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_capsicum.h" 35 #include "opt_compat.h" 36 #include "opt_core.h" 37 38 #include <sys/param.h> 39 #include <sys/capability.h> 40 #include <sys/exec.h> 41 #include <sys/fcntl.h> 42 #include <sys/imgact.h> 43 #include <sys/imgact_elf.h> 44 #include <sys/kernel.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mutex.h> 49 #include <sys/mman.h> 50 #include <sys/namei.h> 51 #include <sys/pioctl.h> 52 #include <sys/proc.h> 53 #include <sys/procfs.h> 54 #include <sys/racct.h> 55 #include <sys/resourcevar.h> 56 #include <sys/sf_buf.h> 57 #include <sys/smp.h> 58 #include <sys/systm.h> 59 #include <sys/signalvar.h> 60 #include <sys/stat.h> 61 #include <sys/sx.h> 62 #include <sys/syscall.h> 63 #include <sys/sysctl.h> 64 #include <sys/sysent.h> 65 #include <sys/vnode.h> 66 #include <sys/syslog.h> 67 #include <sys/eventhandler.h> 68 69 #include <net/zlib.h> 70 71 #include <vm/vm.h> 72 #include <vm/vm_kern.h> 73 #include <vm/vm_param.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_map.h> 76 #include <vm/vm_object.h> 77 #include <vm/vm_extern.h> 78 79 #include <machine/elf.h> 80 #include <machine/md_var.h> 81 82 #define OLD_EI_BRAND 8 83 84 static int __elfN(check_header)(const Elf_Ehdr *hdr); 85 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 86 const char *interp, int interp_name_len, int32_t *osrel); 87 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 88 u_long *entry, size_t pagesize); 89 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, 90 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 91 size_t pagesize); 92 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 93 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, 94 int32_t *osrel); 95 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 96 static boolean_t __elfN(check_note)(struct image_params *imgp, 97 Elf_Brandnote *checknote, int32_t *osrel); 98 static vm_prot_t __elfN(trans_prot)(Elf_Word); 99 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 100 101 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 102 ""); 103 104 #ifdef COMPRESS_USER_CORES 105 static int compress_core(gzFile, char *, char *, unsigned int, 106 struct thread * td); 107 #define CORE_BUF_SIZE (16 * 1024) 108 #endif 109 110 int __elfN(fallback_brand) = -1; 111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 112 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 114 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 115 &__elfN(fallback_brand)); 116 117 static int elf_legacy_coredump = 0; 118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 119 &elf_legacy_coredump, 0, ""); 120 121 int __elfN(nxstack) = 122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ 123 1; 124 #else 125 0; 126 #endif 127 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 128 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 129 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); 130 131 #if __ELF_WORD_SIZE == 32 132 #if defined(__amd64__) || defined(__ia64__) 133 int i386_read_exec = 0; 134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 135 "enable execution from readable segments"); 136 #endif 137 #endif 138 139 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 140 141 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 142 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 143 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 144 145 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 146 147 Elf_Brandnote __elfN(freebsd_brandnote) = { 148 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 149 .hdr.n_descsz = sizeof(int32_t), 150 .hdr.n_type = 1, 151 .vendor = FREEBSD_ABI_VENDOR, 152 .flags = BN_TRANSLATE_OSREL, 153 .trans_osrel = __elfN(freebsd_trans_osrel) 154 }; 155 156 static boolean_t 157 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 158 { 159 uintptr_t p; 160 161 p = (uintptr_t)(note + 1); 162 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 163 *osrel = *(const int32_t *)(p); 164 165 return (TRUE); 166 } 167 168 static const char GNU_ABI_VENDOR[] = "GNU"; 169 static int GNU_KFREEBSD_ABI_DESC = 3; 170 171 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 172 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 173 .hdr.n_descsz = 16, /* XXX at least 16 */ 174 .hdr.n_type = 1, 175 .vendor = GNU_ABI_VENDOR, 176 .flags = BN_TRANSLATE_OSREL, 177 .trans_osrel = kfreebsd_trans_osrel 178 }; 179 180 static boolean_t 181 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 182 { 183 const Elf32_Word *desc; 184 uintptr_t p; 185 186 p = (uintptr_t)(note + 1); 187 p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); 188 189 desc = (const Elf32_Word *)p; 190 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 191 return (FALSE); 192 193 /* 194 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 195 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 196 */ 197 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 198 199 return (TRUE); 200 } 201 202 int 203 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 204 { 205 int i; 206 207 for (i = 0; i < MAX_BRANDS; i++) { 208 if (elf_brand_list[i] == NULL) { 209 elf_brand_list[i] = entry; 210 break; 211 } 212 } 213 if (i == MAX_BRANDS) { 214 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 215 __func__, entry); 216 return (-1); 217 } 218 return (0); 219 } 220 221 int 222 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 223 { 224 int i; 225 226 for (i = 0; i < MAX_BRANDS; i++) { 227 if (elf_brand_list[i] == entry) { 228 elf_brand_list[i] = NULL; 229 break; 230 } 231 } 232 if (i == MAX_BRANDS) 233 return (-1); 234 return (0); 235 } 236 237 int 238 __elfN(brand_inuse)(Elf_Brandinfo *entry) 239 { 240 struct proc *p; 241 int rval = FALSE; 242 243 sx_slock(&allproc_lock); 244 FOREACH_PROC_IN_SYSTEM(p) { 245 if (p->p_sysent == entry->sysvec) { 246 rval = TRUE; 247 break; 248 } 249 } 250 sx_sunlock(&allproc_lock); 251 252 return (rval); 253 } 254 255 static Elf_Brandinfo * 256 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 257 int interp_name_len, int32_t *osrel) 258 { 259 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 260 Elf_Brandinfo *bi; 261 boolean_t ret; 262 int i; 263 264 /* 265 * We support four types of branding -- (1) the ELF EI_OSABI field 266 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 267 * branding w/in the ELF header, (3) path of the `interp_path' 268 * field, and (4) the ".note.ABI-tag" ELF section. 269 */ 270 271 /* Look for an ".note.ABI-tag" ELF section */ 272 for (i = 0; i < MAX_BRANDS; i++) { 273 bi = elf_brand_list[i]; 274 if (bi == NULL) 275 continue; 276 if (hdr->e_machine == bi->machine && (bi->flags & 277 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 278 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 279 if (ret) 280 return (bi); 281 } 282 } 283 284 /* If the executable has a brand, search for it in the brand list. */ 285 for (i = 0; i < MAX_BRANDS; i++) { 286 bi = elf_brand_list[i]; 287 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 288 continue; 289 if (hdr->e_machine == bi->machine && 290 (hdr->e_ident[EI_OSABI] == bi->brand || 291 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 292 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 293 return (bi); 294 } 295 296 /* Lacking a known brand, search for a recognized interpreter. */ 297 if (interp != NULL) { 298 for (i = 0; i < MAX_BRANDS; i++) { 299 bi = elf_brand_list[i]; 300 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 301 continue; 302 if (hdr->e_machine == bi->machine && 303 /* ELF image p_filesz includes terminating zero */ 304 strlen(bi->interp_path) + 1 == interp_name_len && 305 strncmp(interp, bi->interp_path, interp_name_len) 306 == 0) 307 return (bi); 308 } 309 } 310 311 /* Lacking a recognized interpreter, try the default brand */ 312 for (i = 0; i < MAX_BRANDS; i++) { 313 bi = elf_brand_list[i]; 314 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 315 continue; 316 if (hdr->e_machine == bi->machine && 317 __elfN(fallback_brand) == bi->brand) 318 return (bi); 319 } 320 return (NULL); 321 } 322 323 static int 324 __elfN(check_header)(const Elf_Ehdr *hdr) 325 { 326 Elf_Brandinfo *bi; 327 int i; 328 329 if (!IS_ELF(*hdr) || 330 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 331 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 332 hdr->e_ident[EI_VERSION] != EV_CURRENT || 333 hdr->e_phentsize != sizeof(Elf_Phdr) || 334 hdr->e_version != ELF_TARG_VER) 335 return (ENOEXEC); 336 337 /* 338 * Make sure we have at least one brand for this machine. 339 */ 340 341 for (i = 0; i < MAX_BRANDS; i++) { 342 bi = elf_brand_list[i]; 343 if (bi != NULL && bi->machine == hdr->e_machine) 344 break; 345 } 346 if (i == MAX_BRANDS) 347 return (ENOEXEC); 348 349 return (0); 350 } 351 352 static int 353 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 354 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 355 { 356 struct sf_buf *sf; 357 int error; 358 vm_offset_t off; 359 360 /* 361 * Create the page if it doesn't exist yet. Ignore errors. 362 */ 363 vm_map_lock(map); 364 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 365 VM_PROT_ALL, VM_PROT_ALL, 0); 366 vm_map_unlock(map); 367 368 /* 369 * Find the page from the underlying object. 370 */ 371 if (object) { 372 sf = vm_imgact_map_page(object, offset); 373 if (sf == NULL) 374 return (KERN_FAILURE); 375 off = offset - trunc_page(offset); 376 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 377 end - start); 378 vm_imgact_unmap_page(sf); 379 if (error) { 380 return (KERN_FAILURE); 381 } 382 } 383 384 return (KERN_SUCCESS); 385 } 386 387 static int 388 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 389 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 390 { 391 struct sf_buf *sf; 392 vm_offset_t off; 393 vm_size_t sz; 394 int error, rv; 395 396 if (start != trunc_page(start)) { 397 rv = __elfN(map_partial)(map, object, offset, start, 398 round_page(start), prot); 399 if (rv) 400 return (rv); 401 offset += round_page(start) - start; 402 start = round_page(start); 403 } 404 if (end != round_page(end)) { 405 rv = __elfN(map_partial)(map, object, offset + 406 trunc_page(end) - start, trunc_page(end), end, prot); 407 if (rv) 408 return (rv); 409 end = trunc_page(end); 410 } 411 if (end > start) { 412 if (offset & PAGE_MASK) { 413 /* 414 * The mapping is not page aligned. This means we have 415 * to copy the data. Sigh. 416 */ 417 rv = vm_map_find(map, NULL, 0, &start, end - start, 418 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 419 if (rv) 420 return (rv); 421 if (object == NULL) 422 return (KERN_SUCCESS); 423 for (; start < end; start += sz) { 424 sf = vm_imgact_map_page(object, offset); 425 if (sf == NULL) 426 return (KERN_FAILURE); 427 off = offset - trunc_page(offset); 428 sz = end - start; 429 if (sz > PAGE_SIZE - off) 430 sz = PAGE_SIZE - off; 431 error = copyout((caddr_t)sf_buf_kva(sf) + off, 432 (caddr_t)start, sz); 433 vm_imgact_unmap_page(sf); 434 if (error) { 435 return (KERN_FAILURE); 436 } 437 offset += sz; 438 } 439 rv = KERN_SUCCESS; 440 } else { 441 vm_object_reference(object); 442 vm_map_lock(map); 443 rv = vm_map_insert(map, object, offset, start, end, 444 prot, VM_PROT_ALL, cow); 445 vm_map_unlock(map); 446 if (rv != KERN_SUCCESS) 447 vm_object_deallocate(object); 448 } 449 return (rv); 450 } else { 451 return (KERN_SUCCESS); 452 } 453 } 454 455 static int 456 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, 457 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 458 size_t pagesize) 459 { 460 struct sf_buf *sf; 461 size_t map_len; 462 vm_map_t map; 463 vm_object_t object; 464 vm_offset_t map_addr; 465 int error, rv, cow; 466 size_t copy_len; 467 vm_offset_t file_addr; 468 469 /* 470 * It's necessary to fail if the filsz + offset taken from the 471 * header is greater than the actual file pager object's size. 472 * If we were to allow this, then the vm_map_find() below would 473 * walk right off the end of the file object and into the ether. 474 * 475 * While I'm here, might as well check for something else that 476 * is invalid: filsz cannot be greater than memsz. 477 */ 478 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) { 479 uprintf("elf_load_section: truncated ELF file\n"); 480 return (ENOEXEC); 481 } 482 483 object = imgp->object; 484 map = &imgp->proc->p_vmspace->vm_map; 485 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 486 file_addr = trunc_page_ps(offset, pagesize); 487 488 /* 489 * We have two choices. We can either clear the data in the last page 490 * of an oversized mapping, or we can start the anon mapping a page 491 * early and copy the initialized data into that first page. We 492 * choose the second.. 493 */ 494 if (memsz > filsz) 495 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 496 else 497 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 498 499 if (map_len != 0) { 500 /* cow flags: don't dump readonly sections in core */ 501 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 502 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 503 504 rv = __elfN(map_insert)(map, 505 object, 506 file_addr, /* file offset */ 507 map_addr, /* virtual start */ 508 map_addr + map_len,/* virtual end */ 509 prot, 510 cow); 511 if (rv != KERN_SUCCESS) 512 return (EINVAL); 513 514 /* we can stop now if we've covered it all */ 515 if (memsz == filsz) { 516 return (0); 517 } 518 } 519 520 521 /* 522 * We have to get the remaining bit of the file into the first part 523 * of the oversized map segment. This is normally because the .data 524 * segment in the file is extended to provide bss. It's a neat idea 525 * to try and save a page, but it's a pain in the behind to implement. 526 */ 527 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 528 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 529 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 530 map_addr; 531 532 /* This had damn well better be true! */ 533 if (map_len != 0) { 534 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr + 535 map_len, VM_PROT_ALL, 0); 536 if (rv != KERN_SUCCESS) { 537 return (EINVAL); 538 } 539 } 540 541 if (copy_len != 0) { 542 vm_offset_t off; 543 544 sf = vm_imgact_map_page(object, offset + filsz); 545 if (sf == NULL) 546 return (EIO); 547 548 /* send the page fragment to user space */ 549 off = trunc_page_ps(offset + filsz, pagesize) - 550 trunc_page(offset + filsz); 551 error = copyout((caddr_t)sf_buf_kva(sf) + off, 552 (caddr_t)map_addr, copy_len); 553 vm_imgact_unmap_page(sf); 554 if (error) { 555 return (error); 556 } 557 } 558 559 /* 560 * set it to the specified protection. 561 * XXX had better undo the damage from pasting over the cracks here! 562 */ 563 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 564 map_len), prot, FALSE); 565 566 return (0); 567 } 568 569 /* 570 * Load the file "file" into memory. It may be either a shared object 571 * or an executable. 572 * 573 * The "addr" reference parameter is in/out. On entry, it specifies 574 * the address where a shared object should be loaded. If the file is 575 * an executable, this value is ignored. On exit, "addr" specifies 576 * where the file was actually loaded. 577 * 578 * The "entry" reference parameter is out only. On exit, it specifies 579 * the entry point for the loaded file. 580 */ 581 static int 582 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 583 u_long *entry, size_t pagesize) 584 { 585 struct { 586 struct nameidata nd; 587 struct vattr attr; 588 struct image_params image_params; 589 } *tempdata; 590 const Elf_Ehdr *hdr = NULL; 591 const Elf_Phdr *phdr = NULL; 592 struct nameidata *nd; 593 struct vattr *attr; 594 struct image_params *imgp; 595 vm_prot_t prot; 596 u_long rbase; 597 u_long base_addr = 0; 598 int error, i, numsegs; 599 600 #ifdef CAPABILITY_MODE 601 /* 602 * XXXJA: This check can go away once we are sufficiently confident 603 * that the checks in namei() are correct. 604 */ 605 if (IN_CAPABILITY_MODE(curthread)) 606 return (ECAPMODE); 607 #endif 608 609 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 610 nd = &tempdata->nd; 611 attr = &tempdata->attr; 612 imgp = &tempdata->image_params; 613 614 /* 615 * Initialize part of the common data 616 */ 617 imgp->proc = p; 618 imgp->attr = attr; 619 imgp->firstpage = NULL; 620 imgp->image_header = NULL; 621 imgp->object = NULL; 622 imgp->execlabel = NULL; 623 624 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); 625 if ((error = namei(nd)) != 0) { 626 nd->ni_vp = NULL; 627 goto fail; 628 } 629 NDFREE(nd, NDF_ONLY_PNBUF); 630 imgp->vp = nd->ni_vp; 631 632 /* 633 * Check permissions, modes, uid, etc on the file, and "open" it. 634 */ 635 error = exec_check_permissions(imgp); 636 if (error) 637 goto fail; 638 639 error = exec_map_first_page(imgp); 640 if (error) 641 goto fail; 642 643 /* 644 * Also make certain that the interpreter stays the same, so set 645 * its VV_TEXT flag, too. 646 */ 647 VOP_SET_TEXT(nd->ni_vp); 648 649 imgp->object = nd->ni_vp->v_object; 650 651 hdr = (const Elf_Ehdr *)imgp->image_header; 652 if ((error = __elfN(check_header)(hdr)) != 0) 653 goto fail; 654 if (hdr->e_type == ET_DYN) 655 rbase = *addr; 656 else if (hdr->e_type == ET_EXEC) 657 rbase = 0; 658 else { 659 error = ENOEXEC; 660 goto fail; 661 } 662 663 /* Only support headers that fit within first page for now */ 664 /* (multiplication of two Elf_Half fields will not overflow) */ 665 if ((hdr->e_phoff > PAGE_SIZE) || 666 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 667 error = ENOEXEC; 668 goto fail; 669 } 670 671 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 672 if (!aligned(phdr, Elf_Addr)) { 673 error = ENOEXEC; 674 goto fail; 675 } 676 677 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 678 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 679 /* Loadable segment */ 680 prot = __elfN(trans_prot)(phdr[i].p_flags); 681 error = __elfN(load_section)(imgp, phdr[i].p_offset, 682 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 683 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); 684 if (error != 0) 685 goto fail; 686 /* 687 * Establish the base address if this is the 688 * first segment. 689 */ 690 if (numsegs == 0) 691 base_addr = trunc_page(phdr[i].p_vaddr + 692 rbase); 693 numsegs++; 694 } 695 } 696 *addr = base_addr; 697 *entry = (unsigned long)hdr->e_entry + rbase; 698 699 fail: 700 if (imgp->firstpage) 701 exec_unmap_first_page(imgp); 702 703 if (nd->ni_vp) 704 vput(nd->ni_vp); 705 706 free(tempdata, M_TEMP); 707 708 return (error); 709 } 710 711 static int 712 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 713 { 714 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 715 const Elf_Phdr *phdr; 716 Elf_Auxargs *elf_auxargs; 717 struct vmspace *vmspace; 718 vm_prot_t prot; 719 u_long text_size = 0, data_size = 0, total_size = 0; 720 u_long text_addr = 0, data_addr = 0; 721 u_long seg_size, seg_addr; 722 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0; 723 int32_t osrel = 0; 724 int error = 0, i, n, interp_name_len = 0; 725 const char *interp = NULL, *newinterp = NULL; 726 Elf_Brandinfo *brand_info; 727 char *path; 728 struct sysentvec *sv; 729 730 /* 731 * Do we have a valid ELF header ? 732 * 733 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 734 * if particular brand doesn't support it. 735 */ 736 if (__elfN(check_header)(hdr) != 0 || 737 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 738 return (-1); 739 740 /* 741 * From here on down, we return an errno, not -1, as we've 742 * detected an ELF file. 743 */ 744 745 if ((hdr->e_phoff > PAGE_SIZE) || 746 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 747 /* Only support headers in first page for now */ 748 return (ENOEXEC); 749 } 750 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 751 if (!aligned(phdr, Elf_Addr)) 752 return (ENOEXEC); 753 n = 0; 754 baddr = 0; 755 for (i = 0; i < hdr->e_phnum; i++) { 756 switch (phdr[i].p_type) { 757 case PT_LOAD: 758 if (n == 0) 759 baddr = phdr[i].p_vaddr; 760 n++; 761 break; 762 case PT_INTERP: 763 /* Path to interpreter */ 764 if (phdr[i].p_filesz > MAXPATHLEN || 765 phdr[i].p_offset >= PAGE_SIZE || 766 phdr[i].p_offset + phdr[i].p_filesz >= PAGE_SIZE) 767 return (ENOEXEC); 768 interp = imgp->image_header + phdr[i].p_offset; 769 interp_name_len = phdr[i].p_filesz; 770 break; 771 case PT_GNU_STACK: 772 if (__elfN(nxstack)) 773 imgp->stack_prot = 774 __elfN(trans_prot)(phdr[i].p_flags); 775 break; 776 } 777 } 778 779 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 780 &osrel); 781 if (brand_info == NULL) { 782 uprintf("ELF binary type \"%u\" not known.\n", 783 hdr->e_ident[EI_OSABI]); 784 return (ENOEXEC); 785 } 786 if (hdr->e_type == ET_DYN) { 787 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) 788 return (ENOEXEC); 789 /* 790 * Honour the base load address from the dso if it is 791 * non-zero for some reason. 792 */ 793 if (baddr == 0) 794 et_dyn_addr = ET_DYN_LOAD_ADDR; 795 else 796 et_dyn_addr = 0; 797 } else 798 et_dyn_addr = 0; 799 sv = brand_info->sysvec; 800 if (interp != NULL && brand_info->interp_newpath != NULL) 801 newinterp = brand_info->interp_newpath; 802 803 /* 804 * Avoid a possible deadlock if the current address space is destroyed 805 * and that address space maps the locked vnode. In the common case, 806 * the locked vnode's v_usecount is decremented but remains greater 807 * than zero. Consequently, the vnode lock is not needed by vrele(). 808 * However, in cases where the vnode lock is external, such as nullfs, 809 * v_usecount may become zero. 810 * 811 * The VV_TEXT flag prevents modifications to the executable while 812 * the vnode is unlocked. 813 */ 814 VOP_UNLOCK(imgp->vp, 0); 815 816 error = exec_new_vmspace(imgp, sv); 817 imgp->proc->p_sysent = sv; 818 819 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 820 if (error) 821 return (error); 822 823 for (i = 0; i < hdr->e_phnum; i++) { 824 switch (phdr[i].p_type) { 825 case PT_LOAD: /* Loadable segment */ 826 if (phdr[i].p_memsz == 0) 827 break; 828 prot = __elfN(trans_prot)(phdr[i].p_flags); 829 error = __elfN(load_section)(imgp, phdr[i].p_offset, 830 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 831 phdr[i].p_memsz, phdr[i].p_filesz, prot, 832 sv->sv_pagesize); 833 if (error != 0) 834 return (error); 835 836 /* 837 * If this segment contains the program headers, 838 * remember their virtual address for the AT_PHDR 839 * aux entry. Static binaries don't usually include 840 * a PT_PHDR entry. 841 */ 842 if (phdr[i].p_offset == 0 && 843 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 844 <= phdr[i].p_filesz) 845 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 846 et_dyn_addr; 847 848 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 849 seg_size = round_page(phdr[i].p_memsz + 850 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 851 852 /* 853 * Make the largest executable segment the official 854 * text segment and all others data. 855 * 856 * Note that obreak() assumes that data_addr + 857 * data_size == end of data load area, and the ELF 858 * file format expects segments to be sorted by 859 * address. If multiple data segments exist, the 860 * last one will be used. 861 */ 862 863 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 864 text_size = seg_size; 865 text_addr = seg_addr; 866 } else { 867 data_size = seg_size; 868 data_addr = seg_addr; 869 } 870 total_size += seg_size; 871 break; 872 case PT_PHDR: /* Program header table info */ 873 proghdr = phdr[i].p_vaddr + et_dyn_addr; 874 break; 875 default: 876 break; 877 } 878 } 879 880 if (data_addr == 0 && data_size == 0) { 881 data_addr = text_addr; 882 data_size = text_size; 883 } 884 885 entry = (u_long)hdr->e_entry + et_dyn_addr; 886 887 /* 888 * Check limits. It should be safe to check the 889 * limits after loading the segments since we do 890 * not actually fault in all the segments pages. 891 */ 892 PROC_LOCK(imgp->proc); 893 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 894 text_size > maxtsiz || 895 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) || 896 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 || 897 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) { 898 PROC_UNLOCK(imgp->proc); 899 return (ENOMEM); 900 } 901 902 vmspace = imgp->proc->p_vmspace; 903 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 904 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 905 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 906 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 907 908 /* 909 * We load the dynamic linker where a userland call 910 * to mmap(0, ...) would put it. The rationale behind this 911 * calculation is that it leaves room for the heap to grow to 912 * its maximum allowed size. 913 */ 914 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc, 915 RLIMIT_DATA)); 916 PROC_UNLOCK(imgp->proc); 917 918 imgp->entry_addr = entry; 919 920 if (interp != NULL) { 921 int have_interp = FALSE; 922 VOP_UNLOCK(imgp->vp, 0); 923 if (brand_info->emul_path != NULL && 924 brand_info->emul_path[0] != '\0') { 925 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 926 snprintf(path, MAXPATHLEN, "%s%s", 927 brand_info->emul_path, interp); 928 error = __elfN(load_file)(imgp->proc, path, &addr, 929 &imgp->entry_addr, sv->sv_pagesize); 930 free(path, M_TEMP); 931 if (error == 0) 932 have_interp = TRUE; 933 } 934 if (!have_interp && newinterp != NULL) { 935 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 936 &imgp->entry_addr, sv->sv_pagesize); 937 if (error == 0) 938 have_interp = TRUE; 939 } 940 if (!have_interp) { 941 error = __elfN(load_file)(imgp->proc, interp, &addr, 942 &imgp->entry_addr, sv->sv_pagesize); 943 } 944 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 945 if (error != 0) { 946 uprintf("ELF interpreter %s not found\n", interp); 947 return (error); 948 } 949 } else 950 addr = et_dyn_addr; 951 952 /* 953 * Construct auxargs table (used by the fixup routine) 954 */ 955 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 956 elf_auxargs->execfd = -1; 957 elf_auxargs->phdr = proghdr; 958 elf_auxargs->phent = hdr->e_phentsize; 959 elf_auxargs->phnum = hdr->e_phnum; 960 elf_auxargs->pagesz = PAGE_SIZE; 961 elf_auxargs->base = addr; 962 elf_auxargs->flags = 0; 963 elf_auxargs->entry = entry; 964 965 imgp->auxargs = elf_auxargs; 966 imgp->interpreted = 0; 967 imgp->reloc_base = addr; 968 imgp->proc->p_osrel = osrel; 969 970 return (error); 971 } 972 973 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 974 975 int 976 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 977 { 978 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 979 Elf_Addr *base; 980 Elf_Addr *pos; 981 982 base = (Elf_Addr *)*stack_base; 983 pos = base + (imgp->args->argc + imgp->args->envc + 2); 984 985 if (args->execfd != -1) 986 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 987 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 988 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 989 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 990 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 991 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 992 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 993 AUXARGS_ENTRY(pos, AT_BASE, args->base); 994 if (imgp->execpathp != 0) 995 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 996 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate); 997 if (imgp->canary != 0) { 998 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 999 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1000 } 1001 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1002 if (imgp->pagesizes != 0) { 1003 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1004 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1005 } 1006 if (imgp->sysent->sv_timekeep_base != 0) { 1007 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1008 imgp->sysent->sv_timekeep_base); 1009 } 1010 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1011 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1012 imgp->sysent->sv_stackprot); 1013 AUXARGS_ENTRY(pos, AT_NULL, 0); 1014 1015 free(imgp->auxargs, M_TEMP); 1016 imgp->auxargs = NULL; 1017 1018 base--; 1019 suword(base, (long)imgp->args->argc); 1020 *stack_base = (register_t *)base; 1021 return (0); 1022 } 1023 1024 /* 1025 * Code for generating ELF core dumps. 1026 */ 1027 1028 typedef void (*segment_callback)(vm_map_entry_t, void *); 1029 1030 /* Closure for cb_put_phdr(). */ 1031 struct phdr_closure { 1032 Elf_Phdr *phdr; /* Program header to fill in */ 1033 Elf_Off offset; /* Offset of segment in core file */ 1034 }; 1035 1036 /* Closure for cb_size_segment(). */ 1037 struct sseg_closure { 1038 int count; /* Count of writable segments. */ 1039 size_t size; /* Total size of all writable segments. */ 1040 }; 1041 1042 static void cb_put_phdr(vm_map_entry_t, void *); 1043 static void cb_size_segment(vm_map_entry_t, void *); 1044 static void each_writable_segment(struct thread *, segment_callback, void *); 1045 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 1046 int, void *, size_t, gzFile); 1047 static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 1048 static void __elfN(putnote)(void *, size_t *, const char *, int, 1049 const void *, size_t); 1050 1051 #ifdef COMPRESS_USER_CORES 1052 extern int compress_user_cores; 1053 extern int compress_user_cores_gzlevel; 1054 #endif 1055 1056 static int 1057 core_output(struct vnode *vp, void *base, size_t len, off_t offset, 1058 struct ucred *active_cred, struct ucred *file_cred, 1059 struct thread *td, char *core_buf, gzFile gzfile) { 1060 1061 int error; 1062 if (gzfile) { 1063 #ifdef COMPRESS_USER_CORES 1064 error = compress_core(gzfile, base, core_buf, len, td); 1065 #else 1066 panic("shouldn't be here"); 1067 #endif 1068 } else { 1069 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset, 1070 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred, 1071 NULL, td); 1072 } 1073 return (error); 1074 } 1075 1076 int 1077 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1078 { 1079 struct ucred *cred = td->td_ucred; 1080 int error = 0; 1081 struct sseg_closure seginfo; 1082 void *hdr; 1083 size_t hdrsize; 1084 1085 gzFile gzfile = Z_NULL; 1086 char *core_buf = NULL; 1087 #ifdef COMPRESS_USER_CORES 1088 char gzopen_flags[8]; 1089 char *p; 1090 int doing_compress = flags & IMGACT_CORE_COMPRESS; 1091 #endif 1092 1093 hdr = NULL; 1094 1095 #ifdef COMPRESS_USER_CORES 1096 if (doing_compress) { 1097 p = gzopen_flags; 1098 *p++ = 'w'; 1099 if (compress_user_cores_gzlevel >= 0 && 1100 compress_user_cores_gzlevel <= 9) 1101 *p++ = '0' + compress_user_cores_gzlevel; 1102 *p = 0; 1103 gzfile = gz_open("", gzopen_flags, vp); 1104 if (gzfile == Z_NULL) { 1105 error = EFAULT; 1106 goto done; 1107 } 1108 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1109 if (!core_buf) { 1110 error = ENOMEM; 1111 goto done; 1112 } 1113 } 1114 #endif 1115 1116 /* Size the program segments. */ 1117 seginfo.count = 0; 1118 seginfo.size = 0; 1119 each_writable_segment(td, cb_size_segment, &seginfo); 1120 1121 /* 1122 * Calculate the size of the core file header area by making 1123 * a dry run of generating it. Nothing is written, but the 1124 * size is calculated. 1125 */ 1126 hdrsize = 0; 1127 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 1128 1129 #ifdef RACCT 1130 PROC_LOCK(td->td_proc); 1131 error = racct_add(td->td_proc, RACCT_CORE, hdrsize + seginfo.size); 1132 PROC_UNLOCK(td->td_proc); 1133 if (error != 0) { 1134 error = EFAULT; 1135 goto done; 1136 } 1137 #endif 1138 if (hdrsize + seginfo.size >= limit) { 1139 error = EFAULT; 1140 goto done; 1141 } 1142 1143 /* 1144 * Allocate memory for building the header, fill it up, 1145 * and write it out. 1146 */ 1147 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1148 if (hdr == NULL) { 1149 error = EINVAL; 1150 goto done; 1151 } 1152 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize, 1153 gzfile); 1154 1155 /* Write the contents of all of the writable segments. */ 1156 if (error == 0) { 1157 Elf_Phdr *php; 1158 off_t offset; 1159 int i; 1160 1161 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1162 offset = hdrsize; 1163 for (i = 0; i < seginfo.count; i++) { 1164 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr, 1165 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile); 1166 if (error != 0) 1167 break; 1168 offset += php->p_filesz; 1169 php++; 1170 } 1171 } 1172 if (error) { 1173 log(LOG_WARNING, 1174 "Failed to write core file for process %s (error %d)\n", 1175 curproc->p_comm, error); 1176 } 1177 1178 done: 1179 #ifdef COMPRESS_USER_CORES 1180 if (core_buf) 1181 free(core_buf, M_TEMP); 1182 if (gzfile) 1183 gzclose(gzfile); 1184 #endif 1185 1186 free(hdr, M_TEMP); 1187 1188 return (error); 1189 } 1190 1191 /* 1192 * A callback for each_writable_segment() to write out the segment's 1193 * program header entry. 1194 */ 1195 static void 1196 cb_put_phdr(entry, closure) 1197 vm_map_entry_t entry; 1198 void *closure; 1199 { 1200 struct phdr_closure *phc = (struct phdr_closure *)closure; 1201 Elf_Phdr *phdr = phc->phdr; 1202 1203 phc->offset = round_page(phc->offset); 1204 1205 phdr->p_type = PT_LOAD; 1206 phdr->p_offset = phc->offset; 1207 phdr->p_vaddr = entry->start; 1208 phdr->p_paddr = 0; 1209 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1210 phdr->p_align = PAGE_SIZE; 1211 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1212 1213 phc->offset += phdr->p_filesz; 1214 phc->phdr++; 1215 } 1216 1217 /* 1218 * A callback for each_writable_segment() to gather information about 1219 * the number of segments and their total size. 1220 */ 1221 static void 1222 cb_size_segment(entry, closure) 1223 vm_map_entry_t entry; 1224 void *closure; 1225 { 1226 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1227 1228 ssc->count++; 1229 ssc->size += entry->end - entry->start; 1230 } 1231 1232 /* 1233 * For each writable segment in the process's memory map, call the given 1234 * function with a pointer to the map entry and some arbitrary 1235 * caller-supplied data. 1236 */ 1237 static void 1238 each_writable_segment(td, func, closure) 1239 struct thread *td; 1240 segment_callback func; 1241 void *closure; 1242 { 1243 struct proc *p = td->td_proc; 1244 vm_map_t map = &p->p_vmspace->vm_map; 1245 vm_map_entry_t entry; 1246 vm_object_t backing_object, object; 1247 boolean_t ignore_entry; 1248 1249 vm_map_lock_read(map); 1250 for (entry = map->header.next; entry != &map->header; 1251 entry = entry->next) { 1252 /* 1253 * Don't dump inaccessible mappings, deal with legacy 1254 * coredump mode. 1255 * 1256 * Note that read-only segments related to the elf binary 1257 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1258 * need to arbitrarily ignore such segments. 1259 */ 1260 if (elf_legacy_coredump) { 1261 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1262 continue; 1263 } else { 1264 if ((entry->protection & VM_PROT_ALL) == 0) 1265 continue; 1266 } 1267 1268 /* 1269 * Dont include memory segment in the coredump if 1270 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1271 * madvise(2). Do not dump submaps (i.e. parts of the 1272 * kernel map). 1273 */ 1274 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1275 continue; 1276 1277 if ((object = entry->object.vm_object) == NULL) 1278 continue; 1279 1280 /* Ignore memory-mapped devices and such things. */ 1281 VM_OBJECT_LOCK(object); 1282 while ((backing_object = object->backing_object) != NULL) { 1283 VM_OBJECT_LOCK(backing_object); 1284 VM_OBJECT_UNLOCK(object); 1285 object = backing_object; 1286 } 1287 ignore_entry = object->type != OBJT_DEFAULT && 1288 object->type != OBJT_SWAP && object->type != OBJT_VNODE; 1289 VM_OBJECT_UNLOCK(object); 1290 if (ignore_entry) 1291 continue; 1292 1293 (*func)(entry, closure); 1294 } 1295 vm_map_unlock_read(map); 1296 } 1297 1298 /* 1299 * Write the core file header to the file, including padding up to 1300 * the page boundary. 1301 */ 1302 static int 1303 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize, gzfile) 1304 struct thread *td; 1305 struct vnode *vp; 1306 struct ucred *cred; 1307 int numsegs; 1308 size_t hdrsize; 1309 void *hdr; 1310 gzFile gzfile; 1311 { 1312 size_t off; 1313 1314 /* Fill in the header. */ 1315 bzero(hdr, hdrsize); 1316 off = 0; 1317 __elfN(puthdr)(td, hdr, &off, numsegs); 1318 1319 if (!gzfile) { 1320 /* Write it to the core file. */ 1321 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1322 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1323 td)); 1324 } else { 1325 #ifdef COMPRESS_USER_CORES 1326 if (gzwrite(gzfile, hdr, hdrsize) != hdrsize) { 1327 log(LOG_WARNING, 1328 "Failed to compress core file header for process" 1329 " %s.\n", curproc->p_comm); 1330 return (EFAULT); 1331 } 1332 else { 1333 return (0); 1334 } 1335 #else 1336 panic("shouldn't be here"); 1337 #endif 1338 } 1339 } 1340 1341 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1342 #include <compat/freebsd32/freebsd32.h> 1343 1344 typedef struct prstatus32 elf_prstatus_t; 1345 typedef struct prpsinfo32 elf_prpsinfo_t; 1346 typedef struct fpreg32 elf_prfpregset_t; 1347 typedef struct fpreg32 elf_fpregset_t; 1348 typedef struct reg32 elf_gregset_t; 1349 typedef struct thrmisc32 elf_thrmisc_t; 1350 #else 1351 typedef prstatus_t elf_prstatus_t; 1352 typedef prpsinfo_t elf_prpsinfo_t; 1353 typedef prfpregset_t elf_prfpregset_t; 1354 typedef prfpregset_t elf_fpregset_t; 1355 typedef gregset_t elf_gregset_t; 1356 typedef thrmisc_t elf_thrmisc_t; 1357 #endif 1358 1359 static void 1360 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1361 { 1362 struct { 1363 elf_prstatus_t status; 1364 elf_prfpregset_t fpregset; 1365 elf_prpsinfo_t psinfo; 1366 elf_thrmisc_t thrmisc; 1367 } *tempdata; 1368 elf_prstatus_t *status; 1369 elf_prfpregset_t *fpregset; 1370 elf_prpsinfo_t *psinfo; 1371 elf_thrmisc_t *thrmisc; 1372 struct proc *p; 1373 struct thread *thr; 1374 size_t ehoff, noteoff, notesz, phoff; 1375 1376 p = td->td_proc; 1377 1378 ehoff = *off; 1379 *off += sizeof(Elf_Ehdr); 1380 1381 phoff = *off; 1382 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1383 1384 noteoff = *off; 1385 /* 1386 * Don't allocate space for the notes if we're just calculating 1387 * the size of the header. We also don't collect the data. 1388 */ 1389 if (dst != NULL) { 1390 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1391 status = &tempdata->status; 1392 fpregset = &tempdata->fpregset; 1393 psinfo = &tempdata->psinfo; 1394 thrmisc = &tempdata->thrmisc; 1395 } else { 1396 tempdata = NULL; 1397 status = NULL; 1398 fpregset = NULL; 1399 psinfo = NULL; 1400 thrmisc = NULL; 1401 } 1402 1403 if (dst != NULL) { 1404 psinfo->pr_version = PRPSINFO_VERSION; 1405 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1406 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1407 /* 1408 * XXX - We don't fill in the command line arguments properly 1409 * yet. 1410 */ 1411 strlcpy(psinfo->pr_psargs, p->p_comm, 1412 sizeof(psinfo->pr_psargs)); 1413 } 1414 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1415 sizeof *psinfo); 1416 1417 /* 1418 * To have the debugger select the right thread (LWP) as the initial 1419 * thread, we dump the state of the thread passed to us in td first. 1420 * This is the thread that causes the core dump and thus likely to 1421 * be the right thread one wants to have selected in the debugger. 1422 */ 1423 thr = td; 1424 while (thr != NULL) { 1425 if (dst != NULL) { 1426 status->pr_version = PRSTATUS_VERSION; 1427 status->pr_statussz = sizeof(elf_prstatus_t); 1428 status->pr_gregsetsz = sizeof(elf_gregset_t); 1429 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1430 status->pr_osreldate = osreldate; 1431 status->pr_cursig = p->p_sig; 1432 status->pr_pid = thr->td_tid; 1433 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1434 fill_regs32(thr, &status->pr_reg); 1435 fill_fpregs32(thr, fpregset); 1436 #else 1437 fill_regs(thr, &status->pr_reg); 1438 fill_fpregs(thr, fpregset); 1439 #endif 1440 memset(&thrmisc->_pad, 0, sizeof (thrmisc->_pad)); 1441 strcpy(thrmisc->pr_tname, thr->td_name); 1442 } 1443 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1444 sizeof *status); 1445 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1446 sizeof *fpregset); 1447 __elfN(putnote)(dst, off, "FreeBSD", NT_THRMISC, thrmisc, 1448 sizeof *thrmisc); 1449 /* 1450 * Allow for MD specific notes, as well as any MD 1451 * specific preparations for writing MI notes. 1452 */ 1453 __elfN(dump_thread)(thr, dst, off); 1454 1455 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1456 TAILQ_NEXT(thr, td_plist); 1457 if (thr == td) 1458 thr = TAILQ_NEXT(thr, td_plist); 1459 } 1460 1461 notesz = *off - noteoff; 1462 1463 if (dst != NULL) 1464 free(tempdata, M_TEMP); 1465 1466 /* Align up to a page boundary for the program segments. */ 1467 *off = round_page(*off); 1468 1469 if (dst != NULL) { 1470 Elf_Ehdr *ehdr; 1471 Elf_Phdr *phdr; 1472 struct phdr_closure phc; 1473 1474 /* 1475 * Fill in the ELF header. 1476 */ 1477 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1478 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1479 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1480 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1481 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1482 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1483 ehdr->e_ident[EI_DATA] = ELF_DATA; 1484 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1485 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1486 ehdr->e_ident[EI_ABIVERSION] = 0; 1487 ehdr->e_ident[EI_PAD] = 0; 1488 ehdr->e_type = ET_CORE; 1489 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1490 ehdr->e_machine = ELF_ARCH32; 1491 #else 1492 ehdr->e_machine = ELF_ARCH; 1493 #endif 1494 ehdr->e_version = EV_CURRENT; 1495 ehdr->e_entry = 0; 1496 ehdr->e_phoff = phoff; 1497 ehdr->e_flags = 0; 1498 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1499 ehdr->e_phentsize = sizeof(Elf_Phdr); 1500 ehdr->e_phnum = numsegs + 1; 1501 ehdr->e_shentsize = sizeof(Elf_Shdr); 1502 ehdr->e_shnum = 0; 1503 ehdr->e_shstrndx = SHN_UNDEF; 1504 1505 /* 1506 * Fill in the program header entries. 1507 */ 1508 phdr = (Elf_Phdr *)((char *)dst + phoff); 1509 1510 /* The note segement. */ 1511 phdr->p_type = PT_NOTE; 1512 phdr->p_offset = noteoff; 1513 phdr->p_vaddr = 0; 1514 phdr->p_paddr = 0; 1515 phdr->p_filesz = notesz; 1516 phdr->p_memsz = 0; 1517 phdr->p_flags = 0; 1518 phdr->p_align = 0; 1519 phdr++; 1520 1521 /* All the writable segments from the program. */ 1522 phc.phdr = phdr; 1523 phc.offset = *off; 1524 each_writable_segment(td, cb_put_phdr, &phc); 1525 } 1526 } 1527 1528 static void 1529 __elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1530 const void *desc, size_t descsz) 1531 { 1532 Elf_Note note; 1533 1534 note.n_namesz = strlen(name) + 1; 1535 note.n_descsz = descsz; 1536 note.n_type = type; 1537 if (dst != NULL) 1538 bcopy(¬e, (char *)dst + *off, sizeof note); 1539 *off += sizeof note; 1540 if (dst != NULL) 1541 bcopy(name, (char *)dst + *off, note.n_namesz); 1542 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1543 if (dst != NULL) 1544 bcopy(desc, (char *)dst + *off, note.n_descsz); 1545 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1546 } 1547 1548 static boolean_t 1549 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 1550 int32_t *osrel, const Elf_Phdr *pnote) 1551 { 1552 const Elf_Note *note, *note0, *note_end; 1553 const char *note_name; 1554 int i; 1555 1556 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE || 1557 pnote->p_filesz > PAGE_SIZE || 1558 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE) 1559 return (FALSE); 1560 1561 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); 1562 note_end = (const Elf_Note *)(imgp->image_header + 1563 pnote->p_offset + pnote->p_filesz); 1564 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 1565 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 1566 (const char *)note < sizeof(Elf_Note)) 1567 return (FALSE); 1568 if (note->n_namesz != checknote->hdr.n_namesz || 1569 note->n_descsz != checknote->hdr.n_descsz || 1570 note->n_type != checknote->hdr.n_type) 1571 goto nextnote; 1572 note_name = (const char *)(note + 1); 1573 if (note_name + checknote->hdr.n_namesz >= 1574 (const char *)note_end || strncmp(checknote->vendor, 1575 note_name, checknote->hdr.n_namesz) != 0) 1576 goto nextnote; 1577 1578 /* 1579 * Fetch the osreldate for binary 1580 * from the ELF OSABI-note if necessary. 1581 */ 1582 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 1583 checknote->trans_osrel != NULL) 1584 return (checknote->trans_osrel(note, osrel)); 1585 return (TRUE); 1586 1587 nextnote: 1588 note = (const Elf_Note *)((const char *)(note + 1) + 1589 roundup2(note->n_namesz, sizeof(Elf32_Addr)) + 1590 roundup2(note->n_descsz, sizeof(Elf32_Addr))); 1591 } 1592 1593 return (FALSE); 1594 } 1595 1596 /* 1597 * Try to find the appropriate ABI-note section for checknote, 1598 * fetch the osreldate for binary from the ELF OSABI-note. Only the 1599 * first page of the image is searched, the same as for headers. 1600 */ 1601 static boolean_t 1602 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 1603 int32_t *osrel) 1604 { 1605 const Elf_Phdr *phdr; 1606 const Elf_Ehdr *hdr; 1607 int i; 1608 1609 hdr = (const Elf_Ehdr *)imgp->image_header; 1610 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1611 1612 for (i = 0; i < hdr->e_phnum; i++) { 1613 if (phdr[i].p_type == PT_NOTE && 1614 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 1615 return (TRUE); 1616 } 1617 return (FALSE); 1618 1619 } 1620 1621 /* 1622 * Tell kern_execve.c about it, with a little help from the linker. 1623 */ 1624 static struct execsw __elfN(execsw) = { 1625 __CONCAT(exec_, __elfN(imgact)), 1626 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1627 }; 1628 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1629 1630 #ifdef COMPRESS_USER_CORES 1631 /* 1632 * Compress and write out a core segment for a user process. 1633 * 1634 * 'inbuf' is the starting address of a VM segment in the process' address 1635 * space that is to be compressed and written out to the core file. 'dest_buf' 1636 * is a buffer in the kernel's address space. The segment is copied from 1637 * 'inbuf' to 'dest_buf' first before being processed by the compression 1638 * routine gzwrite(). This copying is necessary because the content of the VM 1639 * segment may change between the compression pass and the crc-computation pass 1640 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel. 1641 */ 1642 static int 1643 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len, 1644 struct thread *td) 1645 { 1646 int len_compressed; 1647 int error = 0; 1648 unsigned int chunk_len; 1649 1650 while (len) { 1651 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len; 1652 copyin(inbuf, dest_buf, chunk_len); 1653 len_compressed = gzwrite(file, dest_buf, chunk_len); 1654 1655 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed); 1656 1657 if ((unsigned int)len_compressed != chunk_len) { 1658 log(LOG_WARNING, 1659 "compress_core: length mismatch (0x%x returned, " 1660 "0x%x expected)\n", len_compressed, chunk_len); 1661 EVENTHANDLER_INVOKE(app_coredump_error, td, 1662 "compress_core: length mismatch %x -> %x", 1663 chunk_len, len_compressed); 1664 error = EFAULT; 1665 break; 1666 } 1667 inbuf += chunk_len; 1668 len -= chunk_len; 1669 maybe_yield(); 1670 } 1671 1672 return (error); 1673 } 1674 #endif /* COMPRESS_USER_CORES */ 1675 1676 static vm_prot_t 1677 __elfN(trans_prot)(Elf_Word flags) 1678 { 1679 vm_prot_t prot; 1680 1681 prot = 0; 1682 if (flags & PF_X) 1683 prot |= VM_PROT_EXECUTE; 1684 if (flags & PF_W) 1685 prot |= VM_PROT_WRITE; 1686 if (flags & PF_R) 1687 prot |= VM_PROT_READ; 1688 #if __ELF_WORD_SIZE == 32 1689 #if defined(__amd64__) || defined(__ia64__) 1690 if (i386_read_exec && (flags & PF_R)) 1691 prot |= VM_PROT_EXECUTE; 1692 #endif 1693 #endif 1694 return (prot); 1695 } 1696 1697 static Elf_Word 1698 __elfN(untrans_prot)(vm_prot_t prot) 1699 { 1700 Elf_Word flags; 1701 1702 flags = 0; 1703 if (prot & VM_PROT_EXECUTE) 1704 flags |= PF_X; 1705 if (prot & VM_PROT_READ) 1706 flags |= PF_R; 1707 if (prot & VM_PROT_WRITE) 1708 flags |= PF_W; 1709 return (flags); 1710 } 1711