
681         }
682
683         imgp->proc->p_sysent = brand_info->sysvec;
684         if (interp != NULL) {
685                 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
686                 snprintf(path, MAXPATHLEN, "%s%s",
687                          brand_info->emul_path, interp);
688                 if ((error = elf_load_file(imgp->proc, path, &addr,
689                                            &imgp->entry_addr)) != 0) {
690                         if ((error = elf_load_file(imgp->proc, interp, &addr,
691                                                    &imgp->entry_addr)) != 0) {
692                                 uprintf("ELF interpreter %s not found\n", path);
693                                 free(path, M_TEMP);
694                                 goto fail;
695                         }
696                 }
697                 free(path, M_TEMP);
698         }
699
700         /*
701          * Construct auxargs table (used by the fixup routine)
702          */
703         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
704         elf_auxargs->execfd = -1;
705         elf_auxargs->phdr = proghdr;
706         elf_auxargs->phent = hdr->e_phentsize;
707         elf_auxargs->phnum = hdr->e_phnum;
708         elf_auxargs->pagesz = PAGE_SIZE;
709         elf_auxargs->base = addr;
710         elf_auxargs->flags = 0;
711         elf_auxargs->entry = entry;
712         elf_auxargs->trace = elf_trace;
713
714         imgp->auxargs = elf_auxargs;
715         imgp->interpreted = 0;
716
717 fail:
718         return error;
719 }
720
721 static int
722 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
723 {
724         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
725         register_t *pos;
726
727         pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
728
729         if (args->trace) {
730                 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
731         }
732         if (args->execfd != -1) {
733                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
734         }
735         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
736         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
737         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
738         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
739         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
740         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
741         AUXARGS_ENTRY(pos, AT_BASE, args->base);
742         AUXARGS_ENTRY(pos, AT_NULL, 0);
743
744         free(imgp->auxargs, M_TEMP);
745         imgp->auxargs = NULL;
746
747         (*stack_base)--;
748         suword(*stack_base, (long) imgp->args->argc);
749         return 0;
750
751
752 /*
753  * Code for generating ELF core dumps.
754  */
755
756 typedef int (*segment_callback) (vm_map_entry_t, void *);
757
758 /* Closure for cb_put_phdr(). */
759 struct phdr_closure {
760         Elf_Phdr *phdr;         /* Program header to fill in (incremented) */
761         Elf_Phdr *phdr_max;     /* Pointer bound for error check */
762         Elf_Off offset;         /* Offset of segment in core file */
763 };
764
765 /* Closure for cb_size_segment(). */
766 struct sseg_closure {
767         int count;              /* Count of writable segments. */
768         size_t vsize;           /* Total size of all writable segments. */
769 };
770
771 /* Closure for cb_put_fp(). */
772 struct fp_closure {
773         struct vn_hdr *vnh;
774         struct vn_hdr *vnh_max;
775         int count;
776         struct stat *sb;
777 };
778
779 typedef struct elf_buf {
780         char    *buf;
781         size_t  off;
782         size_t  off_max;
783 } *elf_buf_t;
784
785 static void *target_reserve(elf_buf_t target, size_t bytes, int *error);
786
787 static int cb_put_phdr (vm_map_entry_t, void *);
788 static int cb_size_segment (vm_map_entry_t, void *);
789 static int cb_fpcount_segment(vm_map_entry_t, void *);
790 static int cb_put_fp(vm_map_entry_t, void *);
791
792
793 static int each_segment (struct proc *, segment_callback, void *, int);
794 static int elf_corehdr (struct proc *, struct file *, struct ucred *,
795                         int, elf_buf_t);
796 static int elf_puthdr (struct proc *, elf_buf_t, const prstatus_t *,
797                         const prfpregset_t *, const prpsinfo_t *, int);
798 static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t);
799
800 static int elf_putsigs(struct proc *, elf_buf_t);
801 static int elf_puttextvp(struct proc *, elf_buf_t);
802 static int elf_putfiles(struct proc *, elf_buf_t);
803
804 extern int osreldate;
805
806 int
807 elf_coredump(struct proc *p, struct vnode *vp, off_t limit)
808 {
809         struct file *fp; 
810         int error;
811
812         if ((error = falloc(NULL, &fp, NULL)) != 0)
813                 return (error);
814         fsetcred(fp, p->p_ucred);
815
816         fp->f_data = (caddr_t)vp;
817         fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW;
818         fp->f_ops = &vnops;
819         fp->f_type = DTYPE_VNODE;
820         VOP_UNLOCK(vp, NULL, 0, p->p_thread);
821         
822         error = generic_elf_coredump(p, fp, limit);
823
824         fp->f_data = NULL;
825         fp->f_flag = 0;
826         fp->f_ops = &badfileops;
827         fp->f_type = 0;
828         fdrop(fp, p->p_thread);
829         return (error);
830 }
831
832 int
833 generic_elf_coredump(struct proc *p, struct file *fp, off_t limit)
834 {
835         struct ucred *cred = p->p_ucred;
836         int error = 0;
837         struct sseg_closure seginfo;
838         struct elf_buf target;
839
840         if (!fp)
841                 printf("can't dump core - null fp\n");
842
843         /*
844          * Size the program segments
845          */
846         seginfo.count = 0;
847         seginfo.vsize = 0;
848         each_segment(p, cb_size_segment, &seginfo, 1);
849
850         /*
851          * Calculate the size of the core file header area by making
852          * a dry run of generating it.  Nothing is written, but the
853          * size is calculated.
854          */
855         bzero(&target, sizeof(target));
856         elf_puthdr(p, &target, NULL, NULL, NULL, seginfo.count);
857
858         if (target.off + seginfo.vsize >= limit)
859                 return (EFAULT);
860
861         /*
862          * Allocate memory for building the header, fill it up,
863          * and write it out.
864          */
865         target.off_max = target.off;
866         target.off = 0;
867         target.buf = malloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO);
868
869         if (target.buf == NULL)
870                 return EINVAL;
871         error = elf_corehdr(p, fp, cred, seginfo.count, &target);
872
873         /* Write the contents of all of the writable segments. */
874         if (error == 0) {
875                 Elf_Phdr *php;
876                 int i;
877                 int nbytes;
878
879                 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1;
880                 for (i = 0; i < seginfo.count; i++) {
881                         error = fp_write(fp, (caddr_t)php->p_vaddr,
882                                         php->p_filesz, &nbytes);
883                         if (error != 0)
884                                 break;
885                         php++;
886                 }
887         }
888         free(target.buf, M_TEMP);
889         
890         return error;
891 }
892
893 /*
894  * A callback for each_segment() to write out the segment's
895  * program header entry.
896  */
897 static int
898 cb_put_phdr(vm_map_entry_t entry, void *closure)
899 {
900         struct phdr_closure *phc = closure;
901         Elf_Phdr *phdr = phc->phdr;
902
903         if (phc->phdr == phc->phdr_max)
904                 return EINVAL;
905
906         phc->offset = round_page(phc->offset);
907
908         phdr->p_type = PT_LOAD;
909         phdr->p_offset = phc->offset;
910         phdr->p_vaddr = entry->start;
911         phdr->p_paddr = 0;
912         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
913         phdr->p_align = PAGE_SIZE;
914         phdr->p_flags = 0;
915         if (entry->protection & VM_PROT_READ)
916                 phdr->p_flags |= PF_R;
917         if (entry->protection & VM_PROT_WRITE)
918                 phdr->p_flags |= PF_W;
919         if (entry->protection & VM_PROT_EXECUTE)
920                 phdr->p_flags |= PF_X;
921
922         phc->offset += phdr->p_filesz;
923         ++phc->phdr;
924         return 0;
925 }
926
927 /*
928  * A callback for each_writable_segment() to gather information about
929  * the number of segments and their total size.
930  */
931 static int
932 cb_size_segment(vm_map_entry_t entry, void *closure)
933 {
934         struct sseg_closure *ssc = closure;
935
936         ++ssc->count;
937         ssc->vsize += entry->end - entry->start;
938         return 0;
939 }
940
941 /*
942  * A callback for each_segment() to gather information about
943  * the number of text segments.
944  */
945 static int
946 cb_fpcount_segment(vm_map_entry_t entry, void *closure)
947 {
948         int *count = closure;
949         if (entry->object.vm_object->type == OBJT_VNODE)
950                 ++*count;
951         return 0;
952 }
953
954 static int
955 cb_put_fp(vm_map_entry_t entry, void *closure) 
956 {
957         struct fp_closure *fpc = closure;
958         struct vn_hdr *vnh = fpc->vnh;
959         Elf_Phdr *phdr = &vnh->vnh_phdr;
960         struct vnode *vp;
961         int error;
962
963         if (entry->object.vm_object->type == OBJT_VNODE) {
964                 if (vnh == fpc->vnh_max)
965                         return EINVAL;
966                 vp = (struct vnode *)entry->object.vm_object->handle;
967
968                 if (vp->v_mount)
969                         vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
970                 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid);
971                 if (error) 
972                         return error;
973
974                 phdr->p_type = PT_LOAD;
975                 phdr->p_offset = 0;        /* not written to core */
976                 phdr->p_vaddr = entry->start;
977                 phdr->p_paddr = 0;
978                 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
979                 phdr->p_align = PAGE_SIZE;
980                 phdr->p_flags = 0;
981                 if (entry->protection & VM_PROT_READ)
982                         phdr->p_flags |= PF_R;
983                 if (entry->protection & VM_PROT_WRITE)
984                         phdr->p_flags |= PF_W;
985                 if (entry->protection & VM_PROT_EXECUTE)
986                         phdr->p_flags |= PF_X;
987                 ++fpc->vnh;
988                 ++fpc->count;
989         }
990         return 0;
991 }
992
993 /*
994  * For each writable segment in the process's memory map, call the given
995  * function with a pointer to the map entry and some arbitrary
996  * caller-supplied data.
997  */
998 static int
999 each_segment(struct proc *p, segment_callback func, void *closure, int writable)
1000 {
1001         int error = 0;
1002         vm_map_t map = &p->p_vmspace->vm_map;
1003         vm_map_entry_t entry;
1004
1005         for (entry = map->header.next; error == 0 && entry != &map->header;
1006             entry = entry->next) {
1007                 vm_object_t obj;
1008
1009                 /*
1010                  * Don't dump inaccessible mappings, deal with legacy
1011                  * coredump mode.
1012                  *
1013                  * Note that read-only segments related to the elf binary
1014                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1015                  * need to arbitrarily ignore such segments.
1016                  */
1017                 if (elf_legacy_coredump) {
1018                         if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW)
1019                                 continue;
1020                 } else {
1021                         if (writable && (entry->protection & VM_PROT_ALL) == 0)
1022                                 continue;
1023                 }
1024
1025                 /*
1026                  * Dont include memory segment in the coredump if
1027                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1028                  * madvise(2).  Do not dump submaps (i.e. parts of the
1029                  * kernel map).
1030                  */
1031                 if (writable && entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1032                         continue;
1033
1034                 if ((obj = entry->object.vm_object) == NULL)
1035                         continue;
1036
1037                 /* Find the deepest backing object. */
1038                 while (obj->backing_object != NULL)
1039                         obj = obj->backing_object;
1040
1041                 /* Ignore memory-mapped devices and such things. */
1042                 if (obj->type != OBJT_DEFAULT &&
1043                     obj->type != OBJT_SWAP &&
1044                     obj->type != OBJT_VNODE)
1045                         continue;
1046
1047                 error = (*func)(entry, closure);
1048         }
1049         return error;
1050 }
1051
1052 static
1053 void *
1054 target_reserve(elf_buf_t target, size_t bytes, int *error)
1055 {
1056     void *res = NULL;
1057
1058     if (target->buf) {
1059             if (target->off + bytes > target->off_max)
1060                     *error = EINVAL;
1061             else
1062                     res = target->buf + target->off;
1063     }
1064     target->off += bytes;
1065     return (res);
1066 }
1067
1068 /*
1069  * Write the core file header to the file, including padding up to
1070  * the page boundary.
1071  */
1072 static int
1073 elf_corehdr(struct proc *p, struct file *fp, struct ucred *cred, int numsegs, 
1074             elf_buf_t target)
1075 {
1076         struct {
1077                 prstatus_t status;
1078                 prfpregset_t fpregset;
1079                 prpsinfo_t psinfo;
1080         } *tempdata;
1081         int error;
1082         prstatus_t *status;
1083         prfpregset_t *fpregset;
1084         prpsinfo_t *psinfo;
1085         int nbytes;
1086         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK);
1087         status = &tempdata->status;
1088         fpregset = &tempdata->fpregset;
1089         psinfo = &tempdata->psinfo;
1090
1091         /* Gather the information for the header. */
1092         status->pr_version = PRSTATUS_VERSION;
1093         status->pr_statussz = sizeof(prstatus_t);
1094         status->pr_gregsetsz = sizeof(gregset_t);
1095         status->pr_fpregsetsz = sizeof(fpregset_t);
1096         status->pr_osreldate = osreldate;
1097         status->pr_cursig = p->p_sig;
1098         status->pr_pid = p->p_pid;
1099         fill_regs(p, &status->pr_reg);
1100
1101         fill_fpregs(p, fpregset);
1102
1103         psinfo->pr_version = PRPSINFO_VERSION;
1104         psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1105         strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1);
1106
1107         /* XXX - We don't fill in the command line arguments properly yet. */
1108         strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
1109
1110         /* Fill in the header. */
1111         error = elf_puthdr(p, target, status, fpregset, psinfo, numsegs);
1112
1113         free(tempdata, M_TEMP);
1114
1115         /* Write it to the core file. */
1116         if (error == 0)
1117                 error = fp_write(fp, target->buf, target->off, &nbytes);
1118         return error;
1119 }
1120
1121 static int
1122 elf_puthdr(struct proc *p, elf_buf_t target, const prstatus_t *status,
1123         const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
1124 {
1125         int error = 0;
1126         size_t phoff;
1127         size_t noteoff;
1128         size_t notesz;
1129         Elf_Ehdr *ehdr;
1130         Elf_Phdr *phdr;
1131
1132         ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error);
1133
1134         phoff = target->off;
1135         phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error);
1136
1137         noteoff = target->off;
1138         if (error == 0) {
1139                 error = elf_putnote(target, "FreeBSD", NT_PRSTATUS, 
1140                                         status, sizeof *status);
1141         }
1142         if (error == 0) {
1143                 error = elf_putnote(target, "FreeBSD", NT_FPREGSET,
1144                                         fpregset, sizeof *fpregset);
1145         }
1146         if (error == 0) {
1147                 error = elf_putnote(target, "FreeBSD", NT_PRPSINFO,
1148                                         psinfo, sizeof *psinfo);
1149         }
1150         notesz = target->off - noteoff;
1151
1152         /*
1153          * put extra cruft for dumping process state here 
1154          *  - we really want it be before all the program 
1155          *    mappings
1156          *  - we just need to update the offset accordingly
1157          *    and GDB will be none the wiser.
1158          */
1159         if (error == 0)
1160                 error = elf_puttextvp(p, target);
1161         if (error == 0)
1162                 error = elf_putsigs(p, target);
1163         if (error == 0)
1164                 error = elf_putfiles(p, target);
1165
1166         /*
1167          * Align up to a page boundary for the program segments.  The
1168          * actual data will be written to the outptu file, not to elf_buf_t,
1169          * so we do not have to do any further bounds checking.
1170          */
1171         target->off = round_page(target->off);
1172         if (error == 0 && ehdr != NULL) {
1173                 /*
1174                  * Fill in the ELF header.
1175                  */
1176                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1177                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1178                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1179                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1180                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1181                 ehdr->e_ident[EI_DATA] = ELF_DATA;
1182                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1183                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1184                 ehdr->e_ident[EI_ABIVERSION] = 0;
1185                 ehdr->e_ident[EI_PAD] = 0;
1186                 ehdr->e_type = ET_CORE;
1187                 ehdr->e_machine = ELF_ARCH;
1188                 ehdr->e_version = EV_CURRENT;
1189                 ehdr->e_entry = 0;
1190                 ehdr->e_phoff = phoff;
1191                 ehdr->e_flags = 0;
1192                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1193                 ehdr->e_phentsize = sizeof(Elf_Phdr);
1194                 ehdr->e_phnum = numsegs + 1;
1195                 ehdr->e_shentsize = sizeof(Elf_Shdr);
1196                 ehdr->e_shnum = 0;
1197                 ehdr->e_shstrndx = SHN_UNDEF;
1198         }
1199         if (error == 0 && phdr != NULL) {
1200                 /*
1201                  * Fill in the program header entries.
1202                  */
1203                 struct phdr_closure phc;
1204
1205                 /* The note segement. */
1206                 phdr->p_type = PT_NOTE;
1207                 phdr->p_offset = noteoff;
1208                 phdr->p_vaddr = 0;
1209                 phdr->p_paddr = 0;
1210                 phdr->p_filesz = notesz;
1211                 phdr->p_memsz = 0;
1212                 phdr->p_flags = 0;
1213                 phdr->p_align = 0;
1214                 ++phdr;
1215
1216                 /* All the writable segments from the program. */
1217                 phc.phdr = phdr;
1218                 phc.phdr_max = phdr + numsegs;
1219                 phc.offset = target->off;
1220                 each_segment(p, cb_put_phdr, &phc, 1);
1221         }
1222         return (error);
1223 }
1224
1225 static int
1226 elf_putnote(elf_buf_t target, const char *name, int type,
1227             const void *desc, size_t descsz)
1228 {
1229         int error = 0;
1230         char *dst;
1231         Elf_Note note;
1232
1233         note.n_namesz = strlen(name) + 1;
1234         note.n_descsz = descsz;
1235         note.n_type = type;
1236         dst = target_reserve(target, sizeof(note), &error);
1237         if (dst != NULL)
1238                 bcopy(&note, dst, sizeof note);
1239         dst = target_reserve(target, note.n_namesz, &error);
1240         if (dst != NULL)
1241                 bcopy(name, dst, note.n_namesz);
1242         target->off = roundup2(target->off, sizeof(Elf_Size));
1243         dst = target_reserve(target, note.n_descsz, &error);
1244         if (dst != NULL)
1245                 bcopy(desc, dst, note.n_descsz);
1246         target->off = roundup2(target->off, sizeof(Elf_Size));
1247         return(error);
1248 }
1249
1250
1251 static int
1252 elf_putsigs(struct proc *p, elf_buf_t target)
1253 {
1254         int error = 0;
1255         struct ckpt_siginfo *csi;
1256
1257         csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error);
1258         if (csi) {
1259                 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo);
1260                 bcopy(p->p_procsig, &csi->csi_procsig, sizeof(struct procsig));
1261                 bcopy(p->p_procsig->ps_sigacts, &csi->csi_sigacts, sizeof(struct sigacts));
1262                 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval));
1263                 csi->csi_sigparent = p->p_sigparent;
1264         }
1265         return(error);
1266 }
1267
1268 static int
1269 elf_putfiles(struct proc *p, elf_buf_t target)
1270 {
1271         int error = 0;
1272         int i;
1273         struct ckpt_filehdr *cfh = NULL;
1274         struct ckpt_fileinfo *cfi;
1275         struct file *fp;        
1276         struct vnode *vp;
1277         /*
1278          * the duplicated loop is gross, but it was the only way
1279          * to eliminate uninitialized variable warnings 
1280          */
1281         cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error);
1282         if (cfh) {
1283                 cfh->cfh_nfiles = 0;            
1284         }
1285
1286         /*
1287          * ignore STDIN/STDERR/STDOUT
1288          */
1289         for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) {
1290                 if ((fp = p->p_fd->fd_ofiles[i]) == NULL)
1291                         continue;
1292                 if (fp->f_type != DTYPE_VNODE)
1293                         continue;
1294                 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo), &error);
1295                 if (cfi) {
1296                         cfi->cfi_index = -1;
1297                         vp = (struct vnode *)fp->f_data;
1298                         /*
1299                          * it looks like a bug in ptrace is marking 
1300                          * a non-vnode as a vnode - until we find the 
1301                          * root cause this will at least prevent
1302                          * further panics from truss
1303                          */
1304                         if (vp == NULL || vp->v_mount == NULL)
1305                                 continue;
1306                         cfh->cfh_nfiles++;
1307                         cfi->cfi_index = i;
1308                         cfi->cfi_flags = fp->f_flag;
1309                         cfi->cfi_offset = fp->f_offset;
1310                         cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1311                         error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid);
1312                 }
1313         }
1314         return(error);
1315 }
1316
1317 static int
1318 elf_puttextvp(struct proc *p, elf_buf_t target)
1319 {
1320         int error = 0;
1321         int *vn_count;
1322         struct fp_closure fpc;
1323         struct ckpt_vminfo *vminfo;
1324
1325         vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error);
1326         if (vminfo != NULL) {
1327                 vminfo->cvm_dsize = p->p_vmspace->vm_dsize;
1328                 vminfo->cvm_tsize = p->p_vmspace->vm_tsize;
1329                 vminfo->cvm_daddr = p->p_vmspace->vm_daddr;
1330                 vminfo->cvm_taddr = p->p_vmspace->vm_taddr;
1331         }
1332
1333         fpc.count = 0;
1334         vn_count = target_reserve(target, sizeof(int), &error);
1335         if (target->buf != NULL) {
1336                 fpc.vnh = (struct vn_hdr *)(target->buf + target->off);
1337                 fpc.vnh_max = fpc.vnh + 
1338                         (target->off_max - target->off) / sizeof(struct vn_hdr);
1339                 error = each_segment(p, cb_put_fp, &fpc, 0);
1340                 if (vn_count)
1341                         *vn_count = fpc.count;
1342         } else {
1343                 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0);
1344         }
1345         target->off += fpc.count * sizeof(struct vn_hdr);
1346         return(error);
1347 }
1348
1349
1350 /*
1351  * Tell kern_execve.c about it, with a little help from the linker.
1352  */
1353 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1354 EXEC_SET(elf, elf_execsw);