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/vnode.h> 68 #include <sys/syslog.h> 69 #include <sys/eventhandler.h> 70 #include <sys/user.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_kern.h> 74 #include <vm/vm_param.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_object.h> 78 #include <vm/vm_extern.h> 79 80 #include <machine/elf.h> 81 #include <machine/md_var.h> 82 83 #define ELF_NOTE_ROUNDSIZE 4 84 #define OLD_EI_BRAND 8 85 86 static int __elfN(check_header)(const Elf_Ehdr *hdr); 87 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 88 const char *interp, int32_t *osrel, uint32_t *fctl0); 89 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 90 u_long *entry); 91 static int __elfN(load_section)(const struct image_params *imgp, 92 vm_ooffset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 93 vm_prot_t prot); 94 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 95 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note, 96 int32_t *osrel); 97 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 98 static bool __elfN(check_note)(struct image_params *imgp, 99 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0, 100 uint32_t *fctl0); 101 static vm_prot_t __elfN(trans_prot)(Elf_Word); 102 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 103 static size_t __elfN(prepare_register_notes)(struct thread *td, 104 struct note_info_list *list, struct thread *target_td); 105 106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), 107 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 108 ""); 109 110 int __elfN(fallback_brand) = -1; 111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 112 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, 113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 114 115 static int elf_legacy_coredump = 0; 116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 117 &elf_legacy_coredump, 0, 118 "include all and only RW pages in core dumps"); 119 120 int __elfN(nxstack) = 121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \ 122 defined(__arm__) || defined(__aarch64__) || \ 123 defined(__riscv) 124 1; 125 #else 126 0; 127 #endif 128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": support PT_GNU_STACK for non-executable stack control"); 131 132 #if defined(__amd64__) 133 static int __elfN(vdso) = 1; 134 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 135 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0, 136 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading"); 137 #else 138 static int __elfN(vdso) = 0; 139 #endif 140 141 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 142 int i386_read_exec = 0; 143 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 144 "enable execution from readable segments"); 145 #endif 146 147 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR; 148 static int 149 sysctl_pie_base(SYSCTL_HANDLER_ARGS) 150 { 151 u_long val; 152 int error; 153 154 val = __elfN(pie_base); 155 error = sysctl_handle_long(oidp, &val, 0, req); 156 if (error != 0 || req->newptr == NULL) 157 return (error); 158 if ((val & PAGE_MASK) != 0) 159 return (EINVAL); 160 __elfN(pie_base) = val; 161 return (0); 162 } 163 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base, 164 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 165 sysctl_pie_base, "LU", 166 "PIE load base without randomization"); 167 168 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, 169 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 170 ""); 171 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr) 172 173 /* 174 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures 175 * have limited address space (which can cause issues for applications with 176 * high memory use) so we leave it off there. 177 */ 178 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64; 179 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN, 180 &__elfN(aslr_enabled), 0, 181 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 182 ": enable address map randomization"); 183 184 /* 185 * Enable ASLR by default for 64-bit PIE binaries. 186 */ 187 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64; 188 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN, 189 &__elfN(pie_aslr_enabled), 0, 190 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 191 ": enable address map randomization for PIE binaries"); 192 193 /* 194 * Sbrk is deprecated and it can be assumed that in most cases it will not be 195 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR 196 * to use the bss grow region. 197 */ 198 static int __elfN(aslr_honor_sbrk) = 0; 199 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW, 200 &__elfN(aslr_honor_sbrk), 0, 201 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used"); 202 203 static int __elfN(aslr_stack) = __ELF_WORD_SIZE == 64; 204 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN, 205 &__elfN(aslr_stack), 0, 206 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 207 ": enable stack address randomization"); 208 209 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64; 210 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN, 211 &__elfN(aslr_shared_page), 0, 212 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 213 ": enable shared page address randomization"); 214 215 static int __elfN(sigfastblock) = 1; 216 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock, 217 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0, 218 "enable sigfastblock for new processes"); 219 220 static bool __elfN(allow_wx) = true; 221 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx, 222 CTLFLAG_RWTUN, &__elfN(allow_wx), 0, 223 "Allow pages to be mapped simultaneously writable and executable"); 224 225 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 226 227 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a)) 228 229 Elf_Brandnote __elfN(freebsd_brandnote) = { 230 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 231 .hdr.n_descsz = sizeof(int32_t), 232 .hdr.n_type = NT_FREEBSD_ABI_TAG, 233 .vendor = FREEBSD_ABI_VENDOR, 234 .flags = BN_TRANSLATE_OSREL, 235 .trans_osrel = __elfN(freebsd_trans_osrel) 236 }; 237 238 static bool 239 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 240 { 241 uintptr_t p; 242 243 p = (uintptr_t)(note + 1); 244 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 245 *osrel = *(const int32_t *)(p); 246 247 return (true); 248 } 249 250 static int GNU_KFREEBSD_ABI_DESC = 3; 251 252 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 253 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 254 .hdr.n_descsz = 16, /* XXX at least 16 */ 255 .hdr.n_type = 1, 256 .vendor = GNU_ABI_VENDOR, 257 .flags = BN_TRANSLATE_OSREL, 258 .trans_osrel = kfreebsd_trans_osrel 259 }; 260 261 static bool 262 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 263 { 264 const Elf32_Word *desc; 265 uintptr_t p; 266 267 p = (uintptr_t)(note + 1); 268 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 269 270 desc = (const Elf32_Word *)p; 271 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 272 return (false); 273 274 /* 275 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 276 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 277 */ 278 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 279 280 return (true); 281 } 282 283 int 284 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 285 { 286 int i; 287 288 for (i = 0; i < MAX_BRANDS; i++) { 289 if (elf_brand_list[i] == NULL) { 290 elf_brand_list[i] = entry; 291 break; 292 } 293 } 294 if (i == MAX_BRANDS) { 295 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 296 __func__, entry); 297 return (-1); 298 } 299 return (0); 300 } 301 302 int 303 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 304 { 305 int i; 306 307 for (i = 0; i < MAX_BRANDS; i++) { 308 if (elf_brand_list[i] == entry) { 309 elf_brand_list[i] = NULL; 310 break; 311 } 312 } 313 if (i == MAX_BRANDS) 314 return (-1); 315 return (0); 316 } 317 318 bool 319 __elfN(brand_inuse)(Elf_Brandinfo *entry) 320 { 321 struct proc *p; 322 bool rval = false; 323 324 sx_slock(&allproc_lock); 325 FOREACH_PROC_IN_SYSTEM(p) { 326 if (p->p_sysent == entry->sysvec) { 327 rval = true; 328 break; 329 } 330 } 331 sx_sunlock(&allproc_lock); 332 333 return (rval); 334 } 335 336 static Elf_Brandinfo * 337 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 338 int32_t *osrel, uint32_t *fctl0) 339 { 340 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 341 Elf_Brandinfo *bi, *bi_m; 342 bool ret, has_fctl0; 343 int i, interp_name_len; 344 345 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0; 346 347 /* 348 * We support four types of branding -- (1) the ELF EI_OSABI field 349 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 350 * branding w/in the ELF header, (3) path of the `interp_path' 351 * field, and (4) the ".note.ABI-tag" ELF section. 352 */ 353 354 /* Look for an ".note.ABI-tag" ELF section */ 355 bi_m = NULL; 356 for (i = 0; i < MAX_BRANDS; i++) { 357 bi = elf_brand_list[i]; 358 if (bi == NULL) 359 continue; 360 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 361 continue; 362 if (hdr->e_machine == bi->machine && (bi->flags & 363 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 364 has_fctl0 = false; 365 *fctl0 = 0; 366 *osrel = 0; 367 ret = __elfN(check_note)(imgp, bi->brand_note, osrel, 368 &has_fctl0, fctl0); 369 /* Give brand a chance to veto check_note's guess */ 370 if (ret && bi->header_supported) { 371 ret = bi->header_supported(imgp, osrel, 372 has_fctl0 ? fctl0 : NULL); 373 } 374 /* 375 * If note checker claimed the binary, but the 376 * interpreter path in the image does not 377 * match default one for the brand, try to 378 * search for other brands with the same 379 * interpreter. Either there is better brand 380 * with the right interpreter, or, failing 381 * this, we return first brand which accepted 382 * our note and, optionally, header. 383 */ 384 if (ret && bi_m == NULL && interp != NULL && 385 (bi->interp_path == NULL || 386 (strlen(bi->interp_path) + 1 != interp_name_len || 387 strncmp(interp, bi->interp_path, interp_name_len) 388 != 0))) { 389 bi_m = bi; 390 ret = 0; 391 } 392 if (ret) 393 return (bi); 394 } 395 } 396 if (bi_m != NULL) 397 return (bi_m); 398 399 /* If the executable has a brand, search for it in the brand list. */ 400 for (i = 0; i < MAX_BRANDS; i++) { 401 bi = elf_brand_list[i]; 402 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 403 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 404 continue; 405 if (hdr->e_machine == bi->machine && 406 (hdr->e_ident[EI_OSABI] == bi->brand || 407 (bi->compat_3_brand != NULL && 408 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 409 bi->compat_3_brand) == 0))) { 410 /* Looks good, but give brand a chance to veto */ 411 if (bi->header_supported == NULL || 412 bi->header_supported(imgp, NULL, NULL)) { 413 /* 414 * Again, prefer strictly matching 415 * interpreter path. 416 */ 417 if (interp_name_len == 0 && 418 bi->interp_path == NULL) 419 return (bi); 420 if (bi->interp_path != NULL && 421 strlen(bi->interp_path) + 1 == 422 interp_name_len && strncmp(interp, 423 bi->interp_path, interp_name_len) == 0) 424 return (bi); 425 if (bi_m == NULL) 426 bi_m = bi; 427 } 428 } 429 } 430 if (bi_m != NULL) 431 return (bi_m); 432 433 /* No known brand, see if the header is recognized by any brand */ 434 for (i = 0; i < MAX_BRANDS; i++) { 435 bi = elf_brand_list[i]; 436 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 437 bi->header_supported == NULL) 438 continue; 439 if (hdr->e_machine == bi->machine) { 440 ret = bi->header_supported(imgp, NULL, NULL); 441 if (ret) 442 return (bi); 443 } 444 } 445 446 /* Lacking a known brand, search for a recognized interpreter. */ 447 if (interp != NULL) { 448 for (i = 0; i < MAX_BRANDS; i++) { 449 bi = elf_brand_list[i]; 450 if (bi == NULL || (bi->flags & 451 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 452 != 0) 453 continue; 454 if (hdr->e_machine == bi->machine && 455 bi->interp_path != NULL && 456 /* ELF image p_filesz includes terminating zero */ 457 strlen(bi->interp_path) + 1 == interp_name_len && 458 strncmp(interp, bi->interp_path, interp_name_len) 459 == 0 && (bi->header_supported == NULL || 460 bi->header_supported(imgp, NULL, NULL))) 461 return (bi); 462 } 463 } 464 465 /* Lacking a recognized interpreter, try the default brand */ 466 for (i = 0; i < MAX_BRANDS; i++) { 467 bi = elf_brand_list[i]; 468 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 469 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 470 continue; 471 if (hdr->e_machine == bi->machine && 472 __elfN(fallback_brand) == bi->brand && 473 (bi->header_supported == NULL || 474 bi->header_supported(imgp, NULL, NULL))) 475 return (bi); 476 } 477 return (NULL); 478 } 479 480 static bool 481 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr) 482 { 483 return (hdr->e_phoff <= PAGE_SIZE && 484 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff); 485 } 486 487 static int 488 __elfN(check_header)(const Elf_Ehdr *hdr) 489 { 490 Elf_Brandinfo *bi; 491 int i; 492 493 if (!IS_ELF(*hdr) || 494 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 495 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 496 hdr->e_ident[EI_VERSION] != EV_CURRENT || 497 hdr->e_phentsize != sizeof(Elf_Phdr) || 498 hdr->e_version != ELF_TARG_VER) 499 return (ENOEXEC); 500 501 /* 502 * Make sure we have at least one brand for this machine. 503 */ 504 505 for (i = 0; i < MAX_BRANDS; i++) { 506 bi = elf_brand_list[i]; 507 if (bi != NULL && bi->machine == hdr->e_machine) 508 break; 509 } 510 if (i == MAX_BRANDS) 511 return (ENOEXEC); 512 513 return (0); 514 } 515 516 static int 517 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 518 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 519 { 520 struct sf_buf *sf; 521 int error; 522 vm_offset_t off; 523 524 /* 525 * Create the page if it doesn't exist yet. Ignore errors. 526 */ 527 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 528 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 529 530 /* 531 * Find the page from the underlying object. 532 */ 533 if (object != NULL) { 534 sf = vm_imgact_map_page(object, offset); 535 if (sf == NULL) 536 return (KERN_FAILURE); 537 off = offset - trunc_page(offset); 538 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 539 end - start); 540 vm_imgact_unmap_page(sf); 541 if (error != 0) 542 return (KERN_FAILURE); 543 } 544 545 return (KERN_SUCCESS); 546 } 547 548 static int 549 __elfN(map_insert)(const struct image_params *imgp, vm_map_t map, 550 vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, 551 vm_prot_t prot, int cow) 552 { 553 struct sf_buf *sf; 554 vm_offset_t off; 555 vm_size_t sz; 556 int error, locked, rv; 557 558 if (start != trunc_page(start)) { 559 rv = __elfN(map_partial)(map, object, offset, start, 560 round_page(start), prot); 561 if (rv != KERN_SUCCESS) 562 return (rv); 563 offset += round_page(start) - start; 564 start = round_page(start); 565 } 566 if (end != round_page(end)) { 567 rv = __elfN(map_partial)(map, object, offset + 568 trunc_page(end) - start, trunc_page(end), end, prot); 569 if (rv != KERN_SUCCESS) 570 return (rv); 571 end = trunc_page(end); 572 } 573 if (start >= end) 574 return (KERN_SUCCESS); 575 if ((offset & PAGE_MASK) != 0) { 576 /* 577 * The mapping is not page aligned. This means that we have 578 * to copy the data. 579 */ 580 rv = vm_map_fixed(map, NULL, 0, start, end - start, 581 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 582 if (rv != KERN_SUCCESS) 583 return (rv); 584 if (object == NULL) 585 return (KERN_SUCCESS); 586 for (; start < end; start += sz) { 587 sf = vm_imgact_map_page(object, offset); 588 if (sf == NULL) 589 return (KERN_FAILURE); 590 off = offset - trunc_page(offset); 591 sz = end - start; 592 if (sz > PAGE_SIZE - off) 593 sz = PAGE_SIZE - off; 594 error = copyout((caddr_t)sf_buf_kva(sf) + off, 595 (caddr_t)start, sz); 596 vm_imgact_unmap_page(sf); 597 if (error != 0) 598 return (KERN_FAILURE); 599 offset += sz; 600 } 601 } else { 602 vm_object_reference(object); 603 rv = vm_map_fixed(map, object, offset, start, end - start, 604 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL | 605 (object != NULL ? MAP_VN_EXEC : 0)); 606 if (rv != KERN_SUCCESS) { 607 locked = VOP_ISLOCKED(imgp->vp); 608 VOP_UNLOCK(imgp->vp); 609 vm_object_deallocate(object); 610 vn_lock(imgp->vp, locked | LK_RETRY); 611 return (rv); 612 } else if (object != NULL) { 613 MPASS(imgp->vp->v_object == object); 614 VOP_SET_TEXT_CHECKED(imgp->vp); 615 } 616 } 617 return (KERN_SUCCESS); 618 } 619 620 static int 621 __elfN(load_section)(const struct image_params *imgp, vm_ooffset_t offset, 622 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot) 623 { 624 struct sf_buf *sf; 625 size_t map_len; 626 vm_map_t map; 627 vm_object_t object; 628 vm_offset_t map_addr; 629 int error, rv, cow; 630 size_t copy_len; 631 vm_ooffset_t file_addr; 632 633 /* 634 * It's necessary to fail if the filsz + offset taken from the 635 * header is greater than the actual file pager object's size. 636 * If we were to allow this, then the vm_map_find() below would 637 * walk right off the end of the file object and into the ether. 638 * 639 * While I'm here, might as well check for something else that 640 * is invalid: filsz cannot be greater than memsz. 641 */ 642 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 643 filsz > memsz) { 644 uprintf("elf_load_section: truncated ELF file\n"); 645 return (ENOEXEC); 646 } 647 648 object = imgp->object; 649 map = &imgp->proc->p_vmspace->vm_map; 650 map_addr = trunc_page((vm_offset_t)vmaddr); 651 file_addr = trunc_page(offset); 652 653 /* 654 * We have two choices. We can either clear the data in the last page 655 * of an oversized mapping, or we can start the anon mapping a page 656 * early and copy the initialized data into that first page. We 657 * choose the second. 658 */ 659 if (filsz == 0) 660 map_len = 0; 661 else if (memsz > filsz) 662 map_len = trunc_page(offset + filsz) - file_addr; 663 else 664 map_len = round_page(offset + filsz) - file_addr; 665 666 if (map_len != 0) { 667 /* cow flags: don't dump readonly sections in core */ 668 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 669 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 670 671 rv = __elfN(map_insert)(imgp, map, object, file_addr, 672 map_addr, map_addr + map_len, prot, cow); 673 if (rv != KERN_SUCCESS) 674 return (EINVAL); 675 676 /* we can stop now if we've covered it all */ 677 if (memsz == filsz) 678 return (0); 679 } 680 681 /* 682 * We have to get the remaining bit of the file into the first part 683 * of the oversized map segment. This is normally because the .data 684 * segment in the file is extended to provide bss. It's a neat idea 685 * to try and save a page, but it's a pain in the behind to implement. 686 */ 687 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset + 688 filsz); 689 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 690 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 691 692 /* This had damn well better be true! */ 693 if (map_len != 0) { 694 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 695 map_addr + map_len, prot, 0); 696 if (rv != KERN_SUCCESS) 697 return (EINVAL); 698 } 699 700 if (copy_len != 0) { 701 sf = vm_imgact_map_page(object, offset + filsz); 702 if (sf == NULL) 703 return (EIO); 704 705 /* send the page fragment to user space */ 706 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr, 707 copy_len); 708 vm_imgact_unmap_page(sf); 709 if (error != 0) 710 return (error); 711 } 712 713 /* 714 * Remove write access to the page if it was only granted by map_insert 715 * to allow copyout. 716 */ 717 if ((prot & VM_PROT_WRITE) == 0) 718 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 719 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT); 720 721 return (0); 722 } 723 724 static int 725 __elfN(load_sections)(const struct image_params *imgp, const Elf_Ehdr *hdr, 726 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp) 727 { 728 vm_prot_t prot; 729 u_long base_addr; 730 bool first; 731 int error, i; 732 733 ASSERT_VOP_LOCKED(imgp->vp, __func__); 734 735 base_addr = 0; 736 first = true; 737 738 for (i = 0; i < hdr->e_phnum; i++) { 739 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 740 continue; 741 742 /* Loadable segment */ 743 prot = __elfN(trans_prot)(phdr[i].p_flags); 744 error = __elfN(load_section)(imgp, phdr[i].p_offset, 745 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 746 phdr[i].p_memsz, phdr[i].p_filesz, prot); 747 if (error != 0) 748 return (error); 749 750 /* 751 * Establish the base address if this is the first segment. 752 */ 753 if (first) { 754 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 755 first = false; 756 } 757 } 758 759 if (base_addrp != NULL) 760 *base_addrp = base_addr; 761 762 return (0); 763 } 764 765 /* 766 * Load the file "file" into memory. It may be either a shared object 767 * or an executable. 768 * 769 * The "addr" reference parameter is in/out. On entry, it specifies 770 * the address where a shared object should be loaded. If the file is 771 * an executable, this value is ignored. On exit, "addr" specifies 772 * where the file was actually loaded. 773 * 774 * The "entry" reference parameter is out only. On exit, it specifies 775 * the entry point for the loaded file. 776 */ 777 static int 778 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 779 u_long *entry) 780 { 781 struct { 782 struct nameidata nd; 783 struct vattr attr; 784 struct image_params image_params; 785 } *tempdata; 786 const Elf_Ehdr *hdr = NULL; 787 const Elf_Phdr *phdr = NULL; 788 struct nameidata *nd; 789 struct vattr *attr; 790 struct image_params *imgp; 791 u_long rbase; 792 u_long base_addr = 0; 793 int error; 794 795 #ifdef CAPABILITY_MODE 796 /* 797 * XXXJA: This check can go away once we are sufficiently confident 798 * that the checks in namei() are correct. 799 */ 800 if (IN_CAPABILITY_MODE(curthread)) 801 return (ECAPMODE); 802 #endif 803 804 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO); 805 nd = &tempdata->nd; 806 attr = &tempdata->attr; 807 imgp = &tempdata->image_params; 808 809 /* 810 * Initialize part of the common data 811 */ 812 imgp->proc = p; 813 imgp->attr = attr; 814 815 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF, 816 UIO_SYSSPACE, file); 817 if ((error = namei(nd)) != 0) { 818 nd->ni_vp = NULL; 819 goto fail; 820 } 821 NDFREE_PNBUF(nd); 822 imgp->vp = nd->ni_vp; 823 824 /* 825 * Check permissions, modes, uid, etc on the file, and "open" it. 826 */ 827 error = exec_check_permissions(imgp); 828 if (error) 829 goto fail; 830 831 error = exec_map_first_page(imgp); 832 if (error) 833 goto fail; 834 835 imgp->object = nd->ni_vp->v_object; 836 837 hdr = (const Elf_Ehdr *)imgp->image_header; 838 if ((error = __elfN(check_header)(hdr)) != 0) 839 goto fail; 840 if (hdr->e_type == ET_DYN) 841 rbase = *addr; 842 else if (hdr->e_type == ET_EXEC) 843 rbase = 0; 844 else { 845 error = ENOEXEC; 846 goto fail; 847 } 848 849 /* Only support headers that fit within first page for now */ 850 if (!__elfN(phdr_in_zero_page)(hdr)) { 851 error = ENOEXEC; 852 goto fail; 853 } 854 855 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 856 if (!aligned(phdr, Elf_Addr)) { 857 error = ENOEXEC; 858 goto fail; 859 } 860 861 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr); 862 if (error != 0) 863 goto fail; 864 865 if (p->p_sysent->sv_protect != NULL) 866 p->p_sysent->sv_protect(imgp, SVP_INTERP); 867 868 *addr = base_addr; 869 *entry = (unsigned long)hdr->e_entry + rbase; 870 871 fail: 872 if (imgp->firstpage) 873 exec_unmap_first_page(imgp); 874 875 if (nd->ni_vp) { 876 if (imgp->textset) 877 VOP_UNSET_TEXT_CHECKED(nd->ni_vp); 878 vput(nd->ni_vp); 879 } 880 free(tempdata, M_TEMP); 881 882 return (error); 883 } 884 885 /* 886 * Select randomized valid address in the map map, between minv and 887 * maxv, with specified alignment. The [minv, maxv) range must belong 888 * to the map. Note that function only allocates the address, it is 889 * up to caller to clamp maxv in a way that the final allocation 890 * length fit into the map. 891 * 892 * Result is returned in *resp, error code indicates that arguments 893 * did not pass sanity checks for overflow and range correctness. 894 */ 895 static int 896 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv, 897 u_int align, u_long *resp) 898 { 899 u_long rbase, res; 900 901 MPASS(vm_map_min(map) <= minv); 902 903 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) { 904 uprintf("Invalid ELF segments layout\n"); 905 return (ENOEXEC); 906 } 907 908 arc4rand(&rbase, sizeof(rbase), 0); 909 res = roundup(minv, (u_long)align) + rbase % (maxv - minv); 910 res &= ~((u_long)align - 1); 911 if (res >= maxv) 912 res -= align; 913 914 KASSERT(res >= minv, 915 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx", 916 res, minv, maxv, rbase)); 917 KASSERT(res < maxv, 918 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx", 919 res, maxv, minv, rbase)); 920 921 *resp = res; 922 return (0); 923 } 924 925 static int 926 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr, 927 const Elf_Phdr *phdr) 928 { 929 struct vmspace *vmspace; 930 const char *err_str; 931 u_long text_size, data_size, total_size, text_addr, data_addr; 932 u_long seg_size, seg_addr; 933 int i; 934 935 err_str = NULL; 936 text_size = data_size = total_size = text_addr = data_addr = 0; 937 938 for (i = 0; i < hdr->e_phnum; i++) { 939 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 940 continue; 941 942 seg_addr = trunc_page(phdr[i].p_vaddr + imgp->et_dyn_addr); 943 seg_size = round_page(phdr[i].p_memsz + 944 phdr[i].p_vaddr + imgp->et_dyn_addr - seg_addr); 945 946 /* 947 * Make the largest executable segment the official 948 * text segment and all others data. 949 * 950 * Note that obreak() assumes that data_addr + data_size == end 951 * of data load area, and the ELF file format expects segments 952 * to be sorted by address. If multiple data segments exist, 953 * the last one will be used. 954 */ 955 956 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) { 957 text_size = seg_size; 958 text_addr = seg_addr; 959 } else { 960 data_size = seg_size; 961 data_addr = seg_addr; 962 } 963 total_size += seg_size; 964 } 965 966 if (data_addr == 0 && data_size == 0) { 967 data_addr = text_addr; 968 data_size = text_size; 969 } 970 971 /* 972 * Check limits. It should be safe to check the 973 * limits after loading the segments since we do 974 * not actually fault in all the segments pages. 975 */ 976 PROC_LOCK(imgp->proc); 977 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 978 err_str = "Data segment size exceeds process limit"; 979 else if (text_size > maxtsiz) 980 err_str = "Text segment size exceeds system limit"; 981 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 982 err_str = "Total segment size exceeds process limit"; 983 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 984 err_str = "Data segment size exceeds resource limit"; 985 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 986 err_str = "Total segment size exceeds resource limit"; 987 PROC_UNLOCK(imgp->proc); 988 if (err_str != NULL) { 989 uprintf("%s\n", err_str); 990 return (ENOMEM); 991 } 992 993 vmspace = imgp->proc->p_vmspace; 994 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 995 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 996 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 997 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 998 999 return (0); 1000 } 1001 1002 static int 1003 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr, 1004 char **interpp, bool *free_interpp) 1005 { 1006 struct thread *td; 1007 char *interp; 1008 int error, interp_name_len; 1009 1010 KASSERT(phdr->p_type == PT_INTERP, 1011 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type)); 1012 ASSERT_VOP_LOCKED(imgp->vp, __func__); 1013 1014 td = curthread; 1015 1016 /* Path to interpreter */ 1017 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) { 1018 uprintf("Invalid PT_INTERP\n"); 1019 return (ENOEXEC); 1020 } 1021 1022 interp_name_len = phdr->p_filesz; 1023 if (phdr->p_offset > PAGE_SIZE || 1024 interp_name_len > PAGE_SIZE - phdr->p_offset) { 1025 /* 1026 * The vnode lock might be needed by the pagedaemon to 1027 * clean pages owned by the vnode. Do not allow sleep 1028 * waiting for memory with the vnode locked, instead 1029 * try non-sleepable allocation first, and if it 1030 * fails, go to the slow path were we drop the lock 1031 * and do M_WAITOK. A text reference prevents 1032 * modifications to the vnode content. 1033 */ 1034 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT); 1035 if (interp == NULL) { 1036 VOP_UNLOCK(imgp->vp); 1037 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK); 1038 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1039 } 1040 1041 error = vn_rdwr(UIO_READ, imgp->vp, interp, 1042 interp_name_len, phdr->p_offset, 1043 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 1044 NOCRED, NULL, td); 1045 if (error != 0) { 1046 free(interp, M_TEMP); 1047 uprintf("i/o error PT_INTERP %d\n", error); 1048 return (error); 1049 } 1050 interp[interp_name_len] = '\0'; 1051 1052 *interpp = interp; 1053 *free_interpp = true; 1054 return (0); 1055 } 1056 1057 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset; 1058 if (interp[interp_name_len - 1] != '\0') { 1059 uprintf("Invalid PT_INTERP\n"); 1060 return (ENOEXEC); 1061 } 1062 1063 *interpp = interp; 1064 *free_interpp = false; 1065 return (0); 1066 } 1067 1068 static int 1069 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info, 1070 const char *interp, u_long *addr, u_long *entry) 1071 { 1072 int error; 1073 1074 if (brand_info->interp_newpath != NULL && 1075 (brand_info->interp_path == NULL || 1076 strcmp(interp, brand_info->interp_path) == 0)) { 1077 error = __elfN(load_file)(imgp->proc, 1078 brand_info->interp_newpath, addr, entry); 1079 if (error == 0) 1080 return (0); 1081 } 1082 1083 error = __elfN(load_file)(imgp->proc, interp, addr, entry); 1084 if (error == 0) 1085 return (0); 1086 1087 uprintf("ELF interpreter %s not found, error %d\n", interp, error); 1088 return (error); 1089 } 1090 1091 /* 1092 * Impossible et_dyn_addr initial value indicating that the real base 1093 * must be calculated later with some randomization applied. 1094 */ 1095 #define ET_DYN_ADDR_RAND 1 1096 1097 static int 1098 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 1099 { 1100 struct thread *td; 1101 const Elf_Ehdr *hdr; 1102 const Elf_Phdr *phdr; 1103 Elf_Auxargs *elf_auxargs; 1104 struct vmspace *vmspace; 1105 vm_map_t map; 1106 char *interp; 1107 Elf_Brandinfo *brand_info; 1108 struct sysentvec *sv; 1109 u_long addr, baddr, entry, proghdr; 1110 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc; 1111 uint32_t fctl0; 1112 int32_t osrel; 1113 bool free_interp; 1114 int error, i, n; 1115 1116 hdr = (const Elf_Ehdr *)imgp->image_header; 1117 1118 /* 1119 * Do we have a valid ELF header ? 1120 * 1121 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 1122 * if particular brand doesn't support it. 1123 */ 1124 if (__elfN(check_header)(hdr) != 0 || 1125 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 1126 return (-1); 1127 1128 /* 1129 * From here on down, we return an errno, not -1, as we've 1130 * detected an ELF file. 1131 */ 1132 1133 if (!__elfN(phdr_in_zero_page)(hdr)) { 1134 uprintf("Program headers not in the first page\n"); 1135 return (ENOEXEC); 1136 } 1137 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1138 if (!aligned(phdr, Elf_Addr)) { 1139 uprintf("Unaligned program headers\n"); 1140 return (ENOEXEC); 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 td = curthread; 1151 1152 /* 1153 * Somewhat arbitrary, limit accepted max alignment for the 1154 * loadable segment to the max supported superpage size. Too 1155 * large alignment requests are not useful and are indicators 1156 * of corrupted or outright malicious binary. 1157 */ 1158 maxalign = PAGE_SIZE; 1159 maxsalign = PAGE_SIZE * 1024; 1160 for (i = MAXPAGESIZES - 1; i > 0; i--) { 1161 if (pagesizes[i] > maxsalign) { 1162 maxsalign = pagesizes[i]; 1163 break; 1164 } 1165 } 1166 1167 mapsz = 0; 1168 1169 for (i = 0; i < hdr->e_phnum; i++) { 1170 switch (phdr[i].p_type) { 1171 case PT_LOAD: 1172 if (n == 0) 1173 baddr = phdr[i].p_vaddr; 1174 if (!powerof2(phdr[i].p_align) || 1175 phdr[i].p_align > maxsalign) { 1176 uprintf("Invalid segment alignment\n"); 1177 error = ENOEXEC; 1178 goto ret; 1179 } 1180 if (phdr[i].p_align > maxalign) 1181 maxalign = phdr[i].p_align; 1182 if (mapsz + phdr[i].p_memsz < mapsz) { 1183 uprintf("Mapsize overflow\n"); 1184 error = ENOEXEC; 1185 goto ret; 1186 } 1187 mapsz += phdr[i].p_memsz; 1188 n++; 1189 1190 /* 1191 * If this segment contains the program headers, 1192 * remember their virtual address for the AT_PHDR 1193 * aux entry. Static binaries don't usually include 1194 * a PT_PHDR entry. 1195 */ 1196 if (phdr[i].p_offset == 0 && 1197 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= 1198 phdr[i].p_filesz) 1199 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 1200 break; 1201 case PT_INTERP: 1202 /* Path to interpreter */ 1203 if (interp != NULL) { 1204 uprintf("Multiple PT_INTERP headers\n"); 1205 error = ENOEXEC; 1206 goto ret; 1207 } 1208 error = __elfN(get_interp)(imgp, &phdr[i], &interp, 1209 &free_interp); 1210 if (error != 0) 1211 goto ret; 1212 break; 1213 case PT_GNU_STACK: 1214 if (__elfN(nxstack)) { 1215 imgp->stack_prot = 1216 __elfN(trans_prot)(phdr[i].p_flags); 1217 if ((imgp->stack_prot & VM_PROT_RW) != 1218 VM_PROT_RW) { 1219 uprintf("Invalid PT_GNU_STACK\n"); 1220 error = ENOEXEC; 1221 goto ret; 1222 } 1223 } 1224 imgp->stack_sz = phdr[i].p_memsz; 1225 break; 1226 case PT_PHDR: /* Program header table info */ 1227 proghdr = phdr[i].p_vaddr; 1228 break; 1229 } 1230 } 1231 1232 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0); 1233 if (brand_info == NULL) { 1234 uprintf("ELF binary type \"%u\" not known.\n", 1235 hdr->e_ident[EI_OSABI]); 1236 error = ENOEXEC; 1237 goto ret; 1238 } 1239 sv = brand_info->sysvec; 1240 if (hdr->e_type == ET_DYN) { 1241 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 1242 uprintf("Cannot execute shared object\n"); 1243 error = ENOEXEC; 1244 goto ret; 1245 } 1246 /* 1247 * Honour the base load address from the dso if it is 1248 * non-zero for some reason. 1249 */ 1250 if (baddr == 0) { 1251 if ((sv->sv_flags & SV_ASLR) == 0 || 1252 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) 1253 imgp->et_dyn_addr = __elfN(pie_base); 1254 else if ((__elfN(pie_aslr_enabled) && 1255 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) || 1256 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0) 1257 imgp->et_dyn_addr = ET_DYN_ADDR_RAND; 1258 else 1259 imgp->et_dyn_addr = __elfN(pie_base); 1260 } 1261 } 1262 1263 /* 1264 * Avoid a possible deadlock if the current address space is destroyed 1265 * and that address space maps the locked vnode. In the common case, 1266 * the locked vnode's v_usecount is decremented but remains greater 1267 * than zero. Consequently, the vnode lock is not needed by vrele(). 1268 * However, in cases where the vnode lock is external, such as nullfs, 1269 * v_usecount may become zero. 1270 * 1271 * The VV_TEXT flag prevents modifications to the executable while 1272 * the vnode is unlocked. 1273 */ 1274 VOP_UNLOCK(imgp->vp); 1275 1276 /* 1277 * Decide whether to enable randomization of user mappings. 1278 * First, reset user preferences for the setid binaries. 1279 * Then, account for the support of the randomization by the 1280 * ABI, by user preferences, and make special treatment for 1281 * PIE binaries. 1282 */ 1283 if (imgp->credential_setid) { 1284 PROC_LOCK(imgp->proc); 1285 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE | 1286 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 1287 PROC_UNLOCK(imgp->proc); 1288 } 1289 if ((sv->sv_flags & SV_ASLR) == 0 || 1290 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 || 1291 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) { 1292 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND, 1293 ("imgp->et_dyn_addr == RAND and !ASLR")); 1294 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 || 1295 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) || 1296 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1297 imgp->map_flags |= MAP_ASLR; 1298 /* 1299 * If user does not care about sbrk, utilize the bss 1300 * grow region for mappings as well. We can select 1301 * the base for the image anywere and still not suffer 1302 * from the fragmentation. 1303 */ 1304 if (!__elfN(aslr_honor_sbrk) || 1305 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0) 1306 imgp->map_flags |= MAP_ASLR_IGNSTART; 1307 if (__elfN(aslr_stack)) 1308 imgp->map_flags |= MAP_ASLR_STACK; 1309 if (__elfN(aslr_shared_page)) 1310 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE; 1311 } 1312 1313 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 && 1314 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) || 1315 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0) 1316 imgp->map_flags |= MAP_WXORX; 1317 1318 error = exec_new_vmspace(imgp, sv); 1319 1320 imgp->proc->p_sysent = sv; 1321 imgp->proc->p_elf_brandinfo = brand_info; 1322 1323 vmspace = imgp->proc->p_vmspace; 1324 map = &vmspace->vm_map; 1325 maxv = sv->sv_usrstack; 1326 if ((imgp->map_flags & MAP_ASLR_STACK) == 0) 1327 maxv -= lim_max(td, RLIMIT_STACK); 1328 if (error == 0 && mapsz >= maxv - vm_map_min(map)) { 1329 uprintf("Excessive mapping size\n"); 1330 error = ENOEXEC; 1331 } 1332 1333 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1334 KASSERT((map->flags & MAP_ASLR) != 0, 1335 ("ET_DYN_ADDR_RAND but !MAP_ASLR")); 1336 error = __CONCAT(rnd_, __elfN(base))(map, 1337 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA), 1338 /* reserve half of the address space to interpreter */ 1339 maxv / 2, maxalign, &imgp->et_dyn_addr); 1340 } 1341 1342 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1343 if (error != 0) 1344 goto ret; 1345 1346 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL); 1347 if (error != 0) 1348 goto ret; 1349 1350 error = __elfN(enforce_limits)(imgp, hdr, phdr); 1351 if (error != 0) 1352 goto ret; 1353 1354 /* 1355 * We load the dynamic linker where a userland call 1356 * to mmap(0, ...) would put it. The rationale behind this 1357 * calculation is that it leaves room for the heap to grow to 1358 * its maximum allowed size. 1359 */ 1360 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1361 RLIMIT_DATA)); 1362 if ((map->flags & MAP_ASLR) != 0) { 1363 maxv1 = maxv / 2 + addr / 2; 1364 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1, 1365 #if VM_NRESERVLEVEL > 0 1366 pagesizes[VM_NRESERVLEVEL] != 0 ? 1367 /* Align anon_loc to the largest superpage size. */ 1368 pagesizes[VM_NRESERVLEVEL] : 1369 #endif 1370 pagesizes[0], &anon_loc); 1371 if (error != 0) 1372 goto ret; 1373 map->anon_loc = anon_loc; 1374 } else { 1375 map->anon_loc = addr; 1376 } 1377 1378 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr; 1379 imgp->entry_addr = entry; 1380 1381 if (sv->sv_protect != NULL) 1382 sv->sv_protect(imgp, SVP_IMAGE); 1383 1384 if (interp != NULL) { 1385 VOP_UNLOCK(imgp->vp); 1386 if ((map->flags & MAP_ASLR) != 0) { 1387 /* Assume that interpreter fits into 1/4 of AS */ 1388 maxv1 = maxv / 2 + addr / 2; 1389 error = __CONCAT(rnd_, __elfN(base))(map, addr, 1390 maxv1, PAGE_SIZE, &addr); 1391 } 1392 if (error == 0) { 1393 error = __elfN(load_interp)(imgp, brand_info, interp, 1394 &addr, &imgp->entry_addr); 1395 } 1396 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1397 if (error != 0) 1398 goto ret; 1399 } else 1400 addr = imgp->et_dyn_addr; 1401 1402 error = exec_map_stack(imgp); 1403 if (error != 0) 1404 goto ret; 1405 1406 /* 1407 * Construct auxargs table (used by the copyout_auxargs routine) 1408 */ 1409 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT); 1410 if (elf_auxargs == NULL) { 1411 VOP_UNLOCK(imgp->vp); 1412 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1413 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1414 } 1415 elf_auxargs->execfd = -1; 1416 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr; 1417 elf_auxargs->phent = hdr->e_phentsize; 1418 elf_auxargs->phnum = hdr->e_phnum; 1419 elf_auxargs->pagesz = PAGE_SIZE; 1420 elf_auxargs->base = addr; 1421 elf_auxargs->flags = 0; 1422 elf_auxargs->entry = entry; 1423 elf_auxargs->hdr_eflags = hdr->e_flags; 1424 1425 imgp->auxargs = elf_auxargs; 1426 imgp->interpreted = 0; 1427 imgp->reloc_base = addr; 1428 imgp->proc->p_osrel = osrel; 1429 imgp->proc->p_fctl0 = fctl0; 1430 imgp->proc->p_elf_flags = hdr->e_flags; 1431 1432 ret: 1433 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock"); 1434 if (free_interp) 1435 free(interp, M_TEMP); 1436 return (error); 1437 } 1438 1439 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE) 1440 1441 int 1442 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base) 1443 { 1444 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1445 Elf_Auxinfo *argarray, *pos; 1446 struct vmspace *vmspace; 1447 rlim_t stacksz; 1448 int error, oc; 1449 uint32_t bsdflags; 1450 1451 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1452 M_WAITOK | M_ZERO); 1453 1454 vmspace = imgp->proc->p_vmspace; 1455 1456 if (args->execfd != -1) 1457 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1458 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1459 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1460 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1461 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1462 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1463 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1464 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1465 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1466 if (imgp->execpathp != 0) 1467 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp); 1468 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1469 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1470 if (imgp->canary != 0) { 1471 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary); 1472 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1473 } 1474 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1475 if (imgp->pagesizes != 0) { 1476 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes); 1477 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1478 } 1479 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) { 1480 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1481 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset); 1482 } 1483 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1484 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1485 imgp->sysent->sv_stackprot); 1486 if (imgp->sysent->sv_hwcap != NULL) 1487 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1488 if (imgp->sysent->sv_hwcap2 != NULL) 1489 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1490 bsdflags = 0; 1491 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0; 1492 oc = atomic_load_int(&vm_overcommit); 1493 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) != 1494 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0; 1495 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags); 1496 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc); 1497 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv); 1498 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc); 1499 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv); 1500 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings); 1501 #ifdef RANDOM_FENESTRASX 1502 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) { 1503 AUXARGS_ENTRY(pos, AT_FXRNG, 1504 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset); 1505 } 1506 #endif 1507 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) { 1508 AUXARGS_ENTRY(pos, AT_KPRELOAD, 1509 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset); 1510 } 1511 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop)); 1512 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur; 1513 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz); 1514 AUXARGS_ENTRY(pos, AT_NULL, 0); 1515 1516 free(imgp->auxargs, M_TEMP); 1517 imgp->auxargs = NULL; 1518 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1519 1520 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT); 1521 free(argarray, M_TEMP); 1522 return (error); 1523 } 1524 1525 int 1526 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp) 1527 { 1528 Elf_Addr *base; 1529 1530 base = (Elf_Addr *)*stack_base; 1531 base--; 1532 if (elf_suword(base, imgp->args->argc) == -1) 1533 return (EFAULT); 1534 *stack_base = (uintptr_t)base; 1535 return (0); 1536 } 1537 1538 /* 1539 * Code for generating ELF core dumps. 1540 */ 1541 1542 typedef void (*segment_callback)(vm_map_entry_t, void *); 1543 1544 /* Closure for cb_put_phdr(). */ 1545 struct phdr_closure { 1546 Elf_Phdr *phdr; /* Program header to fill in */ 1547 Elf_Off offset; /* Offset of segment in core file */ 1548 }; 1549 1550 struct note_info { 1551 int type; /* Note type. */ 1552 struct regset *regset; /* Register set. */ 1553 outfunc_t outfunc; /* Output function. */ 1554 void *outarg; /* Argument for the output function. */ 1555 size_t outsize; /* Output size. */ 1556 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1557 }; 1558 1559 TAILQ_HEAD(note_info_list, note_info); 1560 1561 extern int compress_user_cores; 1562 extern int compress_user_cores_level; 1563 1564 static void cb_put_phdr(vm_map_entry_t, void *); 1565 static void cb_size_segment(vm_map_entry_t, void *); 1566 static void each_dumpable_segment(struct thread *, segment_callback, void *, 1567 int); 1568 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1569 struct note_info_list *, size_t, int); 1570 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *); 1571 1572 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1573 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1574 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1575 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1576 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1577 static void __elfN(note_procstat_kqueues)(void *, struct sbuf *, size_t *); 1578 static void note_procstat_files(void *, struct sbuf *, size_t *); 1579 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1580 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1581 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1582 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1583 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1584 1585 static int 1586 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1587 { 1588 1589 return (core_write((struct coredump_params *)arg, base, len, offset, 1590 UIO_SYSSPACE, NULL)); 1591 } 1592 1593 int 1594 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1595 { 1596 struct ucred *cred = td->td_ucred; 1597 int compm, error = 0; 1598 struct sseg_closure seginfo; 1599 struct note_info_list notelst; 1600 struct coredump_params params; 1601 struct note_info *ninfo; 1602 void *hdr, *tmpbuf; 1603 size_t hdrsize, notesz, coresize; 1604 1605 hdr = NULL; 1606 tmpbuf = NULL; 1607 TAILQ_INIT(¬elst); 1608 1609 /* Size the program segments. */ 1610 __elfN(size_segments)(td, &seginfo, flags); 1611 1612 /* 1613 * Collect info about the core file header area. 1614 */ 1615 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1616 if (seginfo.count + 1 >= PN_XNUM) 1617 hdrsize += sizeof(Elf_Shdr); 1618 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz); 1619 coresize = round_page(hdrsize + notesz) + seginfo.size; 1620 1621 /* Set up core dump parameters. */ 1622 params.offset = 0; 1623 params.active_cred = cred; 1624 params.file_cred = NOCRED; 1625 params.td = td; 1626 params.vp = vp; 1627 params.comp = NULL; 1628 1629 #ifdef RACCT 1630 if (racct_enable) { 1631 PROC_LOCK(td->td_proc); 1632 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1633 PROC_UNLOCK(td->td_proc); 1634 if (error != 0) { 1635 error = EFAULT; 1636 goto done; 1637 } 1638 } 1639 #endif 1640 if (coresize >= limit) { 1641 error = EFAULT; 1642 goto done; 1643 } 1644 1645 /* Create a compression stream if necessary. */ 1646 compm = compress_user_cores; 1647 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP && 1648 compm == 0) 1649 compm = COMPRESS_GZIP; 1650 if (compm != 0) { 1651 params.comp = compressor_init(core_compressed_write, 1652 compm, CORE_BUF_SIZE, 1653 compress_user_cores_level, ¶ms); 1654 if (params.comp == NULL) { 1655 error = EFAULT; 1656 goto done; 1657 } 1658 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1659 } 1660 1661 /* 1662 * Allocate memory for building the header, fill it up, 1663 * and write it out following the notes. 1664 */ 1665 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1666 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1667 notesz, flags); 1668 1669 /* Write the contents of all of the writable segments. */ 1670 if (error == 0) { 1671 Elf_Phdr *php; 1672 off_t offset; 1673 int i; 1674 1675 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1676 offset = round_page(hdrsize + notesz); 1677 for (i = 0; i < seginfo.count; i++) { 1678 error = core_output((char *)(uintptr_t)php->p_vaddr, 1679 php->p_filesz, offset, ¶ms, tmpbuf); 1680 if (error != 0) 1681 break; 1682 offset += php->p_filesz; 1683 php++; 1684 } 1685 if (error == 0 && params.comp != NULL) 1686 error = compressor_flush(params.comp); 1687 } 1688 if (error) { 1689 log(LOG_WARNING, 1690 "Failed to write core file for process %s (error %d)\n", 1691 curproc->p_comm, error); 1692 } 1693 1694 done: 1695 free(tmpbuf, M_TEMP); 1696 if (params.comp != NULL) 1697 compressor_fini(params.comp); 1698 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1699 TAILQ_REMOVE(¬elst, ninfo, link); 1700 free(ninfo, M_TEMP); 1701 } 1702 if (hdr != NULL) 1703 free(hdr, M_TEMP); 1704 1705 return (error); 1706 } 1707 1708 /* 1709 * A callback for each_dumpable_segment() to write out the segment's 1710 * program header entry. 1711 */ 1712 static void 1713 cb_put_phdr(vm_map_entry_t entry, void *closure) 1714 { 1715 struct phdr_closure *phc = (struct phdr_closure *)closure; 1716 Elf_Phdr *phdr = phc->phdr; 1717 1718 phc->offset = round_page(phc->offset); 1719 1720 phdr->p_type = PT_LOAD; 1721 phdr->p_offset = phc->offset; 1722 phdr->p_vaddr = entry->start; 1723 phdr->p_paddr = 0; 1724 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1725 phdr->p_align = PAGE_SIZE; 1726 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1727 1728 phc->offset += phdr->p_filesz; 1729 phc->phdr++; 1730 } 1731 1732 /* 1733 * A callback for each_dumpable_segment() to gather information about 1734 * the number of segments and their total size. 1735 */ 1736 static void 1737 cb_size_segment(vm_map_entry_t entry, void *closure) 1738 { 1739 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1740 1741 ssc->count++; 1742 ssc->size += entry->end - entry->start; 1743 } 1744 1745 void 1746 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo, 1747 int flags) 1748 { 1749 seginfo->count = 0; 1750 seginfo->size = 0; 1751 1752 each_dumpable_segment(td, cb_size_segment, seginfo, flags); 1753 } 1754 1755 /* 1756 * For each writable segment in the process's memory map, call the given 1757 * function with a pointer to the map entry and some arbitrary 1758 * caller-supplied data. 1759 */ 1760 static void 1761 each_dumpable_segment(struct thread *td, segment_callback func, void *closure, 1762 int flags) 1763 { 1764 struct proc *p = td->td_proc; 1765 vm_map_t map = &p->p_vmspace->vm_map; 1766 vm_map_entry_t entry; 1767 vm_object_t backing_object, object; 1768 bool ignore_entry; 1769 1770 vm_map_lock_read(map); 1771 VM_MAP_ENTRY_FOREACH(entry, map) { 1772 /* 1773 * Don't dump inaccessible mappings, deal with legacy 1774 * coredump mode. 1775 * 1776 * Note that read-only segments related to the elf binary 1777 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1778 * need to arbitrarily ignore such segments. 1779 */ 1780 if ((flags & SVC_ALL) == 0) { 1781 if (elf_legacy_coredump) { 1782 if ((entry->protection & VM_PROT_RW) != 1783 VM_PROT_RW) 1784 continue; 1785 } else { 1786 if ((entry->protection & VM_PROT_ALL) == 0) 1787 continue; 1788 } 1789 } 1790 1791 /* 1792 * Dont include memory segment in the coredump if 1793 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1794 * madvise(2). Do not dump submaps (i.e. parts of the 1795 * kernel map). 1796 */ 1797 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 1798 continue; 1799 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 && 1800 (flags & SVC_ALL) == 0) 1801 continue; 1802 if ((object = entry->object.vm_object) == NULL) 1803 continue; 1804 1805 /* Ignore memory-mapped devices and such things. */ 1806 VM_OBJECT_RLOCK(object); 1807 while ((backing_object = object->backing_object) != NULL) { 1808 VM_OBJECT_RLOCK(backing_object); 1809 VM_OBJECT_RUNLOCK(object); 1810 object = backing_object; 1811 } 1812 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0; 1813 VM_OBJECT_RUNLOCK(object); 1814 if (ignore_entry) 1815 continue; 1816 1817 (*func)(entry, closure); 1818 } 1819 vm_map_unlock_read(map); 1820 } 1821 1822 /* 1823 * Write the core file header to the file, including padding up to 1824 * the page boundary. 1825 */ 1826 static int 1827 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1828 size_t hdrsize, struct note_info_list *notelst, size_t notesz, 1829 int flags) 1830 { 1831 struct note_info *ninfo; 1832 struct sbuf *sb; 1833 int error; 1834 1835 /* Fill in the header. */ 1836 bzero(hdr, hdrsize); 1837 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags); 1838 1839 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1840 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1841 sbuf_start_section(sb, NULL); 1842 sbuf_bcat(sb, hdr, hdrsize); 1843 TAILQ_FOREACH(ninfo, notelst, link) 1844 __elfN(putnote)(p->td, ninfo, sb); 1845 /* Align up to a page boundary for the program segments. */ 1846 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1847 error = sbuf_finish(sb); 1848 sbuf_delete(sb); 1849 1850 return (error); 1851 } 1852 1853 void 1854 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1855 size_t *sizep) 1856 { 1857 struct proc *p; 1858 struct thread *thr; 1859 size_t size; 1860 1861 p = td->td_proc; 1862 size = 0; 1863 1864 size += __elfN(register_note)(td, list, NT_PRPSINFO, 1865 __elfN(note_prpsinfo), p); 1866 1867 /* 1868 * To have the debugger select the right thread (LWP) as the initial 1869 * thread, we dump the state of the thread passed to us in td first. 1870 * This is the thread that causes the core dump and thus likely to 1871 * be the right thread one wants to have selected in the debugger. 1872 */ 1873 thr = td; 1874 while (thr != NULL) { 1875 size += __elfN(prepare_register_notes)(td, list, thr); 1876 size += __elfN(register_note)(td, list, -1, 1877 __elfN(note_threadmd), thr); 1878 1879 thr = thr == td ? TAILQ_FIRST(&p->p_threads) : 1880 TAILQ_NEXT(thr, td_plist); 1881 if (thr == td) 1882 thr = TAILQ_NEXT(thr, td_plist); 1883 } 1884 1885 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC, 1886 __elfN(note_procstat_proc), p); 1887 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES, 1888 note_procstat_files, p); 1889 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP, 1890 note_procstat_vmmap, p); 1891 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS, 1892 note_procstat_groups, p); 1893 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK, 1894 note_procstat_umask, p); 1895 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT, 1896 note_procstat_rlimit, p); 1897 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL, 1898 note_procstat_osrel, p); 1899 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS, 1900 __elfN(note_procstat_psstrings), p); 1901 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV, 1902 __elfN(note_procstat_auxv), p); 1903 size += __elfN(register_note)(td, list, NT_PROCSTAT_KQUEUES, 1904 __elfN(note_procstat_kqueues), p); 1905 1906 *sizep = size; 1907 } 1908 1909 void 1910 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1911 size_t notesz, int flags) 1912 { 1913 Elf_Ehdr *ehdr; 1914 Elf_Phdr *phdr; 1915 Elf_Shdr *shdr; 1916 struct phdr_closure phc; 1917 Elf_Brandinfo *bi; 1918 1919 ehdr = (Elf_Ehdr *)hdr; 1920 bi = td->td_proc->p_elf_brandinfo; 1921 1922 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1923 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1924 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1925 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1926 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1927 ehdr->e_ident[EI_DATA] = ELF_DATA; 1928 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1929 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi; 1930 ehdr->e_ident[EI_ABIVERSION] = 0; 1931 ehdr->e_ident[EI_PAD] = 0; 1932 ehdr->e_type = ET_CORE; 1933 ehdr->e_machine = bi->machine; 1934 ehdr->e_version = EV_CURRENT; 1935 ehdr->e_entry = 0; 1936 ehdr->e_phoff = sizeof(Elf_Ehdr); 1937 ehdr->e_flags = td->td_proc->p_elf_flags; 1938 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1939 ehdr->e_phentsize = sizeof(Elf_Phdr); 1940 ehdr->e_shentsize = sizeof(Elf_Shdr); 1941 ehdr->e_shstrndx = SHN_UNDEF; 1942 if (numsegs + 1 < PN_XNUM) { 1943 ehdr->e_phnum = numsegs + 1; 1944 ehdr->e_shnum = 0; 1945 } else { 1946 ehdr->e_phnum = PN_XNUM; 1947 ehdr->e_shnum = 1; 1948 1949 ehdr->e_shoff = ehdr->e_phoff + 1950 (numsegs + 1) * ehdr->e_phentsize; 1951 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1952 ("e_shoff: %zu, hdrsize - shdr: %zu", 1953 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1954 1955 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1956 memset(shdr, 0, sizeof(*shdr)); 1957 /* 1958 * A special first section is used to hold large segment and 1959 * section counts. This was proposed by Sun Microsystems in 1960 * Solaris and has been adopted by Linux; the standard ELF 1961 * tools are already familiar with the technique. 1962 * 1963 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1964 * (or 12-7 depending on the version of the document) for more 1965 * details. 1966 */ 1967 shdr->sh_type = SHT_NULL; 1968 shdr->sh_size = ehdr->e_shnum; 1969 shdr->sh_link = ehdr->e_shstrndx; 1970 shdr->sh_info = numsegs + 1; 1971 } 1972 1973 /* 1974 * Fill in the program header entries. 1975 */ 1976 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1977 1978 /* The note segement. */ 1979 phdr->p_type = PT_NOTE; 1980 phdr->p_offset = hdrsize; 1981 phdr->p_vaddr = 0; 1982 phdr->p_paddr = 0; 1983 phdr->p_filesz = notesz; 1984 phdr->p_memsz = 0; 1985 phdr->p_flags = PF_R; 1986 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1987 phdr++; 1988 1989 /* All the writable segments from the program. */ 1990 phc.phdr = phdr; 1991 phc.offset = round_page(hdrsize + notesz); 1992 each_dumpable_segment(td, cb_put_phdr, &phc, flags); 1993 } 1994 1995 static size_t 1996 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list, 1997 struct regset *regset, struct thread *target_td) 1998 { 1999 const struct sysentvec *sv; 2000 struct note_info *ninfo; 2001 size_t size, notesize; 2002 2003 size = 0; 2004 if (!regset->get(regset, target_td, NULL, &size) || size == 0) 2005 return (0); 2006 2007 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2008 ninfo->type = regset->note; 2009 ninfo->regset = regset; 2010 ninfo->outarg = target_td; 2011 ninfo->outsize = size; 2012 TAILQ_INSERT_TAIL(list, ninfo, link); 2013 2014 sv = td->td_proc->p_sysent; 2015 notesize = sizeof(Elf_Note) + /* note header */ 2016 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2017 /* note name */ 2018 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2019 2020 return (notesize); 2021 } 2022 2023 size_t 2024 __elfN(register_note)(struct thread *td, struct note_info_list *list, 2025 int type, outfunc_t out, void *arg) 2026 { 2027 const struct sysentvec *sv; 2028 struct note_info *ninfo; 2029 size_t size, notesize; 2030 2031 sv = td->td_proc->p_sysent; 2032 size = 0; 2033 out(arg, NULL, &size); 2034 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2035 ninfo->type = type; 2036 ninfo->outfunc = out; 2037 ninfo->outarg = arg; 2038 ninfo->outsize = size; 2039 TAILQ_INSERT_TAIL(list, ninfo, link); 2040 2041 if (type == -1) 2042 return (size); 2043 2044 notesize = sizeof(Elf_Note) + /* note header */ 2045 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2046 /* note name */ 2047 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2048 2049 return (notesize); 2050 } 2051 2052 static size_t 2053 append_note_data(const void *src, void *dst, size_t len) 2054 { 2055 size_t padded_len; 2056 2057 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 2058 if (dst != NULL) { 2059 bcopy(src, dst, len); 2060 bzero((char *)dst + len, padded_len - len); 2061 } 2062 return (padded_len); 2063 } 2064 2065 size_t 2066 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 2067 { 2068 Elf_Note *note; 2069 char *buf; 2070 size_t notesize; 2071 2072 buf = dst; 2073 if (buf != NULL) { 2074 note = (Elf_Note *)buf; 2075 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 2076 note->n_descsz = size; 2077 note->n_type = type; 2078 buf += sizeof(*note); 2079 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 2080 sizeof(FREEBSD_ABI_VENDOR)); 2081 append_note_data(src, buf, size); 2082 if (descp != NULL) 2083 *descp = buf; 2084 } 2085 2086 notesize = sizeof(Elf_Note) + /* note header */ 2087 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 2088 /* note name */ 2089 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2090 2091 return (notesize); 2092 } 2093 2094 static void 2095 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb) 2096 { 2097 Elf_Note note; 2098 const struct sysentvec *sv; 2099 ssize_t old_len, sect_len; 2100 size_t new_len, descsz, i; 2101 2102 if (ninfo->type == -1) { 2103 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2104 return; 2105 } 2106 2107 sv = td->td_proc->p_sysent; 2108 2109 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1; 2110 note.n_descsz = ninfo->outsize; 2111 note.n_type = ninfo->type; 2112 2113 sbuf_bcat(sb, ¬e, sizeof(note)); 2114 sbuf_start_section(sb, &old_len); 2115 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor, 2116 strlen(sv->sv_elf_core_abi_vendor) + 1); 2117 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2118 if (note.n_descsz == 0) 2119 return; 2120 sbuf_start_section(sb, &old_len); 2121 if (ninfo->regset != NULL) { 2122 struct regset *regset = ninfo->regset; 2123 void *buf; 2124 2125 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK); 2126 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize); 2127 sbuf_bcat(sb, buf, ninfo->outsize); 2128 free(buf, M_TEMP); 2129 } else 2130 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2131 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2132 if (sect_len < 0) 2133 return; 2134 2135 new_len = (size_t)sect_len; 2136 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 2137 if (new_len < descsz) { 2138 /* 2139 * It is expected that individual note emitters will correctly 2140 * predict their expected output size and fill up to that size 2141 * themselves, padding in a format-specific way if needed. 2142 * However, in case they don't, just do it here with zeros. 2143 */ 2144 for (i = 0; i < descsz - new_len; i++) 2145 sbuf_putc(sb, 0); 2146 } else if (new_len > descsz) { 2147 /* 2148 * We can't always truncate sb -- we may have drained some 2149 * of it already. 2150 */ 2151 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 2152 "read it (%zu > %zu). Since it is longer than " 2153 "expected, this coredump's notes are corrupt. THIS " 2154 "IS A BUG in the note_procstat routine for type %u.\n", 2155 __func__, (unsigned)note.n_type, new_len, descsz, 2156 (unsigned)note.n_type)); 2157 } 2158 } 2159 2160 /* 2161 * Miscellaneous note out functions. 2162 */ 2163 2164 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2165 #include <compat/freebsd32/freebsd32.h> 2166 #include <compat/freebsd32/freebsd32_signal.h> 2167 2168 typedef struct prstatus32 elf_prstatus_t; 2169 typedef struct prpsinfo32 elf_prpsinfo_t; 2170 typedef struct fpreg32 elf_prfpregset_t; 2171 typedef struct fpreg32 elf_fpregset_t; 2172 typedef struct reg32 elf_gregset_t; 2173 typedef struct thrmisc32 elf_thrmisc_t; 2174 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t; 2175 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 2176 typedef struct kinfo_proc32 elf_kinfo_proc_t; 2177 typedef uint32_t elf_ps_strings_t; 2178 #else 2179 typedef prstatus_t elf_prstatus_t; 2180 typedef prpsinfo_t elf_prpsinfo_t; 2181 typedef prfpregset_t elf_prfpregset_t; 2182 typedef prfpregset_t elf_fpregset_t; 2183 typedef gregset_t elf_gregset_t; 2184 typedef thrmisc_t elf_thrmisc_t; 2185 typedef struct ptrace_lwpinfo elf_lwpinfo_t; 2186 #define ELF_KERN_PROC_MASK 0 2187 typedef struct kinfo_proc elf_kinfo_proc_t; 2188 typedef vm_offset_t elf_ps_strings_t; 2189 #endif 2190 2191 static void 2192 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2193 { 2194 struct sbuf sbarg; 2195 size_t len; 2196 char *cp, *end; 2197 struct proc *p; 2198 elf_prpsinfo_t *psinfo; 2199 int error; 2200 2201 p = arg; 2202 if (sb != NULL) { 2203 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 2204 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 2205 psinfo->pr_version = PRPSINFO_VERSION; 2206 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 2207 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 2208 PROC_LOCK(p); 2209 if (p->p_args != NULL) { 2210 len = sizeof(psinfo->pr_psargs) - 1; 2211 if (len > p->p_args->ar_length) 2212 len = p->p_args->ar_length; 2213 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 2214 PROC_UNLOCK(p); 2215 error = 0; 2216 } else { 2217 _PHOLD(p); 2218 PROC_UNLOCK(p); 2219 sbuf_new(&sbarg, psinfo->pr_psargs, 2220 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 2221 error = proc_getargv(curthread, p, &sbarg); 2222 PRELE(p); 2223 if (sbuf_finish(&sbarg) == 0) { 2224 len = sbuf_len(&sbarg); 2225 if (len > 0) 2226 len--; 2227 } else { 2228 len = sizeof(psinfo->pr_psargs) - 1; 2229 } 2230 sbuf_delete(&sbarg); 2231 } 2232 if (error != 0 || len == 0 || (ssize_t)len == -1) 2233 strlcpy(psinfo->pr_psargs, p->p_comm, 2234 sizeof(psinfo->pr_psargs)); 2235 else { 2236 KASSERT(len < sizeof(psinfo->pr_psargs), 2237 ("len is too long: %zu vs %zu", len, 2238 sizeof(psinfo->pr_psargs))); 2239 cp = psinfo->pr_psargs; 2240 end = cp + len - 1; 2241 for (;;) { 2242 cp = memchr(cp, '\0', end - cp); 2243 if (cp == NULL) 2244 break; 2245 *cp = ' '; 2246 } 2247 } 2248 psinfo->pr_pid = p->p_pid; 2249 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 2250 free(psinfo, M_TEMP); 2251 } 2252 *sizep = sizeof(*psinfo); 2253 } 2254 2255 static bool 2256 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf, 2257 size_t *sizep) 2258 { 2259 elf_prstatus_t *status; 2260 2261 if (buf != NULL) { 2262 KASSERT(*sizep == sizeof(*status), ("%s: invalid size", 2263 __func__)); 2264 status = buf; 2265 memset(status, 0, *sizep); 2266 status->pr_version = PRSTATUS_VERSION; 2267 status->pr_statussz = sizeof(elf_prstatus_t); 2268 status->pr_gregsetsz = sizeof(elf_gregset_t); 2269 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 2270 status->pr_osreldate = osreldate; 2271 status->pr_cursig = td->td_proc->p_sig; 2272 status->pr_pid = td->td_tid; 2273 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2274 fill_regs32(td, &status->pr_reg); 2275 #else 2276 fill_regs(td, &status->pr_reg); 2277 #endif 2278 } 2279 *sizep = sizeof(*status); 2280 return (true); 2281 } 2282 2283 static bool 2284 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf, 2285 size_t size) 2286 { 2287 elf_prstatus_t *status; 2288 2289 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__)); 2290 status = buf; 2291 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2292 set_regs32(td, &status->pr_reg); 2293 #else 2294 set_regs(td, &status->pr_reg); 2295 #endif 2296 return (true); 2297 } 2298 2299 static struct regset __elfN(regset_prstatus) = { 2300 .note = NT_PRSTATUS, 2301 .size = sizeof(elf_prstatus_t), 2302 .get = __elfN(get_prstatus), 2303 .set = __elfN(set_prstatus), 2304 }; 2305 ELF_REGSET(__elfN(regset_prstatus)); 2306 2307 static bool 2308 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf, 2309 size_t *sizep) 2310 { 2311 elf_prfpregset_t *fpregset; 2312 2313 if (buf != NULL) { 2314 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size", 2315 __func__)); 2316 fpregset = buf; 2317 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2318 fill_fpregs32(td, fpregset); 2319 #else 2320 fill_fpregs(td, fpregset); 2321 #endif 2322 } 2323 *sizep = sizeof(*fpregset); 2324 return (true); 2325 } 2326 2327 static bool 2328 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf, 2329 size_t size) 2330 { 2331 elf_prfpregset_t *fpregset; 2332 2333 fpregset = buf; 2334 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__)); 2335 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2336 set_fpregs32(td, fpregset); 2337 #else 2338 set_fpregs(td, fpregset); 2339 #endif 2340 return (true); 2341 } 2342 2343 static struct regset __elfN(regset_fpregset) = { 2344 .note = NT_FPREGSET, 2345 .size = sizeof(elf_prfpregset_t), 2346 .get = __elfN(get_fpregset), 2347 .set = __elfN(set_fpregset), 2348 }; 2349 ELF_REGSET(__elfN(regset_fpregset)); 2350 2351 static bool 2352 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf, 2353 size_t *sizep) 2354 { 2355 elf_thrmisc_t *thrmisc; 2356 2357 if (buf != NULL) { 2358 KASSERT(*sizep == sizeof(*thrmisc), 2359 ("%s: invalid size", __func__)); 2360 thrmisc = buf; 2361 bzero(thrmisc, sizeof(*thrmisc)); 2362 strcpy(thrmisc->pr_tname, td->td_name); 2363 } 2364 *sizep = sizeof(*thrmisc); 2365 return (true); 2366 } 2367 2368 static struct regset __elfN(regset_thrmisc) = { 2369 .note = NT_THRMISC, 2370 .size = sizeof(elf_thrmisc_t), 2371 .get = __elfN(get_thrmisc), 2372 }; 2373 ELF_REGSET(__elfN(regset_thrmisc)); 2374 2375 static bool 2376 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf, 2377 size_t *sizep) 2378 { 2379 elf_lwpinfo_t pl; 2380 size_t size; 2381 int structsize; 2382 2383 size = sizeof(structsize) + sizeof(pl); 2384 if (buf != NULL) { 2385 KASSERT(*sizep == size, ("%s: invalid size", __func__)); 2386 structsize = sizeof(pl); 2387 memcpy(buf, &structsize, sizeof(structsize)); 2388 bzero(&pl, sizeof(pl)); 2389 pl.pl_lwpid = td->td_tid; 2390 pl.pl_event = PL_EVENT_NONE; 2391 pl.pl_sigmask = td->td_sigmask; 2392 pl.pl_siglist = td->td_siglist; 2393 if (td->td_si.si_signo != 0) { 2394 pl.pl_event = PL_EVENT_SIGNAL; 2395 pl.pl_flags |= PL_FLAG_SI; 2396 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2397 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2398 #else 2399 pl.pl_siginfo = td->td_si; 2400 #endif 2401 } 2402 strcpy(pl.pl_tdname, td->td_name); 2403 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2404 memcpy((int *)buf + 1, &pl, sizeof(pl)); 2405 } 2406 *sizep = size; 2407 return (true); 2408 } 2409 2410 static struct regset __elfN(regset_lwpinfo) = { 2411 .note = NT_PTLWPINFO, 2412 .size = sizeof(int) + sizeof(elf_lwpinfo_t), 2413 .get = __elfN(get_lwpinfo), 2414 }; 2415 ELF_REGSET(__elfN(regset_lwpinfo)); 2416 2417 static size_t 2418 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list, 2419 struct thread *target_td) 2420 { 2421 struct sysentvec *sv = td->td_proc->p_sysent; 2422 struct regset **regsetp, **regset_end, *regset; 2423 size_t size; 2424 2425 size = 0; 2426 2427 if (target_td == td) 2428 cpu_update_pcb(target_td); 2429 2430 /* NT_PRSTATUS must be the first register set note. */ 2431 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus), 2432 target_td); 2433 2434 regsetp = sv->sv_regset_begin; 2435 if (regsetp == NULL) { 2436 /* XXX: This shouldn't be true for any FreeBSD ABIs. */ 2437 size += __elfN(register_regset_note)(td, list, 2438 &__elfN(regset_fpregset), target_td); 2439 return (size); 2440 } 2441 regset_end = sv->sv_regset_end; 2442 MPASS(regset_end != NULL); 2443 for (; regsetp < regset_end; regsetp++) { 2444 regset = *regsetp; 2445 if (regset->note == NT_PRSTATUS) 2446 continue; 2447 size += __elfN(register_regset_note)(td, list, regset, 2448 target_td); 2449 } 2450 return (size); 2451 } 2452 2453 /* 2454 * Allow for MD specific notes, as well as any MD 2455 * specific preparations for writing MI notes. 2456 */ 2457 static void 2458 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2459 { 2460 struct thread *td; 2461 void *buf; 2462 size_t size; 2463 2464 td = (struct thread *)arg; 2465 size = *sizep; 2466 if (size != 0 && sb != NULL) 2467 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2468 else 2469 buf = NULL; 2470 size = 0; 2471 __elfN(dump_thread)(td, buf, &size); 2472 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2473 if (size != 0 && sb != NULL) 2474 sbuf_bcat(sb, buf, size); 2475 free(buf, M_TEMP); 2476 *sizep = size; 2477 } 2478 2479 #ifdef KINFO_PROC_SIZE 2480 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2481 #endif 2482 2483 static void 2484 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2485 { 2486 struct proc *p; 2487 size_t size; 2488 int structsize; 2489 2490 p = arg; 2491 size = sizeof(structsize) + p->p_numthreads * 2492 sizeof(elf_kinfo_proc_t); 2493 2494 if (sb != NULL) { 2495 KASSERT(*sizep == size, ("invalid size")); 2496 structsize = sizeof(elf_kinfo_proc_t); 2497 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2498 sx_slock(&proctree_lock); 2499 PROC_LOCK(p); 2500 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2501 sx_sunlock(&proctree_lock); 2502 } 2503 *sizep = size; 2504 } 2505 2506 #ifdef KINFO_FILE_SIZE 2507 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2508 #endif 2509 2510 static void 2511 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2512 { 2513 struct proc *p; 2514 size_t size, sect_sz, i; 2515 ssize_t start_len, sect_len; 2516 int structsize, filedesc_flags; 2517 2518 if (coredump_pack_fileinfo) 2519 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2520 else 2521 filedesc_flags = 0; 2522 2523 p = arg; 2524 structsize = sizeof(struct kinfo_file); 2525 if (sb == NULL) { 2526 size = 0; 2527 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2528 sbuf_set_drain(sb, sbuf_count_drain, &size); 2529 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2530 PROC_LOCK(p); 2531 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2532 sbuf_finish(sb); 2533 sbuf_delete(sb); 2534 *sizep = size; 2535 } else { 2536 sbuf_start_section(sb, &start_len); 2537 2538 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2539 PROC_LOCK(p); 2540 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2541 filedesc_flags); 2542 2543 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2544 if (sect_len < 0) 2545 return; 2546 sect_sz = sect_len; 2547 2548 KASSERT(sect_sz <= *sizep, 2549 ("kern_proc_filedesc_out did not respect maxlen; " 2550 "requested %zu, got %zu", *sizep - sizeof(structsize), 2551 sect_sz - sizeof(structsize))); 2552 2553 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2554 sbuf_putc(sb, 0); 2555 } 2556 } 2557 2558 #ifdef KINFO_VMENTRY_SIZE 2559 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2560 #endif 2561 2562 static void 2563 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2564 { 2565 struct proc *p; 2566 size_t size; 2567 int structsize, vmmap_flags; 2568 2569 if (coredump_pack_vmmapinfo) 2570 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2571 else 2572 vmmap_flags = 0; 2573 2574 p = arg; 2575 structsize = sizeof(struct kinfo_vmentry); 2576 if (sb == NULL) { 2577 size = 0; 2578 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2579 sbuf_set_drain(sb, sbuf_count_drain, &size); 2580 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2581 PROC_LOCK(p); 2582 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2583 sbuf_finish(sb); 2584 sbuf_delete(sb); 2585 *sizep = size; 2586 } else { 2587 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2588 PROC_LOCK(p); 2589 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2590 vmmap_flags); 2591 } 2592 } 2593 2594 static void 2595 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2596 { 2597 struct proc *p; 2598 size_t size; 2599 int structsize; 2600 2601 p = arg; 2602 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2603 if (sb != NULL) { 2604 KASSERT(*sizep == size, ("invalid size")); 2605 structsize = sizeof(gid_t); 2606 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2607 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2608 sizeof(gid_t)); 2609 } 2610 *sizep = size; 2611 } 2612 2613 static void 2614 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2615 { 2616 struct proc *p; 2617 size_t size; 2618 int structsize; 2619 2620 p = arg; 2621 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask); 2622 if (sb != NULL) { 2623 KASSERT(*sizep == size, ("invalid size")); 2624 structsize = sizeof(p->p_pd->pd_cmask); 2625 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2626 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask)); 2627 } 2628 *sizep = size; 2629 } 2630 2631 static void 2632 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2633 { 2634 struct proc *p; 2635 struct rlimit rlim[RLIM_NLIMITS]; 2636 size_t size; 2637 int structsize, i; 2638 2639 p = arg; 2640 size = sizeof(structsize) + sizeof(rlim); 2641 if (sb != NULL) { 2642 KASSERT(*sizep == size, ("invalid size")); 2643 structsize = sizeof(rlim); 2644 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2645 PROC_LOCK(p); 2646 for (i = 0; i < RLIM_NLIMITS; i++) 2647 lim_rlimit_proc(p, i, &rlim[i]); 2648 PROC_UNLOCK(p); 2649 sbuf_bcat(sb, rlim, sizeof(rlim)); 2650 } 2651 *sizep = size; 2652 } 2653 2654 static void 2655 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2656 { 2657 struct proc *p; 2658 size_t size; 2659 int structsize; 2660 2661 p = arg; 2662 size = sizeof(structsize) + sizeof(p->p_osrel); 2663 if (sb != NULL) { 2664 KASSERT(*sizep == size, ("invalid size")); 2665 structsize = sizeof(p->p_osrel); 2666 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2667 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2668 } 2669 *sizep = size; 2670 } 2671 2672 static void 2673 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2674 { 2675 struct proc *p; 2676 elf_ps_strings_t ps_strings; 2677 size_t size; 2678 int structsize; 2679 2680 p = arg; 2681 size = sizeof(structsize) + sizeof(ps_strings); 2682 if (sb != NULL) { 2683 KASSERT(*sizep == size, ("invalid size")); 2684 structsize = sizeof(ps_strings); 2685 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2686 ps_strings = PTROUT(PROC_PS_STRINGS(p)); 2687 #else 2688 ps_strings = PROC_PS_STRINGS(p); 2689 #endif 2690 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2691 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2692 } 2693 *sizep = size; 2694 } 2695 2696 static void 2697 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2698 { 2699 struct proc *p; 2700 size_t size; 2701 int structsize; 2702 2703 p = arg; 2704 if (sb == NULL) { 2705 size = 0; 2706 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo), 2707 SBUF_FIXEDLEN); 2708 sbuf_set_drain(sb, sbuf_count_drain, &size); 2709 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2710 PHOLD(p); 2711 proc_getauxv(curthread, p, sb); 2712 PRELE(p); 2713 sbuf_finish(sb); 2714 sbuf_delete(sb); 2715 *sizep = size; 2716 } else { 2717 structsize = sizeof(Elf_Auxinfo); 2718 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2719 PHOLD(p); 2720 proc_getauxv(curthread, p, sb); 2721 PRELE(p); 2722 } 2723 } 2724 2725 static void 2726 __elfN(note_procstat_kqueues)(void *arg, struct sbuf *sb, size_t *sizep) 2727 { 2728 struct proc *p; 2729 size_t size, sect_sz, i; 2730 ssize_t start_len, sect_len; 2731 int structsize; 2732 bool compat32; 2733 2734 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2735 compat32 = true; 2736 structsize = sizeof(struct kinfo_knote32); 2737 #else 2738 compat32 = false; 2739 structsize = sizeof(struct kinfo_knote); 2740 #endif 2741 p = arg; 2742 if (sb == NULL) { 2743 size = 0; 2744 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2745 sbuf_set_drain(sb, sbuf_count_drain, &size); 2746 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2747 kern_proc_kqueues_out(p, sb, -1, compat32); 2748 sbuf_finish(sb); 2749 sbuf_delete(sb); 2750 *sizep = size; 2751 } else { 2752 sbuf_start_section(sb, &start_len); 2753 2754 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2755 kern_proc_kqueues_out(p, sb, *sizep - sizeof(structsize), 2756 compat32); 2757 2758 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2759 if (sect_len < 0) 2760 return; 2761 sect_sz = sect_len; 2762 2763 KASSERT(sect_sz <= *sizep, 2764 ("kern_proc_kqueue_out did not respect maxlen; " 2765 "requested %zu, got %zu", *sizep - sizeof(structsize), 2766 sect_sz - sizeof(structsize))); 2767 2768 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2769 sbuf_putc(sb, 0); 2770 } 2771 } 2772 2773 #define MAX_NOTES_LOOP 4096 2774 bool 2775 __elfN(parse_notes)(const struct image_params *imgp, const Elf_Note *checknote, 2776 const char *note_vendor, const Elf_Phdr *pnote, 2777 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg) 2778 { 2779 const Elf_Note *note, *note0, *note_end; 2780 const char *note_name; 2781 char *buf; 2782 int i, error; 2783 bool res; 2784 2785 /* We need some limit, might as well use PAGE_SIZE. */ 2786 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2787 return (false); 2788 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2789 if (pnote->p_offset > PAGE_SIZE || 2790 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2791 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT); 2792 if (buf == NULL) { 2793 VOP_UNLOCK(imgp->vp); 2794 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2795 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 2796 } 2797 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2798 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2799 curthread->td_ucred, NOCRED, NULL, curthread); 2800 if (error != 0) { 2801 uprintf("i/o error PT_NOTE\n"); 2802 goto retf; 2803 } 2804 note = note0 = (const Elf_Note *)buf; 2805 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2806 } else { 2807 note = note0 = (const Elf_Note *)(imgp->image_header + 2808 pnote->p_offset); 2809 note_end = (const Elf_Note *)(imgp->image_header + 2810 pnote->p_offset + pnote->p_filesz); 2811 buf = NULL; 2812 } 2813 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end; 2814 i++) { 2815 if (!aligned(note, Elf32_Addr)) { 2816 uprintf("Unaligned ELF note\n"); 2817 goto retf; 2818 } 2819 if ((const char *)note_end - (const char *)note < 2820 sizeof(Elf_Note)) { 2821 uprintf("ELF note to short\n"); 2822 goto retf; 2823 } 2824 if (note->n_namesz != checknote->n_namesz || 2825 note->n_descsz != checknote->n_descsz || 2826 note->n_type != checknote->n_type) 2827 goto nextnote; 2828 note_name = (const char *)(note + 1); 2829 if (note_name + checknote->n_namesz >= 2830 (const char *)note_end || strncmp(note_vendor, 2831 note_name, checknote->n_namesz) != 0) 2832 goto nextnote; 2833 2834 if (cb(note, cb_arg, &res)) 2835 goto ret; 2836 nextnote: 2837 note = (const Elf_Note *)((const char *)(note + 1) + 2838 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2839 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2840 } 2841 if (i >= MAX_NOTES_LOOP) 2842 uprintf("ELF note parser reached %d notes\n", i); 2843 retf: 2844 res = false; 2845 ret: 2846 free(buf, M_TEMP); 2847 return (res); 2848 } 2849 2850 struct brandnote_cb_arg { 2851 Elf_Brandnote *brandnote; 2852 int32_t *osrel; 2853 }; 2854 2855 static bool 2856 brandnote_cb(const Elf_Note *note, void *arg0, bool *res) 2857 { 2858 struct brandnote_cb_arg *arg; 2859 2860 arg = arg0; 2861 2862 /* 2863 * Fetch the osreldate for binary from the ELF OSABI-note if 2864 * necessary. 2865 */ 2866 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 && 2867 arg->brandnote->trans_osrel != NULL ? 2868 arg->brandnote->trans_osrel(note, arg->osrel) : true; 2869 2870 return (true); 2871 } 2872 2873 static Elf_Note fctl_note = { 2874 .n_namesz = sizeof(FREEBSD_ABI_VENDOR), 2875 .n_descsz = sizeof(uint32_t), 2876 .n_type = NT_FREEBSD_FEATURE_CTL, 2877 }; 2878 2879 struct fctl_cb_arg { 2880 bool *has_fctl0; 2881 uint32_t *fctl0; 2882 }; 2883 2884 static bool 2885 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res) 2886 { 2887 struct fctl_cb_arg *arg; 2888 const Elf32_Word *desc; 2889 uintptr_t p; 2890 2891 arg = arg0; 2892 p = (uintptr_t)(note + 1); 2893 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 2894 desc = (const Elf32_Word *)p; 2895 *arg->has_fctl0 = true; 2896 *arg->fctl0 = desc[0]; 2897 *res = true; 2898 return (true); 2899 } 2900 2901 /* 2902 * Try to find the appropriate ABI-note section for checknote, fetch 2903 * the osreldate and feature control flags for binary from the ELF 2904 * OSABI-note. Only the first page of the image is searched, the same 2905 * as for headers. 2906 */ 2907 static bool 2908 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote, 2909 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0) 2910 { 2911 const Elf_Phdr *phdr; 2912 const Elf_Ehdr *hdr; 2913 struct brandnote_cb_arg b_arg; 2914 struct fctl_cb_arg f_arg; 2915 int i, j; 2916 2917 hdr = (const Elf_Ehdr *)imgp->image_header; 2918 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2919 b_arg.brandnote = brandnote; 2920 b_arg.osrel = osrel; 2921 f_arg.has_fctl0 = has_fctl0; 2922 f_arg.fctl0 = fctl0; 2923 2924 for (i = 0; i < hdr->e_phnum; i++) { 2925 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, 2926 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb, 2927 &b_arg)) { 2928 for (j = 0; j < hdr->e_phnum; j++) { 2929 if (phdr[j].p_type == PT_NOTE && 2930 __elfN(parse_notes)(imgp, &fctl_note, 2931 FREEBSD_ABI_VENDOR, &phdr[j], 2932 note_fctl_cb, &f_arg)) 2933 break; 2934 } 2935 return (true); 2936 } 2937 } 2938 return (false); 2939 2940 } 2941 2942 /* 2943 * Tell kern_execve.c about it, with a little help from the linker. 2944 */ 2945 static struct execsw __elfN(execsw) = { 2946 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 2947 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2948 }; 2949 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2950 2951 static vm_prot_t 2952 __elfN(trans_prot)(Elf_Word flags) 2953 { 2954 vm_prot_t prot; 2955 2956 prot = 0; 2957 if (flags & PF_X) 2958 prot |= VM_PROT_EXECUTE; 2959 if (flags & PF_W) 2960 prot |= VM_PROT_WRITE; 2961 if (flags & PF_R) 2962 prot |= VM_PROT_READ; 2963 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 2964 if (i386_read_exec && (flags & PF_R)) 2965 prot |= VM_PROT_EXECUTE; 2966 #endif 2967 return (prot); 2968 } 2969 2970 static Elf_Word 2971 __elfN(untrans_prot)(vm_prot_t prot) 2972 { 2973 Elf_Word flags; 2974 2975 flags = 0; 2976 if (prot & VM_PROT_EXECUTE) 2977 flags |= PF_X; 2978 if (prot & VM_PROT_READ) 2979 flags |= PF_R; 2980 if (prot & VM_PROT_WRITE) 2981 flags |= PF_W; 2982 return (flags); 2983 } 2984