sys/kern/imgact_elf.c

   1 /*-
   2  * Copyright (c) 1995-1996 Søren Schmidt
   3  * Copyright (c) 1996 Peter Wemm
   4  * All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer
  11  *    in this position and unchanged.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. The name of the author may not be used to endorse or promote products
  16  *    derived from this software withough specific prior written permission
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28  *
  29  * $FreeBSD: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $
  30  * $DragonFly: src/sys/kern/imgact_elf.c,v 1.9 2003/09/23 05:03:51 dillon Exp $
  31  */
  32
  33 #include <sys/param.h>
  34 #include <sys/exec.h>
  35 #include <sys/fcntl.h>
  36 #include <sys/imgact.h>
  37 #include <sys/imgact_elf.h>
  38 #include <sys/kernel.h>
  39 #include <sys/malloc.h>
  40 #include <sys/mman.h>
  41 #include <sys/systm.h>
  42 #include <sys/proc.h>
  43 #include <sys/namei.h>
  44 #include <sys/pioctl.h>
  45 #include <sys/procfs.h>
  46 #include <sys/resourcevar.h>
  47 #include <sys/signalvar.h>
  48 #include <sys/stat.h>
  49 #include <sys/syscall.h>
  50 #include <sys/sysctl.h>
  51 #include <sys/sysent.h>
  52 #include <sys/vnode.h>
  53
  54 #include <vm/vm.h>
  55 #include <vm/vm_kern.h>
  56 #include <vm/vm_param.h>
  57 #include <vm/pmap.h>
  58 #include <sys/lock.h>
  59 #include <vm/vm_map.h>
  60 #include <vm/vm_object.h>
  61 #include <vm/vm_extern.h>
  62
  63 #include <machine/elf.h>
  64 #include <machine/md_var.h>
  65
  66 #define OLD_EI_BRAND    8
  67
  68 __ElfType(Brandinfo);
  69 __ElfType(Auxargs);
  70
  71 static int elf_check_header (const Elf_Ehdr *hdr);
  72 static int elf_freebsd_fixup (register_t **stack_base,
  73     struct image_params *imgp);
  74 static int elf_load_file (struct proc *p, const char *file, u_long *addr,
  75     u_long *entry);
  76 static int elf_load_section (struct proc *p,
  77     struct vmspace *vmspace, struct vnode *vp,
  78     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
  79     vm_prot_t prot);
  80 static int exec_elf_imgact (struct image_params *imgp);
  81
  82 static int elf_trace = 0;
  83 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
  84 static int elf_legacy_coredump = 0;
  85 SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
  86     &elf_legacy_coredump, 0, "");
  87
  88 static struct sysentvec elf_freebsd_sysvec = {
  89         SYS_MAXSYSCALL,
  90         sysent,
  91         0,
  92         0,
  93         0,
  94         0,
  95         0,
  96         0,
  97         elf_freebsd_fixup,
  98         sendsig,
  99         sigcode,
 100         &szsigcode,
 101         0,
 102         "FreeBSD ELF",
 103         elf_coredump,
 104         NULL,
 105         MINSIGSTKSZ
 106 };
 107
 108 static Elf_Brandinfo freebsd_brand_info = {
 109                                                 ELFOSABI_FREEBSD,
 110                                                 "FreeBSD",
 111                                                 "",
 112                                                 "/usr/libexec/ld-elf.so.1",
 113                                                 &elf_freebsd_sysvec
 114                                           };
 115 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
 116                                                         &freebsd_brand_info,
 117                                                         NULL, NULL, NULL,
 118                                                         NULL, NULL, NULL, NULL
 119                                                     };
 120
 121 int
 122 elf_insert_brand_entry(Elf_Brandinfo *entry)
 123 {
 124         int i;
 125
 126         for (i=1; i<MAX_BRANDS; i++) {
 127                 if (elf_brand_list[i] == NULL) {
 128                         elf_brand_list[i] = entry;
 129                         break;
 130                 }
 131         }
 132         if (i == MAX_BRANDS)
 133                 return -1;
 134         return 0;
 135 }
 136
 137 int
 138 elf_remove_brand_entry(Elf_Brandinfo *entry)
 139 {
 140         int i;
 141
 142         for (i=1; i<MAX_BRANDS; i++) {
 143                 if (elf_brand_list[i] == entry) {
 144                         elf_brand_list[i] = NULL;
 145                         break;
 146                 }
 147         }
 148         if (i == MAX_BRANDS)
 149                 return -1;
 150         return 0;
 151 }
 152
 153 int
 154 elf_brand_inuse(Elf_Brandinfo *entry)
 155 {
 156         struct proc *p;
 157         int rval = FALSE;
 158
 159         FOREACH_PROC_IN_SYSTEM(p) {
 160                 if (p->p_sysent == entry->sysvec) {
 161                         rval = TRUE;
 162                         break;
 163                 }
 164         }
 165
 166         return (rval);
 167 }
 168
 169 static int
 170 elf_check_header(const Elf_Ehdr *hdr)
 171 {
 172         if (!IS_ELF(*hdr) ||
 173             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
 174             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
 175             hdr->e_ident[EI_VERSION] != EV_CURRENT)
 176                 return ENOEXEC;
 177
 178         if (!ELF_MACHINE_OK(hdr->e_machine))
 179                 return ENOEXEC;
 180
 181         if (hdr->e_version != ELF_TARG_VER)
 182                 return ENOEXEC;
 183
 184         return 0;
 185 }
 186
 187 static int
 188 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
 189 {
 190         size_t map_len;
 191         vm_offset_t map_addr;
 192         int error, rv, cow;
 193         int count;
 194         size_t copy_len;
 195         vm_object_t object;
 196         vm_offset_t file_addr;
 197         vm_offset_t data_buf = 0;
 198
 199         VOP_GETVOBJECT(vp, &object);
 200         error = 0;
 201
 202         /*
 203          * It's necessary to fail if the filsz + offset taken from the
 204          * header is greater than the actual file pager object's size.
 205          * If we were to allow this, then the vm_map_find() below would
 206          * walk right off the end of the file object and into the ether.
 207          *
 208          * While I'm here, might as well check for something else that
 209          * is invalid: filsz cannot be greater than memsz.
 210          */
 211         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
 212             filsz > memsz) {
 213                 uprintf("elf_load_section: truncated ELF file\n");
 214                 return (ENOEXEC);
 215         }
 216
 217         map_addr = trunc_page((vm_offset_t)vmaddr);
 218         file_addr = trunc_page(offset);
 219
 220         /*
 221          * We have two choices.  We can either clear the data in the last page
 222          * of an oversized mapping, or we can start the anon mapping a page
 223          * early and copy the initialized data into that first page.  We
 224          * choose the second..
 225          */
 226         if (memsz > filsz)
 227                 map_len = trunc_page(offset+filsz) - file_addr;
 228         else
 229                 map_len = round_page(offset+filsz) - file_addr;
 230
 231         if (map_len != 0) {
 232                 vm_object_reference(object);
 233
 234                 /* cow flags: don't dump readonly sections in core */
 235                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
 236                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
 237
 238                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 239                 vm_map_lock(&vmspace->vm_map);
 240                 rv = vm_map_insert(&vmspace->vm_map, &count,
 241                                       object,
 242                                       file_addr,        /* file offset */
 243                                       map_addr,         /* virtual start */
 244                                       map_addr + map_len,/* virtual end */
 245                                       prot,
 246                                       VM_PROT_ALL,
 247                                       cow);
 248                 vm_map_unlock(&vmspace->vm_map);
 249                 vm_map_entry_release(count);
 250                 if (rv != KERN_SUCCESS) {
 251                         vm_object_deallocate(object);
 252                         return EINVAL;
 253                 }
 254
 255                 /* we can stop now if we've covered it all */
 256                 if (memsz == filsz) {
 257                         return 0;
 258                 }
 259         }
 260
 261
 262         /*
 263          * We have to get the remaining bit of the file into the first part
 264          * of the oversized map segment.  This is normally because the .data
 265          * segment in the file is extended to provide bss.  It's a neat idea
 266          * to try and save a page, but it's a pain in the behind to implement.
 267          */
 268         copy_len = (offset + filsz) - trunc_page(offset + filsz);
 269         map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
 270         map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
 271
 272         /* This had damn well better be true! */
 273         if (map_len != 0) {
 274                 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
 275                 vm_map_lock(&vmspace->vm_map);
 276                 rv = vm_map_insert(&vmspace->vm_map, &count,
 277                                         NULL, 0,
 278                                         map_addr, map_addr + map_len,
 279                                         VM_PROT_ALL, VM_PROT_ALL, 0);
 280                 vm_map_unlock(&vmspace->vm_map);
 281                 vm_map_entry_release(count);
 282                 if (rv != KERN_SUCCESS) {
 283                         return EINVAL;
 284                 }
 285         }
 286
 287         if (copy_len != 0) {
 288                 vm_object_reference(object);
 289                 rv = vm_map_find(exec_map,
 290                                  object,
 291                                  trunc_page(offset + filsz),
 292                                  &data_buf,
 293                                  PAGE_SIZE,
 294                                  TRUE,
 295                                  VM_PROT_READ,
 296                                  VM_PROT_ALL,
 297                                  MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
 298                 if (rv != KERN_SUCCESS) {
 299                         vm_object_deallocate(object);
 300                         return EINVAL;
 301                 }
 302
 303                 /* send the page fragment to user space */
 304                 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
 305                 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
 306                 if (error) {
 307                         return (error);
 308                 }
 309         }
 310
 311         /*
 312          * set it to the specified protection
 313          */
 314         vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len,  prot,
 315                        FALSE);
 316
 317         return error;
 318 }
 319
 320 /*
 321  * Load the file "file" into memory.  It may be either a shared object
 322  * or an executable.
 323  *
 324  * The "addr" reference parameter is in/out.  On entry, it specifies
 325  * the address where a shared object should be loaded.  If the file is
 326  * an executable, this value is ignored.  On exit, "addr" specifies
 327  * where the file was actually loaded.
 328  *
 329  * The "entry" reference parameter is out only.  On exit, it specifies
 330  * the entry point for the loaded file.
 331  */
 332 static int
 333 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
 334 {
 335         struct {
 336                 struct nameidata nd;
 337                 struct vattr attr;
 338                 struct image_params image_params;
 339         } *tempdata;
 340         const Elf_Ehdr *hdr = NULL;
 341         const Elf_Phdr *phdr = NULL;
 342         struct nameidata *nd;
 343         struct vmspace *vmspace = p->p_vmspace;
 344         struct vattr *attr;
 345         struct image_params *imgp;
 346         vm_prot_t prot;
 347         u_long rbase;
 348         u_long base_addr = 0;
 349         int error, i, numsegs;
 350         struct thread *td = p->p_thread;
 351
 352         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
 353         nd = &tempdata->nd;
 354         attr = &tempdata->attr;
 355         imgp = &tempdata->image_params;
 356
 357         /*
 358          * Initialize part of the common data
 359          */
 360         imgp->proc = p;
 361         imgp->uap = NULL;
 362         imgp->attr = attr;
 363         imgp->firstpage = NULL;
 364         imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
 365
 366         if (imgp->image_header == NULL) {
 367                 nd->ni_vp = NULL;
 368                 error = ENOMEM;
 369                 goto fail;
 370         }
 371
 372         NDINIT(nd, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW,
 373             UIO_SYSSPACE, file, td);
 374
 375         if ((error = namei(nd)) != 0) {
 376                 nd->ni_vp = NULL;
 377                 goto fail;
 378         }
 379         NDFREE(nd, NDF_ONLY_PNBUF);
 380         imgp->vp = nd->ni_vp;
 381
 382         /*
 383          * Check permissions, modes, uid, etc on the file, and "open" it.
 384          */
 385         error = exec_check_permissions(imgp);
 386         if (error) {
 387                 VOP_UNLOCK(nd->ni_vp, 0, td);
 388                 goto fail;
 389         }
 390
 391         error = exec_map_first_page(imgp);
 392         /*
 393          * Also make certain that the interpreter stays the same, so set
 394          * its VTEXT flag, too.
 395          */
 396         if (error == 0)
 397                 nd->ni_vp->v_flag |= VTEXT;
 398         VOP_UNLOCK(nd->ni_vp, 0, td);
 399         if (error)
 400                 goto fail;
 401
 402         hdr = (const Elf_Ehdr *)imgp->image_header;
 403         if ((error = elf_check_header(hdr)) != 0)
 404                 goto fail;
 405         if (hdr->e_type == ET_DYN)
 406                 rbase = *addr;
 407         else if (hdr->e_type == ET_EXEC)
 408                 rbase = 0;
 409         else {
 410                 error = ENOEXEC;
 411                 goto fail;
 412         }
 413
 414         /* Only support headers that fit within first page for now */
 415         if ((hdr->e_phoff > PAGE_SIZE) ||
 416             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
 417                 error = ENOEXEC;
 418                 goto fail;
 419         }
 420
 421         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 422
 423         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
 424                 if (phdr[i].p_type == PT_LOAD) {        /* Loadable segment */
 425                         prot = 0;
 426                         if (phdr[i].p_flags & PF_X)
 427                                 prot |= VM_PROT_EXECUTE;
 428                         if (phdr[i].p_flags & PF_W)
 429                                 prot |= VM_PROT_WRITE;
 430                         if (phdr[i].p_flags & PF_R)
 431                                 prot |= VM_PROT_READ;
 432
 433                         error = elf_load_section(
 434                                     p, vmspace, nd->ni_vp,
 435                                     phdr[i].p_offset,
 436                                     (caddr_t)phdr[i].p_vaddr +
 437                                     rbase,
 438                                     phdr[i].p_memsz,
 439                                     phdr[i].p_filesz, prot);
 440                         if (error != 0)
 441                                 goto fail;
 442                         /*
 443                          * Establish the base address if this is the
 444                          * first segment.
 445                          */
 446                         if (numsegs == 0)
 447                                 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
 448                         numsegs++;
 449                 }
 450         }
 451         *addr = base_addr;
 452         *entry=(unsigned long)hdr->e_entry + rbase;
 453
 454 fail:
 455         if (imgp->firstpage)
 456                 exec_unmap_first_page(imgp);
 457         if (imgp->image_header)
 458                 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
 459                         PAGE_SIZE);
 460         if (nd->ni_vp)
 461                 vrele(nd->ni_vp);
 462
 463         free(tempdata, M_TEMP);
 464
 465         return error;
 466 }
 467
 468 /*
 469  * non static, as it can be overridden by start_init()
 470  */
 471 int fallback_elf_brand = -1;
 472 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
 473                 &fallback_elf_brand, -1,
 474                 "ELF brand of last resort");
 475
 476 static int
 477 exec_elf_imgact(struct image_params *imgp)
 478 {
 479         const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
 480         const Elf_Phdr *phdr;
 481         Elf_Auxargs *elf_auxargs = NULL;
 482         struct vmspace *vmspace;
 483         vm_prot_t prot;
 484         u_long text_size = 0, data_size = 0, total_size = 0;
 485         u_long text_addr = 0, data_addr = 0;
 486         u_long seg_size, seg_addr;
 487         u_long addr, entry = 0, proghdr = 0;
 488         int error, i;
 489         const char *interp = NULL;
 490         Elf_Brandinfo *brand_info;
 491         char *path;
 492
 493         /*
 494          * Do we have a valid ELF header ?
 495          */
 496         if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
 497                 return -1;
 498
 499         /*
 500          * From here on down, we return an errno, not -1, as we've
 501          * detected an ELF file.
 502          */
 503
 504         if ((hdr->e_phoff > PAGE_SIZE) ||
 505             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
 506                 /* Only support headers in first page for now */
 507                 return ENOEXEC;
 508         }
 509         phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
 510
 511         /*
 512          * From this point on, we may have resources that need to be freed.
 513          */
 514
 515         if ((error = exec_extract_strings(imgp)) != 0)
 516                 goto fail;
 517
 518         exec_new_vmspace(imgp);
 519
 520         /*
 521          * Yeah, I'm paranoid.  There is every reason in the world to get
 522          * VTEXT now since from here on out, there are places we can have
 523          * a context switch.  Better safe than sorry; I really don't want
 524          * the file to change while it's being loaded.
 525          */
 526         lwkt_gettoken(&imgp->vp->v_interlock);
 527         imgp->vp->v_flag |= VTEXT;
 528         lwkt_reltoken(&imgp->vp->v_interlock);
 529
 530         vmspace = imgp->proc->p_vmspace;
 531
 532         for (i = 0; i < hdr->e_phnum; i++) {
 533                 switch(phdr[i].p_type) {
 534
 535                 case PT_LOAD:   /* Loadable segment */
 536                         prot = 0;
 537                         if (phdr[i].p_flags & PF_X)
 538                                 prot |= VM_PROT_EXECUTE;
 539                         if (phdr[i].p_flags & PF_W)
 540                                 prot |= VM_PROT_WRITE;
 541                         if (phdr[i].p_flags & PF_R)
 542                                 prot |= VM_PROT_READ;
 543
 544                         if ((error = elf_load_section(imgp->proc,
 545                                                      vmspace, imgp->vp,
 546                                                      phdr[i].p_offset,
 547                                                      (caddr_t)phdr[i].p_vaddr,
 548                                                      phdr[i].p_memsz,
 549                                                      phdr[i].p_filesz, prot)) != 0)
 550                                 goto fail;
 551
 552                         seg_addr = trunc_page(phdr[i].p_vaddr);
 553                         seg_size = round_page(phdr[i].p_memsz +
 554                                 phdr[i].p_vaddr - seg_addr);
 555
 556                         /*
 557                          * Is this .text or .data?  We can't use
 558                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
 559                          * alpha terribly and possibly does other bad
 560                          * things so we stick to the old way of figuring
 561                          * it out:  If the segment contains the program
 562                          * entry point, it's a text segment, otherwise it
 563                          * is a data segment.
 564                          *
 565                          * Note that obreak() assumes that data_addr +
 566                          * data_size == end of data load area, and the ELF
 567                          * file format expects segments to be sorted by
 568                          * address.  If multiple data segments exist, the
 569                          * last one will be used.
 570                          */
 571                         if (hdr->e_entry >= phdr[i].p_vaddr &&
 572                             hdr->e_entry < (phdr[i].p_vaddr +
 573                             phdr[i].p_memsz)) {
 574                                 text_size = seg_size;
 575                                 text_addr = seg_addr;
 576                                 entry = (u_long)hdr->e_entry;
 577                         } else {
 578                                 data_size = seg_size;
 579                                 data_addr = seg_addr;
 580                         }
 581                         total_size += seg_size;
 582
 583                         /*
 584                          * Check limits.  It should be safe to check the
 585                          * limits after loading the segment since we do
 586                          * not actually fault in all the segment's pages.
 587                          */
 588                         if (data_size >
 589                             imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
 590                             text_size > maxtsiz ||
 591                             total_size >
 592                             imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
 593                                 error = ENOMEM;
 594                                 goto fail;
 595                         }
 596                         break;
 597                 case PT_INTERP: /* Path to interpreter */
 598                         if (phdr[i].p_filesz > MAXPATHLEN ||
 599                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
 600                                 error = ENOEXEC;
 601                                 goto fail;
 602                         }
 603                         interp = imgp->image_header + phdr[i].p_offset;
 604                         break;
 605                 case PT_PHDR:   /* Program header table info */
 606                         proghdr = phdr[i].p_vaddr;
 607                         break;
 608                 default:
 609                         break;
 610                 }
 611         }
 612
 613         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
 614         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
 615         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
 616         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
 617
 618         addr = ELF_RTLD_ADDR(vmspace);
 619
 620         imgp->entry_addr = entry;
 621
 622         brand_info = NULL;
 623
 624         /* We support three types of branding -- (1) the ELF EI_OSABI field
 625          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
 626          * branding w/in the ELF header, and (3) path of the `interp_path'
 627          * field.  We should also look for an ".note.ABI-tag" ELF section now
 628          * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
 629          */
 630
 631         /* If the executable has a brand, search for it in the brand list. */
 632         if (brand_info == NULL) {
 633                 for (i = 0;  i < MAX_BRANDS;  i++) {
 634                         Elf_Brandinfo *bi = elf_brand_list[i];
 635
 636                         if (bi != NULL &&
 637                             (hdr->e_ident[EI_OSABI] == bi->brand
 638                             || 0 ==
 639                             strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
 640                             bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
 641                                 brand_info = bi;
 642                                 break;
 643                         }
 644                 }
 645         }
 646
 647         /* Lacking a known brand, search for a recognized interpreter. */
 648         if (brand_info == NULL && interp != NULL) {
 649                 for (i = 0;  i < MAX_BRANDS;  i++) {
 650                         Elf_Brandinfo *bi = elf_brand_list[i];
 651
 652                         if (bi != NULL &&
 653                             strcmp(interp, bi->interp_path) == 0) {
 654                                 brand_info = bi;
 655                                 break;
 656                         }
 657                 }
 658         }
 659
 660         /* Lacking a recognized interpreter, try the default brand */
 661         if (brand_info == NULL) {
 662                 for (i = 0; i < MAX_BRANDS; i++) {
 663                         Elf_Brandinfo *bi = elf_brand_list[i];
 664
 665                         if (bi != NULL && fallback_elf_brand == bi->brand) {
 666                                 brand_info = bi;
 667                                 break;
 668                         }
 669                 }
 670         }
 671
 672         if (brand_info == NULL) {
 673                 uprintf("ELF binary type \"%u\" not known.\n",
 674                     hdr->e_ident[EI_OSABI]);
 675                 error = ENOEXEC;
 676                 goto fail;
 677         }
 678
 679         imgp->proc->p_sysent = brand_info->sysvec;
 680         if (interp != NULL) {
 681                 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 682                 snprintf(path, MAXPATHLEN, "%s%s",
 683                          brand_info->emul_path, interp);
 684                 if ((error = elf_load_file(imgp->proc, path, &addr,
 685                                            &imgp->entry_addr)) != 0) {
 686                         if ((error = elf_load_file(imgp->proc, interp, &addr,
 687                                                    &imgp->entry_addr)) != 0) {
 688                                 uprintf("ELF interpreter %s not found\n", path);
 689                                 free(path, M_TEMP);
 690                                 goto fail;
 691                         }
 692                 }
 693                 free(path, M_TEMP);
 694         }
 695
 696         /*
 697          * Construct auxargs table (used by the fixup routine)
 698          */
 699         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
 700         elf_auxargs->execfd = -1;
 701         elf_auxargs->phdr = proghdr;
 702         elf_auxargs->phent = hdr->e_phentsize;
 703         elf_auxargs->phnum = hdr->e_phnum;
 704         elf_auxargs->pagesz = PAGE_SIZE;
 705         elf_auxargs->base = addr;
 706         elf_auxargs->flags = 0;
 707         elf_auxargs->entry = entry;
 708         elf_auxargs->trace = elf_trace;
 709
 710         imgp->auxargs = elf_auxargs;
 711         imgp->interpreted = 0;
 712
 713 fail:
 714         return error;
 715 }
 716
 717 static int
 718 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
 719 {
 720         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
 721         register_t *pos;
 722
 723         pos = *stack_base + (imgp->argc + imgp->envc + 2);
 724
 725         if (args->trace) {
 726                 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
 727         }
 728         if (args->execfd != -1) {
 729                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
 730         }
 731         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
 732         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
 733         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
 734         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
 735         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
 736         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
 737         AUXARGS_ENTRY(pos, AT_BASE, args->base);
 738         AUXARGS_ENTRY(pos, AT_NULL, 0);
 739
 740         free(imgp->auxargs, M_TEMP);
 741         imgp->auxargs = NULL;
 742
 743         (*stack_base)--;
 744         suword(*stack_base, (long) imgp->argc);
 745         return 0;
 746 }
 747
 748 /*
 749  * Code for generating ELF core dumps.
 750  */
 751
 752 typedef void (*segment_callback) (vm_map_entry_t, void *);
 753
 754 /* Closure for cb_put_phdr(). */
 755 struct phdr_closure {
 756         Elf_Phdr *phdr;         /* Program header to fill in */
 757         Elf_Off offset;         /* Offset of segment in core file */
 758 };
 759
 760 /* Closure for cb_size_segment(). */
 761 struct sseg_closure {
 762         int count;              /* Count of writable segments. */
 763         size_t size;            /* Total size of all writable segments. */
 764 };
 765
 766 static void cb_put_phdr (vm_map_entry_t, void *);
 767 static void cb_size_segment (vm_map_entry_t, void *);
 768 static void each_writable_segment (struct proc *, segment_callback,
 769     void *);
 770 static int elf_corehdr (struct proc *, struct vnode *, struct ucred *,
 771     int, void *, size_t);
 772 static void elf_puthdr (struct proc *, void *, size_t *,
 773     const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int);
 774 static void elf_putnote (void *, size_t *, const char *, int,
 775     const void *, size_t);
 776
 777 extern int osreldate;
 778
 779 int
 780 elf_coredump(p, vp, limit)
 781         struct proc *p;
 782         struct vnode *vp;
 783         off_t limit;
 784 {
 785         struct ucred *cred = p->p_ucred;
 786         struct thread *td = p->p_thread;
 787         int error = 0;
 788         struct sseg_closure seginfo;
 789         void *hdr;
 790         size_t hdrsize;
 791
 792         /* Size the program segments. */
 793         seginfo.count = 0;
 794         seginfo.size = 0;
 795         each_writable_segment(p, cb_size_segment, &seginfo);
 796
 797         /*
 798          * Calculate the size of the core file header area by making
 799          * a dry run of generating it.  Nothing is written, but the
 800          * size is calculated.
 801          */
 802         hdrsize = 0;
 803         elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize,
 804             (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
 805             (const prpsinfo_t *)NULL, seginfo.count);
 806
 807         if (hdrsize + seginfo.size >= limit)
 808                 return (EFAULT);
 809
 810         /*
 811          * Allocate memory for building the header, fill it up,
 812          * and write it out.
 813          */
 814         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
 815         if (hdr == NULL) {
 816                 return EINVAL;
 817         }
 818         error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize);
 819
 820         /* Write the contents of all of the writable segments. */
 821         if (error == 0) {
 822                 Elf_Phdr *php;
 823                 off_t offset;
 824                 int i;
 825
 826                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 827                 offset = hdrsize;
 828                 for (i = 0;  i < seginfo.count;  i++) {
 829                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
 830                             (caddr_t)php->p_vaddr,
 831                             php->p_filesz, offset, UIO_USERSPACE,
 832                             IO_UNIT | IO_DIRECT | IO_CORE, cred,
 833                             (int *)NULL, td);
 834                         if (error != 0)
 835                                 break;
 836                         offset += php->p_filesz;
 837                         php++;
 838                 }
 839         }
 840         free(hdr, M_TEMP);
 841
 842         return error;
 843 }
 844
 845 /*
 846  * A callback for each_writable_segment() to write out the segment's
 847  * program header entry.
 848  */
 849 static void
 850 cb_put_phdr(entry, closure)
 851         vm_map_entry_t entry;
 852         void *closure;
 853 {
 854         struct phdr_closure *phc = (struct phdr_closure *)closure;
 855         Elf_Phdr *phdr = phc->phdr;
 856
 857         phc->offset = round_page(phc->offset);
 858
 859         phdr->p_type = PT_LOAD;
 860         phdr->p_offset = phc->offset;
 861         phdr->p_vaddr = entry->start;
 862         phdr->p_paddr = 0;
 863         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
 864         phdr->p_align = PAGE_SIZE;
 865         phdr->p_flags = 0;
 866         if (entry->protection & VM_PROT_READ)
 867                 phdr->p_flags |= PF_R;
 868         if (entry->protection & VM_PROT_WRITE)
 869                 phdr->p_flags |= PF_W;
 870         if (entry->protection & VM_PROT_EXECUTE)
 871                 phdr->p_flags |= PF_X;
 872
 873         phc->offset += phdr->p_filesz;
 874         phc->phdr++;
 875 }
 876
 877 /*
 878  * A callback for each_writable_segment() to gather information about
 879  * the number of segments and their total size.
 880  */
 881 static void
 882 cb_size_segment(entry, closure)
 883         vm_map_entry_t entry;
 884         void *closure;
 885 {
 886         struct sseg_closure *ssc = (struct sseg_closure *)closure;
 887
 888         ssc->count++;
 889         ssc->size += entry->end - entry->start;
 890 }
 891
 892 /*
 893  * For each writable segment in the process's memory map, call the given
 894  * function with a pointer to the map entry and some arbitrary
 895  * caller-supplied data.
 896  */
 897 static void
 898 each_writable_segment(p, func, closure)
 899         struct proc *p;
 900         segment_callback func;
 901         void *closure;
 902 {
 903         vm_map_t map = &p->p_vmspace->vm_map;
 904         vm_map_entry_t entry;
 905
 906         for (entry = map->header.next;  entry != &map->header;
 907             entry = entry->next) {
 908                 vm_object_t obj;
 909
 910                 /*
 911                  * Don't dump inaccessible mappings, deal with legacy
 912                  * coredump mode.
 913                  *
 914                  * Note that read-only segments related to the elf binary
 915                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
 916                  * need to arbitrarily ignore such segments.
 917                  */
 918                 if (elf_legacy_coredump) {
 919                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
 920                                 continue;
 921                 } else {
 922                         if ((entry->protection & VM_PROT_ALL) == 0)
 923                                 continue;
 924                 }
 925
 926                 /*
 927                  * Dont include memory segment in the coredump if
 928                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
 929                  * madvise(2).  Do not dump submaps (i.e. parts of the
 930                  * kernel map).
 931                  */
 932                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
 933                         continue;
 934
 935                 if ((obj = entry->object.vm_object) == NULL)
 936                         continue;
 937
 938                 /* Find the deepest backing object. */
 939                 while (obj->backing_object != NULL)
 940                         obj = obj->backing_object;
 941
 942                 /* Ignore memory-mapped devices and such things. */
 943                 if (obj->type != OBJT_DEFAULT &&
 944                     obj->type != OBJT_SWAP &&
 945                     obj->type != OBJT_VNODE)
 946                         continue;
 947
 948                 (*func)(entry, closure);
 949         }
 950 }
 951
 952 /*
 953  * Write the core file header to the file, including padding up to
 954  * the page boundary.
 955  */
 956 static int
 957 elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize)
 958         struct proc *p;
 959         struct vnode *vp;
 960         struct ucred *cred;
 961         int numsegs;
 962         size_t hdrsize;
 963         void *hdr;
 964 {
 965         struct {
 966                 prstatus_t status;
 967                 prfpregset_t fpregset;
 968                 prpsinfo_t psinfo;
 969         } *tempdata;
 970         size_t off;
 971         prstatus_t *status;
 972         prfpregset_t *fpregset;
 973         prpsinfo_t *psinfo;
 974         struct thread *td = p->p_thread;
 975
 976         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK);
 977         status = &tempdata->status;
 978         fpregset = &tempdata->fpregset;
 979         psinfo = &tempdata->psinfo;
 980
 981         /* Gather the information for the header. */
 982         status->pr_version = PRSTATUS_VERSION;
 983         status->pr_statussz = sizeof(prstatus_t);
 984         status->pr_gregsetsz = sizeof(gregset_t);
 985         status->pr_fpregsetsz = sizeof(fpregset_t);
 986         status->pr_osreldate = osreldate;
 987         status->pr_cursig = p->p_sig;
 988         status->pr_pid = p->p_pid;
 989         fill_regs(p, &status->pr_reg);
 990
 991         fill_fpregs(p, fpregset);
 992
 993         psinfo->pr_version = PRPSINFO_VERSION;
 994         psinfo->pr_psinfosz = sizeof(prpsinfo_t);
 995         strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1);
 996
 997         /* XXX - We don't fill in the command line arguments properly yet. */
 998         strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
 999
1000         /* Fill in the header. */
1001         bzero(hdr, hdrsize);
1002         off = 0;
1003         elf_puthdr(p, hdr, &off, status, fpregset, psinfo, numsegs);
1004
1005         free(tempdata, M_TEMP);
1006
1007         /* Write it to the core file. */
1008         return vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1009             UIO_SYSSPACE, IO_UNIT | IO_DIRECT | IO_CORE, cred, NULL, td);
1010 }
1011
1012 static void
1013 elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
1014     const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
1015 {
1016         size_t ehoff;
1017         size_t phoff;
1018         size_t noteoff;
1019         size_t notesz;
1020
1021         ehoff = *off;
1022         *off += sizeof(Elf_Ehdr);
1023
1024         phoff = *off;
1025         *off += (numsegs + 1) * sizeof(Elf_Phdr);
1026
1027         noteoff = *off;
1028         elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
1029             sizeof *status);
1030         elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1031             sizeof *fpregset);
1032         elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1033             sizeof *psinfo);
1034         notesz = *off - noteoff;
1035
1036         /* Align up to a page boundary for the program segments. */
1037         *off = round_page(*off);
1038
1039         if (dst != NULL) {
1040                 Elf_Ehdr *ehdr;
1041                 Elf_Phdr *phdr;
1042                 struct phdr_closure phc;
1043
1044                 /*
1045                  * Fill in the ELF header.
1046                  */
1047                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1048                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1049                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1050                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1051                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1052                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1053                 ehdr->e_ident[EI_DATA] = ELF_DATA;
1054                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1055                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1056                 ehdr->e_ident[EI_ABIVERSION] = 0;
1057                 ehdr->e_ident[EI_PAD] = 0;
1058                 ehdr->e_type = ET_CORE;
1059                 ehdr->e_machine = ELF_ARCH;
1060                 ehdr->e_version = EV_CURRENT;
1061                 ehdr->e_entry = 0;
1062                 ehdr->e_phoff = phoff;
1063                 ehdr->e_flags = 0;
1064                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1065                 ehdr->e_phentsize = sizeof(Elf_Phdr);
1066                 ehdr->e_phnum = numsegs + 1;
1067                 ehdr->e_shentsize = sizeof(Elf_Shdr);
1068                 ehdr->e_shnum = 0;
1069                 ehdr->e_shstrndx = SHN_UNDEF;
1070
1071                 /*
1072                  * Fill in the program header entries.
1073                  */
1074                 phdr = (Elf_Phdr *)((char *)dst + phoff);
1075
1076                 /* The note segement. */
1077                 phdr->p_type = PT_NOTE;
1078                 phdr->p_offset = noteoff;
1079                 phdr->p_vaddr = 0;
1080                 phdr->p_paddr = 0;
1081                 phdr->p_filesz = notesz;
1082                 phdr->p_memsz = 0;
1083                 phdr->p_flags = 0;
1084                 phdr->p_align = 0;
1085                 phdr++;
1086
1087                 /* All the writable segments from the program. */
1088                 phc.phdr = phdr;
1089                 phc.offset = *off;
1090                 each_writable_segment(p, cb_put_phdr, &phc);
1091         }
1092 }
1093
1094 static void
1095 elf_putnote(void *dst, size_t *off, const char *name, int type,
1096     const void *desc, size_t descsz)
1097 {
1098         Elf_Note note;
1099
1100         note.n_namesz = strlen(name) + 1;
1101         note.n_descsz = descsz;
1102         note.n_type = type;
1103         if (dst != NULL)
1104                 bcopy(&note, (char *)dst + *off, sizeof note);
1105         *off += sizeof note;
1106         if (dst != NULL)
1107                 bcopy(name, (char *)dst + *off, note.n_namesz);
1108         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1109         if (dst != NULL)
1110                 bcopy(desc, (char *)dst + *off, note.n_descsz);
1111         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1112 }
1113
1114 /*
1115  * Tell kern_execve.c about it, with a little help from the linker.
1116  */
1117 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1118 EXEC_SET(elf, elf_execsw);