sys/kern/imgact_elf.c

   1 /*-
   2  * Copyright (c) 2000 David O'Brien
   3  * Copyright (c) 1995-1996 Søren Schmidt
   4  * Copyright (c) 1996 Peter Wemm
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer
  12  *    in this position and unchanged.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. The name of the author may not be used to endorse or promote products
  17  *    derived from this software without specific prior written permission
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29  *
  30  * $FreeBSD: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $
  31  */
  32
  33 #include <sys/param.h>
  34 #include <sys/exec.h>
  35 #include <sys/fcntl.h>
  36 #include <sys/file.h>
  37 #include <sys/imgact.h>
  38 #include <sys/imgact_elf.h>
  39 #include <sys/kernel.h>
  40 #include <sys/malloc.h>
  41 #include <sys/mman.h>
  42 #include <sys/systm.h>
  43 #include <sys/proc.h>
  44 #include <sys/nlookup.h>
  45 #include <sys/pioctl.h>
  46 #include <sys/procfs.h>
  47 #include <sys/resourcevar.h>
  48 #include <sys/signalvar.h>
  49 #include <sys/stat.h>
  50 #include <sys/syscall.h>
  51 #include <sys/sysctl.h>
  52 #include <sys/sysent.h>
  53 #include <sys/vnode.h>
  54 #include <sys/eventhandler.h>
  55
  56 #include <cpu/lwbuf.h>
  57
  58 #include <vm/vm.h>
  59 #include <vm/vm_kern.h>
  60 #include <vm/vm_param.h>
  61 #include <vm/pmap.h>
  62 #include <sys/lock.h>
  63 #include <vm/vm_map.h>
  64 #include <vm/vm_object.h>
  65 #include <vm/vm_extern.h>
  66
  67 #include <machine/elf.h>
  68 #include <machine/md_var.h>
  69 #include <sys/mount.h>
  70 #include <sys/ckpt.h>
  71
  72 #define OLD_EI_BRAND    8
  73 #define truncps(va,ps)  ((va) & ~(ps - 1))
  74 #define aligned(a,t)    (truncps((u_long)(a), sizeof(t)) == (u_long)(a))
  75
  76 static int __elfN(check_header)(const Elf_Ehdr *hdr);
  77 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
  78     const char *interp, int32_t *osrel);
  79 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
  80     u_long *entry);
  81 static int __elfN(load_section)(struct proc *p,
  82     struct vmspace *vmspace, struct vnode *vp,
  83     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
  84     vm_prot_t prot);
  85 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
  86 static boolean_t __elfN(bsd_trans_osrel)(const Elf_Note *note,
  87     int32_t *osrel);
  88 static boolean_t __elfN(check_note)(struct image_params *imgp,
  89     Elf_Brandnote *checknote, int32_t *osrel);
  90 static boolean_t check_PT_NOTE(struct image_params *imgp,
  91     Elf_Brandnote *checknote, int32_t *osrel, const Elf_Phdr * pnote);
  92 static boolean_t extract_interpreter(struct image_params *imgp,
  93     const Elf_Phdr *pinterpreter, char *data);
  94
  95 static int elf_legacy_coredump = 0;
  96 static int __elfN(fallback_brand) = -1;
  97 #if defined(__x86_64__)
  98 SYSCTL_NODE(_kern, OID_AUTO, elf64, CTLFLAG_RW, 0, "");
  99 SYSCTL_INT(_debug, OID_AUTO, elf64_legacy_coredump, CTLFLAG_RW,
 100     &elf_legacy_coredump, 0, "legacy coredump mode");
 101 SYSCTL_INT(_kern_elf64, OID_AUTO, fallback_brand, CTLFLAG_RW,
 102     &elf64_fallback_brand, 0, "ELF64 brand of last resort");
 103 TUNABLE_INT("kern.elf64.fallback_brand", &elf64_fallback_brand);
 104 #else /* i386 assumed */
 105 SYSCTL_NODE(_kern, OID_AUTO, elf32, CTLFLAG_RW, 0, "");
 106 SYSCTL_INT(_debug, OID_AUTO, elf32_legacy_coredump, CTLFLAG_RW,
 107     &elf_legacy_coredump, 0, "legacy coredump mode");
 108 SYSCTL_INT(_kern_elf32, OID_AUTO, fallback_brand, CTLFLAG_RW,
 109     &elf32_fallback_brand, 0, "ELF32 brand of last resort");
 110 TUNABLE_INT("kern.elf32.fallback_brand", &elf32_fallback_brand);
 111 #endif
 112
 113 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
 114
 115 static const char DRAGONFLY_ABI_VENDOR[] = "DragonFly";
 116 static const char FREEBSD_ABI_VENDOR[]   = "FreeBSD";
 117
 118 Elf_Brandnote __elfN(dragonfly_brandnote) = {
 119         .hdr.n_namesz   = sizeof(DRAGONFLY_ABI_VENDOR),
 120         .hdr.n_descsz   = sizeof(int32_t),
 121         .hdr.n_type     = 1,
 122         .vendor         = DRAGONFLY_ABI_VENDOR,
 123         .flags          = BN_TRANSLATE_OSREL,
 124         .trans_osrel    = __elfN(bsd_trans_osrel),
 125 };
 126
 127 Elf_Brandnote __elfN(freebsd_brandnote) = {
 128         .hdr.n_namesz   = sizeof(FREEBSD_ABI_VENDOR),
 129         .hdr.n_descsz   = sizeof(int32_t),
 130         .hdr.n_type     = 1,
 131         .vendor         = FREEBSD_ABI_VENDOR,
 132         .flags          = BN_TRANSLATE_OSREL,
 133         .trans_osrel    = __elfN(bsd_trans_osrel),
 134 };
 135
 136 int
 137 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
 138 {
 139         int i;
 140
 141         for (i = 0; i < MAX_BRANDS; i++) {
 142                 if (elf_brand_list[i] == NULL) {
 143                         elf_brand_list[i] = entry;
 144                         break;
 145                 }
 146         }
 147         if (i == MAX_BRANDS) {
 148                 uprintf("WARNING: %s: could not insert brandinfo entry: %p\n",
 149                         __func__, entry);
 150                 return (-1);
 151         }
 152         return (0);
 153 }
 154
 155 int
 156 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
 157 {
 158         int i;
 159
 160         for (i = 0; i < MAX_BRANDS; i++) {
 161                 if (elf_brand_list[i] == entry) {
 162                         elf_brand_list[i] = NULL;
 163                         break;
 164                 }
 165         }
 166         if (i == MAX_BRANDS)
 167                 return (-1);
 168         return (0);
 169 }
 170
 171 /*
 172  * Check if an elf brand is being used anywhere in the system.
 173  *
 174  * Used by the linux emulation module unloader.  This isn't safe from
 175  * races.
 176  */
 177 struct elf_brand_inuse_info {
 178         int rval;
 179         Elf_Brandinfo *entry;
 180 };
 181
 182 static int elf_brand_inuse_callback(struct proc *p, void *data);
 183
 184 int
 185 __elfN(brand_inuse)(Elf_Brandinfo *entry)
 186 {
 187         struct elf_brand_inuse_info info;
 188
 189         info.rval = FALSE;
 190         info.entry = entry;
 191         allproc_scan(elf_brand_inuse_callback, entry);
 192         return (info.rval);
 193 }
 194
 195 static
 196 int
 197 elf_brand_inuse_callback(struct proc *p, void *data)
 198 {
 199         struct elf_brand_inuse_info *info = data;
 200
 201         if (p->p_sysent == info->entry->sysvec) {
 202                 info->rval = TRUE;
 203                 return (-1);
 204         }
 205         return (0);
 206 }
 207
 208 static int
 209 __elfN(check_header)(const Elf_Ehdr *hdr)
 210 {
 211         Elf_Brandinfo *bi;
 212         int i;
 213
 214         if (!IS_ELF(*hdr) ||
 215             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
 216             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
 217             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
 218             hdr->e_phentsize != sizeof(Elf_Phdr) ||
 219             hdr->e_ehsize != sizeof(Elf_Ehdr) ||
 220             hdr->e_version != ELF_TARG_VER)
 221                 return (ENOEXEC);
 222
 223         /*
 224          * Make sure we have at least one brand for this machine.
 225          */
 226
 227         for (i = 0; i < MAX_BRANDS; i++) {
 228                 bi = elf_brand_list[i];
 229                 if (bi != NULL && bi->machine == hdr->e_machine)
 230                         break;
 231         }
 232         if (i == MAX_BRANDS)
 233                 return (ENOEXEC);
 234
 235         return (0);
 236 }
 237
 238 static int
 239 __elfN(load_section)(struct proc *p, struct vmspace *vmspace, struct vnode *vp,
 240                  vm_offset_t offset, caddr_t vmaddr, size_t memsz,
 241                  size_t filsz, vm_prot_t prot)
 242 {
 243         size_t map_len;
 244         vm_offset_t map_addr;
 245         int error, rv, cow;
 246         int count;
 247         size_t copy_len;
 248         vm_object_t object;
 249         vm_offset_t file_addr;
 250
 251         object = vp->v_object;
 252         error = 0;
 253
 254         vm_object_hold(object);
 255
 256         /*
 257          * It's necessary to fail if the filsz + offset taken from the
 258          * header is greater than the actual file pager object's size.
 259          * If we were to allow this, then the vm_map_find() below would
 260          * walk right off the end of the file object and into the ether.
 261          *
 262          * While I'm here, might as well check for something else that
 263          * is invalid: filsz cannot be greater than memsz.
 264          */
 265         if ((off_t)filsz + offset > vp->v_filesize || filsz > memsz) {
 266                 uprintf("elf_load_section: truncated ELF file\n");
 267                 vm_object_drop(object);
 268                 return (ENOEXEC);
 269         }
 270
 271         map_addr = trunc_page((vm_offset_t)vmaddr);
 272         file_addr = trunc_page(offset);
 273
 274         /*
 275          * We have two choices.  We can either clear the data in the last page
 276          * of an oversized mapping, or we can start the anon mapping a page
 277          * early and copy the initialized data into that first page.  We
 278          * choose the second..
 279          */
 280         if (memsz > filsz)
 281                 map_len = trunc_page(offset+filsz) - file_addr;
 282         else
 283                 map_len = round_page(offset+filsz) - file_addr;
 284
 285         if (map_len != 0) {
 286                 vm_object_reference_locked(object);
 287
 288                 /* cow flags: don't dump readonly sections in core */
 289                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
 290                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
 291
 292                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 293                 vm_map_lock(&vmspace->vm_map);
 294                 rv = vm_map_insert(&vmspace->vm_map, &count,
 295                                       object,
 296                                       file_addr,        /* file offset */
 297                                       map_addr,         /* virtual start */
 298                                       map_addr + map_len,/* virtual end */
 299                                       VM_MAPTYPE_NORMAL,
 300                                       prot, VM_PROT_ALL,
 301                                       cow);
 302                 vm_map_unlock(&vmspace->vm_map);
 303                 vm_map_entry_release(count);
 304                 if (rv != KERN_SUCCESS) {
 305                         vm_object_deallocate(object);
 306                         vm_object_drop(object);
 307                         return (EINVAL);
 308                 }
 309
 310                 /* we can stop now if we've covered it all */
 311                 if (memsz == filsz) {
 312                         vm_object_drop(object);
 313                         return (0);
 314                 }
 315         }
 316
 317
 318         /*
 319          * We have to get the remaining bit of the file into the first part
 320          * of the oversized map segment.  This is normally because the .data
 321          * segment in the file is extended to provide bss.  It's a neat idea
 322          * to try and save a page, but it's a pain in the behind to implement.
 323          */
 324         copy_len = (offset + filsz) - trunc_page(offset + filsz);
 325         map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
 326         map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
 327
 328         /* This had damn well better be true! */
 329         if (map_len != 0) {
 330                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 331                 vm_map_lock(&vmspace->vm_map);
 332                 rv = vm_map_insert(&vmspace->vm_map, &count,
 333                                         NULL, 0,
 334                                         map_addr, map_addr + map_len,
 335                                         VM_MAPTYPE_NORMAL,
 336                                         VM_PROT_ALL, VM_PROT_ALL,
 337                                         0);
 338                 vm_map_unlock(&vmspace->vm_map);
 339                 vm_map_entry_release(count);
 340                 if (rv != KERN_SUCCESS) {
 341                         vm_object_drop(object);
 342                         return (EINVAL);
 343                 }
 344         }
 345
 346         if (copy_len != 0) {
 347                 vm_page_t m;
 348                 struct lwbuf *lwb;
 349                 struct lwbuf lwb_cache;
 350
 351                 m = vm_fault_object_page(object, trunc_page(offset + filsz),
 352                                          VM_PROT_READ, 0, &error);
 353                 if (m) {
 354                         lwb = lwbuf_alloc(m, &lwb_cache);
 355                         error = copyout((caddr_t)lwbuf_kva(lwb),
 356                                         (caddr_t)map_addr, copy_len);
 357                         lwbuf_free(lwb);
 358                         vm_page_unhold(m);
 359                 }
 360                 if (error) {
 361                         vm_object_drop(object);
 362                         return (error);
 363                 }
 364         }
 365
 366         vm_object_drop(object);
 367         /*
 368          * set it to the specified protection
 369          */
 370         vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len,
 371                        prot, FALSE);
 372
 373         return (error);
 374 }
 375
 376 /*
 377  * Load the file "file" into memory.  It may be either a shared object
 378  * or an executable.
 379  *
 380  * The "addr" reference parameter is in/out.  On entry, it specifies
 381  * the address where a shared object should be loaded.  If the file is
 382  * an executable, this value is ignored.  On exit, "addr" specifies
 383  * where the file was actually loaded.
 384  *
 385  * The "entry" reference parameter is out only.  On exit, it specifies
 386  * the entry point for the loaded file.
 387  */
 388 static int
 389 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, u_long *entry)
 390 {
 391         struct {
 392                 struct nlookupdata nd;
 393                 struct vattr attr;
 394                 struct image_params image_params;
 395         } *tempdata;
 396         const Elf_Ehdr *hdr = NULL;
 397         const Elf_Phdr *phdr = NULL;
 398         struct nlookupdata *nd;
 399         struct vmspace *vmspace = p->p_vmspace;
 400         struct vattr *attr;
 401         struct image_params *imgp;
 402         struct mount *topmnt;
 403         vm_prot_t prot;
 404         u_long rbase;
 405         u_long base_addr = 0;
 406         int error, i, numsegs;
 407
 408         tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
 409         nd = &tempdata->nd;
 410         attr = &tempdata->attr;
 411         imgp = &tempdata->image_params;
 412
 413         /*
 414          * Initialize part of the common data
 415          */
 416         imgp->proc = p;
 417         imgp->attr = attr;
 418         imgp->firstpage = NULL;
 419         imgp->image_header = NULL;
 420         imgp->vp = NULL;
 421
 422         error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW);
 423         if (error == 0)
 424                 error = nlookup(nd);
 425         if (error == 0)
 426                 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
 427         topmnt = nd->nl_nch.mount;
 428         nlookup_done(nd);
 429         if (error)
 430                 goto fail;
 431
 432         /*
 433          * Check permissions, modes, uid, etc on the file, and "open" it.
 434          */
 435         error = exec_check_permissions(imgp, topmnt);
 436         if (error) {
 437                 vn_unlock(imgp->vp);
 438                 goto fail;
 439         }
 440
 441         error = exec_map_first_page(imgp);
 442         /*
 443          * Also make certain that the interpreter stays the same, so set
 444          * its VTEXT flag, too.
 445          */
 446         if (error == 0)
 447                 vsetflags(imgp->vp, VTEXT);
 448         vn_unlock(imgp->vp);
 449         if (error)
 450                 goto fail;
 451
 452         hdr = (const Elf_Ehdr *)imgp->image_header;
 453         if ((error = __elfN(check_header)(hdr)) != 0)
 454                 goto fail;
 455         if (hdr->e_type == ET_DYN)
 456                 rbase = *addr;
 457         else if (hdr->e_type == ET_EXEC)
 458                 rbase = 0;
 459         else {
 460                 error = ENOEXEC;
 461                 goto fail;
 462         }
 463
 464         /* Only support headers that fit within first page for now      */
 465         /*    (multiplication of two Elf_Half fields will not overflow) */
 466         if ((hdr->e_phoff > PAGE_SIZE) ||
 467             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
 468                 error = ENOEXEC;
 469                 goto fail;
 470         }
 471
 472         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 473         if (!aligned(phdr, Elf_Addr)) {
 474                 error = ENOEXEC;
 475                 goto fail;
 476         }
 477
 478         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
 479                 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
 480                         /* Loadable segment */
 481                         prot = 0;
 482                         if (phdr[i].p_flags & PF_X)
 483                                 prot |= VM_PROT_EXECUTE;
 484                         if (phdr[i].p_flags & PF_W)
 485                                 prot |= VM_PROT_WRITE;
 486                         if (phdr[i].p_flags & PF_R)
 487                                 prot |= VM_PROT_READ;
 488
 489                         error = __elfN(load_section)(
 490                                     p, vmspace, imgp->vp,
 491                                     phdr[i].p_offset,
 492                                     (caddr_t)phdr[i].p_vaddr +
 493                                     rbase,
 494                                     phdr[i].p_memsz,
 495                                     phdr[i].p_filesz, prot);
 496                         if (error != 0)
 497                                 goto fail;
 498                         /*
 499                          * Establish the base address if this is the
 500                          * first segment.
 501                          */
 502                         if (numsegs == 0)
 503                                 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
 504                         numsegs++;
 505                 }
 506         }
 507         *addr = base_addr;
 508         *entry = (unsigned long)hdr->e_entry + rbase;
 509
 510 fail:
 511         if (imgp->firstpage)
 512                 exec_unmap_first_page(imgp);
 513         if (imgp->vp) {
 514                 vrele(imgp->vp);
 515                 imgp->vp = NULL;
 516         }
 517         kfree(tempdata, M_TEMP);
 518
 519         return (error);
 520 }
 521
 522 static Elf_Brandinfo *
 523 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
 524     int32_t *osrel)
 525 {
 526         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
 527         Elf_Brandinfo *bi;
 528         boolean_t ret;
 529         int i;
 530
 531         /* We support four types of branding -- (1) the ELF EI_OSABI field
 532          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
 533          * branding within the ELF header, (3) path of the `interp_path' field,
 534          * and (4) the ".note.ABI-tag" ELF section.
 535          */
 536
 537         /* Look for an ".note.ABI-tag" ELF section */
 538         for (i = 0; i < MAX_BRANDS; i++) {
 539                 bi = elf_brand_list[i];
 540
 541                 if (bi == NULL)
 542                         continue;
 543                 if (hdr->e_machine == bi->machine && (bi->flags &
 544                     (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
 545                         ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
 546                         if (ret)
 547                                 return (bi);
 548                 }
 549         }
 550
 551         /* If the executable has a brand, search for it in the brand list. */
 552         for (i = 0;  i < MAX_BRANDS;  i++) {
 553                 bi = elf_brand_list[i];
 554
 555                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
 556                         continue;
 557                 if (hdr->e_machine == bi->machine &&
 558                     (hdr->e_ident[EI_OSABI] == bi->brand ||
 559                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
 560                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
 561                         return (bi);
 562         }
 563
 564         /* Lacking a known brand, search for a recognized interpreter. */
 565         if (interp != NULL) {
 566                 for (i = 0;  i < MAX_BRANDS;  i++) {
 567                         bi = elf_brand_list[i];
 568
 569                         if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
 570                                 continue;
 571                         if (hdr->e_machine == bi->machine &&
 572                             strcmp(interp, bi->interp_path) == 0)
 573                                 return (bi);
 574                 }
 575         }
 576
 577         /* Lacking a recognized interpreter, try the default brand */
 578         for (i = 0; i < MAX_BRANDS; i++) {
 579                 bi = elf_brand_list[i];
 580
 581                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
 582                         continue;
 583                 if (hdr->e_machine == bi->machine &&
 584                     __elfN(fallback_brand) == bi->brand)
 585                         return (bi);
 586         }
 587         return (NULL);
 588 }
 589
 590 static int
 591 __CONCAT(exec_,__elfN(imgact))(struct image_params *imgp)
 592 {
 593         const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
 594         const Elf_Phdr *phdr;
 595         Elf_Auxargs *elf_auxargs;
 596         struct vmspace *vmspace;
 597         vm_prot_t prot;
 598         u_long text_size = 0, data_size = 0, total_size = 0;
 599         u_long text_addr = 0, data_addr = 0;
 600         u_long seg_size, seg_addr;
 601         u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
 602         int32_t osrel = 0;
 603         int error = 0, i, n;
 604         boolean_t failure;
 605         char *interp = NULL;
 606         const char *newinterp = NULL;
 607         Elf_Brandinfo *brand_info;
 608         char *path;
 609
 610         /*
 611          * Do we have a valid ELF header ?
 612          *
 613          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later if a particular
 614          * brand doesn't support it.  Both DragonFly platforms do by default.
 615          */
 616         if (__elfN(check_header)(hdr) != 0 ||
 617             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
 618                 return (-1);
 619
 620         /*
 621          * From here on down, we return an errno, not -1, as we've
 622          * detected an ELF file.
 623          */
 624
 625         if ((hdr->e_phoff > PAGE_SIZE) ||
 626             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
 627                 /* Only support headers in first page for now */
 628                 return (ENOEXEC);
 629         }
 630         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 631         if (!aligned(phdr, Elf_Addr))
 632                 return (ENOEXEC);
 633         n = 0;
 634         baddr = 0;
 635         for (i = 0; i < hdr->e_phnum; i++) {
 636                 if (phdr[i].p_type == PT_LOAD) {
 637                         if (n == 0)
 638                                 baddr = phdr[i].p_vaddr;
 639                         n++;
 640                         continue;
 641                 }
 642                 if (phdr[i].p_type == PT_INTERP) {
 643                         /*
 644                          * If interp is already defined there are more than
 645                          * one PT_INTERP program headers present.  Take only
 646                          * the first one and ignore the rest.
 647                          */
 648                         if (interp != NULL)
 649                                 continue;
 650
 651                         if (phdr[i].p_filesz == 0 ||
 652                             phdr[i].p_filesz > PAGE_SIZE ||
 653                             phdr[i].p_filesz > MAXPATHLEN)
 654                                 return (ENOEXEC);
 655
 656                         interp = kmalloc(phdr[i].p_filesz, M_TEMP, M_WAITOK);
 657                         failure = extract_interpreter(imgp, &phdr[i], interp);
 658                         if (failure) {
 659                                 kfree(interp, M_TEMP);
 660                                 return (ENOEXEC);
 661                         }
 662                         continue;
 663                 }
 664         }
 665
 666         brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
 667         if (brand_info == NULL) {
 668                 uprintf("ELF binary type \"%u\" not known.\n",
 669                     hdr->e_ident[EI_OSABI]);
 670                 if (interp != NULL)
 671                         kfree(interp, M_TEMP);
 672                 return (ENOEXEC);
 673         }
 674         if (hdr->e_type == ET_DYN) {
 675                 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
 676                         if (interp != NULL)
 677                                 kfree(interp, M_TEMP);
 678                         return (ENOEXEC);
 679                 }
 680                 /*
 681                  * Honour the base load address from the dso if it is
 682                  * non-zero for some reason.
 683                  */
 684                 if (baddr == 0)
 685                         et_dyn_addr = ET_DYN_LOAD_ADDR;
 686                 else
 687                         et_dyn_addr = 0;
 688         } else
 689                 et_dyn_addr = 0;
 690
 691         if (interp != NULL && brand_info->interp_newpath != NULL)
 692                 newinterp = brand_info->interp_newpath;
 693
 694         exec_new_vmspace(imgp, NULL);
 695
 696         /*
 697          * Yeah, I'm paranoid.  There is every reason in the world to get
 698          * VTEXT now since from here on out, there are places we can have
 699          * a context switch.  Better safe than sorry; I really don't want
 700          * the file to change while it's being loaded.
 701          */
 702         vsetflags(imgp->vp, VTEXT);
 703
 704         vmspace = imgp->proc->p_vmspace;
 705
 706         for (i = 0; i < hdr->e_phnum; i++) {
 707                 switch (phdr[i].p_type) {
 708
 709                 case PT_LOAD:   /* Loadable segment */
 710                         if (phdr[i].p_memsz == 0)
 711                                 break;
 712                         prot = 0;
 713                         if (phdr[i].p_flags & PF_X)
 714                                 prot |= VM_PROT_EXECUTE;
 715                         if (phdr[i].p_flags & PF_W)
 716                                 prot |= VM_PROT_WRITE;
 717                         if (phdr[i].p_flags & PF_R)
 718                                 prot |= VM_PROT_READ;
 719
 720                         if ((error = __elfN(load_section)(
 721                                         imgp->proc,
 722                                         vmspace,
 723                                         imgp->vp,
 724                                         phdr[i].p_offset,
 725                                         (caddr_t)phdr[i].p_vaddr + et_dyn_addr,
 726                                         phdr[i].p_memsz,
 727                                         phdr[i].p_filesz,
 728                                         prot)) != 0) {
 729                                 if (interp != NULL)
 730                                         kfree (interp, M_TEMP);
 731                                 return (error);
 732                         }
 733
 734                         /*
 735                          * If this segment contains the program headers,
 736                          * remember their virtual address for the AT_PHDR
 737                          * aux entry. Static binaries don't usually include
 738                          * a PT_PHDR entry.
 739                          */
 740                         if (phdr[i].p_offset == 0 &&
 741                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
 742                                 <= phdr[i].p_filesz)
 743                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
 744                                     et_dyn_addr;
 745
 746                         seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
 747                         seg_size = round_page(phdr[i].p_memsz +
 748                             phdr[i].p_vaddr + et_dyn_addr - seg_addr);
 749
 750                         /*
 751                          * Is this .text or .data?  We can't use
 752                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
 753                          * alpha terribly and possibly does other bad
 754                          * things so we stick to the old way of figuring
 755                          * it out:  If the segment contains the program
 756                          * entry point, it's a text segment, otherwise it
 757                          * is a data segment.
 758                          *
 759                          * Note that obreak() assumes that data_addr +
 760                          * data_size == end of data load area, and the ELF
 761                          * file format expects segments to be sorted by
 762                          * address.  If multiple data segments exist, the
 763                          * last one will be used.
 764                          */
 765                         if (hdr->e_entry >= phdr[i].p_vaddr &&
 766                             hdr->e_entry < (phdr[i].p_vaddr +
 767                             phdr[i].p_memsz)) {
 768                                 text_size = seg_size;
 769                                 text_addr = seg_addr;
 770                                 entry = (u_long)hdr->e_entry + et_dyn_addr;
 771                         } else {
 772                                 data_size = seg_size;
 773                                 data_addr = seg_addr;
 774                         }
 775                         total_size += seg_size;
 776
 777                         /*
 778                          * Check limits.  It should be safe to check the
 779                          * limits after loading the segment since we do
 780                          * not actually fault in all the segment's pages.
 781                          */
 782                         if (data_size >
 783                             imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
 784                             text_size > maxtsiz ||
 785                             total_size >
 786                             imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
 787                                 if (interp != NULL)
 788                                         kfree(interp, M_TEMP);
 789                                 error = ENOMEM;
 790                                 return (error);
 791                         }
 792                         break;
 793                 case PT_PHDR:   /* Program header table info */
 794                         proghdr = phdr[i].p_vaddr + et_dyn_addr;
 795                         break;
 796                 default:
 797                         break;
 798                 }
 799         }
 800
 801         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
 802         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
 803         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
 804         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
 805
 806         addr = ELF_RTLD_ADDR(vmspace);
 807
 808         imgp->entry_addr = entry;
 809
 810         imgp->proc->p_sysent = brand_info->sysvec;
 811         EVENTHANDLER_INVOKE(process_exec, imgp);
 812
 813         if (interp != NULL) {
 814                 int have_interp = FALSE;
 815                 if (brand_info->emul_path != NULL &&
 816                     brand_info->emul_path[0] != '\0') {
 817                         path = kmalloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 818                         ksnprintf(path, MAXPATHLEN, "%s%s",
 819                             brand_info->emul_path, interp);
 820                         error = __elfN(load_file)(imgp->proc, path, &addr,
 821                             &imgp->entry_addr);
 822                         kfree(path, M_TEMP);
 823                         if (error == 0)
 824                                 have_interp = TRUE;
 825                 }
 826                 if (!have_interp && newinterp != NULL) {
 827                         error = __elfN(load_file)(imgp->proc, newinterp,
 828                             &addr, &imgp->entry_addr);
 829                         if (error == 0)
 830                                 have_interp = TRUE;
 831                 }
 832                 if (!have_interp) {
 833                         error = __elfN(load_file)(imgp->proc, interp, &addr,
 834                             &imgp->entry_addr);
 835                 }
 836                 if (error != 0) {
 837                         uprintf("ELF interpreter %s not found\n", interp);
 838                         kfree(interp, M_TEMP);
 839                         return (error);
 840                 }
 841                 kfree(interp, M_TEMP);
 842         } else
 843                 addr = et_dyn_addr;
 844
 845         /*
 846          * Construct auxargs table (used by the fixup routine)
 847          */
 848         elf_auxargs = kmalloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
 849         elf_auxargs->execfd = -1;
 850         elf_auxargs->phdr = proghdr;
 851         elf_auxargs->phent = hdr->e_phentsize;
 852         elf_auxargs->phnum = hdr->e_phnum;
 853         elf_auxargs->pagesz = PAGE_SIZE;
 854         elf_auxargs->base = addr;
 855         elf_auxargs->flags = 0;
 856         elf_auxargs->entry = entry;
 857
 858         imgp->auxargs = elf_auxargs;
 859         imgp->interpreted = 0;
 860         imgp->proc->p_osrel = osrel;
 861
 862         return (error);
 863 }
 864
 865 int
 866 __elfN(dragonfly_fixup)(register_t **stack_base, struct image_params *imgp)
 867 {
 868         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
 869         Elf_Addr *base;
 870         Elf_Addr *pos;
 871
 872         base = (Elf_Addr *)*stack_base;
 873         pos = base + (imgp->args->argc + imgp->args->envc + 2);
 874
 875         if (args->execfd != -1)
 876                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
 877         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
 878         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
 879         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
 880         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
 881         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
 882         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
 883         AUXARGS_ENTRY(pos, AT_BASE, args->base);
 884         if (imgp->execpathp != 0)
 885                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
 886         AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate);
 887         AUXARGS_ENTRY(pos, AT_NULL, 0);
 888
 889         kfree(imgp->auxargs, M_TEMP);
 890         imgp->auxargs = NULL;
 891
 892         base--;
 893         suword(base, (long)imgp->args->argc);
 894         *stack_base = (register_t *)base;
 895         return (0);
 896 }
 897
 898 /*
 899  * Code for generating ELF core dumps.
 900  */
 901
 902 typedef int (*segment_callback)(vm_map_entry_t, void *);
 903
 904 /* Closure for cb_put_phdr(). */
 905 struct phdr_closure {
 906         Elf_Phdr *phdr;         /* Program header to fill in (incremented) */
 907         Elf_Phdr *phdr_max;     /* Pointer bound for error check */
 908         Elf_Off offset;         /* Offset of segment in core file */
 909 };
 910
 911 /* Closure for cb_size_segment(). */
 912 struct sseg_closure {
 913         int count;              /* Count of writable segments. */
 914         size_t vsize;           /* Total size of all writable segments. */
 915 };
 916
 917 /* Closure for cb_put_fp(). */
 918 struct fp_closure {
 919         struct vn_hdr *vnh;
 920         struct vn_hdr *vnh_max;
 921         int count;
 922         struct stat *sb;
 923 };
 924
 925 typedef struct elf_buf {
 926         char    *buf;
 927         size_t  off;
 928         size_t  off_max;
 929 } *elf_buf_t;
 930
 931 static void *target_reserve(elf_buf_t target, size_t bytes, int *error);
 932
 933 static int cb_put_phdr (vm_map_entry_t, void *);
 934 static int cb_size_segment (vm_map_entry_t, void *);
 935 static int cb_fpcount_segment(vm_map_entry_t, void *);
 936 static int cb_put_fp(vm_map_entry_t, void *);
 937
 938
 939 static int each_segment (struct proc *, segment_callback, void *, int);
 940 static int __elfN(corehdr)(struct lwp *, int, struct file *, struct ucred *,
 941                         int, elf_buf_t);
 942 enum putmode { WRITE, DRYRUN };
 943 static int __elfN(puthdr)(struct lwp *, elf_buf_t, int sig, enum putmode,
 944                         int, struct file *);
 945 static int elf_putallnotes(struct lwp *, elf_buf_t, int, enum putmode);
 946 static int __elfN(putnote)(elf_buf_t, const char *, int, const void *, size_t);
 947
 948 static int elf_putsigs(struct lwp *, elf_buf_t);
 949 static int elf_puttextvp(struct proc *, elf_buf_t);
 950 static int elf_putfiles(struct proc *, elf_buf_t, struct file *);
 951
 952 int
 953 __elfN(coredump)(struct lwp *lp, int sig, struct vnode *vp, off_t limit)
 954 {
 955         struct file *fp;
 956         int error;
 957
 958         if ((error = falloc(NULL, &fp, NULL)) != 0)
 959                 return (error);
 960         fsetcred(fp, lp->lwp_proc->p_ucred);
 961
 962         /*
 963          * XXX fixme.
 964          */
 965         fp->f_type = DTYPE_VNODE;
 966         fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW;
 967         fp->f_ops = &vnode_fileops;
 968         fp->f_data = vp;
 969         vn_unlock(vp);
 970
 971         error = generic_elf_coredump(lp, sig, fp, limit);
 972
 973         fp->f_type = 0;
 974         fp->f_flag = 0;
 975         fp->f_ops = &badfileops;
 976         fp->f_data = NULL;
 977         fdrop(fp);
 978         return (error);
 979 }
 980
 981 int
 982 generic_elf_coredump(struct lwp *lp, int sig, struct file *fp, off_t limit)
 983 {
 984         struct proc *p = lp->lwp_proc;
 985         struct ucred *cred = p->p_ucred;
 986         int error = 0;
 987         struct sseg_closure seginfo;
 988         struct elf_buf target;
 989
 990         if (!fp)
 991                 kprintf("can't dump core - null fp\n");
 992
 993         /*
 994          * Size the program segments
 995          */
 996         seginfo.count = 0;
 997         seginfo.vsize = 0;
 998         each_segment(p, cb_size_segment, &seginfo, 1);
 999
1000         /*
1001          * Calculate the size of the core file header area by making
1002          * a dry run of generating it.  Nothing is written, but the
1003          * size is calculated.
1004          */
1005         bzero(&target, sizeof(target));
1006         __elfN(puthdr)(lp, &target, sig, DRYRUN, seginfo.count, fp);
1007
1008         if (target.off + seginfo.vsize >= limit)
1009                 return (EFAULT);
1010
1011         /*
1012          * Allocate memory for building the header, fill it up,
1013          * and write it out.
1014          */
1015         target.off_max = target.off;
1016         target.off = 0;
1017         target.buf = kmalloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO);
1018
1019         error = __elfN(corehdr)(lp, sig, fp, cred, seginfo.count, &target);
1020
1021         /* Write the contents of all of the writable segments. */
1022         if (error == 0) {
1023                 Elf_Phdr *php;
1024                 int i;
1025                 ssize_t nbytes;
1026
1027                 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1;
1028                 for (i = 0; i < seginfo.count; i++) {
1029                         error = fp_write(fp, (caddr_t)php->p_vaddr,
1030                                         php->p_filesz, &nbytes, UIO_USERSPACE);
1031                         if (error != 0)
1032                                 break;
1033                         php++;
1034                 }
1035         }
1036         kfree(target.buf, M_TEMP);
1037
1038         return (error);
1039 }
1040
1041 /*
1042  * A callback for each_segment() to write out the segment's
1043  * program header entry.
1044  */
1045 static int
1046 cb_put_phdr(vm_map_entry_t entry, void *closure)
1047 {
1048         struct phdr_closure *phc = closure;
1049         Elf_Phdr *phdr = phc->phdr;
1050
1051         if (phc->phdr == phc->phdr_max)
1052                 return (EINVAL);
1053
1054         phc->offset = round_page(phc->offset);
1055
1056         phdr->p_type = PT_LOAD;
1057         phdr->p_offset = phc->offset;
1058         phdr->p_vaddr = entry->start;
1059         phdr->p_paddr = 0;
1060         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1061         phdr->p_align = PAGE_SIZE;
1062         phdr->p_flags = 0;
1063         if (entry->protection & VM_PROT_READ)
1064                 phdr->p_flags |= PF_R;
1065         if (entry->protection & VM_PROT_WRITE)
1066                 phdr->p_flags |= PF_W;
1067         if (entry->protection & VM_PROT_EXECUTE)
1068                 phdr->p_flags |= PF_X;
1069
1070         phc->offset += phdr->p_filesz;
1071         ++phc->phdr;
1072         return (0);
1073 }
1074
1075 /*
1076  * A callback for each_writable_segment() to gather information about
1077  * the number of segments and their total size.
1078  */
1079 static int
1080 cb_size_segment(vm_map_entry_t entry, void *closure)
1081 {
1082         struct sseg_closure *ssc = closure;
1083
1084         ++ssc->count;
1085         ssc->vsize += entry->end - entry->start;
1086         return (0);
1087 }
1088
1089 /*
1090  * A callback for each_segment() to gather information about
1091  * the number of text segments.
1092  */
1093 static int
1094 cb_fpcount_segment(vm_map_entry_t entry, void *closure)
1095 {
1096         int *count = closure;
1097         struct vnode *vp;
1098
1099         if (entry->object.vm_object->type == OBJT_VNODE) {
1100                 vp = (struct vnode *)entry->object.vm_object->handle;
1101                 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1102                         return (0);
1103                 ++*count;
1104         }
1105         return (0);
1106 }
1107
1108 static int
1109 cb_put_fp(vm_map_entry_t entry, void *closure)
1110 {
1111         struct fp_closure *fpc = closure;
1112         struct vn_hdr *vnh = fpc->vnh;
1113         Elf_Phdr *phdr = &vnh->vnh_phdr;
1114         struct vnode *vp;
1115         int error;
1116
1117         /*
1118          * If an entry represents a vnode then write out a file handle.
1119          *
1120          * If we are checkpointing a checkpoint-restored program we do
1121          * NOT record the filehandle for the old checkpoint vnode (which
1122          * is mapped all over the place).  Instead we rely on the fact
1123          * that a checkpoint-restored program does not mmap() the checkpt
1124          * vnode NOCORE, so its contents will be written out to the
1125          * new checkpoint file.  This is necessary because the 'old'
1126          * checkpoint file is typically destroyed when a new one is created
1127          * and thus cannot be used to restore the new checkpoint.
1128          *
1129          * Theoretically we could create a chain of checkpoint files and
1130          * operate the checkpointing operation kinda like an incremental
1131          * checkpoint, but a checkpoint restore would then likely wind up
1132          * referencing many prior checkpoint files and that is a bit over
1133          * the top for the purpose of the checkpoint API.
1134          */
1135         if (entry->object.vm_object->type == OBJT_VNODE) {
1136                 vp = (struct vnode *)entry->object.vm_object->handle;
1137                 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1138                         return (0);
1139                 if (vnh == fpc->vnh_max)
1140                         return (EINVAL);
1141
1142                 if (vp->v_mount)
1143                         vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1144                 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid);
1145                 if (error) {
1146                         char *freepath, *fullpath;
1147
1148                         if (vn_fullpath(curproc, vp, &fullpath, &freepath, 0)) {
1149                                 kprintf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp);
1150                         } else {
1151                                 kprintf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath);
1152                                 kfree(freepath, M_TEMP);
1153                         }
1154                         error = 0;
1155                 }
1156
1157                 phdr->p_type = PT_LOAD;
1158                 phdr->p_offset = 0;        /* not written to core */
1159                 phdr->p_vaddr = entry->start;
1160                 phdr->p_paddr = 0;
1161                 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1162                 phdr->p_align = PAGE_SIZE;
1163                 phdr->p_flags = 0;
1164                 if (entry->protection & VM_PROT_READ)
1165                         phdr->p_flags |= PF_R;
1166                 if (entry->protection & VM_PROT_WRITE)
1167                         phdr->p_flags |= PF_W;
1168                 if (entry->protection & VM_PROT_EXECUTE)
1169                         phdr->p_flags |= PF_X;
1170                 ++fpc->vnh;
1171                 ++fpc->count;
1172         }
1173         return (0);
1174 }
1175
1176 /*
1177  * For each writable segment in the process's memory map, call the given
1178  * function with a pointer to the map entry and some arbitrary
1179  * caller-supplied data.
1180  */
1181 static int
1182 each_segment(struct proc *p, segment_callback func, void *closure, int writable)
1183 {
1184         int error = 0;
1185         vm_map_t map = &p->p_vmspace->vm_map;
1186         vm_map_entry_t entry;
1187
1188         for (entry = map->header.next; error == 0 && entry != &map->header;
1189             entry = entry->next) {
1190                 vm_object_t obj;
1191                 vm_object_t lobj;
1192                 vm_object_t tobj;
1193
1194                 /*
1195                  * Don't dump inaccessible mappings, deal with legacy
1196                  * coredump mode.
1197                  *
1198                  * Note that read-only segments related to the elf binary
1199                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1200                  * need to arbitrarily ignore such segments.
1201                  */
1202                 if (elf_legacy_coredump) {
1203                         if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW)
1204                                 continue;
1205                 } else {
1206                         if (writable && (entry->protection & VM_PROT_ALL) == 0)
1207                                 continue;
1208                 }
1209
1210                 /*
1211                  * Dont include memory segment in the coredump if
1212                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1213                  * madvise(2).
1214                  *
1215                  * Currently we only dump normal VM object maps.  We do
1216                  * not dump submaps or virtual page tables.
1217                  */
1218                 if (writable && (entry->eflags & MAP_ENTRY_NOCOREDUMP))
1219                         continue;
1220                 if (entry->maptype != VM_MAPTYPE_NORMAL)
1221                         continue;
1222                 if ((obj = entry->object.vm_object) == NULL)
1223                         continue;
1224
1225                 /*
1226                  * Find the bottom-most object, leaving the base object
1227                  * and the bottom-most object held (but only one hold
1228                  * if they happen to be the same).
1229                  */
1230                 vm_object_hold(obj);
1231
1232                 lobj = obj;
1233                 while (lobj && (tobj = lobj->backing_object) != NULL) {
1234                         KKASSERT(tobj != obj);
1235                         vm_object_hold(tobj);
1236                         if (tobj == lobj->backing_object) {
1237                                 if (lobj != obj) {
1238                                         vm_object_lock_swap();
1239                                         vm_object_drop(lobj);
1240                                 }
1241                                 lobj = tobj;
1242                         } else {
1243                                 vm_object_drop(tobj);
1244                         }
1245                 }
1246
1247                 /*
1248                  * The callback only applies to default, swap, or vnode
1249                  * objects.  Other types of objects such as memory-mapped
1250                  * devices are ignored.
1251                  */
1252                 if (lobj->type == OBJT_DEFAULT || lobj->type == OBJT_SWAP ||
1253                     lobj->type == OBJT_VNODE) {
1254                         error = (*func)(entry, closure);
1255                 }
1256                 if (lobj != obj)
1257                         vm_object_drop(lobj);
1258                 vm_object_drop(obj);
1259         }
1260         return (error);
1261 }
1262
1263 static
1264 void *
1265 target_reserve(elf_buf_t target, size_t bytes, int *error)
1266 {
1267     void *res = NULL;
1268
1269     if (target->buf) {
1270             if (target->off + bytes > target->off_max)
1271                     *error = EINVAL;
1272             else
1273                     res = target->buf + target->off;
1274     }
1275     target->off += bytes;
1276     return (res);
1277 }
1278
1279 /*
1280  * Write the core file header to the file, including padding up to
1281  * the page boundary.
1282  */
1283 static int
1284 __elfN(corehdr)(struct lwp *lp, int sig, struct file *fp, struct ucred *cred,
1285             int numsegs, elf_buf_t target)
1286 {
1287         int error;
1288         ssize_t nbytes;
1289
1290         /*
1291          * Fill in the header.  The fp is passed so we can detect and flag
1292          * a checkpoint file pointer within the core file itself, because
1293          * it may not be restored from the same file handle.
1294          */
1295         error = __elfN(puthdr)(lp, target, sig, WRITE, numsegs, fp);
1296
1297         /* Write it to the core file. */
1298         if (error == 0) {
1299                 error = fp_write(fp, target->buf, target->off, &nbytes,
1300                                  UIO_SYSSPACE);
1301         }
1302         return (error);
1303 }
1304
1305 static int
1306 __elfN(puthdr)(struct lwp *lp, elf_buf_t target, int sig, enum putmode mode,
1307     int numsegs, struct file *fp)
1308 {
1309         struct proc *p = lp->lwp_proc;
1310         int error = 0;
1311         size_t phoff;
1312         size_t noteoff;
1313         size_t notesz;
1314         Elf_Ehdr *ehdr;
1315         Elf_Phdr *phdr;
1316
1317         ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error);
1318
1319         phoff = target->off;
1320         phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error);
1321
1322         noteoff = target->off;
1323         if (error == 0)
1324                 elf_putallnotes(lp, target, sig, mode);
1325         notesz = target->off - noteoff;
1326
1327         /*
1328          * put extra cruft for dumping process state here
1329          *  - we really want it be before all the program
1330          *    mappings
1331          *  - we just need to update the offset accordingly
1332          *    and GDB will be none the wiser.
1333          */
1334         if (error == 0)
1335                 error = elf_puttextvp(p, target);
1336         if (error == 0)
1337                 error = elf_putsigs(lp, target);
1338         if (error == 0)
1339                 error = elf_putfiles(p, target, fp);
1340
1341         /*
1342          * Align up to a page boundary for the program segments.  The
1343          * actual data will be written to the outptu file, not to elf_buf_t,
1344          * so we do not have to do any further bounds checking.
1345          */
1346         target->off = round_page(target->off);
1347         if (error == 0 && ehdr != NULL) {
1348                 /*
1349                  * Fill in the ELF header.
1350                  */
1351                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1352                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1353                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1354                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1355                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1356                 ehdr->e_ident[EI_DATA] = ELF_DATA;
1357                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1358                 ehdr->e_ident[EI_OSABI] = ELFOSABI_NONE;
1359                 ehdr->e_ident[EI_ABIVERSION] = 0;
1360                 ehdr->e_ident[EI_PAD] = 0;
1361                 ehdr->e_type = ET_CORE;
1362                 ehdr->e_machine = ELF_ARCH;
1363                 ehdr->e_version = EV_CURRENT;
1364                 ehdr->e_entry = 0;
1365                 ehdr->e_phoff = phoff;
1366                 ehdr->e_flags = 0;
1367                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1368                 ehdr->e_phentsize = sizeof(Elf_Phdr);
1369                 ehdr->e_phnum = numsegs + 1;
1370                 ehdr->e_shentsize = sizeof(Elf_Shdr);
1371                 ehdr->e_shnum = 0;
1372                 ehdr->e_shstrndx = SHN_UNDEF;
1373         }
1374         if (error == 0 && phdr != NULL) {
1375                 /*
1376                  * Fill in the program header entries.
1377                  */
1378                 struct phdr_closure phc;
1379
1380                 /* The note segement. */
1381                 phdr->p_type = PT_NOTE;
1382                 phdr->p_offset = noteoff;
1383                 phdr->p_vaddr = 0;
1384                 phdr->p_paddr = 0;
1385                 phdr->p_filesz = notesz;
1386                 phdr->p_memsz = 0;
1387                 phdr->p_flags = 0;
1388                 phdr->p_align = 0;
1389                 ++phdr;
1390
1391                 /* All the writable segments from the program. */
1392                 phc.phdr = phdr;
1393                 phc.phdr_max = phdr + numsegs;
1394                 phc.offset = target->off;
1395                 each_segment(p, cb_put_phdr, &phc, 1);
1396         }
1397         return (error);
1398 }
1399
1400 /*
1401  * Append core dump notes to target ELF buffer or simply update target size
1402  * if dryrun selected.
1403  */
1404 static int
1405 elf_putallnotes(struct lwp *corelp, elf_buf_t target, int sig,
1406     enum putmode mode)
1407 {
1408         struct proc *p = corelp->lwp_proc;
1409         int error;
1410         struct {
1411                 prstatus_t status;
1412                 prfpregset_t fpregs;
1413                 prpsinfo_t psinfo;
1414         } *tmpdata;
1415         prstatus_t *status;
1416         prfpregset_t *fpregs;
1417         prpsinfo_t *psinfo;
1418         struct lwp *lp;
1419
1420         /*
1421          * Allocate temporary storage for notes on heap to avoid stack overflow.
1422          */
1423         if (mode != DRYRUN) {
1424                 tmpdata = kmalloc(sizeof(*tmpdata), M_TEMP, M_ZERO | M_WAITOK);
1425                 status = &tmpdata->status;
1426                 fpregs = &tmpdata->fpregs;
1427                 psinfo = &tmpdata->psinfo;
1428         } else {
1429                 tmpdata = NULL;
1430                 status = NULL;
1431                 fpregs = NULL;
1432                 psinfo = NULL;
1433         }
1434
1435         /*
1436          * Append LWP-agnostic note.
1437          */
1438         if (mode != DRYRUN) {
1439                 psinfo->pr_version = PRPSINFO_VERSION;
1440                 psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1441                 strlcpy(psinfo->pr_fname, p->p_comm,
1442                         sizeof(psinfo->pr_fname));
1443                 /*
1444                  * XXX - We don't fill in the command line arguments
1445                  * properly yet.
1446                  */
1447                 strlcpy(psinfo->pr_psargs, p->p_comm,
1448                         sizeof(psinfo->pr_psargs));
1449         }
1450         error =
1451             __elfN(putnote)(target, "CORE", NT_PRPSINFO, psinfo, sizeof *psinfo);
1452         if (error)
1453                 goto exit;
1454
1455         /*
1456          * Append first note for LWP that triggered core so that it is
1457          * the selected one when the debugger starts.
1458          */
1459         if (mode != DRYRUN) {
1460                 status->pr_version = PRSTATUS_VERSION;
1461                 status->pr_statussz = sizeof(prstatus_t);
1462                 status->pr_gregsetsz = sizeof(gregset_t);
1463                 status->pr_fpregsetsz = sizeof(fpregset_t);
1464                 status->pr_osreldate = osreldate;
1465                 status->pr_cursig = sig;
1466                 /*
1467                  * XXX GDB needs unique pr_pid for each LWP and does not
1468                  * not support pr_pid==0 but lwp_tid can be 0, so hack unique
1469                  * value.
1470                  */
1471                 status->pr_pid = corelp->lwp_tid;
1472                 fill_regs(corelp, &status->pr_reg);
1473                 fill_fpregs(corelp, fpregs);
1474         }
1475         error =
1476             __elfN(putnote)(target, "CORE", NT_PRSTATUS, status, sizeof *status);
1477         if (error)
1478                 goto exit;
1479         error =
1480             __elfN(putnote)(target, "CORE", NT_FPREGSET, fpregs, sizeof *fpregs);
1481         if (error)
1482                 goto exit;
1483
1484         /*
1485          * Then append notes for other LWPs.
1486          */
1487         FOREACH_LWP_IN_PROC(lp, p) {
1488                 if (lp == corelp)
1489                         continue;
1490                 /* skip lwps being created */
1491                 if (lp->lwp_thread == NULL)
1492                         continue;
1493                 if (mode != DRYRUN) {
1494                         status->pr_pid = lp->lwp_tid;
1495                         fill_regs(lp, &status->pr_reg);
1496                         fill_fpregs(lp, fpregs);
1497                 }
1498                 error = __elfN(putnote)(target, "CORE", NT_PRSTATUS,
1499                                         status, sizeof *status);
1500                 if (error)
1501                         goto exit;
1502                 error = __elfN(putnote)(target, "CORE", NT_FPREGSET,
1503                                         fpregs, sizeof *fpregs);
1504                 if (error)
1505                         goto exit;
1506         }
1507
1508 exit:
1509         if (tmpdata != NULL)
1510                 kfree(tmpdata, M_TEMP);
1511         return (error);
1512 }
1513
1514 /*
1515  * Generate a note sub-structure.
1516  *
1517  * NOTE: 4-byte alignment.
1518  */
1519 static int
1520 __elfN(putnote)(elf_buf_t target, const char *name, int type,
1521             const void *desc, size_t descsz)
1522 {
1523         int error = 0;
1524         char *dst;
1525         Elf_Note note;
1526
1527         note.n_namesz = strlen(name) + 1;
1528         note.n_descsz = descsz;
1529         note.n_type = type;
1530         dst = target_reserve(target, sizeof(note), &error);
1531         if (dst != NULL)
1532                 bcopy(&note, dst, sizeof note);
1533         dst = target_reserve(target, note.n_namesz, &error);
1534         if (dst != NULL)
1535                 bcopy(name, dst, note.n_namesz);
1536         target->off = roundup2(target->off, sizeof(Elf_Word));
1537         dst = target_reserve(target, note.n_descsz, &error);
1538         if (dst != NULL)
1539                 bcopy(desc, dst, note.n_descsz);
1540         target->off = roundup2(target->off, sizeof(Elf_Word));
1541         return (error);
1542 }
1543
1544
1545 static int
1546 elf_putsigs(struct lwp *lp, elf_buf_t target)
1547 {
1548         /* XXX lwp handle more than one lwp */
1549         struct proc *p = lp->lwp_proc;
1550         int error = 0;
1551         struct ckpt_siginfo *csi;
1552
1553         csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error);
1554         if (csi) {
1555                 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo);
1556                 bcopy(p->p_sigacts, &csi->csi_sigacts, sizeof(*p->p_sigacts));
1557                 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval));
1558                 bcopy(&lp->lwp_sigmask, &csi->csi_sigmask,
1559                         sizeof(sigset_t));
1560                 csi->csi_sigparent = p->p_sigparent;
1561         }
1562         return (error);
1563 }
1564
1565 static int
1566 elf_putfiles(struct proc *p, elf_buf_t target, struct file *ckfp)
1567 {
1568         int error = 0;
1569         int i;
1570         struct ckpt_filehdr *cfh = NULL;
1571         struct ckpt_fileinfo *cfi;
1572         struct file *fp;
1573         struct vnode *vp;
1574         /*
1575          * the duplicated loop is gross, but it was the only way
1576          * to eliminate uninitialized variable warnings
1577          */
1578         cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error);
1579         if (cfh) {
1580                 cfh->cfh_nfiles = 0;
1581         }
1582
1583         /*
1584          * ignore STDIN/STDERR/STDOUT.
1585          */
1586         for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) {
1587                 fp = holdfp(p->p_fd, i, -1);
1588                 if (fp == NULL)
1589                         continue;
1590                 /*
1591                  * XXX Only checkpoint vnodes for now.
1592                  */
1593                 if (fp->f_type != DTYPE_VNODE) {
1594                         fdrop(fp);
1595                         continue;
1596                 }
1597                 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo),
1598                                         &error);
1599                 if (cfi == NULL) {
1600                         fdrop(fp);
1601                         continue;
1602                 }
1603                 cfi->cfi_index = -1;
1604                 cfi->cfi_type = fp->f_type;
1605                 cfi->cfi_flags = fp->f_flag;
1606                 cfi->cfi_offset = fp->f_offset;
1607                 cfi->cfi_ckflags = 0;
1608
1609                 if (fp == ckfp)
1610                         cfi->cfi_ckflags |= CKFIF_ISCKPTFD;
1611                 /* f_count and f_msgcount should not be saved/restored */
1612                 /* XXX save cred info */
1613
1614                 switch(fp->f_type) {
1615                 case DTYPE_VNODE:
1616                         vp = (struct vnode *)fp->f_data;
1617                         /*
1618                          * it looks like a bug in ptrace is marking
1619                          * a non-vnode as a vnode - until we find the
1620                          * root cause this will at least prevent
1621                          * further panics from truss
1622                          */
1623                         if (vp == NULL || vp->v_mount == NULL)
1624                                 break;
1625                         cfh->cfh_nfiles++;
1626                         cfi->cfi_index = i;
1627                         cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1628                         error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid);
1629                         break;
1630                 default:
1631                         break;
1632                 }
1633                 fdrop(fp);
1634         }
1635         return (error);
1636 }
1637
1638 static int
1639 elf_puttextvp(struct proc *p, elf_buf_t target)
1640 {
1641         int error = 0;
1642         int *vn_count;
1643         struct fp_closure fpc;
1644         struct ckpt_vminfo *vminfo;
1645
1646         vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error);
1647         if (vminfo != NULL) {
1648                 vminfo->cvm_dsize = p->p_vmspace->vm_dsize;
1649                 vminfo->cvm_tsize = p->p_vmspace->vm_tsize;
1650                 vminfo->cvm_daddr = p->p_vmspace->vm_daddr;
1651                 vminfo->cvm_taddr = p->p_vmspace->vm_taddr;
1652         }
1653
1654         fpc.count = 0;
1655         vn_count = target_reserve(target, sizeof(int), &error);
1656         if (target->buf != NULL) {
1657                 fpc.vnh = (struct vn_hdr *)(target->buf + target->off);
1658                 fpc.vnh_max = fpc.vnh +
1659                         (target->off_max - target->off) / sizeof(struct vn_hdr);
1660                 error = each_segment(p, cb_put_fp, &fpc, 0);
1661                 if (vn_count)
1662                         *vn_count = fpc.count;
1663         } else {
1664                 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0);
1665         }
1666         target->off += fpc.count * sizeof(struct vn_hdr);
1667         return (error);
1668 }
1669
1670 /*
1671  * Try to find the appropriate ABI-note section for checknote,
1672  * The entire image is searched if necessary, not only the first page.
1673  */
1674 static boolean_t
1675 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1676     int32_t *osrel)
1677 {
1678         boolean_t valid_note_found;
1679         const Elf_Phdr *phdr, *pnote;
1680         const Elf_Ehdr *hdr;
1681         int i;
1682
1683         valid_note_found = FALSE;
1684         hdr = (const Elf_Ehdr *)imgp->image_header;
1685         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1686
1687         for (i = 0; i < hdr->e_phnum; i++) {
1688                 if (phdr[i].p_type == PT_NOTE) {
1689                         pnote = &phdr[i];
1690                         valid_note_found = check_PT_NOTE (imgp, checknote,
1691                                 osrel, pnote);
1692                         if (valid_note_found)
1693                                 break;
1694                 }
1695         }
1696         return valid_note_found;
1697 }
1698
1699 static boolean_t
1700 check_PT_NOTE(struct image_params *imgp, Elf_Brandnote *checknote,
1701     int32_t *osrel, const Elf_Phdr * pnote)
1702 {
1703         boolean_t limited_to_first_page;
1704         boolean_t found = FALSE;
1705         const Elf_Note *note, *note0, *note_end;
1706         const char *note_name;
1707         __ElfN(Off) noteloc, firstloc;
1708         __ElfN(Size) notesz, firstlen, endbyte;
1709         struct lwbuf *lwb;
1710         struct lwbuf lwb_cache;
1711         const char *page;
1712         char *data = NULL;
1713         int n;
1714
1715         notesz = pnote->p_filesz;
1716         noteloc = pnote->p_offset;
1717         endbyte = noteloc + notesz;
1718         limited_to_first_page = noteloc < PAGE_SIZE && endbyte < PAGE_SIZE;
1719
1720         if (limited_to_first_page) {
1721                 note = (const Elf_Note *)(imgp->image_header + noteloc);
1722                 note_end = (const Elf_Note *)(imgp->image_header + endbyte);
1723                 note0 = note;
1724         } else {
1725                 firstloc = noteloc & PAGE_MASK;
1726                 firstlen = PAGE_SIZE - firstloc;
1727                 if (notesz < sizeof(Elf_Note) || notesz > PAGE_SIZE)
1728                         return (FALSE);
1729
1730                 lwb = &lwb_cache;
1731                 if (exec_map_page(imgp, noteloc >> PAGE_SHIFT, &lwb, &page))
1732                         return (FALSE);
1733                 if (firstlen < notesz) {         /* crosses page boundary */
1734                         data = kmalloc(notesz, M_TEMP, M_WAITOK);
1735                         bcopy(page + firstloc, data, firstlen);
1736
1737                         exec_unmap_page(lwb);
1738                         lwb = &lwb_cache;
1739                         if (exec_map_page(imgp, (noteloc >> PAGE_SHIFT) + 1,
1740                                 &lwb, &page)) {
1741                                 kfree(data, M_TEMP);
1742                                 return (FALSE);
1743                         }
1744                         bcopy(page, data + firstlen, notesz - firstlen);
1745                         note = note0 = (const Elf_Note *)(data);
1746                         note_end = (const Elf_Note *)(data + notesz);
1747                 } else {
1748                         note = note0 = (const Elf_Note *)(page + firstloc);
1749                         note_end = (const Elf_Note *)(page + firstloc +
1750                                 firstlen);
1751                 }
1752         }
1753
1754         for (n = 0; n < 100 && note >= note0 && note < note_end; n++) {
1755                 if (!aligned(note, Elf32_Addr))
1756                         break;
1757                 note_name = (const char *)(note + 1);
1758
1759                 if (note->n_namesz == checknote->hdr.n_namesz
1760                     && note->n_descsz == checknote->hdr.n_descsz
1761                     && note->n_type == checknote->hdr.n_type
1762                     && (strncmp(checknote->vendor, note_name,
1763                         checknote->hdr.n_namesz) == 0)) {
1764                         /* Fetch osreldata from ABI.note-tag */
1765                         if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
1766                             checknote->trans_osrel != NULL)
1767                                 checknote->trans_osrel(note, osrel);
1768                         found = TRUE;
1769                         break;
1770                 }
1771                 note = (const Elf_Note *)((const char *)(note + 1) +
1772                     roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1773                     roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1774         }
1775
1776         if (!limited_to_first_page) {
1777                 if (data != NULL)
1778                         kfree(data, M_TEMP);
1779                 exec_unmap_page(lwb);
1780         }
1781         return (found);
1782 }
1783
1784 /*
1785  * The interpreter program header may be located beyond the first page, so
1786  * regardless of its location, a copy of the interpreter path is created so
1787  * that it may be safely referenced by the calling function in all case.  The
1788  * memory is allocated by calling function, and the copying is done here.
1789  */
1790 static boolean_t
1791 extract_interpreter(struct image_params *imgp, const Elf_Phdr *pinterpreter,
1792     char *data)
1793 {
1794         boolean_t limited_to_first_page;
1795         const boolean_t result_success = FALSE;
1796         const boolean_t result_failure = TRUE;
1797         __ElfN(Off) pathloc, firstloc;
1798         __ElfN(Size) pathsz, firstlen, endbyte;
1799         struct lwbuf *lwb;
1800         struct lwbuf lwb_cache;
1801         const char *page;
1802
1803         pathsz  = pinterpreter->p_filesz;
1804         pathloc = pinterpreter->p_offset;
1805         endbyte = pathloc + pathsz;
1806
1807         limited_to_first_page = pathloc < PAGE_SIZE && endbyte < PAGE_SIZE;
1808         if (limited_to_first_page) {
1809                 bcopy(imgp->image_header + pathloc, data, pathsz);
1810                 return (result_success);
1811         }
1812
1813         firstloc = pathloc & PAGE_MASK;
1814         firstlen = PAGE_SIZE - firstloc;
1815
1816         lwb = &lwb_cache;
1817         if (exec_map_page(imgp, pathloc >> PAGE_SHIFT, &lwb, &page))
1818                 return (result_failure);
1819
1820         if (firstlen < pathsz) {         /* crosses page boundary */
1821                 bcopy(page + firstloc, data, firstlen);
1822
1823                 exec_unmap_page(lwb);
1824                 lwb = &lwb_cache;
1825                 if (exec_map_page(imgp, (pathloc >> PAGE_SHIFT) + 1, &lwb,
1826                         &page))
1827                         return (result_failure);
1828                 bcopy(page, data + firstlen, pathsz - firstlen);
1829         } else
1830                 bcopy(page + firstloc, data, pathsz);
1831
1832         exec_unmap_page(lwb);
1833         return (result_success);
1834 }
1835
1836 static boolean_t
1837 __elfN(bsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
1838 {
1839         uintptr_t p;
1840
1841         p = (uintptr_t)(note + 1);
1842         p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
1843         *osrel = *(const int32_t *)(p);
1844
1845         return (TRUE);
1846 }
1847
1848 /*
1849  * Tell kern_execve.c about it, with a little help from the linker.
1850  */
1851 #if defined(__x86_64__)
1852 static struct execsw elf_execsw = {exec_elf64_imgact, "ELF64"};
1853 EXEC_SET_ORDERED(elf64, elf_execsw, SI_ORDER_FIRST);
1854 #else /* i386 assumed */
1855 static struct execsw elf_execsw = {exec_elf32_imgact, "ELF32"};
1856 EXEC_SET_ORDERED(elf32, elf_execsw, SI_ORDER_FIRST);
1857 #endif