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 static int __elfN(check_header)(const Elf_Ehdr *hdr); 88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 89 const char *interp, int32_t *osrel, uint32_t *fctl0); 90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 91 u_long *entry); 92 static int __elfN(load_section)(const struct image_params *imgp, 93 vm_ooffset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 94 vm_prot_t prot); 95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 96 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note, 97 int32_t *osrel); 98 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 99 static bool __elfN(check_note)(struct image_params *imgp, 100 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0, 101 uint32_t *fctl0); 102 static vm_prot_t __elfN(trans_prot)(Elf_Word); 103 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 104 static size_t __elfN(prepare_register_notes)(struct thread *td, 105 struct note_info_list *list, struct thread *target_td); 106 107 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), 108 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 109 ""); 110 111 int __elfN(fallback_brand) = -1; 112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 113 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, 114 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 115 116 static int elf_legacy_coredump = 0; 117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 118 &elf_legacy_coredump, 0, 119 "include all and only RW pages in core dumps"); 120 121 int __elfN(nxstack) = 122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \ 123 defined(__arm__) || defined(__aarch64__) || \ 124 defined(__riscv) 125 1; 126 #else 127 0; 128 #endif 129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 130 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 131 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": support PT_GNU_STACK for non-executable stack control"); 132 133 #if defined(__amd64__) 134 static int __elfN(vdso) = 1; 135 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 136 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0, 137 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading"); 138 #else 139 static int __elfN(vdso) = 0; 140 #endif 141 142 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 143 int i386_read_exec = 0; 144 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 145 "enable execution from readable segments"); 146 #endif 147 148 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR; 149 static int 150 sysctl_pie_base(SYSCTL_HANDLER_ARGS) 151 { 152 u_long val; 153 int error; 154 155 val = __elfN(pie_base); 156 error = sysctl_handle_long(oidp, &val, 0, req); 157 if (error != 0 || req->newptr == NULL) 158 return (error); 159 if ((val & PAGE_MASK) != 0) 160 return (EINVAL); 161 __elfN(pie_base) = val; 162 return (0); 163 } 164 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base, 165 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 166 sysctl_pie_base, "LU", 167 "PIE load base without randomization"); 168 169 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, 170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 171 ""); 172 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr) 173 174 /* 175 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures 176 * have limited address space (which can cause issues for applications with 177 * high memory use) so we leave it off there. 178 */ 179 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64; 180 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN, 181 &__elfN(aslr_enabled), 0, 182 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 183 ": enable address map randomization"); 184 185 /* 186 * Enable ASLR by default for 64-bit PIE binaries. 187 */ 188 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64; 189 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN, 190 &__elfN(pie_aslr_enabled), 0, 191 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 192 ": enable address map randomization for PIE binaries"); 193 194 /* 195 * Sbrk is deprecated and it can be assumed that in most cases it will not be 196 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR 197 * to use the bss grow region. 198 */ 199 static int __elfN(aslr_honor_sbrk) = 0; 200 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW, 201 &__elfN(aslr_honor_sbrk), 0, 202 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used"); 203 204 static int __elfN(aslr_stack) = __ELF_WORD_SIZE == 64; 205 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN, 206 &__elfN(aslr_stack), 0, 207 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 208 ": enable stack address randomization"); 209 210 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64; 211 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN, 212 &__elfN(aslr_shared_page), 0, 213 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 214 ": enable shared page address randomization"); 215 216 static int __elfN(sigfastblock) = 1; 217 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock, 218 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0, 219 "enable sigfastblock for new processes"); 220 221 static bool __elfN(allow_wx) = true; 222 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx, 223 CTLFLAG_RWTUN, &__elfN(allow_wx), 0, 224 "Allow pages to be mapped simultaneously writable and executable"); 225 226 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 227 228 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a)) 229 230 Elf_Brandnote __elfN(freebsd_brandnote) = { 231 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 232 .hdr.n_descsz = sizeof(int32_t), 233 .hdr.n_type = NT_FREEBSD_ABI_TAG, 234 .vendor = FREEBSD_ABI_VENDOR, 235 .flags = BN_TRANSLATE_OSREL, 236 .trans_osrel = __elfN(freebsd_trans_osrel) 237 }; 238 239 static bool 240 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 241 { 242 uintptr_t p; 243 244 p = (uintptr_t)(note + 1); 245 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 246 *osrel = *(const int32_t *)(p); 247 248 return (true); 249 } 250 251 static int GNU_KFREEBSD_ABI_DESC = 3; 252 253 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 254 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 255 .hdr.n_descsz = 16, /* XXX at least 16 */ 256 .hdr.n_type = 1, 257 .vendor = GNU_ABI_VENDOR, 258 .flags = BN_TRANSLATE_OSREL, 259 .trans_osrel = kfreebsd_trans_osrel 260 }; 261 262 static bool 263 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 264 { 265 const Elf32_Word *desc; 266 uintptr_t p; 267 268 p = (uintptr_t)(note + 1); 269 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 270 271 desc = (const Elf32_Word *)p; 272 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 273 return (false); 274 275 /* 276 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 277 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 278 */ 279 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 280 281 return (true); 282 } 283 284 int 285 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 286 { 287 int i; 288 289 for (i = 0; i < MAX_BRANDS; i++) { 290 if (elf_brand_list[i] == NULL) { 291 elf_brand_list[i] = entry; 292 break; 293 } 294 } 295 if (i == MAX_BRANDS) { 296 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 297 __func__, entry); 298 return (-1); 299 } 300 return (0); 301 } 302 303 int 304 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 305 { 306 int i; 307 308 for (i = 0; i < MAX_BRANDS; i++) { 309 if (elf_brand_list[i] == entry) { 310 elf_brand_list[i] = NULL; 311 break; 312 } 313 } 314 if (i == MAX_BRANDS) 315 return (-1); 316 return (0); 317 } 318 319 bool 320 __elfN(brand_inuse)(Elf_Brandinfo *entry) 321 { 322 struct proc *p; 323 bool rval = false; 324 325 sx_slock(&allproc_lock); 326 FOREACH_PROC_IN_SYSTEM(p) { 327 if (p->p_sysent == entry->sysvec) { 328 rval = true; 329 break; 330 } 331 } 332 sx_sunlock(&allproc_lock); 333 334 return (rval); 335 } 336 337 static Elf_Brandinfo * 338 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 339 int32_t *osrel, uint32_t *fctl0) 340 { 341 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 342 Elf_Brandinfo *bi, *bi_m; 343 bool ret, has_fctl0; 344 int i, interp_name_len; 345 346 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0; 347 348 /* 349 * We support four types of branding -- (1) the ELF EI_OSABI field 350 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 351 * branding w/in the ELF header, (3) path of the `interp_path' 352 * field, and (4) the ".note.ABI-tag" ELF section. 353 */ 354 355 /* Look for an ".note.ABI-tag" ELF section */ 356 bi_m = NULL; 357 for (i = 0; i < MAX_BRANDS; i++) { 358 bi = elf_brand_list[i]; 359 if (bi == NULL) 360 continue; 361 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 362 continue; 363 if (hdr->e_machine == bi->machine && (bi->flags & 364 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 365 has_fctl0 = false; 366 *fctl0 = 0; 367 *osrel = 0; 368 ret = __elfN(check_note)(imgp, bi->brand_note, osrel, 369 &has_fctl0, fctl0); 370 /* Give brand a chance to veto check_note's guess */ 371 if (ret && bi->header_supported) { 372 ret = bi->header_supported(imgp, osrel, 373 has_fctl0 ? fctl0 : NULL); 374 } 375 /* 376 * If note checker claimed the binary, but the 377 * interpreter path in the image does not 378 * match default one for the brand, try to 379 * search for other brands with the same 380 * interpreter. Either there is better brand 381 * with the right interpreter, or, failing 382 * this, we return first brand which accepted 383 * our note and, optionally, header. 384 */ 385 if (ret && bi_m == NULL && interp != NULL && 386 (bi->interp_path == NULL || 387 (strlen(bi->interp_path) + 1 != interp_name_len || 388 strncmp(interp, bi->interp_path, interp_name_len) 389 != 0))) { 390 bi_m = bi; 391 ret = 0; 392 } 393 if (ret) 394 return (bi); 395 } 396 } 397 if (bi_m != NULL) 398 return (bi_m); 399 400 /* If the executable has a brand, search for it in the brand list. */ 401 for (i = 0; i < MAX_BRANDS; i++) { 402 bi = elf_brand_list[i]; 403 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 404 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 405 continue; 406 if (hdr->e_machine == bi->machine && 407 (hdr->e_ident[EI_OSABI] == bi->brand || 408 (bi->compat_3_brand != NULL && 409 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 410 bi->compat_3_brand) == 0))) { 411 /* Looks good, but give brand a chance to veto */ 412 if (bi->header_supported == NULL || 413 bi->header_supported(imgp, NULL, NULL)) { 414 /* 415 * Again, prefer strictly matching 416 * interpreter path. 417 */ 418 if (interp_name_len == 0 && 419 bi->interp_path == NULL) 420 return (bi); 421 if (bi->interp_path != NULL && 422 strlen(bi->interp_path) + 1 == 423 interp_name_len && strncmp(interp, 424 bi->interp_path, interp_name_len) == 0) 425 return (bi); 426 if (bi_m == NULL) 427 bi_m = bi; 428 } 429 } 430 } 431 if (bi_m != NULL) 432 return (bi_m); 433 434 /* No known brand, see if the header is recognized by any brand */ 435 for (i = 0; i < MAX_BRANDS; i++) { 436 bi = elf_brand_list[i]; 437 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 438 bi->header_supported == NULL) 439 continue; 440 if (hdr->e_machine == bi->machine) { 441 ret = bi->header_supported(imgp, NULL, NULL); 442 if (ret) 443 return (bi); 444 } 445 } 446 447 /* Lacking a known brand, search for a recognized interpreter. */ 448 if (interp != NULL) { 449 for (i = 0; i < MAX_BRANDS; i++) { 450 bi = elf_brand_list[i]; 451 if (bi == NULL || (bi->flags & 452 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 453 != 0) 454 continue; 455 if (hdr->e_machine == bi->machine && 456 bi->interp_path != NULL && 457 /* ELF image p_filesz includes terminating zero */ 458 strlen(bi->interp_path) + 1 == interp_name_len && 459 strncmp(interp, bi->interp_path, interp_name_len) 460 == 0 && (bi->header_supported == NULL || 461 bi->header_supported(imgp, NULL, NULL))) 462 return (bi); 463 } 464 } 465 466 /* Lacking a recognized interpreter, try the default brand */ 467 for (i = 0; i < MAX_BRANDS; i++) { 468 bi = elf_brand_list[i]; 469 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 470 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 471 continue; 472 if (hdr->e_machine == bi->machine && 473 __elfN(fallback_brand) == bi->brand && 474 (bi->header_supported == NULL || 475 bi->header_supported(imgp, NULL, NULL))) 476 return (bi); 477 } 478 return (NULL); 479 } 480 481 static bool 482 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr) 483 { 484 return (hdr->e_phoff <= PAGE_SIZE && 485 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff); 486 } 487 488 static int 489 __elfN(check_header)(const Elf_Ehdr *hdr) 490 { 491 Elf_Brandinfo *bi; 492 int i; 493 494 if (!IS_ELF(*hdr) || 495 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 496 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 497 hdr->e_ident[EI_VERSION] != EV_CURRENT || 498 hdr->e_phentsize != sizeof(Elf_Phdr) || 499 hdr->e_version != ELF_TARG_VER) 500 return (ENOEXEC); 501 502 /* 503 * Make sure we have at least one brand for this machine. 504 */ 505 506 for (i = 0; i < MAX_BRANDS; i++) { 507 bi = elf_brand_list[i]; 508 if (bi != NULL && bi->machine == hdr->e_machine) 509 break; 510 } 511 if (i == MAX_BRANDS) 512 return (ENOEXEC); 513 514 return (0); 515 } 516 517 static int 518 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 519 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 520 { 521 struct sf_buf *sf; 522 int error; 523 vm_offset_t off; 524 525 /* 526 * Create the page if it doesn't exist yet. Ignore errors. 527 */ 528 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 529 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 530 531 /* 532 * Find the page from the underlying object. 533 */ 534 if (object != NULL) { 535 sf = vm_imgact_map_page(object, offset); 536 if (sf == NULL) 537 return (KERN_FAILURE); 538 off = offset - trunc_page(offset); 539 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 540 end - start); 541 vm_imgact_unmap_page(sf); 542 if (error != 0) 543 return (KERN_FAILURE); 544 } 545 546 return (KERN_SUCCESS); 547 } 548 549 static int 550 __elfN(map_insert)(const struct image_params *imgp, vm_map_t map, 551 vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, 552 vm_prot_t prot, int cow) 553 { 554 struct sf_buf *sf; 555 vm_offset_t off; 556 vm_size_t sz; 557 int error, locked, rv; 558 559 if (start != trunc_page(start)) { 560 rv = __elfN(map_partial)(map, object, offset, start, 561 round_page(start), prot); 562 if (rv != KERN_SUCCESS) 563 return (rv); 564 offset += round_page(start) - start; 565 start = round_page(start); 566 } 567 if (end != round_page(end)) { 568 rv = __elfN(map_partial)(map, object, offset + 569 trunc_page(end) - start, trunc_page(end), end, prot); 570 if (rv != KERN_SUCCESS) 571 return (rv); 572 end = trunc_page(end); 573 } 574 if (start >= end) 575 return (KERN_SUCCESS); 576 if ((offset & PAGE_MASK) != 0) { 577 /* 578 * The mapping is not page aligned. This means that we have 579 * to copy the data. 580 */ 581 rv = vm_map_fixed(map, NULL, 0, start, end - start, 582 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 583 if (rv != KERN_SUCCESS) 584 return (rv); 585 if (object == NULL) 586 return (KERN_SUCCESS); 587 for (; start < end; start += sz) { 588 sf = vm_imgact_map_page(object, offset); 589 if (sf == NULL) 590 return (KERN_FAILURE); 591 off = offset - trunc_page(offset); 592 sz = end - start; 593 if (sz > PAGE_SIZE - off) 594 sz = PAGE_SIZE - off; 595 error = copyout((caddr_t)sf_buf_kva(sf) + off, 596 (caddr_t)start, sz); 597 vm_imgact_unmap_page(sf); 598 if (error != 0) 599 return (KERN_FAILURE); 600 offset += sz; 601 } 602 } else { 603 vm_object_reference(object); 604 rv = vm_map_fixed(map, object, offset, start, end - start, 605 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL | 606 (object != NULL ? MAP_VN_EXEC : 0)); 607 if (rv != KERN_SUCCESS) { 608 locked = VOP_ISLOCKED(imgp->vp); 609 VOP_UNLOCK(imgp->vp); 610 vm_object_deallocate(object); 611 vn_lock(imgp->vp, locked | LK_RETRY); 612 return (rv); 613 } else if (object != NULL) { 614 MPASS(imgp->vp->v_object == object); 615 VOP_SET_TEXT_CHECKED(imgp->vp); 616 } 617 } 618 return (KERN_SUCCESS); 619 } 620 621 static int 622 __elfN(load_section)(const struct image_params *imgp, vm_ooffset_t offset, 623 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot) 624 { 625 struct sf_buf *sf; 626 size_t map_len; 627 vm_map_t map; 628 vm_object_t object; 629 vm_offset_t map_addr; 630 int error, rv, cow; 631 size_t copy_len; 632 vm_ooffset_t file_addr; 633 634 /* 635 * It's necessary to fail if the filsz + offset taken from the 636 * header is greater than the actual file pager object's size. 637 * If we were to allow this, then the vm_map_find() below would 638 * walk right off the end of the file object and into the ether. 639 * 640 * While I'm here, might as well check for something else that 641 * is invalid: filsz cannot be greater than memsz. 642 */ 643 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 644 filsz > memsz) { 645 uprintf("elf_load_section: truncated ELF file\n"); 646 return (ENOEXEC); 647 } 648 649 object = imgp->object; 650 map = &imgp->proc->p_vmspace->vm_map; 651 map_addr = trunc_page((vm_offset_t)vmaddr); 652 file_addr = trunc_page(offset); 653 654 /* 655 * We have two choices. We can either clear the data in the last page 656 * of an oversized mapping, or we can start the anon mapping a page 657 * early and copy the initialized data into that first page. We 658 * choose the second. 659 */ 660 if (filsz == 0) 661 map_len = 0; 662 else if (memsz > filsz) 663 map_len = trunc_page(offset + filsz) - file_addr; 664 else 665 map_len = round_page(offset + filsz) - file_addr; 666 667 if (map_len != 0) { 668 /* cow flags: don't dump readonly sections in core */ 669 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 670 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 671 672 rv = __elfN(map_insert)(imgp, map, object, file_addr, 673 map_addr, map_addr + map_len, prot, cow); 674 if (rv != KERN_SUCCESS) 675 return (EINVAL); 676 677 /* we can stop now if we've covered it all */ 678 if (memsz == filsz) 679 return (0); 680 } 681 682 /* 683 * We have to get the remaining bit of the file into the first part 684 * of the oversized map segment. This is normally because the .data 685 * segment in the file is extended to provide bss. It's a neat idea 686 * to try and save a page, but it's a pain in the behind to implement. 687 */ 688 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset + 689 filsz); 690 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 691 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 692 693 /* This had damn well better be true! */ 694 if (map_len != 0) { 695 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 696 map_addr + map_len, prot, 0); 697 if (rv != KERN_SUCCESS) 698 return (EINVAL); 699 } 700 701 if (copy_len != 0) { 702 sf = vm_imgact_map_page(object, offset + filsz); 703 if (sf == NULL) 704 return (EIO); 705 706 /* send the page fragment to user space */ 707 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr, 708 copy_len); 709 vm_imgact_unmap_page(sf); 710 if (error != 0) 711 return (error); 712 } 713 714 /* 715 * Remove write access to the page if it was only granted by map_insert 716 * to allow copyout. 717 */ 718 if ((prot & VM_PROT_WRITE) == 0) 719 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 720 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT); 721 722 return (0); 723 } 724 725 static int 726 __elfN(load_sections)(const struct image_params *imgp, const Elf_Ehdr *hdr, 727 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp) 728 { 729 vm_prot_t prot; 730 u_long base_addr; 731 bool first; 732 int error, i; 733 734 ASSERT_VOP_LOCKED(imgp->vp, __func__); 735 736 base_addr = 0; 737 first = true; 738 739 for (i = 0; i < hdr->e_phnum; i++) { 740 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 741 continue; 742 743 /* Loadable segment */ 744 prot = __elfN(trans_prot)(phdr[i].p_flags); 745 error = __elfN(load_section)(imgp, phdr[i].p_offset, 746 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 747 phdr[i].p_memsz, phdr[i].p_filesz, prot); 748 if (error != 0) 749 return (error); 750 751 /* 752 * Establish the base address if this is the first segment. 753 */ 754 if (first) { 755 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 756 first = false; 757 } 758 } 759 760 if (base_addrp != NULL) 761 *base_addrp = base_addr; 762 763 return (0); 764 } 765 766 /* 767 * Load the file "file" into memory. It may be either a shared object 768 * or an executable. 769 * 770 * The "addr" reference parameter is in/out. On entry, it specifies 771 * the address where a shared object should be loaded. If the file is 772 * an executable, this value is ignored. On exit, "addr" specifies 773 * where the file was actually loaded. 774 * 775 * The "entry" reference parameter is out only. On exit, it specifies 776 * the entry point for the loaded file. 777 */ 778 static int 779 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 780 u_long *entry) 781 { 782 struct { 783 struct nameidata nd; 784 struct vattr attr; 785 struct image_params image_params; 786 } *tempdata; 787 const Elf_Ehdr *hdr = NULL; 788 const Elf_Phdr *phdr = NULL; 789 struct nameidata *nd; 790 struct vattr *attr; 791 struct image_params *imgp; 792 u_long rbase; 793 u_long base_addr = 0; 794 int error; 795 796 #ifdef CAPABILITY_MODE 797 /* 798 * XXXJA: This check can go away once we are sufficiently confident 799 * that the checks in namei() are correct. 800 */ 801 if (IN_CAPABILITY_MODE(curthread)) 802 return (ECAPMODE); 803 #endif 804 805 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO); 806 nd = &tempdata->nd; 807 attr = &tempdata->attr; 808 imgp = &tempdata->image_params; 809 810 /* 811 * Initialize part of the common data 812 */ 813 imgp->proc = p; 814 imgp->attr = attr; 815 816 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF, 817 UIO_SYSSPACE, file); 818 if ((error = namei(nd)) != 0) { 819 nd->ni_vp = NULL; 820 goto fail; 821 } 822 NDFREE_PNBUF(nd); 823 imgp->vp = nd->ni_vp; 824 825 /* 826 * Check permissions, modes, uid, etc on the file, and "open" it. 827 */ 828 error = exec_check_permissions(imgp); 829 if (error) 830 goto fail; 831 832 error = exec_map_first_page(imgp); 833 if (error) 834 goto fail; 835 836 imgp->object = nd->ni_vp->v_object; 837 838 hdr = (const Elf_Ehdr *)imgp->image_header; 839 if ((error = __elfN(check_header)(hdr)) != 0) 840 goto fail; 841 if (hdr->e_type == ET_DYN) 842 rbase = *addr; 843 else if (hdr->e_type == ET_EXEC) 844 rbase = 0; 845 else { 846 error = ENOEXEC; 847 goto fail; 848 } 849 850 /* Only support headers that fit within first page for now */ 851 if (!__elfN(phdr_in_zero_page)(hdr)) { 852 error = ENOEXEC; 853 goto fail; 854 } 855 856 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 857 if (!aligned(phdr, Elf_Addr)) { 858 error = ENOEXEC; 859 goto fail; 860 } 861 862 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr); 863 if (error != 0) 864 goto fail; 865 866 if (p->p_sysent->sv_protect != NULL) 867 p->p_sysent->sv_protect(imgp, SVP_INTERP); 868 869 *addr = base_addr; 870 *entry = (unsigned long)hdr->e_entry + rbase; 871 872 fail: 873 if (imgp->firstpage) 874 exec_unmap_first_page(imgp); 875 876 if (nd->ni_vp) { 877 if (imgp->textset) 878 VOP_UNSET_TEXT_CHECKED(nd->ni_vp); 879 vput(nd->ni_vp); 880 } 881 free(tempdata, M_TEMP); 882 883 return (error); 884 } 885 886 /* 887 * Select randomized valid address in the map map, between minv and 888 * maxv, with specified alignment. The [minv, maxv) range must belong 889 * to the map. Note that function only allocates the address, it is 890 * up to caller to clamp maxv in a way that the final allocation 891 * length fit into the map. 892 * 893 * Result is returned in *resp, error code indicates that arguments 894 * did not pass sanity checks for overflow and range correctness. 895 */ 896 static int 897 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv, 898 u_int align, u_long *resp) 899 { 900 u_long rbase, res; 901 902 MPASS(vm_map_min(map) <= minv); 903 904 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) { 905 uprintf("Invalid ELF segments layout\n"); 906 return (ENOEXEC); 907 } 908 909 arc4rand(&rbase, sizeof(rbase), 0); 910 res = roundup(minv, (u_long)align) + rbase % (maxv - minv); 911 res &= ~((u_long)align - 1); 912 if (res >= maxv) 913 res -= align; 914 915 KASSERT(res >= minv, 916 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx", 917 res, minv, maxv, rbase)); 918 KASSERT(res < maxv, 919 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx", 920 res, maxv, minv, rbase)); 921 922 *resp = res; 923 return (0); 924 } 925 926 static int 927 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr, 928 const Elf_Phdr *phdr) 929 { 930 struct vmspace *vmspace; 931 const char *err_str; 932 u_long text_size, data_size, total_size, text_addr, data_addr; 933 u_long seg_size, seg_addr; 934 int i; 935 936 err_str = NULL; 937 text_size = data_size = total_size = text_addr = data_addr = 0; 938 939 for (i = 0; i < hdr->e_phnum; i++) { 940 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 941 continue; 942 943 seg_addr = trunc_page(phdr[i].p_vaddr + imgp->et_dyn_addr); 944 seg_size = round_page(phdr[i].p_memsz + 945 phdr[i].p_vaddr + imgp->et_dyn_addr - seg_addr); 946 947 /* 948 * Make the largest executable segment the official 949 * text segment and all others data. 950 * 951 * Note that obreak() assumes that data_addr + data_size == end 952 * of data load area, and the ELF file format expects segments 953 * to be sorted by address. If multiple data segments exist, 954 * the last one will be used. 955 */ 956 957 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) { 958 text_size = seg_size; 959 text_addr = seg_addr; 960 } else { 961 data_size = seg_size; 962 data_addr = seg_addr; 963 } 964 total_size += seg_size; 965 } 966 967 if (data_addr == 0 && data_size == 0) { 968 data_addr = text_addr; 969 data_size = text_size; 970 } 971 972 /* 973 * Check limits. It should be safe to check the 974 * limits after loading the segments since we do 975 * not actually fault in all the segments pages. 976 */ 977 PROC_LOCK(imgp->proc); 978 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 979 err_str = "Data segment size exceeds process limit"; 980 else if (text_size > maxtsiz) 981 err_str = "Text segment size exceeds system limit"; 982 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 983 err_str = "Total segment size exceeds process limit"; 984 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 985 err_str = "Data segment size exceeds resource limit"; 986 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 987 err_str = "Total segment size exceeds resource limit"; 988 PROC_UNLOCK(imgp->proc); 989 if (err_str != NULL) { 990 uprintf("%s\n", err_str); 991 return (ENOMEM); 992 } 993 994 vmspace = imgp->proc->p_vmspace; 995 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 996 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 997 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 998 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 999 1000 return (0); 1001 } 1002 1003 static int 1004 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr, 1005 char **interpp, bool *free_interpp) 1006 { 1007 struct thread *td; 1008 char *interp; 1009 int error, interp_name_len; 1010 1011 KASSERT(phdr->p_type == PT_INTERP, 1012 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type)); 1013 ASSERT_VOP_LOCKED(imgp->vp, __func__); 1014 1015 td = curthread; 1016 1017 /* Path to interpreter */ 1018 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) { 1019 uprintf("Invalid PT_INTERP\n"); 1020 return (ENOEXEC); 1021 } 1022 1023 interp_name_len = phdr->p_filesz; 1024 if (phdr->p_offset > PAGE_SIZE || 1025 interp_name_len > PAGE_SIZE - phdr->p_offset) { 1026 /* 1027 * The vnode lock might be needed by the pagedaemon to 1028 * clean pages owned by the vnode. Do not allow sleep 1029 * waiting for memory with the vnode locked, instead 1030 * try non-sleepable allocation first, and if it 1031 * fails, go to the slow path were we drop the lock 1032 * and do M_WAITOK. A text reference prevents 1033 * modifications to the vnode content. 1034 */ 1035 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT); 1036 if (interp == NULL) { 1037 VOP_UNLOCK(imgp->vp); 1038 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK); 1039 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1040 } 1041 1042 error = vn_rdwr(UIO_READ, imgp->vp, interp, 1043 interp_name_len, phdr->p_offset, 1044 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 1045 NOCRED, NULL, td); 1046 if (error != 0) { 1047 free(interp, M_TEMP); 1048 uprintf("i/o error PT_INTERP %d\n", error); 1049 return (error); 1050 } 1051 interp[interp_name_len] = '\0'; 1052 1053 *interpp = interp; 1054 *free_interpp = true; 1055 return (0); 1056 } 1057 1058 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset; 1059 if (interp[interp_name_len - 1] != '\0') { 1060 uprintf("Invalid PT_INTERP\n"); 1061 return (ENOEXEC); 1062 } 1063 1064 *interpp = interp; 1065 *free_interpp = false; 1066 return (0); 1067 } 1068 1069 static int 1070 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info, 1071 const char *interp, u_long *addr, u_long *entry) 1072 { 1073 int error; 1074 1075 if (brand_info->interp_newpath != NULL && 1076 (brand_info->interp_path == NULL || 1077 strcmp(interp, brand_info->interp_path) == 0)) { 1078 error = __elfN(load_file)(imgp->proc, 1079 brand_info->interp_newpath, addr, entry); 1080 if (error == 0) 1081 return (0); 1082 } 1083 1084 error = __elfN(load_file)(imgp->proc, interp, addr, entry); 1085 if (error == 0) 1086 return (0); 1087 1088 uprintf("ELF interpreter %s not found, error %d\n", interp, error); 1089 return (error); 1090 } 1091 1092 /* 1093 * Impossible et_dyn_addr initial value indicating that the real base 1094 * must be calculated later with some randomization applied. 1095 */ 1096 #define ET_DYN_ADDR_RAND 1 1097 1098 static int 1099 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 1100 { 1101 struct thread *td; 1102 const Elf_Ehdr *hdr; 1103 const Elf_Phdr *phdr; 1104 Elf_Auxargs *elf_auxargs; 1105 struct vmspace *vmspace; 1106 vm_map_t map; 1107 char *interp; 1108 Elf_Brandinfo *brand_info; 1109 struct sysentvec *sv; 1110 u_long addr, baddr, entry, proghdr; 1111 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc; 1112 uint32_t fctl0; 1113 int32_t osrel; 1114 bool free_interp; 1115 int error, i, n; 1116 1117 hdr = (const Elf_Ehdr *)imgp->image_header; 1118 1119 /* 1120 * Do we have a valid ELF header ? 1121 * 1122 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 1123 * if particular brand doesn't support it. 1124 */ 1125 if (__elfN(check_header)(hdr) != 0 || 1126 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 1127 return (-1); 1128 1129 /* 1130 * From here on down, we return an errno, not -1, as we've 1131 * detected an ELF file. 1132 */ 1133 1134 if (!__elfN(phdr_in_zero_page)(hdr)) { 1135 uprintf("Program headers not in the first page\n"); 1136 return (ENOEXEC); 1137 } 1138 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1139 if (!aligned(phdr, Elf_Addr)) { 1140 uprintf("Unaligned program headers\n"); 1141 return (ENOEXEC); 1142 } 1143 1144 n = error = 0; 1145 baddr = 0; 1146 osrel = 0; 1147 fctl0 = 0; 1148 entry = proghdr = 0; 1149 interp = NULL; 1150 free_interp = false; 1151 td = curthread; 1152 1153 /* 1154 * Somewhat arbitrary, limit accepted max alignment for the 1155 * loadable segment to the max supported superpage size. Too 1156 * large alignment requests are not useful and are indicators 1157 * of corrupted or outright malicious binary. 1158 */ 1159 maxalign = PAGE_SIZE; 1160 maxsalign = PAGE_SIZE * 1024; 1161 for (i = MAXPAGESIZES - 1; i > 0; i--) { 1162 if (pagesizes[i] > maxsalign) { 1163 maxsalign = pagesizes[i]; 1164 break; 1165 } 1166 } 1167 1168 mapsz = 0; 1169 1170 for (i = 0; i < hdr->e_phnum; i++) { 1171 switch (phdr[i].p_type) { 1172 case PT_LOAD: 1173 if (n == 0) 1174 baddr = phdr[i].p_vaddr; 1175 if (!powerof2(phdr[i].p_align) || 1176 phdr[i].p_align > maxsalign) { 1177 uprintf("Invalid segment alignment\n"); 1178 error = ENOEXEC; 1179 goto ret; 1180 } 1181 if (phdr[i].p_align > maxalign) 1182 maxalign = phdr[i].p_align; 1183 if (mapsz + phdr[i].p_memsz < mapsz) { 1184 uprintf("Mapsize overflow\n"); 1185 error = ENOEXEC; 1186 goto ret; 1187 } 1188 mapsz += phdr[i].p_memsz; 1189 n++; 1190 1191 /* 1192 * If this segment contains the program headers, 1193 * remember their virtual address for the AT_PHDR 1194 * aux entry. Static binaries don't usually include 1195 * a PT_PHDR entry. 1196 */ 1197 if (phdr[i].p_offset == 0 && 1198 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= 1199 phdr[i].p_filesz) 1200 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 1201 break; 1202 case PT_INTERP: 1203 /* Path to interpreter */ 1204 if (interp != NULL) { 1205 uprintf("Multiple PT_INTERP headers\n"); 1206 error = ENOEXEC; 1207 goto ret; 1208 } 1209 error = __elfN(get_interp)(imgp, &phdr[i], &interp, 1210 &free_interp); 1211 if (error != 0) 1212 goto ret; 1213 break; 1214 case PT_GNU_STACK: 1215 if (__elfN(nxstack)) { 1216 imgp->stack_prot = 1217 __elfN(trans_prot)(phdr[i].p_flags); 1218 if ((imgp->stack_prot & VM_PROT_RW) != 1219 VM_PROT_RW) { 1220 uprintf("Invalid PT_GNU_STACK\n"); 1221 error = ENOEXEC; 1222 goto ret; 1223 } 1224 } 1225 imgp->stack_sz = phdr[i].p_memsz; 1226 break; 1227 case PT_PHDR: /* Program header table info */ 1228 proghdr = phdr[i].p_vaddr; 1229 break; 1230 } 1231 } 1232 1233 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0); 1234 if (brand_info == NULL) { 1235 uprintf("ELF binary type \"%u\" not known.\n", 1236 hdr->e_ident[EI_OSABI]); 1237 error = ENOEXEC; 1238 goto ret; 1239 } 1240 sv = brand_info->sysvec; 1241 if (hdr->e_type == ET_DYN) { 1242 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 1243 uprintf("Cannot execute shared object\n"); 1244 error = ENOEXEC; 1245 goto ret; 1246 } 1247 /* 1248 * Honour the base load address from the dso if it is 1249 * non-zero for some reason. 1250 */ 1251 if (baddr == 0) { 1252 if ((sv->sv_flags & SV_ASLR) == 0 || 1253 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) 1254 imgp->et_dyn_addr = __elfN(pie_base); 1255 else if ((__elfN(pie_aslr_enabled) && 1256 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) || 1257 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0) 1258 imgp->et_dyn_addr = ET_DYN_ADDR_RAND; 1259 else 1260 imgp->et_dyn_addr = __elfN(pie_base); 1261 } 1262 } 1263 1264 /* 1265 * Avoid a possible deadlock if the current address space is destroyed 1266 * and that address space maps the locked vnode. In the common case, 1267 * the locked vnode's v_usecount is decremented but remains greater 1268 * than zero. Consequently, the vnode lock is not needed by vrele(). 1269 * However, in cases where the vnode lock is external, such as nullfs, 1270 * v_usecount may become zero. 1271 * 1272 * The VV_TEXT flag prevents modifications to the executable while 1273 * the vnode is unlocked. 1274 */ 1275 VOP_UNLOCK(imgp->vp); 1276 1277 /* 1278 * Decide whether to enable randomization of user mappings. 1279 * First, reset user preferences for the setid binaries. 1280 * Then, account for the support of the randomization by the 1281 * ABI, by user preferences, and make special treatment for 1282 * 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 if ((sv->sv_flags & SV_ASLR) == 0 || 1291 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 || 1292 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) { 1293 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND, 1294 ("imgp->et_dyn_addr == RAND and !ASLR")); 1295 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 || 1296 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) || 1297 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1298 imgp->map_flags |= MAP_ASLR; 1299 /* 1300 * If user does not care about sbrk, utilize the bss 1301 * grow region for mappings as well. We can select 1302 * the base for the image anywere and still not suffer 1303 * from the fragmentation. 1304 */ 1305 if (!__elfN(aslr_honor_sbrk) || 1306 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0) 1307 imgp->map_flags |= MAP_ASLR_IGNSTART; 1308 if (__elfN(aslr_stack)) 1309 imgp->map_flags |= MAP_ASLR_STACK; 1310 if (__elfN(aslr_shared_page)) 1311 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE; 1312 } 1313 1314 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 && 1315 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) || 1316 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0) 1317 imgp->map_flags |= MAP_WXORX; 1318 1319 error = exec_new_vmspace(imgp, sv); 1320 1321 imgp->proc->p_sysent = sv; 1322 imgp->proc->p_elf_brandinfo = brand_info; 1323 1324 vmspace = imgp->proc->p_vmspace; 1325 map = &vmspace->vm_map; 1326 maxv = sv->sv_usrstack; 1327 if ((imgp->map_flags & MAP_ASLR_STACK) == 0) 1328 maxv -= lim_max(td, RLIMIT_STACK); 1329 if (error == 0 && mapsz >= maxv - vm_map_min(map)) { 1330 uprintf("Excessive mapping size\n"); 1331 error = ENOEXEC; 1332 } 1333 1334 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1335 KASSERT((map->flags & MAP_ASLR) != 0, 1336 ("ET_DYN_ADDR_RAND but !MAP_ASLR")); 1337 error = __CONCAT(rnd_, __elfN(base))(map, 1338 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA), 1339 /* reserve half of the address space to interpreter */ 1340 maxv / 2, maxalign, &imgp->et_dyn_addr); 1341 } 1342 1343 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1344 if (error != 0) 1345 goto ret; 1346 1347 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL); 1348 if (error != 0) 1349 goto ret; 1350 1351 error = __elfN(enforce_limits)(imgp, hdr, phdr); 1352 if (error != 0) 1353 goto ret; 1354 1355 /* 1356 * We load the dynamic linker where a userland call 1357 * to mmap(0, ...) would put it. The rationale behind this 1358 * calculation is that it leaves room for the heap to grow to 1359 * its maximum allowed size. 1360 */ 1361 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1362 RLIMIT_DATA)); 1363 if ((map->flags & MAP_ASLR) != 0) { 1364 maxv1 = maxv / 2 + addr / 2; 1365 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1, 1366 #if VM_NRESERVLEVEL > 0 1367 pagesizes[VM_NRESERVLEVEL] != 0 ? 1368 /* Align anon_loc to the largest superpage size. */ 1369 pagesizes[VM_NRESERVLEVEL] : 1370 #endif 1371 pagesizes[0], &anon_loc); 1372 if (error != 0) 1373 goto ret; 1374 map->anon_loc = anon_loc; 1375 } else { 1376 map->anon_loc = addr; 1377 } 1378 1379 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr; 1380 imgp->entry_addr = entry; 1381 1382 if (sv->sv_protect != NULL) 1383 sv->sv_protect(imgp, SVP_IMAGE); 1384 1385 if (interp != NULL) { 1386 VOP_UNLOCK(imgp->vp); 1387 if ((map->flags & MAP_ASLR) != 0) { 1388 /* Assume that interpreter fits into 1/4 of AS */ 1389 maxv1 = maxv / 2 + addr / 2; 1390 error = __CONCAT(rnd_, __elfN(base))(map, addr, 1391 maxv1, PAGE_SIZE, &addr); 1392 } 1393 if (error == 0) { 1394 error = __elfN(load_interp)(imgp, brand_info, interp, 1395 &addr, &imgp->entry_addr); 1396 } 1397 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1398 if (error != 0) 1399 goto ret; 1400 } else 1401 addr = imgp->et_dyn_addr; 1402 1403 error = exec_map_stack(imgp); 1404 if (error != 0) 1405 goto ret; 1406 1407 /* 1408 * Construct auxargs table (used by the copyout_auxargs routine) 1409 */ 1410 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT); 1411 if (elf_auxargs == NULL) { 1412 VOP_UNLOCK(imgp->vp); 1413 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1414 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1415 } 1416 elf_auxargs->execfd = -1; 1417 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr; 1418 elf_auxargs->phent = hdr->e_phentsize; 1419 elf_auxargs->phnum = hdr->e_phnum; 1420 elf_auxargs->pagesz = PAGE_SIZE; 1421 elf_auxargs->base = addr; 1422 elf_auxargs->flags = 0; 1423 elf_auxargs->entry = entry; 1424 elf_auxargs->hdr_eflags = hdr->e_flags; 1425 1426 imgp->auxargs = elf_auxargs; 1427 imgp->interpreted = 0; 1428 imgp->reloc_base = addr; 1429 imgp->proc->p_osrel = osrel; 1430 imgp->proc->p_fctl0 = fctl0; 1431 imgp->proc->p_elf_flags = hdr->e_flags; 1432 1433 ret: 1434 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock"); 1435 if (free_interp) 1436 free(interp, M_TEMP); 1437 return (error); 1438 } 1439 1440 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE) 1441 1442 int 1443 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base) 1444 { 1445 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1446 Elf_Auxinfo *argarray, *pos; 1447 struct vmspace *vmspace; 1448 rlim_t stacksz; 1449 int error, oc; 1450 uint32_t bsdflags; 1451 1452 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1453 M_WAITOK | M_ZERO); 1454 1455 vmspace = imgp->proc->p_vmspace; 1456 1457 if (args->execfd != -1) 1458 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1459 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1460 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1461 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1462 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1463 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1464 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1465 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1466 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1467 if (imgp->execpathp != 0) 1468 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp); 1469 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1470 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1471 if (imgp->canary != 0) { 1472 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary); 1473 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1474 } 1475 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1476 if (imgp->pagesizes != 0) { 1477 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes); 1478 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1479 } 1480 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) { 1481 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1482 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset); 1483 } 1484 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1485 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1486 imgp->sysent->sv_stackprot); 1487 if (imgp->sysent->sv_hwcap != NULL) 1488 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1489 if (imgp->sysent->sv_hwcap2 != NULL) 1490 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1491 if (imgp->sysent->sv_hwcap3 != NULL) 1492 AUXARGS_ENTRY(pos, AT_HWCAP3, *imgp->sysent->sv_hwcap3); 1493 if (imgp->sysent->sv_hwcap4 != NULL) 1494 AUXARGS_ENTRY(pos, AT_HWCAP4, *imgp->sysent->sv_hwcap4); 1495 bsdflags = 0; 1496 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0; 1497 oc = atomic_load_int(&vm_overcommit); 1498 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) != 1499 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0; 1500 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags); 1501 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc); 1502 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv); 1503 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc); 1504 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv); 1505 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings); 1506 #ifdef RANDOM_FENESTRASX 1507 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) { 1508 AUXARGS_ENTRY(pos, AT_FXRNG, 1509 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset); 1510 } 1511 #endif 1512 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) { 1513 AUXARGS_ENTRY(pos, AT_KPRELOAD, 1514 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset); 1515 } 1516 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop)); 1517 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur; 1518 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz); 1519 AUXARGS_ENTRY(pos, AT_NULL, 0); 1520 1521 free(imgp->auxargs, M_TEMP); 1522 imgp->auxargs = NULL; 1523 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1524 1525 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT); 1526 free(argarray, M_TEMP); 1527 return (error); 1528 } 1529 1530 int 1531 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp) 1532 { 1533 Elf_Addr *base; 1534 1535 base = (Elf_Addr *)*stack_base; 1536 base--; 1537 if (elf_suword(base, imgp->args->argc) == -1) 1538 return (EFAULT); 1539 *stack_base = (uintptr_t)base; 1540 return (0); 1541 } 1542 1543 /* 1544 * Code for generating ELF core dumps. 1545 */ 1546 1547 typedef void (*segment_callback)(vm_map_entry_t, void *); 1548 1549 /* Closure for cb_put_phdr(). */ 1550 struct phdr_closure { 1551 Elf_Phdr *phdr; /* Program header to fill in */ 1552 Elf_Off offset; /* Offset of segment in core file */ 1553 }; 1554 1555 struct note_info { 1556 int type; /* Note type. */ 1557 struct regset *regset; /* Register set. */ 1558 outfunc_t outfunc; /* Output function. */ 1559 void *outarg; /* Argument for the output function. */ 1560 size_t outsize; /* Output size. */ 1561 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1562 }; 1563 1564 TAILQ_HEAD(note_info_list, note_info); 1565 1566 static void cb_put_phdr(vm_map_entry_t, void *); 1567 static void cb_size_segment(vm_map_entry_t, void *); 1568 static void each_dumpable_segment(struct thread *, segment_callback, void *, 1569 int); 1570 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1571 struct note_info_list *, size_t, int); 1572 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *); 1573 1574 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1575 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1576 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1577 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1578 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1579 static void __elfN(note_procstat_kqueues)(void *, struct sbuf *, size_t *); 1580 static void note_procstat_files(void *, struct sbuf *, size_t *); 1581 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1582 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1583 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1584 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1585 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1586 1587 static int 1588 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1589 { 1590 1591 return (core_write((struct coredump_params *)arg, base, len, offset, 1592 UIO_SYSSPACE, NULL)); 1593 } 1594 1595 int 1596 __elfN(coredump)(struct thread *td, struct coredump_writer *cdw, off_t limit, int flags) 1597 { 1598 struct ucred *cred = td->td_ucred; 1599 int compm, error = 0; 1600 struct sseg_closure seginfo; 1601 struct note_info_list notelst; 1602 struct coredump_params params; 1603 struct note_info *ninfo; 1604 void *hdr, *tmpbuf; 1605 size_t hdrsize, notesz, coresize; 1606 1607 hdr = NULL; 1608 tmpbuf = NULL; 1609 TAILQ_INIT(¬elst); 1610 1611 /* Size the program segments. */ 1612 __elfN(size_segments)(td, &seginfo, flags); 1613 1614 /* 1615 * Collect info about the core file header area. 1616 */ 1617 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1618 if (seginfo.count + 1 >= PN_XNUM) 1619 hdrsize += sizeof(Elf_Shdr); 1620 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz); 1621 coresize = round_page(hdrsize + notesz) + seginfo.size; 1622 1623 /* Set up core dump parameters. */ 1624 params.offset = 0; 1625 params.active_cred = cred; 1626 params.td = td; 1627 params.cdw = cdw; 1628 params.comp = NULL; 1629 1630 #ifdef RACCT 1631 if (racct_enable) { 1632 PROC_LOCK(td->td_proc); 1633 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1634 PROC_UNLOCK(td->td_proc); 1635 if (error != 0) { 1636 error = EFAULT; 1637 goto done; 1638 } 1639 } 1640 #endif 1641 if (coresize >= limit) { 1642 error = EFAULT; 1643 goto done; 1644 } 1645 1646 /* Create a compression stream if necessary. */ 1647 compm = compress_user_cores; 1648 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP && 1649 compm == 0) 1650 compm = COMPRESS_GZIP; 1651 if (compm != 0) { 1652 params.comp = compressor_init(core_compressed_write, 1653 compm, CORE_BUF_SIZE, 1654 compress_user_cores_level, ¶ms); 1655 if (params.comp == NULL) { 1656 error = EFAULT; 1657 goto done; 1658 } 1659 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1660 } 1661 1662 if (cdw->init_fn != NULL) { 1663 error = (*cdw->init_fn)(cdw, ¶ms); 1664 if (error != 0) 1665 goto done; 1666 } 1667 1668 /* 1669 * Allocate memory for building the header, fill it up, 1670 * and write it out following the notes. 1671 */ 1672 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1673 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1674 notesz, flags); 1675 1676 /* Write the contents of all of the writable segments. */ 1677 if (error == 0) { 1678 Elf_Phdr *php; 1679 off_t offset; 1680 int i; 1681 1682 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1683 offset = round_page(hdrsize + notesz); 1684 for (i = 0; i < seginfo.count; i++) { 1685 error = core_output((char *)(uintptr_t)php->p_vaddr, 1686 php->p_filesz, offset, ¶ms, tmpbuf); 1687 if (error != 0) 1688 break; 1689 offset += php->p_filesz; 1690 php++; 1691 } 1692 if (error == 0 && params.comp != NULL) 1693 error = compressor_flush(params.comp); 1694 } 1695 if (error) { 1696 log(LOG_WARNING, 1697 "Failed to write core file for process %s (error %d)\n", 1698 curproc->p_comm, error); 1699 } 1700 1701 done: 1702 free(tmpbuf, M_TEMP); 1703 if (params.comp != NULL) 1704 compressor_fini(params.comp); 1705 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1706 TAILQ_REMOVE(¬elst, ninfo, link); 1707 free(ninfo, M_TEMP); 1708 } 1709 if (hdr != NULL) 1710 free(hdr, M_TEMP); 1711 1712 return (error); 1713 } 1714 1715 /* 1716 * A callback for each_dumpable_segment() to write out the segment's 1717 * program header entry. 1718 */ 1719 static void 1720 cb_put_phdr(vm_map_entry_t entry, void *closure) 1721 { 1722 struct phdr_closure *phc = (struct phdr_closure *)closure; 1723 Elf_Phdr *phdr = phc->phdr; 1724 1725 phc->offset = round_page(phc->offset); 1726 1727 phdr->p_type = PT_LOAD; 1728 phdr->p_offset = phc->offset; 1729 phdr->p_vaddr = entry->start; 1730 phdr->p_paddr = 0; 1731 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1732 phdr->p_align = PAGE_SIZE; 1733 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1734 1735 phc->offset += phdr->p_filesz; 1736 phc->phdr++; 1737 } 1738 1739 /* 1740 * A callback for each_dumpable_segment() to gather information about 1741 * the number of segments and their total size. 1742 */ 1743 static void 1744 cb_size_segment(vm_map_entry_t entry, void *closure) 1745 { 1746 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1747 1748 ssc->count++; 1749 ssc->size += entry->end - entry->start; 1750 } 1751 1752 void 1753 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo, 1754 int flags) 1755 { 1756 seginfo->count = 0; 1757 seginfo->size = 0; 1758 1759 each_dumpable_segment(td, cb_size_segment, seginfo, flags); 1760 } 1761 1762 /* 1763 * For each writable segment in the process's memory map, call the given 1764 * function with a pointer to the map entry and some arbitrary 1765 * caller-supplied data. 1766 */ 1767 static void 1768 each_dumpable_segment(struct thread *td, segment_callback func, void *closure, 1769 int flags) 1770 { 1771 struct proc *p = td->td_proc; 1772 vm_map_t map = &p->p_vmspace->vm_map; 1773 vm_map_entry_t entry; 1774 vm_object_t backing_object, object; 1775 bool ignore_entry; 1776 1777 vm_map_lock_read(map); 1778 VM_MAP_ENTRY_FOREACH(entry, map) { 1779 /* 1780 * Don't dump inaccessible mappings, deal with legacy 1781 * coredump mode. 1782 * 1783 * Note that read-only segments related to the elf binary 1784 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1785 * need to arbitrarily ignore such segments. 1786 */ 1787 if ((flags & SVC_ALL) == 0) { 1788 if (elf_legacy_coredump) { 1789 if ((entry->protection & VM_PROT_RW) != 1790 VM_PROT_RW) 1791 continue; 1792 } else { 1793 if ((entry->protection & VM_PROT_ALL) == 0) 1794 continue; 1795 } 1796 } 1797 1798 /* 1799 * Dont include memory segment in the coredump if 1800 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1801 * madvise(2). Do not dump submaps (i.e. parts of the 1802 * kernel map). 1803 */ 1804 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 1805 continue; 1806 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 && 1807 (flags & SVC_ALL) == 0) 1808 continue; 1809 if ((object = entry->object.vm_object) == NULL) 1810 continue; 1811 1812 /* Ignore memory-mapped devices and such things. */ 1813 VM_OBJECT_RLOCK(object); 1814 while ((backing_object = object->backing_object) != NULL) { 1815 VM_OBJECT_RLOCK(backing_object); 1816 VM_OBJECT_RUNLOCK(object); 1817 object = backing_object; 1818 } 1819 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0; 1820 VM_OBJECT_RUNLOCK(object); 1821 if (ignore_entry) 1822 continue; 1823 1824 (*func)(entry, closure); 1825 } 1826 vm_map_unlock_read(map); 1827 } 1828 1829 /* 1830 * Write the core file header to the file, including padding up to 1831 * the page boundary. 1832 */ 1833 static int 1834 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1835 size_t hdrsize, struct note_info_list *notelst, size_t notesz, 1836 int flags) 1837 { 1838 struct note_info *ninfo; 1839 struct sbuf *sb; 1840 int error; 1841 1842 /* Fill in the header. */ 1843 bzero(hdr, hdrsize); 1844 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags); 1845 1846 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1847 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1848 sbuf_start_section(sb, NULL); 1849 sbuf_bcat(sb, hdr, hdrsize); 1850 TAILQ_FOREACH(ninfo, notelst, link) 1851 __elfN(putnote)(p->td, ninfo, sb); 1852 /* Align up to a page boundary for the program segments. */ 1853 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1854 error = sbuf_finish(sb); 1855 sbuf_delete(sb); 1856 1857 return (error); 1858 } 1859 1860 void 1861 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1862 size_t *sizep) 1863 { 1864 struct proc *p; 1865 struct thread *thr; 1866 size_t size; 1867 1868 p = td->td_proc; 1869 size = 0; 1870 1871 size += __elfN(register_note)(td, list, NT_PRPSINFO, 1872 __elfN(note_prpsinfo), p); 1873 1874 /* 1875 * To have the debugger select the right thread (LWP) as the initial 1876 * thread, we dump the state of the thread passed to us in td first. 1877 * This is the thread that causes the core dump and thus likely to 1878 * be the right thread one wants to have selected in the debugger. 1879 */ 1880 thr = td; 1881 while (thr != NULL) { 1882 size += __elfN(prepare_register_notes)(td, list, thr); 1883 size += __elfN(register_note)(td, list, -1, 1884 __elfN(note_threadmd), thr); 1885 1886 thr = thr == td ? TAILQ_FIRST(&p->p_threads) : 1887 TAILQ_NEXT(thr, td_plist); 1888 if (thr == td) 1889 thr = TAILQ_NEXT(thr, td_plist); 1890 } 1891 1892 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC, 1893 __elfN(note_procstat_proc), p); 1894 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES, 1895 note_procstat_files, p); 1896 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP, 1897 note_procstat_vmmap, p); 1898 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS, 1899 note_procstat_groups, p); 1900 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK, 1901 note_procstat_umask, p); 1902 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT, 1903 note_procstat_rlimit, p); 1904 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL, 1905 note_procstat_osrel, p); 1906 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS, 1907 __elfN(note_procstat_psstrings), p); 1908 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV, 1909 __elfN(note_procstat_auxv), p); 1910 size += __elfN(register_note)(td, list, NT_PROCSTAT_KQUEUES, 1911 __elfN(note_procstat_kqueues), p); 1912 1913 *sizep = size; 1914 } 1915 1916 void 1917 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1918 size_t notesz, int flags) 1919 { 1920 Elf_Ehdr *ehdr; 1921 Elf_Phdr *phdr; 1922 Elf_Shdr *shdr; 1923 struct phdr_closure phc; 1924 Elf_Brandinfo *bi; 1925 1926 ehdr = (Elf_Ehdr *)hdr; 1927 bi = td->td_proc->p_elf_brandinfo; 1928 1929 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1930 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1931 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1932 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1933 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1934 ehdr->e_ident[EI_DATA] = ELF_DATA; 1935 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1936 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi; 1937 ehdr->e_ident[EI_ABIVERSION] = 0; 1938 ehdr->e_ident[EI_PAD] = 0; 1939 ehdr->e_type = ET_CORE; 1940 ehdr->e_machine = bi->machine; 1941 ehdr->e_version = EV_CURRENT; 1942 ehdr->e_entry = 0; 1943 ehdr->e_phoff = sizeof(Elf_Ehdr); 1944 ehdr->e_flags = td->td_proc->p_elf_flags; 1945 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1946 ehdr->e_phentsize = sizeof(Elf_Phdr); 1947 ehdr->e_shentsize = sizeof(Elf_Shdr); 1948 ehdr->e_shstrndx = SHN_UNDEF; 1949 if (numsegs + 1 < PN_XNUM) { 1950 ehdr->e_phnum = numsegs + 1; 1951 ehdr->e_shnum = 0; 1952 } else { 1953 ehdr->e_phnum = PN_XNUM; 1954 ehdr->e_shnum = 1; 1955 1956 ehdr->e_shoff = ehdr->e_phoff + 1957 (numsegs + 1) * ehdr->e_phentsize; 1958 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1959 ("e_shoff: %zu, hdrsize - shdr: %zu", 1960 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1961 1962 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1963 memset(shdr, 0, sizeof(*shdr)); 1964 /* 1965 * A special first section is used to hold large segment and 1966 * section counts. This was proposed by Sun Microsystems in 1967 * Solaris and has been adopted by Linux; the standard ELF 1968 * tools are already familiar with the technique. 1969 * 1970 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1971 * (or 12-7 depending on the version of the document) for more 1972 * details. 1973 */ 1974 shdr->sh_type = SHT_NULL; 1975 shdr->sh_size = ehdr->e_shnum; 1976 shdr->sh_link = ehdr->e_shstrndx; 1977 shdr->sh_info = numsegs + 1; 1978 } 1979 1980 /* 1981 * Fill in the program header entries. 1982 */ 1983 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1984 1985 /* The note segement. */ 1986 phdr->p_type = PT_NOTE; 1987 phdr->p_offset = hdrsize; 1988 phdr->p_vaddr = 0; 1989 phdr->p_paddr = 0; 1990 phdr->p_filesz = notesz; 1991 phdr->p_memsz = 0; 1992 phdr->p_flags = PF_R; 1993 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1994 phdr++; 1995 1996 /* All the writable segments from the program. */ 1997 phc.phdr = phdr; 1998 phc.offset = round_page(hdrsize + notesz); 1999 each_dumpable_segment(td, cb_put_phdr, &phc, flags); 2000 } 2001 2002 static size_t 2003 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list, 2004 struct regset *regset, struct thread *target_td) 2005 { 2006 const struct sysentvec *sv; 2007 struct note_info *ninfo; 2008 size_t size, notesize; 2009 2010 size = 0; 2011 if (!regset->get(regset, target_td, NULL, &size) || size == 0) 2012 return (0); 2013 2014 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2015 ninfo->type = regset->note; 2016 ninfo->regset = regset; 2017 ninfo->outarg = target_td; 2018 ninfo->outsize = size; 2019 TAILQ_INSERT_TAIL(list, ninfo, link); 2020 2021 sv = td->td_proc->p_sysent; 2022 notesize = sizeof(Elf_Note) + /* note header */ 2023 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2024 /* note name */ 2025 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2026 2027 return (notesize); 2028 } 2029 2030 size_t 2031 __elfN(register_note)(struct thread *td, struct note_info_list *list, 2032 int type, outfunc_t out, void *arg) 2033 { 2034 const struct sysentvec *sv; 2035 struct note_info *ninfo; 2036 size_t size, notesize; 2037 2038 sv = td->td_proc->p_sysent; 2039 size = 0; 2040 out(arg, NULL, &size); 2041 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2042 ninfo->type = type; 2043 ninfo->outfunc = out; 2044 ninfo->outarg = arg; 2045 ninfo->outsize = size; 2046 TAILQ_INSERT_TAIL(list, ninfo, link); 2047 2048 if (type == -1) 2049 return (size); 2050 2051 notesize = sizeof(Elf_Note) + /* note header */ 2052 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2053 /* note name */ 2054 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2055 2056 return (notesize); 2057 } 2058 2059 static size_t 2060 append_note_data(const void *src, void *dst, size_t len) 2061 { 2062 size_t padded_len; 2063 2064 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 2065 if (dst != NULL) { 2066 bcopy(src, dst, len); 2067 bzero((char *)dst + len, padded_len - len); 2068 } 2069 return (padded_len); 2070 } 2071 2072 size_t 2073 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 2074 { 2075 Elf_Note *note; 2076 char *buf; 2077 size_t notesize; 2078 2079 buf = dst; 2080 if (buf != NULL) { 2081 note = (Elf_Note *)buf; 2082 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 2083 note->n_descsz = size; 2084 note->n_type = type; 2085 buf += sizeof(*note); 2086 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 2087 sizeof(FREEBSD_ABI_VENDOR)); 2088 append_note_data(src, buf, size); 2089 if (descp != NULL) 2090 *descp = buf; 2091 } 2092 2093 notesize = sizeof(Elf_Note) + /* note header */ 2094 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 2095 /* note name */ 2096 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2097 2098 return (notesize); 2099 } 2100 2101 static void 2102 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb) 2103 { 2104 Elf_Note note; 2105 const struct sysentvec *sv; 2106 ssize_t old_len, sect_len; 2107 size_t new_len, descsz, i; 2108 2109 if (ninfo->type == -1) { 2110 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2111 return; 2112 } 2113 2114 sv = td->td_proc->p_sysent; 2115 2116 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1; 2117 note.n_descsz = ninfo->outsize; 2118 note.n_type = ninfo->type; 2119 2120 sbuf_bcat(sb, ¬e, sizeof(note)); 2121 sbuf_start_section(sb, &old_len); 2122 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor, 2123 strlen(sv->sv_elf_core_abi_vendor) + 1); 2124 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2125 if (note.n_descsz == 0) 2126 return; 2127 sbuf_start_section(sb, &old_len); 2128 if (ninfo->regset != NULL) { 2129 struct regset *regset = ninfo->regset; 2130 void *buf; 2131 2132 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK); 2133 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize); 2134 sbuf_bcat(sb, buf, ninfo->outsize); 2135 free(buf, M_TEMP); 2136 } else 2137 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2138 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2139 if (sect_len < 0) 2140 return; 2141 2142 new_len = (size_t)sect_len; 2143 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 2144 if (new_len < descsz) { 2145 /* 2146 * It is expected that individual note emitters will correctly 2147 * predict their expected output size and fill up to that size 2148 * themselves, padding in a format-specific way if needed. 2149 * However, in case they don't, just do it here with zeros. 2150 */ 2151 for (i = 0; i < descsz - new_len; i++) 2152 sbuf_putc(sb, 0); 2153 } else if (new_len > descsz) { 2154 /* 2155 * We can't always truncate sb -- we may have drained some 2156 * of it already. 2157 */ 2158 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 2159 "read it (%zu > %zu). Since it is longer than " 2160 "expected, this coredump's notes are corrupt. THIS " 2161 "IS A BUG in the note_procstat routine for type %u.\n", 2162 __func__, (unsigned)note.n_type, new_len, descsz, 2163 (unsigned)note.n_type)); 2164 } 2165 } 2166 2167 /* 2168 * Miscellaneous note out functions. 2169 */ 2170 2171 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2172 #include <compat/freebsd32/freebsd32.h> 2173 #include <compat/freebsd32/freebsd32_signal.h> 2174 2175 typedef struct prstatus32 elf_prstatus_t; 2176 typedef struct prpsinfo32 elf_prpsinfo_t; 2177 typedef struct fpreg32 elf_prfpregset_t; 2178 typedef struct fpreg32 elf_fpregset_t; 2179 typedef struct reg32 elf_gregset_t; 2180 typedef struct thrmisc32 elf_thrmisc_t; 2181 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t; 2182 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 2183 typedef struct kinfo_proc32 elf_kinfo_proc_t; 2184 typedef uint32_t elf_ps_strings_t; 2185 #else 2186 typedef prstatus_t elf_prstatus_t; 2187 typedef prpsinfo_t elf_prpsinfo_t; 2188 typedef prfpregset_t elf_prfpregset_t; 2189 typedef prfpregset_t elf_fpregset_t; 2190 typedef gregset_t elf_gregset_t; 2191 typedef thrmisc_t elf_thrmisc_t; 2192 typedef struct ptrace_lwpinfo elf_lwpinfo_t; 2193 #define ELF_KERN_PROC_MASK 0 2194 typedef struct kinfo_proc elf_kinfo_proc_t; 2195 typedef vm_offset_t elf_ps_strings_t; 2196 #endif 2197 2198 static void 2199 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2200 { 2201 struct sbuf sbarg; 2202 size_t len; 2203 char *cp, *end; 2204 struct proc *p; 2205 elf_prpsinfo_t *psinfo; 2206 int error; 2207 2208 p = arg; 2209 if (sb != NULL) { 2210 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 2211 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 2212 psinfo->pr_version = PRPSINFO_VERSION; 2213 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 2214 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 2215 PROC_LOCK(p); 2216 if (p->p_args != NULL) { 2217 len = sizeof(psinfo->pr_psargs) - 1; 2218 if (len > p->p_args->ar_length) 2219 len = p->p_args->ar_length; 2220 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 2221 PROC_UNLOCK(p); 2222 error = 0; 2223 } else { 2224 _PHOLD(p); 2225 PROC_UNLOCK(p); 2226 sbuf_new(&sbarg, psinfo->pr_psargs, 2227 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 2228 error = proc_getargv(curthread, p, &sbarg); 2229 PRELE(p); 2230 if (sbuf_finish(&sbarg) == 0) { 2231 len = sbuf_len(&sbarg); 2232 if (len > 0) 2233 len--; 2234 } else { 2235 len = sizeof(psinfo->pr_psargs) - 1; 2236 } 2237 sbuf_delete(&sbarg); 2238 } 2239 if (error != 0 || len == 0 || (ssize_t)len == -1) 2240 strlcpy(psinfo->pr_psargs, p->p_comm, 2241 sizeof(psinfo->pr_psargs)); 2242 else { 2243 KASSERT(len < sizeof(psinfo->pr_psargs), 2244 ("len is too long: %zu vs %zu", len, 2245 sizeof(psinfo->pr_psargs))); 2246 cp = psinfo->pr_psargs; 2247 end = cp + len - 1; 2248 for (;;) { 2249 cp = memchr(cp, '\0', end - cp); 2250 if (cp == NULL) 2251 break; 2252 *cp = ' '; 2253 } 2254 } 2255 psinfo->pr_pid = p->p_pid; 2256 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 2257 free(psinfo, M_TEMP); 2258 } 2259 *sizep = sizeof(*psinfo); 2260 } 2261 2262 static bool 2263 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf, 2264 size_t *sizep) 2265 { 2266 elf_prstatus_t *status; 2267 2268 if (buf != NULL) { 2269 KASSERT(*sizep == sizeof(*status), ("%s: invalid size", 2270 __func__)); 2271 status = buf; 2272 memset(status, 0, *sizep); 2273 status->pr_version = PRSTATUS_VERSION; 2274 status->pr_statussz = sizeof(elf_prstatus_t); 2275 status->pr_gregsetsz = sizeof(elf_gregset_t); 2276 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 2277 status->pr_osreldate = osreldate; 2278 status->pr_cursig = td->td_proc->p_sig; 2279 status->pr_pid = td->td_tid; 2280 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2281 fill_regs32(td, &status->pr_reg); 2282 #else 2283 fill_regs(td, &status->pr_reg); 2284 #endif 2285 } 2286 *sizep = sizeof(*status); 2287 return (true); 2288 } 2289 2290 static bool 2291 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf, 2292 size_t size) 2293 { 2294 elf_prstatus_t *status; 2295 2296 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__)); 2297 status = buf; 2298 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2299 set_regs32(td, &status->pr_reg); 2300 #else 2301 set_regs(td, &status->pr_reg); 2302 #endif 2303 return (true); 2304 } 2305 2306 static struct regset __elfN(regset_prstatus) = { 2307 .note = NT_PRSTATUS, 2308 .size = sizeof(elf_prstatus_t), 2309 .get = __elfN(get_prstatus), 2310 .set = __elfN(set_prstatus), 2311 }; 2312 ELF_REGSET(__elfN(regset_prstatus)); 2313 2314 static bool 2315 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf, 2316 size_t *sizep) 2317 { 2318 elf_prfpregset_t *fpregset; 2319 2320 if (buf != NULL) { 2321 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size", 2322 __func__)); 2323 fpregset = buf; 2324 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2325 fill_fpregs32(td, fpregset); 2326 #else 2327 fill_fpregs(td, fpregset); 2328 #endif 2329 } 2330 *sizep = sizeof(*fpregset); 2331 return (true); 2332 } 2333 2334 static bool 2335 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf, 2336 size_t size) 2337 { 2338 elf_prfpregset_t *fpregset; 2339 2340 fpregset = buf; 2341 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__)); 2342 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2343 set_fpregs32(td, fpregset); 2344 #else 2345 set_fpregs(td, fpregset); 2346 #endif 2347 return (true); 2348 } 2349 2350 static struct regset __elfN(regset_fpregset) = { 2351 .note = NT_FPREGSET, 2352 .size = sizeof(elf_prfpregset_t), 2353 .get = __elfN(get_fpregset), 2354 .set = __elfN(set_fpregset), 2355 }; 2356 ELF_REGSET(__elfN(regset_fpregset)); 2357 2358 static bool 2359 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf, 2360 size_t *sizep) 2361 { 2362 elf_thrmisc_t *thrmisc; 2363 2364 if (buf != NULL) { 2365 KASSERT(*sizep == sizeof(*thrmisc), 2366 ("%s: invalid size", __func__)); 2367 thrmisc = buf; 2368 bzero(thrmisc, sizeof(*thrmisc)); 2369 strcpy(thrmisc->pr_tname, td->td_name); 2370 } 2371 *sizep = sizeof(*thrmisc); 2372 return (true); 2373 } 2374 2375 static struct regset __elfN(regset_thrmisc) = { 2376 .note = NT_THRMISC, 2377 .size = sizeof(elf_thrmisc_t), 2378 .get = __elfN(get_thrmisc), 2379 }; 2380 ELF_REGSET(__elfN(regset_thrmisc)); 2381 2382 static bool 2383 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf, 2384 size_t *sizep) 2385 { 2386 elf_lwpinfo_t pl; 2387 size_t size; 2388 int structsize; 2389 2390 size = sizeof(structsize) + sizeof(pl); 2391 if (buf != NULL) { 2392 KASSERT(*sizep == size, ("%s: invalid size", __func__)); 2393 structsize = sizeof(pl); 2394 memcpy(buf, &structsize, sizeof(structsize)); 2395 bzero(&pl, sizeof(pl)); 2396 pl.pl_lwpid = td->td_tid; 2397 pl.pl_event = PL_EVENT_NONE; 2398 pl.pl_sigmask = td->td_sigmask; 2399 pl.pl_siglist = td->td_siglist; 2400 if (td->td_si.si_signo != 0) { 2401 pl.pl_event = PL_EVENT_SIGNAL; 2402 pl.pl_flags |= PL_FLAG_SI; 2403 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2404 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2405 #else 2406 pl.pl_siginfo = td->td_si; 2407 #endif 2408 } 2409 strcpy(pl.pl_tdname, td->td_name); 2410 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2411 memcpy((int *)buf + 1, &pl, sizeof(pl)); 2412 } 2413 *sizep = size; 2414 return (true); 2415 } 2416 2417 static struct regset __elfN(regset_lwpinfo) = { 2418 .note = NT_PTLWPINFO, 2419 .size = sizeof(int) + sizeof(elf_lwpinfo_t), 2420 .get = __elfN(get_lwpinfo), 2421 }; 2422 ELF_REGSET(__elfN(regset_lwpinfo)); 2423 2424 static size_t 2425 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list, 2426 struct thread *target_td) 2427 { 2428 struct sysentvec *sv = td->td_proc->p_sysent; 2429 struct regset **regsetp, **regset_end, *regset; 2430 size_t size; 2431 2432 size = 0; 2433 2434 if (target_td == td) 2435 cpu_update_pcb(target_td); 2436 2437 /* NT_PRSTATUS must be the first register set note. */ 2438 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus), 2439 target_td); 2440 2441 regsetp = sv->sv_regset_begin; 2442 if (regsetp == NULL) { 2443 /* XXX: This shouldn't be true for any FreeBSD ABIs. */ 2444 size += __elfN(register_regset_note)(td, list, 2445 &__elfN(regset_fpregset), target_td); 2446 return (size); 2447 } 2448 regset_end = sv->sv_regset_end; 2449 MPASS(regset_end != NULL); 2450 for (; regsetp < regset_end; regsetp++) { 2451 regset = *regsetp; 2452 if (regset->note == NT_PRSTATUS) 2453 continue; 2454 size += __elfN(register_regset_note)(td, list, regset, 2455 target_td); 2456 } 2457 return (size); 2458 } 2459 2460 /* 2461 * Allow for MD specific notes, as well as any MD 2462 * specific preparations for writing MI notes. 2463 */ 2464 static void 2465 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2466 { 2467 struct thread *td; 2468 void *buf; 2469 size_t size; 2470 2471 td = (struct thread *)arg; 2472 size = *sizep; 2473 if (size != 0 && sb != NULL) 2474 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2475 else 2476 buf = NULL; 2477 size = 0; 2478 __elfN(dump_thread)(td, buf, &size); 2479 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2480 if (size != 0 && sb != NULL) 2481 sbuf_bcat(sb, buf, size); 2482 free(buf, M_TEMP); 2483 *sizep = size; 2484 } 2485 2486 #ifdef KINFO_PROC_SIZE 2487 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2488 #endif 2489 2490 static void 2491 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2492 { 2493 struct proc *p; 2494 size_t size; 2495 int structsize; 2496 2497 p = arg; 2498 size = sizeof(structsize) + p->p_numthreads * 2499 sizeof(elf_kinfo_proc_t); 2500 2501 if (sb != NULL) { 2502 KASSERT(*sizep == size, ("invalid size")); 2503 structsize = sizeof(elf_kinfo_proc_t); 2504 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2505 sx_slock(&proctree_lock); 2506 PROC_LOCK(p); 2507 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2508 sx_sunlock(&proctree_lock); 2509 } 2510 *sizep = size; 2511 } 2512 2513 #ifdef KINFO_FILE_SIZE 2514 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2515 #endif 2516 2517 static void 2518 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2519 { 2520 struct proc *p; 2521 size_t size, sect_sz, i; 2522 ssize_t start_len, sect_len; 2523 int structsize, filedesc_flags; 2524 2525 if (coredump_pack_fileinfo) 2526 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2527 else 2528 filedesc_flags = 0; 2529 2530 p = arg; 2531 structsize = sizeof(struct kinfo_file); 2532 if (sb == NULL) { 2533 size = 0; 2534 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2535 sbuf_set_drain(sb, sbuf_count_drain, &size); 2536 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2537 PROC_LOCK(p); 2538 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2539 sbuf_finish(sb); 2540 sbuf_delete(sb); 2541 *sizep = size; 2542 } else { 2543 sbuf_start_section(sb, &start_len); 2544 2545 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2546 PROC_LOCK(p); 2547 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2548 filedesc_flags); 2549 2550 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2551 if (sect_len < 0) 2552 return; 2553 sect_sz = sect_len; 2554 2555 KASSERT(sect_sz <= *sizep, 2556 ("kern_proc_filedesc_out did not respect maxlen; " 2557 "requested %zu, got %zu", *sizep - sizeof(structsize), 2558 sect_sz - sizeof(structsize))); 2559 2560 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2561 sbuf_putc(sb, 0); 2562 } 2563 } 2564 2565 #ifdef KINFO_VMENTRY_SIZE 2566 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2567 #endif 2568 2569 static void 2570 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2571 { 2572 struct proc *p; 2573 size_t size; 2574 int structsize, vmmap_flags; 2575 2576 if (coredump_pack_vmmapinfo) 2577 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2578 else 2579 vmmap_flags = 0; 2580 2581 p = arg; 2582 structsize = sizeof(struct kinfo_vmentry); 2583 if (sb == NULL) { 2584 size = 0; 2585 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2586 sbuf_set_drain(sb, sbuf_count_drain, &size); 2587 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2588 PROC_LOCK(p); 2589 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2590 sbuf_finish(sb); 2591 sbuf_delete(sb); 2592 *sizep = size; 2593 } else { 2594 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2595 PROC_LOCK(p); 2596 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2597 vmmap_flags); 2598 } 2599 } 2600 2601 static void 2602 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2603 { 2604 struct proc *p; 2605 size_t size; 2606 int structsize; 2607 2608 p = arg; 2609 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2610 if (sb != NULL) { 2611 KASSERT(*sizep == size, ("invalid size")); 2612 structsize = sizeof(gid_t); 2613 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2614 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2615 sizeof(gid_t)); 2616 } 2617 *sizep = size; 2618 } 2619 2620 static void 2621 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2622 { 2623 struct proc *p; 2624 size_t size; 2625 int structsize; 2626 2627 p = arg; 2628 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask); 2629 if (sb != NULL) { 2630 KASSERT(*sizep == size, ("invalid size")); 2631 structsize = sizeof(p->p_pd->pd_cmask); 2632 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2633 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask)); 2634 } 2635 *sizep = size; 2636 } 2637 2638 static void 2639 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2640 { 2641 struct proc *p; 2642 struct rlimit rlim[RLIM_NLIMITS]; 2643 size_t size; 2644 int structsize, i; 2645 2646 p = arg; 2647 size = sizeof(structsize) + sizeof(rlim); 2648 if (sb != NULL) { 2649 KASSERT(*sizep == size, ("invalid size")); 2650 structsize = sizeof(rlim); 2651 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2652 PROC_LOCK(p); 2653 for (i = 0; i < RLIM_NLIMITS; i++) 2654 lim_rlimit_proc(p, i, &rlim[i]); 2655 PROC_UNLOCK(p); 2656 sbuf_bcat(sb, rlim, sizeof(rlim)); 2657 } 2658 *sizep = size; 2659 } 2660 2661 static void 2662 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2663 { 2664 struct proc *p; 2665 size_t size; 2666 int structsize; 2667 2668 p = arg; 2669 size = sizeof(structsize) + sizeof(p->p_osrel); 2670 if (sb != NULL) { 2671 KASSERT(*sizep == size, ("invalid size")); 2672 structsize = sizeof(p->p_osrel); 2673 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2674 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2675 } 2676 *sizep = size; 2677 } 2678 2679 static void 2680 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2681 { 2682 struct proc *p; 2683 elf_ps_strings_t ps_strings; 2684 size_t size; 2685 int structsize; 2686 2687 p = arg; 2688 size = sizeof(structsize) + sizeof(ps_strings); 2689 if (sb != NULL) { 2690 KASSERT(*sizep == size, ("invalid size")); 2691 structsize = sizeof(ps_strings); 2692 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2693 ps_strings = PTROUT(PROC_PS_STRINGS(p)); 2694 #else 2695 ps_strings = PROC_PS_STRINGS(p); 2696 #endif 2697 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2698 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2699 } 2700 *sizep = size; 2701 } 2702 2703 static void 2704 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2705 { 2706 struct proc *p; 2707 size_t size; 2708 int structsize; 2709 2710 p = arg; 2711 if (sb == NULL) { 2712 size = 0; 2713 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo), 2714 SBUF_FIXEDLEN); 2715 sbuf_set_drain(sb, sbuf_count_drain, &size); 2716 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2717 PHOLD(p); 2718 proc_getauxv(curthread, p, sb); 2719 PRELE(p); 2720 sbuf_finish(sb); 2721 sbuf_delete(sb); 2722 *sizep = size; 2723 } else { 2724 structsize = sizeof(Elf_Auxinfo); 2725 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2726 PHOLD(p); 2727 proc_getauxv(curthread, p, sb); 2728 PRELE(p); 2729 } 2730 } 2731 2732 static void 2733 __elfN(note_procstat_kqueues)(void *arg, struct sbuf *sb, size_t *sizep) 2734 { 2735 struct proc *p; 2736 size_t size, sect_sz, i; 2737 ssize_t start_len, sect_len; 2738 int structsize; 2739 bool compat32; 2740 2741 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2742 compat32 = true; 2743 structsize = sizeof(struct kinfo_knote32); 2744 #else 2745 compat32 = false; 2746 structsize = sizeof(struct kinfo_knote); 2747 #endif 2748 p = arg; 2749 if (sb == NULL) { 2750 size = 0; 2751 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2752 sbuf_set_drain(sb, sbuf_count_drain, &size); 2753 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2754 kern_proc_kqueues_out(p, sb, -1, compat32); 2755 sbuf_finish(sb); 2756 sbuf_delete(sb); 2757 *sizep = size; 2758 } else { 2759 sbuf_start_section(sb, &start_len); 2760 2761 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2762 kern_proc_kqueues_out(p, sb, *sizep - sizeof(structsize), 2763 compat32); 2764 2765 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2766 if (sect_len < 0) 2767 return; 2768 sect_sz = sect_len; 2769 2770 KASSERT(sect_sz <= *sizep, 2771 ("kern_proc_kqueue_out did not respect maxlen; " 2772 "requested %zu, got %zu", *sizep - sizeof(structsize), 2773 sect_sz - sizeof(structsize))); 2774 2775 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2776 sbuf_putc(sb, 0); 2777 } 2778 } 2779 2780 #define MAX_NOTES_LOOP 4096 2781 bool 2782 __elfN(parse_notes)(const struct image_params *imgp, const Elf_Note *checknote, 2783 const char *note_vendor, const Elf_Phdr *pnote, 2784 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg) 2785 { 2786 const Elf_Note *note, *note0, *note_end; 2787 const char *note_name; 2788 char *buf; 2789 int i, error; 2790 bool res; 2791 2792 /* We need some limit, might as well use PAGE_SIZE. */ 2793 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2794 return (false); 2795 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2796 if (pnote->p_offset > PAGE_SIZE || 2797 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2798 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT); 2799 if (buf == NULL) { 2800 VOP_UNLOCK(imgp->vp); 2801 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2802 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 2803 } 2804 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2805 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2806 curthread->td_ucred, NOCRED, NULL, curthread); 2807 if (error != 0) { 2808 uprintf("i/o error PT_NOTE\n"); 2809 goto retf; 2810 } 2811 note = note0 = (const Elf_Note *)buf; 2812 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2813 } else { 2814 note = note0 = (const Elf_Note *)(imgp->image_header + 2815 pnote->p_offset); 2816 note_end = (const Elf_Note *)(imgp->image_header + 2817 pnote->p_offset + pnote->p_filesz); 2818 buf = NULL; 2819 } 2820 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end; 2821 i++) { 2822 if (!aligned(note, Elf32_Addr)) { 2823 uprintf("Unaligned ELF note\n"); 2824 goto retf; 2825 } 2826 if ((const char *)note_end - (const char *)note < 2827 sizeof(Elf_Note)) { 2828 uprintf("ELF note to short\n"); 2829 goto retf; 2830 } 2831 if (note->n_namesz != checknote->n_namesz || 2832 note->n_descsz != checknote->n_descsz || 2833 note->n_type != checknote->n_type) 2834 goto nextnote; 2835 note_name = (const char *)(note + 1); 2836 if (note_name + checknote->n_namesz >= 2837 (const char *)note_end || strncmp(note_vendor, 2838 note_name, checknote->n_namesz) != 0) 2839 goto nextnote; 2840 2841 if (cb(note, cb_arg, &res)) 2842 goto ret; 2843 nextnote: 2844 note = (const Elf_Note *)((const char *)(note + 1) + 2845 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2846 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2847 } 2848 if (i >= MAX_NOTES_LOOP) 2849 uprintf("ELF note parser reached %d notes\n", i); 2850 retf: 2851 res = false; 2852 ret: 2853 free(buf, M_TEMP); 2854 return (res); 2855 } 2856 2857 struct brandnote_cb_arg { 2858 Elf_Brandnote *brandnote; 2859 int32_t *osrel; 2860 }; 2861 2862 static bool 2863 brandnote_cb(const Elf_Note *note, void *arg0, bool *res) 2864 { 2865 struct brandnote_cb_arg *arg; 2866 2867 arg = arg0; 2868 2869 /* 2870 * Fetch the osreldate for binary from the ELF OSABI-note if 2871 * necessary. 2872 */ 2873 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 && 2874 arg->brandnote->trans_osrel != NULL ? 2875 arg->brandnote->trans_osrel(note, arg->osrel) : true; 2876 2877 return (true); 2878 } 2879 2880 static Elf_Note fctl_note = { 2881 .n_namesz = sizeof(FREEBSD_ABI_VENDOR), 2882 .n_descsz = sizeof(uint32_t), 2883 .n_type = NT_FREEBSD_FEATURE_CTL, 2884 }; 2885 2886 struct fctl_cb_arg { 2887 bool *has_fctl0; 2888 uint32_t *fctl0; 2889 }; 2890 2891 static bool 2892 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res) 2893 { 2894 struct fctl_cb_arg *arg; 2895 const Elf32_Word *desc; 2896 uintptr_t p; 2897 2898 arg = arg0; 2899 p = (uintptr_t)(note + 1); 2900 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 2901 desc = (const Elf32_Word *)p; 2902 *arg->has_fctl0 = true; 2903 *arg->fctl0 = desc[0]; 2904 *res = true; 2905 return (true); 2906 } 2907 2908 /* 2909 * Try to find the appropriate ABI-note section for checknote, fetch 2910 * the osreldate and feature control flags for binary from the ELF 2911 * OSABI-note. Only the first page of the image is searched, the same 2912 * as for headers. 2913 */ 2914 static bool 2915 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote, 2916 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0) 2917 { 2918 const Elf_Phdr *phdr; 2919 const Elf_Ehdr *hdr; 2920 struct brandnote_cb_arg b_arg; 2921 struct fctl_cb_arg f_arg; 2922 int i, j; 2923 2924 hdr = (const Elf_Ehdr *)imgp->image_header; 2925 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2926 b_arg.brandnote = brandnote; 2927 b_arg.osrel = osrel; 2928 f_arg.has_fctl0 = has_fctl0; 2929 f_arg.fctl0 = fctl0; 2930 2931 for (i = 0; i < hdr->e_phnum; i++) { 2932 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, 2933 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb, 2934 &b_arg)) { 2935 for (j = 0; j < hdr->e_phnum; j++) { 2936 if (phdr[j].p_type == PT_NOTE && 2937 __elfN(parse_notes)(imgp, &fctl_note, 2938 FREEBSD_ABI_VENDOR, &phdr[j], 2939 note_fctl_cb, &f_arg)) 2940 break; 2941 } 2942 return (true); 2943 } 2944 } 2945 return (false); 2946 2947 } 2948 2949 /* 2950 * Tell kern_execve.c about it, with a little help from the linker. 2951 */ 2952 static struct execsw __elfN(execsw) = { 2953 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 2954 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2955 }; 2956 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2957 2958 static vm_prot_t 2959 __elfN(trans_prot)(Elf_Word flags) 2960 { 2961 vm_prot_t prot; 2962 2963 prot = 0; 2964 if (flags & PF_X) 2965 prot |= VM_PROT_EXECUTE; 2966 if (flags & PF_W) 2967 prot |= VM_PROT_WRITE; 2968 if (flags & PF_R) 2969 prot |= VM_PROT_READ; 2970 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 2971 if (i386_read_exec && (flags & PF_R)) 2972 prot |= VM_PROT_EXECUTE; 2973 #endif 2974 return (prot); 2975 } 2976 2977 static Elf_Word 2978 __elfN(untrans_prot)(vm_prot_t prot) 2979 { 2980 Elf_Word flags; 2981 2982 flags = 0; 2983 if (prot & VM_PROT_EXECUTE) 2984 flags |= PF_X; 2985 if (prot & VM_PROT_READ) 2986 flags |= PF_R; 2987 if (prot & VM_PROT_WRITE) 2988 flags |= PF_W; 2989 return (flags); 2990 } 2991