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 1.15 2003/11/12 01:00:33 daver 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                 return ENOEXEC;
 179
 180         if (!ELF_MACHINE_OK(hdr->e_machine))
 181                 return ENOEXEC;
 182
 183         if (hdr->e_version != ELF_TARG_VER)
 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, 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, 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         if ((hdr->e_phoff > PAGE_SIZE) ||
 419             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
 420                 error = ENOEXEC;
 421                 goto fail;
 422         }
 423
 424         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 425
 426         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
 427                 if (phdr[i].p_type == PT_LOAD) {        /* Loadable segment */
 428                         prot = 0;
 429                         if (phdr[i].p_flags & PF_X)
 430                                 prot |= VM_PROT_EXECUTE;
 431                         if (phdr[i].p_flags & PF_W)
 432                                 prot |= VM_PROT_WRITE;
 433                         if (phdr[i].p_flags & PF_R)
 434                                 prot |= VM_PROT_READ;
 435
 436                         error = elf_load_section(
 437                                     p, vmspace, nd->ni_vp,
 438                                     phdr[i].p_offset,
 439                                     (caddr_t)phdr[i].p_vaddr +
 440                                     rbase,
 441                                     phdr[i].p_memsz,
 442                                     phdr[i].p_filesz, prot);
 443                         if (error != 0)
 444                                 goto fail;
 445                         /*
 446                          * Establish the base address if this is the
 447                          * first segment.
 448                          */
 449                         if (numsegs == 0)
 450                                 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
 451                         numsegs++;
 452                 }
 453         }
 454         *addr = base_addr;
 455         *entry=(unsigned long)hdr->e_entry + rbase;
 456
 457 fail:
 458         if (imgp->firstpage)
 459                 exec_unmap_first_page(imgp);
 460         if (imgp->image_header)
 461                 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
 462                         PAGE_SIZE);
 463         if (nd->ni_vp)
 464                 vrele(nd->ni_vp);
 465
 466         free(tempdata, M_TEMP);
 467
 468         return error;
 469 }
 470
 471 /*
 472  * non static, as it can be overridden by start_init()
 473  */
 474 int fallback_elf_brand = -1;
 475 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
 476                 &fallback_elf_brand, -1,
 477                 "ELF brand of last resort");
 478
 479 static int
 480 exec_elf_imgact(struct image_params *imgp)
 481 {
 482         const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
 483         const Elf_Phdr *phdr;
 484         Elf_Auxargs *elf_auxargs = NULL;
 485         struct vmspace *vmspace;
 486         vm_prot_t prot;
 487         u_long text_size = 0, data_size = 0, total_size = 0;
 488         u_long text_addr = 0, data_addr = 0;
 489         u_long seg_size, seg_addr;
 490         u_long addr, entry = 0, proghdr = 0;
 491         int error, i;
 492         const char *interp = NULL;
 493         Elf_Brandinfo *brand_info;
 494         char *path;
 495
 496         error = 0;
 497
 498         /*
 499          * Do we have a valid ELF header ?
 500          */
 501         if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
 502                 return -1;
 503
 504         /*
 505          * From here on down, we return an errno, not -1, as we've
 506          * detected an ELF file.
 507          */
 508
 509         if ((hdr->e_phoff > PAGE_SIZE) ||
 510             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
 511                 /* Only support headers in first page for now */
 512                 return ENOEXEC;
 513         }
 514         phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
 515
 516         /*
 517          * From this point on, we may have resources that need to be freed.
 518          */
 519
 520         exec_new_vmspace(imgp);
 521
 522         /*
 523          * Yeah, I'm paranoid.  There is every reason in the world to get
 524          * VTEXT now since from here on out, there are places we can have
 525          * a context switch.  Better safe than sorry; I really don't want
 526          * the file to change while it's being loaded.
 527          */
 528         lwkt_gettoken(&imgp->vp->v_interlock);
 529         imgp->vp->v_flag |= VTEXT;
 530         lwkt_reltoken(&imgp->vp->v_interlock);
 531
 532         vmspace = imgp->proc->p_vmspace;
 533
 534         for (i = 0; i < hdr->e_phnum; i++) {
 535                 switch(phdr[i].p_type) {
 536
 537                 case PT_LOAD:   /* Loadable segment */
 538                         prot = 0;
 539                         if (phdr[i].p_flags & PF_X)
 540                                 prot |= VM_PROT_EXECUTE;
 541                         if (phdr[i].p_flags & PF_W)
 542                                 prot |= VM_PROT_WRITE;
 543                         if (phdr[i].p_flags & PF_R)
 544                                 prot |= VM_PROT_READ;
 545
 546                         if ((error = elf_load_section(imgp->proc,
 547                                                      vmspace, imgp->vp,
 548                                                      phdr[i].p_offset,
 549                                                      (caddr_t)phdr[i].p_vaddr,
 550                                                      phdr[i].p_memsz,
 551                                                      phdr[i].p_filesz, prot)) != 0)
 552                                 goto fail;
 553
 554                         seg_addr = trunc_page(phdr[i].p_vaddr);
 555                         seg_size = round_page(phdr[i].p_memsz +
 556                                 phdr[i].p_vaddr - seg_addr);
 557
 558                         /*
 559                          * Is this .text or .data?  We can't use
 560                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
 561                          * alpha terribly and possibly does other bad
 562                          * things so we stick to the old way of figuring
 563                          * it out:  If the segment contains the program
 564                          * entry point, it's a text segment, otherwise it
 565                          * is a data segment.
 566                          *
 567                          * Note that obreak() assumes that data_addr +
 568                          * data_size == end of data load area, and the ELF
 569                          * file format expects segments to be sorted by
 570                          * address.  If multiple data segments exist, the
 571                          * last one will be used.
 572                          */
 573                         if (hdr->e_entry >= phdr[i].p_vaddr &&
 574                             hdr->e_entry < (phdr[i].p_vaddr +
 575                             phdr[i].p_memsz)) {
 576                                 text_size = seg_size;
 577                                 text_addr = seg_addr;
 578                                 entry = (u_long)hdr->e_entry;
 579                         } else {
 580                                 data_size = seg_size;
 581                                 data_addr = seg_addr;
 582                         }
 583                         total_size += seg_size;
 584
 585                         /*
 586                          * Check limits.  It should be safe to check the
 587                          * limits after loading the segment since we do
 588                          * not actually fault in all the segment's pages.
 589                          */
 590                         if (data_size >
 591                             imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
 592                             text_size > maxtsiz ||
 593                             total_size >
 594                             imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
 595                                 error = ENOMEM;
 596                                 goto fail;
 597                         }
 598                         break;
 599                 case PT_INTERP: /* Path to interpreter */
 600                         if (phdr[i].p_filesz > MAXPATHLEN ||
 601                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
 602                                 error = ENOEXEC;
 603                                 goto fail;
 604                         }
 605                         interp = imgp->image_header + phdr[i].p_offset;
 606                         break;
 607                 case PT_PHDR:   /* Program header table info */
 608                         proghdr = phdr[i].p_vaddr;
 609                         break;
 610                 default:
 611                         break;
 612                 }
 613         }
 614
 615         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
 616         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
 617         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
 618         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
 619
 620         addr = ELF_RTLD_ADDR(vmspace);
 621
 622         imgp->entry_addr = entry;
 623
 624         brand_info = NULL;
 625
 626         /* We support three types of branding -- (1) the ELF EI_OSABI field
 627          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
 628          * branding w/in the ELF header, and (3) path of the `interp_path'
 629          * field.  We should also look for an ".note.ABI-tag" ELF section now
 630          * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
 631          */
 632
 633         /* If the executable has a brand, search for it in the brand list. */
 634         if (brand_info == NULL) {
 635                 for (i = 0;  i < MAX_BRANDS;  i++) {
 636                         Elf_Brandinfo *bi = elf_brand_list[i];
 637
 638                         if (bi != NULL &&
 639                             (hdr->e_ident[EI_OSABI] == bi->brand
 640                             || 0 ==
 641                             strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
 642                             bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
 643                                 brand_info = bi;
 644                                 break;
 645                         }
 646                 }
 647         }
 648
 649         /* Lacking a known brand, search for a recognized interpreter. */
 650         if (brand_info == NULL && interp != NULL) {
 651                 for (i = 0;  i < MAX_BRANDS;  i++) {
 652                         Elf_Brandinfo *bi = elf_brand_list[i];
 653
 654                         if (bi != NULL &&
 655                             strcmp(interp, bi->interp_path) == 0) {
 656                                 brand_info = bi;
 657                                 break;
 658                         }
 659                 }
 660         }
 661
 662         /* Lacking a recognized interpreter, try the default brand */
 663         if (brand_info == NULL) {
 664                 for (i = 0; i < MAX_BRANDS; i++) {
 665                         Elf_Brandinfo *bi = elf_brand_list[i];
 666
 667                         if (bi != NULL && fallback_elf_brand == bi->brand) {
 668                                 brand_info = bi;
 669                                 break;
 670                         }
 671                 }
 672         }
 673
 674         if (brand_info == NULL) {
 675                 uprintf("ELF binary type \"%u\" not known.\n",
 676                     hdr->e_ident[EI_OSABI]);
 677                 error = ENOEXEC;
 678                 goto fail;
 679         }
 680
 681         imgp->proc->p_sysent = brand_info->sysvec;
 682         if (interp != NULL) {
 683                 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 684                 snprintf(path, MAXPATHLEN, "%s%s",
 685                          brand_info->emul_path, interp);
 686                 if ((error = elf_load_file(imgp->proc, path, &addr,
 687                                            &imgp->entry_addr)) != 0) {
 688                         if ((error = elf_load_file(imgp->proc, interp, &addr,
 689                                                    &imgp->entry_addr)) != 0) {
 690                                 uprintf("ELF interpreter %s not found\n", path);
 691                                 free(path, M_TEMP);
 692                                 goto fail;
 693                         }
 694                 }
 695                 free(path, M_TEMP);
 696         }
 697
 698         /*
 699          * Construct auxargs table (used by the fixup routine)
 700          */
 701         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
 702         elf_auxargs->execfd = -1;
 703         elf_auxargs->phdr = proghdr;
 704         elf_auxargs->phent = hdr->e_phentsize;
 705         elf_auxargs->phnum = hdr->e_phnum;
 706         elf_auxargs->pagesz = PAGE_SIZE;
 707         elf_auxargs->base = addr;
 708         elf_auxargs->flags = 0;
 709         elf_auxargs->entry = entry;
 710         elf_auxargs->trace = elf_trace;
 711
 712         imgp->auxargs = elf_auxargs;
 713         imgp->interpreted = 0;
 714
 715 fail:
 716         return error;
 717 }
 718
 719 static int
 720 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
 721 {
 722         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
 723         register_t *pos;
 724
 725         pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
 726
 727         if (args->trace) {
 728                 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
 729         }
 730         if (args->execfd != -1) {
 731                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
 732         }
 733         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
 734         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
 735         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
 736         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
 737         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
 738         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
 739         AUXARGS_ENTRY(pos, AT_BASE, args->base);
 740         AUXARGS_ENTRY(pos, AT_NULL, 0);
 741
 742         free(imgp->auxargs, M_TEMP);
 743         imgp->auxargs = NULL;
 744
 745         (*stack_base)--;
 746         suword(*stack_base, (long) imgp->args->argc);
 747         return 0;
 748 }
 749
 750 /*
 751  * Code for generating ELF core dumps.
 752  */
 753
 754 typedef int (*segment_callback) (vm_map_entry_t, void *);
 755
 756 /* Closure for cb_put_phdr(). */
 757 struct phdr_closure {
 758         Elf_Phdr *phdr;         /* Program header to fill in (incremented) */
 759         Elf_Phdr *phdr_max;     /* Pointer bound for error check */
 760         Elf_Off offset;         /* Offset of segment in core file */
 761 };
 762
 763 /* Closure for cb_size_segment(). */
 764 struct sseg_closure {
 765         int count;              /* Count of writable segments. */
 766         size_t vsize;           /* Total size of all writable segments. */
 767 };
 768
 769 /* Closure for cb_put_fp(). */
 770 struct fp_closure {
 771         struct vn_hdr *vnh;
 772         struct vn_hdr *vnh_max;
 773         int count;
 774         struct stat *sb;
 775 };
 776
 777 typedef struct elf_buf {
 778         char    *buf;
 779         size_t  off;
 780         size_t  off_max;
 781 } *elf_buf_t;
 782
 783 static void *target_reserve(elf_buf_t target, size_t bytes, int *error);
 784
 785 static int cb_put_phdr (vm_map_entry_t, void *);
 786 static int cb_size_segment (vm_map_entry_t, void *);
 787 static int cb_fpcount_segment(vm_map_entry_t, void *);
 788 static int cb_put_fp(vm_map_entry_t, void *);
 789
 790
 791 static int each_segment (struct proc *, segment_callback, void *, int);
 792 static int elf_corehdr (struct proc *, struct file *, struct ucred *,
 793                         int, elf_buf_t);
 794 static int elf_puthdr (struct proc *, elf_buf_t, const prstatus_t *,
 795                         const prfpregset_t *, const prpsinfo_t *, int);
 796 static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t);
 797
 798 static int elf_putsigs(struct proc *, elf_buf_t);
 799 static int elf_puttextvp(struct proc *, elf_buf_t);
 800 static int elf_putfiles(struct proc *, elf_buf_t);
 801
 802 extern int osreldate;
 803
 804 int
 805 elf_coredump(struct proc *p, struct vnode *vp, off_t limit)
 806 {
 807         struct file *fp;
 808         int error;
 809
 810         if ((error = falloc(NULL, &fp, NULL)) != 0)
 811                 return (error);
 812         fsetcred(fp, p->p_ucred);
 813
 814         fp->f_data = (caddr_t)vp;
 815         fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW;
 816         fp->f_ops = &vnops;
 817         fp->f_type = DTYPE_VNODE;
 818         VOP_UNLOCK(vp, 0, p->p_thread);
 819
 820         error = generic_elf_coredump(p, fp, limit);
 821
 822         fp->f_data = NULL;
 823         fp->f_flag = 0;
 824         fp->f_ops = &badfileops;
 825         fp->f_type = 0;
 826         fdrop(fp, p->p_thread);
 827         return (error);
 828 }
 829
 830 int
 831 generic_elf_coredump(struct proc *p, struct file *fp, off_t limit)
 832 {
 833         struct ucred *cred = p->p_ucred;
 834         int error = 0;
 835         struct sseg_closure seginfo;
 836         struct elf_buf target;
 837
 838         if (!fp)
 839                 printf("can't dump core - null fp\n");
 840
 841         /*
 842          * Size the program segments
 843          */
 844         seginfo.count = 0;
 845         seginfo.vsize = 0;
 846         each_segment(p, cb_size_segment, &seginfo, 1);
 847
 848         /*
 849          * Calculate the size of the core file header area by making
 850          * a dry run of generating it.  Nothing is written, but the
 851          * size is calculated.
 852          */
 853         bzero(&target, sizeof(target));
 854         elf_puthdr(p, &target, NULL, NULL, NULL, seginfo.count);
 855
 856         if (target.off + seginfo.vsize >= limit)
 857                 return (EFAULT);
 858
 859         /*
 860          * Allocate memory for building the header, fill it up,
 861          * and write it out.
 862          */
 863         target.off_max = target.off;
 864         target.off = 0;
 865         target.buf = malloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO);
 866
 867         if (target.buf == NULL)
 868                 return EINVAL;
 869         error = elf_corehdr(p, fp, cred, seginfo.count, &target);
 870
 871         /* Write the contents of all of the writable segments. */
 872         if (error == 0) {
 873                 Elf_Phdr *php;
 874                 int i;
 875                 int nbytes;
 876
 877                 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1;
 878                 for (i = 0; i < seginfo.count; i++) {
 879                         error = fp_write(fp, (caddr_t)php->p_vaddr,
 880                                         php->p_filesz, &nbytes);
 881                         if (error != 0)
 882                                 break;
 883                         php++;
 884                 }
 885         }
 886         free(target.buf, M_TEMP);
 887
 888         return error;
 889 }
 890
 891 /*
 892  * A callback for each_segment() to write out the segment's
 893  * program header entry.
 894  */
 895 static int
 896 cb_put_phdr(vm_map_entry_t entry, void *closure)
 897 {
 898         struct phdr_closure *phc = closure;
 899         Elf_Phdr *phdr = phc->phdr;
 900
 901         if (phc->phdr == phc->phdr_max)
 902                 return EINVAL;
 903
 904         phc->offset = round_page(phc->offset);
 905
 906         phdr->p_type = PT_LOAD;
 907         phdr->p_offset = phc->offset;
 908         phdr->p_vaddr = entry->start;
 909         phdr->p_paddr = 0;
 910         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
 911         phdr->p_align = PAGE_SIZE;
 912         phdr->p_flags = 0;
 913         if (entry->protection & VM_PROT_READ)
 914                 phdr->p_flags |= PF_R;
 915         if (entry->protection & VM_PROT_WRITE)
 916                 phdr->p_flags |= PF_W;
 917         if (entry->protection & VM_PROT_EXECUTE)
 918                 phdr->p_flags |= PF_X;
 919
 920         phc->offset += phdr->p_filesz;
 921         ++phc->phdr;
 922         return 0;
 923 }
 924
 925 /*
 926  * A callback for each_writable_segment() to gather information about
 927  * the number of segments and their total size.
 928  */
 929 static int
 930 cb_size_segment(vm_map_entry_t entry, void *closure)
 931 {
 932         struct sseg_closure *ssc = closure;
 933
 934         ++ssc->count;
 935         ssc->vsize += entry->end - entry->start;
 936         return 0;
 937 }
 938
 939 /*
 940  * A callback for each_segment() to gather information about
 941  * the number of text segments.
 942  */
 943 static int
 944 cb_fpcount_segment(vm_map_entry_t entry, void *closure)
 945 {
 946         int *count = closure;
 947         if (entry->object.vm_object->type == OBJT_VNODE)
 948                 ++*count;
 949         return 0;
 950 }
 951
 952 static int
 953 cb_put_fp(vm_map_entry_t entry, void *closure)
 954 {
 955         struct fp_closure *fpc = closure;
 956         struct vn_hdr *vnh = fpc->vnh;
 957         Elf_Phdr *phdr = &vnh->vnh_phdr;
 958         struct vnode *vp;
 959         int error;
 960
 961         if (entry->object.vm_object->type == OBJT_VNODE) {
 962                 if (vnh == fpc->vnh_max)
 963                         return EINVAL;
 964                 vp = (struct vnode *)entry->object.vm_object->handle;
 965
 966                 if (vp->v_mount)
 967                         vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
 968                 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid);
 969                 if (error)
 970                         return error;
 971
 972                 phdr->p_type = PT_LOAD;
 973                 phdr->p_offset = 0;        /* not written to core */
 974                 phdr->p_vaddr = entry->start;
 975                 phdr->p_paddr = 0;
 976                 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
 977                 phdr->p_align = PAGE_SIZE;
 978                 phdr->p_flags = 0;
 979                 if (entry->protection & VM_PROT_READ)
 980                         phdr->p_flags |= PF_R;
 981                 if (entry->protection & VM_PROT_WRITE)
 982                         phdr->p_flags |= PF_W;
 983                 if (entry->protection & VM_PROT_EXECUTE)
 984                         phdr->p_flags |= PF_X;
 985                 ++fpc->vnh;
 986                 ++fpc->count;
 987         }
 988         return 0;
 989 }
 990
 991 /*
 992  * For each writable segment in the process's memory map, call the given
 993  * function with a pointer to the map entry and some arbitrary
 994  * caller-supplied data.
 995  */
 996 static int
 997 each_segment(struct proc *p, segment_callback func, void *closure, int writable)
 998 {
 999         int error = 0;
1000         vm_map_t map = &p->p_vmspace->vm_map;
1001         vm_map_entry_t entry;
1002
1003         for (entry = map->header.next; error == 0 && entry != &map->header;
1004             entry = entry->next) {
1005                 vm_object_t obj;
1006
1007                 /*
1008                  * Don't dump inaccessible mappings, deal with legacy
1009                  * coredump mode.
1010                  *
1011                  * Note that read-only segments related to the elf binary
1012                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1013                  * need to arbitrarily ignore such segments.
1014                  */
1015                 if (elf_legacy_coredump) {
1016                         if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW)
1017                                 continue;
1018                 } else {
1019                         if (writable && (entry->protection & VM_PROT_ALL) == 0)
1020                                 continue;
1021                 }
1022
1023                 /*
1024                  * Dont include memory segment in the coredump if
1025                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1026                  * madvise(2).  Do not dump submaps (i.e. parts of the
1027                  * kernel map).
1028                  */
1029                 if (writable && entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1030                         continue;
1031
1032                 if ((obj = entry->object.vm_object) == NULL)
1033                         continue;
1034
1035                 /* Find the deepest backing object. */
1036                 while (obj->backing_object != NULL)
1037                         obj = obj->backing_object;
1038
1039                 /* Ignore memory-mapped devices and such things. */
1040                 if (obj->type != OBJT_DEFAULT &&
1041                     obj->type != OBJT_SWAP &&
1042                     obj->type != OBJT_VNODE)
1043                         continue;
1044
1045                 error = (*func)(entry, closure);
1046         }
1047         return error;
1048 }
1049
1050 static
1051 void *
1052 target_reserve(elf_buf_t target, size_t bytes, int *error)
1053 {
1054     void *res = NULL;
1055
1056     if (target->buf) {
1057             if (target->off + bytes > target->off_max)
1058                     *error = EINVAL;
1059             else
1060                     res = target->buf + target->off;
1061     }
1062     target->off += bytes;
1063     return (res);
1064 }
1065
1066 /*
1067  * Write the core file header to the file, including padding up to
1068  * the page boundary.
1069  */
1070 static int
1071 elf_corehdr(struct proc *p, struct file *fp, struct ucred *cred, int numsegs,
1072             elf_buf_t target)
1073 {
1074         struct {
1075                 prstatus_t status;
1076                 prfpregset_t fpregset;
1077                 prpsinfo_t psinfo;
1078         } *tempdata;
1079         int error;
1080         prstatus_t *status;
1081         prfpregset_t *fpregset;
1082         prpsinfo_t *psinfo;
1083         int nbytes;
1084         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK);
1085         status = &tempdata->status;
1086         fpregset = &tempdata->fpregset;
1087         psinfo = &tempdata->psinfo;
1088
1089         /* Gather the information for the header. */
1090         status->pr_version = PRSTATUS_VERSION;
1091         status->pr_statussz = sizeof(prstatus_t);
1092         status->pr_gregsetsz = sizeof(gregset_t);
1093         status->pr_fpregsetsz = sizeof(fpregset_t);
1094         status->pr_osreldate = osreldate;
1095         status->pr_cursig = p->p_sig;
1096         status->pr_pid = p->p_pid;
1097         fill_regs(p, &status->pr_reg);
1098
1099         fill_fpregs(p, fpregset);
1100
1101         psinfo->pr_version = PRPSINFO_VERSION;
1102         psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1103         strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1);
1104
1105         /* XXX - We don't fill in the command line arguments properly yet. */
1106         strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
1107
1108         /* Fill in the header. */
1109         error = elf_puthdr(p, target, status, fpregset, psinfo, numsegs);
1110
1111         free(tempdata, M_TEMP);
1112
1113         /* Write it to the core file. */
1114         if (error == 0)
1115                 error = fp_write(fp, target->buf, target->off, &nbytes);
1116         return error;
1117 }
1118
1119 static int
1120 elf_puthdr(struct proc *p, elf_buf_t target, const prstatus_t *status,
1121         const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
1122 {
1123         int error = 0;
1124         size_t phoff;
1125         size_t noteoff;
1126         size_t notesz;
1127         Elf_Ehdr *ehdr;
1128         Elf_Phdr *phdr;
1129
1130         ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error);
1131
1132         phoff = target->off;
1133         phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error);
1134
1135         noteoff = target->off;
1136         if (error == 0) {
1137                 error = elf_putnote(target, "FreeBSD", NT_PRSTATUS,
1138                                         status, sizeof *status);
1139         }
1140         if (error == 0) {
1141                 error = elf_putnote(target, "FreeBSD", NT_FPREGSET,
1142                                         fpregset, sizeof *fpregset);
1143         }
1144         if (error == 0) {
1145                 error = elf_putnote(target, "FreeBSD", NT_PRPSINFO,
1146                                         psinfo, sizeof *psinfo);
1147         }
1148         notesz = target->off - noteoff;
1149
1150         /*
1151          * put extra cruft for dumping process state here
1152          *  - we really want it be before all the program
1153          *    mappings
1154          *  - we just need to update the offset accordingly
1155          *    and GDB will be none the wiser.
1156          */
1157         if (error == 0)
1158                 error = elf_puttextvp(p, target);
1159         if (error == 0)
1160                 error = elf_putsigs(p, target);
1161         if (error == 0)
1162                 error = elf_putfiles(p, target);
1163
1164         /*
1165          * Align up to a page boundary for the program segments.  The
1166          * actual data will be written to the outptu file, not to elf_buf_t,
1167          * so we do not have to do any further bounds checking.
1168          */
1169         target->off = round_page(target->off);
1170         if (error == 0 && ehdr != NULL) {
1171                 /*
1172                  * Fill in the ELF header.
1173                  */
1174                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1175                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1176                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1177                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1178                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1179                 ehdr->e_ident[EI_DATA] = ELF_DATA;
1180                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1181                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1182                 ehdr->e_ident[EI_ABIVERSION] = 0;
1183                 ehdr->e_ident[EI_PAD] = 0;
1184                 ehdr->e_type = ET_CORE;
1185                 ehdr->e_machine = ELF_ARCH;
1186                 ehdr->e_version = EV_CURRENT;
1187                 ehdr->e_entry = 0;
1188                 ehdr->e_phoff = phoff;
1189                 ehdr->e_flags = 0;
1190                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1191                 ehdr->e_phentsize = sizeof(Elf_Phdr);
1192                 ehdr->e_phnum = numsegs + 1;
1193                 ehdr->e_shentsize = sizeof(Elf_Shdr);
1194                 ehdr->e_shnum = 0;
1195                 ehdr->e_shstrndx = SHN_UNDEF;
1196         }
1197         if (error == 0 && phdr != NULL) {
1198                 /*
1199                  * Fill in the program header entries.
1200                  */
1201                 struct phdr_closure phc;
1202
1203                 /* The note segement. */
1204                 phdr->p_type = PT_NOTE;
1205                 phdr->p_offset = noteoff;
1206                 phdr->p_vaddr = 0;
1207                 phdr->p_paddr = 0;
1208                 phdr->p_filesz = notesz;
1209                 phdr->p_memsz = 0;
1210                 phdr->p_flags = 0;
1211                 phdr->p_align = 0;
1212                 ++phdr;
1213
1214                 /* All the writable segments from the program. */
1215                 phc.phdr = phdr;
1216                 phc.phdr_max = phdr + numsegs;
1217                 phc.offset = target->off;
1218                 each_segment(p, cb_put_phdr, &phc, 1);
1219         }
1220         return (error);
1221 }
1222
1223 static int
1224 elf_putnote(elf_buf_t target, const char *name, int type,
1225             const void *desc, size_t descsz)
1226 {
1227         int error = 0;
1228         char *dst;
1229         Elf_Note note;
1230
1231         note.n_namesz = strlen(name) + 1;
1232         note.n_descsz = descsz;
1233         note.n_type = type;
1234         dst = target_reserve(target, sizeof(note), &error);
1235         if (dst != NULL)
1236                 bcopy(&note, dst, sizeof note);
1237         dst = target_reserve(target, note.n_namesz, &error);
1238         if (dst != NULL)
1239                 bcopy(name, dst, note.n_namesz);
1240         target->off = roundup2(target->off, sizeof(Elf_Size));
1241         dst = target_reserve(target, note.n_descsz, &error);
1242         if (dst != NULL)
1243                 bcopy(desc, dst, note.n_descsz);
1244         target->off = roundup2(target->off, sizeof(Elf_Size));
1245         return(error);
1246 }
1247
1248
1249 static int
1250 elf_putsigs(struct proc *p, elf_buf_t target)
1251 {
1252         int error = 0;
1253         struct ckpt_siginfo *csi;
1254
1255         csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error);
1256         if (csi) {
1257                 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo);
1258                 bcopy(p->p_procsig, &csi->csi_procsig, sizeof(struct procsig));
1259                 bcopy(p->p_procsig->ps_sigacts, &csi->csi_sigacts, sizeof(struct sigacts));
1260                 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval));
1261                 csi->csi_sigparent = p->p_sigparent;
1262         }
1263         return(error);
1264 }
1265
1266 static int
1267 elf_putfiles(struct proc *p, elf_buf_t target)
1268 {
1269         int error = 0;
1270         int i;
1271         struct ckpt_filehdr *cfh = NULL;
1272         struct ckpt_fileinfo *cfi;
1273         struct file *fp;
1274         struct vnode *vp;
1275         /*
1276          * the duplicated loop is gross, but it was the only way
1277          * to eliminate uninitialized variable warnings
1278          */
1279         cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error);
1280         if (cfh) {
1281                 cfh->cfh_nfiles = 0;
1282         }
1283
1284         /*
1285          * ignore STDIN/STDERR/STDOUT
1286          */
1287         for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) {
1288                 if ((fp = p->p_fd->fd_ofiles[i]) == NULL)
1289                         continue;
1290                 if (fp->f_type != DTYPE_VNODE)
1291                         continue;
1292                 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo), &error);
1293                 if (cfi) {
1294                         cfi->cfi_index = -1;
1295                         vp = (struct vnode *)fp->f_data;
1296                         /*
1297                          * it looks like a bug in ptrace is marking
1298                          * a non-vnode as a vnode - until we find the
1299                          * root cause this will at least prevent
1300                          * further panics from truss
1301                          */
1302                         if (vp == NULL || vp->v_mount == NULL)
1303                                 continue;
1304                         cfh->cfh_nfiles++;
1305                         cfi->cfi_index = i;
1306                         cfi->cfi_flags = fp->f_flag;
1307                         cfi->cfi_offset = fp->f_offset;
1308                         cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1309                         error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid);
1310                 }
1311         }
1312         return(error);
1313 }
1314
1315 static int
1316 elf_puttextvp(struct proc *p, elf_buf_t target)
1317 {
1318         int error = 0;
1319         int *vn_count;
1320         struct fp_closure fpc;
1321         struct ckpt_vminfo *vminfo;
1322
1323         vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error);
1324         if (vminfo != NULL) {
1325                 vminfo->cvm_dsize = p->p_vmspace->vm_dsize;
1326                 vminfo->cvm_tsize = p->p_vmspace->vm_tsize;
1327                 vminfo->cvm_daddr = p->p_vmspace->vm_daddr;
1328                 vminfo->cvm_taddr = p->p_vmspace->vm_taddr;
1329         }
1330
1331         fpc.count = 0;
1332         vn_count = target_reserve(target, sizeof(int), &error);
1333         if (target->buf != NULL) {
1334                 fpc.vnh = (struct vn_hdr *)(target->buf + target->off);
1335                 fpc.vnh_max = fpc.vnh +
1336                         (target->off_max - target->off) / sizeof(struct vn_hdr);
1337                 error = each_segment(p, cb_put_fp, &fpc, 0);
1338                 if (vn_count)
1339                         *vn_count = fpc.count;
1340         } else {
1341                 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0);
1342         }
1343         target->off += fpc.count * sizeof(struct vn_hdr);
1344         return(error);
1345 }
1346
1347
1348 /*
1349  * Tell kern_execve.c about it, with a little help from the linker.
1350  */
1351 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1352 EXEC_SET(elf, elf_execsw);