1 /*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 S�ren 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/sf_buf.h> 53 #include <sys/systm.h> 54 #include <sys/signalvar.h> 55 #include <sys/stat.h> 56 #include <sys/sx.h> 57 #include <sys/syscall.h> 58 #include <sys/sysctl.h> 59 #include <sys/sysent.h> 60 #include <sys/vnode.h> 61 62 #include <vm/vm.h> 63 #include <vm/vm_kern.h> 64 #include <vm/vm_param.h> 65 #include <vm/pmap.h> 66 #include <vm/vm_map.h> 67 #include <vm/vm_object.h> 68 #include <vm/vm_extern.h> 69 70 #include <machine/elf.h> 71 #include <machine/md_var.h> 72 73 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 74 #include <machine/fpu.h> 75 #include <compat/ia32/ia32_reg.h> 76 #endif 77 78 #define OLD_EI_BRAND 8 79 80 static int __elfN(check_header)(const Elf_Ehdr *hdr); 81 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 82 const char *interp, int32_t *osrel); 83 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 84 u_long *entry, size_t pagesize); 85 static int __elfN(load_section)(struct vmspace *vmspace, 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 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, 90 int32_t *osrel); 91 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 92 static boolean_t __elfN(check_note)(struct image_params *imgp, 93 Elf_Brandnote *checknote, int32_t *osrel); 94 95 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 96 ""); 97 98 int __elfN(fallback_brand) = -1; 99 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 100 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 101 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 102 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 103 &__elfN(fallback_brand)); 104 105 static int elf_legacy_coredump = 0; 106 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 107 &elf_legacy_coredump, 0, ""); 108 109 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 110 111 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 112 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 113 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 114 115 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 116 117 Elf_Brandnote __elfN(freebsd_brandnote) = { 118 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 119 .hdr.n_descsz = sizeof(int32_t), 120 .hdr.n_type = 1, 121 .vendor = FREEBSD_ABI_VENDOR, 122 .flags = BN_TRANSLATE_OSREL, 123 .trans_osrel = __elfN(freebsd_trans_osrel) 124 }; 125 126 static boolean_t 127 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 128 { 129 uintptr_t p; 130 131 p = (uintptr_t)(note + 1); 132 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 133 *osrel = *(const int32_t *)(p); 134 135 return (TRUE); 136 } 137 138 static const char GNU_ABI_VENDOR[] = "GNU"; 139 static int GNU_KFREEBSD_ABI_DESC = 3; 140 141 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 142 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 143 .hdr.n_descsz = 16, /* XXX at least 16 */ 144 .hdr.n_type = 1, 145 .vendor = GNU_ABI_VENDOR, 146 .flags = BN_TRANSLATE_OSREL, 147 .trans_osrel = kfreebsd_trans_osrel 148 }; 149 150 static boolean_t 151 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 152 { 153 const Elf32_Word *desc; 154 uintptr_t p; 155 156 p = (uintptr_t)(note + 1); 157 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 158 159 desc = (const Elf32_Word *)p; 160 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 161 return (FALSE); 162 163 /* 164 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 165 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 166 */ 167 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 168 169 return (TRUE); 170 } 171 172 int 173 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 174 { 175 int i; 176 177 for (i = 0; i < MAX_BRANDS; i++) { 178 if (elf_brand_list[i] == NULL) { 179 elf_brand_list[i] = entry; 180 break; 181 } 182 } 183 if (i == MAX_BRANDS) 184 return (-1); 185 return (0); 186 } 187 188 int 189 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 190 { 191 int i; 192 193 for (i = 0; i < MAX_BRANDS; i++) { 194 if (elf_brand_list[i] == entry) { 195 elf_brand_list[i] = NULL; 196 break; 197 } 198 } 199 if (i == MAX_BRANDS) 200 return (-1); 201 return (0); 202 } 203 204 int 205 __elfN(brand_inuse)(Elf_Brandinfo *entry) 206 { 207 struct proc *p; 208 int rval = FALSE; 209 210 sx_slock(&allproc_lock); 211 FOREACH_PROC_IN_SYSTEM(p) { 212 if (p->p_sysent == entry->sysvec) { 213 rval = TRUE; 214 break; 215 } 216 } 217 sx_sunlock(&allproc_lock); 218 219 return (rval); 220 } 221 222 static Elf_Brandinfo * 223 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 224 int32_t *osrel) 225 { 226 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 227 Elf_Brandinfo *bi; 228 boolean_t ret; 229 int i; 230 231 /* 232 * We support four types of branding -- (1) the ELF EI_OSABI field 233 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 234 * branding w/in the ELF header, (3) path of the `interp_path' 235 * field, and (4) the ".note.ABI-tag" ELF section. 236 */ 237 238 /* Look for an ".note.ABI-tag" ELF section */ 239 for (i = 0; i < MAX_BRANDS; i++) { 240 bi = elf_brand_list[i]; 241 if (bi == NULL) 242 continue; 243 if (hdr->e_machine == bi->machine && (bi->flags & 244 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 245 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 246 if (ret) 247 return (bi); 248 } 249 } 250 251 /* If the executable has a brand, search for it in the brand list. */ 252 for (i = 0; i < MAX_BRANDS; i++) { 253 bi = elf_brand_list[i]; 254 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 255 continue; 256 if (hdr->e_machine == bi->machine && 257 (hdr->e_ident[EI_OSABI] == bi->brand || 258 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 259 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 260 return (bi); 261 } 262 263 /* Lacking a known brand, search for a recognized interpreter. */ 264 if (interp != NULL) { 265 for (i = 0; i < MAX_BRANDS; i++) { 266 bi = elf_brand_list[i]; 267 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 268 continue; 269 if (hdr->e_machine == bi->machine && 270 strcmp(interp, bi->interp_path) == 0) 271 return (bi); 272 } 273 } 274 275 /* Lacking a recognized interpreter, try the default brand */ 276 for (i = 0; i < MAX_BRANDS; i++) { 277 bi = elf_brand_list[i]; 278 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 279 continue; 280 if (hdr->e_machine == bi->machine && 281 __elfN(fallback_brand) == bi->brand) 282 return (bi); 283 } 284 return (NULL); 285 } 286 287 static int 288 __elfN(check_header)(const Elf_Ehdr *hdr) 289 { 290 Elf_Brandinfo *bi; 291 int i; 292 293 if (!IS_ELF(*hdr) || 294 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 295 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 296 hdr->e_ident[EI_VERSION] != EV_CURRENT || 297 hdr->e_phentsize != sizeof(Elf_Phdr) || 298 hdr->e_version != ELF_TARG_VER) 299 return (ENOEXEC); 300 301 /* 302 * Make sure we have at least one brand for this machine. 303 */ 304 305 for (i = 0; i < MAX_BRANDS; i++) { 306 bi = elf_brand_list[i]; 307 if (bi != NULL && bi->machine == hdr->e_machine) 308 break; 309 } 310 if (i == MAX_BRANDS) 311 return (ENOEXEC); 312 313 return (0); 314 } 315 316 static int 317 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 318 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 319 { 320 struct sf_buf *sf; 321 int error; 322 vm_offset_t off; 323 324 /* 325 * Create the page if it doesn't exist yet. Ignore errors. 326 */ 327 vm_map_lock(map); 328 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 329 VM_PROT_ALL, VM_PROT_ALL, 0); 330 vm_map_unlock(map); 331 332 /* 333 * Find the page from the underlying object. 334 */ 335 if (object) { 336 sf = vm_imgact_map_page(object, offset); 337 if (sf == NULL) 338 return (KERN_FAILURE); 339 off = offset - trunc_page(offset); 340 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 341 end - start); 342 vm_imgact_unmap_page(sf); 343 if (error) { 344 return (KERN_FAILURE); 345 } 346 } 347 348 return (KERN_SUCCESS); 349 } 350 351 static int 352 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 353 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 354 { 355 struct sf_buf *sf; 356 vm_offset_t off; 357 vm_size_t sz; 358 int error, rv; 359 360 if (start != trunc_page(start)) { 361 rv = __elfN(map_partial)(map, object, offset, start, 362 round_page(start), prot); 363 if (rv) 364 return (rv); 365 offset += round_page(start) - start; 366 start = round_page(start); 367 } 368 if (end != round_page(end)) { 369 rv = __elfN(map_partial)(map, object, offset + 370 trunc_page(end) - start, trunc_page(end), end, prot); 371 if (rv) 372 return (rv); 373 end = trunc_page(end); 374 } 375 if (end > start) { 376 if (offset & PAGE_MASK) { 377 /* 378 * The mapping is not page aligned. This means we have 379 * to copy the data. Sigh. 380 */ 381 rv = vm_map_find(map, NULL, 0, &start, end - start, 382 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 383 if (rv) 384 return (rv); 385 if (object == NULL) 386 return (KERN_SUCCESS); 387 for (; start < end; start += sz) { 388 sf = vm_imgact_map_page(object, offset); 389 if (sf == NULL) 390 return (KERN_FAILURE); 391 off = offset - trunc_page(offset); 392 sz = end - start; 393 if (sz > PAGE_SIZE - off) 394 sz = PAGE_SIZE - off; 395 error = copyout((caddr_t)sf_buf_kva(sf) + off, 396 (caddr_t)start, sz); 397 vm_imgact_unmap_page(sf); 398 if (error) { 399 return (KERN_FAILURE); 400 } 401 offset += sz; 402 } 403 rv = KERN_SUCCESS; 404 } else { 405 vm_object_reference(object); 406 vm_map_lock(map); 407 rv = vm_map_insert(map, object, offset, start, end, 408 prot, VM_PROT_ALL, cow); 409 vm_map_unlock(map); 410 if (rv != KERN_SUCCESS) 411 vm_object_deallocate(object); 412 } 413 return (rv); 414 } else { 415 return (KERN_SUCCESS); 416 } 417 } 418 419 static int 420 __elfN(load_section)(struct vmspace *vmspace, 421 vm_object_t object, vm_offset_t offset, 422 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 423 size_t pagesize) 424 { 425 struct sf_buf *sf; 426 size_t map_len; 427 vm_offset_t map_addr; 428 int error, rv, cow; 429 size_t copy_len; 430 vm_offset_t file_addr; 431 432 /* 433 * It's necessary to fail if the filsz + offset taken from the 434 * header is greater than the actual file pager object's size. 435 * If we were to allow this, then the vm_map_find() below would 436 * walk right off the end of the file object and into the ether. 437 * 438 * While I'm here, might as well check for something else that 439 * is invalid: filsz cannot be greater than memsz. 440 */ 441 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 442 filsz > memsz) { 443 uprintf("elf_load_section: truncated ELF file\n"); 444 return (ENOEXEC); 445 } 446 447 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 448 file_addr = trunc_page_ps(offset, pagesize); 449 450 /* 451 * We have two choices. We can either clear the data in the last page 452 * of an oversized mapping, or we can start the anon mapping a page 453 * early and copy the initialized data into that first page. We 454 * choose the second.. 455 */ 456 if (memsz > filsz) 457 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 458 else 459 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 460 461 if (map_len != 0) { 462 /* cow flags: don't dump readonly sections in core */ 463 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 464 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 465 466 rv = __elfN(map_insert)(&vmspace->vm_map, 467 object, 468 file_addr, /* file offset */ 469 map_addr, /* virtual start */ 470 map_addr + map_len,/* virtual end */ 471 prot, 472 cow); 473 if (rv != KERN_SUCCESS) 474 return (EINVAL); 475 476 /* we can stop now if we've covered it all */ 477 if (memsz == filsz) { 478 return (0); 479 } 480 } 481 482 483 /* 484 * We have to get the remaining bit of the file into the first part 485 * of the oversized map segment. This is normally because the .data 486 * segment in the file is extended to provide bss. It's a neat idea 487 * to try and save a page, but it's a pain in the behind to implement. 488 */ 489 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 490 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 491 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 492 map_addr; 493 494 /* This had damn well better be true! */ 495 if (map_len != 0) { 496 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 497 map_addr + map_len, VM_PROT_ALL, 0); 498 if (rv != KERN_SUCCESS) { 499 return (EINVAL); 500 } 501 } 502 503 if (copy_len != 0) { 504 vm_offset_t off; 505 506 sf = vm_imgact_map_page(object, offset + filsz); 507 if (sf == NULL) 508 return (EIO); 509 510 /* send the page fragment to user space */ 511 off = trunc_page_ps(offset + filsz, pagesize) - 512 trunc_page(offset + filsz); 513 error = copyout((caddr_t)sf_buf_kva(sf) + off, 514 (caddr_t)map_addr, copy_len); 515 vm_imgact_unmap_page(sf); 516 if (error) { 517 return (error); 518 } 519 } 520 521 /* 522 * set it to the specified protection. 523 * XXX had better undo the damage from pasting over the cracks here! 524 */ 525 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 526 round_page(map_addr + map_len), prot, FALSE); 527 528 return (0); 529 } 530 531 /* 532 * Load the file "file" into memory. It may be either a shared object 533 * or an executable. 534 * 535 * The "addr" reference parameter is in/out. On entry, it specifies 536 * the address where a shared object should be loaded. If the file is 537 * an executable, this value is ignored. On exit, "addr" specifies 538 * where the file was actually loaded. 539 * 540 * The "entry" reference parameter is out only. On exit, it specifies 541 * the entry point for the loaded file. 542 */ 543 static int 544 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 545 u_long *entry, size_t pagesize) 546 { 547 struct { 548 struct nameidata nd; 549 struct vattr attr; 550 struct image_params image_params; 551 } *tempdata; 552 const Elf_Ehdr *hdr = NULL; 553 const Elf_Phdr *phdr = NULL; 554 struct nameidata *nd; 555 struct vmspace *vmspace = p->p_vmspace; 556 struct vattr *attr; 557 struct image_params *imgp; 558 vm_prot_t prot; 559 u_long rbase; 560 u_long base_addr = 0; 561 int vfslocked, error, i, numsegs; 562 563 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 564 nd = &tempdata->nd; 565 attr = &tempdata->attr; 566 imgp = &tempdata->image_params; 567 568 /* 569 * Initialize part of the common data 570 */ 571 imgp->proc = p; 572 imgp->attr = attr; 573 imgp->firstpage = NULL; 574 imgp->image_header = NULL; 575 imgp->object = NULL; 576 imgp->execlabel = NULL; 577 578 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, 579 curthread); 580 vfslocked = 0; 581 if ((error = namei(nd)) != 0) { 582 nd->ni_vp = NULL; 583 goto fail; 584 } 585 vfslocked = NDHASGIANT(nd); 586 NDFREE(nd, NDF_ONLY_PNBUF); 587 imgp->vp = nd->ni_vp; 588 589 /* 590 * Check permissions, modes, uid, etc on the file, and "open" it. 591 */ 592 error = exec_check_permissions(imgp); 593 if (error) 594 goto fail; 595 596 error = exec_map_first_page(imgp); 597 if (error) 598 goto fail; 599 600 /* 601 * Also make certain that the interpreter stays the same, so set 602 * its VV_TEXT flag, too. 603 */ 604 nd->ni_vp->v_vflag |= VV_TEXT; 605 606 imgp->object = nd->ni_vp->v_object; 607 608 hdr = (const Elf_Ehdr *)imgp->image_header; 609 if ((error = __elfN(check_header)(hdr)) != 0) 610 goto fail; 611 if (hdr->e_type == ET_DYN) 612 rbase = *addr; 613 else if (hdr->e_type == ET_EXEC) 614 rbase = 0; 615 else { 616 error = ENOEXEC; 617 goto fail; 618 } 619 620 /* Only support headers that fit within first page for now */ 621 /* (multiplication of two Elf_Half fields will not overflow) */ 622 if ((hdr->e_phoff > PAGE_SIZE) || 623 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 624 error = ENOEXEC; 625 goto fail; 626 } 627 628 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 629 if (!aligned(phdr, Elf_Addr)) { 630 error = ENOEXEC; 631 goto fail; 632 } 633 634 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 635 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 636 prot = 0; 637 if (phdr[i].p_flags & PF_X) 638 prot |= VM_PROT_EXECUTE; 639 if (phdr[i].p_flags & PF_W) 640 prot |= VM_PROT_WRITE; 641 if (phdr[i].p_flags & PF_R) 642 prot |= VM_PROT_READ; 643 644 if ((error = __elfN(load_section)(vmspace, 645 imgp->object, phdr[i].p_offset, 646 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 647 phdr[i].p_memsz, phdr[i].p_filesz, prot, 648 pagesize)) != 0) 649 goto fail; 650 /* 651 * Establish the base address if this is the 652 * first segment. 653 */ 654 if (numsegs == 0) 655 base_addr = trunc_page(phdr[i].p_vaddr + 656 rbase); 657 numsegs++; 658 } 659 } 660 *addr = base_addr; 661 *entry = (unsigned long)hdr->e_entry + rbase; 662 663 fail: 664 if (imgp->firstpage) 665 exec_unmap_first_page(imgp); 666 667 if (nd->ni_vp) 668 vput(nd->ni_vp); 669 670 VFS_UNLOCK_GIANT(vfslocked); 671 free(tempdata, M_TEMP); 672 673 return (error); 674 } 675 676 static int 677 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 678 { 679 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 680 const Elf_Phdr *phdr; 681 Elf_Auxargs *elf_auxargs; 682 struct vmspace *vmspace; 683 vm_prot_t prot; 684 u_long text_size = 0, data_size = 0, total_size = 0; 685 u_long text_addr = 0, data_addr = 0; 686 u_long seg_size, seg_addr; 687 u_long addr, entry = 0, proghdr = 0; 688 int32_t osrel = 0; 689 int error = 0, i; 690 const char *interp = NULL, *newinterp = NULL; 691 Elf_Brandinfo *brand_info; 692 char *path; 693 struct sysentvec *sv; 694 695 /* 696 * Do we have a valid ELF header ? 697 * 698 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 699 * if particular brand doesn't support it. 700 */ 701 if (__elfN(check_header)(hdr) != 0 || 702 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 703 return (-1); 704 705 /* 706 * From here on down, we return an errno, not -1, as we've 707 * detected an ELF file. 708 */ 709 710 if ((hdr->e_phoff > PAGE_SIZE) || 711 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 712 /* Only support headers in first page for now */ 713 return (ENOEXEC); 714 } 715 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 716 if (!aligned(phdr, Elf_Addr)) 717 return (ENOEXEC); 718 for (i = 0; i < hdr->e_phnum; i++) { 719 if (phdr[i].p_type == PT_INTERP) { 720 /* Path to interpreter */ 721 if (phdr[i].p_filesz > MAXPATHLEN || 722 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 723 return (ENOEXEC); 724 interp = imgp->image_header + phdr[i].p_offset; 725 break; 726 } 727 } 728 729 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel); 730 if (brand_info == NULL) { 731 uprintf("ELF binary type \"%u\" not known.\n", 732 hdr->e_ident[EI_OSABI]); 733 return (ENOEXEC); 734 } 735 if (hdr->e_type == ET_DYN && 736 (brand_info->flags & BI_CAN_EXEC_DYN) == 0) 737 return (ENOEXEC); 738 sv = brand_info->sysvec; 739 if (interp != NULL && brand_info->interp_newpath != NULL) 740 newinterp = brand_info->interp_newpath; 741 742 /* 743 * Avoid a possible deadlock if the current address space is destroyed 744 * and that address space maps the locked vnode. In the common case, 745 * the locked vnode's v_usecount is decremented but remains greater 746 * than zero. Consequently, the vnode lock is not needed by vrele(). 747 * However, in cases where the vnode lock is external, such as nullfs, 748 * v_usecount may become zero. 749 */ 750 VOP_UNLOCK(imgp->vp, 0); 751 752 error = exec_new_vmspace(imgp, sv); 753 imgp->proc->p_sysent = sv; 754 755 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 756 if (error) 757 return (error); 758 759 vmspace = imgp->proc->p_vmspace; 760 761 for (i = 0; i < hdr->e_phnum; i++) { 762 switch (phdr[i].p_type) { 763 case PT_LOAD: /* Loadable segment */ 764 prot = 0; 765 if (phdr[i].p_flags & PF_X) 766 prot |= VM_PROT_EXECUTE; 767 if (phdr[i].p_flags & PF_W) 768 prot |= VM_PROT_WRITE; 769 if (phdr[i].p_flags & PF_R) 770 prot |= VM_PROT_READ; 771 772 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 773 /* 774 * Some x86 binaries assume read == executable, 775 * notably the M3 runtime and therefore cvsup 776 */ 777 if (prot & VM_PROT_READ) 778 prot |= VM_PROT_EXECUTE; 779 #endif 780 781 if ((error = __elfN(load_section)(vmspace, 782 imgp->object, phdr[i].p_offset, 783 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 784 phdr[i].p_memsz, phdr[i].p_filesz, prot, 785 sv->sv_pagesize)) != 0) 786 return (error); 787 788 /* 789 * If this segment contains the program headers, 790 * remember their virtual address for the AT_PHDR 791 * aux entry. Static binaries don't usually include 792 * a PT_PHDR entry. 793 */ 794 if (phdr[i].p_offset == 0 && 795 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 796 <= phdr[i].p_filesz) 797 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 798 799 seg_addr = trunc_page(phdr[i].p_vaddr); 800 seg_size = round_page(phdr[i].p_memsz + 801 phdr[i].p_vaddr - seg_addr); 802 803 /* 804 * Is this .text or .data? We can't use 805 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 806 * alpha terribly and possibly does other bad 807 * things so we stick to the old way of figuring 808 * it out: If the segment contains the program 809 * entry point, it's a text segment, otherwise it 810 * is a data segment. 811 * 812 * Note that obreak() assumes that data_addr + 813 * data_size == end of data load area, and the ELF 814 * file format expects segments to be sorted by 815 * address. If multiple data segments exist, the 816 * last one will be used. 817 */ 818 if (hdr->e_entry >= phdr[i].p_vaddr && 819 hdr->e_entry < (phdr[i].p_vaddr + 820 phdr[i].p_memsz)) { 821 text_size = seg_size; 822 text_addr = seg_addr; 823 entry = (u_long)hdr->e_entry; 824 } else { 825 data_size = seg_size; 826 data_addr = seg_addr; 827 } 828 total_size += seg_size; 829 break; 830 case PT_PHDR: /* Program header table info */ 831 proghdr = phdr[i].p_vaddr; 832 break; 833 default: 834 break; 835 } 836 } 837 838 if (data_addr == 0 && data_size == 0) { 839 data_addr = text_addr; 840 data_size = text_size; 841 } 842 843 /* 844 * Check limits. It should be safe to check the 845 * limits after loading the segments since we do 846 * not actually fault in all the segments pages. 847 */ 848 PROC_LOCK(imgp->proc); 849 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 850 text_size > maxtsiz || 851 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) { 852 PROC_UNLOCK(imgp->proc); 853 return (ENOMEM); 854 } 855 856 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 857 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 858 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 859 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 860 861 /* 862 * We load the dynamic linker where a userland call 863 * to mmap(0, ...) would put it. The rationale behind this 864 * calculation is that it leaves room for the heap to grow to 865 * its maximum allowed size. 866 */ 867 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 868 lim_max(imgp->proc, RLIMIT_DATA)); 869 PROC_UNLOCK(imgp->proc); 870 871 imgp->entry_addr = entry; 872 873 if (interp != NULL) { 874 int have_interp = FALSE; 875 VOP_UNLOCK(imgp->vp, 0); 876 if (brand_info->emul_path != NULL && 877 brand_info->emul_path[0] != '\0') { 878 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 879 snprintf(path, MAXPATHLEN, "%s%s", 880 brand_info->emul_path, interp); 881 error = __elfN(load_file)(imgp->proc, path, &addr, 882 &imgp->entry_addr, sv->sv_pagesize); 883 free(path, M_TEMP); 884 if (error == 0) 885 have_interp = TRUE; 886 } 887 if (!have_interp && newinterp != NULL) { 888 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 889 &imgp->entry_addr, sv->sv_pagesize); 890 if (error == 0) 891 have_interp = TRUE; 892 } 893 if (!have_interp) { 894 error = __elfN(load_file)(imgp->proc, interp, &addr, 895 &imgp->entry_addr, sv->sv_pagesize); 896 } 897 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 898 if (error != 0) { 899 uprintf("ELF interpreter %s not found\n", interp); 900 return (error); 901 } 902 } else 903 addr = 0; 904 905 /* 906 * Construct auxargs table (used by the fixup routine) 907 */ 908 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 909 elf_auxargs->execfd = -1; 910 elf_auxargs->phdr = proghdr; 911 elf_auxargs->phent = hdr->e_phentsize; 912 elf_auxargs->phnum = hdr->e_phnum; 913 elf_auxargs->pagesz = PAGE_SIZE; 914 elf_auxargs->base = addr; 915 elf_auxargs->flags = 0; 916 elf_auxargs->entry = entry; 917 918 imgp->auxargs = elf_auxargs; 919 imgp->interpreted = 0; 920 imgp->proc->p_osrel = osrel; 921 922 return (error); 923 } 924 925 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 926 927 int 928 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 929 { 930 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 931 Elf_Addr *base; 932 Elf_Addr *pos; 933 934 base = (Elf_Addr *)*stack_base; 935 pos = base + (imgp->args->argc + imgp->args->envc + 2); 936 937 if (args->execfd != -1) 938 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 939 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 940 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 941 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 942 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 943 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 944 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 945 AUXARGS_ENTRY(pos, AT_BASE, args->base); 946 if (imgp->execpathp != 0) 947 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 948 AUXARGS_ENTRY(pos, AT_NULL, 0); 949 950 free(imgp->auxargs, M_TEMP); 951 imgp->auxargs = NULL; 952 953 base--; 954 suword(base, (long)imgp->args->argc); 955 *stack_base = (register_t *)base; 956 return (0); 957 } 958 959 /* 960 * Code for generating ELF core dumps. 961 */ 962 963 typedef void (*segment_callback)(vm_map_entry_t, void *); 964 965 /* Closure for cb_put_phdr(). */ 966 struct phdr_closure { 967 Elf_Phdr *phdr; /* Program header to fill in */ 968 Elf_Off offset; /* Offset of segment in core file */ 969 }; 970 971 /* Closure for cb_size_segment(). */ 972 struct sseg_closure { 973 int count; /* Count of writable segments. */ 974 size_t size; /* Total size of all writable segments. */ 975 }; 976 977 static void cb_put_phdr(vm_map_entry_t, void *); 978 static void cb_size_segment(vm_map_entry_t, void *); 979 static void each_writable_segment(struct thread *, segment_callback, void *); 980 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 981 int, void *, size_t); 982 static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 983 static void __elfN(putnote)(void *, size_t *, const char *, int, 984 const void *, size_t); 985 986 int 987 __elfN(coredump)(td, vp, limit) 988 struct thread *td; 989 struct vnode *vp; 990 off_t limit; 991 { 992 struct ucred *cred = td->td_ucred; 993 int error = 0; 994 struct sseg_closure seginfo; 995 void *hdr; 996 size_t hdrsize; 997 998 /* Size the program segments. */ 999 seginfo.count = 0; 1000 seginfo.size = 0; 1001 each_writable_segment(td, cb_size_segment, &seginfo); 1002 1003 /* 1004 * Calculate the size of the core file header area by making 1005 * a dry run of generating it. Nothing is written, but the 1006 * size is calculated. 1007 */ 1008 hdrsize = 0; 1009 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 1010 1011 if (hdrsize + seginfo.size >= limit) 1012 return (EFAULT); 1013 1014 /* 1015 * Allocate memory for building the header, fill it up, 1016 * and write it out. 1017 */ 1018 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1019 if (hdr == NULL) { 1020 return (EINVAL); 1021 } 1022 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 1023 1024 /* Write the contents of all of the writable segments. */ 1025 if (error == 0) { 1026 Elf_Phdr *php; 1027 off_t offset; 1028 int i; 1029 1030 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1031 offset = hdrsize; 1032 for (i = 0; i < seginfo.count; i++) { 1033 error = vn_rdwr_inchunks(UIO_WRITE, vp, 1034 (caddr_t)(uintptr_t)php->p_vaddr, 1035 php->p_filesz, offset, UIO_USERSPACE, 1036 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1037 curthread); 1038 if (error != 0) 1039 break; 1040 offset += php->p_filesz; 1041 php++; 1042 } 1043 } 1044 free(hdr, M_TEMP); 1045 1046 return (error); 1047 } 1048 1049 /* 1050 * A callback for each_writable_segment() to write out the segment's 1051 * program header entry. 1052 */ 1053 static void 1054 cb_put_phdr(entry, closure) 1055 vm_map_entry_t entry; 1056 void *closure; 1057 { 1058 struct phdr_closure *phc = (struct phdr_closure *)closure; 1059 Elf_Phdr *phdr = phc->phdr; 1060 1061 phc->offset = round_page(phc->offset); 1062 1063 phdr->p_type = PT_LOAD; 1064 phdr->p_offset = phc->offset; 1065 phdr->p_vaddr = entry->start; 1066 phdr->p_paddr = 0; 1067 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1068 phdr->p_align = PAGE_SIZE; 1069 phdr->p_flags = 0; 1070 if (entry->protection & VM_PROT_READ) 1071 phdr->p_flags |= PF_R; 1072 if (entry->protection & VM_PROT_WRITE) 1073 phdr->p_flags |= PF_W; 1074 if (entry->protection & VM_PROT_EXECUTE) 1075 phdr->p_flags |= PF_X; 1076 1077 phc->offset += phdr->p_filesz; 1078 phc->phdr++; 1079 } 1080 1081 /* 1082 * A callback for each_writable_segment() to gather information about 1083 * the number of segments and their total size. 1084 */ 1085 static void 1086 cb_size_segment(entry, closure) 1087 vm_map_entry_t entry; 1088 void *closure; 1089 { 1090 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1091 1092 ssc->count++; 1093 ssc->size += entry->end - entry->start; 1094 } 1095 1096 /* 1097 * For each writable segment in the process's memory map, call the given 1098 * function with a pointer to the map entry and some arbitrary 1099 * caller-supplied data. 1100 */ 1101 static void 1102 each_writable_segment(td, func, closure) 1103 struct thread *td; 1104 segment_callback func; 1105 void *closure; 1106 { 1107 struct proc *p = td->td_proc; 1108 vm_map_t map = &p->p_vmspace->vm_map; 1109 vm_map_entry_t entry; 1110 vm_object_t backing_object, object; 1111 boolean_t ignore_entry; 1112 1113 vm_map_lock_read(map); 1114 for (entry = map->header.next; entry != &map->header; 1115 entry = entry->next) { 1116 /* 1117 * Don't dump inaccessible mappings, deal with legacy 1118 * coredump mode. 1119 * 1120 * Note that read-only segments related to the elf binary 1121 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1122 * need to arbitrarily ignore such segments. 1123 */ 1124 if (elf_legacy_coredump) { 1125 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1126 continue; 1127 } else { 1128 if ((entry->protection & VM_PROT_ALL) == 0) 1129 continue; 1130 } 1131 1132 /* 1133 * Dont include memory segment in the coredump if 1134 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1135 * madvise(2). Do not dump submaps (i.e. parts of the 1136 * kernel map). 1137 */ 1138 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1139 continue; 1140 1141 if ((object = entry->object.vm_object) == NULL) 1142 continue; 1143 1144 /* Ignore memory-mapped devices and such things. */ 1145 VM_OBJECT_LOCK(object); 1146 while ((backing_object = object->backing_object) != NULL) { 1147 VM_OBJECT_LOCK(backing_object); 1148 VM_OBJECT_UNLOCK(object); 1149 object = backing_object; 1150 } 1151 ignore_entry = object->type != OBJT_DEFAULT && 1152 object->type != OBJT_SWAP && object->type != OBJT_VNODE; 1153 VM_OBJECT_UNLOCK(object); 1154 if (ignore_entry) 1155 continue; 1156 1157 (*func)(entry, closure); 1158 } 1159 vm_map_unlock_read(map); 1160 } 1161 1162 /* 1163 * Write the core file header to the file, including padding up to 1164 * the page boundary. 1165 */ 1166 static int 1167 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1168 struct thread *td; 1169 struct vnode *vp; 1170 struct ucred *cred; 1171 int numsegs; 1172 size_t hdrsize; 1173 void *hdr; 1174 { 1175 size_t off; 1176 1177 /* Fill in the header. */ 1178 bzero(hdr, hdrsize); 1179 off = 0; 1180 __elfN(puthdr)(td, hdr, &off, numsegs); 1181 1182 /* Write it to the core file. */ 1183 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1184 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1185 td)); 1186 } 1187 1188 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1189 typedef struct prstatus32 elf_prstatus_t; 1190 typedef struct prpsinfo32 elf_prpsinfo_t; 1191 typedef struct fpreg32 elf_prfpregset_t; 1192 typedef struct fpreg32 elf_fpregset_t; 1193 typedef struct reg32 elf_gregset_t; 1194 #else 1195 typedef prstatus_t elf_prstatus_t; 1196 typedef prpsinfo_t elf_prpsinfo_t; 1197 typedef prfpregset_t elf_prfpregset_t; 1198 typedef prfpregset_t elf_fpregset_t; 1199 typedef gregset_t elf_gregset_t; 1200 #endif 1201 1202 static void 1203 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1204 { 1205 struct { 1206 elf_prstatus_t status; 1207 elf_prfpregset_t fpregset; 1208 elf_prpsinfo_t psinfo; 1209 } *tempdata; 1210 elf_prstatus_t *status; 1211 elf_prfpregset_t *fpregset; 1212 elf_prpsinfo_t *psinfo; 1213 struct proc *p; 1214 struct thread *thr; 1215 size_t ehoff, noteoff, notesz, phoff; 1216 1217 p = td->td_proc; 1218 1219 ehoff = *off; 1220 *off += sizeof(Elf_Ehdr); 1221 1222 phoff = *off; 1223 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1224 1225 noteoff = *off; 1226 /* 1227 * Don't allocate space for the notes if we're just calculating 1228 * the size of the header. We also don't collect the data. 1229 */ 1230 if (dst != NULL) { 1231 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1232 status = &tempdata->status; 1233 fpregset = &tempdata->fpregset; 1234 psinfo = &tempdata->psinfo; 1235 } else { 1236 tempdata = NULL; 1237 status = NULL; 1238 fpregset = NULL; 1239 psinfo = NULL; 1240 } 1241 1242 if (dst != NULL) { 1243 psinfo->pr_version = PRPSINFO_VERSION; 1244 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1245 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1246 /* 1247 * XXX - We don't fill in the command line arguments properly 1248 * yet. 1249 */ 1250 strlcpy(psinfo->pr_psargs, p->p_comm, 1251 sizeof(psinfo->pr_psargs)); 1252 } 1253 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1254 sizeof *psinfo); 1255 1256 /* 1257 * To have the debugger select the right thread (LWP) as the initial 1258 * thread, we dump the state of the thread passed to us in td first. 1259 * This is the thread that causes the core dump and thus likely to 1260 * be the right thread one wants to have selected in the debugger. 1261 */ 1262 thr = td; 1263 while (thr != NULL) { 1264 if (dst != NULL) { 1265 status->pr_version = PRSTATUS_VERSION; 1266 status->pr_statussz = sizeof(elf_prstatus_t); 1267 status->pr_gregsetsz = sizeof(elf_gregset_t); 1268 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1269 status->pr_osreldate = osreldate; 1270 status->pr_cursig = p->p_sig; 1271 status->pr_pid = thr->td_tid; 1272 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1273 fill_regs32(thr, &status->pr_reg); 1274 fill_fpregs32(thr, fpregset); 1275 #else 1276 fill_regs(thr, &status->pr_reg); 1277 fill_fpregs(thr, fpregset); 1278 #endif 1279 } 1280 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1281 sizeof *status); 1282 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1283 sizeof *fpregset); 1284 /* 1285 * Allow for MD specific notes, as well as any MD 1286 * specific preparations for writing MI notes. 1287 */ 1288 __elfN(dump_thread)(thr, dst, off); 1289 1290 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1291 TAILQ_NEXT(thr, td_plist); 1292 if (thr == td) 1293 thr = TAILQ_NEXT(thr, td_plist); 1294 } 1295 1296 notesz = *off - noteoff; 1297 1298 if (dst != NULL) 1299 free(tempdata, M_TEMP); 1300 1301 /* Align up to a page boundary for the program segments. */ 1302 *off = round_page(*off); 1303 1304 if (dst != NULL) { 1305 Elf_Ehdr *ehdr; 1306 Elf_Phdr *phdr; 1307 struct phdr_closure phc; 1308 1309 /* 1310 * Fill in the ELF header. 1311 */ 1312 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1313 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1314 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1315 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1316 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1317 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1318 ehdr->e_ident[EI_DATA] = ELF_DATA; 1319 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1320 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1321 ehdr->e_ident[EI_ABIVERSION] = 0; 1322 ehdr->e_ident[EI_PAD] = 0; 1323 ehdr->e_type = ET_CORE; 1324 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1325 ehdr->e_machine = EM_386; 1326 #else 1327 ehdr->e_machine = ELF_ARCH; 1328 #endif 1329 ehdr->e_version = EV_CURRENT; 1330 ehdr->e_entry = 0; 1331 ehdr->e_phoff = phoff; 1332 ehdr->e_flags = 0; 1333 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1334 ehdr->e_phentsize = sizeof(Elf_Phdr); 1335 ehdr->e_phnum = numsegs + 1; 1336 ehdr->e_shentsize = sizeof(Elf_Shdr); 1337 ehdr->e_shnum = 0; 1338 ehdr->e_shstrndx = SHN_UNDEF; 1339 1340 /* 1341 * Fill in the program header entries. 1342 */ 1343 phdr = (Elf_Phdr *)((char *)dst + phoff); 1344 1345 /* The note segement. */ 1346 phdr->p_type = PT_NOTE; 1347 phdr->p_offset = noteoff; 1348 phdr->p_vaddr = 0; 1349 phdr->p_paddr = 0; 1350 phdr->p_filesz = notesz; 1351 phdr->p_memsz = 0; 1352 phdr->p_flags = 0; 1353 phdr->p_align = 0; 1354 phdr++; 1355 1356 /* All the writable segments from the program. */ 1357 phc.phdr = phdr; 1358 phc.offset = *off; 1359 each_writable_segment(td, cb_put_phdr, &phc); 1360 } 1361 } 1362 1363 static void 1364 __elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1365 const void *desc, size_t descsz) 1366 { 1367 Elf_Note note; 1368 1369 note.n_namesz = strlen(name) + 1; 1370 note.n_descsz = descsz; 1371 note.n_type = type; 1372 if (dst != NULL) 1373 bcopy(¬e, (char *)dst + *off, sizeof note); 1374 *off += sizeof note; 1375 if (dst != NULL) 1376 bcopy(name, (char *)dst + *off, note.n_namesz); 1377 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1378 if (dst != NULL) 1379 bcopy(desc, (char *)dst + *off, note.n_descsz); 1380 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1381 } 1382 1383 /* 1384 * Try to find the appropriate ABI-note section for checknote, 1385 * fetch the osreldate for binary from the ELF OSABI-note. Only the 1386 * first page of the image is searched, the same as for headers. 1387 */ 1388 static boolean_t 1389 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 1390 int32_t *osrel) 1391 { 1392 const Elf_Note *note, *note0, *note_end; 1393 const Elf_Phdr *phdr, *pnote; 1394 const Elf_Ehdr *hdr; 1395 const char *note_name; 1396 int i; 1397 1398 pnote = NULL; 1399 hdr = (const Elf_Ehdr *)imgp->image_header; 1400 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1401 1402 for (i = 0; i < hdr->e_phnum; i++) { 1403 if (phdr[i].p_type == PT_NOTE) { 1404 pnote = &phdr[i]; 1405 break; 1406 } 1407 } 1408 1409 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE || 1410 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE) 1411 return (FALSE); 1412 1413 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 1414 note_end = (const Elf_Note *)(imgp->image_header + 1415 pnote->p_offset + pnote->p_filesz); 1416 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 1417 if (!aligned(note, Elf32_Addr)) 1418 return (FALSE); 1419 if (note->n_namesz != checknote->hdr.n_namesz || 1420 note->n_descsz != checknote->hdr.n_descsz || 1421 note->n_type != checknote->hdr.n_type) 1422 goto nextnote; 1423 note_name = (const char *)(note + 1); 1424 if (strncmp(checknote->vendor, note_name, 1425 checknote->hdr.n_namesz) != 0) 1426 goto nextnote; 1427 1428 /* 1429 * Fetch the osreldate for binary 1430 * from the ELF OSABI-note if necessary. 1431 */ 1432 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 1433 checknote->trans_osrel != NULL) 1434 return (checknote->trans_osrel(note, osrel)); 1435 return (TRUE); 1436 1437 nextnote: 1438 note = (const Elf_Note *)((const char *)(note + 1) + 1439 roundup2(note->n_namesz, sizeof(Elf32_Addr)) + 1440 roundup2(note->n_descsz, sizeof(Elf32_Addr))); 1441 } 1442 1443 return (FALSE); 1444 } 1445 1446 /* 1447 * Tell kern_execve.c about it, with a little help from the linker. 1448 */ 1449 static struct execsw __elfN(execsw) = { 1450 __CONCAT(exec_, __elfN(imgact)), 1451 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1452 }; 1453 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1454