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
  31 #include <sys/cdefs.h>
  32 __FBSDID("$FreeBSD$");
  33
  34 #include "opt_compat.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/exec.h>
  38 #include <sys/fcntl.h>
  39 #include <sys/imgact.h>
  40 #include <sys/imgact_elf.h>
  41 #include <sys/kernel.h>
  42 #include <sys/lock.h>
  43 #include <sys/malloc.h>
  44 #include <sys/mount.h>
  45 #include <sys/mutex.h>
  46 #include <sys/mman.h>
  47 #include <sys/namei.h>
  48 #include <sys/pioctl.h>
  49 #include <sys/proc.h>
  50 #include <sys/procfs.h>
  51 #include <sys/resourcevar.h>
  52 #include <sys/systm.h>
  53 #include <sys/signalvar.h>
  54 #include <sys/stat.h>
  55 #include <sys/sx.h>
  56 #include <sys/syscall.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/sysent.h>
  59 #include <sys/vnode.h>
  60
  61 #include <vm/vm.h>
  62 #include <vm/vm_kern.h>
  63 #include <vm/vm_param.h>
  64 #include <vm/pmap.h>
  65 #include <vm/vm_map.h>
  66 #include <vm/vm_object.h>
  67 #include <vm/vm_extern.h>
  68
  69 #include <machine/elf.h>
  70 #include <machine/md_var.h>
  71
  72 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
  73 #include <machine/fpu.h>
  74 #include <compat/ia32/ia32_reg.h>
  75 #endif
  76
  77 #define OLD_EI_BRAND    8
  78
  79 static int __elfN(check_header)(const Elf_Ehdr *hdr);
  80 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr,
  81     const char *interp);
  82 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
  83     u_long *entry, size_t pagesize);
  84 static int __elfN(load_section)(struct proc *p,
  85     struct vmspace *vmspace, struct vnode *vp, vm_object_t object,
  86     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
  87     vm_prot_t prot, size_t pagesize);
  88 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
  89
  90 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
  91     "");
  92
  93 int __elfN(fallback_brand) = -1;
  94 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
  95     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
  96     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
  97 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
  98     &__elfN(fallback_brand));
  99
 100 static int elf_trace = 0;
 101 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
 102
 103 static int elf_legacy_coredump = 0;
 104 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
 105     &elf_legacy_coredump, 0, "");
 106
 107 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
 108
 109 int
 110 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
 111 {
 112         int i;
 113
 114         for (i = 0; i < MAX_BRANDS; i++) {
 115                 if (elf_brand_list[i] == NULL) {
 116                         elf_brand_list[i] = entry;
 117                         break;
 118                 }
 119         }
 120         if (i == MAX_BRANDS)
 121                 return (-1);
 122         return (0);
 123 }
 124
 125 int
 126 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
 127 {
 128         int i;
 129
 130         for (i = 0; i < MAX_BRANDS; i++) {
 131                 if (elf_brand_list[i] == entry) {
 132                         elf_brand_list[i] = NULL;
 133                         break;
 134                 }
 135         }
 136         if (i == MAX_BRANDS)
 137                 return (-1);
 138         return (0);
 139 }
 140
 141 int
 142 __elfN(brand_inuse)(Elf_Brandinfo *entry)
 143 {
 144         struct proc *p;
 145         int rval = FALSE;
 146
 147         sx_slock(&allproc_lock);
 148         LIST_FOREACH(p, &allproc, p_list) {
 149                 if (p->p_sysent == entry->sysvec) {
 150                         rval = TRUE;
 151                         break;
 152                 }
 153         }
 154         sx_sunlock(&allproc_lock);
 155
 156         return (rval);
 157 }
 158
 159 static Elf_Brandinfo *
 160 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
 161 {
 162         Elf_Brandinfo *bi;
 163         int i;
 164
 165         /*
 166          * We support three types of branding -- (1) the ELF EI_OSABI field
 167          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
 168          * branding w/in the ELF header, and (3) path of the `interp_path'
 169          * field.  We should also look for an ".note.ABI-tag" ELF section now
 170          * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
 171          */
 172
 173         /* If the executable has a brand, search for it in the brand list. */
 174         for (i = 0; i < MAX_BRANDS; i++) {
 175                 bi = elf_brand_list[i];
 176                 if (bi != NULL && hdr->e_machine == bi->machine &&
 177                     (hdr->e_ident[EI_OSABI] == bi->brand ||
 178                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
 179                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
 180                         return (bi);
 181         }
 182
 183         /* Lacking a known brand, search for a recognized interpreter. */
 184         if (interp != NULL) {
 185                 for (i = 0; i < MAX_BRANDS; i++) {
 186                         bi = elf_brand_list[i];
 187                         if (bi != NULL && hdr->e_machine == bi->machine &&
 188                             strcmp(interp, bi->interp_path) == 0)
 189                                 return (bi);
 190                 }
 191         }
 192
 193         /* Lacking a recognized interpreter, try the default brand */
 194         for (i = 0; i < MAX_BRANDS; i++) {
 195                 bi = elf_brand_list[i];
 196                 if (bi != NULL && hdr->e_machine == bi->machine &&
 197                     __elfN(fallback_brand) == bi->brand)
 198                         return (bi);
 199         }
 200         return (NULL);
 201 }
 202
 203 static int
 204 __elfN(check_header)(const Elf_Ehdr *hdr)
 205 {
 206         Elf_Brandinfo *bi;
 207         int i;
 208
 209         if (!IS_ELF(*hdr) ||
 210             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
 211             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
 212             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
 213             hdr->e_phentsize != sizeof(Elf_Phdr) ||
 214             hdr->e_version != ELF_TARG_VER)
 215                 return (ENOEXEC);
 216
 217         /*
 218          * Make sure we have at least one brand for this machine.
 219          */
 220
 221         for (i = 0; i < MAX_BRANDS; i++) {
 222                 bi = elf_brand_list[i];
 223                 if (bi != NULL && bi->machine == hdr->e_machine)
 224                         break;
 225         }
 226         if (i == MAX_BRANDS)
 227                 return (ENOEXEC);
 228
 229         return (0);
 230 }
 231
 232 static int
 233 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
 234         vm_offset_t start, vm_offset_t end, vm_prot_t prot,
 235         vm_prot_t max)
 236 {
 237         int error, rv;
 238         vm_offset_t off;
 239         vm_offset_t data_buf = 0;
 240
 241         /*
 242          * Create the page if it doesn't exist yet. Ignore errors.
 243          */
 244         vm_map_lock(map);
 245         vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), max,
 246             max, 0);
 247         vm_map_unlock(map);
 248
 249         /*
 250          * Find the page from the underlying object.
 251          */
 252         if (object) {
 253                 vm_object_reference(object);
 254                 rv = vm_map_find(exec_map,
 255                                  object,
 256                                  trunc_page(offset),
 257                                  &data_buf,
 258                                  PAGE_SIZE,
 259                                  TRUE,
 260                                  VM_PROT_READ,
 261                                  VM_PROT_ALL,
 262                                  MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
 263                 if (rv != KERN_SUCCESS) {
 264                         vm_object_deallocate(object);
 265                         return (rv);
 266                 }
 267
 268                 off = offset - trunc_page(offset);
 269                 error = copyout((caddr_t)data_buf + off, (caddr_t)start,
 270                     end - start);
 271                 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
 272                 if (error) {
 273                         return (KERN_FAILURE);
 274                 }
 275         }
 276
 277         return (KERN_SUCCESS);
 278 }
 279
 280 static int
 281 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
 282         vm_offset_t start, vm_offset_t end, vm_prot_t prot,
 283         vm_prot_t max, int cow)
 284 {
 285         vm_offset_t data_buf, off;
 286         vm_size_t sz;
 287         int error, rv;
 288
 289         if (start != trunc_page(start)) {
 290                 rv = __elfN(map_partial)(map, object, offset, start,
 291                     round_page(start), prot, max);
 292                 if (rv)
 293                         return (rv);
 294                 offset += round_page(start) - start;
 295                 start = round_page(start);
 296         }
 297         if (end != round_page(end)) {
 298                 rv = __elfN(map_partial)(map, object, offset +
 299                     trunc_page(end) - start, trunc_page(end), end, prot, max);
 300                 if (rv)
 301                         return (rv);
 302                 end = trunc_page(end);
 303         }
 304         if (end > start) {
 305                 if (offset & PAGE_MASK) {
 306                         /*
 307                          * The mapping is not page aligned. This means we have
 308                          * to copy the data. Sigh.
 309                          */
 310                         rv = vm_map_find(map, 0, 0, &start, end - start,
 311                             FALSE, prot, max, 0);
 312                         if (rv)
 313                                 return (rv);
 314                         data_buf = 0;
 315                         while (start < end) {
 316                                 vm_object_reference(object);
 317                                 rv = vm_map_find(exec_map,
 318                                                  object,
 319                                                  trunc_page(offset),
 320                                                  &data_buf,
 321                                                  2 * PAGE_SIZE,
 322                                                  TRUE,
 323                                                  VM_PROT_READ,
 324                                                  VM_PROT_ALL,
 325                                                  (MAP_COPY_ON_WRITE
 326                                                   | MAP_PREFAULT_PARTIAL));
 327                                 if (rv != KERN_SUCCESS) {
 328                                         vm_object_deallocate(object);
 329                                         return (rv);
 330                                 }
 331                                 off = offset - trunc_page(offset);
 332                                 sz = end - start;
 333                                 if (sz > PAGE_SIZE)
 334                                         sz = PAGE_SIZE;
 335                                 error = copyout((caddr_t)data_buf + off,
 336                                     (caddr_t)start, sz);
 337                                 vm_map_remove(exec_map, data_buf,
 338                                     data_buf + 2 * PAGE_SIZE);
 339                                 if (error) {
 340                                         return (KERN_FAILURE);
 341                                 }
 342                                 start += sz;
 343                         }
 344                         rv = KERN_SUCCESS;
 345                 } else {
 346                         vm_map_lock(map);
 347                         rv = vm_map_insert(map, object, offset, start, end,
 348                             prot, max, cow);
 349                         vm_map_unlock(map);
 350                 }
 351                 return (rv);
 352         } else {
 353                 return (KERN_SUCCESS);
 354         }
 355 }
 356
 357 static int
 358 __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
 359         struct vnode *vp, vm_object_t object, vm_offset_t offset,
 360         caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
 361         size_t pagesize)
 362 {
 363         size_t map_len;
 364         vm_offset_t map_addr;
 365         int error, rv, cow;
 366         size_t copy_len;
 367         vm_offset_t file_addr;
 368         vm_offset_t data_buf = 0;
 369
 370         error = 0;
 371
 372         /*
 373          * It's necessary to fail if the filsz + offset taken from the
 374          * header is greater than the actual file pager object's size.
 375          * If we were to allow this, then the vm_map_find() below would
 376          * walk right off the end of the file object and into the ether.
 377          *
 378          * While I'm here, might as well check for something else that
 379          * is invalid: filsz cannot be greater than memsz.
 380          */
 381         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
 382             filsz > memsz) {
 383                 uprintf("elf_load_section: truncated ELF file\n");
 384                 return (ENOEXEC);
 385         }
 386
 387 #define trunc_page_ps(va, ps)   ((va) & ~(ps - 1))
 388 #define round_page_ps(va, ps)   (((va) + (ps - 1)) & ~(ps - 1))
 389
 390         map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
 391         file_addr = trunc_page_ps(offset, pagesize);
 392
 393         /*
 394          * We have two choices.  We can either clear the data in the last page
 395          * of an oversized mapping, or we can start the anon mapping a page
 396          * early and copy the initialized data into that first page.  We
 397          * choose the second..
 398          */
 399         if (memsz > filsz)
 400                 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
 401         else
 402                 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
 403
 404         if (map_len != 0) {
 405                 vm_object_reference(object);
 406
 407                 /* cow flags: don't dump readonly sections in core */
 408                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
 409                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
 410
 411                 rv = __elfN(map_insert)(&vmspace->vm_map,
 412                                       object,
 413                                       file_addr,        /* file offset */
 414                                       map_addr,         /* virtual start */
 415                                       map_addr + map_len,/* virtual end */
 416                                       prot,
 417                                       VM_PROT_ALL,
 418                                       cow);
 419                 if (rv != KERN_SUCCESS) {
 420                         vm_object_deallocate(object);
 421                         return (EINVAL);
 422                 }
 423
 424                 /* we can stop now if we've covered it all */
 425                 if (memsz == filsz) {
 426                         return (0);
 427                 }
 428         }
 429
 430
 431         /*
 432          * We have to get the remaining bit of the file into the first part
 433          * of the oversized map segment.  This is normally because the .data
 434          * segment in the file is extended to provide bss.  It's a neat idea
 435          * to try and save a page, but it's a pain in the behind to implement.
 436          */
 437         copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
 438         map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
 439         map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
 440             map_addr;
 441
 442         /* This had damn well better be true! */
 443         if (map_len != 0) {
 444                 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
 445                     map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0);
 446                 if (rv != KERN_SUCCESS) {
 447                         return (EINVAL);
 448                 }
 449         }
 450
 451         if (copy_len != 0) {
 452                 vm_offset_t off;
 453                 vm_object_reference(object);
 454                 rv = vm_map_find(exec_map,
 455                                  object,
 456                                  trunc_page(offset + filsz),
 457                                  &data_buf,
 458                                  PAGE_SIZE,
 459                                  TRUE,
 460                                  VM_PROT_READ,
 461                                  VM_PROT_ALL,
 462                                  MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
 463                 if (rv != KERN_SUCCESS) {
 464                         vm_object_deallocate(object);
 465                         return (EINVAL);
 466                 }
 467
 468                 /* send the page fragment to user space */
 469                 off = trunc_page_ps(offset + filsz, pagesize) -
 470                     trunc_page(offset + filsz);
 471                 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr,
 472                     copy_len);
 473                 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
 474                 if (error) {
 475                         return (error);
 476                 }
 477         }
 478
 479         /*
 480          * set it to the specified protection.
 481          * XXX had better undo the damage from pasting over the cracks here!
 482          */
 483         vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
 484             round_page(map_addr + map_len),  prot, FALSE);
 485
 486         return (error);
 487 }
 488
 489 /*
 490  * Load the file "file" into memory.  It may be either a shared object
 491  * or an executable.
 492  *
 493  * The "addr" reference parameter is in/out.  On entry, it specifies
 494  * the address where a shared object should be loaded.  If the file is
 495  * an executable, this value is ignored.  On exit, "addr" specifies
 496  * where the file was actually loaded.
 497  *
 498  * The "entry" reference parameter is out only.  On exit, it specifies
 499  * the entry point for the loaded file.
 500  */
 501 static int
 502 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
 503         u_long *entry, size_t pagesize)
 504 {
 505         struct {
 506                 struct nameidata nd;
 507                 struct vattr attr;
 508                 struct image_params image_params;
 509         } *tempdata;
 510         const Elf_Ehdr *hdr = NULL;
 511         const Elf_Phdr *phdr = NULL;
 512         struct nameidata *nd;
 513         struct vmspace *vmspace = p->p_vmspace;
 514         struct vattr *attr;
 515         struct image_params *imgp;
 516         vm_prot_t prot;
 517         u_long rbase;
 518         u_long base_addr = 0;
 519         int vfslocked, error, i, numsegs;
 520
 521         if (curthread->td_proc != p)
 522                 panic("elf_load_file - thread");        /* XXXKSE DIAGNOSTIC */
 523
 524         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
 525         nd = &tempdata->nd;
 526         attr = &tempdata->attr;
 527         imgp = &tempdata->image_params;
 528
 529         /*
 530          * Initialize part of the common data
 531          */
 532         imgp->proc = p;
 533         imgp->attr = attr;
 534         imgp->firstpage = NULL;
 535         imgp->image_header = NULL;
 536         imgp->object = NULL;
 537         imgp->execlabel = NULL;
 538
 539         /* XXXKSE */
 540         NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
 541             curthread);
 542         vfslocked = 0;
 543         if ((error = namei(nd)) != 0) {
 544                 nd->ni_vp = NULL;
 545                 goto fail;
 546         }
 547         vfslocked = NDHASGIANT(nd);
 548         NDFREE(nd, NDF_ONLY_PNBUF);
 549         imgp->vp = nd->ni_vp;
 550
 551         /*
 552          * Check permissions, modes, uid, etc on the file, and "open" it.
 553          */
 554         error = exec_check_permissions(imgp);
 555         if (error) {
 556                 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
 557                 goto fail;
 558         }
 559
 560         error = exec_map_first_page(imgp);
 561         /*
 562          * Also make certain that the interpreter stays the same, so set
 563          * its VV_TEXT flag, too.
 564          */
 565         if (error == 0)
 566                 nd->ni_vp->v_vflag |= VV_TEXT;
 567
 568         imgp->object = nd->ni_vp->v_object;
 569         vm_object_reference(imgp->object);
 570
 571         VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
 572         if (error)
 573                 goto fail;
 574
 575         hdr = (const Elf_Ehdr *)imgp->image_header;
 576         if ((error = __elfN(check_header)(hdr)) != 0)
 577                 goto fail;
 578         if (hdr->e_type == ET_DYN)
 579                 rbase = *addr;
 580         else if (hdr->e_type == ET_EXEC)
 581                 rbase = 0;
 582         else {
 583                 error = ENOEXEC;
 584                 goto fail;
 585         }
 586
 587         /* Only support headers that fit within first page for now      */
 588         /*    (multiplication of two Elf_Half fields will not overflow) */
 589         if ((hdr->e_phoff > PAGE_SIZE) ||
 590             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
 591                 error = ENOEXEC;
 592                 goto fail;
 593         }
 594
 595         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 596
 597         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
 598                 if (phdr[i].p_type == PT_LOAD) {        /* Loadable segment */
 599                         prot = 0;
 600                         if (phdr[i].p_flags & PF_X)
 601                                 prot |= VM_PROT_EXECUTE;
 602                         if (phdr[i].p_flags & PF_W)
 603                                 prot |= VM_PROT_WRITE;
 604                         if (phdr[i].p_flags & PF_R)
 605                                 prot |= VM_PROT_READ;
 606
 607                         if ((error = __elfN(load_section)(p, vmspace,
 608                             nd->ni_vp, imgp->object, phdr[i].p_offset,
 609                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
 610                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
 611                             pagesize)) != 0)
 612                                 goto fail;
 613                         /*
 614                          * Establish the base address if this is the
 615                          * first segment.
 616                          */
 617                         if (numsegs == 0)
 618                                 base_addr = trunc_page(phdr[i].p_vaddr +
 619                                     rbase);
 620                         numsegs++;
 621                 }
 622         }
 623         *addr = base_addr;
 624         *entry = (unsigned long)hdr->e_entry + rbase;
 625
 626 fail:
 627         if (imgp->firstpage)
 628                 exec_unmap_first_page(imgp);
 629         if (imgp->object)
 630                 vm_object_deallocate(imgp->object);
 631
 632         if (nd->ni_vp)
 633                 vrele(nd->ni_vp);
 634
 635         VFS_UNLOCK_GIANT(vfslocked);
 636         free(tempdata, M_TEMP);
 637
 638         return (error);
 639 }
 640
 641 static int
 642 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
 643 {
 644         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
 645         const Elf_Phdr *phdr;
 646         Elf_Auxargs *elf_auxargs = NULL;
 647         struct vmspace *vmspace;
 648         vm_prot_t prot;
 649         u_long text_size = 0, data_size = 0, total_size = 0;
 650         u_long text_addr = 0, data_addr = 0;
 651         u_long seg_size, seg_addr;
 652         u_long addr, entry = 0, proghdr = 0;
 653         int error = 0, i;
 654         const char *interp = NULL;
 655         Elf_Brandinfo *brand_info;
 656         char *path;
 657         struct thread *td = curthread;
 658         struct sysentvec *sv;
 659
 660         /*
 661          * Do we have a valid ELF header ?
 662          */
 663         if (__elfN(check_header)(hdr) != 0 || hdr->e_type != ET_EXEC)
 664                 return (-1);
 665
 666         /*
 667          * From here on down, we return an errno, not -1, as we've
 668          * detected an ELF file.
 669          */
 670
 671         if ((hdr->e_phoff > PAGE_SIZE) ||
 672             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
 673                 /* Only support headers in first page for now */
 674                 return (ENOEXEC);
 675         }
 676         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 677
 678         /*
 679          * From this point on, we may have resources that need to be freed.
 680          */
 681
 682         VOP_UNLOCK(imgp->vp, 0, td);
 683
 684         for (i = 0; i < hdr->e_phnum; i++) {
 685                 switch (phdr[i].p_type) {
 686                 case PT_INTERP: /* Path to interpreter */
 687                         if (phdr[i].p_filesz > MAXPATHLEN ||
 688                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
 689                                 error = ENOEXEC;
 690                                 goto fail;
 691                         }
 692                         interp = imgp->image_header + phdr[i].p_offset;
 693                         break;
 694                 default:
 695                         break;
 696                 }
 697         }
 698
 699         brand_info = __elfN(get_brandinfo)(hdr, interp);
 700         if (brand_info == NULL) {
 701                 uprintf("ELF binary type \"%u\" not known.\n",
 702                     hdr->e_ident[EI_OSABI]);
 703                 error = ENOEXEC;
 704                 goto fail;
 705         }
 706         sv = brand_info->sysvec;
 707         if (interp != NULL && brand_info->interp_newpath != NULL)
 708                 interp = brand_info->interp_newpath;
 709
 710         exec_new_vmspace(imgp, sv);
 711
 712         vmspace = imgp->proc->p_vmspace;
 713
 714         for (i = 0; i < hdr->e_phnum; i++) {
 715                 switch (phdr[i].p_type) {
 716                 case PT_LOAD:   /* Loadable segment */
 717                         prot = 0;
 718                         if (phdr[i].p_flags & PF_X)
 719                                 prot |= VM_PROT_EXECUTE;
 720                         if (phdr[i].p_flags & PF_W)
 721                                 prot |= VM_PROT_WRITE;
 722                         if (phdr[i].p_flags & PF_R)
 723                                 prot |= VM_PROT_READ;
 724
 725 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
 726                         /*
 727                          * Some x86 binaries assume read == executable,
 728                          * notably the M3 runtime and therefore cvsup
 729                          */
 730                         if (prot & VM_PROT_READ)
 731                                 prot |= VM_PROT_EXECUTE;
 732 #endif
 733
 734                         if ((error = __elfN(load_section)(imgp->proc, vmspace,
 735                             imgp->vp, imgp->object, phdr[i].p_offset,
 736                             (caddr_t)(uintptr_t)phdr[i].p_vaddr,
 737                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
 738                             sv->sv_pagesize)) != 0)
 739                                 goto fail;
 740
 741                         /*
 742                          * If this segment contains the program headers,
 743                          * remember their virtual address for the AT_PHDR
 744                          * aux entry. Static binaries don't usually include
 745                          * a PT_PHDR entry.
 746                          */
 747                         if (phdr[i].p_offset == 0 &&
 748                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
 749                                 <= phdr[i].p_filesz)
 750                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
 751
 752                         seg_addr = trunc_page(phdr[i].p_vaddr);
 753                         seg_size = round_page(phdr[i].p_memsz +
 754                             phdr[i].p_vaddr - seg_addr);
 755
 756                         /*
 757                          * Is this .text or .data?  We can't use
 758                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
 759                          * alpha terribly and possibly does other bad
 760                          * things so we stick to the old way of figuring
 761                          * it out:  If the segment contains the program
 762                          * entry point, it's a text segment, otherwise it
 763                          * is a data segment.
 764                          *
 765                          * Note that obreak() assumes that data_addr +
 766                          * data_size == end of data load area, and the ELF
 767                          * file format expects segments to be sorted by
 768                          * address.  If multiple data segments exist, the
 769                          * last one will be used.
 770                          */
 771                         if (hdr->e_entry >= phdr[i].p_vaddr &&
 772                             hdr->e_entry < (phdr[i].p_vaddr +
 773                             phdr[i].p_memsz)) {
 774                                 text_size = seg_size;
 775                                 text_addr = seg_addr;
 776                                 entry = (u_long)hdr->e_entry;
 777                         } else {
 778                                 data_size = seg_size;
 779                                 data_addr = seg_addr;
 780                         }
 781                         total_size += seg_size;
 782                         break;
 783                 case PT_PHDR:   /* Program header table info */
 784                         proghdr = phdr[i].p_vaddr;
 785                         break;
 786                 default:
 787                         break;
 788                 }
 789         }
 790
 791         if (data_addr == 0 && data_size == 0) {
 792                 data_addr = text_addr;
 793                 data_size = text_size;
 794         }
 795
 796         /*
 797          * Check limits.  It should be safe to check the
 798          * limits after loading the segments since we do
 799          * not actually fault in all the segments pages.
 800          */
 801         PROC_LOCK(imgp->proc);
 802         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
 803             text_size > maxtsiz ||
 804             total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
 805                 PROC_UNLOCK(imgp->proc);
 806                 error = ENOMEM;
 807                 goto fail;
 808         }
 809
 810         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
 811         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
 812         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
 813         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
 814
 815         /*
 816          * We load the dynamic linker where a userland call
 817          * to mmap(0, ...) would put it.  The rationale behind this
 818          * calculation is that it leaves room for the heap to grow to
 819          * its maximum allowed size.
 820          */
 821         addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
 822             lim_max(imgp->proc, RLIMIT_DATA));
 823         PROC_UNLOCK(imgp->proc);
 824
 825         imgp->entry_addr = entry;
 826
 827         imgp->proc->p_sysent = sv;
 828         if (interp != NULL && brand_info->emul_path != NULL &&
 829             brand_info->emul_path[0] != '\0') {
 830                 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 831                 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path,
 832                     interp);
 833                 error = __elfN(load_file)(imgp->proc, path, &addr,
 834                     &imgp->entry_addr, sv->sv_pagesize);
 835                 free(path, M_TEMP);
 836                 if (error == 0)
 837                         interp = NULL;
 838         }
 839         if (interp != NULL) {
 840                 error = __elfN(load_file)(imgp->proc, interp, &addr,
 841                     &imgp->entry_addr, sv->sv_pagesize);
 842                 if (error != 0) {
 843                         uprintf("ELF interpreter %s not found\n", interp);
 844                         goto fail;
 845                 }
 846         }
 847
 848         /*
 849          * Construct auxargs table (used by the fixup routine)
 850          */
 851         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
 852         elf_auxargs->execfd = -1;
 853         elf_auxargs->phdr = proghdr;
 854         elf_auxargs->phent = hdr->e_phentsize;
 855         elf_auxargs->phnum = hdr->e_phnum;
 856         elf_auxargs->pagesz = PAGE_SIZE;
 857         elf_auxargs->base = addr;
 858         elf_auxargs->flags = 0;
 859         elf_auxargs->entry = entry;
 860         elf_auxargs->trace = elf_trace;
 861
 862         imgp->auxargs = elf_auxargs;
 863         imgp->interpreted = 0;
 864
 865 fail:
 866         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
 867         return (error);
 868 }
 869
 870 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
 871
 872 int
 873 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
 874 {
 875         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
 876         Elf_Addr *base;
 877         Elf_Addr *pos;
 878
 879         base = (Elf_Addr *)*stack_base;
 880         pos = base + (imgp->args->argc + imgp->args->envc + 2);
 881
 882         if (args->trace) {
 883                 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
 884         }
 885         if (args->execfd != -1) {
 886                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
 887         }
 888         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
 889         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
 890         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
 891         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
 892         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
 893         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
 894         AUXARGS_ENTRY(pos, AT_BASE, args->base);
 895         AUXARGS_ENTRY(pos, AT_NULL, 0);
 896
 897         free(imgp->auxargs, M_TEMP);
 898         imgp->auxargs = NULL;
 899
 900         base--;
 901         suword(base, (long)imgp->args->argc);
 902         *stack_base = (register_t *)base;
 903         return (0);
 904 }
 905
 906 /*
 907  * Code for generating ELF core dumps.
 908  */
 909
 910 typedef void (*segment_callback)(vm_map_entry_t, void *);
 911
 912 /* Closure for cb_put_phdr(). */
 913 struct phdr_closure {
 914         Elf_Phdr *phdr;         /* Program header to fill in */
 915         Elf_Off offset;         /* Offset of segment in core file */
 916 };
 917
 918 /* Closure for cb_size_segment(). */
 919 struct sseg_closure {
 920         int count;              /* Count of writable segments. */
 921         size_t size;            /* Total size of all writable segments. */
 922 };
 923
 924 static void cb_put_phdr(vm_map_entry_t, void *);
 925 static void cb_size_segment(vm_map_entry_t, void *);
 926 static void each_writable_segment(struct thread *, segment_callback, void *);
 927 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
 928     int, void *, size_t);
 929 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
 930 static void __elfN(putnote)(void *, size_t *, const char *, int,
 931     const void *, size_t);
 932
 933 extern int osreldate;
 934
 935 int
 936 __elfN(coredump)(td, vp, limit)
 937         struct thread *td;
 938         struct vnode *vp;
 939         off_t limit;
 940 {
 941         struct ucred *cred = td->td_ucred;
 942         int error = 0;
 943         struct sseg_closure seginfo;
 944         void *hdr;
 945         size_t hdrsize;
 946
 947         /* Size the program segments. */
 948         seginfo.count = 0;
 949         seginfo.size = 0;
 950         each_writable_segment(td, cb_size_segment, &seginfo);
 951
 952         /*
 953          * Calculate the size of the core file header area by making
 954          * a dry run of generating it.  Nothing is written, but the
 955          * size is calculated.
 956          */
 957         hdrsize = 0;
 958         __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
 959
 960         if (hdrsize + seginfo.size >= limit)
 961                 return (EFAULT);
 962
 963         /*
 964          * Allocate memory for building the header, fill it up,
 965          * and write it out.
 966          */
 967         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
 968         if (hdr == NULL) {
 969                 return (EINVAL);
 970         }
 971         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
 972
 973         /* Write the contents of all of the writable segments. */
 974         if (error == 0) {
 975                 Elf_Phdr *php;
 976                 off_t offset;
 977                 int i;
 978
 979                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 980                 offset = hdrsize;
 981                 for (i = 0; i < seginfo.count; i++) {
 982                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
 983                             (caddr_t)(uintptr_t)php->p_vaddr,
 984                             php->p_filesz, offset, UIO_USERSPACE,
 985                             IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 986                             curthread); /* XXXKSE */
 987                         if (error != 0)
 988                                 break;
 989                         offset += php->p_filesz;
 990                         php++;
 991                 }
 992         }
 993         free(hdr, M_TEMP);
 994
 995         return (error);
 996 }
 997
 998 /*
 999  * A callback for each_writable_segment() to write out the segment's
1000  * program header entry.
1001  */
1002 static void
1003 cb_put_phdr(entry, closure)
1004         vm_map_entry_t entry;
1005         void *closure;
1006 {
1007         struct phdr_closure *phc = (struct phdr_closure *)closure;
1008         Elf_Phdr *phdr = phc->phdr;
1009
1010         phc->offset = round_page(phc->offset);
1011
1012         phdr->p_type = PT_LOAD;
1013         phdr->p_offset = phc->offset;
1014         phdr->p_vaddr = entry->start;
1015         phdr->p_paddr = 0;
1016         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1017         phdr->p_align = PAGE_SIZE;
1018         phdr->p_flags = 0;
1019         if (entry->protection & VM_PROT_READ)
1020                 phdr->p_flags |= PF_R;
1021         if (entry->protection & VM_PROT_WRITE)
1022                 phdr->p_flags |= PF_W;
1023         if (entry->protection & VM_PROT_EXECUTE)
1024                 phdr->p_flags |= PF_X;
1025
1026         phc->offset += phdr->p_filesz;
1027         phc->phdr++;
1028 }
1029
1030 /*
1031  * A callback for each_writable_segment() to gather information about
1032  * the number of segments and their total size.
1033  */
1034 static void
1035 cb_size_segment(entry, closure)
1036         vm_map_entry_t entry;
1037         void *closure;
1038 {
1039         struct sseg_closure *ssc = (struct sseg_closure *)closure;
1040
1041         ssc->count++;
1042         ssc->size += entry->end - entry->start;
1043 }
1044
1045 /*
1046  * For each writable segment in the process's memory map, call the given
1047  * function with a pointer to the map entry and some arbitrary
1048  * caller-supplied data.
1049  */
1050 static void
1051 each_writable_segment(td, func, closure)
1052         struct thread *td;
1053         segment_callback func;
1054         void *closure;
1055 {
1056         struct proc *p = td->td_proc;
1057         vm_map_t map = &p->p_vmspace->vm_map;
1058         vm_map_entry_t entry;
1059
1060         for (entry = map->header.next; entry != &map->header;
1061             entry = entry->next) {
1062                 vm_object_t obj;
1063
1064                 /*
1065                  * Don't dump inaccessible mappings, deal with legacy
1066                  * coredump mode.
1067                  *
1068                  * Note that read-only segments related to the elf binary
1069                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1070                  * need to arbitrarily ignore such segments.
1071                  */
1072                 if (elf_legacy_coredump) {
1073                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1074                                 continue;
1075                 } else {
1076                         if ((entry->protection & VM_PROT_ALL) == 0)
1077                                 continue;
1078                 }
1079
1080                 /*
1081                  * Dont include memory segment in the coredump if
1082                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1083                  * madvise(2).  Do not dump submaps (i.e. parts of the
1084                  * kernel map).
1085                  */
1086                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1087                         continue;
1088
1089                 if ((obj = entry->object.vm_object) == NULL)
1090                         continue;
1091
1092                 /* Find the deepest backing object. */
1093                 while (obj->backing_object != NULL)
1094                         obj = obj->backing_object;
1095
1096                 /* Ignore memory-mapped devices and such things. */
1097                 if (obj->type != OBJT_DEFAULT &&
1098                     obj->type != OBJT_SWAP &&
1099                     obj->type != OBJT_VNODE)
1100                         continue;
1101
1102                 (*func)(entry, closure);
1103         }
1104 }
1105
1106 /*
1107  * Write the core file header to the file, including padding up to
1108  * the page boundary.
1109  */
1110 static int
1111 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1112         struct thread *td;
1113         struct vnode *vp;
1114         struct ucred *cred;
1115         int numsegs;
1116         size_t hdrsize;
1117         void *hdr;
1118 {
1119         size_t off;
1120
1121         /* Fill in the header. */
1122         bzero(hdr, hdrsize);
1123         off = 0;
1124         __elfN(puthdr)(td, hdr, &off, numsegs);
1125
1126         /* Write it to the core file. */
1127         return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1128             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1129             td)); /* XXXKSE */
1130 }
1131
1132 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1133 typedef struct prstatus32 elf_prstatus_t;
1134 typedef struct prpsinfo32 elf_prpsinfo_t;
1135 typedef struct fpreg32 elf_prfpregset_t;
1136 typedef struct fpreg32 elf_fpregset_t;
1137 typedef struct reg32 elf_gregset_t;
1138 #else
1139 typedef prstatus_t elf_prstatus_t;
1140 typedef prpsinfo_t elf_prpsinfo_t;
1141 typedef prfpregset_t elf_prfpregset_t;
1142 typedef prfpregset_t elf_fpregset_t;
1143 typedef gregset_t elf_gregset_t;
1144 #endif
1145
1146 static void
1147 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1148 {
1149         struct {
1150                 elf_prstatus_t status;
1151                 elf_prfpregset_t fpregset;
1152                 elf_prpsinfo_t psinfo;
1153         } *tempdata;
1154         elf_prstatus_t *status;
1155         elf_prfpregset_t *fpregset;
1156         elf_prpsinfo_t *psinfo;
1157         struct proc *p;
1158         struct thread *thr;
1159         size_t ehoff, noteoff, notesz, phoff;
1160
1161         p = td->td_proc;
1162
1163         ehoff = *off;
1164         *off += sizeof(Elf_Ehdr);
1165
1166         phoff = *off;
1167         *off += (numsegs + 1) * sizeof(Elf_Phdr);
1168
1169         noteoff = *off;
1170         /*
1171          * Don't allocate space for the notes if we're just calculating
1172          * the size of the header. We also don't collect the data.
1173          */
1174         if (dst != NULL) {
1175                 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1176                 status = &tempdata->status;
1177                 fpregset = &tempdata->fpregset;
1178                 psinfo = &tempdata->psinfo;
1179         } else {
1180                 tempdata = NULL;
1181                 status = NULL;
1182                 fpregset = NULL;
1183                 psinfo = NULL;
1184         }
1185
1186         if (dst != NULL) {
1187                 psinfo->pr_version = PRPSINFO_VERSION;
1188                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1189                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1190                 /*
1191                  * XXX - We don't fill in the command line arguments properly
1192                  * yet.
1193                  */
1194                 strlcpy(psinfo->pr_psargs, p->p_comm,
1195                     sizeof(psinfo->pr_psargs));
1196         }
1197         __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1198             sizeof *psinfo);
1199
1200         /*
1201          * To have the debugger select the right thread (LWP) as the initial
1202          * thread, we dump the state of the thread passed to us in td first.
1203          * This is the thread that causes the core dump and thus likely to
1204          * be the right thread one wants to have selected in the debugger.
1205          */
1206         thr = td;
1207         while (thr != NULL) {
1208                 if (dst != NULL) {
1209                         status->pr_version = PRSTATUS_VERSION;
1210                         status->pr_statussz = sizeof(elf_prstatus_t);
1211                         status->pr_gregsetsz = sizeof(elf_gregset_t);
1212                         status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1213                         status->pr_osreldate = osreldate;
1214                         status->pr_cursig = p->p_sig;
1215                         status->pr_pid = thr->td_tid;
1216 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1217                         fill_regs32(thr, &status->pr_reg);
1218                         fill_fpregs32(thr, fpregset);
1219 #else
1220                         fill_regs(thr, &status->pr_reg);
1221                         fill_fpregs(thr, fpregset);
1222 #endif
1223                 }
1224                 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1225                     sizeof *status);
1226                 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1227                     sizeof *fpregset);
1228                 /*
1229                  * Allow for MD specific notes, as well as any MD
1230                  * specific preparations for writing MI notes.
1231                  */
1232                 __elfN(dump_thread)(thr, dst, off);
1233
1234                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1235                     TAILQ_NEXT(thr, td_plist);
1236                 if (thr == td)
1237                         thr = TAILQ_NEXT(thr, td_plist);
1238         }
1239
1240         notesz = *off - noteoff;
1241
1242         if (dst != NULL)
1243                 free(tempdata, M_TEMP);
1244
1245         /* Align up to a page boundary for the program segments. */
1246         *off = round_page(*off);
1247
1248         if (dst != NULL) {
1249                 Elf_Ehdr *ehdr;
1250                 Elf_Phdr *phdr;
1251                 struct phdr_closure phc;
1252
1253                 /*
1254                  * Fill in the ELF header.
1255                  */
1256                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1257                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1258                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1259                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1260                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1261                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1262                 ehdr->e_ident[EI_DATA] = ELF_DATA;
1263                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1264                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1265                 ehdr->e_ident[EI_ABIVERSION] = 0;
1266                 ehdr->e_ident[EI_PAD] = 0;
1267                 ehdr->e_type = ET_CORE;
1268 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1269                 ehdr->e_machine = EM_386;
1270 #else
1271                 ehdr->e_machine = ELF_ARCH;
1272 #endif
1273                 ehdr->e_version = EV_CURRENT;
1274                 ehdr->e_entry = 0;
1275                 ehdr->e_phoff = phoff;
1276                 ehdr->e_flags = 0;
1277                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1278                 ehdr->e_phentsize = sizeof(Elf_Phdr);
1279                 ehdr->e_phnum = numsegs + 1;
1280                 ehdr->e_shentsize = sizeof(Elf_Shdr);
1281                 ehdr->e_shnum = 0;
1282                 ehdr->e_shstrndx = SHN_UNDEF;
1283
1284                 /*
1285                  * Fill in the program header entries.
1286                  */
1287                 phdr = (Elf_Phdr *)((char *)dst + phoff);
1288
1289                 /* The note segement. */
1290                 phdr->p_type = PT_NOTE;
1291                 phdr->p_offset = noteoff;
1292                 phdr->p_vaddr = 0;
1293                 phdr->p_paddr = 0;
1294                 phdr->p_filesz = notesz;
1295                 phdr->p_memsz = 0;
1296                 phdr->p_flags = 0;
1297                 phdr->p_align = 0;
1298                 phdr++;
1299
1300                 /* All the writable segments from the program. */
1301                 phc.phdr = phdr;
1302                 phc.offset = *off;
1303                 each_writable_segment(td, cb_put_phdr, &phc);
1304         }
1305 }
1306
1307 static void
1308 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1309     const void *desc, size_t descsz)
1310 {
1311         Elf_Note note;
1312
1313         note.n_namesz = strlen(name) + 1;
1314         note.n_descsz = descsz;
1315         note.n_type = type;
1316         if (dst != NULL)
1317                 bcopy(&note, (char *)dst + *off, sizeof note);
1318         *off += sizeof note;
1319         if (dst != NULL)
1320                 bcopy(name, (char *)dst + *off, note.n_namesz);
1321         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1322         if (dst != NULL)
1323                 bcopy(desc, (char *)dst + *off, note.n_descsz);
1324         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1325 }
1326
1327 /*
1328  * Tell kern_execve.c about it, with a little help from the linker.
1329  */
1330 static struct execsw __elfN(execsw) = {
1331         __CONCAT(exec_, __elfN(imgact)),
1332         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1333 };
1334 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));