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 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO); 807 nd = &tempdata->nd; 808 attr = &tempdata->attr; 809 imgp = &tempdata->image_params; 810 811 /* 812 * Initialize part of the common data 813 */ 814 imgp->td = td; 815 imgp->proc = td->td_proc; 816 imgp->attr = attr; 817 imgp->interpreted = IMGACT_INTERP_ELF; /* ignored by do_execve */ 818 819 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF, 820 UIO_SYSSPACE, file); 821 if ((error = namei(nd)) != 0) { 822 nd->ni_vp = NULL; 823 goto fail; 824 } 825 NDFREE_PNBUF(nd); 826 imgp->vp = nd->ni_vp; 827 828 /* 829 * Check permissions, modes, uid, etc on the file, and "open" it. 830 */ 831 error = exec_check_permissions(imgp); 832 if (error) 833 goto fail; 834 835 error = exec_map_first_page(imgp); 836 if (error) 837 goto fail; 838 839 imgp->object = nd->ni_vp->v_object; 840 841 hdr = (const Elf_Ehdr *)imgp->image_header; 842 if ((error = __elfN(check_header)(hdr)) != 0) 843 goto fail; 844 if (hdr->e_type == ET_DYN) 845 rbase = *addr; 846 else if (hdr->e_type == ET_EXEC) 847 rbase = 0; 848 else { 849 error = ENOEXEC; 850 goto fail; 851 } 852 853 if (hdr->e_phnum > __elfN(phnums)) { 854 error = ENOEXEC; 855 goto fail; 856 } 857 if (__elfN(phdr_in_zero_page)(hdr) && 858 aligned(imgp->image_header + hdr->e_phoff, Elf_Addr)) { 859 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 860 } else { 861 VOP_UNLOCK(imgp->vp); 862 phdr = m_phdrs = malloc(hdr->e_phnum * sizeof(Elf_Phdr), 863 M_TEMP, M_WAITOK | M_ZERO); 864 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 865 error = vn_rdwr(UIO_READ, imgp->vp, m_phdrs, 866 hdr->e_phnum * sizeof(Elf_Phdr), hdr->e_phoff, 867 UIO_SYSSPACE, IO_NODELOCKED, imgp->td->td_ucred, 868 NOCRED, NULL, imgp->td); 869 if (error != 0) 870 goto fail; 871 } 872 873 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr); 874 if (error != 0) 875 goto fail; 876 877 if (imgp->proc->p_sysent->sv_protect != NULL) 878 imgp->proc->p_sysent->sv_protect(imgp, SVP_INTERP); 879 880 *addr = base_addr; 881 *entry = (unsigned long)hdr->e_entry + rbase; 882 883 fail: 884 if (imgp->firstpage) 885 exec_unmap_first_page(imgp); 886 887 if (nd->ni_vp) { 888 if (imgp->textset) 889 VOP_UNSET_TEXT_CHECKED(nd->ni_vp); 890 vput(nd->ni_vp); 891 } 892 free(m_phdrs, M_TEMP); 893 free(tempdata, M_TEMP); 894 895 return (error); 896 } 897 898 /* 899 * Select randomized valid address in the map map, between minv and 900 * maxv, with specified alignment. The [minv, maxv) range must belong 901 * to the map. Note that function only allocates the address, it is 902 * up to caller to clamp maxv in a way that the final allocation 903 * length fit into the map. 904 * 905 * Result is returned in *resp, error code indicates that arguments 906 * did not pass sanity checks for overflow and range correctness. 907 */ 908 static int 909 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv, 910 u_int align, u_long *resp) 911 { 912 u_long rbase, res; 913 914 MPASS(vm_map_min(map) <= minv); 915 916 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) { 917 uprintf("Invalid ELF segments layout\n"); 918 return (ENOEXEC); 919 } 920 921 arc4rand(&rbase, sizeof(rbase), 0); 922 res = roundup(minv, (u_long)align) + rbase % (maxv - minv); 923 res &= ~((u_long)align - 1); 924 if (res >= maxv) 925 res -= align; 926 927 KASSERT(res >= minv, 928 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx", 929 res, minv, maxv, rbase)); 930 KASSERT(res < maxv, 931 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx", 932 res, maxv, minv, rbase)); 933 934 *resp = res; 935 return (0); 936 } 937 938 static int 939 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr, 940 const Elf_Phdr *phdr) 941 { 942 struct vmspace *vmspace; 943 const char *err_str; 944 u_long text_size, data_size, total_size, text_addr, data_addr; 945 u_long seg_size, seg_addr; 946 int i; 947 948 err_str = NULL; 949 text_size = data_size = total_size = text_addr = data_addr = 0; 950 951 for (i = 0; i < hdr->e_phnum; i++) { 952 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 953 continue; 954 955 seg_addr = trunc_page(phdr[i].p_vaddr + imgp->et_dyn_addr); 956 seg_size = round_page(phdr[i].p_memsz + 957 phdr[i].p_vaddr + imgp->et_dyn_addr - seg_addr); 958 959 /* 960 * Make the largest executable segment the official 961 * text segment and all others data. 962 * 963 * Note that obreak() assumes that data_addr + data_size == end 964 * of data load area, and the ELF file format expects segments 965 * to be sorted by address. If multiple data segments exist, 966 * the last one will be used. 967 */ 968 969 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) { 970 text_size = seg_size; 971 text_addr = seg_addr; 972 } else { 973 data_size = seg_size; 974 data_addr = seg_addr; 975 } 976 total_size += seg_size; 977 } 978 979 if (data_addr == 0 && data_size == 0) { 980 data_addr = text_addr; 981 data_size = text_size; 982 } 983 984 /* 985 * Check limits. It should be safe to check the 986 * limits after loading the segments since we do 987 * not actually fault in all the segments pages. 988 */ 989 PROC_LOCK(imgp->proc); 990 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 991 err_str = "Data segment size exceeds process limit"; 992 else if (text_size > maxtsiz) 993 err_str = "Text segment size exceeds system limit"; 994 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 995 err_str = "Total segment size exceeds process limit"; 996 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 997 err_str = "Data segment size exceeds resource limit"; 998 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 999 err_str = "Total segment size exceeds resource limit"; 1000 PROC_UNLOCK(imgp->proc); 1001 if (err_str != NULL) { 1002 uprintf("%s\n", err_str); 1003 return (ENOMEM); 1004 } 1005 1006 vmspace = imgp->proc->p_vmspace; 1007 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 1008 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 1009 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 1010 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 1011 1012 return (0); 1013 } 1014 1015 static int 1016 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr, 1017 char **interpp, bool *free_interpp) 1018 { 1019 char *interp; 1020 int error, interp_name_len; 1021 1022 KASSERT(phdr->p_type == PT_INTERP, 1023 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type)); 1024 ASSERT_VOP_LOCKED(imgp->vp, __func__); 1025 1026 /* Path to interpreter */ 1027 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) { 1028 uprintf("Invalid PT_INTERP\n"); 1029 return (ENOEXEC); 1030 } 1031 1032 interp_name_len = phdr->p_filesz; 1033 if (phdr->p_offset > PAGE_SIZE || 1034 interp_name_len > PAGE_SIZE - phdr->p_offset) { 1035 /* 1036 * The vnode lock might be needed by the pagedaemon to 1037 * clean pages owned by the vnode. Do not allow sleep 1038 * waiting for memory with the vnode locked, instead 1039 * try non-sleepable allocation first, and if it 1040 * fails, go to the slow path were we drop the lock 1041 * and do M_WAITOK. A text reference prevents 1042 * modifications to the vnode content. 1043 */ 1044 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT); 1045 if (interp == NULL) { 1046 VOP_UNLOCK(imgp->vp); 1047 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK); 1048 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1049 } 1050 1051 error = vn_rdwr(UIO_READ, imgp->vp, interp, 1052 interp_name_len, phdr->p_offset, 1053 UIO_SYSSPACE, IO_NODELOCKED, imgp->td->td_ucred, 1054 NOCRED, NULL, imgp->td); 1055 if (error != 0) { 1056 free(interp, M_TEMP); 1057 uprintf("i/o error PT_INTERP %d\n", error); 1058 return (error); 1059 } 1060 interp[interp_name_len] = '\0'; 1061 1062 *interpp = interp; 1063 *free_interpp = true; 1064 return (0); 1065 } 1066 1067 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset; 1068 if (interp[interp_name_len - 1] != '\0') { 1069 uprintf("Invalid PT_INTERP\n"); 1070 return (ENOEXEC); 1071 } 1072 1073 *interpp = interp; 1074 *free_interpp = false; 1075 return (0); 1076 } 1077 1078 static int 1079 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info, 1080 const char *interp, u_long *addr, u_long *entry) 1081 { 1082 int error; 1083 1084 if (brand_info->interp_newpath != NULL && 1085 (brand_info->interp_path == NULL || 1086 strcmp(interp, brand_info->interp_path) == 0)) { 1087 error = __elfN(load_interp_file)(imgp->td, 1088 brand_info->interp_newpath, addr, entry); 1089 if (error == 0) 1090 return (0); 1091 } 1092 1093 error = __elfN(load_interp_file)(imgp->td, interp, addr, entry); 1094 if (error == 0) 1095 return (0); 1096 1097 uprintf("ELF interpreter %s not found, error %d\n", interp, error); 1098 return (error); 1099 } 1100 1101 /* 1102 * Impossible et_dyn_addr initial value indicating that the real base 1103 * must be calculated later with some randomization applied. 1104 */ 1105 #define ET_DYN_ADDR_RAND 1 1106 1107 static int 1108 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 1109 { 1110 const Elf_Ehdr *hdr; 1111 const Elf_Phdr *phdr; 1112 Elf_Auxargs *elf_auxargs; 1113 struct vmspace *vmspace; 1114 vm_map_t map; 1115 char *interp; 1116 const Elf_Brandinfo *brand_info; 1117 struct sysentvec *sv; 1118 void *m_phdrs; 1119 u_long addr, baddr, entry, proghdr; 1120 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc; 1121 uint32_t fctl0; 1122 int32_t osrel; 1123 bool free_interp; 1124 int error, i, n; 1125 1126 hdr = (const Elf_Ehdr *)imgp->image_header; 1127 1128 /* 1129 * Do we have a valid ELF header ? 1130 * 1131 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 1132 * if particular brand doesn't support it. 1133 */ 1134 if (__elfN(check_header)(hdr) != 0 || 1135 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 1136 return (-1); 1137 1138 /* 1139 * From here on down, we return an errno, not -1, as we've 1140 * detected an ELF file. 1141 */ 1142 1143 n = error = 0; 1144 baddr = 0; 1145 osrel = 0; 1146 fctl0 = 0; 1147 entry = proghdr = 0; 1148 interp = NULL; 1149 free_interp = false; 1150 m_phdrs = NULL; 1151 1152 if (hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize < hdr->e_phoff) { 1153 uprintf("PHDRS wrap\n"); 1154 return (ENOEXEC); 1155 } 1156 if (hdr->e_phnum > __elfN(phnums)) { 1157 uprintf("Too many program headers (%u, %u max)\n", 1158 hdr->e_phnum, __elfN(phnums)); 1159 return (ENOEXEC); 1160 } 1161 if (__elfN(phdr_in_zero_page)(hdr) && 1162 aligned(imgp->image_header + hdr->e_phoff, Elf_Addr)) { 1163 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1164 } else { 1165 VOP_UNLOCK(imgp->vp); 1166 phdr = m_phdrs = malloc(hdr->e_phnum * sizeof(Elf_Phdr), 1167 M_TEMP, M_WAITOK | M_ZERO); 1168 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1169 error = vn_rdwr(UIO_READ, imgp->vp, m_phdrs, 1170 hdr->e_phnum * sizeof(Elf_Phdr), hdr->e_phoff, 1171 UIO_SYSSPACE, IO_NODELOCKED, imgp->td->td_ucred, 1172 NOCRED, NULL, imgp->td); 1173 if (error != 0) 1174 goto ret; 1175 } 1176 1177 /* 1178 * Somewhat arbitrary, limit accepted max alignment for the 1179 * loadable segment to the max supported superpage size. Too 1180 * large alignment requests are not useful and are indicators 1181 * of corrupted or outright malicious binary. 1182 */ 1183 maxalign = PAGE_SIZE; 1184 maxsalign = PAGE_SIZE * 1024; 1185 for (i = MAXPAGESIZES - 1; i > 0; i--) { 1186 if (pagesizes[i] > maxsalign) { 1187 maxsalign = pagesizes[i]; 1188 break; 1189 } 1190 } 1191 1192 mapsz = 0; 1193 1194 for (i = 0; i < hdr->e_phnum; i++) { 1195 switch (phdr[i].p_type) { 1196 case PT_LOAD: 1197 if (n == 0) 1198 baddr = phdr[i].p_vaddr; 1199 if (!powerof2(phdr[i].p_align) || 1200 phdr[i].p_align > maxsalign) { 1201 uprintf("Invalid segment alignment\n"); 1202 error = ENOEXEC; 1203 goto ret; 1204 } 1205 if (phdr[i].p_align > maxalign) 1206 maxalign = phdr[i].p_align; 1207 if (mapsz + phdr[i].p_memsz < mapsz) { 1208 uprintf("Mapsize overflow\n"); 1209 error = ENOEXEC; 1210 goto ret; 1211 } 1212 mapsz += phdr[i].p_memsz; 1213 n++; 1214 1215 /* 1216 * If this segment contains the program headers, 1217 * remember their virtual address for the AT_PHDR 1218 * aux entry. Static binaries don't usually include 1219 * a PT_PHDR entry. 1220 */ 1221 if (phdr[i].p_offset == 0 && 1222 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= 1223 phdr[i].p_filesz) 1224 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 1225 break; 1226 case PT_INTERP: 1227 /* Path to interpreter */ 1228 if (interp != NULL) { 1229 uprintf("Multiple PT_INTERP headers\n"); 1230 error = ENOEXEC; 1231 goto ret; 1232 } 1233 error = __elfN(get_interp)(imgp, &phdr[i], &interp, 1234 &free_interp); 1235 if (error != 0) 1236 goto ret; 1237 break; 1238 case PT_GNU_STACK: 1239 if (__elfN(nxstack)) { 1240 imgp->stack_prot = 1241 __elfN(trans_prot)(phdr[i].p_flags); 1242 if ((imgp->stack_prot & VM_PROT_RW) != 1243 VM_PROT_RW) { 1244 uprintf("Invalid PT_GNU_STACK\n"); 1245 error = ENOEXEC; 1246 goto ret; 1247 } 1248 } 1249 imgp->stack_sz = phdr[i].p_memsz; 1250 break; 1251 case PT_PHDR: /* Program header table info */ 1252 proghdr = phdr[i].p_vaddr; 1253 break; 1254 } 1255 } 1256 1257 brand_info = __elfN(get_brandinfo)(imgp, phdr, interp, &osrel, &fctl0); 1258 if (brand_info == NULL) { 1259 uprintf("ELF binary type \"%u\" not known.\n", 1260 hdr->e_ident[EI_OSABI]); 1261 error = ENOEXEC; 1262 goto ret; 1263 } 1264 1265 /* 1266 * Avoid a possible deadlock if the current address space is destroyed 1267 * and that address space maps the locked vnode. In the common case, 1268 * the locked vnode's v_usecount is decremented but remains greater 1269 * than zero. Consequently, the vnode lock is not needed by vrele(). 1270 * However, in cases where the vnode lock is external, such as nullfs, 1271 * v_usecount may become zero. 1272 * 1273 * The VV_TEXT flag prevents modifications to the executable while 1274 * the vnode is unlocked. 1275 */ 1276 VOP_UNLOCK(imgp->vp); 1277 1278 /* 1279 * Decide whether to enable randomization of user mappings. First, 1280 * reset user preferences for the setid binaries. Then, account for the 1281 * support of randomization by the ABI, by user preferences, and make 1282 * special treatment for PIE binaries. 1283 */ 1284 if (imgp->credential_setid) { 1285 PROC_LOCK(imgp->proc); 1286 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE | 1287 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 1288 PROC_UNLOCK(imgp->proc); 1289 } 1290 1291 sv = brand_info->sysvec; 1292 if (hdr->e_type == ET_DYN) { 1293 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 1294 uprintf("Cannot execute shared object\n"); 1295 error = ENOEXEC; 1296 (void)vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1297 goto ret; 1298 } 1299 /* 1300 * Honour the base load address from the dso if it is 1301 * non-zero for some reason. 1302 */ 1303 if (baddr == 0) { 1304 if ((sv->sv_flags & SV_ASLR) == 0 || 1305 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) 1306 imgp->et_dyn_addr = __elfN(pie_base); 1307 else if ((__elfN(pie_aslr_enabled) && 1308 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) || 1309 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0) 1310 imgp->et_dyn_addr = ET_DYN_ADDR_RAND; 1311 else 1312 imgp->et_dyn_addr = __elfN(pie_base); 1313 } 1314 } 1315 if ((sv->sv_flags & SV_ASLR) == 0 || 1316 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 || 1317 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) { 1318 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND, 1319 ("imgp->et_dyn_addr == RAND and !ASLR")); 1320 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 || 1321 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) || 1322 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1323 imgp->map_flags |= MAP_ASLR; 1324 /* 1325 * If user does not care about sbrk, utilize the bss 1326 * grow region for mappings as well. We can select 1327 * the base for the image anywere and still not suffer 1328 * from the fragmentation. 1329 */ 1330 if (!__elfN(aslr_honor_sbrk) || 1331 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0) 1332 imgp->map_flags |= MAP_ASLR_IGNSTART; 1333 if (__elfN(aslr_stack)) 1334 imgp->map_flags |= MAP_ASLR_STACK; 1335 if (__elfN(aslr_shared_page)) 1336 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE; 1337 } 1338 1339 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 && 1340 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) || 1341 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0) 1342 imgp->map_flags |= MAP_WXORX; 1343 1344 error = exec_new_vmspace(imgp, sv); 1345 1346 imgp->proc->p_sysent = sv; 1347 imgp->proc->p_elf_brandinfo = brand_info; 1348 1349 vmspace = imgp->proc->p_vmspace; 1350 map = &vmspace->vm_map; 1351 maxv = sv->sv_usrstack; 1352 if ((imgp->map_flags & MAP_ASLR_STACK) == 0) 1353 maxv -= lim_max(imgp->td, RLIMIT_STACK); 1354 if (error == 0 && mapsz >= maxv - vm_map_min(map)) { 1355 uprintf("Excessive mapping size\n"); 1356 error = ENOEXEC; 1357 } 1358 1359 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1360 KASSERT((map->flags & MAP_ASLR) != 0, 1361 ("ET_DYN_ADDR_RAND but !MAP_ASLR")); 1362 error = __CONCAT(rnd_, __elfN(base))(map, 1363 vm_map_min(map) + mapsz + lim_max(imgp->td, RLIMIT_DATA), 1364 /* reserve half of the address space to interpreter */ 1365 maxv / 2, maxalign, &imgp->et_dyn_addr); 1366 } 1367 1368 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1369 if (error != 0) 1370 goto ret; 1371 1372 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL); 1373 if (error != 0) 1374 goto ret; 1375 1376 error = __elfN(enforce_limits)(imgp, hdr, phdr); 1377 if (error != 0) 1378 goto ret; 1379 1380 /* 1381 * We load the dynamic linker where a userland call 1382 * to mmap(0, ...) would put it. The rationale behind this 1383 * calculation is that it leaves room for the heap to grow to 1384 * its maximum allowed size. 1385 */ 1386 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->td, 1387 RLIMIT_DATA)); 1388 if ((map->flags & MAP_ASLR) != 0) { 1389 maxv1 = maxv / 2 + addr / 2; 1390 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1, 1391 #if VM_NRESERVLEVEL > 0 1392 pagesizes[VM_NRESERVLEVEL] != 0 ? 1393 /* Align anon_loc to the largest superpage size. */ 1394 pagesizes[VM_NRESERVLEVEL] : 1395 #endif 1396 pagesizes[0], &anon_loc); 1397 if (error != 0) 1398 goto ret; 1399 map->anon_loc = anon_loc; 1400 } else { 1401 map->anon_loc = addr; 1402 } 1403 1404 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr; 1405 imgp->entry_addr = entry; 1406 1407 if (sv->sv_protect != NULL) 1408 sv->sv_protect(imgp, SVP_IMAGE); 1409 1410 if (interp != NULL) { 1411 VOP_UNLOCK(imgp->vp); 1412 if ((map->flags & MAP_ASLR) != 0) { 1413 /* Assume that interpreter fits into 1/4 of AS */ 1414 maxv1 = maxv / 2 + addr / 2; 1415 error = __CONCAT(rnd_, __elfN(base))(map, addr, 1416 maxv1, PAGE_SIZE, &addr); 1417 } 1418 if (error == 0) { 1419 error = __elfN(load_interp)(imgp, brand_info, interp, 1420 &addr, &imgp->entry_addr); 1421 } 1422 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1423 if (error != 0) 1424 goto ret; 1425 } else 1426 addr = imgp->et_dyn_addr; 1427 1428 error = exec_map_stack(imgp); 1429 if (error != 0) 1430 goto ret; 1431 1432 /* 1433 * Construct auxargs table (used by the copyout_auxargs routine) 1434 */ 1435 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT); 1436 if (elf_auxargs == NULL) { 1437 VOP_UNLOCK(imgp->vp); 1438 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1439 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1440 } 1441 elf_auxargs->execfd = -1; 1442 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr; 1443 elf_auxargs->phent = hdr->e_phentsize; 1444 elf_auxargs->phnum = hdr->e_phnum; 1445 elf_auxargs->pagesz = PAGE_SIZE; 1446 elf_auxargs->base = addr; 1447 elf_auxargs->flags = 0; 1448 elf_auxargs->entry = entry; 1449 elf_auxargs->hdr_eflags = hdr->e_flags; 1450 1451 imgp->auxargs = elf_auxargs; 1452 imgp->interpreted = 0; 1453 imgp->reloc_base = addr; 1454 imgp->proc->p_osrel = osrel; 1455 imgp->proc->p_fctl0 = fctl0; 1456 imgp->proc->p_elf_flags = hdr->e_flags; 1457 1458 ret: 1459 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock"); 1460 if (free_interp) 1461 free(interp, M_TEMP); 1462 free(m_phdrs, M_TEMP); 1463 return (error); 1464 } 1465 1466 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE) 1467 1468 int 1469 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base) 1470 { 1471 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1472 Elf_Auxinfo *argarray, *pos; 1473 struct vmspace *vmspace; 1474 rlim_t stacksz; 1475 int error, oc; 1476 uint32_t bsdflags; 1477 1478 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1479 M_WAITOK | M_ZERO); 1480 1481 vmspace = imgp->proc->p_vmspace; 1482 1483 if (args->execfd != -1) 1484 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1485 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1486 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1487 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1488 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1489 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1490 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1491 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1492 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1493 if (imgp->execpathp != 0) 1494 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp); 1495 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1496 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1497 if (imgp->canary != 0) { 1498 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary); 1499 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1500 } 1501 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1502 if (imgp->pagesizes != 0) { 1503 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes); 1504 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1505 } 1506 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) { 1507 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1508 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset); 1509 } 1510 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1511 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1512 imgp->sysent->sv_stackprot); 1513 if (imgp->sysent->sv_hwcap != NULL) 1514 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1515 if (imgp->sysent->sv_hwcap2 != NULL) 1516 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1517 if (imgp->sysent->sv_hwcap3 != NULL) 1518 AUXARGS_ENTRY(pos, AT_HWCAP3, *imgp->sysent->sv_hwcap3); 1519 if (imgp->sysent->sv_hwcap4 != NULL) 1520 AUXARGS_ENTRY(pos, AT_HWCAP4, *imgp->sysent->sv_hwcap4); 1521 bsdflags = 0; 1522 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0; 1523 oc = atomic_load_int(&vm_overcommit); 1524 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) != 1525 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0; 1526 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags); 1527 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc); 1528 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv); 1529 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc); 1530 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv); 1531 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings); 1532 #ifdef RANDOM_FENESTRASX 1533 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) { 1534 AUXARGS_ENTRY(pos, AT_FXRNG, 1535 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset); 1536 } 1537 #endif 1538 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) { 1539 AUXARGS_ENTRY(pos, AT_KPRELOAD, 1540 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset); 1541 } 1542 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop)); 1543 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur; 1544 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz); 1545 AUXARGS_ENTRY(pos, AT_NULL, 0); 1546 1547 free(imgp->auxargs, M_TEMP); 1548 imgp->auxargs = NULL; 1549 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1550 1551 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT); 1552 free(argarray, M_TEMP); 1553 return (error); 1554 } 1555 1556 int 1557 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp) 1558 { 1559 Elf_Addr *base; 1560 1561 base = (Elf_Addr *)*stack_base; 1562 base--; 1563 if (elf_suword(base, imgp->args->argc) == -1) 1564 return (EFAULT); 1565 *stack_base = (uintptr_t)base; 1566 return (0); 1567 } 1568 1569 /* 1570 * Code for generating ELF core dumps. 1571 */ 1572 1573 typedef void (*segment_callback)(vm_map_entry_t, void *); 1574 1575 /* Closure for cb_put_phdr(). */ 1576 struct phdr_closure { 1577 Elf_Phdr *phdr; /* Program header to fill in */ 1578 Elf_Off offset; /* Offset of segment in core file */ 1579 int numsegs; /* Maximum number of segments */ 1580 int nextseg; /* Next segment to fill in */ 1581 }; 1582 1583 struct note_info { 1584 int type; /* Note type. */ 1585 struct regset *regset; /* Register set. */ 1586 outfunc_t outfunc; /* Output function. */ 1587 void *outarg; /* Argument for the output function. */ 1588 size_t outsize; /* Output size. */ 1589 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1590 }; 1591 1592 TAILQ_HEAD(note_info_list, note_info); 1593 1594 static void cb_put_phdr(vm_map_entry_t, void *); 1595 static void cb_size_segment(vm_map_entry_t, void *); 1596 static void each_dumpable_segment(struct thread *, segment_callback, void *, 1597 int); 1598 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1599 struct note_info_list *, size_t, int); 1600 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *); 1601 1602 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1603 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1604 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1605 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1606 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1607 static void __elfN(note_procstat_kqueues)(void *, struct sbuf *, size_t *); 1608 static void note_procstat_files(void *, struct sbuf *, size_t *); 1609 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1610 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1611 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1612 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1613 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1614 1615 static int 1616 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1617 { 1618 1619 return (core_write((struct coredump_params *)arg, base, len, offset, 1620 UIO_SYSSPACE, NULL)); 1621 } 1622 1623 int 1624 __elfN(coredump)(struct thread *td, struct coredump_writer *cdw, off_t limit, int flags) 1625 { 1626 struct ucred *cred = td->td_ucred; 1627 int compm, error = 0; 1628 struct sseg_closure seginfo; 1629 struct note_info_list notelst; 1630 struct coredump_params params; 1631 struct note_info *ninfo; 1632 void *hdr, *tmpbuf; 1633 size_t hdrsize, notesz, coresize; 1634 1635 hdr = NULL; 1636 tmpbuf = NULL; 1637 TAILQ_INIT(¬elst); 1638 1639 /* Size the program segments. */ 1640 __elfN(size_segments)(td, &seginfo, flags); 1641 1642 /* 1643 * Collect info about the core file header area. 1644 */ 1645 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1646 if (seginfo.count + 1 >= PN_XNUM) 1647 hdrsize += sizeof(Elf_Shdr); 1648 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz); 1649 coresize = round_page(hdrsize + notesz) + seginfo.size; 1650 1651 /* Set up core dump parameters. */ 1652 params.offset = 0; 1653 params.active_cred = cred; 1654 params.td = td; 1655 params.cdw = cdw; 1656 params.comp = NULL; 1657 1658 #ifdef RACCT 1659 if (racct_enable) { 1660 PROC_LOCK(td->td_proc); 1661 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1662 PROC_UNLOCK(td->td_proc); 1663 if (error != 0) { 1664 error = EFAULT; 1665 goto done; 1666 } 1667 } 1668 #endif 1669 if (coresize >= limit) { 1670 error = EFAULT; 1671 goto done; 1672 } 1673 1674 /* Create a compression stream if necessary. */ 1675 compm = compress_user_cores; 1676 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP && 1677 compm == 0) 1678 compm = COMPRESS_GZIP; 1679 if (compm != 0) { 1680 params.comp = compressor_init(core_compressed_write, 1681 compm, CORE_BUF_SIZE, 1682 compress_user_cores_level, ¶ms); 1683 if (params.comp == NULL) { 1684 error = EFAULT; 1685 goto done; 1686 } 1687 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1688 } 1689 1690 if (cdw->init_fn != NULL) { 1691 error = (*cdw->init_fn)(cdw, ¶ms); 1692 if (error != 0) 1693 goto done; 1694 } 1695 1696 /* 1697 * Allocate memory for building the header, fill it up, and write it out 1698 * following the notes. 1699 * 1700 * Note that a process sharing our vmspace might be concurrently 1701 * mutating the map, in which case we could populate fewer than 1702 * seginfo.count headers. Zero the buffer to ensure that unpopulated 1703 * headers are still initialized. 1704 */ 1705 hdr = malloc(hdrsize, M_TEMP, M_WAITOK | M_ZERO); 1706 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1707 notesz, flags); 1708 1709 /* Write the contents of all of the writable segments. */ 1710 if (error == 0) { 1711 Elf_Phdr *php; 1712 off_t offset; 1713 int i; 1714 1715 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1716 offset = round_page(hdrsize + notesz); 1717 for (i = 0; i < seginfo.count; i++) { 1718 error = core_output((char *)(uintptr_t)php->p_vaddr, 1719 php->p_filesz, offset, ¶ms, tmpbuf); 1720 if (error != 0) 1721 break; 1722 offset += php->p_filesz; 1723 php++; 1724 } 1725 if (error == 0 && params.comp != NULL) 1726 error = compressor_flush(params.comp); 1727 } 1728 if (error) { 1729 log(LOG_WARNING, 1730 "Failed to write core file for process %s (error %d)\n", 1731 curproc->p_comm, error); 1732 } 1733 1734 done: 1735 free(tmpbuf, M_TEMP); 1736 if (params.comp != NULL) 1737 compressor_fini(params.comp); 1738 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1739 TAILQ_REMOVE(¬elst, ninfo, link); 1740 free(ninfo, M_TEMP); 1741 } 1742 if (hdr != NULL) 1743 free(hdr, M_TEMP); 1744 1745 return (error); 1746 } 1747 1748 /* 1749 * A callback for each_dumpable_segment() to write out the segment's 1750 * program header entry. 1751 */ 1752 static void 1753 cb_put_phdr(vm_map_entry_t entry, void *closure) 1754 { 1755 struct phdr_closure *phc = (struct phdr_closure *)closure; 1756 Elf_Phdr *phdr = phc->phdr; 1757 1758 if (phc->nextseg >= phc->numsegs) { 1759 /* Only write as many headers as we have space for. */ 1760 return; 1761 } 1762 1763 phc->offset = round_page(phc->offset); 1764 1765 phdr->p_type = PT_LOAD; 1766 phdr->p_offset = phc->offset; 1767 phdr->p_vaddr = entry->start; 1768 phdr->p_paddr = 0; 1769 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1770 phdr->p_align = PAGE_SIZE; 1771 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1772 1773 phc->offset += phdr->p_filesz; 1774 phc->phdr++; 1775 1776 phc->nextseg++; 1777 } 1778 1779 /* 1780 * A callback for each_dumpable_segment() to gather information about 1781 * the number of segments and their total size. 1782 */ 1783 static void 1784 cb_size_segment(vm_map_entry_t entry, void *closure) 1785 { 1786 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1787 1788 ssc->count++; 1789 ssc->size += entry->end - entry->start; 1790 } 1791 1792 void 1793 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo, 1794 int flags) 1795 { 1796 seginfo->count = 0; 1797 seginfo->size = 0; 1798 1799 each_dumpable_segment(td, cb_size_segment, seginfo, flags); 1800 } 1801 1802 /* 1803 * For each writable segment in the process's memory map, call the given 1804 * function with a pointer to the map entry and some arbitrary 1805 * caller-supplied data. 1806 */ 1807 static void 1808 each_dumpable_segment(struct thread *td, segment_callback func, void *closure, 1809 int flags) 1810 { 1811 struct proc *p = td->td_proc; 1812 vm_map_t map = &p->p_vmspace->vm_map; 1813 vm_map_entry_t entry; 1814 vm_object_t backing_object, object; 1815 bool ignore_entry; 1816 1817 vm_map_lock_read(map); 1818 VM_MAP_ENTRY_FOREACH(entry, map) { 1819 /* 1820 * Don't dump inaccessible mappings, deal with legacy 1821 * coredump mode. 1822 * 1823 * Note that read-only segments related to the elf binary 1824 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1825 * need to arbitrarily ignore such segments. 1826 */ 1827 if ((flags & SVC_ALL) == 0) { 1828 if (elf_legacy_coredump) { 1829 if ((entry->protection & VM_PROT_RW) != 1830 VM_PROT_RW) 1831 continue; 1832 } else { 1833 if ((entry->protection & VM_PROT_ALL) == 0) 1834 continue; 1835 } 1836 } 1837 1838 /* 1839 * Dont include memory segment in the coredump if 1840 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1841 * madvise(2). Do not dump submaps (i.e. parts of the 1842 * kernel map). 1843 */ 1844 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 1845 continue; 1846 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 && 1847 (flags & SVC_ALL) == 0) 1848 continue; 1849 if ((object = entry->object.vm_object) == NULL) 1850 continue; 1851 1852 /* Ignore memory-mapped devices and such things. */ 1853 VM_OBJECT_RLOCK(object); 1854 while ((backing_object = object->backing_object) != NULL) { 1855 VM_OBJECT_RLOCK(backing_object); 1856 VM_OBJECT_RUNLOCK(object); 1857 object = backing_object; 1858 } 1859 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0; 1860 VM_OBJECT_RUNLOCK(object); 1861 if (ignore_entry) 1862 continue; 1863 1864 (*func)(entry, closure); 1865 } 1866 vm_map_unlock_read(map); 1867 } 1868 1869 /* 1870 * Write the core file header to the file, including padding up to 1871 * the page boundary. 1872 */ 1873 static int 1874 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1875 size_t hdrsize, struct note_info_list *notelst, size_t notesz, 1876 int flags) 1877 { 1878 struct note_info *ninfo; 1879 struct sbuf *sb; 1880 int error; 1881 1882 /* Fill in the header. */ 1883 bzero(hdr, hdrsize); 1884 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags); 1885 1886 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1887 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1888 sbuf_start_section(sb, NULL); 1889 sbuf_bcat(sb, hdr, hdrsize); 1890 TAILQ_FOREACH(ninfo, notelst, link) 1891 __elfN(putnote)(p->td, ninfo, sb); 1892 /* Align up to a page boundary for the program segments. */ 1893 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1894 error = sbuf_finish(sb); 1895 sbuf_delete(sb); 1896 1897 return (error); 1898 } 1899 1900 void 1901 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1902 size_t *sizep) 1903 { 1904 struct proc *p; 1905 struct thread *thr; 1906 size_t size; 1907 1908 p = td->td_proc; 1909 size = 0; 1910 1911 size += __elfN(register_note)(td, list, NT_PRPSINFO, 1912 __elfN(note_prpsinfo), p); 1913 1914 /* 1915 * To have the debugger select the right thread (LWP) as the initial 1916 * thread, we dump the state of the thread passed to us in td first. 1917 * This is the thread that causes the core dump and thus likely to 1918 * be the right thread one wants to have selected in the debugger. 1919 */ 1920 thr = td; 1921 while (thr != NULL) { 1922 size += __elfN(prepare_register_notes)(td, list, thr); 1923 size += __elfN(register_note)(td, list, -1, 1924 __elfN(note_threadmd), thr); 1925 1926 thr = thr == td ? TAILQ_FIRST(&p->p_threads) : 1927 TAILQ_NEXT(thr, td_plist); 1928 if (thr == td) 1929 thr = TAILQ_NEXT(thr, td_plist); 1930 } 1931 1932 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC, 1933 __elfN(note_procstat_proc), p); 1934 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES, 1935 note_procstat_files, p); 1936 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP, 1937 note_procstat_vmmap, p); 1938 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS, 1939 note_procstat_groups, p); 1940 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK, 1941 note_procstat_umask, p); 1942 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT, 1943 note_procstat_rlimit, p); 1944 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL, 1945 note_procstat_osrel, p); 1946 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS, 1947 __elfN(note_procstat_psstrings), p); 1948 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV, 1949 __elfN(note_procstat_auxv), p); 1950 size += __elfN(register_note)(td, list, NT_PROCSTAT_KQUEUES, 1951 __elfN(note_procstat_kqueues), p); 1952 1953 *sizep = size; 1954 } 1955 1956 void 1957 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1958 size_t notesz, int flags) 1959 { 1960 Elf_Ehdr *ehdr; 1961 Elf_Phdr *phdr; 1962 Elf_Shdr *shdr; 1963 struct phdr_closure phc; 1964 const Elf_Brandinfo *bi; 1965 1966 ehdr = (Elf_Ehdr *)hdr; 1967 bi = td->td_proc->p_elf_brandinfo; 1968 1969 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1970 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1971 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1972 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1973 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1974 ehdr->e_ident[EI_DATA] = ELF_DATA; 1975 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1976 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi; 1977 ehdr->e_ident[EI_ABIVERSION] = 0; 1978 ehdr->e_ident[EI_PAD] = 0; 1979 ehdr->e_type = ET_CORE; 1980 ehdr->e_machine = bi->machine; 1981 ehdr->e_version = EV_CURRENT; 1982 ehdr->e_entry = 0; 1983 ehdr->e_phoff = sizeof(Elf_Ehdr); 1984 ehdr->e_flags = td->td_proc->p_elf_flags; 1985 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1986 ehdr->e_phentsize = sizeof(Elf_Phdr); 1987 ehdr->e_shentsize = sizeof(Elf_Shdr); 1988 ehdr->e_shstrndx = SHN_UNDEF; 1989 if (numsegs + 1 < PN_XNUM) { 1990 ehdr->e_phnum = numsegs + 1; 1991 ehdr->e_shnum = 0; 1992 } else { 1993 ehdr->e_phnum = PN_XNUM; 1994 ehdr->e_shnum = 1; 1995 1996 ehdr->e_shoff = ehdr->e_phoff + 1997 (numsegs + 1) * ehdr->e_phentsize; 1998 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1999 ("e_shoff: %zu, hdrsize - shdr: %zu", 2000 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 2001 2002 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 2003 memset(shdr, 0, sizeof(*shdr)); 2004 /* 2005 * A special first section is used to hold large segment and 2006 * section counts. This was proposed by Sun Microsystems in 2007 * Solaris and has been adopted by Linux; the standard ELF 2008 * tools are already familiar with the technique. 2009 * 2010 * See table 7-7 of the Solaris "Linker and Libraries Guide" 2011 * (or 12-7 depending on the version of the document) for more 2012 * details. 2013 */ 2014 shdr->sh_type = SHT_NULL; 2015 shdr->sh_size = ehdr->e_shnum; 2016 shdr->sh_link = ehdr->e_shstrndx; 2017 shdr->sh_info = numsegs + 1; 2018 } 2019 2020 /* 2021 * Fill in the program header entries. 2022 */ 2023 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 2024 2025 /* The note segment. */ 2026 phdr->p_type = PT_NOTE; 2027 phdr->p_offset = hdrsize; 2028 phdr->p_vaddr = 0; 2029 phdr->p_paddr = 0; 2030 phdr->p_filesz = notesz; 2031 phdr->p_memsz = 0; 2032 phdr->p_flags = PF_R; 2033 phdr->p_align = ELF_NOTE_ROUNDSIZE; 2034 phdr++; 2035 2036 /* All the writable segments from the program. */ 2037 phc.phdr = phdr; 2038 phc.offset = round_page(hdrsize + notesz); 2039 phc.numsegs = numsegs; 2040 phc.nextseg = 0; 2041 each_dumpable_segment(td, cb_put_phdr, &phc, flags); 2042 } 2043 2044 static size_t 2045 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list, 2046 struct regset *regset, struct thread *target_td) 2047 { 2048 const struct sysentvec *sv; 2049 struct note_info *ninfo; 2050 size_t size, notesize; 2051 2052 size = 0; 2053 if (!regset->get(regset, target_td, NULL, &size) || size == 0) 2054 return (0); 2055 2056 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2057 ninfo->type = regset->note; 2058 ninfo->regset = regset; 2059 ninfo->outarg = target_td; 2060 ninfo->outsize = size; 2061 TAILQ_INSERT_TAIL(list, ninfo, link); 2062 2063 sv = td->td_proc->p_sysent; 2064 notesize = sizeof(Elf_Note) + /* note header */ 2065 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2066 /* note name */ 2067 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2068 2069 return (notesize); 2070 } 2071 2072 size_t 2073 __elfN(register_note)(struct thread *td, struct note_info_list *list, 2074 int type, outfunc_t out, void *arg) 2075 { 2076 const struct sysentvec *sv; 2077 struct note_info *ninfo; 2078 size_t size, notesize; 2079 2080 sv = td->td_proc->p_sysent; 2081 size = 0; 2082 out(arg, NULL, &size); 2083 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2084 ninfo->type = type; 2085 ninfo->outfunc = out; 2086 ninfo->outarg = arg; 2087 ninfo->outsize = size; 2088 TAILQ_INSERT_TAIL(list, ninfo, link); 2089 2090 if (type == -1) 2091 return (size); 2092 2093 notesize = sizeof(Elf_Note) + /* note header */ 2094 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2095 /* note name */ 2096 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2097 2098 return (notesize); 2099 } 2100 2101 static size_t 2102 append_note_data(const void *src, void *dst, size_t len) 2103 { 2104 size_t padded_len; 2105 2106 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 2107 if (dst != NULL) { 2108 bcopy(src, dst, len); 2109 bzero((char *)dst + len, padded_len - len); 2110 } 2111 return (padded_len); 2112 } 2113 2114 size_t 2115 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 2116 { 2117 Elf_Note *note; 2118 char *buf; 2119 size_t notesize; 2120 2121 buf = dst; 2122 if (buf != NULL) { 2123 note = (Elf_Note *)buf; 2124 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 2125 note->n_descsz = size; 2126 note->n_type = type; 2127 buf += sizeof(*note); 2128 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 2129 sizeof(FREEBSD_ABI_VENDOR)); 2130 append_note_data(src, buf, size); 2131 if (descp != NULL) 2132 *descp = buf; 2133 } 2134 2135 notesize = sizeof(Elf_Note) + /* note header */ 2136 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 2137 /* note name */ 2138 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2139 2140 return (notesize); 2141 } 2142 2143 static void 2144 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb) 2145 { 2146 Elf_Note note; 2147 const struct sysentvec *sv; 2148 ssize_t old_len, sect_len; 2149 size_t new_len, descsz, i; 2150 2151 if (ninfo->type == -1) { 2152 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2153 return; 2154 } 2155 2156 sv = td->td_proc->p_sysent; 2157 2158 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1; 2159 note.n_descsz = ninfo->outsize; 2160 note.n_type = ninfo->type; 2161 2162 sbuf_bcat(sb, ¬e, sizeof(note)); 2163 sbuf_start_section(sb, &old_len); 2164 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor, 2165 strlen(sv->sv_elf_core_abi_vendor) + 1); 2166 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2167 if (note.n_descsz == 0) 2168 return; 2169 sbuf_start_section(sb, &old_len); 2170 if (ninfo->regset != NULL) { 2171 struct regset *regset = ninfo->regset; 2172 void *buf; 2173 2174 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK); 2175 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize); 2176 sbuf_bcat(sb, buf, ninfo->outsize); 2177 free(buf, M_TEMP); 2178 } else 2179 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2180 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2181 if (sect_len < 0) 2182 return; 2183 2184 new_len = (size_t)sect_len; 2185 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 2186 if (new_len < descsz) { 2187 /* 2188 * It is expected that individual note emitters will correctly 2189 * predict their expected output size and fill up to that size 2190 * themselves, padding in a format-specific way if needed. 2191 * However, in case they don't, just do it here with zeros. 2192 */ 2193 for (i = 0; i < descsz - new_len; i++) 2194 sbuf_putc(sb, 0); 2195 } else if (new_len > descsz) { 2196 /* 2197 * We can't always truncate sb -- we may have drained some 2198 * of it already. 2199 */ 2200 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 2201 "read it (%zu > %zu). Since it is longer than " 2202 "expected, this coredump's notes are corrupt. THIS " 2203 "IS A BUG in the note_procstat routine for type %u.\n", 2204 __func__, (unsigned)note.n_type, new_len, descsz, 2205 (unsigned)note.n_type)); 2206 } 2207 } 2208 2209 /* 2210 * Miscellaneous note out functions. 2211 */ 2212 2213 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2214 #include <compat/freebsd32/freebsd32.h> 2215 #include <compat/freebsd32/freebsd32_signal.h> 2216 2217 typedef struct prstatus32 elf_prstatus_t; 2218 typedef struct prpsinfo32 elf_prpsinfo_t; 2219 typedef struct fpreg32 elf_prfpregset_t; 2220 typedef struct fpreg32 elf_fpregset_t; 2221 typedef struct reg32 elf_gregset_t; 2222 typedef struct thrmisc32 elf_thrmisc_t; 2223 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t; 2224 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 2225 typedef struct kinfo_proc32 elf_kinfo_proc_t; 2226 typedef uint32_t elf_ps_strings_t; 2227 #else 2228 typedef prstatus_t elf_prstatus_t; 2229 typedef prpsinfo_t elf_prpsinfo_t; 2230 typedef prfpregset_t elf_prfpregset_t; 2231 typedef prfpregset_t elf_fpregset_t; 2232 typedef gregset_t elf_gregset_t; 2233 typedef thrmisc_t elf_thrmisc_t; 2234 typedef struct ptrace_lwpinfo elf_lwpinfo_t; 2235 #define ELF_KERN_PROC_MASK 0 2236 typedef struct kinfo_proc elf_kinfo_proc_t; 2237 typedef vm_offset_t elf_ps_strings_t; 2238 #endif 2239 2240 static void 2241 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2242 { 2243 struct sbuf sbarg; 2244 size_t len; 2245 char *cp, *end; 2246 struct proc *p; 2247 elf_prpsinfo_t *psinfo; 2248 int error; 2249 2250 p = arg; 2251 if (sb != NULL) { 2252 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 2253 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 2254 psinfo->pr_version = PRPSINFO_VERSION; 2255 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 2256 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 2257 PROC_LOCK(p); 2258 if (p->p_args != NULL) { 2259 len = sizeof(psinfo->pr_psargs) - 1; 2260 if (len > p->p_args->ar_length) 2261 len = p->p_args->ar_length; 2262 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 2263 PROC_UNLOCK(p); 2264 error = 0; 2265 } else { 2266 _PHOLD(p); 2267 PROC_UNLOCK(p); 2268 sbuf_new(&sbarg, psinfo->pr_psargs, 2269 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 2270 error = proc_getargv(curthread, p, &sbarg); 2271 PRELE(p); 2272 if (sbuf_finish(&sbarg) == 0) { 2273 len = sbuf_len(&sbarg); 2274 if (len > 0) 2275 len--; 2276 } else { 2277 len = sizeof(psinfo->pr_psargs) - 1; 2278 } 2279 sbuf_delete(&sbarg); 2280 } 2281 if (error != 0 || len == 0 || (ssize_t)len == -1) 2282 strlcpy(psinfo->pr_psargs, p->p_comm, 2283 sizeof(psinfo->pr_psargs)); 2284 else { 2285 KASSERT(len < sizeof(psinfo->pr_psargs), 2286 ("len is too long: %zu vs %zu", len, 2287 sizeof(psinfo->pr_psargs))); 2288 cp = psinfo->pr_psargs; 2289 end = cp + len - 1; 2290 for (;;) { 2291 cp = memchr(cp, '\0', end - cp); 2292 if (cp == NULL) 2293 break; 2294 *cp = ' '; 2295 } 2296 } 2297 psinfo->pr_pid = p->p_pid; 2298 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 2299 free(psinfo, M_TEMP); 2300 } 2301 *sizep = sizeof(*psinfo); 2302 } 2303 2304 static bool 2305 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf, 2306 size_t *sizep) 2307 { 2308 elf_prstatus_t *status; 2309 2310 if (buf != NULL) { 2311 KASSERT(*sizep == sizeof(*status), ("%s: invalid size", 2312 __func__)); 2313 status = buf; 2314 memset(status, 0, *sizep); 2315 status->pr_version = PRSTATUS_VERSION; 2316 status->pr_statussz = sizeof(elf_prstatus_t); 2317 status->pr_gregsetsz = sizeof(elf_gregset_t); 2318 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 2319 status->pr_osreldate = osreldate; 2320 status->pr_cursig = td->td_proc->p_sig; 2321 status->pr_pid = td->td_tid; 2322 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2323 fill_regs32(td, &status->pr_reg); 2324 #else 2325 fill_regs(td, &status->pr_reg); 2326 #endif 2327 } 2328 *sizep = sizeof(*status); 2329 return (true); 2330 } 2331 2332 static bool 2333 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf, 2334 size_t size) 2335 { 2336 elf_prstatus_t *status; 2337 2338 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__)); 2339 status = buf; 2340 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2341 set_regs32(td, &status->pr_reg); 2342 #else 2343 set_regs(td, &status->pr_reg); 2344 #endif 2345 return (true); 2346 } 2347 2348 static struct regset __elfN(regset_prstatus) = { 2349 .note = NT_PRSTATUS, 2350 .size = sizeof(elf_prstatus_t), 2351 .get = __elfN(get_prstatus), 2352 .set = __elfN(set_prstatus), 2353 }; 2354 ELF_REGSET(__elfN(regset_prstatus)); 2355 2356 static bool 2357 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf, 2358 size_t *sizep) 2359 { 2360 elf_prfpregset_t *fpregset; 2361 2362 if (buf != NULL) { 2363 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size", 2364 __func__)); 2365 fpregset = buf; 2366 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2367 fill_fpregs32(td, fpregset); 2368 #else 2369 fill_fpregs(td, fpregset); 2370 #endif 2371 } 2372 *sizep = sizeof(*fpregset); 2373 return (true); 2374 } 2375 2376 static bool 2377 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf, 2378 size_t size) 2379 { 2380 elf_prfpregset_t *fpregset; 2381 2382 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__)); 2383 2384 fpregset = buf; 2385 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2386 return (set_fpregs32(td, fpregset) == 0); 2387 #else 2388 return (set_fpregs(td, fpregset) == 0); 2389 #endif 2390 } 2391 2392 static struct regset __elfN(regset_fpregset) = { 2393 .note = NT_FPREGSET, 2394 .size = sizeof(elf_prfpregset_t), 2395 .get = __elfN(get_fpregset), 2396 .set = __elfN(set_fpregset), 2397 }; 2398 ELF_REGSET(__elfN(regset_fpregset)); 2399 2400 static bool 2401 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf, 2402 size_t *sizep) 2403 { 2404 elf_thrmisc_t *thrmisc; 2405 2406 if (buf != NULL) { 2407 KASSERT(*sizep == sizeof(*thrmisc), 2408 ("%s: invalid size", __func__)); 2409 thrmisc = buf; 2410 bzero(thrmisc, sizeof(*thrmisc)); 2411 strcpy(thrmisc->pr_tname, td->td_name); 2412 } 2413 *sizep = sizeof(*thrmisc); 2414 return (true); 2415 } 2416 2417 static struct regset __elfN(regset_thrmisc) = { 2418 .note = NT_THRMISC, 2419 .size = sizeof(elf_thrmisc_t), 2420 .get = __elfN(get_thrmisc), 2421 }; 2422 ELF_REGSET(__elfN(regset_thrmisc)); 2423 2424 static bool 2425 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf, 2426 size_t *sizep) 2427 { 2428 elf_lwpinfo_t pl; 2429 size_t size; 2430 int structsize; 2431 2432 size = sizeof(structsize) + sizeof(pl); 2433 if (buf != NULL) { 2434 KASSERT(*sizep == size, ("%s: invalid size", __func__)); 2435 structsize = sizeof(pl); 2436 memcpy(buf, &structsize, sizeof(structsize)); 2437 bzero(&pl, sizeof(pl)); 2438 pl.pl_lwpid = td->td_tid; 2439 pl.pl_event = PL_EVENT_NONE; 2440 pl.pl_sigmask = td->td_sigmask; 2441 pl.pl_siglist = td->td_siglist; 2442 if (td->td_si.si_signo != 0) { 2443 pl.pl_event = PL_EVENT_SIGNAL; 2444 pl.pl_flags |= PL_FLAG_SI; 2445 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2446 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2447 #else 2448 pl.pl_siginfo = td->td_si; 2449 #endif 2450 } 2451 strcpy(pl.pl_tdname, td->td_name); 2452 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2453 memcpy((int *)buf + 1, &pl, sizeof(pl)); 2454 } 2455 *sizep = size; 2456 return (true); 2457 } 2458 2459 static struct regset __elfN(regset_lwpinfo) = { 2460 .note = NT_PTLWPINFO, 2461 .size = sizeof(int) + sizeof(elf_lwpinfo_t), 2462 .get = __elfN(get_lwpinfo), 2463 }; 2464 ELF_REGSET(__elfN(regset_lwpinfo)); 2465 2466 static size_t 2467 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list, 2468 struct thread *target_td) 2469 { 2470 struct sysentvec *sv = td->td_proc->p_sysent; 2471 struct regset **regsetp, **regset_end, *regset; 2472 size_t size; 2473 2474 size = 0; 2475 2476 if (target_td == td) 2477 cpu_update_pcb(target_td); 2478 2479 /* NT_PRSTATUS must be the first register set note. */ 2480 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus), 2481 target_td); 2482 2483 regsetp = sv->sv_regset_begin; 2484 if (regsetp == NULL) { 2485 /* XXX: This shouldn't be true for any FreeBSD ABIs. */ 2486 size += __elfN(register_regset_note)(td, list, 2487 &__elfN(regset_fpregset), target_td); 2488 return (size); 2489 } 2490 regset_end = sv->sv_regset_end; 2491 MPASS(regset_end != NULL); 2492 for (; regsetp < regset_end; regsetp++) { 2493 regset = *regsetp; 2494 if (regset->note == NT_PRSTATUS) 2495 continue; 2496 size += __elfN(register_regset_note)(td, list, regset, 2497 target_td); 2498 } 2499 return (size); 2500 } 2501 2502 /* 2503 * Allow for MD specific notes, as well as any MD 2504 * specific preparations for writing MI notes. 2505 */ 2506 static void 2507 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2508 { 2509 struct thread *td; 2510 void *buf; 2511 size_t size; 2512 2513 td = (struct thread *)arg; 2514 size = *sizep; 2515 if (size != 0 && sb != NULL) 2516 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2517 else 2518 buf = NULL; 2519 size = 0; 2520 __elfN(dump_thread)(td, buf, &size); 2521 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2522 if (size != 0 && sb != NULL) 2523 sbuf_bcat(sb, buf, size); 2524 free(buf, M_TEMP); 2525 *sizep = size; 2526 } 2527 2528 #ifdef KINFO_PROC_SIZE 2529 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2530 #endif 2531 2532 static void 2533 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2534 { 2535 struct proc *p; 2536 size_t size; 2537 int structsize; 2538 2539 p = arg; 2540 size = sizeof(structsize) + p->p_numthreads * 2541 sizeof(elf_kinfo_proc_t); 2542 2543 if (sb != NULL) { 2544 KASSERT(*sizep == size, ("invalid size")); 2545 structsize = sizeof(elf_kinfo_proc_t); 2546 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2547 sx_slock(&proctree_lock); 2548 PROC_LOCK(p); 2549 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2550 sx_sunlock(&proctree_lock); 2551 } 2552 *sizep = size; 2553 } 2554 2555 #ifdef KINFO_FILE_SIZE 2556 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2557 #endif 2558 2559 static void 2560 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2561 { 2562 struct proc *p; 2563 size_t size, sect_sz, i; 2564 ssize_t start_len, sect_len; 2565 int structsize, filedesc_flags; 2566 2567 if (coredump_pack_fileinfo) 2568 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2569 else 2570 filedesc_flags = 0; 2571 2572 p = arg; 2573 structsize = sizeof(struct kinfo_file); 2574 if (sb == NULL) { 2575 size = 0; 2576 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2577 sbuf_set_drain(sb, sbuf_count_drain, &size); 2578 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2579 PROC_LOCK(p); 2580 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2581 sbuf_finish(sb); 2582 sbuf_delete(sb); 2583 *sizep = size; 2584 } else { 2585 sbuf_start_section(sb, &start_len); 2586 2587 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2588 PROC_LOCK(p); 2589 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2590 filedesc_flags); 2591 2592 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2593 if (sect_len < 0) 2594 return; 2595 sect_sz = sect_len; 2596 2597 KASSERT(sect_sz <= *sizep, 2598 ("kern_proc_filedesc_out did not respect maxlen; " 2599 "requested %zu, got %zu", *sizep - sizeof(structsize), 2600 sect_sz - sizeof(structsize))); 2601 2602 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2603 sbuf_putc(sb, 0); 2604 } 2605 } 2606 2607 #ifdef KINFO_VMENTRY_SIZE 2608 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2609 #endif 2610 2611 static void 2612 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2613 { 2614 struct proc *p; 2615 size_t size; 2616 int structsize, vmmap_flags; 2617 2618 if (coredump_pack_vmmapinfo) 2619 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2620 else 2621 vmmap_flags = 0; 2622 2623 p = arg; 2624 structsize = sizeof(struct kinfo_vmentry); 2625 if (sb == NULL) { 2626 size = 0; 2627 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2628 sbuf_set_drain(sb, sbuf_count_drain, &size); 2629 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2630 PROC_LOCK(p); 2631 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2632 sbuf_finish(sb); 2633 sbuf_delete(sb); 2634 *sizep = size; 2635 } else { 2636 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2637 PROC_LOCK(p); 2638 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2639 vmmap_flags); 2640 } 2641 } 2642 2643 static void 2644 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2645 { 2646 struct proc *p; 2647 size_t size; 2648 int structsize; 2649 2650 p = arg; 2651 size = sizeof(structsize) + 2652 (1 + p->p_ucred->cr_ngroups) * sizeof(gid_t); 2653 if (sb != NULL) { 2654 KASSERT(*sizep == size, ("invalid size")); 2655 structsize = sizeof(gid_t); 2656 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2657 sbuf_bcat(sb, &p->p_ucred->cr_gid, sizeof(gid_t)); 2658 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2659 sizeof(gid_t)); 2660 } 2661 *sizep = size; 2662 } 2663 2664 static void 2665 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2666 { 2667 struct proc *p; 2668 size_t size; 2669 int structsize; 2670 2671 p = arg; 2672 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask); 2673 if (sb != NULL) { 2674 KASSERT(*sizep == size, ("invalid size")); 2675 structsize = sizeof(p->p_pd->pd_cmask); 2676 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2677 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask)); 2678 } 2679 *sizep = size; 2680 } 2681 2682 static void 2683 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2684 { 2685 struct proc *p; 2686 struct rlimit rlim[RLIM_NLIMITS]; 2687 size_t size; 2688 int structsize, i; 2689 2690 p = arg; 2691 size = sizeof(structsize) + sizeof(rlim); 2692 if (sb != NULL) { 2693 KASSERT(*sizep == size, ("invalid size")); 2694 structsize = sizeof(rlim); 2695 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2696 PROC_LOCK(p); 2697 for (i = 0; i < RLIM_NLIMITS; i++) 2698 lim_rlimit_proc(p, i, &rlim[i]); 2699 PROC_UNLOCK(p); 2700 sbuf_bcat(sb, rlim, sizeof(rlim)); 2701 } 2702 *sizep = size; 2703 } 2704 2705 static void 2706 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2707 { 2708 struct proc *p; 2709 size_t size; 2710 int structsize; 2711 2712 p = arg; 2713 size = sizeof(structsize) + sizeof(p->p_osrel); 2714 if (sb != NULL) { 2715 KASSERT(*sizep == size, ("invalid size")); 2716 structsize = sizeof(p->p_osrel); 2717 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2718 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2719 } 2720 *sizep = size; 2721 } 2722 2723 static void 2724 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2725 { 2726 struct proc *p; 2727 elf_ps_strings_t ps_strings; 2728 size_t size; 2729 int structsize; 2730 2731 p = arg; 2732 size = sizeof(structsize) + sizeof(ps_strings); 2733 if (sb != NULL) { 2734 KASSERT(*sizep == size, ("invalid size")); 2735 structsize = sizeof(ps_strings); 2736 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2737 ps_strings = PTROUT(PROC_PS_STRINGS(p)); 2738 #else 2739 ps_strings = PROC_PS_STRINGS(p); 2740 #endif 2741 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2742 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2743 } 2744 *sizep = size; 2745 } 2746 2747 static void 2748 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2749 { 2750 struct proc *p; 2751 size_t size; 2752 int structsize; 2753 2754 p = arg; 2755 structsize = sizeof(Elf_Auxinfo); 2756 if (sb == NULL) { 2757 size = 0; 2758 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo), 2759 SBUF_FIXEDLEN); 2760 sbuf_set_drain(sb, sbuf_count_drain, &size); 2761 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2762 PHOLD(p); 2763 proc_getauxv(curthread, p, sb); 2764 PRELE(p); 2765 sbuf_finish(sb); 2766 sbuf_delete(sb); 2767 *sizep = size; 2768 } else { 2769 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2770 PHOLD(p); 2771 proc_getauxv(curthread, p, sb); 2772 PRELE(p); 2773 } 2774 } 2775 2776 static void 2777 __elfN(note_procstat_kqueues)(void *arg, struct sbuf *sb, size_t *sizep) 2778 { 2779 struct proc *p; 2780 size_t size, sect_sz, i; 2781 ssize_t start_len, sect_len; 2782 int structsize; 2783 bool compat32; 2784 2785 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2786 compat32 = true; 2787 structsize = sizeof(struct kinfo_knote32); 2788 #else 2789 compat32 = false; 2790 structsize = sizeof(struct kinfo_knote); 2791 #endif 2792 p = arg; 2793 if (sb == NULL) { 2794 size = 0; 2795 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2796 sbuf_set_drain(sb, sbuf_count_drain, &size); 2797 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2798 kern_proc_kqueues_out(p, sb, -1, compat32); 2799 sbuf_finish(sb); 2800 sbuf_delete(sb); 2801 *sizep = size; 2802 } else { 2803 sbuf_start_section(sb, &start_len); 2804 2805 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2806 kern_proc_kqueues_out(p, sb, *sizep - sizeof(structsize), 2807 compat32); 2808 2809 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2810 if (sect_len < 0) 2811 return; 2812 sect_sz = sect_len; 2813 2814 KASSERT(sect_sz <= *sizep, 2815 ("kern_proc_kqueue_out did not respect maxlen; " 2816 "requested %zu, got %zu", *sizep - sizeof(structsize), 2817 sect_sz - sizeof(structsize))); 2818 2819 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2820 sbuf_putc(sb, 0); 2821 } 2822 } 2823 2824 #define MAX_NOTES_LOOP 4096 2825 bool 2826 __elfN(parse_notes)(const struct image_params *imgp, const Elf_Note *checknote, 2827 const char *note_vendor, const Elf_Phdr *pnote, 2828 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg) 2829 { 2830 const Elf_Note *note, *note0, *note_end; 2831 const char *note_name; 2832 char *buf; 2833 int i, error; 2834 bool res; 2835 2836 /* We need some limit, might as well use PAGE_SIZE. */ 2837 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2838 return (false); 2839 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2840 if (pnote->p_offset > PAGE_SIZE || 2841 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2842 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT); 2843 if (buf == NULL) { 2844 VOP_UNLOCK(imgp->vp); 2845 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2846 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 2847 } 2848 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2849 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2850 imgp->td->td_ucred, NOCRED, NULL, imgp->td); 2851 if (error != 0) { 2852 uprintf("i/o error PT_NOTE\n"); 2853 goto retf; 2854 } 2855 note = note0 = (const Elf_Note *)buf; 2856 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2857 } else { 2858 note = note0 = (const Elf_Note *)(imgp->image_header + 2859 pnote->p_offset); 2860 note_end = (const Elf_Note *)(imgp->image_header + 2861 pnote->p_offset + pnote->p_filesz); 2862 buf = NULL; 2863 } 2864 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end; 2865 i++) { 2866 if (!aligned(note, Elf32_Addr)) { 2867 uprintf("Unaligned ELF note\n"); 2868 goto retf; 2869 } 2870 if ((const char *)note_end - (const char *)note < 2871 sizeof(Elf_Note)) { 2872 uprintf("ELF note too short\n"); 2873 goto retf; 2874 } 2875 if (note->n_namesz != checknote->n_namesz || 2876 note->n_descsz != checknote->n_descsz || 2877 note->n_type != checknote->n_type) 2878 goto nextnote; 2879 note_name = (const char *)(note + 1); 2880 if (note_name + roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2881 note->n_descsz > (const char *)note_end || 2882 strncmp(note_vendor, note_name, checknote->n_namesz) != 0) 2883 goto nextnote; 2884 2885 if (cb(note, cb_arg, &res)) 2886 goto ret; 2887 nextnote: 2888 note = (const Elf_Note *)((const char *)(note + 1) + 2889 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2890 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2891 } 2892 if (i >= MAX_NOTES_LOOP) 2893 uprintf("ELF note parser reached %d notes\n", i); 2894 retf: 2895 res = false; 2896 ret: 2897 free(buf, M_TEMP); 2898 return (res); 2899 } 2900 2901 struct brandnote_cb_arg { 2902 const Elf_Brandnote *brandnote; 2903 int32_t *osrel; 2904 }; 2905 2906 static bool 2907 brandnote_cb(const Elf_Note *note, void *arg0, bool *res) 2908 { 2909 struct brandnote_cb_arg *arg; 2910 2911 arg = arg0; 2912 2913 /* 2914 * Fetch the osreldate for binary from the ELF OSABI-note if 2915 * necessary. 2916 */ 2917 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 && 2918 arg->brandnote->trans_osrel != NULL ? 2919 arg->brandnote->trans_osrel(note, arg->osrel) : true; 2920 2921 return (true); 2922 } 2923 2924 static const Elf_Note fctl_note = { 2925 .n_namesz = sizeof(FREEBSD_ABI_VENDOR), 2926 .n_descsz = sizeof(uint32_t), 2927 .n_type = NT_FREEBSD_FEATURE_CTL, 2928 }; 2929 2930 struct fctl_cb_arg { 2931 bool *has_fctl0; 2932 uint32_t *fctl0; 2933 }; 2934 2935 static bool 2936 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res) 2937 { 2938 struct fctl_cb_arg *arg; 2939 const Elf32_Word *desc; 2940 uintptr_t p; 2941 2942 arg = arg0; 2943 p = (uintptr_t)(note + 1); 2944 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 2945 desc = (const Elf32_Word *)p; 2946 *arg->has_fctl0 = true; 2947 *arg->fctl0 = desc[0]; 2948 *res = true; 2949 return (true); 2950 } 2951 2952 /* 2953 * Try to find the appropriate ABI-note section for checknote, fetch 2954 * the osreldate and feature control flags for binary from the ELF 2955 * OSABI-note. Only the first page of the image is searched, the same 2956 * as for headers. 2957 */ 2958 static bool 2959 __elfN(check_note)(struct image_params *imgp, const Elf_Phdr *phdr, 2960 const Elf_Brandnote *brandnote, int32_t *osrel, bool *has_fctl0, 2961 uint32_t *fctl0) 2962 { 2963 const Elf_Ehdr *hdr; 2964 struct brandnote_cb_arg b_arg; 2965 struct fctl_cb_arg f_arg; 2966 int i, j; 2967 2968 hdr = (const Elf_Ehdr *)imgp->image_header; 2969 b_arg.brandnote = brandnote; 2970 b_arg.osrel = osrel; 2971 f_arg.has_fctl0 = has_fctl0; 2972 f_arg.fctl0 = fctl0; 2973 2974 for (i = 0; i < hdr->e_phnum; i++) { 2975 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, 2976 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb, 2977 &b_arg)) { 2978 for (j = 0; j < hdr->e_phnum; j++) { 2979 if (phdr[j].p_type == PT_NOTE && 2980 __elfN(parse_notes)(imgp, &fctl_note, 2981 FREEBSD_ABI_VENDOR, &phdr[j], 2982 note_fctl_cb, &f_arg)) 2983 break; 2984 } 2985 return (true); 2986 } 2987 } 2988 return (false); 2989 2990 } 2991 2992 /* 2993 * Tell kern_execve.c about it, with a little help from the linker. 2994 */ 2995 static struct execsw __elfN(execsw) = { 2996 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 2997 .ex_name = ELF_ABI_NAME 2998 }; 2999 EXEC_SET(ELF_ABI_ID, __elfN(execsw)); 3000 3001 static vm_prot_t 3002 __elfN(trans_prot)(Elf_Word flags) 3003 { 3004 vm_prot_t prot; 3005 3006 prot = 0; 3007 if (flags & PF_X) 3008 prot |= VM_PROT_EXECUTE; 3009 if (flags & PF_W) 3010 prot |= VM_PROT_WRITE; 3011 if (flags & PF_R) 3012 prot |= VM_PROT_READ; 3013 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 3014 if (i386_read_exec && (flags & PF_R)) 3015 prot |= VM_PROT_EXECUTE; 3016 #endif 3017 return (prot); 3018 } 3019 3020 static Elf_Word 3021 __elfN(untrans_prot)(vm_prot_t prot) 3022 { 3023 Elf_Word flags; 3024 3025 flags = 0; 3026 if (prot & VM_PROT_EXECUTE) 3027 flags |= PF_X; 3028 if (prot & VM_PROT_READ) 3029 flags |= PF_R; 3030 if (prot & VM_PROT_WRITE) 3031 flags |= PF_W; 3032 return (flags); 3033 } 3034