1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2017 Dell EMC 5 * Copyright (c) 2000-2001, 2003 David O'Brien 6 * Copyright (c) 1995-1996 Søren Schmidt 7 * Copyright (c) 1996 Peter Wemm 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer 15 * in this position and unchanged. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. The name of the author may not be used to endorse or promote products 20 * derived from this software without specific prior written permission 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_capsicum.h" 38 39 #include <sys/param.h> 40 #include <sys/capsicum.h> 41 #include <sys/compressor.h> 42 #include <sys/exec.h> 43 #include <sys/fcntl.h> 44 #include <sys/imgact.h> 45 #include <sys/imgact_elf.h> 46 #include <sys/jail.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/mount.h> 51 #include <sys/mman.h> 52 #include <sys/namei.h> 53 #include <sys/proc.h> 54 #include <sys/procfs.h> 55 #include <sys/ptrace.h> 56 #include <sys/racct.h> 57 #include <sys/reg.h> 58 #include <sys/resourcevar.h> 59 #include <sys/rwlock.h> 60 #include <sys/sbuf.h> 61 #include <sys/sf_buf.h> 62 #include <sys/smp.h> 63 #include <sys/systm.h> 64 #include <sys/signalvar.h> 65 #include <sys/stat.h> 66 #include <sys/sx.h> 67 #include <sys/syscall.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/vnode.h> 71 #include <sys/syslog.h> 72 #include <sys/eventhandler.h> 73 #include <sys/user.h> 74 75 #include <vm/vm.h> 76 #include <vm/vm_kern.h> 77 #include <vm/vm_param.h> 78 #include <vm/pmap.h> 79 #include <vm/vm_map.h> 80 #include <vm/vm_object.h> 81 #include <vm/vm_extern.h> 82 83 #include <machine/elf.h> 84 #include <machine/md_var.h> 85 86 #define ELF_NOTE_ROUNDSIZE 4 87 #define OLD_EI_BRAND 8 88 89 static int __elfN(check_header)(const Elf_Ehdr *hdr); 90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 91 const char *interp, int32_t *osrel, uint32_t *fctl0); 92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 93 u_long *entry); 94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset, 95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot); 96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note, 98 int32_t *osrel); 99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 100 static bool __elfN(check_note)(struct image_params *imgp, 101 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0, 102 uint32_t *fctl0); 103 static vm_prot_t __elfN(trans_prot)(Elf_Word); 104 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 105 static size_t __elfN(prepare_register_notes)(struct thread *td, 106 struct note_info_list *list, struct thread *target_td); 107 108 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), 109 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 110 ""); 111 112 int __elfN(fallback_brand) = -1; 113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 114 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, 115 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 116 117 static int elf_legacy_coredump = 0; 118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 119 &elf_legacy_coredump, 0, 120 "include all and only RW pages in core dumps"); 121 122 int __elfN(nxstack) = 123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \ 124 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \ 125 defined(__riscv) 126 1; 127 #else 128 0; 129 #endif 130 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 131 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 132 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); 133 134 #if defined(__amd64__) 135 static int __elfN(vdso) = 1; 136 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 137 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0, 138 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading"); 139 #else 140 static int __elfN(vdso) = 0; 141 #endif 142 143 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 144 int i386_read_exec = 0; 145 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 146 "enable execution from readable segments"); 147 #endif 148 149 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR; 150 static int 151 sysctl_pie_base(SYSCTL_HANDLER_ARGS) 152 { 153 u_long val; 154 int error; 155 156 val = __elfN(pie_base); 157 error = sysctl_handle_long(oidp, &val, 0, req); 158 if (error != 0 || req->newptr == NULL) 159 return (error); 160 if ((val & PAGE_MASK) != 0) 161 return (EINVAL); 162 __elfN(pie_base) = val; 163 return (0); 164 } 165 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base, 166 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 167 sysctl_pie_base, "LU", 168 "PIE load base without randomization"); 169 170 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, 171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 172 ""); 173 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr) 174 175 /* 176 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures 177 * have limited address space (which can cause issues for applications with 178 * high memory use) so we leave it off there. 179 */ 180 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64; 181 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN, 182 &__elfN(aslr_enabled), 0, 183 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 184 ": enable address map randomization"); 185 186 /* 187 * Enable ASLR by default for 64-bit PIE binaries. 188 */ 189 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64; 190 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN, 191 &__elfN(pie_aslr_enabled), 0, 192 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 193 ": enable address map randomization for PIE binaries"); 194 195 /* 196 * Sbrk is deprecated and it can be assumed that in most cases it will not be 197 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR 198 * to use the bss grow region. 199 */ 200 static int __elfN(aslr_honor_sbrk) = 0; 201 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW, 202 &__elfN(aslr_honor_sbrk), 0, 203 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used"); 204 205 static int __elfN(aslr_stack) = 1; 206 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN, 207 &__elfN(aslr_stack), 0, 208 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 209 ": enable stack address randomization"); 210 211 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64; 212 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN, 213 &__elfN(aslr_shared_page), 0, 214 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 215 ": enable shared page address randomization"); 216 217 static int __elfN(sigfastblock) = 1; 218 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock, 219 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0, 220 "enable sigfastblock for new processes"); 221 222 static bool __elfN(allow_wx) = true; 223 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx, 224 CTLFLAG_RWTUN, &__elfN(allow_wx), 0, 225 "Allow pages to be mapped simultaneously writable and executable"); 226 227 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 228 229 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a)) 230 231 Elf_Brandnote __elfN(freebsd_brandnote) = { 232 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 233 .hdr.n_descsz = sizeof(int32_t), 234 .hdr.n_type = NT_FREEBSD_ABI_TAG, 235 .vendor = FREEBSD_ABI_VENDOR, 236 .flags = BN_TRANSLATE_OSREL, 237 .trans_osrel = __elfN(freebsd_trans_osrel) 238 }; 239 240 static bool 241 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 242 { 243 uintptr_t p; 244 245 p = (uintptr_t)(note + 1); 246 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 247 *osrel = *(const int32_t *)(p); 248 249 return (true); 250 } 251 252 static const char GNU_ABI_VENDOR[] = "GNU"; 253 static int GNU_KFREEBSD_ABI_DESC = 3; 254 255 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 256 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 257 .hdr.n_descsz = 16, /* XXX at least 16 */ 258 .hdr.n_type = 1, 259 .vendor = GNU_ABI_VENDOR, 260 .flags = BN_TRANSLATE_OSREL, 261 .trans_osrel = kfreebsd_trans_osrel 262 }; 263 264 static bool 265 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 266 { 267 const Elf32_Word *desc; 268 uintptr_t p; 269 270 p = (uintptr_t)(note + 1); 271 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 272 273 desc = (const Elf32_Word *)p; 274 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 275 return (false); 276 277 /* 278 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 279 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 280 */ 281 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 282 283 return (true); 284 } 285 286 int 287 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 288 { 289 int i; 290 291 for (i = 0; i < MAX_BRANDS; i++) { 292 if (elf_brand_list[i] == NULL) { 293 elf_brand_list[i] = entry; 294 break; 295 } 296 } 297 if (i == MAX_BRANDS) { 298 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 299 __func__, entry); 300 return (-1); 301 } 302 return (0); 303 } 304 305 int 306 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 307 { 308 int i; 309 310 for (i = 0; i < MAX_BRANDS; i++) { 311 if (elf_brand_list[i] == entry) { 312 elf_brand_list[i] = NULL; 313 break; 314 } 315 } 316 if (i == MAX_BRANDS) 317 return (-1); 318 return (0); 319 } 320 321 bool 322 __elfN(brand_inuse)(Elf_Brandinfo *entry) 323 { 324 struct proc *p; 325 bool rval = false; 326 327 sx_slock(&allproc_lock); 328 FOREACH_PROC_IN_SYSTEM(p) { 329 if (p->p_sysent == entry->sysvec) { 330 rval = true; 331 break; 332 } 333 } 334 sx_sunlock(&allproc_lock); 335 336 return (rval); 337 } 338 339 static Elf_Brandinfo * 340 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 341 int32_t *osrel, uint32_t *fctl0) 342 { 343 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 344 Elf_Brandinfo *bi, *bi_m; 345 bool ret, has_fctl0; 346 int i, interp_name_len; 347 348 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0; 349 350 /* 351 * We support four types of branding -- (1) the ELF EI_OSABI field 352 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 353 * branding w/in the ELF header, (3) path of the `interp_path' 354 * field, and (4) the ".note.ABI-tag" ELF section. 355 */ 356 357 /* Look for an ".note.ABI-tag" ELF section */ 358 bi_m = NULL; 359 for (i = 0; i < MAX_BRANDS; i++) { 360 bi = elf_brand_list[i]; 361 if (bi == NULL) 362 continue; 363 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 364 continue; 365 if (hdr->e_machine == bi->machine && (bi->flags & 366 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 367 has_fctl0 = false; 368 *fctl0 = 0; 369 *osrel = 0; 370 ret = __elfN(check_note)(imgp, bi->brand_note, osrel, 371 &has_fctl0, fctl0); 372 /* Give brand a chance to veto check_note's guess */ 373 if (ret && bi->header_supported) { 374 ret = bi->header_supported(imgp, osrel, 375 has_fctl0 ? fctl0 : NULL); 376 } 377 /* 378 * If note checker claimed the binary, but the 379 * interpreter path in the image does not 380 * match default one for the brand, try to 381 * search for other brands with the same 382 * interpreter. Either there is better brand 383 * with the right interpreter, or, failing 384 * this, we return first brand which accepted 385 * our note and, optionally, header. 386 */ 387 if (ret && bi_m == NULL && interp != NULL && 388 (bi->interp_path == NULL || 389 (strlen(bi->interp_path) + 1 != interp_name_len || 390 strncmp(interp, bi->interp_path, interp_name_len) 391 != 0))) { 392 bi_m = bi; 393 ret = 0; 394 } 395 if (ret) 396 return (bi); 397 } 398 } 399 if (bi_m != NULL) 400 return (bi_m); 401 402 /* If the executable has a brand, search for it in the brand list. */ 403 for (i = 0; i < MAX_BRANDS; i++) { 404 bi = elf_brand_list[i]; 405 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 406 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 407 continue; 408 if (hdr->e_machine == bi->machine && 409 (hdr->e_ident[EI_OSABI] == bi->brand || 410 (bi->compat_3_brand != NULL && 411 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 412 bi->compat_3_brand) == 0))) { 413 /* Looks good, but give brand a chance to veto */ 414 if (bi->header_supported == NULL || 415 bi->header_supported(imgp, NULL, NULL)) { 416 /* 417 * Again, prefer strictly matching 418 * interpreter path. 419 */ 420 if (interp_name_len == 0 && 421 bi->interp_path == NULL) 422 return (bi); 423 if (bi->interp_path != NULL && 424 strlen(bi->interp_path) + 1 == 425 interp_name_len && strncmp(interp, 426 bi->interp_path, interp_name_len) == 0) 427 return (bi); 428 if (bi_m == NULL) 429 bi_m = bi; 430 } 431 } 432 } 433 if (bi_m != NULL) 434 return (bi_m); 435 436 /* No known brand, see if the header is recognized by any brand */ 437 for (i = 0; i < MAX_BRANDS; i++) { 438 bi = elf_brand_list[i]; 439 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 440 bi->header_supported == NULL) 441 continue; 442 if (hdr->e_machine == bi->machine) { 443 ret = bi->header_supported(imgp, NULL, NULL); 444 if (ret) 445 return (bi); 446 } 447 } 448 449 /* Lacking a known brand, search for a recognized interpreter. */ 450 if (interp != NULL) { 451 for (i = 0; i < MAX_BRANDS; i++) { 452 bi = elf_brand_list[i]; 453 if (bi == NULL || (bi->flags & 454 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 455 != 0) 456 continue; 457 if (hdr->e_machine == bi->machine && 458 bi->interp_path != NULL && 459 /* ELF image p_filesz includes terminating zero */ 460 strlen(bi->interp_path) + 1 == interp_name_len && 461 strncmp(interp, bi->interp_path, interp_name_len) 462 == 0 && (bi->header_supported == NULL || 463 bi->header_supported(imgp, NULL, NULL))) 464 return (bi); 465 } 466 } 467 468 /* Lacking a recognized interpreter, try the default brand */ 469 for (i = 0; i < MAX_BRANDS; i++) { 470 bi = elf_brand_list[i]; 471 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 472 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 473 continue; 474 if (hdr->e_machine == bi->machine && 475 __elfN(fallback_brand) == bi->brand && 476 (bi->header_supported == NULL || 477 bi->header_supported(imgp, NULL, NULL))) 478 return (bi); 479 } 480 return (NULL); 481 } 482 483 static bool 484 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr) 485 { 486 return (hdr->e_phoff <= PAGE_SIZE && 487 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff); 488 } 489 490 static int 491 __elfN(check_header)(const Elf_Ehdr *hdr) 492 { 493 Elf_Brandinfo *bi; 494 int i; 495 496 if (!IS_ELF(*hdr) || 497 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 498 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 499 hdr->e_ident[EI_VERSION] != EV_CURRENT || 500 hdr->e_phentsize != sizeof(Elf_Phdr) || 501 hdr->e_version != ELF_TARG_VER) 502 return (ENOEXEC); 503 504 /* 505 * Make sure we have at least one brand for this machine. 506 */ 507 508 for (i = 0; i < MAX_BRANDS; i++) { 509 bi = elf_brand_list[i]; 510 if (bi != NULL && bi->machine == hdr->e_machine) 511 break; 512 } 513 if (i == MAX_BRANDS) 514 return (ENOEXEC); 515 516 return (0); 517 } 518 519 static int 520 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 521 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 522 { 523 struct sf_buf *sf; 524 int error; 525 vm_offset_t off; 526 527 /* 528 * Create the page if it doesn't exist yet. Ignore errors. 529 */ 530 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 531 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 532 533 /* 534 * Find the page from the underlying object. 535 */ 536 if (object != NULL) { 537 sf = vm_imgact_map_page(object, offset); 538 if (sf == NULL) 539 return (KERN_FAILURE); 540 off = offset - trunc_page(offset); 541 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 542 end - start); 543 vm_imgact_unmap_page(sf); 544 if (error != 0) 545 return (KERN_FAILURE); 546 } 547 548 return (KERN_SUCCESS); 549 } 550 551 static int 552 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object, 553 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot, 554 int cow) 555 { 556 struct sf_buf *sf; 557 vm_offset_t off; 558 vm_size_t sz; 559 int error, locked, rv; 560 561 if (start != trunc_page(start)) { 562 rv = __elfN(map_partial)(map, object, offset, start, 563 round_page(start), prot); 564 if (rv != KERN_SUCCESS) 565 return (rv); 566 offset += round_page(start) - start; 567 start = round_page(start); 568 } 569 if (end != round_page(end)) { 570 rv = __elfN(map_partial)(map, object, offset + 571 trunc_page(end) - start, trunc_page(end), end, prot); 572 if (rv != KERN_SUCCESS) 573 return (rv); 574 end = trunc_page(end); 575 } 576 if (start >= end) 577 return (KERN_SUCCESS); 578 if ((offset & PAGE_MASK) != 0) { 579 /* 580 * The mapping is not page aligned. This means that we have 581 * to copy the data. 582 */ 583 rv = vm_map_fixed(map, NULL, 0, start, end - start, 584 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 585 if (rv != KERN_SUCCESS) 586 return (rv); 587 if (object == NULL) 588 return (KERN_SUCCESS); 589 for (; start < end; start += sz) { 590 sf = vm_imgact_map_page(object, offset); 591 if (sf == NULL) 592 return (KERN_FAILURE); 593 off = offset - trunc_page(offset); 594 sz = end - start; 595 if (sz > PAGE_SIZE - off) 596 sz = PAGE_SIZE - off; 597 error = copyout((caddr_t)sf_buf_kva(sf) + off, 598 (caddr_t)start, sz); 599 vm_imgact_unmap_page(sf); 600 if (error != 0) 601 return (KERN_FAILURE); 602 offset += sz; 603 } 604 } else { 605 vm_object_reference(object); 606 rv = vm_map_fixed(map, object, offset, start, end - start, 607 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL | 608 (object != NULL ? MAP_VN_EXEC : 0)); 609 if (rv != KERN_SUCCESS) { 610 locked = VOP_ISLOCKED(imgp->vp); 611 VOP_UNLOCK(imgp->vp); 612 vm_object_deallocate(object); 613 vn_lock(imgp->vp, locked | LK_RETRY); 614 return (rv); 615 } else if (object != NULL) { 616 MPASS(imgp->vp->v_object == object); 617 VOP_SET_TEXT_CHECKED(imgp->vp); 618 } 619 } 620 return (KERN_SUCCESS); 621 } 622 623 static int 624 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset, 625 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot) 626 { 627 struct sf_buf *sf; 628 size_t map_len; 629 vm_map_t map; 630 vm_object_t object; 631 vm_offset_t map_addr; 632 int error, rv, cow; 633 size_t copy_len; 634 vm_ooffset_t file_addr; 635 636 /* 637 * It's necessary to fail if the filsz + offset taken from the 638 * header is greater than the actual file pager object's size. 639 * If we were to allow this, then the vm_map_find() below would 640 * walk right off the end of the file object and into the ether. 641 * 642 * While I'm here, might as well check for something else that 643 * is invalid: filsz cannot be greater than memsz. 644 */ 645 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 646 filsz > memsz) { 647 uprintf("elf_load_section: truncated ELF file\n"); 648 return (ENOEXEC); 649 } 650 651 object = imgp->object; 652 map = &imgp->proc->p_vmspace->vm_map; 653 map_addr = trunc_page((vm_offset_t)vmaddr); 654 file_addr = trunc_page(offset); 655 656 /* 657 * We have two choices. We can either clear the data in the last page 658 * of an oversized mapping, or we can start the anon mapping a page 659 * early and copy the initialized data into that first page. We 660 * choose the second. 661 */ 662 if (filsz == 0) 663 map_len = 0; 664 else if (memsz > filsz) 665 map_len = trunc_page(offset + filsz) - file_addr; 666 else 667 map_len = round_page(offset + filsz) - file_addr; 668 669 if (map_len != 0) { 670 /* cow flags: don't dump readonly sections in core */ 671 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 672 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 673 674 rv = __elfN(map_insert)(imgp, map, object, file_addr, 675 map_addr, map_addr + map_len, prot, cow); 676 if (rv != KERN_SUCCESS) 677 return (EINVAL); 678 679 /* we can stop now if we've covered it all */ 680 if (memsz == filsz) 681 return (0); 682 } 683 684 /* 685 * We have to get the remaining bit of the file into the first part 686 * of the oversized map segment. This is normally because the .data 687 * segment in the file is extended to provide bss. It's a neat idea 688 * to try and save a page, but it's a pain in the behind to implement. 689 */ 690 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset + 691 filsz); 692 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 693 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 694 695 /* This had damn well better be true! */ 696 if (map_len != 0) { 697 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 698 map_addr + map_len, prot, 0); 699 if (rv != KERN_SUCCESS) 700 return (EINVAL); 701 } 702 703 if (copy_len != 0) { 704 sf = vm_imgact_map_page(object, offset + filsz); 705 if (sf == NULL) 706 return (EIO); 707 708 /* send the page fragment to user space */ 709 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr, 710 copy_len); 711 vm_imgact_unmap_page(sf); 712 if (error != 0) 713 return (error); 714 } 715 716 /* 717 * Remove write access to the page if it was only granted by map_insert 718 * to allow copyout. 719 */ 720 if ((prot & VM_PROT_WRITE) == 0) 721 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 722 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT); 723 724 return (0); 725 } 726 727 static int 728 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr, 729 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp) 730 { 731 vm_prot_t prot; 732 u_long base_addr; 733 bool first; 734 int error, i; 735 736 ASSERT_VOP_LOCKED(imgp->vp, __func__); 737 738 base_addr = 0; 739 first = true; 740 741 for (i = 0; i < hdr->e_phnum; i++) { 742 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 743 continue; 744 745 /* Loadable segment */ 746 prot = __elfN(trans_prot)(phdr[i].p_flags); 747 error = __elfN(load_section)(imgp, phdr[i].p_offset, 748 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 749 phdr[i].p_memsz, phdr[i].p_filesz, prot); 750 if (error != 0) 751 return (error); 752 753 /* 754 * Establish the base address if this is the first segment. 755 */ 756 if (first) { 757 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 758 first = false; 759 } 760 } 761 762 if (base_addrp != NULL) 763 *base_addrp = base_addr; 764 765 return (0); 766 } 767 768 /* 769 * Load the file "file" into memory. It may be either a shared object 770 * or an executable. 771 * 772 * The "addr" reference parameter is in/out. On entry, it specifies 773 * the address where a shared object should be loaded. If the file is 774 * an executable, this value is ignored. On exit, "addr" specifies 775 * where the file was actually loaded. 776 * 777 * The "entry" reference parameter is out only. On exit, it specifies 778 * the entry point for the loaded file. 779 */ 780 static int 781 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 782 u_long *entry) 783 { 784 struct { 785 struct nameidata nd; 786 struct vattr attr; 787 struct image_params image_params; 788 } *tempdata; 789 const Elf_Ehdr *hdr = NULL; 790 const Elf_Phdr *phdr = NULL; 791 struct nameidata *nd; 792 struct vattr *attr; 793 struct image_params *imgp; 794 u_long rbase; 795 u_long base_addr = 0; 796 int error; 797 798 #ifdef CAPABILITY_MODE 799 /* 800 * XXXJA: This check can go away once we are sufficiently confident 801 * that the checks in namei() are correct. 802 */ 803 if (IN_CAPABILITY_MODE(curthread)) 804 return (ECAPMODE); 805 #endif 806 807 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO); 808 nd = &tempdata->nd; 809 attr = &tempdata->attr; 810 imgp = &tempdata->image_params; 811 812 /* 813 * Initialize part of the common data 814 */ 815 imgp->proc = p; 816 imgp->attr = attr; 817 818 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF, 819 UIO_SYSSPACE, file); 820 if ((error = namei(nd)) != 0) { 821 nd->ni_vp = NULL; 822 goto fail; 823 } 824 NDFREE_PNBUF(nd); 825 imgp->vp = nd->ni_vp; 826 827 /* 828 * Check permissions, modes, uid, etc on the file, and "open" it. 829 */ 830 error = exec_check_permissions(imgp); 831 if (error) 832 goto fail; 833 834 error = exec_map_first_page(imgp); 835 if (error) 836 goto fail; 837 838 imgp->object = nd->ni_vp->v_object; 839 840 hdr = (const Elf_Ehdr *)imgp->image_header; 841 if ((error = __elfN(check_header)(hdr)) != 0) 842 goto fail; 843 if (hdr->e_type == ET_DYN) 844 rbase = *addr; 845 else if (hdr->e_type == ET_EXEC) 846 rbase = 0; 847 else { 848 error = ENOEXEC; 849 goto fail; 850 } 851 852 /* Only support headers that fit within first page for now */ 853 if (!__elfN(phdr_in_zero_page)(hdr)) { 854 error = ENOEXEC; 855 goto fail; 856 } 857 858 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 859 if (!aligned(phdr, Elf_Addr)) { 860 error = ENOEXEC; 861 goto fail; 862 } 863 864 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr); 865 if (error != 0) 866 goto fail; 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 } 1164 1165 mapsz = 0; 1166 1167 for (i = 0; i < hdr->e_phnum; i++) { 1168 switch (phdr[i].p_type) { 1169 case PT_LOAD: 1170 if (n == 0) 1171 baddr = phdr[i].p_vaddr; 1172 if (!powerof2(phdr[i].p_align) || 1173 phdr[i].p_align > maxsalign) { 1174 uprintf("Invalid segment alignment\n"); 1175 error = ENOEXEC; 1176 goto ret; 1177 } 1178 if (phdr[i].p_align > maxalign) 1179 maxalign = phdr[i].p_align; 1180 if (mapsz + phdr[i].p_memsz < mapsz) { 1181 uprintf("Mapsize overflow\n"); 1182 error = ENOEXEC; 1183 goto ret; 1184 } 1185 mapsz += phdr[i].p_memsz; 1186 n++; 1187 1188 /* 1189 * If this segment contains the program headers, 1190 * remember their virtual address for the AT_PHDR 1191 * aux entry. Static binaries don't usually include 1192 * a PT_PHDR entry. 1193 */ 1194 if (phdr[i].p_offset == 0 && 1195 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= 1196 phdr[i].p_filesz) 1197 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 1198 break; 1199 case PT_INTERP: 1200 /* Path to interpreter */ 1201 if (interp != NULL) { 1202 uprintf("Multiple PT_INTERP headers\n"); 1203 error = ENOEXEC; 1204 goto ret; 1205 } 1206 error = __elfN(get_interp)(imgp, &phdr[i], &interp, 1207 &free_interp); 1208 if (error != 0) 1209 goto ret; 1210 break; 1211 case PT_GNU_STACK: 1212 if (__elfN(nxstack)) { 1213 imgp->stack_prot = 1214 __elfN(trans_prot)(phdr[i].p_flags); 1215 if ((imgp->stack_prot & VM_PROT_RW) != 1216 VM_PROT_RW) { 1217 uprintf("Invalid PT_GNU_STACK\n"); 1218 error = ENOEXEC; 1219 goto ret; 1220 } 1221 } 1222 imgp->stack_sz = phdr[i].p_memsz; 1223 break; 1224 case PT_PHDR: /* Program header table info */ 1225 proghdr = phdr[i].p_vaddr; 1226 break; 1227 } 1228 } 1229 1230 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0); 1231 if (brand_info == NULL) { 1232 uprintf("ELF binary type \"%u\" not known.\n", 1233 hdr->e_ident[EI_OSABI]); 1234 error = ENOEXEC; 1235 goto ret; 1236 } 1237 sv = brand_info->sysvec; 1238 if (hdr->e_type == ET_DYN) { 1239 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 1240 uprintf("Cannot execute shared object\n"); 1241 error = ENOEXEC; 1242 goto ret; 1243 } 1244 /* 1245 * Honour the base load address from the dso if it is 1246 * non-zero for some reason. 1247 */ 1248 if (baddr == 0) { 1249 if ((sv->sv_flags & SV_ASLR) == 0 || 1250 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) 1251 imgp->et_dyn_addr = __elfN(pie_base); 1252 else if ((__elfN(pie_aslr_enabled) && 1253 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) || 1254 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0) 1255 imgp->et_dyn_addr = ET_DYN_ADDR_RAND; 1256 else 1257 imgp->et_dyn_addr = __elfN(pie_base); 1258 } 1259 } 1260 1261 /* 1262 * Avoid a possible deadlock if the current address space is destroyed 1263 * and that address space maps the locked vnode. In the common case, 1264 * the locked vnode's v_usecount is decremented but remains greater 1265 * than zero. Consequently, the vnode lock is not needed by vrele(). 1266 * However, in cases where the vnode lock is external, such as nullfs, 1267 * v_usecount may become zero. 1268 * 1269 * The VV_TEXT flag prevents modifications to the executable while 1270 * the vnode is unlocked. 1271 */ 1272 VOP_UNLOCK(imgp->vp); 1273 1274 /* 1275 * Decide whether to enable randomization of user mappings. 1276 * First, reset user preferences for the setid binaries. 1277 * Then, account for the support of the randomization by the 1278 * ABI, by user preferences, and make special treatment for 1279 * PIE binaries. 1280 */ 1281 if (imgp->credential_setid) { 1282 PROC_LOCK(imgp->proc); 1283 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE | 1284 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 1285 PROC_UNLOCK(imgp->proc); 1286 } 1287 if ((sv->sv_flags & SV_ASLR) == 0 || 1288 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 || 1289 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) { 1290 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND, 1291 ("imgp->et_dyn_addr == RAND and !ASLR")); 1292 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 || 1293 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) || 1294 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1295 imgp->map_flags |= MAP_ASLR; 1296 /* 1297 * If user does not care about sbrk, utilize the bss 1298 * grow region for mappings as well. We can select 1299 * the base for the image anywere and still not suffer 1300 * from the fragmentation. 1301 */ 1302 if (!__elfN(aslr_honor_sbrk) || 1303 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0) 1304 imgp->map_flags |= MAP_ASLR_IGNSTART; 1305 if (__elfN(aslr_stack)) 1306 imgp->map_flags |= MAP_ASLR_STACK; 1307 if (__elfN(aslr_shared_page)) 1308 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE; 1309 } 1310 1311 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 && 1312 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) || 1313 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0) 1314 imgp->map_flags |= MAP_WXORX; 1315 1316 error = exec_new_vmspace(imgp, sv); 1317 1318 imgp->proc->p_sysent = sv; 1319 imgp->proc->p_elf_brandinfo = brand_info; 1320 1321 vmspace = imgp->proc->p_vmspace; 1322 map = &vmspace->vm_map; 1323 maxv = sv->sv_usrstack; 1324 if ((imgp->map_flags & MAP_ASLR_STACK) == 0) 1325 maxv -= lim_max(td, RLIMIT_STACK); 1326 if (error == 0 && mapsz >= maxv - vm_map_min(map)) { 1327 uprintf("Excessive mapping size\n"); 1328 error = ENOEXEC; 1329 } 1330 1331 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1332 KASSERT((map->flags & MAP_ASLR) != 0, 1333 ("ET_DYN_ADDR_RAND but !MAP_ASLR")); 1334 error = __CONCAT(rnd_, __elfN(base))(map, 1335 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA), 1336 /* reserve half of the address space to interpreter */ 1337 maxv / 2, maxalign, &imgp->et_dyn_addr); 1338 } 1339 1340 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1341 if (error != 0) 1342 goto ret; 1343 1344 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL); 1345 if (error != 0) 1346 goto ret; 1347 1348 error = __elfN(enforce_limits)(imgp, hdr, phdr); 1349 if (error != 0) 1350 goto ret; 1351 1352 /* 1353 * We load the dynamic linker where a userland call 1354 * to mmap(0, ...) would put it. The rationale behind this 1355 * calculation is that it leaves room for the heap to grow to 1356 * its maximum allowed size. 1357 */ 1358 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1359 RLIMIT_DATA)); 1360 if ((map->flags & MAP_ASLR) != 0) { 1361 maxv1 = maxv / 2 + addr / 2; 1362 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1, 1363 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ? 1364 pagesizes[1] : pagesizes[0], &anon_loc); 1365 if (error != 0) 1366 goto ret; 1367 map->anon_loc = anon_loc; 1368 } else { 1369 map->anon_loc = addr; 1370 } 1371 1372 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr; 1373 imgp->entry_addr = entry; 1374 1375 if (interp != NULL) { 1376 VOP_UNLOCK(imgp->vp); 1377 if ((map->flags & MAP_ASLR) != 0) { 1378 /* Assume that interpreter fits into 1/4 of AS */ 1379 maxv1 = maxv / 2 + addr / 2; 1380 error = __CONCAT(rnd_, __elfN(base))(map, addr, 1381 maxv1, PAGE_SIZE, &addr); 1382 } 1383 if (error == 0) { 1384 error = __elfN(load_interp)(imgp, brand_info, interp, 1385 &addr, &imgp->entry_addr); 1386 } 1387 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1388 if (error != 0) 1389 goto ret; 1390 } else 1391 addr = imgp->et_dyn_addr; 1392 1393 error = exec_map_stack(imgp); 1394 if (error != 0) 1395 goto ret; 1396 1397 /* 1398 * Construct auxargs table (used by the copyout_auxargs routine) 1399 */ 1400 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT); 1401 if (elf_auxargs == NULL) { 1402 VOP_UNLOCK(imgp->vp); 1403 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1404 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1405 } 1406 elf_auxargs->execfd = -1; 1407 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr; 1408 elf_auxargs->phent = hdr->e_phentsize; 1409 elf_auxargs->phnum = hdr->e_phnum; 1410 elf_auxargs->pagesz = PAGE_SIZE; 1411 elf_auxargs->base = addr; 1412 elf_auxargs->flags = 0; 1413 elf_auxargs->entry = entry; 1414 elf_auxargs->hdr_eflags = hdr->e_flags; 1415 1416 imgp->auxargs = elf_auxargs; 1417 imgp->interpreted = 0; 1418 imgp->reloc_base = addr; 1419 imgp->proc->p_osrel = osrel; 1420 imgp->proc->p_fctl0 = fctl0; 1421 imgp->proc->p_elf_flags = hdr->e_flags; 1422 1423 ret: 1424 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock"); 1425 if (free_interp) 1426 free(interp, M_TEMP); 1427 return (error); 1428 } 1429 1430 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE) 1431 1432 int 1433 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base) 1434 { 1435 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1436 Elf_Auxinfo *argarray, *pos; 1437 struct vmspace *vmspace; 1438 rlim_t stacksz; 1439 int error, bsdflags, oc; 1440 1441 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1442 M_WAITOK | M_ZERO); 1443 1444 vmspace = imgp->proc->p_vmspace; 1445 1446 if (args->execfd != -1) 1447 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1448 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1449 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1450 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1451 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1452 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1453 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1454 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1455 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1456 if (imgp->execpathp != 0) 1457 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp); 1458 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1459 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1460 if (imgp->canary != 0) { 1461 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary); 1462 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1463 } 1464 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1465 if (imgp->pagesizes != 0) { 1466 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes); 1467 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1468 } 1469 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) { 1470 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1471 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset); 1472 } 1473 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1474 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1475 imgp->sysent->sv_stackprot); 1476 if (imgp->sysent->sv_hwcap != NULL) 1477 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1478 if (imgp->sysent->sv_hwcap2 != NULL) 1479 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1480 bsdflags = 0; 1481 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0; 1482 oc = atomic_load_int(&vm_overcommit); 1483 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) != 1484 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0; 1485 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags); 1486 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc); 1487 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv); 1488 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc); 1489 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv); 1490 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings); 1491 #ifdef RANDOM_FENESTRASX 1492 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) { 1493 AUXARGS_ENTRY(pos, AT_FXRNG, 1494 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset); 1495 } 1496 #endif 1497 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) { 1498 AUXARGS_ENTRY(pos, AT_KPRELOAD, 1499 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset); 1500 } 1501 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop)); 1502 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur; 1503 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz); 1504 AUXARGS_ENTRY(pos, AT_NULL, 0); 1505 1506 free(imgp->auxargs, M_TEMP); 1507 imgp->auxargs = NULL; 1508 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1509 1510 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT); 1511 free(argarray, M_TEMP); 1512 return (error); 1513 } 1514 1515 int 1516 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp) 1517 { 1518 Elf_Addr *base; 1519 1520 base = (Elf_Addr *)*stack_base; 1521 base--; 1522 if (elf_suword(base, imgp->args->argc) == -1) 1523 return (EFAULT); 1524 *stack_base = (uintptr_t)base; 1525 return (0); 1526 } 1527 1528 /* 1529 * Code for generating ELF core dumps. 1530 */ 1531 1532 typedef void (*segment_callback)(vm_map_entry_t, void *); 1533 1534 /* Closure for cb_put_phdr(). */ 1535 struct phdr_closure { 1536 Elf_Phdr *phdr; /* Program header to fill in */ 1537 Elf_Off offset; /* Offset of segment in core file */ 1538 }; 1539 1540 struct note_info { 1541 int type; /* Note type. */ 1542 struct regset *regset; /* Register set. */ 1543 outfunc_t outfunc; /* Output function. */ 1544 void *outarg; /* Argument for the output function. */ 1545 size_t outsize; /* Output size. */ 1546 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1547 }; 1548 1549 TAILQ_HEAD(note_info_list, note_info); 1550 1551 extern int compress_user_cores; 1552 extern int compress_user_cores_level; 1553 1554 static void cb_put_phdr(vm_map_entry_t, void *); 1555 static void cb_size_segment(vm_map_entry_t, void *); 1556 static void each_dumpable_segment(struct thread *, segment_callback, void *, 1557 int); 1558 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1559 struct note_info_list *, size_t, int); 1560 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *); 1561 1562 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1563 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1564 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1565 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1566 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1567 static void note_procstat_files(void *, struct sbuf *, size_t *); 1568 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1569 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1570 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1571 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1572 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1573 1574 static int 1575 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1576 { 1577 1578 return (core_write((struct coredump_params *)arg, base, len, offset, 1579 UIO_SYSSPACE, NULL)); 1580 } 1581 1582 int 1583 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1584 { 1585 struct ucred *cred = td->td_ucred; 1586 int compm, error = 0; 1587 struct sseg_closure seginfo; 1588 struct note_info_list notelst; 1589 struct coredump_params params; 1590 struct note_info *ninfo; 1591 void *hdr, *tmpbuf; 1592 size_t hdrsize, notesz, coresize; 1593 1594 hdr = NULL; 1595 tmpbuf = NULL; 1596 TAILQ_INIT(¬elst); 1597 1598 /* Size the program segments. */ 1599 __elfN(size_segments)(td, &seginfo, flags); 1600 1601 /* 1602 * Collect info about the core file header area. 1603 */ 1604 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1605 if (seginfo.count + 1 >= PN_XNUM) 1606 hdrsize += sizeof(Elf_Shdr); 1607 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz); 1608 coresize = round_page(hdrsize + notesz) + seginfo.size; 1609 1610 /* Set up core dump parameters. */ 1611 params.offset = 0; 1612 params.active_cred = cred; 1613 params.file_cred = NOCRED; 1614 params.td = td; 1615 params.vp = vp; 1616 params.comp = NULL; 1617 1618 #ifdef RACCT 1619 if (racct_enable) { 1620 PROC_LOCK(td->td_proc); 1621 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1622 PROC_UNLOCK(td->td_proc); 1623 if (error != 0) { 1624 error = EFAULT; 1625 goto done; 1626 } 1627 } 1628 #endif 1629 if (coresize >= limit) { 1630 error = EFAULT; 1631 goto done; 1632 } 1633 1634 /* Create a compression stream if necessary. */ 1635 compm = compress_user_cores; 1636 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP && 1637 compm == 0) 1638 compm = COMPRESS_GZIP; 1639 if (compm != 0) { 1640 params.comp = compressor_init(core_compressed_write, 1641 compm, CORE_BUF_SIZE, 1642 compress_user_cores_level, ¶ms); 1643 if (params.comp == NULL) { 1644 error = EFAULT; 1645 goto done; 1646 } 1647 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1648 } 1649 1650 /* 1651 * Allocate memory for building the header, fill it up, 1652 * and write it out following the notes. 1653 */ 1654 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1655 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1656 notesz, flags); 1657 1658 /* Write the contents of all of the writable segments. */ 1659 if (error == 0) { 1660 Elf_Phdr *php; 1661 off_t offset; 1662 int i; 1663 1664 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1665 offset = round_page(hdrsize + notesz); 1666 for (i = 0; i < seginfo.count; i++) { 1667 error = core_output((char *)(uintptr_t)php->p_vaddr, 1668 php->p_filesz, offset, ¶ms, tmpbuf); 1669 if (error != 0) 1670 break; 1671 offset += php->p_filesz; 1672 php++; 1673 } 1674 if (error == 0 && params.comp != NULL) 1675 error = compressor_flush(params.comp); 1676 } 1677 if (error) { 1678 log(LOG_WARNING, 1679 "Failed to write core file for process %s (error %d)\n", 1680 curproc->p_comm, error); 1681 } 1682 1683 done: 1684 free(tmpbuf, M_TEMP); 1685 if (params.comp != NULL) 1686 compressor_fini(params.comp); 1687 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1688 TAILQ_REMOVE(¬elst, ninfo, link); 1689 free(ninfo, M_TEMP); 1690 } 1691 if (hdr != NULL) 1692 free(hdr, M_TEMP); 1693 1694 return (error); 1695 } 1696 1697 /* 1698 * A callback for each_dumpable_segment() to write out the segment's 1699 * program header entry. 1700 */ 1701 static void 1702 cb_put_phdr(vm_map_entry_t entry, void *closure) 1703 { 1704 struct phdr_closure *phc = (struct phdr_closure *)closure; 1705 Elf_Phdr *phdr = phc->phdr; 1706 1707 phc->offset = round_page(phc->offset); 1708 1709 phdr->p_type = PT_LOAD; 1710 phdr->p_offset = phc->offset; 1711 phdr->p_vaddr = entry->start; 1712 phdr->p_paddr = 0; 1713 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1714 phdr->p_align = PAGE_SIZE; 1715 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1716 1717 phc->offset += phdr->p_filesz; 1718 phc->phdr++; 1719 } 1720 1721 /* 1722 * A callback for each_dumpable_segment() to gather information about 1723 * the number of segments and their total size. 1724 */ 1725 static void 1726 cb_size_segment(vm_map_entry_t entry, void *closure) 1727 { 1728 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1729 1730 ssc->count++; 1731 ssc->size += entry->end - entry->start; 1732 } 1733 1734 void 1735 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo, 1736 int flags) 1737 { 1738 seginfo->count = 0; 1739 seginfo->size = 0; 1740 1741 each_dumpable_segment(td, cb_size_segment, seginfo, flags); 1742 } 1743 1744 /* 1745 * For each writable segment in the process's memory map, call the given 1746 * function with a pointer to the map entry and some arbitrary 1747 * caller-supplied data. 1748 */ 1749 static void 1750 each_dumpable_segment(struct thread *td, segment_callback func, void *closure, 1751 int flags) 1752 { 1753 struct proc *p = td->td_proc; 1754 vm_map_t map = &p->p_vmspace->vm_map; 1755 vm_map_entry_t entry; 1756 vm_object_t backing_object, object; 1757 bool ignore_entry; 1758 1759 vm_map_lock_read(map); 1760 VM_MAP_ENTRY_FOREACH(entry, map) { 1761 /* 1762 * Don't dump inaccessible mappings, deal with legacy 1763 * coredump mode. 1764 * 1765 * Note that read-only segments related to the elf binary 1766 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1767 * need to arbitrarily ignore such segments. 1768 */ 1769 if ((flags & SVC_ALL) == 0) { 1770 if (elf_legacy_coredump) { 1771 if ((entry->protection & VM_PROT_RW) != 1772 VM_PROT_RW) 1773 continue; 1774 } else { 1775 if ((entry->protection & VM_PROT_ALL) == 0) 1776 continue; 1777 } 1778 } 1779 1780 /* 1781 * Dont include memory segment in the coredump if 1782 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1783 * madvise(2). Do not dump submaps (i.e. parts of the 1784 * kernel map). 1785 */ 1786 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 1787 continue; 1788 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 && 1789 (flags & SVC_ALL) == 0) 1790 continue; 1791 if ((object = entry->object.vm_object) == NULL) 1792 continue; 1793 1794 /* Ignore memory-mapped devices and such things. */ 1795 VM_OBJECT_RLOCK(object); 1796 while ((backing_object = object->backing_object) != NULL) { 1797 VM_OBJECT_RLOCK(backing_object); 1798 VM_OBJECT_RUNLOCK(object); 1799 object = backing_object; 1800 } 1801 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0; 1802 VM_OBJECT_RUNLOCK(object); 1803 if (ignore_entry) 1804 continue; 1805 1806 (*func)(entry, closure); 1807 } 1808 vm_map_unlock_read(map); 1809 } 1810 1811 /* 1812 * Write the core file header to the file, including padding up to 1813 * the page boundary. 1814 */ 1815 static int 1816 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1817 size_t hdrsize, struct note_info_list *notelst, size_t notesz, 1818 int flags) 1819 { 1820 struct note_info *ninfo; 1821 struct sbuf *sb; 1822 int error; 1823 1824 /* Fill in the header. */ 1825 bzero(hdr, hdrsize); 1826 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags); 1827 1828 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1829 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1830 sbuf_start_section(sb, NULL); 1831 sbuf_bcat(sb, hdr, hdrsize); 1832 TAILQ_FOREACH(ninfo, notelst, link) 1833 __elfN(putnote)(p->td, ninfo, sb); 1834 /* Align up to a page boundary for the program segments. */ 1835 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1836 error = sbuf_finish(sb); 1837 sbuf_delete(sb); 1838 1839 return (error); 1840 } 1841 1842 void 1843 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1844 size_t *sizep) 1845 { 1846 struct proc *p; 1847 struct thread *thr; 1848 size_t size; 1849 1850 p = td->td_proc; 1851 size = 0; 1852 1853 size += __elfN(register_note)(td, list, NT_PRPSINFO, 1854 __elfN(note_prpsinfo), p); 1855 1856 /* 1857 * To have the debugger select the right thread (LWP) as the initial 1858 * thread, we dump the state of the thread passed to us in td first. 1859 * This is the thread that causes the core dump and thus likely to 1860 * be the right thread one wants to have selected in the debugger. 1861 */ 1862 thr = td; 1863 while (thr != NULL) { 1864 size += __elfN(prepare_register_notes)(td, list, thr); 1865 size += __elfN(register_note)(td, list, -1, 1866 __elfN(note_threadmd), thr); 1867 1868 thr = thr == td ? TAILQ_FIRST(&p->p_threads) : 1869 TAILQ_NEXT(thr, td_plist); 1870 if (thr == td) 1871 thr = TAILQ_NEXT(thr, td_plist); 1872 } 1873 1874 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC, 1875 __elfN(note_procstat_proc), p); 1876 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES, 1877 note_procstat_files, p); 1878 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP, 1879 note_procstat_vmmap, p); 1880 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS, 1881 note_procstat_groups, p); 1882 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK, 1883 note_procstat_umask, p); 1884 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT, 1885 note_procstat_rlimit, p); 1886 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL, 1887 note_procstat_osrel, p); 1888 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS, 1889 __elfN(note_procstat_psstrings), p); 1890 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV, 1891 __elfN(note_procstat_auxv), p); 1892 1893 *sizep = size; 1894 } 1895 1896 void 1897 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1898 size_t notesz, int flags) 1899 { 1900 Elf_Ehdr *ehdr; 1901 Elf_Phdr *phdr; 1902 Elf_Shdr *shdr; 1903 struct phdr_closure phc; 1904 Elf_Brandinfo *bi; 1905 1906 ehdr = (Elf_Ehdr *)hdr; 1907 bi = td->td_proc->p_elf_brandinfo; 1908 1909 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1910 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1911 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1912 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1913 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1914 ehdr->e_ident[EI_DATA] = ELF_DATA; 1915 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1916 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi; 1917 ehdr->e_ident[EI_ABIVERSION] = 0; 1918 ehdr->e_ident[EI_PAD] = 0; 1919 ehdr->e_type = ET_CORE; 1920 ehdr->e_machine = bi->machine; 1921 ehdr->e_version = EV_CURRENT; 1922 ehdr->e_entry = 0; 1923 ehdr->e_phoff = sizeof(Elf_Ehdr); 1924 ehdr->e_flags = td->td_proc->p_elf_flags; 1925 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1926 ehdr->e_phentsize = sizeof(Elf_Phdr); 1927 ehdr->e_shentsize = sizeof(Elf_Shdr); 1928 ehdr->e_shstrndx = SHN_UNDEF; 1929 if (numsegs + 1 < PN_XNUM) { 1930 ehdr->e_phnum = numsegs + 1; 1931 ehdr->e_shnum = 0; 1932 } else { 1933 ehdr->e_phnum = PN_XNUM; 1934 ehdr->e_shnum = 1; 1935 1936 ehdr->e_shoff = ehdr->e_phoff + 1937 (numsegs + 1) * ehdr->e_phentsize; 1938 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1939 ("e_shoff: %zu, hdrsize - shdr: %zu", 1940 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1941 1942 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1943 memset(shdr, 0, sizeof(*shdr)); 1944 /* 1945 * A special first section is used to hold large segment and 1946 * section counts. This was proposed by Sun Microsystems in 1947 * Solaris and has been adopted by Linux; the standard ELF 1948 * tools are already familiar with the technique. 1949 * 1950 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1951 * (or 12-7 depending on the version of the document) for more 1952 * details. 1953 */ 1954 shdr->sh_type = SHT_NULL; 1955 shdr->sh_size = ehdr->e_shnum; 1956 shdr->sh_link = ehdr->e_shstrndx; 1957 shdr->sh_info = numsegs + 1; 1958 } 1959 1960 /* 1961 * Fill in the program header entries. 1962 */ 1963 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1964 1965 /* The note segement. */ 1966 phdr->p_type = PT_NOTE; 1967 phdr->p_offset = hdrsize; 1968 phdr->p_vaddr = 0; 1969 phdr->p_paddr = 0; 1970 phdr->p_filesz = notesz; 1971 phdr->p_memsz = 0; 1972 phdr->p_flags = PF_R; 1973 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1974 phdr++; 1975 1976 /* All the writable segments from the program. */ 1977 phc.phdr = phdr; 1978 phc.offset = round_page(hdrsize + notesz); 1979 each_dumpable_segment(td, cb_put_phdr, &phc, flags); 1980 } 1981 1982 static size_t 1983 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list, 1984 struct regset *regset, struct thread *target_td) 1985 { 1986 const struct sysentvec *sv; 1987 struct note_info *ninfo; 1988 size_t size, notesize; 1989 1990 size = 0; 1991 if (!regset->get(regset, target_td, NULL, &size) || size == 0) 1992 return (0); 1993 1994 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1995 ninfo->type = regset->note; 1996 ninfo->regset = regset; 1997 ninfo->outarg = target_td; 1998 ninfo->outsize = size; 1999 TAILQ_INSERT_TAIL(list, ninfo, link); 2000 2001 sv = td->td_proc->p_sysent; 2002 notesize = sizeof(Elf_Note) + /* note header */ 2003 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2004 /* note name */ 2005 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2006 2007 return (notesize); 2008 } 2009 2010 size_t 2011 __elfN(register_note)(struct thread *td, struct note_info_list *list, 2012 int type, outfunc_t out, void *arg) 2013 { 2014 const struct sysentvec *sv; 2015 struct note_info *ninfo; 2016 size_t size, notesize; 2017 2018 sv = td->td_proc->p_sysent; 2019 size = 0; 2020 out(arg, NULL, &size); 2021 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2022 ninfo->type = type; 2023 ninfo->outfunc = out; 2024 ninfo->outarg = arg; 2025 ninfo->outsize = size; 2026 TAILQ_INSERT_TAIL(list, ninfo, link); 2027 2028 if (type == -1) 2029 return (size); 2030 2031 notesize = sizeof(Elf_Note) + /* note header */ 2032 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2033 /* note name */ 2034 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2035 2036 return (notesize); 2037 } 2038 2039 static size_t 2040 append_note_data(const void *src, void *dst, size_t len) 2041 { 2042 size_t padded_len; 2043 2044 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 2045 if (dst != NULL) { 2046 bcopy(src, dst, len); 2047 bzero((char *)dst + len, padded_len - len); 2048 } 2049 return (padded_len); 2050 } 2051 2052 size_t 2053 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 2054 { 2055 Elf_Note *note; 2056 char *buf; 2057 size_t notesize; 2058 2059 buf = dst; 2060 if (buf != NULL) { 2061 note = (Elf_Note *)buf; 2062 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 2063 note->n_descsz = size; 2064 note->n_type = type; 2065 buf += sizeof(*note); 2066 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 2067 sizeof(FREEBSD_ABI_VENDOR)); 2068 append_note_data(src, buf, size); 2069 if (descp != NULL) 2070 *descp = buf; 2071 } 2072 2073 notesize = sizeof(Elf_Note) + /* note header */ 2074 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 2075 /* note name */ 2076 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2077 2078 return (notesize); 2079 } 2080 2081 static void 2082 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb) 2083 { 2084 Elf_Note note; 2085 const struct sysentvec *sv; 2086 ssize_t old_len, sect_len; 2087 size_t new_len, descsz, i; 2088 2089 if (ninfo->type == -1) { 2090 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2091 return; 2092 } 2093 2094 sv = td->td_proc->p_sysent; 2095 2096 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1; 2097 note.n_descsz = ninfo->outsize; 2098 note.n_type = ninfo->type; 2099 2100 sbuf_bcat(sb, ¬e, sizeof(note)); 2101 sbuf_start_section(sb, &old_len); 2102 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor, 2103 strlen(sv->sv_elf_core_abi_vendor) + 1); 2104 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2105 if (note.n_descsz == 0) 2106 return; 2107 sbuf_start_section(sb, &old_len); 2108 if (ninfo->regset != NULL) { 2109 struct regset *regset = ninfo->regset; 2110 void *buf; 2111 2112 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK); 2113 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize); 2114 sbuf_bcat(sb, buf, ninfo->outsize); 2115 free(buf, M_TEMP); 2116 } else 2117 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2118 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2119 if (sect_len < 0) 2120 return; 2121 2122 new_len = (size_t)sect_len; 2123 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 2124 if (new_len < descsz) { 2125 /* 2126 * It is expected that individual note emitters will correctly 2127 * predict their expected output size and fill up to that size 2128 * themselves, padding in a format-specific way if needed. 2129 * However, in case they don't, just do it here with zeros. 2130 */ 2131 for (i = 0; i < descsz - new_len; i++) 2132 sbuf_putc(sb, 0); 2133 } else if (new_len > descsz) { 2134 /* 2135 * We can't always truncate sb -- we may have drained some 2136 * of it already. 2137 */ 2138 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 2139 "read it (%zu > %zu). Since it is longer than " 2140 "expected, this coredump's notes are corrupt. THIS " 2141 "IS A BUG in the note_procstat routine for type %u.\n", 2142 __func__, (unsigned)note.n_type, new_len, descsz, 2143 (unsigned)note.n_type)); 2144 } 2145 } 2146 2147 /* 2148 * Miscellaneous note out functions. 2149 */ 2150 2151 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2152 #include <compat/freebsd32/freebsd32.h> 2153 #include <compat/freebsd32/freebsd32_signal.h> 2154 2155 typedef struct prstatus32 elf_prstatus_t; 2156 typedef struct prpsinfo32 elf_prpsinfo_t; 2157 typedef struct fpreg32 elf_prfpregset_t; 2158 typedef struct fpreg32 elf_fpregset_t; 2159 typedef struct reg32 elf_gregset_t; 2160 typedef struct thrmisc32 elf_thrmisc_t; 2161 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t; 2162 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 2163 typedef struct kinfo_proc32 elf_kinfo_proc_t; 2164 typedef uint32_t elf_ps_strings_t; 2165 #else 2166 typedef prstatus_t elf_prstatus_t; 2167 typedef prpsinfo_t elf_prpsinfo_t; 2168 typedef prfpregset_t elf_prfpregset_t; 2169 typedef prfpregset_t elf_fpregset_t; 2170 typedef gregset_t elf_gregset_t; 2171 typedef thrmisc_t elf_thrmisc_t; 2172 typedef struct ptrace_lwpinfo elf_lwpinfo_t; 2173 #define ELF_KERN_PROC_MASK 0 2174 typedef struct kinfo_proc elf_kinfo_proc_t; 2175 typedef vm_offset_t elf_ps_strings_t; 2176 #endif 2177 2178 static void 2179 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2180 { 2181 struct sbuf sbarg; 2182 size_t len; 2183 char *cp, *end; 2184 struct proc *p; 2185 elf_prpsinfo_t *psinfo; 2186 int error; 2187 2188 p = arg; 2189 if (sb != NULL) { 2190 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 2191 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 2192 psinfo->pr_version = PRPSINFO_VERSION; 2193 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 2194 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 2195 PROC_LOCK(p); 2196 if (p->p_args != NULL) { 2197 len = sizeof(psinfo->pr_psargs) - 1; 2198 if (len > p->p_args->ar_length) 2199 len = p->p_args->ar_length; 2200 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 2201 PROC_UNLOCK(p); 2202 error = 0; 2203 } else { 2204 _PHOLD(p); 2205 PROC_UNLOCK(p); 2206 sbuf_new(&sbarg, psinfo->pr_psargs, 2207 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 2208 error = proc_getargv(curthread, p, &sbarg); 2209 PRELE(p); 2210 if (sbuf_finish(&sbarg) == 0) { 2211 len = sbuf_len(&sbarg); 2212 if (len > 0) 2213 len--; 2214 } else { 2215 len = sizeof(psinfo->pr_psargs) - 1; 2216 } 2217 sbuf_delete(&sbarg); 2218 } 2219 if (error != 0 || len == 0 || (ssize_t)len == -1) 2220 strlcpy(psinfo->pr_psargs, p->p_comm, 2221 sizeof(psinfo->pr_psargs)); 2222 else { 2223 KASSERT(len < sizeof(psinfo->pr_psargs), 2224 ("len is too long: %zu vs %zu", len, 2225 sizeof(psinfo->pr_psargs))); 2226 cp = psinfo->pr_psargs; 2227 end = cp + len - 1; 2228 for (;;) { 2229 cp = memchr(cp, '\0', end - cp); 2230 if (cp == NULL) 2231 break; 2232 *cp = ' '; 2233 } 2234 } 2235 psinfo->pr_pid = p->p_pid; 2236 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 2237 free(psinfo, M_TEMP); 2238 } 2239 *sizep = sizeof(*psinfo); 2240 } 2241 2242 static bool 2243 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf, 2244 size_t *sizep) 2245 { 2246 elf_prstatus_t *status; 2247 2248 if (buf != NULL) { 2249 KASSERT(*sizep == sizeof(*status), ("%s: invalid size", 2250 __func__)); 2251 status = buf; 2252 memset(status, 0, *sizep); 2253 status->pr_version = PRSTATUS_VERSION; 2254 status->pr_statussz = sizeof(elf_prstatus_t); 2255 status->pr_gregsetsz = sizeof(elf_gregset_t); 2256 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 2257 status->pr_osreldate = osreldate; 2258 status->pr_cursig = td->td_proc->p_sig; 2259 status->pr_pid = td->td_tid; 2260 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2261 fill_regs32(td, &status->pr_reg); 2262 #else 2263 fill_regs(td, &status->pr_reg); 2264 #endif 2265 } 2266 *sizep = sizeof(*status); 2267 return (true); 2268 } 2269 2270 static bool 2271 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf, 2272 size_t size) 2273 { 2274 elf_prstatus_t *status; 2275 2276 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__)); 2277 status = buf; 2278 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2279 set_regs32(td, &status->pr_reg); 2280 #else 2281 set_regs(td, &status->pr_reg); 2282 #endif 2283 return (true); 2284 } 2285 2286 static struct regset __elfN(regset_prstatus) = { 2287 .note = NT_PRSTATUS, 2288 .size = sizeof(elf_prstatus_t), 2289 .get = __elfN(get_prstatus), 2290 .set = __elfN(set_prstatus), 2291 }; 2292 ELF_REGSET(__elfN(regset_prstatus)); 2293 2294 static bool 2295 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf, 2296 size_t *sizep) 2297 { 2298 elf_prfpregset_t *fpregset; 2299 2300 if (buf != NULL) { 2301 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size", 2302 __func__)); 2303 fpregset = buf; 2304 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2305 fill_fpregs32(td, fpregset); 2306 #else 2307 fill_fpregs(td, fpregset); 2308 #endif 2309 } 2310 *sizep = sizeof(*fpregset); 2311 return (true); 2312 } 2313 2314 static bool 2315 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf, 2316 size_t size) 2317 { 2318 elf_prfpregset_t *fpregset; 2319 2320 fpregset = buf; 2321 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__)); 2322 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2323 set_fpregs32(td, fpregset); 2324 #else 2325 set_fpregs(td, fpregset); 2326 #endif 2327 return (true); 2328 } 2329 2330 static struct regset __elfN(regset_fpregset) = { 2331 .note = NT_FPREGSET, 2332 .size = sizeof(elf_prfpregset_t), 2333 .get = __elfN(get_fpregset), 2334 .set = __elfN(set_fpregset), 2335 }; 2336 ELF_REGSET(__elfN(regset_fpregset)); 2337 2338 static bool 2339 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf, 2340 size_t *sizep) 2341 { 2342 elf_thrmisc_t *thrmisc; 2343 2344 if (buf != NULL) { 2345 KASSERT(*sizep == sizeof(*thrmisc), 2346 ("%s: invalid size", __func__)); 2347 thrmisc = buf; 2348 bzero(thrmisc, sizeof(*thrmisc)); 2349 strcpy(thrmisc->pr_tname, td->td_name); 2350 } 2351 *sizep = sizeof(*thrmisc); 2352 return (true); 2353 } 2354 2355 static struct regset __elfN(regset_thrmisc) = { 2356 .note = NT_THRMISC, 2357 .size = sizeof(elf_thrmisc_t), 2358 .get = __elfN(get_thrmisc), 2359 }; 2360 ELF_REGSET(__elfN(regset_thrmisc)); 2361 2362 static bool 2363 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf, 2364 size_t *sizep) 2365 { 2366 elf_lwpinfo_t pl; 2367 size_t size; 2368 int structsize; 2369 2370 size = sizeof(structsize) + sizeof(pl); 2371 if (buf != NULL) { 2372 KASSERT(*sizep == size, ("%s: invalid size", __func__)); 2373 structsize = sizeof(pl); 2374 memcpy(buf, &structsize, sizeof(structsize)); 2375 bzero(&pl, sizeof(pl)); 2376 pl.pl_lwpid = td->td_tid; 2377 pl.pl_event = PL_EVENT_NONE; 2378 pl.pl_sigmask = td->td_sigmask; 2379 pl.pl_siglist = td->td_siglist; 2380 if (td->td_si.si_signo != 0) { 2381 pl.pl_event = PL_EVENT_SIGNAL; 2382 pl.pl_flags |= PL_FLAG_SI; 2383 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2384 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2385 #else 2386 pl.pl_siginfo = td->td_si; 2387 #endif 2388 } 2389 strcpy(pl.pl_tdname, td->td_name); 2390 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2391 memcpy((int *)buf + 1, &pl, sizeof(pl)); 2392 } 2393 *sizep = size; 2394 return (true); 2395 } 2396 2397 static struct regset __elfN(regset_lwpinfo) = { 2398 .note = NT_PTLWPINFO, 2399 .size = sizeof(int) + sizeof(elf_lwpinfo_t), 2400 .get = __elfN(get_lwpinfo), 2401 }; 2402 ELF_REGSET(__elfN(regset_lwpinfo)); 2403 2404 static size_t 2405 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list, 2406 struct thread *target_td) 2407 { 2408 struct sysentvec *sv = td->td_proc->p_sysent; 2409 struct regset **regsetp, **regset_end, *regset; 2410 size_t size; 2411 2412 size = 0; 2413 2414 /* NT_PRSTATUS must be the first register set note. */ 2415 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus), 2416 target_td); 2417 2418 regsetp = sv->sv_regset_begin; 2419 if (regsetp == NULL) { 2420 /* XXX: This shouldn't be true for any FreeBSD ABIs. */ 2421 size += __elfN(register_regset_note)(td, list, 2422 &__elfN(regset_fpregset), target_td); 2423 return (size); 2424 } 2425 regset_end = sv->sv_regset_end; 2426 MPASS(regset_end != NULL); 2427 for (; regsetp < regset_end; regsetp++) { 2428 regset = *regsetp; 2429 if (regset->note == NT_PRSTATUS) 2430 continue; 2431 size += __elfN(register_regset_note)(td, list, regset, 2432 target_td); 2433 } 2434 return (size); 2435 } 2436 2437 /* 2438 * Allow for MD specific notes, as well as any MD 2439 * specific preparations for writing MI notes. 2440 */ 2441 static void 2442 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2443 { 2444 struct thread *td; 2445 void *buf; 2446 size_t size; 2447 2448 td = (struct thread *)arg; 2449 size = *sizep; 2450 if (size != 0 && sb != NULL) 2451 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2452 else 2453 buf = NULL; 2454 size = 0; 2455 __elfN(dump_thread)(td, buf, &size); 2456 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2457 if (size != 0 && sb != NULL) 2458 sbuf_bcat(sb, buf, size); 2459 free(buf, M_TEMP); 2460 *sizep = size; 2461 } 2462 2463 #ifdef KINFO_PROC_SIZE 2464 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2465 #endif 2466 2467 static void 2468 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2469 { 2470 struct proc *p; 2471 size_t size; 2472 int structsize; 2473 2474 p = arg; 2475 size = sizeof(structsize) + p->p_numthreads * 2476 sizeof(elf_kinfo_proc_t); 2477 2478 if (sb != NULL) { 2479 KASSERT(*sizep == size, ("invalid size")); 2480 structsize = sizeof(elf_kinfo_proc_t); 2481 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2482 sx_slock(&proctree_lock); 2483 PROC_LOCK(p); 2484 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2485 sx_sunlock(&proctree_lock); 2486 } 2487 *sizep = size; 2488 } 2489 2490 #ifdef KINFO_FILE_SIZE 2491 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2492 #endif 2493 2494 static void 2495 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2496 { 2497 struct proc *p; 2498 size_t size, sect_sz, i; 2499 ssize_t start_len, sect_len; 2500 int structsize, filedesc_flags; 2501 2502 if (coredump_pack_fileinfo) 2503 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2504 else 2505 filedesc_flags = 0; 2506 2507 p = arg; 2508 structsize = sizeof(struct kinfo_file); 2509 if (sb == NULL) { 2510 size = 0; 2511 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2512 sbuf_set_drain(sb, sbuf_count_drain, &size); 2513 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2514 PROC_LOCK(p); 2515 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2516 sbuf_finish(sb); 2517 sbuf_delete(sb); 2518 *sizep = size; 2519 } else { 2520 sbuf_start_section(sb, &start_len); 2521 2522 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2523 PROC_LOCK(p); 2524 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2525 filedesc_flags); 2526 2527 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2528 if (sect_len < 0) 2529 return; 2530 sect_sz = sect_len; 2531 2532 KASSERT(sect_sz <= *sizep, 2533 ("kern_proc_filedesc_out did not respect maxlen; " 2534 "requested %zu, got %zu", *sizep - sizeof(structsize), 2535 sect_sz - sizeof(structsize))); 2536 2537 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2538 sbuf_putc(sb, 0); 2539 } 2540 } 2541 2542 #ifdef KINFO_VMENTRY_SIZE 2543 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2544 #endif 2545 2546 static void 2547 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2548 { 2549 struct proc *p; 2550 size_t size; 2551 int structsize, vmmap_flags; 2552 2553 if (coredump_pack_vmmapinfo) 2554 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2555 else 2556 vmmap_flags = 0; 2557 2558 p = arg; 2559 structsize = sizeof(struct kinfo_vmentry); 2560 if (sb == NULL) { 2561 size = 0; 2562 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2563 sbuf_set_drain(sb, sbuf_count_drain, &size); 2564 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2565 PROC_LOCK(p); 2566 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2567 sbuf_finish(sb); 2568 sbuf_delete(sb); 2569 *sizep = size; 2570 } else { 2571 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2572 PROC_LOCK(p); 2573 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2574 vmmap_flags); 2575 } 2576 } 2577 2578 static void 2579 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2580 { 2581 struct proc *p; 2582 size_t size; 2583 int structsize; 2584 2585 p = arg; 2586 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2587 if (sb != NULL) { 2588 KASSERT(*sizep == size, ("invalid size")); 2589 structsize = sizeof(gid_t); 2590 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2591 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2592 sizeof(gid_t)); 2593 } 2594 *sizep = size; 2595 } 2596 2597 static void 2598 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2599 { 2600 struct proc *p; 2601 size_t size; 2602 int structsize; 2603 2604 p = arg; 2605 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask); 2606 if (sb != NULL) { 2607 KASSERT(*sizep == size, ("invalid size")); 2608 structsize = sizeof(p->p_pd->pd_cmask); 2609 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2610 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask)); 2611 } 2612 *sizep = size; 2613 } 2614 2615 static void 2616 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2617 { 2618 struct proc *p; 2619 struct rlimit rlim[RLIM_NLIMITS]; 2620 size_t size; 2621 int structsize, i; 2622 2623 p = arg; 2624 size = sizeof(structsize) + sizeof(rlim); 2625 if (sb != NULL) { 2626 KASSERT(*sizep == size, ("invalid size")); 2627 structsize = sizeof(rlim); 2628 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2629 PROC_LOCK(p); 2630 for (i = 0; i < RLIM_NLIMITS; i++) 2631 lim_rlimit_proc(p, i, &rlim[i]); 2632 PROC_UNLOCK(p); 2633 sbuf_bcat(sb, rlim, sizeof(rlim)); 2634 } 2635 *sizep = size; 2636 } 2637 2638 static void 2639 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2640 { 2641 struct proc *p; 2642 size_t size; 2643 int structsize; 2644 2645 p = arg; 2646 size = sizeof(structsize) + sizeof(p->p_osrel); 2647 if (sb != NULL) { 2648 KASSERT(*sizep == size, ("invalid size")); 2649 structsize = sizeof(p->p_osrel); 2650 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2651 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2652 } 2653 *sizep = size; 2654 } 2655 2656 static void 2657 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2658 { 2659 struct proc *p; 2660 elf_ps_strings_t ps_strings; 2661 size_t size; 2662 int structsize; 2663 2664 p = arg; 2665 size = sizeof(structsize) + sizeof(ps_strings); 2666 if (sb != NULL) { 2667 KASSERT(*sizep == size, ("invalid size")); 2668 structsize = sizeof(ps_strings); 2669 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2670 ps_strings = PTROUT(PROC_PS_STRINGS(p)); 2671 #else 2672 ps_strings = PROC_PS_STRINGS(p); 2673 #endif 2674 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2675 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2676 } 2677 *sizep = size; 2678 } 2679 2680 static void 2681 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2682 { 2683 struct proc *p; 2684 size_t size; 2685 int structsize; 2686 2687 p = arg; 2688 if (sb == NULL) { 2689 size = 0; 2690 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo), 2691 SBUF_FIXEDLEN); 2692 sbuf_set_drain(sb, sbuf_count_drain, &size); 2693 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2694 PHOLD(p); 2695 proc_getauxv(curthread, p, sb); 2696 PRELE(p); 2697 sbuf_finish(sb); 2698 sbuf_delete(sb); 2699 *sizep = size; 2700 } else { 2701 structsize = sizeof(Elf_Auxinfo); 2702 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2703 PHOLD(p); 2704 proc_getauxv(curthread, p, sb); 2705 PRELE(p); 2706 } 2707 } 2708 2709 static bool 2710 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote, 2711 const char *note_vendor, const Elf_Phdr *pnote, 2712 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg) 2713 { 2714 const Elf_Note *note, *note0, *note_end; 2715 const char *note_name; 2716 char *buf; 2717 int i, error; 2718 bool res; 2719 2720 /* We need some limit, might as well use PAGE_SIZE. */ 2721 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2722 return (false); 2723 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2724 if (pnote->p_offset > PAGE_SIZE || 2725 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2726 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT); 2727 if (buf == NULL) { 2728 VOP_UNLOCK(imgp->vp); 2729 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2730 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 2731 } 2732 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2733 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2734 curthread->td_ucred, NOCRED, NULL, curthread); 2735 if (error != 0) { 2736 uprintf("i/o error PT_NOTE\n"); 2737 goto retf; 2738 } 2739 note = note0 = (const Elf_Note *)buf; 2740 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2741 } else { 2742 note = note0 = (const Elf_Note *)(imgp->image_header + 2743 pnote->p_offset); 2744 note_end = (const Elf_Note *)(imgp->image_header + 2745 pnote->p_offset + pnote->p_filesz); 2746 buf = NULL; 2747 } 2748 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 2749 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 2750 (const char *)note < sizeof(Elf_Note)) { 2751 goto retf; 2752 } 2753 if (note->n_namesz != checknote->n_namesz || 2754 note->n_descsz != checknote->n_descsz || 2755 note->n_type != checknote->n_type) 2756 goto nextnote; 2757 note_name = (const char *)(note + 1); 2758 if (note_name + checknote->n_namesz >= 2759 (const char *)note_end || strncmp(note_vendor, 2760 note_name, checknote->n_namesz) != 0) 2761 goto nextnote; 2762 2763 if (cb(note, cb_arg, &res)) 2764 goto ret; 2765 nextnote: 2766 note = (const Elf_Note *)((const char *)(note + 1) + 2767 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2768 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2769 } 2770 retf: 2771 res = false; 2772 ret: 2773 free(buf, M_TEMP); 2774 return (res); 2775 } 2776 2777 struct brandnote_cb_arg { 2778 Elf_Brandnote *brandnote; 2779 int32_t *osrel; 2780 }; 2781 2782 static bool 2783 brandnote_cb(const Elf_Note *note, void *arg0, bool *res) 2784 { 2785 struct brandnote_cb_arg *arg; 2786 2787 arg = arg0; 2788 2789 /* 2790 * Fetch the osreldate for binary from the ELF OSABI-note if 2791 * necessary. 2792 */ 2793 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 && 2794 arg->brandnote->trans_osrel != NULL ? 2795 arg->brandnote->trans_osrel(note, arg->osrel) : true; 2796 2797 return (true); 2798 } 2799 2800 static Elf_Note fctl_note = { 2801 .n_namesz = sizeof(FREEBSD_ABI_VENDOR), 2802 .n_descsz = sizeof(uint32_t), 2803 .n_type = NT_FREEBSD_FEATURE_CTL, 2804 }; 2805 2806 struct fctl_cb_arg { 2807 bool *has_fctl0; 2808 uint32_t *fctl0; 2809 }; 2810 2811 static bool 2812 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res) 2813 { 2814 struct fctl_cb_arg *arg; 2815 const Elf32_Word *desc; 2816 uintptr_t p; 2817 2818 arg = arg0; 2819 p = (uintptr_t)(note + 1); 2820 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 2821 desc = (const Elf32_Word *)p; 2822 *arg->has_fctl0 = true; 2823 *arg->fctl0 = desc[0]; 2824 *res = true; 2825 return (true); 2826 } 2827 2828 /* 2829 * Try to find the appropriate ABI-note section for checknote, fetch 2830 * the osreldate and feature control flags for binary from the ELF 2831 * OSABI-note. Only the first page of the image is searched, the same 2832 * as for headers. 2833 */ 2834 static bool 2835 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote, 2836 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0) 2837 { 2838 const Elf_Phdr *phdr; 2839 const Elf_Ehdr *hdr; 2840 struct brandnote_cb_arg b_arg; 2841 struct fctl_cb_arg f_arg; 2842 int i, j; 2843 2844 hdr = (const Elf_Ehdr *)imgp->image_header; 2845 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2846 b_arg.brandnote = brandnote; 2847 b_arg.osrel = osrel; 2848 f_arg.has_fctl0 = has_fctl0; 2849 f_arg.fctl0 = fctl0; 2850 2851 for (i = 0; i < hdr->e_phnum; i++) { 2852 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, 2853 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb, 2854 &b_arg)) { 2855 for (j = 0; j < hdr->e_phnum; j++) { 2856 if (phdr[j].p_type == PT_NOTE && 2857 __elfN(parse_notes)(imgp, &fctl_note, 2858 FREEBSD_ABI_VENDOR, &phdr[j], 2859 note_fctl_cb, &f_arg)) 2860 break; 2861 } 2862 return (true); 2863 } 2864 } 2865 return (false); 2866 2867 } 2868 2869 /* 2870 * Tell kern_execve.c about it, with a little help from the linker. 2871 */ 2872 static struct execsw __elfN(execsw) = { 2873 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 2874 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2875 }; 2876 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2877 2878 static vm_prot_t 2879 __elfN(trans_prot)(Elf_Word flags) 2880 { 2881 vm_prot_t prot; 2882 2883 prot = 0; 2884 if (flags & PF_X) 2885 prot |= VM_PROT_EXECUTE; 2886 if (flags & PF_W) 2887 prot |= VM_PROT_WRITE; 2888 if (flags & PF_R) 2889 prot |= VM_PROT_READ; 2890 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 2891 if (i386_read_exec && (flags & PF_R)) 2892 prot |= VM_PROT_EXECUTE; 2893 #endif 2894 return (prot); 2895 } 2896 2897 static Elf_Word 2898 __elfN(untrans_prot)(vm_prot_t prot) 2899 { 2900 Elf_Word flags; 2901 2902 flags = 0; 2903 if (prot & VM_PROT_EXECUTE) 2904 flags |= PF_X; 2905 if (prot & VM_PROT_READ) 2906 flags |= PF_R; 2907 if (prot & VM_PROT_WRITE) 2908 flags |= PF_W; 2909 return (flags); 2910 } 2911