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