1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2017 Dell EMC 5 * Copyright (c) 2000-2001, 2003 David O'Brien 6 * Copyright (c) 1995-1996 Søren Schmidt 7 * Copyright (c) 1996 Peter Wemm 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer 15 * in this position and unchanged. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. The name of the author may not be used to endorse or promote products 20 * derived from this software without specific prior written permission 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_capsicum.h" 38 39 #include <sys/param.h> 40 #include <sys/capsicum.h> 41 #include <sys/compressor.h> 42 #include <sys/exec.h> 43 #include <sys/fcntl.h> 44 #include <sys/imgact.h> 45 #include <sys/imgact_elf.h> 46 #include <sys/jail.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/mount.h> 51 #include <sys/mman.h> 52 #include <sys/namei.h> 53 #include <sys/pioctl.h> 54 #include <sys/proc.h> 55 #include <sys/procfs.h> 56 #include <sys/ptrace.h> 57 #include <sys/racct.h> 58 #include <sys/resourcevar.h> 59 #include <sys/rwlock.h> 60 #include <sys/sbuf.h> 61 #include <sys/sf_buf.h> 62 #include <sys/smp.h> 63 #include <sys/systm.h> 64 #include <sys/signalvar.h> 65 #include <sys/stat.h> 66 #include <sys/sx.h> 67 #include <sys/syscall.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/vnode.h> 71 #include <sys/syslog.h> 72 #include <sys/eventhandler.h> 73 #include <sys/user.h> 74 75 #include <vm/vm.h> 76 #include <vm/vm_kern.h> 77 #include <vm/vm_param.h> 78 #include <vm/pmap.h> 79 #include <vm/vm_map.h> 80 #include <vm/vm_object.h> 81 #include <vm/vm_extern.h> 82 83 #include <machine/elf.h> 84 #include <machine/md_var.h> 85 86 #define ELF_NOTE_ROUNDSIZE 4 87 #define OLD_EI_BRAND 8 88 89 static int __elfN(check_header)(const Elf_Ehdr *hdr); 90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 91 const char *interp, int interp_name_len, int32_t *osrel); 92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 93 u_long *entry, size_t pagesize); 94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset, 95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 96 size_t pagesize); 97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 98 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note, 99 int32_t *osrel); 100 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 101 static boolean_t __elfN(check_note)(struct image_params *imgp, 102 Elf_Brandnote *checknote, int32_t *osrel); 103 static vm_prot_t __elfN(trans_prot)(Elf_Word); 104 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 105 106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 107 ""); 108 109 #define CORE_BUF_SIZE (16 * 1024) 110 111 int __elfN(fallback_brand) = -1; 112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 113 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, 114 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 115 116 static int elf_legacy_coredump = 0; 117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 118 &elf_legacy_coredump, 0, 119 "include all and only RW pages in core dumps"); 120 121 int __elfN(nxstack) = 122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \ 123 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) 124 1; 125 #else 126 0; 127 #endif 128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); 131 132 #if __ELF_WORD_SIZE == 32 133 #if defined(__amd64__) 134 int i386_read_exec = 0; 135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 136 "enable execution from readable segments"); 137 #endif 138 #endif 139 140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 141 142 #define trunc_page_ps(va, ps) rounddown2(va, ps) 143 #define round_page_ps(va, ps) roundup2(va, ps) 144 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) 145 146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 147 148 Elf_Brandnote __elfN(freebsd_brandnote) = { 149 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 150 .hdr.n_descsz = sizeof(int32_t), 151 .hdr.n_type = NT_FREEBSD_ABI_TAG, 152 .vendor = FREEBSD_ABI_VENDOR, 153 .flags = BN_TRANSLATE_OSREL, 154 .trans_osrel = __elfN(freebsd_trans_osrel) 155 }; 156 157 static bool 158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 159 { 160 uintptr_t p; 161 162 p = (uintptr_t)(note + 1); 163 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 164 *osrel = *(const int32_t *)(p); 165 166 return (true); 167 } 168 169 static const char GNU_ABI_VENDOR[] = "GNU"; 170 static int GNU_KFREEBSD_ABI_DESC = 3; 171 172 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 173 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 174 .hdr.n_descsz = 16, /* XXX at least 16 */ 175 .hdr.n_type = 1, 176 .vendor = GNU_ABI_VENDOR, 177 .flags = BN_TRANSLATE_OSREL, 178 .trans_osrel = kfreebsd_trans_osrel 179 }; 180 181 static bool 182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 183 { 184 const Elf32_Word *desc; 185 uintptr_t p; 186 187 p = (uintptr_t)(note + 1); 188 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 189 190 desc = (const Elf32_Word *)p; 191 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 192 return (false); 193 194 /* 195 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 196 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 197 */ 198 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 199 200 return (true); 201 } 202 203 int 204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 205 { 206 int i; 207 208 for (i = 0; i < MAX_BRANDS; i++) { 209 if (elf_brand_list[i] == NULL) { 210 elf_brand_list[i] = entry; 211 break; 212 } 213 } 214 if (i == MAX_BRANDS) { 215 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 216 __func__, entry); 217 return (-1); 218 } 219 return (0); 220 } 221 222 int 223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 224 { 225 int i; 226 227 for (i = 0; i < MAX_BRANDS; i++) { 228 if (elf_brand_list[i] == entry) { 229 elf_brand_list[i] = NULL; 230 break; 231 } 232 } 233 if (i == MAX_BRANDS) 234 return (-1); 235 return (0); 236 } 237 238 int 239 __elfN(brand_inuse)(Elf_Brandinfo *entry) 240 { 241 struct proc *p; 242 int rval = FALSE; 243 244 sx_slock(&allproc_lock); 245 FOREACH_PROC_IN_SYSTEM(p) { 246 if (p->p_sysent == entry->sysvec) { 247 rval = TRUE; 248 break; 249 } 250 } 251 sx_sunlock(&allproc_lock); 252 253 return (rval); 254 } 255 256 static Elf_Brandinfo * 257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 258 int interp_name_len, int32_t *osrel) 259 { 260 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 261 Elf_Brandinfo *bi, *bi_m; 262 boolean_t ret; 263 int i; 264 265 /* 266 * We support four types of branding -- (1) the ELF EI_OSABI field 267 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 268 * branding w/in the ELF header, (3) path of the `interp_path' 269 * field, and (4) the ".note.ABI-tag" ELF section. 270 */ 271 272 /* Look for an ".note.ABI-tag" ELF section */ 273 bi_m = NULL; 274 for (i = 0; i < MAX_BRANDS; i++) { 275 bi = elf_brand_list[i]; 276 if (bi == NULL) 277 continue; 278 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 279 continue; 280 if (hdr->e_machine == bi->machine && (bi->flags & 281 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 282 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 283 /* Give brand a chance to veto check_note's guess */ 284 if (ret && bi->header_supported) 285 ret = bi->header_supported(imgp); 286 /* 287 * If note checker claimed the binary, but the 288 * interpreter path in the image does not 289 * match default one for the brand, try to 290 * search for other brands with the same 291 * interpreter. Either there is better brand 292 * with the right interpreter, or, failing 293 * this, we return first brand which accepted 294 * our note and, optionally, header. 295 */ 296 if (ret && bi_m == NULL && interp != NULL && 297 (bi->interp_path == NULL || 298 (strlen(bi->interp_path) + 1 != interp_name_len || 299 strncmp(interp, bi->interp_path, interp_name_len) 300 != 0))) { 301 bi_m = bi; 302 ret = 0; 303 } 304 if (ret) 305 return (bi); 306 } 307 } 308 if (bi_m != NULL) 309 return (bi_m); 310 311 /* If the executable has a brand, search for it in the brand list. */ 312 for (i = 0; i < MAX_BRANDS; i++) { 313 bi = elf_brand_list[i]; 314 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 315 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 316 continue; 317 if (hdr->e_machine == bi->machine && 318 (hdr->e_ident[EI_OSABI] == bi->brand || 319 (bi->compat_3_brand != NULL && 320 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 321 bi->compat_3_brand) == 0))) { 322 /* Looks good, but give brand a chance to veto */ 323 if (bi->header_supported == NULL || 324 bi->header_supported(imgp)) { 325 /* 326 * Again, prefer strictly matching 327 * interpreter path. 328 */ 329 if (interp_name_len == 0 && 330 bi->interp_path == NULL) 331 return (bi); 332 if (bi->interp_path != NULL && 333 strlen(bi->interp_path) + 1 == 334 interp_name_len && strncmp(interp, 335 bi->interp_path, interp_name_len) == 0) 336 return (bi); 337 if (bi_m == NULL) 338 bi_m = bi; 339 } 340 } 341 } 342 if (bi_m != NULL) 343 return (bi_m); 344 345 /* No known brand, see if the header is recognized by any brand */ 346 for (i = 0; i < MAX_BRANDS; i++) { 347 bi = elf_brand_list[i]; 348 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 349 bi->header_supported == NULL) 350 continue; 351 if (hdr->e_machine == bi->machine) { 352 ret = bi->header_supported(imgp); 353 if (ret) 354 return (bi); 355 } 356 } 357 358 /* Lacking a known brand, search for a recognized interpreter. */ 359 if (interp != NULL) { 360 for (i = 0; i < MAX_BRANDS; i++) { 361 bi = elf_brand_list[i]; 362 if (bi == NULL || (bi->flags & 363 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 364 != 0) 365 continue; 366 if (hdr->e_machine == bi->machine && 367 bi->interp_path != NULL && 368 /* ELF image p_filesz includes terminating zero */ 369 strlen(bi->interp_path) + 1 == interp_name_len && 370 strncmp(interp, bi->interp_path, interp_name_len) 371 == 0 && (bi->header_supported == NULL || 372 bi->header_supported(imgp))) 373 return (bi); 374 } 375 } 376 377 /* Lacking a recognized interpreter, try the default brand */ 378 for (i = 0; i < MAX_BRANDS; i++) { 379 bi = elf_brand_list[i]; 380 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 381 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 382 continue; 383 if (hdr->e_machine == bi->machine && 384 __elfN(fallback_brand) == bi->brand && 385 (bi->header_supported == NULL || 386 bi->header_supported(imgp))) 387 return (bi); 388 } 389 return (NULL); 390 } 391 392 static int 393 __elfN(check_header)(const Elf_Ehdr *hdr) 394 { 395 Elf_Brandinfo *bi; 396 int i; 397 398 if (!IS_ELF(*hdr) || 399 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 400 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 401 hdr->e_ident[EI_VERSION] != EV_CURRENT || 402 hdr->e_phentsize != sizeof(Elf_Phdr) || 403 hdr->e_version != ELF_TARG_VER) 404 return (ENOEXEC); 405 406 /* 407 * Make sure we have at least one brand for this machine. 408 */ 409 410 for (i = 0; i < MAX_BRANDS; i++) { 411 bi = elf_brand_list[i]; 412 if (bi != NULL && bi->machine == hdr->e_machine) 413 break; 414 } 415 if (i == MAX_BRANDS) 416 return (ENOEXEC); 417 418 return (0); 419 } 420 421 static int 422 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 423 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 424 { 425 struct sf_buf *sf; 426 int error; 427 vm_offset_t off; 428 429 /* 430 * Create the page if it doesn't exist yet. Ignore errors. 431 */ 432 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 433 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 434 435 /* 436 * Find the page from the underlying object. 437 */ 438 if (object != NULL) { 439 sf = vm_imgact_map_page(object, offset); 440 if (sf == NULL) 441 return (KERN_FAILURE); 442 off = offset - trunc_page(offset); 443 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 444 end - start); 445 vm_imgact_unmap_page(sf); 446 if (error != 0) 447 return (KERN_FAILURE); 448 } 449 450 return (KERN_SUCCESS); 451 } 452 453 static int 454 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object, 455 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot, 456 int cow) 457 { 458 struct sf_buf *sf; 459 vm_offset_t off; 460 vm_size_t sz; 461 int error, locked, rv; 462 463 if (start != trunc_page(start)) { 464 rv = __elfN(map_partial)(map, object, offset, start, 465 round_page(start), prot); 466 if (rv != KERN_SUCCESS) 467 return (rv); 468 offset += round_page(start) - start; 469 start = round_page(start); 470 } 471 if (end != round_page(end)) { 472 rv = __elfN(map_partial)(map, object, offset + 473 trunc_page(end) - start, trunc_page(end), end, prot); 474 if (rv != KERN_SUCCESS) 475 return (rv); 476 end = trunc_page(end); 477 } 478 if (start >= end) 479 return (KERN_SUCCESS); 480 if ((offset & PAGE_MASK) != 0) { 481 /* 482 * The mapping is not page aligned. This means that we have 483 * to copy the data. 484 */ 485 rv = vm_map_fixed(map, NULL, 0, start, end - start, 486 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 487 if (rv != KERN_SUCCESS) 488 return (rv); 489 if (object == NULL) 490 return (KERN_SUCCESS); 491 for (; start < end; start += sz) { 492 sf = vm_imgact_map_page(object, offset); 493 if (sf == NULL) 494 return (KERN_FAILURE); 495 off = offset - trunc_page(offset); 496 sz = end - start; 497 if (sz > PAGE_SIZE - off) 498 sz = PAGE_SIZE - off; 499 error = copyout((caddr_t)sf_buf_kva(sf) + off, 500 (caddr_t)start, sz); 501 vm_imgact_unmap_page(sf); 502 if (error != 0) 503 return (KERN_FAILURE); 504 offset += sz; 505 } 506 } else { 507 vm_object_reference(object); 508 rv = vm_map_fixed(map, object, offset, start, end - start, 509 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL); 510 if (rv != KERN_SUCCESS) { 511 locked = VOP_ISLOCKED(imgp->vp); 512 VOP_UNLOCK(imgp->vp, 0); 513 vm_object_deallocate(object); 514 vn_lock(imgp->vp, locked | LK_RETRY); 515 return (rv); 516 } 517 } 518 return (KERN_SUCCESS); 519 } 520 521 static int 522 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset, 523 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 524 size_t pagesize) 525 { 526 struct sf_buf *sf; 527 size_t map_len; 528 vm_map_t map; 529 vm_object_t object; 530 vm_offset_t off, map_addr; 531 int error, rv, cow; 532 size_t copy_len; 533 vm_ooffset_t file_addr; 534 535 /* 536 * It's necessary to fail if the filsz + offset taken from the 537 * header is greater than the actual file pager object's size. 538 * If we were to allow this, then the vm_map_find() below would 539 * walk right off the end of the file object and into the ether. 540 * 541 * While I'm here, might as well check for something else that 542 * is invalid: filsz cannot be greater than memsz. 543 */ 544 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 545 filsz > memsz) { 546 uprintf("elf_load_section: truncated ELF file\n"); 547 return (ENOEXEC); 548 } 549 550 object = imgp->object; 551 map = &imgp->proc->p_vmspace->vm_map; 552 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 553 file_addr = trunc_page_ps(offset, pagesize); 554 555 /* 556 * We have two choices. We can either clear the data in the last page 557 * of an oversized mapping, or we can start the anon mapping a page 558 * early and copy the initialized data into that first page. We 559 * choose the second. 560 */ 561 if (filsz == 0) 562 map_len = 0; 563 else if (memsz > filsz) 564 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 565 else 566 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 567 568 if (map_len != 0) { 569 /* cow flags: don't dump readonly sections in core */ 570 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 571 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 572 573 rv = __elfN(map_insert)(imgp, map, 574 object, 575 file_addr, /* file offset */ 576 map_addr, /* virtual start */ 577 map_addr + map_len,/* virtual end */ 578 prot, 579 cow); 580 if (rv != KERN_SUCCESS) 581 return (EINVAL); 582 583 /* we can stop now if we've covered it all */ 584 if (memsz == filsz) 585 return (0); 586 } 587 588 589 /* 590 * We have to get the remaining bit of the file into the first part 591 * of the oversized map segment. This is normally because the .data 592 * segment in the file is extended to provide bss. It's a neat idea 593 * to try and save a page, but it's a pain in the behind to implement. 594 */ 595 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset + 596 filsz, pagesize); 597 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 598 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 599 map_addr; 600 601 /* This had damn well better be true! */ 602 if (map_len != 0) { 603 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 604 map_addr + map_len, prot, 0); 605 if (rv != KERN_SUCCESS) 606 return (EINVAL); 607 } 608 609 if (copy_len != 0) { 610 sf = vm_imgact_map_page(object, offset + filsz); 611 if (sf == NULL) 612 return (EIO); 613 614 /* send the page fragment to user space */ 615 off = trunc_page_ps(offset + filsz, pagesize) - 616 trunc_page(offset + filsz); 617 error = copyout((caddr_t)sf_buf_kva(sf) + off, 618 (caddr_t)map_addr, copy_len); 619 vm_imgact_unmap_page(sf); 620 if (error != 0) 621 return (error); 622 } 623 624 /* 625 * Remove write access to the page if it was only granted by map_insert 626 * to allow copyout. 627 */ 628 if ((prot & VM_PROT_WRITE) == 0) 629 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 630 map_len), prot, FALSE); 631 632 return (0); 633 } 634 635 /* 636 * Load the file "file" into memory. It may be either a shared object 637 * or an executable. 638 * 639 * The "addr" reference parameter is in/out. On entry, it specifies 640 * the address where a shared object should be loaded. If the file is 641 * an executable, this value is ignored. On exit, "addr" specifies 642 * where the file was actually loaded. 643 * 644 * The "entry" reference parameter is out only. On exit, it specifies 645 * the entry point for the loaded file. 646 */ 647 static int 648 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 649 u_long *entry, size_t pagesize) 650 { 651 struct { 652 struct nameidata nd; 653 struct vattr attr; 654 struct image_params image_params; 655 } *tempdata; 656 const Elf_Ehdr *hdr = NULL; 657 const Elf_Phdr *phdr = NULL; 658 struct nameidata *nd; 659 struct vattr *attr; 660 struct image_params *imgp; 661 vm_prot_t prot; 662 u_long rbase; 663 u_long base_addr = 0; 664 int error, i, numsegs; 665 666 #ifdef CAPABILITY_MODE 667 /* 668 * XXXJA: This check can go away once we are sufficiently confident 669 * that the checks in namei() are correct. 670 */ 671 if (IN_CAPABILITY_MODE(curthread)) 672 return (ECAPMODE); 673 #endif 674 675 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 676 nd = &tempdata->nd; 677 attr = &tempdata->attr; 678 imgp = &tempdata->image_params; 679 680 /* 681 * Initialize part of the common data 682 */ 683 imgp->proc = p; 684 imgp->attr = attr; 685 imgp->firstpage = NULL; 686 imgp->image_header = NULL; 687 imgp->object = NULL; 688 imgp->execlabel = NULL; 689 690 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); 691 if ((error = namei(nd)) != 0) { 692 nd->ni_vp = NULL; 693 goto fail; 694 } 695 NDFREE(nd, NDF_ONLY_PNBUF); 696 imgp->vp = nd->ni_vp; 697 698 /* 699 * Check permissions, modes, uid, etc on the file, and "open" it. 700 */ 701 error = exec_check_permissions(imgp); 702 if (error) 703 goto fail; 704 705 error = exec_map_first_page(imgp); 706 if (error) 707 goto fail; 708 709 /* 710 * Also make certain that the interpreter stays the same, so set 711 * its VV_TEXT flag, too. 712 */ 713 VOP_SET_TEXT(nd->ni_vp); 714 715 imgp->object = nd->ni_vp->v_object; 716 717 hdr = (const Elf_Ehdr *)imgp->image_header; 718 if ((error = __elfN(check_header)(hdr)) != 0) 719 goto fail; 720 if (hdr->e_type == ET_DYN) 721 rbase = *addr; 722 else if (hdr->e_type == ET_EXEC) 723 rbase = 0; 724 else { 725 error = ENOEXEC; 726 goto fail; 727 } 728 729 /* Only support headers that fit within first page for now */ 730 if ((hdr->e_phoff > PAGE_SIZE) || 731 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 732 error = ENOEXEC; 733 goto fail; 734 } 735 736 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 737 if (!aligned(phdr, Elf_Addr)) { 738 error = ENOEXEC; 739 goto fail; 740 } 741 742 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 743 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 744 /* Loadable segment */ 745 prot = __elfN(trans_prot)(phdr[i].p_flags); 746 error = __elfN(load_section)(imgp, phdr[i].p_offset, 747 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 748 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); 749 if (error != 0) 750 goto fail; 751 /* 752 * Establish the base address if this is the 753 * first segment. 754 */ 755 if (numsegs == 0) 756 base_addr = trunc_page(phdr[i].p_vaddr + 757 rbase); 758 numsegs++; 759 } 760 } 761 *addr = base_addr; 762 *entry = (unsigned long)hdr->e_entry + rbase; 763 764 fail: 765 if (imgp->firstpage) 766 exec_unmap_first_page(imgp); 767 768 if (nd->ni_vp) 769 vput(nd->ni_vp); 770 771 free(tempdata, M_TEMP); 772 773 return (error); 774 } 775 776 static int 777 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 778 { 779 struct thread *td; 780 const Elf_Ehdr *hdr; 781 const Elf_Phdr *phdr; 782 Elf_Auxargs *elf_auxargs; 783 struct vmspace *vmspace; 784 const char *err_str, *newinterp; 785 char *interp, *interp_buf, *path; 786 Elf_Brandinfo *brand_info; 787 struct sysentvec *sv; 788 vm_prot_t prot; 789 u_long text_size, data_size, total_size, text_addr, data_addr; 790 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr; 791 int32_t osrel; 792 int error, i, n, interp_name_len, have_interp; 793 794 hdr = (const Elf_Ehdr *)imgp->image_header; 795 796 /* 797 * Do we have a valid ELF header ? 798 * 799 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 800 * if particular brand doesn't support it. 801 */ 802 if (__elfN(check_header)(hdr) != 0 || 803 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 804 return (-1); 805 806 /* 807 * From here on down, we return an errno, not -1, as we've 808 * detected an ELF file. 809 */ 810 811 if ((hdr->e_phoff > PAGE_SIZE) || 812 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 813 /* Only support headers in first page for now */ 814 uprintf("Program headers not in the first page\n"); 815 return (ENOEXEC); 816 } 817 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 818 if (!aligned(phdr, Elf_Addr)) { 819 uprintf("Unaligned program headers\n"); 820 return (ENOEXEC); 821 } 822 823 n = error = 0; 824 baddr = 0; 825 osrel = 0; 826 text_size = data_size = total_size = text_addr = data_addr = 0; 827 entry = proghdr = 0; 828 interp_name_len = 0; 829 err_str = newinterp = NULL; 830 interp = interp_buf = NULL; 831 td = curthread; 832 833 for (i = 0; i < hdr->e_phnum; i++) { 834 switch (phdr[i].p_type) { 835 case PT_LOAD: 836 if (n == 0) 837 baddr = phdr[i].p_vaddr; 838 n++; 839 break; 840 case PT_INTERP: 841 /* Path to interpreter */ 842 if (phdr[i].p_filesz > MAXPATHLEN) { 843 uprintf("Invalid PT_INTERP\n"); 844 error = ENOEXEC; 845 goto ret; 846 } 847 if (interp != NULL) { 848 uprintf("Multiple PT_INTERP headers\n"); 849 error = ENOEXEC; 850 goto ret; 851 } 852 interp_name_len = phdr[i].p_filesz; 853 if (phdr[i].p_offset > PAGE_SIZE || 854 interp_name_len > PAGE_SIZE - phdr[i].p_offset) { 855 VOP_UNLOCK(imgp->vp, 0); 856 interp_buf = malloc(interp_name_len + 1, M_TEMP, 857 M_WAITOK); 858 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 859 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf, 860 interp_name_len, phdr[i].p_offset, 861 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 862 NOCRED, NULL, td); 863 if (error != 0) { 864 uprintf("i/o error PT_INTERP %d\n", 865 error); 866 goto ret; 867 } 868 interp_buf[interp_name_len] = '\0'; 869 interp = interp_buf; 870 } else { 871 interp = __DECONST(char *, imgp->image_header) + 872 phdr[i].p_offset; 873 } 874 break; 875 case PT_GNU_STACK: 876 if (__elfN(nxstack)) 877 imgp->stack_prot = 878 __elfN(trans_prot)(phdr[i].p_flags); 879 imgp->stack_sz = phdr[i].p_memsz; 880 break; 881 } 882 } 883 884 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 885 &osrel); 886 if (brand_info == NULL) { 887 uprintf("ELF binary type \"%u\" not known.\n", 888 hdr->e_ident[EI_OSABI]); 889 error = ENOEXEC; 890 goto ret; 891 } 892 et_dyn_addr = 0; 893 if (hdr->e_type == ET_DYN) { 894 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 895 uprintf("Cannot execute shared object\n"); 896 error = ENOEXEC; 897 goto ret; 898 } 899 /* 900 * Honour the base load address from the dso if it is 901 * non-zero for some reason. 902 */ 903 if (baddr == 0) 904 et_dyn_addr = ET_DYN_LOAD_ADDR; 905 } 906 sv = brand_info->sysvec; 907 if (interp != NULL && brand_info->interp_newpath != NULL) 908 newinterp = brand_info->interp_newpath; 909 910 /* 911 * Avoid a possible deadlock if the current address space is destroyed 912 * and that address space maps the locked vnode. In the common case, 913 * the locked vnode's v_usecount is decremented but remains greater 914 * than zero. Consequently, the vnode lock is not needed by vrele(). 915 * However, in cases where the vnode lock is external, such as nullfs, 916 * v_usecount may become zero. 917 * 918 * The VV_TEXT flag prevents modifications to the executable while 919 * the vnode is unlocked. 920 */ 921 VOP_UNLOCK(imgp->vp, 0); 922 923 error = exec_new_vmspace(imgp, sv); 924 imgp->proc->p_sysent = sv; 925 926 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 927 if (error != 0) 928 goto ret; 929 930 for (i = 0; i < hdr->e_phnum; i++) { 931 switch (phdr[i].p_type) { 932 case PT_LOAD: /* Loadable segment */ 933 if (phdr[i].p_memsz == 0) 934 break; 935 prot = __elfN(trans_prot)(phdr[i].p_flags); 936 error = __elfN(load_section)(imgp, phdr[i].p_offset, 937 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 938 phdr[i].p_memsz, phdr[i].p_filesz, prot, 939 sv->sv_pagesize); 940 if (error != 0) 941 goto ret; 942 943 /* 944 * If this segment contains the program headers, 945 * remember their virtual address for the AT_PHDR 946 * aux entry. Static binaries don't usually include 947 * a PT_PHDR entry. 948 */ 949 if (phdr[i].p_offset == 0 && 950 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 951 <= phdr[i].p_filesz) 952 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 953 et_dyn_addr; 954 955 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 956 seg_size = round_page(phdr[i].p_memsz + 957 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 958 959 /* 960 * Make the largest executable segment the official 961 * text segment and all others data. 962 * 963 * Note that obreak() assumes that data_addr + 964 * data_size == end of data load area, and the ELF 965 * file format expects segments to be sorted by 966 * address. If multiple data segments exist, the 967 * last one will be used. 968 */ 969 970 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 971 text_size = seg_size; 972 text_addr = seg_addr; 973 } else { 974 data_size = seg_size; 975 data_addr = seg_addr; 976 } 977 total_size += seg_size; 978 break; 979 case PT_PHDR: /* Program header table info */ 980 proghdr = phdr[i].p_vaddr + et_dyn_addr; 981 break; 982 default: 983 break; 984 } 985 } 986 987 if (data_addr == 0 && data_size == 0) { 988 data_addr = text_addr; 989 data_size = text_size; 990 } 991 992 entry = (u_long)hdr->e_entry + et_dyn_addr; 993 994 /* 995 * Check limits. It should be safe to check the 996 * limits after loading the segments since we do 997 * not actually fault in all the segments pages. 998 */ 999 PROC_LOCK(imgp->proc); 1000 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 1001 err_str = "Data segment size exceeds process limit"; 1002 else if (text_size > maxtsiz) 1003 err_str = "Text segment size exceeds system limit"; 1004 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 1005 err_str = "Total segment size exceeds process limit"; 1006 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 1007 err_str = "Data segment size exceeds resource limit"; 1008 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 1009 err_str = "Total segment size exceeds resource limit"; 1010 if (err_str != NULL) { 1011 PROC_UNLOCK(imgp->proc); 1012 uprintf("%s\n", err_str); 1013 error = ENOMEM; 1014 goto ret; 1015 } 1016 1017 vmspace = imgp->proc->p_vmspace; 1018 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 1019 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 1020 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 1021 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 1022 1023 /* 1024 * We load the dynamic linker where a userland call 1025 * to mmap(0, ...) would put it. The rationale behind this 1026 * calculation is that it leaves room for the heap to grow to 1027 * its maximum allowed size. 1028 */ 1029 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1030 RLIMIT_DATA)); 1031 PROC_UNLOCK(imgp->proc); 1032 1033 imgp->entry_addr = entry; 1034 1035 if (interp != NULL) { 1036 have_interp = FALSE; 1037 VOP_UNLOCK(imgp->vp, 0); 1038 if (brand_info->emul_path != NULL && 1039 brand_info->emul_path[0] != '\0') { 1040 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1041 snprintf(path, MAXPATHLEN, "%s%s", 1042 brand_info->emul_path, interp); 1043 error = __elfN(load_file)(imgp->proc, path, &addr, 1044 &imgp->entry_addr, sv->sv_pagesize); 1045 free(path, M_TEMP); 1046 if (error == 0) 1047 have_interp = TRUE; 1048 } 1049 if (!have_interp && newinterp != NULL && 1050 (brand_info->interp_path == NULL || 1051 strcmp(interp, brand_info->interp_path) == 0)) { 1052 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 1053 &imgp->entry_addr, sv->sv_pagesize); 1054 if (error == 0) 1055 have_interp = TRUE; 1056 } 1057 if (!have_interp) { 1058 error = __elfN(load_file)(imgp->proc, interp, &addr, 1059 &imgp->entry_addr, sv->sv_pagesize); 1060 } 1061 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 1062 if (error != 0) { 1063 uprintf("ELF interpreter %s not found, error %d\n", 1064 interp, error); 1065 goto ret; 1066 } 1067 } else 1068 addr = et_dyn_addr; 1069 1070 /* 1071 * Construct auxargs table (used by the fixup routine) 1072 */ 1073 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1074 elf_auxargs->execfd = -1; 1075 elf_auxargs->phdr = proghdr; 1076 elf_auxargs->phent = hdr->e_phentsize; 1077 elf_auxargs->phnum = hdr->e_phnum; 1078 elf_auxargs->pagesz = PAGE_SIZE; 1079 elf_auxargs->base = addr; 1080 elf_auxargs->flags = 0; 1081 elf_auxargs->entry = entry; 1082 elf_auxargs->hdr_eflags = hdr->e_flags; 1083 1084 imgp->auxargs = elf_auxargs; 1085 imgp->interpreted = 0; 1086 imgp->reloc_base = addr; 1087 imgp->proc->p_osrel = osrel; 1088 imgp->proc->p_elf_machine = hdr->e_machine; 1089 imgp->proc->p_elf_flags = hdr->e_flags; 1090 1091 ret: 1092 free(interp_buf, M_TEMP); 1093 return (error); 1094 } 1095 1096 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 1097 1098 int 1099 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 1100 { 1101 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1102 Elf_Auxinfo *argarray, *pos; 1103 Elf_Addr *base, *auxbase; 1104 int error; 1105 1106 base = (Elf_Addr *)*stack_base; 1107 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1; 1108 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1109 M_WAITOK | M_ZERO); 1110 1111 if (args->execfd != -1) 1112 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1113 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1114 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1115 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1116 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1117 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1118 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1119 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1120 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1121 if (imgp->execpathp != 0) 1122 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 1123 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1124 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1125 if (imgp->canary != 0) { 1126 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 1127 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1128 } 1129 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1130 if (imgp->pagesizes != 0) { 1131 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1132 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1133 } 1134 if (imgp->sysent->sv_timekeep_base != 0) { 1135 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1136 imgp->sysent->sv_timekeep_base); 1137 } 1138 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1139 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1140 imgp->sysent->sv_stackprot); 1141 if (imgp->sysent->sv_hwcap != NULL) 1142 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1143 if (imgp->sysent->sv_hwcap2 != NULL) 1144 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1145 AUXARGS_ENTRY(pos, AT_NULL, 0); 1146 1147 free(imgp->auxargs, M_TEMP); 1148 imgp->auxargs = NULL; 1149 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1150 1151 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT); 1152 free(argarray, M_TEMP); 1153 if (error != 0) 1154 return (error); 1155 1156 base--; 1157 if (suword(base, imgp->args->argc) == -1) 1158 return (EFAULT); 1159 *stack_base = (register_t *)base; 1160 return (0); 1161 } 1162 1163 /* 1164 * Code for generating ELF core dumps. 1165 */ 1166 1167 typedef void (*segment_callback)(vm_map_entry_t, void *); 1168 1169 /* Closure for cb_put_phdr(). */ 1170 struct phdr_closure { 1171 Elf_Phdr *phdr; /* Program header to fill in */ 1172 Elf_Off offset; /* Offset of segment in core file */ 1173 }; 1174 1175 /* Closure for cb_size_segment(). */ 1176 struct sseg_closure { 1177 int count; /* Count of writable segments. */ 1178 size_t size; /* Total size of all writable segments. */ 1179 }; 1180 1181 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); 1182 1183 struct note_info { 1184 int type; /* Note type. */ 1185 outfunc_t outfunc; /* Output function. */ 1186 void *outarg; /* Argument for the output function. */ 1187 size_t outsize; /* Output size. */ 1188 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1189 }; 1190 1191 TAILQ_HEAD(note_info_list, note_info); 1192 1193 /* Coredump output parameters. */ 1194 struct coredump_params { 1195 off_t offset; 1196 struct ucred *active_cred; 1197 struct ucred *file_cred; 1198 struct thread *td; 1199 struct vnode *vp; 1200 struct compressor *comp; 1201 }; 1202 1203 extern int compress_user_cores; 1204 extern int compress_user_cores_level; 1205 1206 static void cb_put_phdr(vm_map_entry_t, void *); 1207 static void cb_size_segment(vm_map_entry_t, void *); 1208 static int core_write(struct coredump_params *, const void *, size_t, off_t, 1209 enum uio_seg); 1210 static void each_dumpable_segment(struct thread *, segment_callback, void *); 1211 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1212 struct note_info_list *, size_t); 1213 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, 1214 size_t *); 1215 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); 1216 static void __elfN(putnote)(struct note_info *, struct sbuf *); 1217 static size_t register_note(struct note_info_list *, int, outfunc_t, void *); 1218 static int sbuf_drain_core_output(void *, const char *, int); 1219 static int sbuf_drain_count(void *arg, const char *data, int len); 1220 1221 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); 1222 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1223 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); 1224 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1225 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); 1226 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *); 1227 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1228 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1229 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1230 static void note_procstat_files(void *, struct sbuf *, size_t *); 1231 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1232 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1233 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1234 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1235 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1236 1237 /* 1238 * Write out a core segment to the compression stream. 1239 */ 1240 static int 1241 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len) 1242 { 1243 u_int chunk_len; 1244 int error; 1245 1246 while (len > 0) { 1247 chunk_len = MIN(len, CORE_BUF_SIZE); 1248 1249 /* 1250 * We can get EFAULT error here. 1251 * In that case zero out the current chunk of the segment. 1252 */ 1253 error = copyin(base, buf, chunk_len); 1254 if (error != 0) 1255 bzero(buf, chunk_len); 1256 error = compressor_write(p->comp, buf, chunk_len); 1257 if (error != 0) 1258 break; 1259 base += chunk_len; 1260 len -= chunk_len; 1261 } 1262 return (error); 1263 } 1264 1265 static int 1266 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1267 { 1268 1269 return (core_write((struct coredump_params *)arg, base, len, offset, 1270 UIO_SYSSPACE)); 1271 } 1272 1273 static int 1274 core_write(struct coredump_params *p, const void *base, size_t len, 1275 off_t offset, enum uio_seg seg) 1276 { 1277 1278 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base), 1279 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1280 p->active_cred, p->file_cred, NULL, p->td)); 1281 } 1282 1283 static int 1284 core_output(void *base, size_t len, off_t offset, struct coredump_params *p, 1285 void *tmpbuf) 1286 { 1287 int error; 1288 1289 if (p->comp != NULL) 1290 return (compress_chunk(p, base, tmpbuf, len)); 1291 1292 /* 1293 * EFAULT is a non-fatal error that we can get, for example, 1294 * if the segment is backed by a file but extends beyond its 1295 * end. 1296 */ 1297 error = core_write(p, base, len, offset, UIO_USERSPACE); 1298 if (error == EFAULT) { 1299 log(LOG_WARNING, "Failed to fully fault in a core file segment " 1300 "at VA %p with size 0x%zx to be written at offset 0x%jx " 1301 "for process %s\n", base, len, offset, curproc->p_comm); 1302 1303 /* 1304 * Write a "real" zero byte at the end of the target region 1305 * in the case this is the last segment. 1306 * The intermediate space will be implicitly zero-filled. 1307 */ 1308 error = core_write(p, zero_region, 1, offset + len - 1, 1309 UIO_SYSSPACE); 1310 } 1311 return (error); 1312 } 1313 1314 /* 1315 * Drain into a core file. 1316 */ 1317 static int 1318 sbuf_drain_core_output(void *arg, const char *data, int len) 1319 { 1320 struct coredump_params *p; 1321 int error, locked; 1322 1323 p = (struct coredump_params *)arg; 1324 1325 /* 1326 * Some kern_proc out routines that print to this sbuf may 1327 * call us with the process lock held. Draining with the 1328 * non-sleepable lock held is unsafe. The lock is needed for 1329 * those routines when dumping a live process. In our case we 1330 * can safely release the lock before draining and acquire 1331 * again after. 1332 */ 1333 locked = PROC_LOCKED(p->td->td_proc); 1334 if (locked) 1335 PROC_UNLOCK(p->td->td_proc); 1336 if (p->comp != NULL) 1337 error = compressor_write(p->comp, __DECONST(char *, data), len); 1338 else 1339 error = core_write(p, __DECONST(void *, data), len, p->offset, 1340 UIO_SYSSPACE); 1341 if (locked) 1342 PROC_LOCK(p->td->td_proc); 1343 if (error != 0) 1344 return (-error); 1345 p->offset += len; 1346 return (len); 1347 } 1348 1349 /* 1350 * Drain into a counter. 1351 */ 1352 static int 1353 sbuf_drain_count(void *arg, const char *data __unused, int len) 1354 { 1355 size_t *sizep; 1356 1357 sizep = (size_t *)arg; 1358 *sizep += len; 1359 return (len); 1360 } 1361 1362 int 1363 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1364 { 1365 struct ucred *cred = td->td_ucred; 1366 int error = 0; 1367 struct sseg_closure seginfo; 1368 struct note_info_list notelst; 1369 struct coredump_params params; 1370 struct note_info *ninfo; 1371 void *hdr, *tmpbuf; 1372 size_t hdrsize, notesz, coresize; 1373 1374 hdr = NULL; 1375 tmpbuf = NULL; 1376 TAILQ_INIT(¬elst); 1377 1378 /* Size the program segments. */ 1379 seginfo.count = 0; 1380 seginfo.size = 0; 1381 each_dumpable_segment(td, cb_size_segment, &seginfo); 1382 1383 /* 1384 * Collect info about the core file header area. 1385 */ 1386 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1387 if (seginfo.count + 1 >= PN_XNUM) 1388 hdrsize += sizeof(Elf_Shdr); 1389 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1390 coresize = round_page(hdrsize + notesz) + seginfo.size; 1391 1392 /* Set up core dump parameters. */ 1393 params.offset = 0; 1394 params.active_cred = cred; 1395 params.file_cred = NOCRED; 1396 params.td = td; 1397 params.vp = vp; 1398 params.comp = NULL; 1399 1400 #ifdef RACCT 1401 if (racct_enable) { 1402 PROC_LOCK(td->td_proc); 1403 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1404 PROC_UNLOCK(td->td_proc); 1405 if (error != 0) { 1406 error = EFAULT; 1407 goto done; 1408 } 1409 } 1410 #endif 1411 if (coresize >= limit) { 1412 error = EFAULT; 1413 goto done; 1414 } 1415 1416 /* Create a compression stream if necessary. */ 1417 if (compress_user_cores != 0) { 1418 params.comp = compressor_init(core_compressed_write, 1419 compress_user_cores, CORE_BUF_SIZE, 1420 compress_user_cores_level, ¶ms); 1421 if (params.comp == NULL) { 1422 error = EFAULT; 1423 goto done; 1424 } 1425 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1426 } 1427 1428 /* 1429 * Allocate memory for building the header, fill it up, 1430 * and write it out following the notes. 1431 */ 1432 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1433 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1434 notesz); 1435 1436 /* Write the contents of all of the writable segments. */ 1437 if (error == 0) { 1438 Elf_Phdr *php; 1439 off_t offset; 1440 int i; 1441 1442 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1443 offset = round_page(hdrsize + notesz); 1444 for (i = 0; i < seginfo.count; i++) { 1445 error = core_output((caddr_t)(uintptr_t)php->p_vaddr, 1446 php->p_filesz, offset, ¶ms, tmpbuf); 1447 if (error != 0) 1448 break; 1449 offset += php->p_filesz; 1450 php++; 1451 } 1452 if (error == 0 && params.comp != NULL) 1453 error = compressor_flush(params.comp); 1454 } 1455 if (error) { 1456 log(LOG_WARNING, 1457 "Failed to write core file for process %s (error %d)\n", 1458 curproc->p_comm, error); 1459 } 1460 1461 done: 1462 free(tmpbuf, M_TEMP); 1463 if (params.comp != NULL) 1464 compressor_fini(params.comp); 1465 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1466 TAILQ_REMOVE(¬elst, ninfo, link); 1467 free(ninfo, M_TEMP); 1468 } 1469 if (hdr != NULL) 1470 free(hdr, M_TEMP); 1471 1472 return (error); 1473 } 1474 1475 /* 1476 * A callback for each_dumpable_segment() to write out the segment's 1477 * program header entry. 1478 */ 1479 static void 1480 cb_put_phdr(vm_map_entry_t entry, void *closure) 1481 { 1482 struct phdr_closure *phc = (struct phdr_closure *)closure; 1483 Elf_Phdr *phdr = phc->phdr; 1484 1485 phc->offset = round_page(phc->offset); 1486 1487 phdr->p_type = PT_LOAD; 1488 phdr->p_offset = phc->offset; 1489 phdr->p_vaddr = entry->start; 1490 phdr->p_paddr = 0; 1491 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1492 phdr->p_align = PAGE_SIZE; 1493 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1494 1495 phc->offset += phdr->p_filesz; 1496 phc->phdr++; 1497 } 1498 1499 /* 1500 * A callback for each_dumpable_segment() to gather information about 1501 * the number of segments and their total size. 1502 */ 1503 static void 1504 cb_size_segment(vm_map_entry_t entry, void *closure) 1505 { 1506 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1507 1508 ssc->count++; 1509 ssc->size += entry->end - entry->start; 1510 } 1511 1512 /* 1513 * For each writable segment in the process's memory map, call the given 1514 * function with a pointer to the map entry and some arbitrary 1515 * caller-supplied data. 1516 */ 1517 static void 1518 each_dumpable_segment(struct thread *td, segment_callback func, void *closure) 1519 { 1520 struct proc *p = td->td_proc; 1521 vm_map_t map = &p->p_vmspace->vm_map; 1522 vm_map_entry_t entry; 1523 vm_object_t backing_object, object; 1524 boolean_t ignore_entry; 1525 1526 vm_map_lock_read(map); 1527 for (entry = map->header.next; entry != &map->header; 1528 entry = entry->next) { 1529 /* 1530 * Don't dump inaccessible mappings, deal with legacy 1531 * coredump mode. 1532 * 1533 * Note that read-only segments related to the elf binary 1534 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1535 * need to arbitrarily ignore such segments. 1536 */ 1537 if (elf_legacy_coredump) { 1538 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1539 continue; 1540 } else { 1541 if ((entry->protection & VM_PROT_ALL) == 0) 1542 continue; 1543 } 1544 1545 /* 1546 * Dont include memory segment in the coredump if 1547 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1548 * madvise(2). Do not dump submaps (i.e. parts of the 1549 * kernel map). 1550 */ 1551 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1552 continue; 1553 1554 if ((object = entry->object.vm_object) == NULL) 1555 continue; 1556 1557 /* Ignore memory-mapped devices and such things. */ 1558 VM_OBJECT_RLOCK(object); 1559 while ((backing_object = object->backing_object) != NULL) { 1560 VM_OBJECT_RLOCK(backing_object); 1561 VM_OBJECT_RUNLOCK(object); 1562 object = backing_object; 1563 } 1564 ignore_entry = object->type != OBJT_DEFAULT && 1565 object->type != OBJT_SWAP && object->type != OBJT_VNODE && 1566 object->type != OBJT_PHYS; 1567 VM_OBJECT_RUNLOCK(object); 1568 if (ignore_entry) 1569 continue; 1570 1571 (*func)(entry, closure); 1572 } 1573 vm_map_unlock_read(map); 1574 } 1575 1576 /* 1577 * Write the core file header to the file, including padding up to 1578 * the page boundary. 1579 */ 1580 static int 1581 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1582 size_t hdrsize, struct note_info_list *notelst, size_t notesz) 1583 { 1584 struct note_info *ninfo; 1585 struct sbuf *sb; 1586 int error; 1587 1588 /* Fill in the header. */ 1589 bzero(hdr, hdrsize); 1590 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz); 1591 1592 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1593 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1594 sbuf_start_section(sb, NULL); 1595 sbuf_bcat(sb, hdr, hdrsize); 1596 TAILQ_FOREACH(ninfo, notelst, link) 1597 __elfN(putnote)(ninfo, sb); 1598 /* Align up to a page boundary for the program segments. */ 1599 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1600 error = sbuf_finish(sb); 1601 sbuf_delete(sb); 1602 1603 return (error); 1604 } 1605 1606 static void 1607 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1608 size_t *sizep) 1609 { 1610 struct proc *p; 1611 struct thread *thr; 1612 size_t size; 1613 1614 p = td->td_proc; 1615 size = 0; 1616 1617 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1618 1619 /* 1620 * To have the debugger select the right thread (LWP) as the initial 1621 * thread, we dump the state of the thread passed to us in td first. 1622 * This is the thread that causes the core dump and thus likely to 1623 * be the right thread one wants to have selected in the debugger. 1624 */ 1625 thr = td; 1626 while (thr != NULL) { 1627 size += register_note(list, NT_PRSTATUS, 1628 __elfN(note_prstatus), thr); 1629 size += register_note(list, NT_FPREGSET, 1630 __elfN(note_fpregset), thr); 1631 size += register_note(list, NT_THRMISC, 1632 __elfN(note_thrmisc), thr); 1633 size += register_note(list, NT_PTLWPINFO, 1634 __elfN(note_ptlwpinfo), thr); 1635 size += register_note(list, -1, 1636 __elfN(note_threadmd), thr); 1637 1638 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1639 TAILQ_NEXT(thr, td_plist); 1640 if (thr == td) 1641 thr = TAILQ_NEXT(thr, td_plist); 1642 } 1643 1644 size += register_note(list, NT_PROCSTAT_PROC, 1645 __elfN(note_procstat_proc), p); 1646 size += register_note(list, NT_PROCSTAT_FILES, 1647 note_procstat_files, p); 1648 size += register_note(list, NT_PROCSTAT_VMMAP, 1649 note_procstat_vmmap, p); 1650 size += register_note(list, NT_PROCSTAT_GROUPS, 1651 note_procstat_groups, p); 1652 size += register_note(list, NT_PROCSTAT_UMASK, 1653 note_procstat_umask, p); 1654 size += register_note(list, NT_PROCSTAT_RLIMIT, 1655 note_procstat_rlimit, p); 1656 size += register_note(list, NT_PROCSTAT_OSREL, 1657 note_procstat_osrel, p); 1658 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1659 __elfN(note_procstat_psstrings), p); 1660 size += register_note(list, NT_PROCSTAT_AUXV, 1661 __elfN(note_procstat_auxv), p); 1662 1663 *sizep = size; 1664 } 1665 1666 static void 1667 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1668 size_t notesz) 1669 { 1670 Elf_Ehdr *ehdr; 1671 Elf_Phdr *phdr; 1672 Elf_Shdr *shdr; 1673 struct phdr_closure phc; 1674 1675 ehdr = (Elf_Ehdr *)hdr; 1676 1677 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1678 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1679 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1680 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1681 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1682 ehdr->e_ident[EI_DATA] = ELF_DATA; 1683 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1684 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1685 ehdr->e_ident[EI_ABIVERSION] = 0; 1686 ehdr->e_ident[EI_PAD] = 0; 1687 ehdr->e_type = ET_CORE; 1688 ehdr->e_machine = td->td_proc->p_elf_machine; 1689 ehdr->e_version = EV_CURRENT; 1690 ehdr->e_entry = 0; 1691 ehdr->e_phoff = sizeof(Elf_Ehdr); 1692 ehdr->e_flags = td->td_proc->p_elf_flags; 1693 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1694 ehdr->e_phentsize = sizeof(Elf_Phdr); 1695 ehdr->e_shentsize = sizeof(Elf_Shdr); 1696 ehdr->e_shstrndx = SHN_UNDEF; 1697 if (numsegs + 1 < PN_XNUM) { 1698 ehdr->e_phnum = numsegs + 1; 1699 ehdr->e_shnum = 0; 1700 } else { 1701 ehdr->e_phnum = PN_XNUM; 1702 ehdr->e_shnum = 1; 1703 1704 ehdr->e_shoff = ehdr->e_phoff + 1705 (numsegs + 1) * ehdr->e_phentsize; 1706 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1707 ("e_shoff: %zu, hdrsize - shdr: %zu", 1708 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1709 1710 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1711 memset(shdr, 0, sizeof(*shdr)); 1712 /* 1713 * A special first section is used to hold large segment and 1714 * section counts. This was proposed by Sun Microsystems in 1715 * Solaris and has been adopted by Linux; the standard ELF 1716 * tools are already familiar with the technique. 1717 * 1718 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1719 * (or 12-7 depending on the version of the document) for more 1720 * details. 1721 */ 1722 shdr->sh_type = SHT_NULL; 1723 shdr->sh_size = ehdr->e_shnum; 1724 shdr->sh_link = ehdr->e_shstrndx; 1725 shdr->sh_info = numsegs + 1; 1726 } 1727 1728 /* 1729 * Fill in the program header entries. 1730 */ 1731 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1732 1733 /* The note segement. */ 1734 phdr->p_type = PT_NOTE; 1735 phdr->p_offset = hdrsize; 1736 phdr->p_vaddr = 0; 1737 phdr->p_paddr = 0; 1738 phdr->p_filesz = notesz; 1739 phdr->p_memsz = 0; 1740 phdr->p_flags = PF_R; 1741 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1742 phdr++; 1743 1744 /* All the writable segments from the program. */ 1745 phc.phdr = phdr; 1746 phc.offset = round_page(hdrsize + notesz); 1747 each_dumpable_segment(td, cb_put_phdr, &phc); 1748 } 1749 1750 static size_t 1751 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) 1752 { 1753 struct note_info *ninfo; 1754 size_t size, notesize; 1755 1756 size = 0; 1757 out(arg, NULL, &size); 1758 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1759 ninfo->type = type; 1760 ninfo->outfunc = out; 1761 ninfo->outarg = arg; 1762 ninfo->outsize = size; 1763 TAILQ_INSERT_TAIL(list, ninfo, link); 1764 1765 if (type == -1) 1766 return (size); 1767 1768 notesize = sizeof(Elf_Note) + /* note header */ 1769 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1770 /* note name */ 1771 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1772 1773 return (notesize); 1774 } 1775 1776 static size_t 1777 append_note_data(const void *src, void *dst, size_t len) 1778 { 1779 size_t padded_len; 1780 1781 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 1782 if (dst != NULL) { 1783 bcopy(src, dst, len); 1784 bzero((char *)dst + len, padded_len - len); 1785 } 1786 return (padded_len); 1787 } 1788 1789 size_t 1790 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 1791 { 1792 Elf_Note *note; 1793 char *buf; 1794 size_t notesize; 1795 1796 buf = dst; 1797 if (buf != NULL) { 1798 note = (Elf_Note *)buf; 1799 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1800 note->n_descsz = size; 1801 note->n_type = type; 1802 buf += sizeof(*note); 1803 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 1804 sizeof(FREEBSD_ABI_VENDOR)); 1805 append_note_data(src, buf, size); 1806 if (descp != NULL) 1807 *descp = buf; 1808 } 1809 1810 notesize = sizeof(Elf_Note) + /* note header */ 1811 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1812 /* note name */ 1813 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1814 1815 return (notesize); 1816 } 1817 1818 static void 1819 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) 1820 { 1821 Elf_Note note; 1822 ssize_t old_len, sect_len; 1823 size_t new_len, descsz, i; 1824 1825 if (ninfo->type == -1) { 1826 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1827 return; 1828 } 1829 1830 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1831 note.n_descsz = ninfo->outsize; 1832 note.n_type = ninfo->type; 1833 1834 sbuf_bcat(sb, ¬e, sizeof(note)); 1835 sbuf_start_section(sb, &old_len); 1836 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR)); 1837 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1838 if (note.n_descsz == 0) 1839 return; 1840 sbuf_start_section(sb, &old_len); 1841 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1842 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1843 if (sect_len < 0) 1844 return; 1845 1846 new_len = (size_t)sect_len; 1847 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 1848 if (new_len < descsz) { 1849 /* 1850 * It is expected that individual note emitters will correctly 1851 * predict their expected output size and fill up to that size 1852 * themselves, padding in a format-specific way if needed. 1853 * However, in case they don't, just do it here with zeros. 1854 */ 1855 for (i = 0; i < descsz - new_len; i++) 1856 sbuf_putc(sb, 0); 1857 } else if (new_len > descsz) { 1858 /* 1859 * We can't always truncate sb -- we may have drained some 1860 * of it already. 1861 */ 1862 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 1863 "read it (%zu > %zu). Since it is longer than " 1864 "expected, this coredump's notes are corrupt. THIS " 1865 "IS A BUG in the note_procstat routine for type %u.\n", 1866 __func__, (unsigned)note.n_type, new_len, descsz, 1867 (unsigned)note.n_type)); 1868 } 1869 } 1870 1871 /* 1872 * Miscellaneous note out functions. 1873 */ 1874 1875 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1876 #include <compat/freebsd32/freebsd32.h> 1877 #include <compat/freebsd32/freebsd32_signal.h> 1878 1879 typedef struct prstatus32 elf_prstatus_t; 1880 typedef struct prpsinfo32 elf_prpsinfo_t; 1881 typedef struct fpreg32 elf_prfpregset_t; 1882 typedef struct fpreg32 elf_fpregset_t; 1883 typedef struct reg32 elf_gregset_t; 1884 typedef struct thrmisc32 elf_thrmisc_t; 1885 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1886 typedef struct kinfo_proc32 elf_kinfo_proc_t; 1887 typedef uint32_t elf_ps_strings_t; 1888 #else 1889 typedef prstatus_t elf_prstatus_t; 1890 typedef prpsinfo_t elf_prpsinfo_t; 1891 typedef prfpregset_t elf_prfpregset_t; 1892 typedef prfpregset_t elf_fpregset_t; 1893 typedef gregset_t elf_gregset_t; 1894 typedef thrmisc_t elf_thrmisc_t; 1895 #define ELF_KERN_PROC_MASK 0 1896 typedef struct kinfo_proc elf_kinfo_proc_t; 1897 typedef vm_offset_t elf_ps_strings_t; 1898 #endif 1899 1900 static void 1901 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 1902 { 1903 struct sbuf sbarg; 1904 size_t len; 1905 char *cp, *end; 1906 struct proc *p; 1907 elf_prpsinfo_t *psinfo; 1908 int error; 1909 1910 p = (struct proc *)arg; 1911 if (sb != NULL) { 1912 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 1913 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 1914 psinfo->pr_version = PRPSINFO_VERSION; 1915 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1916 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1917 PROC_LOCK(p); 1918 if (p->p_args != NULL) { 1919 len = sizeof(psinfo->pr_psargs) - 1; 1920 if (len > p->p_args->ar_length) 1921 len = p->p_args->ar_length; 1922 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 1923 PROC_UNLOCK(p); 1924 error = 0; 1925 } else { 1926 _PHOLD(p); 1927 PROC_UNLOCK(p); 1928 sbuf_new(&sbarg, psinfo->pr_psargs, 1929 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 1930 error = proc_getargv(curthread, p, &sbarg); 1931 PRELE(p); 1932 if (sbuf_finish(&sbarg) == 0) 1933 len = sbuf_len(&sbarg) - 1; 1934 else 1935 len = sizeof(psinfo->pr_psargs) - 1; 1936 sbuf_delete(&sbarg); 1937 } 1938 if (error || len == 0) 1939 strlcpy(psinfo->pr_psargs, p->p_comm, 1940 sizeof(psinfo->pr_psargs)); 1941 else { 1942 KASSERT(len < sizeof(psinfo->pr_psargs), 1943 ("len is too long: %zu vs %zu", len, 1944 sizeof(psinfo->pr_psargs))); 1945 cp = psinfo->pr_psargs; 1946 end = cp + len - 1; 1947 for (;;) { 1948 cp = memchr(cp, '\0', end - cp); 1949 if (cp == NULL) 1950 break; 1951 *cp = ' '; 1952 } 1953 } 1954 psinfo->pr_pid = p->p_pid; 1955 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 1956 free(psinfo, M_TEMP); 1957 } 1958 *sizep = sizeof(*psinfo); 1959 } 1960 1961 static void 1962 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) 1963 { 1964 struct thread *td; 1965 elf_prstatus_t *status; 1966 1967 td = (struct thread *)arg; 1968 if (sb != NULL) { 1969 KASSERT(*sizep == sizeof(*status), ("invalid size")); 1970 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); 1971 status->pr_version = PRSTATUS_VERSION; 1972 status->pr_statussz = sizeof(elf_prstatus_t); 1973 status->pr_gregsetsz = sizeof(elf_gregset_t); 1974 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1975 status->pr_osreldate = osreldate; 1976 status->pr_cursig = td->td_proc->p_sig; 1977 status->pr_pid = td->td_tid; 1978 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1979 fill_regs32(td, &status->pr_reg); 1980 #else 1981 fill_regs(td, &status->pr_reg); 1982 #endif 1983 sbuf_bcat(sb, status, sizeof(*status)); 1984 free(status, M_TEMP); 1985 } 1986 *sizep = sizeof(*status); 1987 } 1988 1989 static void 1990 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) 1991 { 1992 struct thread *td; 1993 elf_prfpregset_t *fpregset; 1994 1995 td = (struct thread *)arg; 1996 if (sb != NULL) { 1997 KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); 1998 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); 1999 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2000 fill_fpregs32(td, fpregset); 2001 #else 2002 fill_fpregs(td, fpregset); 2003 #endif 2004 sbuf_bcat(sb, fpregset, sizeof(*fpregset)); 2005 free(fpregset, M_TEMP); 2006 } 2007 *sizep = sizeof(*fpregset); 2008 } 2009 2010 static void 2011 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) 2012 { 2013 struct thread *td; 2014 elf_thrmisc_t thrmisc; 2015 2016 td = (struct thread *)arg; 2017 if (sb != NULL) { 2018 KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); 2019 bzero(&thrmisc._pad, sizeof(thrmisc._pad)); 2020 strcpy(thrmisc.pr_tname, td->td_name); 2021 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); 2022 } 2023 *sizep = sizeof(thrmisc); 2024 } 2025 2026 static void 2027 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2028 { 2029 struct thread *td; 2030 size_t size; 2031 int structsize; 2032 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2033 struct ptrace_lwpinfo32 pl; 2034 #else 2035 struct ptrace_lwpinfo pl; 2036 #endif 2037 2038 td = (struct thread *)arg; 2039 size = sizeof(structsize) + sizeof(pl); 2040 if (sb != NULL) { 2041 KASSERT(*sizep == size, ("invalid size")); 2042 structsize = sizeof(pl); 2043 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2044 bzero(&pl, sizeof(pl)); 2045 pl.pl_lwpid = td->td_tid; 2046 pl.pl_event = PL_EVENT_NONE; 2047 pl.pl_sigmask = td->td_sigmask; 2048 pl.pl_siglist = td->td_siglist; 2049 if (td->td_si.si_signo != 0) { 2050 pl.pl_event = PL_EVENT_SIGNAL; 2051 pl.pl_flags |= PL_FLAG_SI; 2052 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2053 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2054 #else 2055 pl.pl_siginfo = td->td_si; 2056 #endif 2057 } 2058 strcpy(pl.pl_tdname, td->td_name); 2059 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2060 sbuf_bcat(sb, &pl, sizeof(pl)); 2061 } 2062 *sizep = size; 2063 } 2064 2065 /* 2066 * Allow for MD specific notes, as well as any MD 2067 * specific preparations for writing MI notes. 2068 */ 2069 static void 2070 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2071 { 2072 struct thread *td; 2073 void *buf; 2074 size_t size; 2075 2076 td = (struct thread *)arg; 2077 size = *sizep; 2078 if (size != 0 && sb != NULL) 2079 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2080 else 2081 buf = NULL; 2082 size = 0; 2083 __elfN(dump_thread)(td, buf, &size); 2084 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2085 if (size != 0 && sb != NULL) 2086 sbuf_bcat(sb, buf, size); 2087 free(buf, M_TEMP); 2088 *sizep = size; 2089 } 2090 2091 #ifdef KINFO_PROC_SIZE 2092 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2093 #endif 2094 2095 static void 2096 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2097 { 2098 struct proc *p; 2099 size_t size; 2100 int structsize; 2101 2102 p = (struct proc *)arg; 2103 size = sizeof(structsize) + p->p_numthreads * 2104 sizeof(elf_kinfo_proc_t); 2105 2106 if (sb != NULL) { 2107 KASSERT(*sizep == size, ("invalid size")); 2108 structsize = sizeof(elf_kinfo_proc_t); 2109 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2110 sx_slock(&proctree_lock); 2111 PROC_LOCK(p); 2112 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2113 sx_sunlock(&proctree_lock); 2114 } 2115 *sizep = size; 2116 } 2117 2118 #ifdef KINFO_FILE_SIZE 2119 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2120 #endif 2121 2122 static void 2123 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2124 { 2125 struct proc *p; 2126 size_t size, sect_sz, i; 2127 ssize_t start_len, sect_len; 2128 int structsize, filedesc_flags; 2129 2130 if (coredump_pack_fileinfo) 2131 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2132 else 2133 filedesc_flags = 0; 2134 2135 p = (struct proc *)arg; 2136 structsize = sizeof(struct kinfo_file); 2137 if (sb == NULL) { 2138 size = 0; 2139 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2140 sbuf_set_drain(sb, sbuf_drain_count, &size); 2141 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2142 PROC_LOCK(p); 2143 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2144 sbuf_finish(sb); 2145 sbuf_delete(sb); 2146 *sizep = size; 2147 } else { 2148 sbuf_start_section(sb, &start_len); 2149 2150 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2151 PROC_LOCK(p); 2152 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2153 filedesc_flags); 2154 2155 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2156 if (sect_len < 0) 2157 return; 2158 sect_sz = sect_len; 2159 2160 KASSERT(sect_sz <= *sizep, 2161 ("kern_proc_filedesc_out did not respect maxlen; " 2162 "requested %zu, got %zu", *sizep - sizeof(structsize), 2163 sect_sz - sizeof(structsize))); 2164 2165 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2166 sbuf_putc(sb, 0); 2167 } 2168 } 2169 2170 #ifdef KINFO_VMENTRY_SIZE 2171 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2172 #endif 2173 2174 static void 2175 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2176 { 2177 struct proc *p; 2178 size_t size; 2179 int structsize, vmmap_flags; 2180 2181 if (coredump_pack_vmmapinfo) 2182 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2183 else 2184 vmmap_flags = 0; 2185 2186 p = (struct proc *)arg; 2187 structsize = sizeof(struct kinfo_vmentry); 2188 if (sb == NULL) { 2189 size = 0; 2190 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2191 sbuf_set_drain(sb, sbuf_drain_count, &size); 2192 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2193 PROC_LOCK(p); 2194 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2195 sbuf_finish(sb); 2196 sbuf_delete(sb); 2197 *sizep = size; 2198 } else { 2199 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2200 PROC_LOCK(p); 2201 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2202 vmmap_flags); 2203 } 2204 } 2205 2206 static void 2207 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2208 { 2209 struct proc *p; 2210 size_t size; 2211 int structsize; 2212 2213 p = (struct proc *)arg; 2214 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2215 if (sb != NULL) { 2216 KASSERT(*sizep == size, ("invalid size")); 2217 structsize = sizeof(gid_t); 2218 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2219 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2220 sizeof(gid_t)); 2221 } 2222 *sizep = size; 2223 } 2224 2225 static void 2226 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2227 { 2228 struct proc *p; 2229 size_t size; 2230 int structsize; 2231 2232 p = (struct proc *)arg; 2233 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); 2234 if (sb != NULL) { 2235 KASSERT(*sizep == size, ("invalid size")); 2236 structsize = sizeof(p->p_fd->fd_cmask); 2237 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2238 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); 2239 } 2240 *sizep = size; 2241 } 2242 2243 static void 2244 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2245 { 2246 struct proc *p; 2247 struct rlimit rlim[RLIM_NLIMITS]; 2248 size_t size; 2249 int structsize, i; 2250 2251 p = (struct proc *)arg; 2252 size = sizeof(structsize) + sizeof(rlim); 2253 if (sb != NULL) { 2254 KASSERT(*sizep == size, ("invalid size")); 2255 structsize = sizeof(rlim); 2256 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2257 PROC_LOCK(p); 2258 for (i = 0; i < RLIM_NLIMITS; i++) 2259 lim_rlimit_proc(p, i, &rlim[i]); 2260 PROC_UNLOCK(p); 2261 sbuf_bcat(sb, rlim, sizeof(rlim)); 2262 } 2263 *sizep = size; 2264 } 2265 2266 static void 2267 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2268 { 2269 struct proc *p; 2270 size_t size; 2271 int structsize; 2272 2273 p = (struct proc *)arg; 2274 size = sizeof(structsize) + sizeof(p->p_osrel); 2275 if (sb != NULL) { 2276 KASSERT(*sizep == size, ("invalid size")); 2277 structsize = sizeof(p->p_osrel); 2278 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2279 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2280 } 2281 *sizep = size; 2282 } 2283 2284 static void 2285 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2286 { 2287 struct proc *p; 2288 elf_ps_strings_t ps_strings; 2289 size_t size; 2290 int structsize; 2291 2292 p = (struct proc *)arg; 2293 size = sizeof(structsize) + sizeof(ps_strings); 2294 if (sb != NULL) { 2295 KASSERT(*sizep == size, ("invalid size")); 2296 structsize = sizeof(ps_strings); 2297 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2298 ps_strings = PTROUT(p->p_sysent->sv_psstrings); 2299 #else 2300 ps_strings = p->p_sysent->sv_psstrings; 2301 #endif 2302 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2303 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2304 } 2305 *sizep = size; 2306 } 2307 2308 static void 2309 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2310 { 2311 struct proc *p; 2312 size_t size; 2313 int structsize; 2314 2315 p = (struct proc *)arg; 2316 if (sb == NULL) { 2317 size = 0; 2318 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2319 sbuf_set_drain(sb, sbuf_drain_count, &size); 2320 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2321 PHOLD(p); 2322 proc_getauxv(curthread, p, sb); 2323 PRELE(p); 2324 sbuf_finish(sb); 2325 sbuf_delete(sb); 2326 *sizep = size; 2327 } else { 2328 structsize = sizeof(Elf_Auxinfo); 2329 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2330 PHOLD(p); 2331 proc_getauxv(curthread, p, sb); 2332 PRELE(p); 2333 } 2334 } 2335 2336 static boolean_t 2337 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 2338 int32_t *osrel, const Elf_Phdr *pnote) 2339 { 2340 const Elf_Note *note, *note0, *note_end; 2341 const char *note_name; 2342 char *buf; 2343 int i, error; 2344 boolean_t res; 2345 2346 /* We need some limit, might as well use PAGE_SIZE. */ 2347 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2348 return (FALSE); 2349 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2350 if (pnote->p_offset > PAGE_SIZE || 2351 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2352 VOP_UNLOCK(imgp->vp, 0); 2353 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2354 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 2355 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2356 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2357 curthread->td_ucred, NOCRED, NULL, curthread); 2358 if (error != 0) { 2359 uprintf("i/o error PT_NOTE\n"); 2360 res = FALSE; 2361 goto ret; 2362 } 2363 note = note0 = (const Elf_Note *)buf; 2364 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2365 } else { 2366 note = note0 = (const Elf_Note *)(imgp->image_header + 2367 pnote->p_offset); 2368 note_end = (const Elf_Note *)(imgp->image_header + 2369 pnote->p_offset + pnote->p_filesz); 2370 buf = NULL; 2371 } 2372 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 2373 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 2374 (const char *)note < sizeof(Elf_Note)) { 2375 res = FALSE; 2376 goto ret; 2377 } 2378 if (note->n_namesz != checknote->hdr.n_namesz || 2379 note->n_descsz != checknote->hdr.n_descsz || 2380 note->n_type != checknote->hdr.n_type) 2381 goto nextnote; 2382 note_name = (const char *)(note + 1); 2383 if (note_name + checknote->hdr.n_namesz >= 2384 (const char *)note_end || strncmp(checknote->vendor, 2385 note_name, checknote->hdr.n_namesz) != 0) 2386 goto nextnote; 2387 2388 /* 2389 * Fetch the osreldate for binary 2390 * from the ELF OSABI-note if necessary. 2391 */ 2392 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 2393 checknote->trans_osrel != NULL) { 2394 res = checknote->trans_osrel(note, osrel); 2395 goto ret; 2396 } 2397 res = TRUE; 2398 goto ret; 2399 nextnote: 2400 note = (const Elf_Note *)((const char *)(note + 1) + 2401 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2402 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2403 } 2404 res = FALSE; 2405 ret: 2406 free(buf, M_TEMP); 2407 return (res); 2408 } 2409 2410 /* 2411 * Try to find the appropriate ABI-note section for checknote, 2412 * fetch the osreldate for binary from the ELF OSABI-note. Only the 2413 * first page of the image is searched, the same as for headers. 2414 */ 2415 static boolean_t 2416 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 2417 int32_t *osrel) 2418 { 2419 const Elf_Phdr *phdr; 2420 const Elf_Ehdr *hdr; 2421 int i; 2422 2423 hdr = (const Elf_Ehdr *)imgp->image_header; 2424 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2425 2426 for (i = 0; i < hdr->e_phnum; i++) { 2427 if (phdr[i].p_type == PT_NOTE && 2428 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 2429 return (TRUE); 2430 } 2431 return (FALSE); 2432 2433 } 2434 2435 /* 2436 * Tell kern_execve.c about it, with a little help from the linker. 2437 */ 2438 static struct execsw __elfN(execsw) = { 2439 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 2440 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2441 }; 2442 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2443 2444 static vm_prot_t 2445 __elfN(trans_prot)(Elf_Word flags) 2446 { 2447 vm_prot_t prot; 2448 2449 prot = 0; 2450 if (flags & PF_X) 2451 prot |= VM_PROT_EXECUTE; 2452 if (flags & PF_W) 2453 prot |= VM_PROT_WRITE; 2454 if (flags & PF_R) 2455 prot |= VM_PROT_READ; 2456 #if __ELF_WORD_SIZE == 32 2457 #if defined(__amd64__) 2458 if (i386_read_exec && (flags & PF_R)) 2459 prot |= VM_PROT_EXECUTE; 2460 #endif 2461 #endif 2462 return (prot); 2463 } 2464 2465 static Elf_Word 2466 __elfN(untrans_prot)(vm_prot_t prot) 2467 { 2468 Elf_Word flags; 2469 2470 flags = 0; 2471 if (prot & VM_PROT_EXECUTE) 2472 flags |= PF_X; 2473 if (prot & VM_PROT_READ) 2474 flags |= PF_R; 2475 if (prot & VM_PROT_WRITE) 2476 flags |= PF_W; 2477 return (flags); 2478 } 2479