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 "opt_capsicum.h" 35 36 #include <sys/param.h> 37 #include <sys/capsicum.h> 38 #include <sys/compressor.h> 39 #include <sys/exec.h> 40 #include <sys/fcntl.h> 41 #include <sys/imgact.h> 42 #include <sys/imgact_elf.h> 43 #include <sys/jail.h> 44 #include <sys/kernel.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mman.h> 49 #include <sys/namei.h> 50 #include <sys/proc.h> 51 #include <sys/procfs.h> 52 #include <sys/ptrace.h> 53 #include <sys/racct.h> 54 #include <sys/reg.h> 55 #include <sys/resourcevar.h> 56 #include <sys/rwlock.h> 57 #include <sys/sbuf.h> 58 #include <sys/sf_buf.h> 59 #include <sys/smp.h> 60 #include <sys/systm.h> 61 #include <sys/signalvar.h> 62 #include <sys/stat.h> 63 #include <sys/sx.h> 64 #include <sys/syscall.h> 65 #include <sys/sysctl.h> 66 #include <sys/sysent.h> 67 #include <sys/ucoredump.h> 68 #include <sys/vnode.h> 69 #include <sys/syslog.h> 70 #include <sys/eventhandler.h> 71 #include <sys/user.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_kern.h> 75 #include <vm/vm_param.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_object.h> 79 #include <vm/vm_extern.h> 80 81 #include <machine/elf.h> 82 #include <machine/md_var.h> 83 84 #define ELF_NOTE_ROUNDSIZE 4 85 #define OLD_EI_BRAND 8 86 87 /* 88 * ELF_ABI_NAME is a string name of the ELF ABI. ELF_ABI_ID is used 89 * to build variable names. 90 */ 91 #define ELF_ABI_NAME __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 92 #define ELF_ABI_ID __CONCAT(elf, __ELF_WORD_SIZE) 93 94 static int __elfN(check_header)(const Elf_Ehdr *hdr); 95 static const Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 96 const Elf_Phdr *phdr, const char *interp, int32_t *osrel, uint32_t *fctl0); 97 static int __elfN(load_interp_file)(struct thread *td, const char *file, u_long *addr, 98 u_long *entry); 99 static int __elfN(load_section)(const struct image_params *imgp, 100 vm_ooffset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 101 vm_prot_t prot); 102 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 103 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note, 104 int32_t *osrel); 105 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 106 static bool __elfN(check_note)(struct image_params *imgp, const Elf_Phdr *phdr, 107 const Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0, 108 uint32_t *fctl0); 109 static vm_prot_t __elfN(trans_prot)(Elf_Word); 110 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 111 static size_t __elfN(prepare_register_notes)(struct thread *td, 112 struct note_info_list *list, struct thread *target_td); 113 114 SYSCTL_NODE(_kern, OID_AUTO, ELF_ABI_ID, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 115 ""); 116 117 #define ELF_NODE_OID __CONCAT(_kern_, ELF_ABI_ID) 118 119 static int __elfN(fallback_brand) = -1; 120 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, 121 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, 122 ELF_ABI_NAME " brand of last resort"); 123 124 static int elf_legacy_coredump = 0; 125 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 126 &elf_legacy_coredump, 0, 127 "include all and only RW pages in core dumps"); 128 129 int __elfN(nxstack) = 130 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \ 131 defined(__arm__) || defined(__aarch64__) || \ 132 defined(__riscv) 133 1; 134 #else 135 0; 136 #endif 137 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, 138 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 139 ELF_ABI_NAME ": support PT_GNU_STACK for non-executable stack control"); 140 141 #if defined(__amd64__) 142 static int __elfN(vdso) = 1; 143 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, 144 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0, 145 ELF_ABI_NAME ": enable vdso preloading"); 146 #else 147 static int __elfN(vdso) = 0; 148 #endif 149 150 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 151 int i386_read_exec = 0; 152 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 153 "enable execution from readable segments"); 154 #endif 155 156 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR; 157 static int 158 sysctl_pie_base(SYSCTL_HANDLER_ARGS) 159 { 160 u_long val; 161 int error; 162 163 val = __elfN(pie_base); 164 error = sysctl_handle_long(oidp, &val, 0, req); 165 if (error != 0 || req->newptr == NULL) 166 return (error); 167 if ((val & PAGE_MASK) != 0) 168 return (EINVAL); 169 __elfN(pie_base) = val; 170 return (0); 171 } 172 SYSCTL_PROC(ELF_NODE_OID, OID_AUTO, pie_base, 173 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 174 sysctl_pie_base, "LU", 175 "PIE load base without randomization"); 176 177 SYSCTL_NODE(ELF_NODE_OID, OID_AUTO, aslr, 178 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 179 ""); 180 #define ASLR_NODE_OID __CONCAT(ELF_NODE_OID, _aslr) 181 182 /* 183 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures 184 * have limited address space (which can cause issues for applications with 185 * high memory use) so we leave it off there. 186 */ 187 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64; 188 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN, 189 &__elfN(aslr_enabled), 0, 190 ELF_ABI_NAME ": enable address map randomization"); 191 192 /* 193 * Enable ASLR by default for 64-bit PIE binaries. 194 */ 195 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64; 196 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN, 197 &__elfN(pie_aslr_enabled), 0, 198 ELF_ABI_NAME ": enable address map randomization for PIE binaries"); 199 200 /* 201 * Sbrk is deprecated and it can be assumed that in most cases it will not be 202 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR 203 * to use the bss grow region. 204 */ 205 static int __elfN(aslr_honor_sbrk) = 0; 206 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW, 207 &__elfN(aslr_honor_sbrk), 0, 208 ELF_ABI_NAME ": assume sbrk is used"); 209 210 static int __elfN(aslr_stack) = __ELF_WORD_SIZE == 64; 211 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN, 212 &__elfN(aslr_stack), 0, 213 ELF_ABI_NAME ": enable stack address randomization"); 214 215 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64; 216 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN, 217 &__elfN(aslr_shared_page), 0, 218 ELF_ABI_NAME ": enable shared page address randomization"); 219 220 static int __elfN(sigfastblock) = 1; 221 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, sigfastblock, 222 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0, 223 "enable sigfastblock for new processes"); 224 225 static bool __elfN(allow_wx) = true; 226 SYSCTL_BOOL(ELF_NODE_OID, OID_AUTO, allow_wx, 227 CTLFLAG_RWTUN, &__elfN(allow_wx), 0, 228 "Allow pages to be mapped simultaneously writable and executable"); 229 230 static u_int __elfN(phnums) = 128; 231 SYSCTL_UINT(ELF_NODE_OID, OID_AUTO, phnums, 232 CTLFLAG_RWTUN, &__elfN(phnums), 0, 233 "Max number of program headers to accept"); 234 235 static const Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 236 237 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a)) 238 239 const Elf_Brandnote __elfN(freebsd_brandnote) = { 240 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 241 .hdr.n_descsz = sizeof(int32_t), 242 .hdr.n_type = NT_FREEBSD_ABI_TAG, 243 .vendor = FREEBSD_ABI_VENDOR, 244 .flags = BN_TRANSLATE_OSREL, 245 .trans_osrel = __elfN(freebsd_trans_osrel) 246 }; 247 248 static bool 249 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 250 { 251 uintptr_t p; 252 253 p = (uintptr_t)(note + 1); 254 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 255 *osrel = *(const int32_t *)(p); 256 257 return (true); 258 } 259 260 static int GNU_KFREEBSD_ABI_DESC = 3; 261 262 const Elf_Brandnote __elfN(kfreebsd_brandnote) = { 263 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 264 .hdr.n_descsz = 16, /* XXX at least 16 */ 265 .hdr.n_type = 1, 266 .vendor = GNU_ABI_VENDOR, 267 .flags = BN_TRANSLATE_OSREL, 268 .trans_osrel = kfreebsd_trans_osrel 269 }; 270 271 static bool 272 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 273 { 274 const Elf32_Word *desc; 275 uintptr_t p; 276 277 p = (uintptr_t)(note + 1); 278 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 279 280 desc = (const Elf32_Word *)p; 281 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 282 return (false); 283 284 /* 285 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 286 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 287 */ 288 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 289 290 return (true); 291 } 292 293 int 294 __elfN(insert_brand_entry)(const Elf_Brandinfo *entry) 295 { 296 int i; 297 298 for (i = 0; i < MAX_BRANDS; i++) { 299 if (elf_brand_list[i] == NULL) { 300 elf_brand_list[i] = entry; 301 break; 302 } 303 } 304 if (i == MAX_BRANDS) { 305 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 306 __func__, entry); 307 return (-1); 308 } 309 return (0); 310 } 311 312 int 313 __elfN(remove_brand_entry)(const Elf_Brandinfo *entry) 314 { 315 int i; 316 317 for (i = 0; i < MAX_BRANDS; i++) { 318 if (elf_brand_list[i] == entry) { 319 elf_brand_list[i] = NULL; 320 break; 321 } 322 } 323 if (i == MAX_BRANDS) 324 return (-1); 325 return (0); 326 } 327 328 bool 329 __elfN(brand_inuse)(const Elf_Brandinfo *entry) 330 { 331 struct proc *p; 332 bool rval = false; 333 334 sx_slock(&allproc_lock); 335 FOREACH_PROC_IN_SYSTEM(p) { 336 if (p->p_sysent == entry->sysvec) { 337 rval = true; 338 break; 339 } 340 } 341 sx_sunlock(&allproc_lock); 342 343 return (rval); 344 } 345 346 static const Elf_Brandinfo * 347 __elfN(get_brandinfo)(struct image_params *imgp, const Elf_Phdr *phdr, 348 const char *interp, int32_t *osrel, uint32_t *fctl0) 349 { 350 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 351 const Elf_Brandinfo *bi, *bi_m; 352 bool ret, has_fctl0; 353 int i, interp_name_len; 354 355 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0; 356 357 /* 358 * We support four types of branding -- (1) the ELF EI_OSABI field 359 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 360 * branding w/in the ELF header, (3) path of the `interp_path' 361 * field, and (4) the ".note.ABI-tag" ELF section. 362 */ 363 364 /* Look for an ".note.ABI-tag" ELF section */ 365 bi_m = NULL; 366 for (i = 0; i < MAX_BRANDS; i++) { 367 bi = elf_brand_list[i]; 368 if (bi == NULL) 369 continue; 370 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 371 continue; 372 if (hdr->e_machine == bi->machine && (bi->flags & 373 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 374 has_fctl0 = false; 375 *fctl0 = 0; 376 *osrel = 0; 377 ret = __elfN(check_note)(imgp, phdr, bi->brand_note, 378 osrel, &has_fctl0, fctl0); 379 /* Give brand a chance to veto check_note's guess */ 380 if (ret && bi->header_supported) { 381 ret = bi->header_supported(imgp, osrel, 382 has_fctl0 ? fctl0 : NULL); 383 } 384 /* 385 * If note checker claimed the binary, but the 386 * interpreter path in the image does not 387 * match default one for the brand, try to 388 * search for other brands with the same 389 * interpreter. Either there is better brand 390 * with the right interpreter, or, failing 391 * this, we return first brand which accepted 392 * our note and, optionally, header. 393 */ 394 if (ret && bi_m == NULL && interp != NULL && 395 (bi->interp_path == NULL || 396 (strlen(bi->interp_path) + 1 != interp_name_len || 397 strncmp(interp, bi->interp_path, interp_name_len) 398 != 0))) { 399 bi_m = bi; 400 ret = 0; 401 } 402 if (ret) 403 return (bi); 404 } 405 } 406 if (bi_m != NULL) 407 return (bi_m); 408 409 /* If the executable has a brand, search for it in the brand list. */ 410 for (i = 0; i < MAX_BRANDS; i++) { 411 bi = elf_brand_list[i]; 412 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 413 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 414 continue; 415 if (hdr->e_machine == bi->machine && 416 (hdr->e_ident[EI_OSABI] == bi->brand || 417 (bi->compat_3_brand != NULL && 418 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 419 bi->compat_3_brand) == 0))) { 420 /* Looks good, but give brand a chance to veto */ 421 if (bi->header_supported == NULL || 422 bi->header_supported(imgp, NULL, NULL)) { 423 /* 424 * Again, prefer strictly matching 425 * interpreter path. 426 */ 427 if (interp_name_len == 0 && 428 bi->interp_path == NULL) 429 return (bi); 430 if (bi->interp_path != NULL && 431 strlen(bi->interp_path) + 1 == 432 interp_name_len && strncmp(interp, 433 bi->interp_path, interp_name_len) == 0) 434 return (bi); 435 if (bi_m == NULL) 436 bi_m = bi; 437 } 438 } 439 } 440 if (bi_m != NULL) 441 return (bi_m); 442 443 /* No known brand, see if the header is recognized by any brand */ 444 for (i = 0; i < MAX_BRANDS; i++) { 445 bi = elf_brand_list[i]; 446 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 447 bi->header_supported == NULL) 448 continue; 449 if (hdr->e_machine == bi->machine) { 450 ret = bi->header_supported(imgp, NULL, NULL); 451 if (ret) 452 return (bi); 453 } 454 } 455 456 /* Lacking a known brand, search for a recognized interpreter. */ 457 if (interp != NULL) { 458 for (i = 0; i < MAX_BRANDS; i++) { 459 bi = elf_brand_list[i]; 460 if (bi == NULL || (bi->flags & 461 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 462 != 0) 463 continue; 464 if (hdr->e_machine == bi->machine && 465 bi->interp_path != NULL && 466 /* ELF image p_filesz includes terminating zero */ 467 strlen(bi->interp_path) + 1 == interp_name_len && 468 strncmp(interp, bi->interp_path, interp_name_len) 469 == 0 && (bi->header_supported == NULL || 470 bi->header_supported(imgp, NULL, NULL))) 471 return (bi); 472 } 473 } 474 475 /* Lacking a recognized interpreter, try the default brand */ 476 for (i = 0; i < MAX_BRANDS; i++) { 477 bi = elf_brand_list[i]; 478 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 479 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 480 continue; 481 if (hdr->e_machine == bi->machine && 482 __elfN(fallback_brand) == bi->brand && 483 (bi->header_supported == NULL || 484 bi->header_supported(imgp, NULL, NULL))) 485 return (bi); 486 } 487 return (NULL); 488 } 489 490 static bool 491 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr) 492 { 493 return (hdr->e_phoff <= PAGE_SIZE && 494 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff); 495 } 496 497 static int 498 __elfN(check_header)(const Elf_Ehdr *hdr) 499 { 500 const Elf_Brandinfo *bi; 501 int i; 502 503 if (!IS_ELF(*hdr) || 504 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 505 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 506 hdr->e_ident[EI_VERSION] != EV_CURRENT || 507 hdr->e_phentsize != sizeof(Elf_Phdr) || 508 hdr->e_version != ELF_TARG_VER) 509 return (ENOEXEC); 510 511 /* 512 * Make sure we have at least one brand for this machine. 513 */ 514 515 for (i = 0; i < MAX_BRANDS; i++) { 516 bi = elf_brand_list[i]; 517 if (bi != NULL && bi->machine == hdr->e_machine) 518 break; 519 } 520 if (i == MAX_BRANDS) 521 return (ENOEXEC); 522 523 return (0); 524 } 525 526 static int 527 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 528 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 529 { 530 struct sf_buf *sf; 531 int error; 532 vm_offset_t off; 533 534 /* 535 * Create the page if it doesn't exist yet. Ignore errors. 536 */ 537 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 538 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 539 540 /* 541 * Find the page from the underlying object. 542 */ 543 if (object != NULL) { 544 sf = vm_imgact_map_page(object, offset); 545 if (sf == NULL) 546 return (KERN_FAILURE); 547 off = offset - trunc_page(offset); 548 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 549 end - start); 550 vm_imgact_unmap_page(sf); 551 if (error != 0) 552 return (KERN_FAILURE); 553 } 554 555 return (KERN_SUCCESS); 556 } 557 558 static int 559 __elfN(map_insert)(const struct image_params *imgp, vm_map_t map, 560 vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, 561 vm_prot_t prot, int cow) 562 { 563 struct sf_buf *sf; 564 vm_offset_t off; 565 vm_size_t sz; 566 int error, locked, rv; 567 568 if (start != trunc_page(start)) { 569 rv = __elfN(map_partial)(map, object, offset, start, 570 round_page(start), prot); 571 if (rv != KERN_SUCCESS) 572 return (rv); 573 offset += round_page(start) - start; 574 start = round_page(start); 575 } 576 if (end != round_page(end)) { 577 rv = __elfN(map_partial)(map, object, offset + 578 trunc_page(end) - start, trunc_page(end), end, prot); 579 if (rv != KERN_SUCCESS) 580 return (rv); 581 end = trunc_page(end); 582 } 583 if (start >= end) 584 return (KERN_SUCCESS); 585 if ((offset & PAGE_MASK) != 0) { 586 /* 587 * The mapping is not page aligned. This means that we have 588 * to copy the data. 589 */ 590 rv = vm_map_fixed(map, NULL, 0, start, end - start, 591 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 592 if (rv != KERN_SUCCESS) 593 return (rv); 594 if (object == NULL) 595 return (KERN_SUCCESS); 596 for (; start < end; start += sz) { 597 sf = vm_imgact_map_page(object, offset); 598 if (sf == NULL) 599 return (KERN_FAILURE); 600 off = offset - trunc_page(offset); 601 sz = end - start; 602 if (sz > PAGE_SIZE - off) 603 sz = PAGE_SIZE - off; 604 error = copyout((caddr_t)sf_buf_kva(sf) + off, 605 (caddr_t)start, sz); 606 vm_imgact_unmap_page(sf); 607 if (error != 0) 608 return (KERN_FAILURE); 609 offset += sz; 610 } 611 } else { 612 vm_object_reference(object); 613 rv = vm_map_fixed(map, object, offset, start, end - start, 614 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL | 615 (object != NULL ? MAP_VN_EXEC : 0)); 616 if (rv != KERN_SUCCESS) { 617 locked = VOP_ISLOCKED(imgp->vp); 618 VOP_UNLOCK(imgp->vp); 619 vm_object_deallocate(object); 620 vn_lock(imgp->vp, locked | LK_RETRY); 621 return (rv); 622 } else if (object != NULL) { 623 MPASS(imgp->vp->v_object == object); 624 VOP_SET_TEXT_CHECKED(imgp->vp); 625 } 626 } 627 return (KERN_SUCCESS); 628 } 629 630 static int 631 __elfN(load_section)(const struct image_params *imgp, vm_ooffset_t offset, 632 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot) 633 { 634 struct sf_buf *sf; 635 size_t map_len; 636 vm_map_t map; 637 vm_object_t object; 638 vm_offset_t map_addr; 639 int error, rv, cow; 640 size_t copy_len; 641 vm_ooffset_t file_addr; 642 643 /* 644 * It's necessary to fail if the filsz + offset taken from the 645 * header is greater than the actual file pager object's size. 646 * If we were to allow this, then the vm_map_find() below would 647 * walk right off the end of the file object and into the ether. 648 * 649 * While I'm here, might as well check for something else that 650 * is invalid: filsz cannot be greater than memsz. 651 */ 652 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 653 filsz > memsz) { 654 uprintf("elf_load_section: truncated ELF file\n"); 655 return (ENOEXEC); 656 } 657 658 object = imgp->object; 659 map = &imgp->proc->p_vmspace->vm_map; 660 map_addr = trunc_page((vm_offset_t)vmaddr); 661 file_addr = trunc_page(offset); 662 663 /* 664 * We have two choices. We can either clear the data in the last page 665 * of an oversized mapping, or we can start the anon mapping a page 666 * early and copy the initialized data into that first page. We 667 * choose the second. 668 */ 669 if (filsz == 0) 670 map_len = 0; 671 else if (memsz > filsz) 672 map_len = trunc_page(offset + filsz) - file_addr; 673 else 674 map_len = round_page(offset + filsz) - file_addr; 675 676 if (map_len != 0) { 677 /* cow flags: don't dump readonly sections in core */ 678 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 679 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 680 681 rv = __elfN(map_insert)(imgp, map, object, file_addr, 682 map_addr, map_addr + map_len, prot, cow); 683 if (rv != KERN_SUCCESS) 684 return (EINVAL); 685 686 /* we can stop now if we've covered it all */ 687 if (memsz == filsz) 688 return (0); 689 } 690 691 /* 692 * We have to get the remaining bit of the file into the first part 693 * of the oversized map segment. This is normally because the .data 694 * segment in the file is extended to provide bss. It's a neat idea 695 * to try and save a page, but it's a pain in the behind to implement. 696 */ 697 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset + 698 filsz); 699 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 700 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 701 702 /* This had damn well better be true! */ 703 if (map_len != 0) { 704 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 705 map_addr + map_len, prot, 0); 706 if (rv != KERN_SUCCESS) 707 return (EINVAL); 708 } 709 710 if (copy_len != 0) { 711 sf = vm_imgact_map_page(object, offset + filsz); 712 if (sf == NULL) 713 return (EIO); 714 715 /* send the page fragment to user space */ 716 error = copyout(sf_buf_kva(sf), (caddr_t)map_addr, 717 copy_len); 718 vm_imgact_unmap_page(sf); 719 if (error != 0) 720 return (error); 721 } 722 723 /* 724 * Remove write access to the page if it was only granted by map_insert 725 * to allow copyout. 726 */ 727 if ((prot & VM_PROT_WRITE) == 0) 728 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 729 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT); 730 731 return (0); 732 } 733 734 static int 735 __elfN(load_sections)(const struct image_params *imgp, const Elf_Ehdr *hdr, 736 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp) 737 { 738 vm_prot_t prot; 739 u_long base_addr; 740 bool first; 741 int error, i; 742 743 ASSERT_VOP_LOCKED(imgp->vp, __func__); 744 745 base_addr = 0; 746 first = true; 747 748 for (i = 0; i < hdr->e_phnum; i++) { 749 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 750 continue; 751 752 /* Loadable segment */ 753 prot = __elfN(trans_prot)(phdr[i].p_flags); 754 error = __elfN(load_section)(imgp, phdr[i].p_offset, 755 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 756 phdr[i].p_memsz, phdr[i].p_filesz, prot); 757 if (error != 0) 758 return (error); 759 760 /* 761 * Establish the base address if this is the first segment. 762 */ 763 if (first) { 764 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 765 first = false; 766 } 767 } 768 769 if (base_addrp != NULL) 770 *base_addrp = base_addr; 771 772 return (0); 773 } 774 775 /* 776 * Load the file "file" into memory. It may be either a shared object 777 * or an executable. 778 * 779 * The "addr" reference parameter is in/out. On entry, it specifies 780 * the address where a shared object should be loaded. If the file is 781 * an executable, this value is ignored. On exit, "addr" specifies 782 * where the file was actually loaded. 783 * 784 * The "entry" reference parameter is out only. On exit, it specifies 785 * the entry point for the loaded file. 786 */ 787 static int 788 __elfN(load_interp_file)(struct thread *td, const char *file, u_long *addr, 789 u_long *entry) 790 { 791 struct { 792 struct nameidata nd; 793 struct vattr attr; 794 struct image_params image_params; 795 } *tempdata = NULL; 796 const Elf_Ehdr *hdr = NULL; 797 const Elf_Phdr *phdr = NULL; 798 struct nameidata *nd; 799 struct vattr *attr; 800 struct image_params *imgp; 801 void *m_phdrs = NULL; 802 u_long rbase; 803 u_long base_addr = 0; 804 int error; 805 806 #ifdef CAPABILITY_MODE 807 /* 808 * XXXJA: This check can go away once we are sufficiently confident 809 * that the checks in namei() are correct. 810 */ 811 if (IN_CAPABILITY_MODE(td)) 812 return (ECAPMODE); 813 #endif 814 815 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO); 816 nd = &tempdata->nd; 817 attr = &tempdata->attr; 818 imgp = &tempdata->image_params; 819 820 /* 821 * Initialize part of the common data 822 */ 823 imgp->td = td; 824 imgp->proc = td->td_proc; 825 imgp->attr = attr; 826 imgp->interpreted = IMGACT_INTERP_ELF; /* ignored by do_execve */ 827 828 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF, 829 UIO_SYSSPACE, file); 830 if ((error = namei(nd)) != 0) { 831 nd->ni_vp = NULL; 832 goto fail; 833 } 834 NDFREE_PNBUF(nd); 835 imgp->vp = nd->ni_vp; 836 837 /* 838 * Check permissions, modes, uid, etc on the file, and "open" it. 839 */ 840 error = exec_check_permissions(imgp); 841 if (error) 842 goto fail; 843 844 error = exec_map_first_page(imgp); 845 if (error) 846 goto fail; 847 848 imgp->object = nd->ni_vp->v_object; 849 850 hdr = (const Elf_Ehdr *)imgp->image_header; 851 if ((error = __elfN(check_header)(hdr)) != 0) 852 goto fail; 853 if (hdr->e_type == ET_DYN) 854 rbase = *addr; 855 else if (hdr->e_type == ET_EXEC) 856 rbase = 0; 857 else { 858 error = ENOEXEC; 859 goto fail; 860 } 861 862 if (hdr->e_phnum > __elfN(phnums)) { 863 error = ENOEXEC; 864 goto fail; 865 } 866 if (__elfN(phdr_in_zero_page)(hdr) && 867 aligned(imgp->image_header + hdr->e_phoff, Elf_Addr)) { 868 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 869 } else { 870 VOP_UNLOCK(imgp->vp); 871 phdr = m_phdrs = malloc(hdr->e_phnum * sizeof(Elf_Phdr), 872 M_TEMP, M_WAITOK | M_ZERO); 873 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 874 error = vn_rdwr(UIO_READ, imgp->vp, m_phdrs, 875 hdr->e_phnum * sizeof(Elf_Phdr), hdr->e_phoff, 876 UIO_SYSSPACE, IO_NODELOCKED, imgp->td->td_ucred, 877 NOCRED, NULL, imgp->td); 878 if (error != 0) 879 goto fail; 880 } 881 882 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr); 883 if (error != 0) 884 goto fail; 885 886 if (imgp->proc->p_sysent->sv_protect != NULL) 887 imgp->proc->p_sysent->sv_protect(imgp, SVP_INTERP); 888 889 *addr = base_addr; 890 *entry = (unsigned long)hdr->e_entry + rbase; 891 892 fail: 893 if (imgp->firstpage) 894 exec_unmap_first_page(imgp); 895 896 if (nd->ni_vp) { 897 if (imgp->textset) 898 VOP_UNSET_TEXT_CHECKED(nd->ni_vp); 899 vput(nd->ni_vp); 900 } 901 free(m_phdrs, M_TEMP); 902 free(tempdata, M_TEMP); 903 904 return (error); 905 } 906 907 /* 908 * Select randomized valid address in the map map, between minv and 909 * maxv, with specified alignment. The [minv, maxv) range must belong 910 * to the map. Note that function only allocates the address, it is 911 * up to caller to clamp maxv in a way that the final allocation 912 * length fit into the map. 913 * 914 * Result is returned in *resp, error code indicates that arguments 915 * did not pass sanity checks for overflow and range correctness. 916 */ 917 static int 918 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv, 919 u_int align, u_long *resp) 920 { 921 u_long rbase, res; 922 923 MPASS(vm_map_min(map) <= minv); 924 925 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) { 926 uprintf("Invalid ELF segments layout\n"); 927 return (ENOEXEC); 928 } 929 930 arc4rand(&rbase, sizeof(rbase), 0); 931 res = roundup(minv, (u_long)align) + rbase % (maxv - minv); 932 res &= ~((u_long)align - 1); 933 if (res >= maxv) 934 res -= align; 935 936 KASSERT(res >= minv, 937 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx", 938 res, minv, maxv, rbase)); 939 KASSERT(res < maxv, 940 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx", 941 res, maxv, minv, rbase)); 942 943 *resp = res; 944 return (0); 945 } 946 947 static int 948 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr, 949 const Elf_Phdr *phdr) 950 { 951 struct vmspace *vmspace; 952 const char *err_str; 953 u_long text_size, data_size, total_size, text_addr, data_addr; 954 u_long seg_size, seg_addr; 955 int i; 956 957 err_str = NULL; 958 text_size = data_size = total_size = text_addr = data_addr = 0; 959 960 for (i = 0; i < hdr->e_phnum; i++) { 961 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 962 continue; 963 964 seg_addr = trunc_page(phdr[i].p_vaddr + imgp->et_dyn_addr); 965 seg_size = round_page(phdr[i].p_memsz + 966 phdr[i].p_vaddr + imgp->et_dyn_addr - seg_addr); 967 968 /* 969 * Make the largest executable segment the official 970 * text segment and all others data. 971 * 972 * Note that obreak() assumes that data_addr + data_size == end 973 * of data load area, and the ELF file format expects segments 974 * to be sorted by address. If multiple data segments exist, 975 * the last one will be used. 976 */ 977 978 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) { 979 text_size = seg_size; 980 text_addr = seg_addr; 981 } else { 982 data_size = seg_size; 983 data_addr = seg_addr; 984 } 985 total_size += seg_size; 986 } 987 988 if (data_addr == 0 && data_size == 0) { 989 data_addr = text_addr; 990 data_size = text_size; 991 } 992 993 /* 994 * Check limits. It should be safe to check the 995 * limits after loading the segments since we do 996 * not actually fault in all the segments pages. 997 */ 998 PROC_LOCK(imgp->proc); 999 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 1000 err_str = "Data segment size exceeds process limit"; 1001 else if (text_size > maxtsiz) 1002 err_str = "Text segment size exceeds system limit"; 1003 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 1004 err_str = "Total segment size exceeds process limit"; 1005 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 1006 err_str = "Data segment size exceeds resource limit"; 1007 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 1008 err_str = "Total segment size exceeds resource limit"; 1009 PROC_UNLOCK(imgp->proc); 1010 if (err_str != NULL) { 1011 uprintf("%s\n", err_str); 1012 return (ENOMEM); 1013 } 1014 1015 vmspace = imgp->proc->p_vmspace; 1016 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 1017 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 1018 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 1019 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 1020 1021 return (0); 1022 } 1023 1024 static int 1025 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr, 1026 char **interpp, bool *free_interpp) 1027 { 1028 char *interp; 1029 int error, interp_name_len; 1030 1031 KASSERT(phdr->p_type == PT_INTERP, 1032 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type)); 1033 ASSERT_VOP_LOCKED(imgp->vp, __func__); 1034 1035 /* Path to interpreter */ 1036 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) { 1037 uprintf("Invalid PT_INTERP\n"); 1038 return (ENOEXEC); 1039 } 1040 1041 interp_name_len = phdr->p_filesz; 1042 if (phdr->p_offset > PAGE_SIZE || 1043 interp_name_len > PAGE_SIZE - phdr->p_offset) { 1044 /* 1045 * The vnode lock might be needed by the pagedaemon to 1046 * clean pages owned by the vnode. Do not allow sleep 1047 * waiting for memory with the vnode locked, instead 1048 * try non-sleepable allocation first, and if it 1049 * fails, go to the slow path were we drop the lock 1050 * and do M_WAITOK. A text reference prevents 1051 * modifications to the vnode content. 1052 */ 1053 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT); 1054 if (interp == NULL) { 1055 VOP_UNLOCK(imgp->vp); 1056 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK); 1057 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1058 } 1059 1060 error = vn_rdwr(UIO_READ, imgp->vp, interp, 1061 interp_name_len, phdr->p_offset, 1062 UIO_SYSSPACE, IO_NODELOCKED, imgp->td->td_ucred, 1063 NOCRED, NULL, imgp->td); 1064 if (error != 0) { 1065 free(interp, M_TEMP); 1066 uprintf("i/o error PT_INTERP %d\n", error); 1067 return (error); 1068 } 1069 interp[interp_name_len] = '\0'; 1070 1071 *interpp = interp; 1072 *free_interpp = true; 1073 return (0); 1074 } 1075 1076 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset; 1077 if (interp[interp_name_len - 1] != '\0') { 1078 uprintf("Invalid PT_INTERP\n"); 1079 return (ENOEXEC); 1080 } 1081 1082 *interpp = interp; 1083 *free_interpp = false; 1084 return (0); 1085 } 1086 1087 static int 1088 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info, 1089 const char *interp, u_long *addr, u_long *entry) 1090 { 1091 int error; 1092 1093 if (brand_info->interp_newpath != NULL && 1094 (brand_info->interp_path == NULL || 1095 strcmp(interp, brand_info->interp_path) == 0)) { 1096 error = __elfN(load_interp_file)(imgp->td, 1097 brand_info->interp_newpath, addr, entry); 1098 if (error == 0) 1099 return (0); 1100 } 1101 1102 error = __elfN(load_interp_file)(imgp->td, interp, addr, entry); 1103 if (error == 0) 1104 return (0); 1105 1106 uprintf("ELF interpreter %s not found, error %d\n", interp, error); 1107 return (error); 1108 } 1109 1110 /* 1111 * Impossible et_dyn_addr initial value indicating that the real base 1112 * must be calculated later with some randomization applied. 1113 */ 1114 #define ET_DYN_ADDR_RAND 1 1115 1116 static int 1117 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 1118 { 1119 const Elf_Ehdr *hdr; 1120 const Elf_Phdr *phdr; 1121 Elf_Auxargs *elf_auxargs; 1122 struct vmspace *vmspace; 1123 vm_map_t map; 1124 char *interp; 1125 const Elf_Brandinfo *brand_info; 1126 struct sysentvec *sv; 1127 void *m_phdrs; 1128 u_long addr, baddr, entry, proghdr; 1129 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc; 1130 uint32_t fctl0; 1131 int32_t osrel; 1132 bool free_interp; 1133 int error, i, n; 1134 1135 hdr = (const Elf_Ehdr *)imgp->image_header; 1136 1137 /* 1138 * Do we have a valid ELF header ? 1139 * 1140 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 1141 * if particular brand doesn't support it. 1142 */ 1143 if (__elfN(check_header)(hdr) != 0 || 1144 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 1145 return (-1); 1146 1147 /* 1148 * From here on down, we return an errno, not -1, as we've 1149 * detected an ELF file. 1150 */ 1151 1152 n = error = 0; 1153 baddr = 0; 1154 osrel = 0; 1155 fctl0 = 0; 1156 entry = proghdr = 0; 1157 interp = NULL; 1158 free_interp = false; 1159 m_phdrs = NULL; 1160 1161 if (hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize < hdr->e_phoff) { 1162 uprintf("PHDRS wrap\n"); 1163 return (ENOEXEC); 1164 } 1165 if (hdr->e_phnum > __elfN(phnums)) { 1166 uprintf("Too many program headers (%u, %u max)\n", 1167 hdr->e_phnum, __elfN(phnums)); 1168 return (ENOEXEC); 1169 } 1170 if (__elfN(phdr_in_zero_page)(hdr) && 1171 aligned(imgp->image_header + hdr->e_phoff, Elf_Addr)) { 1172 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1173 } else { 1174 VOP_UNLOCK(imgp->vp); 1175 phdr = m_phdrs = malloc(hdr->e_phnum * sizeof(Elf_Phdr), 1176 M_TEMP, M_WAITOK | M_ZERO); 1177 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1178 error = vn_rdwr(UIO_READ, imgp->vp, m_phdrs, 1179 hdr->e_phnum * sizeof(Elf_Phdr), hdr->e_phoff, 1180 UIO_SYSSPACE, IO_NODELOCKED, imgp->td->td_ucred, 1181 NOCRED, NULL, imgp->td); 1182 if (error != 0) 1183 goto ret; 1184 } 1185 1186 /* 1187 * Somewhat arbitrary, limit accepted max alignment for the 1188 * loadable segment to the max supported superpage size. Too 1189 * large alignment requests are not useful and are indicators 1190 * of corrupted or outright malicious binary. 1191 */ 1192 maxalign = PAGE_SIZE; 1193 maxsalign = PAGE_SIZE * 1024; 1194 for (i = MAXPAGESIZES - 1; i > 0; i--) { 1195 if (pagesizes[i] > maxsalign) { 1196 maxsalign = pagesizes[i]; 1197 break; 1198 } 1199 } 1200 1201 mapsz = 0; 1202 1203 for (i = 0; i < hdr->e_phnum; i++) { 1204 switch (phdr[i].p_type) { 1205 case PT_LOAD: 1206 if (n == 0) 1207 baddr = phdr[i].p_vaddr; 1208 if (!powerof2(phdr[i].p_align) || 1209 phdr[i].p_align > maxsalign) { 1210 uprintf("Invalid segment alignment\n"); 1211 error = ENOEXEC; 1212 goto ret; 1213 } 1214 if (phdr[i].p_align > maxalign) 1215 maxalign = phdr[i].p_align; 1216 if (mapsz + phdr[i].p_memsz < mapsz) { 1217 uprintf("Mapsize overflow\n"); 1218 error = ENOEXEC; 1219 goto ret; 1220 } 1221 mapsz += phdr[i].p_memsz; 1222 n++; 1223 1224 /* 1225 * If this segment contains the program headers, 1226 * remember their virtual address for the AT_PHDR 1227 * aux entry. Static binaries don't usually include 1228 * a PT_PHDR entry. 1229 */ 1230 if (phdr[i].p_offset == 0 && 1231 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= 1232 phdr[i].p_filesz) 1233 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 1234 break; 1235 case PT_INTERP: 1236 /* Path to interpreter */ 1237 if (interp != NULL) { 1238 uprintf("Multiple PT_INTERP headers\n"); 1239 error = ENOEXEC; 1240 goto ret; 1241 } 1242 error = __elfN(get_interp)(imgp, &phdr[i], &interp, 1243 &free_interp); 1244 if (error != 0) 1245 goto ret; 1246 break; 1247 case PT_GNU_STACK: 1248 if (__elfN(nxstack)) { 1249 imgp->stack_prot = 1250 __elfN(trans_prot)(phdr[i].p_flags); 1251 if ((imgp->stack_prot & VM_PROT_RW) != 1252 VM_PROT_RW) { 1253 uprintf("Invalid PT_GNU_STACK\n"); 1254 error = ENOEXEC; 1255 goto ret; 1256 } 1257 } 1258 imgp->stack_sz = phdr[i].p_memsz; 1259 break; 1260 case PT_PHDR: /* Program header table info */ 1261 proghdr = phdr[i].p_vaddr; 1262 break; 1263 } 1264 } 1265 1266 brand_info = __elfN(get_brandinfo)(imgp, phdr, interp, &osrel, &fctl0); 1267 if (brand_info == NULL) { 1268 uprintf("ELF binary type \"%u\" not known.\n", 1269 hdr->e_ident[EI_OSABI]); 1270 error = ENOEXEC; 1271 goto ret; 1272 } 1273 1274 /* 1275 * Avoid a possible deadlock if the current address space is destroyed 1276 * and that address space maps the locked vnode. In the common case, 1277 * the locked vnode's v_usecount is decremented but remains greater 1278 * than zero. Consequently, the vnode lock is not needed by vrele(). 1279 * However, in cases where the vnode lock is external, such as nullfs, 1280 * v_usecount may become zero. 1281 * 1282 * The VV_TEXT flag prevents modifications to the executable while 1283 * the vnode is unlocked. 1284 */ 1285 VOP_UNLOCK(imgp->vp); 1286 1287 /* 1288 * Decide whether to enable randomization of user mappings. First, 1289 * reset user preferences for the setid binaries. Then, account for the 1290 * support of randomization by the ABI, by user preferences, and make 1291 * special treatment for PIE binaries. 1292 */ 1293 if (imgp->credential_setid) { 1294 PROC_LOCK(imgp->proc); 1295 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE | 1296 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 1297 PROC_UNLOCK(imgp->proc); 1298 } 1299 1300 sv = brand_info->sysvec; 1301 if (hdr->e_type == ET_DYN) { 1302 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 1303 uprintf("Cannot execute shared object\n"); 1304 error = ENOEXEC; 1305 (void)vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1306 goto ret; 1307 } 1308 /* 1309 * Honour the base load address from the dso if it is 1310 * non-zero for some reason. 1311 */ 1312 if (baddr == 0) { 1313 if ((sv->sv_flags & SV_ASLR) == 0 || 1314 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) 1315 imgp->et_dyn_addr = __elfN(pie_base); 1316 else if ((__elfN(pie_aslr_enabled) && 1317 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) || 1318 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0) 1319 imgp->et_dyn_addr = ET_DYN_ADDR_RAND; 1320 else 1321 imgp->et_dyn_addr = __elfN(pie_base); 1322 } 1323 } 1324 if ((sv->sv_flags & SV_ASLR) == 0 || 1325 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 || 1326 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) { 1327 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND, 1328 ("imgp->et_dyn_addr == RAND and !ASLR")); 1329 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 || 1330 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) || 1331 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1332 imgp->map_flags |= MAP_ASLR; 1333 /* 1334 * If user does not care about sbrk, utilize the bss 1335 * grow region for mappings as well. We can select 1336 * the base for the image anywere and still not suffer 1337 * from the fragmentation. 1338 */ 1339 if (!__elfN(aslr_honor_sbrk) || 1340 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0) 1341 imgp->map_flags |= MAP_ASLR_IGNSTART; 1342 if (__elfN(aslr_stack)) 1343 imgp->map_flags |= MAP_ASLR_STACK; 1344 if (__elfN(aslr_shared_page)) 1345 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE; 1346 } 1347 1348 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 && 1349 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) || 1350 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0) 1351 imgp->map_flags |= MAP_WXORX; 1352 1353 error = exec_new_vmspace(imgp, sv); 1354 1355 imgp->proc->p_sysent = sv; 1356 imgp->proc->p_elf_brandinfo = brand_info; 1357 1358 vmspace = imgp->proc->p_vmspace; 1359 map = &vmspace->vm_map; 1360 maxv = sv->sv_usrstack; 1361 if ((imgp->map_flags & MAP_ASLR_STACK) == 0) 1362 maxv -= lim_max(imgp->td, RLIMIT_STACK); 1363 if (error == 0 && mapsz >= maxv - vm_map_min(map)) { 1364 uprintf("Excessive mapping size\n"); 1365 error = ENOEXEC; 1366 } 1367 1368 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1369 KASSERT((map->flags & MAP_ASLR) != 0, 1370 ("ET_DYN_ADDR_RAND but !MAP_ASLR")); 1371 error = __CONCAT(rnd_, __elfN(base))(map, 1372 vm_map_min(map) + mapsz + lim_max(imgp->td, RLIMIT_DATA), 1373 /* reserve half of the address space to interpreter */ 1374 maxv / 2, maxalign, &imgp->et_dyn_addr); 1375 } 1376 1377 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1378 if (error != 0) 1379 goto ret; 1380 1381 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL); 1382 if (error != 0) 1383 goto ret; 1384 1385 error = __elfN(enforce_limits)(imgp, hdr, phdr); 1386 if (error != 0) 1387 goto ret; 1388 1389 /* 1390 * We load the dynamic linker where a userland call 1391 * to mmap(0, ...) would put it. The rationale behind this 1392 * calculation is that it leaves room for the heap to grow to 1393 * its maximum allowed size. 1394 */ 1395 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->td, 1396 RLIMIT_DATA)); 1397 if ((map->flags & MAP_ASLR) != 0) { 1398 maxv1 = maxv / 2 + addr / 2; 1399 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1, 1400 #if VM_NRESERVLEVEL > 0 1401 pagesizes[VM_NRESERVLEVEL] != 0 ? 1402 /* Align anon_loc to the largest superpage size. */ 1403 pagesizes[VM_NRESERVLEVEL] : 1404 #endif 1405 pagesizes[0], &anon_loc); 1406 if (error != 0) 1407 goto ret; 1408 map->anon_loc = anon_loc; 1409 } else { 1410 map->anon_loc = addr; 1411 } 1412 1413 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr; 1414 imgp->entry_addr = entry; 1415 1416 if (sv->sv_protect != NULL) 1417 sv->sv_protect(imgp, SVP_IMAGE); 1418 1419 if (interp != NULL) { 1420 VOP_UNLOCK(imgp->vp); 1421 if ((map->flags & MAP_ASLR) != 0) { 1422 /* Assume that interpreter fits into 1/4 of AS */ 1423 maxv1 = maxv / 2 + addr / 2; 1424 error = __CONCAT(rnd_, __elfN(base))(map, addr, 1425 maxv1, PAGE_SIZE, &addr); 1426 } 1427 if (error == 0) { 1428 error = __elfN(load_interp)(imgp, brand_info, interp, 1429 &addr, &imgp->entry_addr); 1430 } 1431 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1432 if (error != 0) 1433 goto ret; 1434 } else 1435 addr = imgp->et_dyn_addr; 1436 1437 error = exec_map_stack(imgp); 1438 if (error != 0) 1439 goto ret; 1440 1441 /* 1442 * Construct auxargs table (used by the copyout_auxargs routine) 1443 */ 1444 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT); 1445 if (elf_auxargs == NULL) { 1446 VOP_UNLOCK(imgp->vp); 1447 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1448 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1449 } 1450 elf_auxargs->execfd = -1; 1451 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr; 1452 elf_auxargs->phent = hdr->e_phentsize; 1453 elf_auxargs->phnum = hdr->e_phnum; 1454 elf_auxargs->pagesz = PAGE_SIZE; 1455 elf_auxargs->base = addr; 1456 elf_auxargs->flags = 0; 1457 elf_auxargs->entry = entry; 1458 elf_auxargs->hdr_eflags = hdr->e_flags; 1459 1460 imgp->auxargs = elf_auxargs; 1461 imgp->interpreted = 0; 1462 imgp->reloc_base = addr; 1463 imgp->proc->p_osrel = osrel; 1464 imgp->proc->p_fctl0 = fctl0; 1465 imgp->proc->p_elf_flags = hdr->e_flags; 1466 1467 ret: 1468 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock"); 1469 if (free_interp) 1470 free(interp, M_TEMP); 1471 free(m_phdrs, M_TEMP); 1472 return (error); 1473 } 1474 1475 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE) 1476 1477 int 1478 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base) 1479 { 1480 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1481 Elf_Auxinfo *argarray, *pos; 1482 struct vmspace *vmspace; 1483 rlim_t stacksz; 1484 int error, oc; 1485 uint32_t bsdflags; 1486 1487 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1488 M_WAITOK | M_ZERO); 1489 1490 vmspace = imgp->proc->p_vmspace; 1491 1492 if (args->execfd != -1) 1493 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1494 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1495 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1496 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1497 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1498 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1499 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1500 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1501 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1502 if (imgp->execpathp != 0) 1503 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp); 1504 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1505 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1506 if (imgp->canary != 0) { 1507 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary); 1508 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1509 } 1510 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1511 if (imgp->pagesizes != 0) { 1512 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes); 1513 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1514 } 1515 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) { 1516 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1517 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset); 1518 } 1519 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1520 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1521 imgp->sysent->sv_stackprot); 1522 if (imgp->sysent->sv_hwcap != NULL) 1523 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1524 if (imgp->sysent->sv_hwcap2 != NULL) 1525 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1526 if (imgp->sysent->sv_hwcap3 != NULL) 1527 AUXARGS_ENTRY(pos, AT_HWCAP3, *imgp->sysent->sv_hwcap3); 1528 if (imgp->sysent->sv_hwcap4 != NULL) 1529 AUXARGS_ENTRY(pos, AT_HWCAP4, *imgp->sysent->sv_hwcap4); 1530 bsdflags = 0; 1531 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0; 1532 oc = atomic_load_int(&vm_overcommit); 1533 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) != 1534 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0; 1535 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags); 1536 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc); 1537 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv); 1538 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc); 1539 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv); 1540 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings); 1541 #ifdef RANDOM_FENESTRASX 1542 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) { 1543 AUXARGS_ENTRY(pos, AT_FXRNG, 1544 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset); 1545 } 1546 #endif 1547 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) { 1548 AUXARGS_ENTRY(pos, AT_KPRELOAD, 1549 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset); 1550 } 1551 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop)); 1552 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur; 1553 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz); 1554 AUXARGS_ENTRY(pos, AT_NULL, 0); 1555 1556 free(imgp->auxargs, M_TEMP); 1557 imgp->auxargs = NULL; 1558 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1559 1560 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT); 1561 free(argarray, M_TEMP); 1562 return (error); 1563 } 1564 1565 int 1566 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp) 1567 { 1568 Elf_Addr *base; 1569 1570 base = (Elf_Addr *)*stack_base; 1571 base--; 1572 if (elf_suword(base, imgp->args->argc) == -1) 1573 return (EFAULT); 1574 *stack_base = (uintptr_t)base; 1575 return (0); 1576 } 1577 1578 /* 1579 * Code for generating ELF core dumps. 1580 */ 1581 1582 typedef void (*segment_callback)(vm_map_entry_t, void *); 1583 1584 /* Closure for cb_put_phdr(). */ 1585 struct phdr_closure { 1586 Elf_Phdr *phdr; /* Program header to fill in */ 1587 Elf_Off offset; /* Offset of segment in core file */ 1588 int numsegs; /* Maximum number of segments */ 1589 int nextseg; /* Next segment to fill in */ 1590 }; 1591 1592 struct note_info { 1593 int type; /* Note type. */ 1594 struct regset *regset; /* Register set. */ 1595 outfunc_t outfunc; /* Output function. */ 1596 void *outarg; /* Argument for the output function. */ 1597 size_t outsize; /* Output size. */ 1598 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1599 }; 1600 1601 TAILQ_HEAD(note_info_list, note_info); 1602 1603 static void cb_put_phdr(vm_map_entry_t, void *); 1604 static void cb_size_segment(vm_map_entry_t, void *); 1605 static void each_dumpable_segment(struct thread *, segment_callback, void *, 1606 int); 1607 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1608 struct note_info_list *, size_t, int); 1609 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *); 1610 1611 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1612 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1613 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1614 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1615 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1616 static void __elfN(note_procstat_kqueues)(void *, struct sbuf *, size_t *); 1617 static void note_procstat_files(void *, struct sbuf *, size_t *); 1618 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1619 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1620 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1621 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1622 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1623 1624 static int 1625 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1626 { 1627 1628 return (core_write((struct coredump_params *)arg, base, len, offset, 1629 UIO_SYSSPACE, NULL)); 1630 } 1631 1632 int 1633 __elfN(coredump)(struct thread *td, struct coredump_writer *cdw, off_t limit, int flags) 1634 { 1635 struct ucred *cred = td->td_ucred; 1636 int compm, error = 0; 1637 struct sseg_closure seginfo; 1638 struct note_info_list notelst; 1639 struct coredump_params params; 1640 struct note_info *ninfo; 1641 void *hdr, *tmpbuf; 1642 size_t hdrsize, notesz, coresize; 1643 1644 hdr = NULL; 1645 tmpbuf = NULL; 1646 TAILQ_INIT(¬elst); 1647 1648 /* Size the program segments. */ 1649 __elfN(size_segments)(td, &seginfo, flags); 1650 1651 /* 1652 * Collect info about the core file header area. 1653 */ 1654 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1655 if (seginfo.count + 1 >= PN_XNUM) 1656 hdrsize += sizeof(Elf_Shdr); 1657 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz); 1658 coresize = round_page(hdrsize + notesz) + seginfo.size; 1659 1660 /* Set up core dump parameters. */ 1661 params.offset = 0; 1662 params.active_cred = cred; 1663 params.td = td; 1664 params.cdw = cdw; 1665 params.comp = NULL; 1666 1667 #ifdef RACCT 1668 if (racct_enable) { 1669 PROC_LOCK(td->td_proc); 1670 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1671 PROC_UNLOCK(td->td_proc); 1672 if (error != 0) { 1673 error = EFAULT; 1674 goto done; 1675 } 1676 } 1677 #endif 1678 if (coresize >= limit) { 1679 error = EFAULT; 1680 goto done; 1681 } 1682 1683 /* Create a compression stream if necessary. */ 1684 compm = compress_user_cores; 1685 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP && 1686 compm == 0) 1687 compm = COMPRESS_GZIP; 1688 if (compm != 0) { 1689 params.comp = compressor_init(core_compressed_write, 1690 compm, CORE_BUF_SIZE, 1691 compress_user_cores_level, ¶ms); 1692 if (params.comp == NULL) { 1693 error = EFAULT; 1694 goto done; 1695 } 1696 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1697 } 1698 1699 if (cdw->init_fn != NULL) { 1700 error = (*cdw->init_fn)(cdw, ¶ms); 1701 if (error != 0) 1702 goto done; 1703 } 1704 1705 /* 1706 * Allocate memory for building the header, fill it up, and write it out 1707 * following the notes. 1708 * 1709 * Note that a process sharing our vmspace might be concurrently 1710 * mutating the map, in which case we could populate fewer than 1711 * seginfo.count headers. Zero the buffer to ensure that unpopulated 1712 * headers are still initialized. 1713 */ 1714 hdr = malloc(hdrsize, M_TEMP, M_WAITOK | M_ZERO); 1715 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1716 notesz, flags); 1717 1718 /* Write the contents of all of the writable segments. */ 1719 if (error == 0) { 1720 Elf_Phdr *php; 1721 off_t offset; 1722 int i; 1723 1724 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1725 offset = round_page(hdrsize + notesz); 1726 for (i = 0; i < seginfo.count; i++) { 1727 error = core_output((char *)(uintptr_t)php->p_vaddr, 1728 php->p_filesz, offset, ¶ms, tmpbuf); 1729 if (error != 0) 1730 break; 1731 offset += php->p_filesz; 1732 php++; 1733 } 1734 if (error == 0 && params.comp != NULL) 1735 error = compressor_flush(params.comp); 1736 } 1737 if (error) { 1738 log(LOG_WARNING, 1739 "Failed to write core file for process %s (error %d)\n", 1740 curproc->p_comm, error); 1741 } 1742 1743 done: 1744 free(tmpbuf, M_TEMP); 1745 if (params.comp != NULL) 1746 compressor_fini(params.comp); 1747 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1748 TAILQ_REMOVE(¬elst, ninfo, link); 1749 free(ninfo, M_TEMP); 1750 } 1751 if (hdr != NULL) 1752 free(hdr, M_TEMP); 1753 1754 return (error); 1755 } 1756 1757 /* 1758 * A callback for each_dumpable_segment() to write out the segment's 1759 * program header entry. 1760 */ 1761 static void 1762 cb_put_phdr(vm_map_entry_t entry, void *closure) 1763 { 1764 struct phdr_closure *phc = (struct phdr_closure *)closure; 1765 Elf_Phdr *phdr = phc->phdr; 1766 1767 if (phc->nextseg >= phc->numsegs) { 1768 /* Only write as many headers as we have space for. */ 1769 return; 1770 } 1771 1772 phc->offset = round_page(phc->offset); 1773 1774 phdr->p_type = PT_LOAD; 1775 phdr->p_offset = phc->offset; 1776 phdr->p_vaddr = entry->start; 1777 phdr->p_paddr = 0; 1778 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1779 phdr->p_align = PAGE_SIZE; 1780 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1781 1782 phc->offset += phdr->p_filesz; 1783 phc->phdr++; 1784 1785 phc->nextseg++; 1786 } 1787 1788 /* 1789 * A callback for each_dumpable_segment() to gather information about 1790 * the number of segments and their total size. 1791 */ 1792 static void 1793 cb_size_segment(vm_map_entry_t entry, void *closure) 1794 { 1795 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1796 1797 ssc->count++; 1798 ssc->size += entry->end - entry->start; 1799 } 1800 1801 void 1802 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo, 1803 int flags) 1804 { 1805 seginfo->count = 0; 1806 seginfo->size = 0; 1807 1808 each_dumpable_segment(td, cb_size_segment, seginfo, flags); 1809 } 1810 1811 /* 1812 * For each writable segment in the process's memory map, call the given 1813 * function with a pointer to the map entry and some arbitrary 1814 * caller-supplied data. 1815 */ 1816 static void 1817 each_dumpable_segment(struct thread *td, segment_callback func, void *closure, 1818 int flags) 1819 { 1820 struct proc *p = td->td_proc; 1821 vm_map_t map = &p->p_vmspace->vm_map; 1822 vm_map_entry_t entry; 1823 vm_object_t backing_object, object; 1824 bool ignore_entry; 1825 1826 vm_map_lock_read(map); 1827 VM_MAP_ENTRY_FOREACH(entry, map) { 1828 /* 1829 * Don't dump inaccessible mappings, deal with legacy 1830 * coredump mode. 1831 * 1832 * Note that read-only segments related to the elf binary 1833 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1834 * need to arbitrarily ignore such segments. 1835 */ 1836 if ((flags & SVC_ALL) == 0) { 1837 if (elf_legacy_coredump) { 1838 if ((entry->protection & VM_PROT_RW) != 1839 VM_PROT_RW) 1840 continue; 1841 } else { 1842 if ((entry->protection & VM_PROT_ALL) == 0) 1843 continue; 1844 } 1845 } 1846 1847 /* 1848 * Dont include memory segment in the coredump if 1849 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1850 * madvise(2). Do not dump submaps (i.e. parts of the 1851 * kernel map). 1852 */ 1853 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 1854 continue; 1855 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 && 1856 (flags & SVC_ALL) == 0) 1857 continue; 1858 if ((object = entry->object.vm_object) == NULL) 1859 continue; 1860 1861 /* Ignore memory-mapped devices and such things. */ 1862 VM_OBJECT_RLOCK(object); 1863 while ((backing_object = object->backing_object) != NULL) { 1864 VM_OBJECT_RLOCK(backing_object); 1865 VM_OBJECT_RUNLOCK(object); 1866 object = backing_object; 1867 } 1868 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0; 1869 VM_OBJECT_RUNLOCK(object); 1870 if (ignore_entry) 1871 continue; 1872 1873 (*func)(entry, closure); 1874 } 1875 vm_map_unlock_read(map); 1876 } 1877 1878 /* 1879 * Write the core file header to the file, including padding up to 1880 * the page boundary. 1881 */ 1882 static int 1883 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1884 size_t hdrsize, struct note_info_list *notelst, size_t notesz, 1885 int flags) 1886 { 1887 struct note_info *ninfo; 1888 struct sbuf *sb; 1889 int error; 1890 1891 /* Fill in the header. */ 1892 bzero(hdr, hdrsize); 1893 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags); 1894 1895 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1896 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1897 sbuf_start_section(sb, NULL); 1898 sbuf_bcat(sb, hdr, hdrsize); 1899 TAILQ_FOREACH(ninfo, notelst, link) 1900 __elfN(putnote)(p->td, ninfo, sb); 1901 /* Align up to a page boundary for the program segments. */ 1902 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1903 error = sbuf_finish(sb); 1904 sbuf_delete(sb); 1905 1906 return (error); 1907 } 1908 1909 void 1910 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1911 size_t *sizep) 1912 { 1913 struct proc *p; 1914 struct thread *thr; 1915 size_t size; 1916 1917 p = td->td_proc; 1918 size = 0; 1919 1920 size += __elfN(register_note)(td, list, NT_PRPSINFO, 1921 __elfN(note_prpsinfo), p); 1922 1923 /* 1924 * To have the debugger select the right thread (LWP) as the initial 1925 * thread, we dump the state of the thread passed to us in td first. 1926 * This is the thread that causes the core dump and thus likely to 1927 * be the right thread one wants to have selected in the debugger. 1928 */ 1929 thr = td; 1930 while (thr != NULL) { 1931 size += __elfN(prepare_register_notes)(td, list, thr); 1932 size += __elfN(register_note)(td, list, -1, 1933 __elfN(note_threadmd), thr); 1934 1935 thr = thr == td ? TAILQ_FIRST(&p->p_threads) : 1936 TAILQ_NEXT(thr, td_plist); 1937 if (thr == td) 1938 thr = TAILQ_NEXT(thr, td_plist); 1939 } 1940 1941 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC, 1942 __elfN(note_procstat_proc), p); 1943 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES, 1944 note_procstat_files, p); 1945 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP, 1946 note_procstat_vmmap, p); 1947 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS, 1948 note_procstat_groups, p); 1949 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK, 1950 note_procstat_umask, p); 1951 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT, 1952 note_procstat_rlimit, p); 1953 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL, 1954 note_procstat_osrel, p); 1955 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS, 1956 __elfN(note_procstat_psstrings), p); 1957 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV, 1958 __elfN(note_procstat_auxv), p); 1959 size += __elfN(register_note)(td, list, NT_PROCSTAT_KQUEUES, 1960 __elfN(note_procstat_kqueues), p); 1961 1962 *sizep = size; 1963 } 1964 1965 void 1966 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1967 size_t notesz, int flags) 1968 { 1969 Elf_Ehdr *ehdr; 1970 Elf_Phdr *phdr; 1971 Elf_Shdr *shdr; 1972 struct phdr_closure phc; 1973 const Elf_Brandinfo *bi; 1974 1975 ehdr = (Elf_Ehdr *)hdr; 1976 bi = td->td_proc->p_elf_brandinfo; 1977 1978 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1979 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1980 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1981 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1982 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1983 ehdr->e_ident[EI_DATA] = ELF_DATA; 1984 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1985 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi; 1986 ehdr->e_ident[EI_ABIVERSION] = 0; 1987 ehdr->e_ident[EI_PAD] = 0; 1988 ehdr->e_type = ET_CORE; 1989 ehdr->e_machine = bi->machine; 1990 ehdr->e_version = EV_CURRENT; 1991 ehdr->e_entry = 0; 1992 ehdr->e_phoff = sizeof(Elf_Ehdr); 1993 ehdr->e_flags = td->td_proc->p_elf_flags; 1994 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1995 ehdr->e_phentsize = sizeof(Elf_Phdr); 1996 ehdr->e_shentsize = sizeof(Elf_Shdr); 1997 ehdr->e_shstrndx = SHN_UNDEF; 1998 if (numsegs + 1 < PN_XNUM) { 1999 ehdr->e_phnum = numsegs + 1; 2000 ehdr->e_shnum = 0; 2001 } else { 2002 ehdr->e_phnum = PN_XNUM; 2003 ehdr->e_shnum = 1; 2004 2005 ehdr->e_shoff = ehdr->e_phoff + 2006 (numsegs + 1) * ehdr->e_phentsize; 2007 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 2008 ("e_shoff: %zu, hdrsize - shdr: %zu", 2009 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 2010 2011 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 2012 memset(shdr, 0, sizeof(*shdr)); 2013 /* 2014 * A special first section is used to hold large segment and 2015 * section counts. This was proposed by Sun Microsystems in 2016 * Solaris and has been adopted by Linux; the standard ELF 2017 * tools are already familiar with the technique. 2018 * 2019 * See table 7-7 of the Solaris "Linker and Libraries Guide" 2020 * (or 12-7 depending on the version of the document) for more 2021 * details. 2022 */ 2023 shdr->sh_type = SHT_NULL; 2024 shdr->sh_size = ehdr->e_shnum; 2025 shdr->sh_link = ehdr->e_shstrndx; 2026 shdr->sh_info = numsegs + 1; 2027 } 2028 2029 /* 2030 * Fill in the program header entries. 2031 */ 2032 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 2033 2034 /* The note segment. */ 2035 phdr->p_type = PT_NOTE; 2036 phdr->p_offset = hdrsize; 2037 phdr->p_vaddr = 0; 2038 phdr->p_paddr = 0; 2039 phdr->p_filesz = notesz; 2040 phdr->p_memsz = 0; 2041 phdr->p_flags = PF_R; 2042 phdr->p_align = ELF_NOTE_ROUNDSIZE; 2043 phdr++; 2044 2045 /* All the writable segments from the program. */ 2046 phc.phdr = phdr; 2047 phc.offset = round_page(hdrsize + notesz); 2048 phc.numsegs = numsegs; 2049 phc.nextseg = 0; 2050 each_dumpable_segment(td, cb_put_phdr, &phc, flags); 2051 } 2052 2053 static size_t 2054 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list, 2055 struct regset *regset, struct thread *target_td) 2056 { 2057 const struct sysentvec *sv; 2058 struct note_info *ninfo; 2059 size_t size, notesize; 2060 2061 size = 0; 2062 if (!regset->get(regset, target_td, NULL, &size) || size == 0) 2063 return (0); 2064 2065 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2066 ninfo->type = regset->note; 2067 ninfo->regset = regset; 2068 ninfo->outarg = target_td; 2069 ninfo->outsize = size; 2070 TAILQ_INSERT_TAIL(list, ninfo, link); 2071 2072 sv = td->td_proc->p_sysent; 2073 notesize = sizeof(Elf_Note) + /* note header */ 2074 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2075 /* note name */ 2076 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2077 2078 return (notesize); 2079 } 2080 2081 size_t 2082 __elfN(register_note)(struct thread *td, struct note_info_list *list, 2083 int type, outfunc_t out, void *arg) 2084 { 2085 const struct sysentvec *sv; 2086 struct note_info *ninfo; 2087 size_t size, notesize; 2088 2089 sv = td->td_proc->p_sysent; 2090 size = 0; 2091 out(arg, NULL, &size); 2092 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2093 ninfo->type = type; 2094 ninfo->outfunc = out; 2095 ninfo->outarg = arg; 2096 ninfo->outsize = size; 2097 TAILQ_INSERT_TAIL(list, ninfo, link); 2098 2099 if (type == -1) 2100 return (size); 2101 2102 notesize = sizeof(Elf_Note) + /* note header */ 2103 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2104 /* note name */ 2105 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2106 2107 return (notesize); 2108 } 2109 2110 static size_t 2111 append_note_data(const void *src, void *dst, size_t len) 2112 { 2113 size_t padded_len; 2114 2115 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 2116 if (dst != NULL) { 2117 bcopy(src, dst, len); 2118 bzero((char *)dst + len, padded_len - len); 2119 } 2120 return (padded_len); 2121 } 2122 2123 size_t 2124 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 2125 { 2126 Elf_Note *note; 2127 char *buf; 2128 size_t notesize; 2129 2130 buf = dst; 2131 if (buf != NULL) { 2132 note = (Elf_Note *)buf; 2133 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 2134 note->n_descsz = size; 2135 note->n_type = type; 2136 buf += sizeof(*note); 2137 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 2138 sizeof(FREEBSD_ABI_VENDOR)); 2139 append_note_data(src, buf, size); 2140 if (descp != NULL) 2141 *descp = buf; 2142 } 2143 2144 notesize = sizeof(Elf_Note) + /* note header */ 2145 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 2146 /* note name */ 2147 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2148 2149 return (notesize); 2150 } 2151 2152 static void 2153 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb) 2154 { 2155 Elf_Note note; 2156 const struct sysentvec *sv; 2157 ssize_t old_len, sect_len; 2158 size_t new_len, descsz, i; 2159 2160 if (ninfo->type == -1) { 2161 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2162 return; 2163 } 2164 2165 sv = td->td_proc->p_sysent; 2166 2167 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1; 2168 note.n_descsz = ninfo->outsize; 2169 note.n_type = ninfo->type; 2170 2171 sbuf_bcat(sb, ¬e, sizeof(note)); 2172 sbuf_start_section(sb, &old_len); 2173 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor, 2174 strlen(sv->sv_elf_core_abi_vendor) + 1); 2175 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2176 if (note.n_descsz == 0) 2177 return; 2178 sbuf_start_section(sb, &old_len); 2179 if (ninfo->regset != NULL) { 2180 struct regset *regset = ninfo->regset; 2181 void *buf; 2182 2183 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK); 2184 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize); 2185 sbuf_bcat(sb, buf, ninfo->outsize); 2186 free(buf, M_TEMP); 2187 } else 2188 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2189 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2190 if (sect_len < 0) 2191 return; 2192 2193 new_len = (size_t)sect_len; 2194 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 2195 if (new_len < descsz) { 2196 /* 2197 * It is expected that individual note emitters will correctly 2198 * predict their expected output size and fill up to that size 2199 * themselves, padding in a format-specific way if needed. 2200 * However, in case they don't, just do it here with zeros. 2201 */ 2202 for (i = 0; i < descsz - new_len; i++) 2203 sbuf_putc(sb, 0); 2204 } else if (new_len > descsz) { 2205 /* 2206 * We can't always truncate sb -- we may have drained some 2207 * of it already. 2208 */ 2209 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 2210 "read it (%zu > %zu). Since it is longer than " 2211 "expected, this coredump's notes are corrupt. THIS " 2212 "IS A BUG in the note_procstat routine for type %u.\n", 2213 __func__, (unsigned)note.n_type, new_len, descsz, 2214 (unsigned)note.n_type)); 2215 } 2216 } 2217 2218 /* 2219 * Miscellaneous note out functions. 2220 */ 2221 2222 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2223 #include <compat/freebsd32/freebsd32.h> 2224 #include <compat/freebsd32/freebsd32_signal.h> 2225 2226 typedef struct prstatus32 elf_prstatus_t; 2227 typedef struct prpsinfo32 elf_prpsinfo_t; 2228 typedef struct fpreg32 elf_prfpregset_t; 2229 typedef struct fpreg32 elf_fpregset_t; 2230 typedef struct reg32 elf_gregset_t; 2231 typedef struct thrmisc32 elf_thrmisc_t; 2232 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t; 2233 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 2234 typedef struct kinfo_proc32 elf_kinfo_proc_t; 2235 typedef uint32_t elf_ps_strings_t; 2236 #else 2237 typedef prstatus_t elf_prstatus_t; 2238 typedef prpsinfo_t elf_prpsinfo_t; 2239 typedef prfpregset_t elf_prfpregset_t; 2240 typedef prfpregset_t elf_fpregset_t; 2241 typedef gregset_t elf_gregset_t; 2242 typedef thrmisc_t elf_thrmisc_t; 2243 typedef struct ptrace_lwpinfo elf_lwpinfo_t; 2244 #define ELF_KERN_PROC_MASK 0 2245 typedef struct kinfo_proc elf_kinfo_proc_t; 2246 typedef vm_offset_t elf_ps_strings_t; 2247 #endif 2248 2249 static void 2250 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2251 { 2252 struct sbuf sbarg; 2253 size_t len; 2254 char *cp, *end; 2255 struct proc *p; 2256 elf_prpsinfo_t *psinfo; 2257 int error; 2258 2259 p = arg; 2260 if (sb != NULL) { 2261 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 2262 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 2263 psinfo->pr_version = PRPSINFO_VERSION; 2264 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 2265 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 2266 PROC_LOCK(p); 2267 if (p->p_args != NULL) { 2268 len = sizeof(psinfo->pr_psargs) - 1; 2269 if (len > p->p_args->ar_length) 2270 len = p->p_args->ar_length; 2271 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 2272 PROC_UNLOCK(p); 2273 error = 0; 2274 } else { 2275 _PHOLD(p); 2276 PROC_UNLOCK(p); 2277 sbuf_new(&sbarg, psinfo->pr_psargs, 2278 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 2279 error = proc_getargv(curthread, p, &sbarg); 2280 PRELE(p); 2281 if (sbuf_finish(&sbarg) == 0) { 2282 len = sbuf_len(&sbarg); 2283 if (len > 0) 2284 len--; 2285 } else { 2286 len = sizeof(psinfo->pr_psargs) - 1; 2287 } 2288 sbuf_delete(&sbarg); 2289 } 2290 if (error != 0 || len == 0 || (ssize_t)len == -1) 2291 strlcpy(psinfo->pr_psargs, p->p_comm, 2292 sizeof(psinfo->pr_psargs)); 2293 else { 2294 KASSERT(len < sizeof(psinfo->pr_psargs), 2295 ("len is too long: %zu vs %zu", len, 2296 sizeof(psinfo->pr_psargs))); 2297 cp = psinfo->pr_psargs; 2298 end = cp + len - 1; 2299 for (;;) { 2300 cp = memchr(cp, '\0', end - cp); 2301 if (cp == NULL) 2302 break; 2303 *cp = ' '; 2304 } 2305 } 2306 psinfo->pr_pid = p->p_pid; 2307 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 2308 free(psinfo, M_TEMP); 2309 } 2310 *sizep = sizeof(*psinfo); 2311 } 2312 2313 static bool 2314 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf, 2315 size_t *sizep) 2316 { 2317 elf_prstatus_t *status; 2318 2319 if (buf != NULL) { 2320 KASSERT(*sizep == sizeof(*status), ("%s: invalid size", 2321 __func__)); 2322 status = buf; 2323 memset(status, 0, *sizep); 2324 status->pr_version = PRSTATUS_VERSION; 2325 status->pr_statussz = sizeof(elf_prstatus_t); 2326 status->pr_gregsetsz = sizeof(elf_gregset_t); 2327 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 2328 status->pr_osreldate = osreldate; 2329 status->pr_cursig = td->td_proc->p_sig; 2330 status->pr_pid = td->td_tid; 2331 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2332 fill_regs32(td, &status->pr_reg); 2333 #else 2334 fill_regs(td, &status->pr_reg); 2335 #endif 2336 } 2337 *sizep = sizeof(*status); 2338 return (true); 2339 } 2340 2341 static bool 2342 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf, 2343 size_t size) 2344 { 2345 elf_prstatus_t *status; 2346 2347 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__)); 2348 status = buf; 2349 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2350 set_regs32(td, &status->pr_reg); 2351 #else 2352 set_regs(td, &status->pr_reg); 2353 #endif 2354 return (true); 2355 } 2356 2357 static struct regset __elfN(regset_prstatus) = { 2358 .note = NT_PRSTATUS, 2359 .size = sizeof(elf_prstatus_t), 2360 .get = __elfN(get_prstatus), 2361 .set = __elfN(set_prstatus), 2362 }; 2363 ELF_REGSET(__elfN(regset_prstatus)); 2364 2365 static bool 2366 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf, 2367 size_t *sizep) 2368 { 2369 elf_prfpregset_t *fpregset; 2370 2371 if (buf != NULL) { 2372 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size", 2373 __func__)); 2374 fpregset = buf; 2375 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2376 fill_fpregs32(td, fpregset); 2377 #else 2378 fill_fpregs(td, fpregset); 2379 #endif 2380 } 2381 *sizep = sizeof(*fpregset); 2382 return (true); 2383 } 2384 2385 static bool 2386 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf, 2387 size_t size) 2388 { 2389 elf_prfpregset_t *fpregset; 2390 2391 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__)); 2392 2393 fpregset = buf; 2394 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2395 return (set_fpregs32(td, fpregset) == 0); 2396 #else 2397 return (set_fpregs(td, fpregset) == 0); 2398 #endif 2399 } 2400 2401 static struct regset __elfN(regset_fpregset) = { 2402 .note = NT_FPREGSET, 2403 .size = sizeof(elf_prfpregset_t), 2404 .get = __elfN(get_fpregset), 2405 .set = __elfN(set_fpregset), 2406 }; 2407 ELF_REGSET(__elfN(regset_fpregset)); 2408 2409 static bool 2410 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf, 2411 size_t *sizep) 2412 { 2413 elf_thrmisc_t *thrmisc; 2414 2415 if (buf != NULL) { 2416 KASSERT(*sizep == sizeof(*thrmisc), 2417 ("%s: invalid size", __func__)); 2418 thrmisc = buf; 2419 bzero(thrmisc, sizeof(*thrmisc)); 2420 strcpy(thrmisc->pr_tname, td->td_name); 2421 } 2422 *sizep = sizeof(*thrmisc); 2423 return (true); 2424 } 2425 2426 static struct regset __elfN(regset_thrmisc) = { 2427 .note = NT_THRMISC, 2428 .size = sizeof(elf_thrmisc_t), 2429 .get = __elfN(get_thrmisc), 2430 }; 2431 ELF_REGSET(__elfN(regset_thrmisc)); 2432 2433 static bool 2434 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf, 2435 size_t *sizep) 2436 { 2437 elf_lwpinfo_t pl; 2438 size_t size; 2439 int structsize; 2440 2441 size = sizeof(structsize) + sizeof(pl); 2442 if (buf != NULL) { 2443 KASSERT(*sizep == size, ("%s: invalid size", __func__)); 2444 structsize = sizeof(pl); 2445 memcpy(buf, &structsize, sizeof(structsize)); 2446 bzero(&pl, sizeof(pl)); 2447 pl.pl_lwpid = td->td_tid; 2448 pl.pl_event = PL_EVENT_NONE; 2449 pl.pl_sigmask = td->td_sigmask; 2450 pl.pl_siglist = td->td_siglist; 2451 if (td->td_si.si_signo != 0) { 2452 pl.pl_event = PL_EVENT_SIGNAL; 2453 pl.pl_flags |= PL_FLAG_SI; 2454 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2455 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2456 #else 2457 pl.pl_siginfo = td->td_si; 2458 #endif 2459 } 2460 strcpy(pl.pl_tdname, td->td_name); 2461 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2462 memcpy((int *)buf + 1, &pl, sizeof(pl)); 2463 } 2464 *sizep = size; 2465 return (true); 2466 } 2467 2468 static struct regset __elfN(regset_lwpinfo) = { 2469 .note = NT_PTLWPINFO, 2470 .size = sizeof(int) + sizeof(elf_lwpinfo_t), 2471 .get = __elfN(get_lwpinfo), 2472 }; 2473 ELF_REGSET(__elfN(regset_lwpinfo)); 2474 2475 static size_t 2476 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list, 2477 struct thread *target_td) 2478 { 2479 struct sysentvec *sv = td->td_proc->p_sysent; 2480 struct regset **regsetp, **regset_end, *regset; 2481 size_t size; 2482 2483 size = 0; 2484 2485 if (target_td == td) 2486 cpu_update_pcb(target_td); 2487 2488 /* NT_PRSTATUS must be the first register set note. */ 2489 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus), 2490 target_td); 2491 2492 regsetp = sv->sv_regset_begin; 2493 if (regsetp == NULL) { 2494 /* XXX: This shouldn't be true for any FreeBSD ABIs. */ 2495 size += __elfN(register_regset_note)(td, list, 2496 &__elfN(regset_fpregset), target_td); 2497 return (size); 2498 } 2499 regset_end = sv->sv_regset_end; 2500 MPASS(regset_end != NULL); 2501 for (; regsetp < regset_end; regsetp++) { 2502 regset = *regsetp; 2503 if (regset->note == NT_PRSTATUS) 2504 continue; 2505 size += __elfN(register_regset_note)(td, list, regset, 2506 target_td); 2507 } 2508 return (size); 2509 } 2510 2511 /* 2512 * Allow for MD specific notes, as well as any MD 2513 * specific preparations for writing MI notes. 2514 */ 2515 static void 2516 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2517 { 2518 struct thread *td; 2519 void *buf; 2520 size_t size; 2521 2522 td = (struct thread *)arg; 2523 size = *sizep; 2524 if (size != 0 && sb != NULL) 2525 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2526 else 2527 buf = NULL; 2528 size = 0; 2529 __elfN(dump_thread)(td, buf, &size); 2530 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2531 if (size != 0 && sb != NULL) 2532 sbuf_bcat(sb, buf, size); 2533 free(buf, M_TEMP); 2534 *sizep = size; 2535 } 2536 2537 #ifdef KINFO_PROC_SIZE 2538 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2539 #endif 2540 2541 static void 2542 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2543 { 2544 struct proc *p; 2545 size_t size; 2546 int structsize; 2547 2548 p = arg; 2549 size = sizeof(structsize) + p->p_numthreads * 2550 sizeof(elf_kinfo_proc_t); 2551 2552 if (sb != NULL) { 2553 KASSERT(*sizep == size, ("invalid size")); 2554 structsize = sizeof(elf_kinfo_proc_t); 2555 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2556 sx_slock(&proctree_lock); 2557 PROC_LOCK(p); 2558 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2559 sx_sunlock(&proctree_lock); 2560 } 2561 *sizep = size; 2562 } 2563 2564 #ifdef KINFO_FILE_SIZE 2565 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2566 #endif 2567 2568 static void 2569 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2570 { 2571 struct proc *p; 2572 size_t size, sect_sz, i; 2573 ssize_t start_len, sect_len; 2574 int structsize, filedesc_flags; 2575 2576 if (coredump_pack_fileinfo) 2577 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2578 else 2579 filedesc_flags = 0; 2580 2581 p = arg; 2582 structsize = sizeof(struct kinfo_file); 2583 if (sb == NULL) { 2584 size = 0; 2585 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2586 sbuf_set_drain(sb, sbuf_count_drain, &size); 2587 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2588 PROC_LOCK(p); 2589 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2590 sbuf_finish(sb); 2591 sbuf_delete(sb); 2592 *sizep = size; 2593 } else { 2594 sbuf_start_section(sb, &start_len); 2595 2596 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2597 PROC_LOCK(p); 2598 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2599 filedesc_flags); 2600 2601 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2602 if (sect_len < 0) 2603 return; 2604 sect_sz = sect_len; 2605 2606 KASSERT(sect_sz <= *sizep, 2607 ("kern_proc_filedesc_out did not respect maxlen; " 2608 "requested %zu, got %zu", *sizep - sizeof(structsize), 2609 sect_sz - sizeof(structsize))); 2610 2611 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2612 sbuf_putc(sb, 0); 2613 } 2614 } 2615 2616 #ifdef KINFO_VMENTRY_SIZE 2617 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2618 #endif 2619 2620 static void 2621 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2622 { 2623 struct proc *p; 2624 size_t size; 2625 int structsize, vmmap_flags; 2626 2627 if (coredump_pack_vmmapinfo) 2628 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2629 else 2630 vmmap_flags = 0; 2631 2632 p = arg; 2633 structsize = sizeof(struct kinfo_vmentry); 2634 if (sb == NULL) { 2635 size = 0; 2636 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2637 sbuf_set_drain(sb, sbuf_count_drain, &size); 2638 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2639 PROC_LOCK(p); 2640 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2641 sbuf_finish(sb); 2642 sbuf_delete(sb); 2643 *sizep = size; 2644 } else { 2645 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2646 PROC_LOCK(p); 2647 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2648 vmmap_flags); 2649 } 2650 } 2651 2652 static void 2653 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2654 { 2655 struct proc *p; 2656 size_t size; 2657 int structsize; 2658 2659 p = arg; 2660 size = sizeof(structsize) + 2661 (1 + p->p_ucred->cr_ngroups) * sizeof(gid_t); 2662 if (sb != NULL) { 2663 KASSERT(*sizep == size, ("invalid size")); 2664 structsize = sizeof(gid_t); 2665 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2666 sbuf_bcat(sb, &p->p_ucred->cr_gid, sizeof(gid_t)); 2667 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2668 sizeof(gid_t)); 2669 } 2670 *sizep = size; 2671 } 2672 2673 static void 2674 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2675 { 2676 struct proc *p; 2677 size_t size; 2678 int structsize; 2679 2680 p = arg; 2681 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask); 2682 if (sb != NULL) { 2683 KASSERT(*sizep == size, ("invalid size")); 2684 structsize = sizeof(p->p_pd->pd_cmask); 2685 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2686 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask)); 2687 } 2688 *sizep = size; 2689 } 2690 2691 static void 2692 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2693 { 2694 struct proc *p; 2695 struct rlimit rlim[RLIM_NLIMITS]; 2696 size_t size; 2697 int structsize, i; 2698 2699 p = arg; 2700 size = sizeof(structsize) + sizeof(rlim); 2701 if (sb != NULL) { 2702 KASSERT(*sizep == size, ("invalid size")); 2703 structsize = sizeof(rlim); 2704 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2705 PROC_LOCK(p); 2706 for (i = 0; i < RLIM_NLIMITS; i++) 2707 lim_rlimit_proc(p, i, &rlim[i]); 2708 PROC_UNLOCK(p); 2709 sbuf_bcat(sb, rlim, sizeof(rlim)); 2710 } 2711 *sizep = size; 2712 } 2713 2714 static void 2715 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2716 { 2717 struct proc *p; 2718 size_t size; 2719 int structsize; 2720 2721 p = arg; 2722 size = sizeof(structsize) + sizeof(p->p_osrel); 2723 if (sb != NULL) { 2724 KASSERT(*sizep == size, ("invalid size")); 2725 structsize = sizeof(p->p_osrel); 2726 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2727 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2728 } 2729 *sizep = size; 2730 } 2731 2732 static void 2733 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2734 { 2735 struct proc *p; 2736 elf_ps_strings_t ps_strings; 2737 size_t size; 2738 int structsize; 2739 2740 p = arg; 2741 size = sizeof(structsize) + sizeof(ps_strings); 2742 if (sb != NULL) { 2743 KASSERT(*sizep == size, ("invalid size")); 2744 structsize = sizeof(ps_strings); 2745 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2746 ps_strings = PTROUT(PROC_PS_STRINGS(p)); 2747 #else 2748 ps_strings = PROC_PS_STRINGS(p); 2749 #endif 2750 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2751 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2752 } 2753 *sizep = size; 2754 } 2755 2756 static void 2757 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2758 { 2759 struct proc *p; 2760 size_t size; 2761 int structsize; 2762 2763 p = arg; 2764 structsize = sizeof(Elf_Auxinfo); 2765 if (sb == NULL) { 2766 size = 0; 2767 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo), 2768 SBUF_FIXEDLEN); 2769 sbuf_set_drain(sb, sbuf_count_drain, &size); 2770 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2771 PHOLD(p); 2772 proc_getauxv(curthread, p, sb); 2773 PRELE(p); 2774 sbuf_finish(sb); 2775 sbuf_delete(sb); 2776 *sizep = size; 2777 } else { 2778 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2779 PHOLD(p); 2780 proc_getauxv(curthread, p, sb); 2781 PRELE(p); 2782 } 2783 } 2784 2785 static void 2786 __elfN(note_procstat_kqueues)(void *arg, struct sbuf *sb, size_t *sizep) 2787 { 2788 struct proc *p; 2789 size_t size, sect_sz, i; 2790 ssize_t start_len, sect_len; 2791 int structsize; 2792 bool compat32; 2793 2794 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2795 compat32 = true; 2796 structsize = sizeof(struct kinfo_knote32); 2797 #else 2798 compat32 = false; 2799 structsize = sizeof(struct kinfo_knote); 2800 #endif 2801 p = arg; 2802 if (sb == NULL) { 2803 size = 0; 2804 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2805 sbuf_set_drain(sb, sbuf_count_drain, &size); 2806 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2807 kern_proc_kqueues_out(p, sb, -1, compat32); 2808 sbuf_finish(sb); 2809 sbuf_delete(sb); 2810 *sizep = size; 2811 } else { 2812 sbuf_start_section(sb, &start_len); 2813 2814 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2815 kern_proc_kqueues_out(p, sb, *sizep - sizeof(structsize), 2816 compat32); 2817 2818 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2819 if (sect_len < 0) 2820 return; 2821 sect_sz = sect_len; 2822 2823 KASSERT(sect_sz <= *sizep, 2824 ("kern_proc_kqueue_out did not respect maxlen; " 2825 "requested %zu, got %zu", *sizep - sizeof(structsize), 2826 sect_sz - sizeof(structsize))); 2827 2828 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2829 sbuf_putc(sb, 0); 2830 } 2831 } 2832 2833 #define MAX_NOTES_LOOP 4096 2834 bool 2835 __elfN(parse_notes)(const struct image_params *imgp, const Elf_Note *checknote, 2836 const char *note_vendor, const Elf_Phdr *pnote, 2837 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg) 2838 { 2839 const Elf_Note *note, *note0, *note_end; 2840 const char *note_name; 2841 char *buf; 2842 int i, error; 2843 bool res; 2844 2845 /* We need some limit, might as well use PAGE_SIZE. */ 2846 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2847 return (false); 2848 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2849 if (pnote->p_offset > PAGE_SIZE || 2850 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2851 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT); 2852 if (buf == NULL) { 2853 VOP_UNLOCK(imgp->vp); 2854 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2855 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 2856 } 2857 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2858 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2859 imgp->td->td_ucred, NOCRED, NULL, imgp->td); 2860 if (error != 0) { 2861 uprintf("i/o error PT_NOTE\n"); 2862 goto retf; 2863 } 2864 note = note0 = (const Elf_Note *)buf; 2865 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2866 } else { 2867 note = note0 = (const Elf_Note *)(imgp->image_header + 2868 pnote->p_offset); 2869 note_end = (const Elf_Note *)(imgp->image_header + 2870 pnote->p_offset + pnote->p_filesz); 2871 buf = NULL; 2872 } 2873 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end; 2874 i++) { 2875 if (!aligned(note, Elf32_Addr)) { 2876 uprintf("Unaligned ELF note\n"); 2877 goto retf; 2878 } 2879 if ((const char *)note_end - (const char *)note < 2880 sizeof(Elf_Note)) { 2881 uprintf("ELF note too short\n"); 2882 goto retf; 2883 } 2884 if (note->n_namesz != checknote->n_namesz || 2885 note->n_descsz != checknote->n_descsz || 2886 note->n_type != checknote->n_type) 2887 goto nextnote; 2888 note_name = (const char *)(note + 1); 2889 if (note_name + roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2890 note->n_descsz > (const char *)note_end || 2891 strncmp(note_vendor, note_name, checknote->n_namesz) != 0) 2892 goto nextnote; 2893 2894 if (cb(note, cb_arg, &res)) 2895 goto ret; 2896 nextnote: 2897 note = (const Elf_Note *)((const char *)(note + 1) + 2898 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2899 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2900 } 2901 if (i >= MAX_NOTES_LOOP) 2902 uprintf("ELF note parser reached %d notes\n", i); 2903 retf: 2904 res = false; 2905 ret: 2906 free(buf, M_TEMP); 2907 return (res); 2908 } 2909 2910 struct brandnote_cb_arg { 2911 const Elf_Brandnote *brandnote; 2912 int32_t *osrel; 2913 }; 2914 2915 static bool 2916 brandnote_cb(const Elf_Note *note, void *arg0, bool *res) 2917 { 2918 struct brandnote_cb_arg *arg; 2919 2920 arg = arg0; 2921 2922 /* 2923 * Fetch the osreldate for binary from the ELF OSABI-note if 2924 * necessary. 2925 */ 2926 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 && 2927 arg->brandnote->trans_osrel != NULL ? 2928 arg->brandnote->trans_osrel(note, arg->osrel) : true; 2929 2930 return (true); 2931 } 2932 2933 static const Elf_Note fctl_note = { 2934 .n_namesz = sizeof(FREEBSD_ABI_VENDOR), 2935 .n_descsz = sizeof(uint32_t), 2936 .n_type = NT_FREEBSD_FEATURE_CTL, 2937 }; 2938 2939 struct fctl_cb_arg { 2940 bool *has_fctl0; 2941 uint32_t *fctl0; 2942 }; 2943 2944 static bool 2945 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res) 2946 { 2947 struct fctl_cb_arg *arg; 2948 const Elf32_Word *desc; 2949 uintptr_t p; 2950 2951 arg = arg0; 2952 p = (uintptr_t)(note + 1); 2953 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 2954 desc = (const Elf32_Word *)p; 2955 *arg->has_fctl0 = true; 2956 *arg->fctl0 = desc[0]; 2957 *res = true; 2958 return (true); 2959 } 2960 2961 /* 2962 * Try to find the appropriate ABI-note section for checknote, fetch 2963 * the osreldate and feature control flags for binary from the ELF 2964 * OSABI-note. Only the first page of the image is searched, the same 2965 * as for headers. 2966 */ 2967 static bool 2968 __elfN(check_note)(struct image_params *imgp, const Elf_Phdr *phdr, 2969 const Elf_Brandnote *brandnote, int32_t *osrel, bool *has_fctl0, 2970 uint32_t *fctl0) 2971 { 2972 const Elf_Ehdr *hdr; 2973 struct brandnote_cb_arg b_arg; 2974 struct fctl_cb_arg f_arg; 2975 int i, j; 2976 2977 hdr = (const Elf_Ehdr *)imgp->image_header; 2978 b_arg.brandnote = brandnote; 2979 b_arg.osrel = osrel; 2980 f_arg.has_fctl0 = has_fctl0; 2981 f_arg.fctl0 = fctl0; 2982 2983 for (i = 0; i < hdr->e_phnum; i++) { 2984 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, 2985 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb, 2986 &b_arg)) { 2987 for (j = 0; j < hdr->e_phnum; j++) { 2988 if (phdr[j].p_type == PT_NOTE && 2989 __elfN(parse_notes)(imgp, &fctl_note, 2990 FREEBSD_ABI_VENDOR, &phdr[j], 2991 note_fctl_cb, &f_arg)) 2992 break; 2993 } 2994 return (true); 2995 } 2996 } 2997 return (false); 2998 2999 } 3000 3001 /* 3002 * Tell kern_execve.c about it, with a little help from the linker. 3003 */ 3004 static struct execsw __elfN(execsw) = { 3005 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 3006 .ex_name = ELF_ABI_NAME 3007 }; 3008 EXEC_SET(ELF_ABI_ID, __elfN(execsw)); 3009 3010 static vm_prot_t 3011 __elfN(trans_prot)(Elf_Word flags) 3012 { 3013 vm_prot_t prot; 3014 3015 prot = 0; 3016 if (flags & PF_X) 3017 prot |= VM_PROT_EXECUTE; 3018 if (flags & PF_W) 3019 prot |= VM_PROT_WRITE; 3020 if (flags & PF_R) 3021 prot |= VM_PROT_READ; 3022 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 3023 if (i386_read_exec && (flags & PF_R)) 3024 prot |= VM_PROT_EXECUTE; 3025 #endif 3026 return (prot); 3027 } 3028 3029 static Elf_Word 3030 __elfN(untrans_prot)(vm_prot_t prot) 3031 { 3032 Elf_Word flags; 3033 3034 flags = 0; 3035 if (prot & VM_PROT_EXECUTE) 3036 flags |= PF_X; 3037 if (prot & VM_PROT_READ) 3038 flags |= PF_R; 3039 if (prot & VM_PROT_WRITE) 3040 flags |= PF_W; 3041 return (flags); 3042 } 3043