1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2017 Dell EMC 5 * Copyright (c) 2000-2001, 2003 David O'Brien 6 * Copyright (c) 1995-1996 Søren Schmidt 7 * Copyright (c) 1996 Peter Wemm 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer 15 * in this position and unchanged. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. The name of the author may not be used to endorse or promote products 20 * derived from this software without specific prior written permission 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "opt_capsicum.h" 35 36 #include <sys/param.h> 37 #include <sys/capsicum.h> 38 #include <sys/compressor.h> 39 #include <sys/exec.h> 40 #include <sys/fcntl.h> 41 #include <sys/imgact.h> 42 #include <sys/imgact_elf.h> 43 #include <sys/jail.h> 44 #include <sys/kernel.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mman.h> 49 #include <sys/namei.h> 50 #include <sys/proc.h> 51 #include <sys/procfs.h> 52 #include <sys/ptrace.h> 53 #include <sys/racct.h> 54 #include <sys/reg.h> 55 #include <sys/resourcevar.h> 56 #include <sys/rwlock.h> 57 #include <sys/sbuf.h> 58 #include <sys/sf_buf.h> 59 #include <sys/smp.h> 60 #include <sys/systm.h> 61 #include <sys/signalvar.h> 62 #include <sys/stat.h> 63 #include <sys/sx.h> 64 #include <sys/syscall.h> 65 #include <sys/sysctl.h> 66 #include <sys/sysent.h> 67 #include <sys/vnode.h> 68 #include <sys/syslog.h> 69 #include <sys/eventhandler.h> 70 #include <sys/user.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_kern.h> 74 #include <vm/vm_param.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_object.h> 78 #include <vm/vm_extern.h> 79 80 #include <machine/elf.h> 81 #include <machine/md_var.h> 82 83 #define ELF_NOTE_ROUNDSIZE 4 84 #define OLD_EI_BRAND 8 85 86 static int __elfN(check_header)(const Elf_Ehdr *hdr); 87 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 88 const char *interp, int32_t *osrel, uint32_t *fctl0); 89 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 90 u_long *entry); 91 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset, 92 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot); 93 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 94 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note, 95 int32_t *osrel); 96 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 97 static bool __elfN(check_note)(struct image_params *imgp, 98 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0, 99 uint32_t *fctl0); 100 static vm_prot_t __elfN(trans_prot)(Elf_Word); 101 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 102 static size_t __elfN(prepare_register_notes)(struct thread *td, 103 struct note_info_list *list, struct thread *target_td); 104 105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), 106 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 107 ""); 108 109 int __elfN(fallback_brand) = -1; 110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, 112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 113 114 static int elf_legacy_coredump = 0; 115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 116 &elf_legacy_coredump, 0, 117 "include all and only RW pages in core dumps"); 118 119 int __elfN(nxstack) = 120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \ 121 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \ 122 defined(__riscv) 123 1; 124 #else 125 0; 126 #endif 127 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 128 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 129 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); 130 131 #if defined(__amd64__) 132 static int __elfN(vdso) = 1; 133 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 134 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0, 135 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading"); 136 #else 137 static int __elfN(vdso) = 0; 138 #endif 139 140 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 141 int i386_read_exec = 0; 142 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 143 "enable execution from readable segments"); 144 #endif 145 146 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR; 147 static int 148 sysctl_pie_base(SYSCTL_HANDLER_ARGS) 149 { 150 u_long val; 151 int error; 152 153 val = __elfN(pie_base); 154 error = sysctl_handle_long(oidp, &val, 0, req); 155 if (error != 0 || req->newptr == NULL) 156 return (error); 157 if ((val & PAGE_MASK) != 0) 158 return (EINVAL); 159 __elfN(pie_base) = val; 160 return (0); 161 } 162 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base, 163 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 164 sysctl_pie_base, "LU", 165 "PIE load base without randomization"); 166 167 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, 168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 169 ""); 170 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr) 171 172 /* 173 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures 174 * have limited address space (which can cause issues for applications with 175 * high memory use) so we leave it off there. 176 */ 177 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64; 178 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN, 179 &__elfN(aslr_enabled), 0, 180 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 181 ": enable address map randomization"); 182 183 /* 184 * Enable ASLR by default for 64-bit PIE binaries. 185 */ 186 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64; 187 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN, 188 &__elfN(pie_aslr_enabled), 0, 189 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 190 ": enable address map randomization for PIE binaries"); 191 192 /* 193 * Sbrk is deprecated and it can be assumed that in most cases it will not be 194 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR 195 * to use the bss grow region. 196 */ 197 static int __elfN(aslr_honor_sbrk) = 0; 198 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW, 199 &__elfN(aslr_honor_sbrk), 0, 200 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used"); 201 202 static int __elfN(aslr_stack) = __ELF_WORD_SIZE == 64; 203 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN, 204 &__elfN(aslr_stack), 0, 205 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 206 ": enable stack address randomization"); 207 208 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64; 209 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN, 210 &__elfN(aslr_shared_page), 0, 211 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 212 ": enable shared page address randomization"); 213 214 static int __elfN(sigfastblock) = 1; 215 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock, 216 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0, 217 "enable sigfastblock for new processes"); 218 219 static bool __elfN(allow_wx) = true; 220 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx, 221 CTLFLAG_RWTUN, &__elfN(allow_wx), 0, 222 "Allow pages to be mapped simultaneously writable and executable"); 223 224 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 225 226 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a)) 227 228 Elf_Brandnote __elfN(freebsd_brandnote) = { 229 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 230 .hdr.n_descsz = sizeof(int32_t), 231 .hdr.n_type = NT_FREEBSD_ABI_TAG, 232 .vendor = FREEBSD_ABI_VENDOR, 233 .flags = BN_TRANSLATE_OSREL, 234 .trans_osrel = __elfN(freebsd_trans_osrel) 235 }; 236 237 static bool 238 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 239 { 240 uintptr_t p; 241 242 p = (uintptr_t)(note + 1); 243 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 244 *osrel = *(const int32_t *)(p); 245 246 return (true); 247 } 248 249 static int GNU_KFREEBSD_ABI_DESC = 3; 250 251 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 252 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 253 .hdr.n_descsz = 16, /* XXX at least 16 */ 254 .hdr.n_type = 1, 255 .vendor = GNU_ABI_VENDOR, 256 .flags = BN_TRANSLATE_OSREL, 257 .trans_osrel = kfreebsd_trans_osrel 258 }; 259 260 static bool 261 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 262 { 263 const Elf32_Word *desc; 264 uintptr_t p; 265 266 p = (uintptr_t)(note + 1); 267 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 268 269 desc = (const Elf32_Word *)p; 270 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 271 return (false); 272 273 /* 274 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 275 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 276 */ 277 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 278 279 return (true); 280 } 281 282 int 283 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 284 { 285 int i; 286 287 for (i = 0; i < MAX_BRANDS; i++) { 288 if (elf_brand_list[i] == NULL) { 289 elf_brand_list[i] = entry; 290 break; 291 } 292 } 293 if (i == MAX_BRANDS) { 294 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 295 __func__, entry); 296 return (-1); 297 } 298 return (0); 299 } 300 301 int 302 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 303 { 304 int i; 305 306 for (i = 0; i < MAX_BRANDS; i++) { 307 if (elf_brand_list[i] == entry) { 308 elf_brand_list[i] = NULL; 309 break; 310 } 311 } 312 if (i == MAX_BRANDS) 313 return (-1); 314 return (0); 315 } 316 317 bool 318 __elfN(brand_inuse)(Elf_Brandinfo *entry) 319 { 320 struct proc *p; 321 bool rval = false; 322 323 sx_slock(&allproc_lock); 324 FOREACH_PROC_IN_SYSTEM(p) { 325 if (p->p_sysent == entry->sysvec) { 326 rval = true; 327 break; 328 } 329 } 330 sx_sunlock(&allproc_lock); 331 332 return (rval); 333 } 334 335 static Elf_Brandinfo * 336 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 337 int32_t *osrel, uint32_t *fctl0) 338 { 339 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 340 Elf_Brandinfo *bi, *bi_m; 341 bool ret, has_fctl0; 342 int i, interp_name_len; 343 344 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0; 345 346 /* 347 * We support four types of branding -- (1) the ELF EI_OSABI field 348 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 349 * branding w/in the ELF header, (3) path of the `interp_path' 350 * field, and (4) the ".note.ABI-tag" ELF section. 351 */ 352 353 /* Look for an ".note.ABI-tag" ELF section */ 354 bi_m = NULL; 355 for (i = 0; i < MAX_BRANDS; i++) { 356 bi = elf_brand_list[i]; 357 if (bi == NULL) 358 continue; 359 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 360 continue; 361 if (hdr->e_machine == bi->machine && (bi->flags & 362 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 363 has_fctl0 = false; 364 *fctl0 = 0; 365 *osrel = 0; 366 ret = __elfN(check_note)(imgp, bi->brand_note, osrel, 367 &has_fctl0, fctl0); 368 /* Give brand a chance to veto check_note's guess */ 369 if (ret && bi->header_supported) { 370 ret = bi->header_supported(imgp, osrel, 371 has_fctl0 ? fctl0 : NULL); 372 } 373 /* 374 * If note checker claimed the binary, but the 375 * interpreter path in the image does not 376 * match default one for the brand, try to 377 * search for other brands with the same 378 * interpreter. Either there is better brand 379 * with the right interpreter, or, failing 380 * this, we return first brand which accepted 381 * our note and, optionally, header. 382 */ 383 if (ret && bi_m == NULL && interp != NULL && 384 (bi->interp_path == NULL || 385 (strlen(bi->interp_path) + 1 != interp_name_len || 386 strncmp(interp, bi->interp_path, interp_name_len) 387 != 0))) { 388 bi_m = bi; 389 ret = 0; 390 } 391 if (ret) 392 return (bi); 393 } 394 } 395 if (bi_m != NULL) 396 return (bi_m); 397 398 /* If the executable has a brand, search for it in the brand list. */ 399 for (i = 0; i < MAX_BRANDS; i++) { 400 bi = elf_brand_list[i]; 401 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 402 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 403 continue; 404 if (hdr->e_machine == bi->machine && 405 (hdr->e_ident[EI_OSABI] == bi->brand || 406 (bi->compat_3_brand != NULL && 407 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 408 bi->compat_3_brand) == 0))) { 409 /* Looks good, but give brand a chance to veto */ 410 if (bi->header_supported == NULL || 411 bi->header_supported(imgp, NULL, NULL)) { 412 /* 413 * Again, prefer strictly matching 414 * interpreter path. 415 */ 416 if (interp_name_len == 0 && 417 bi->interp_path == NULL) 418 return (bi); 419 if (bi->interp_path != NULL && 420 strlen(bi->interp_path) + 1 == 421 interp_name_len && strncmp(interp, 422 bi->interp_path, interp_name_len) == 0) 423 return (bi); 424 if (bi_m == NULL) 425 bi_m = bi; 426 } 427 } 428 } 429 if (bi_m != NULL) 430 return (bi_m); 431 432 /* No known brand, see if the header is recognized by any brand */ 433 for (i = 0; i < MAX_BRANDS; i++) { 434 bi = elf_brand_list[i]; 435 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 436 bi->header_supported == NULL) 437 continue; 438 if (hdr->e_machine == bi->machine) { 439 ret = bi->header_supported(imgp, NULL, NULL); 440 if (ret) 441 return (bi); 442 } 443 } 444 445 /* Lacking a known brand, search for a recognized interpreter. */ 446 if (interp != NULL) { 447 for (i = 0; i < MAX_BRANDS; i++) { 448 bi = elf_brand_list[i]; 449 if (bi == NULL || (bi->flags & 450 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 451 != 0) 452 continue; 453 if (hdr->e_machine == bi->machine && 454 bi->interp_path != NULL && 455 /* ELF image p_filesz includes terminating zero */ 456 strlen(bi->interp_path) + 1 == interp_name_len && 457 strncmp(interp, bi->interp_path, interp_name_len) 458 == 0 && (bi->header_supported == NULL || 459 bi->header_supported(imgp, NULL, NULL))) 460 return (bi); 461 } 462 } 463 464 /* Lacking a recognized interpreter, try the default brand */ 465 for (i = 0; i < MAX_BRANDS; i++) { 466 bi = elf_brand_list[i]; 467 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 468 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 469 continue; 470 if (hdr->e_machine == bi->machine && 471 __elfN(fallback_brand) == bi->brand && 472 (bi->header_supported == NULL || 473 bi->header_supported(imgp, NULL, NULL))) 474 return (bi); 475 } 476 return (NULL); 477 } 478 479 static bool 480 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr) 481 { 482 return (hdr->e_phoff <= PAGE_SIZE && 483 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff); 484 } 485 486 static int 487 __elfN(check_header)(const Elf_Ehdr *hdr) 488 { 489 Elf_Brandinfo *bi; 490 int i; 491 492 if (!IS_ELF(*hdr) || 493 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 494 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 495 hdr->e_ident[EI_VERSION] != EV_CURRENT || 496 hdr->e_phentsize != sizeof(Elf_Phdr) || 497 hdr->e_version != ELF_TARG_VER) 498 return (ENOEXEC); 499 500 /* 501 * Make sure we have at least one brand for this machine. 502 */ 503 504 for (i = 0; i < MAX_BRANDS; i++) { 505 bi = elf_brand_list[i]; 506 if (bi != NULL && bi->machine == hdr->e_machine) 507 break; 508 } 509 if (i == MAX_BRANDS) 510 return (ENOEXEC); 511 512 return (0); 513 } 514 515 static int 516 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 517 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 518 { 519 struct sf_buf *sf; 520 int error; 521 vm_offset_t off; 522 523 /* 524 * Create the page if it doesn't exist yet. Ignore errors. 525 */ 526 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 527 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 528 529 /* 530 * Find the page from the underlying object. 531 */ 532 if (object != NULL) { 533 sf = vm_imgact_map_page(object, offset); 534 if (sf == NULL) 535 return (KERN_FAILURE); 536 off = offset - trunc_page(offset); 537 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 538 end - start); 539 vm_imgact_unmap_page(sf); 540 if (error != 0) 541 return (KERN_FAILURE); 542 } 543 544 return (KERN_SUCCESS); 545 } 546 547 static int 548 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object, 549 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot, 550 int cow) 551 { 552 struct sf_buf *sf; 553 vm_offset_t off; 554 vm_size_t sz; 555 int error, locked, rv; 556 557 if (start != trunc_page(start)) { 558 rv = __elfN(map_partial)(map, object, offset, start, 559 round_page(start), prot); 560 if (rv != KERN_SUCCESS) 561 return (rv); 562 offset += round_page(start) - start; 563 start = round_page(start); 564 } 565 if (end != round_page(end)) { 566 rv = __elfN(map_partial)(map, object, offset + 567 trunc_page(end) - start, trunc_page(end), end, prot); 568 if (rv != KERN_SUCCESS) 569 return (rv); 570 end = trunc_page(end); 571 } 572 if (start >= end) 573 return (KERN_SUCCESS); 574 if ((offset & PAGE_MASK) != 0) { 575 /* 576 * The mapping is not page aligned. This means that we have 577 * to copy the data. 578 */ 579 rv = vm_map_fixed(map, NULL, 0, start, end - start, 580 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 581 if (rv != KERN_SUCCESS) 582 return (rv); 583 if (object == NULL) 584 return (KERN_SUCCESS); 585 for (; start < end; start += sz) { 586 sf = vm_imgact_map_page(object, offset); 587 if (sf == NULL) 588 return (KERN_FAILURE); 589 off = offset - trunc_page(offset); 590 sz = end - start; 591 if (sz > PAGE_SIZE - off) 592 sz = PAGE_SIZE - off; 593 error = copyout((caddr_t)sf_buf_kva(sf) + off, 594 (caddr_t)start, sz); 595 vm_imgact_unmap_page(sf); 596 if (error != 0) 597 return (KERN_FAILURE); 598 offset += sz; 599 } 600 } else { 601 vm_object_reference(object); 602 rv = vm_map_fixed(map, object, offset, start, end - start, 603 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL | 604 (object != NULL ? MAP_VN_EXEC : 0)); 605 if (rv != KERN_SUCCESS) { 606 locked = VOP_ISLOCKED(imgp->vp); 607 VOP_UNLOCK(imgp->vp); 608 vm_object_deallocate(object); 609 vn_lock(imgp->vp, locked | LK_RETRY); 610 return (rv); 611 } else if (object != NULL) { 612 MPASS(imgp->vp->v_object == object); 613 VOP_SET_TEXT_CHECKED(imgp->vp); 614 } 615 } 616 return (KERN_SUCCESS); 617 } 618 619 static int 620 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset, 621 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot) 622 { 623 struct sf_buf *sf; 624 size_t map_len; 625 vm_map_t map; 626 vm_object_t object; 627 vm_offset_t map_addr; 628 int error, rv, cow; 629 size_t copy_len; 630 vm_ooffset_t file_addr; 631 632 /* 633 * It's necessary to fail if the filsz + offset taken from the 634 * header is greater than the actual file pager object's size. 635 * If we were to allow this, then the vm_map_find() below would 636 * walk right off the end of the file object and into the ether. 637 * 638 * While I'm here, might as well check for something else that 639 * is invalid: filsz cannot be greater than memsz. 640 */ 641 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 642 filsz > memsz) { 643 uprintf("elf_load_section: truncated ELF file\n"); 644 return (ENOEXEC); 645 } 646 647 object = imgp->object; 648 map = &imgp->proc->p_vmspace->vm_map; 649 map_addr = trunc_page((vm_offset_t)vmaddr); 650 file_addr = trunc_page(offset); 651 652 /* 653 * We have two choices. We can either clear the data in the last page 654 * of an oversized mapping, or we can start the anon mapping a page 655 * early and copy the initialized data into that first page. We 656 * choose the second. 657 */ 658 if (filsz == 0) 659 map_len = 0; 660 else if (memsz > filsz) 661 map_len = trunc_page(offset + filsz) - file_addr; 662 else 663 map_len = round_page(offset + filsz) - file_addr; 664 665 if (map_len != 0) { 666 /* cow flags: don't dump readonly sections in core */ 667 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 668 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 669 670 rv = __elfN(map_insert)(imgp, map, object, file_addr, 671 map_addr, map_addr + map_len, prot, cow); 672 if (rv != KERN_SUCCESS) 673 return (EINVAL); 674 675 /* we can stop now if we've covered it all */ 676 if (memsz == filsz) 677 return (0); 678 } 679 680 /* 681 * We have to get the remaining bit of the file into the first part 682 * of the oversized map segment. This is normally because the .data 683 * segment in the file is extended to provide bss. It's a neat idea 684 * to try and save a page, but it's a pain in the behind to implement. 685 */ 686 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset + 687 filsz); 688 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 689 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 690 691 /* This had damn well better be true! */ 692 if (map_len != 0) { 693 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 694 map_addr + map_len, prot, 0); 695 if (rv != KERN_SUCCESS) 696 return (EINVAL); 697 } 698 699 if (copy_len != 0) { 700 sf = vm_imgact_map_page(object, offset + filsz); 701 if (sf == NULL) 702 return (EIO); 703 704 /* send the page fragment to user space */ 705 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr, 706 copy_len); 707 vm_imgact_unmap_page(sf); 708 if (error != 0) 709 return (error); 710 } 711 712 /* 713 * Remove write access to the page if it was only granted by map_insert 714 * to allow copyout. 715 */ 716 if ((prot & VM_PROT_WRITE) == 0) 717 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 718 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT); 719 720 return (0); 721 } 722 723 static int 724 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr, 725 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp) 726 { 727 vm_prot_t prot; 728 u_long base_addr; 729 bool first; 730 int error, i; 731 732 ASSERT_VOP_LOCKED(imgp->vp, __func__); 733 734 base_addr = 0; 735 first = true; 736 737 for (i = 0; i < hdr->e_phnum; i++) { 738 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 739 continue; 740 741 /* Loadable segment */ 742 prot = __elfN(trans_prot)(phdr[i].p_flags); 743 error = __elfN(load_section)(imgp, phdr[i].p_offset, 744 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 745 phdr[i].p_memsz, phdr[i].p_filesz, prot); 746 if (error != 0) 747 return (error); 748 749 /* 750 * Establish the base address if this is the first segment. 751 */ 752 if (first) { 753 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 754 first = false; 755 } 756 } 757 758 if (base_addrp != NULL) 759 *base_addrp = base_addr; 760 761 return (0); 762 } 763 764 /* 765 * Load the file "file" into memory. It may be either a shared object 766 * or an executable. 767 * 768 * The "addr" reference parameter is in/out. On entry, it specifies 769 * the address where a shared object should be loaded. If the file is 770 * an executable, this value is ignored. On exit, "addr" specifies 771 * where the file was actually loaded. 772 * 773 * The "entry" reference parameter is out only. On exit, it specifies 774 * the entry point for the loaded file. 775 */ 776 static int 777 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 778 u_long *entry) 779 { 780 struct { 781 struct nameidata nd; 782 struct vattr attr; 783 struct image_params image_params; 784 } *tempdata; 785 const Elf_Ehdr *hdr = NULL; 786 const Elf_Phdr *phdr = NULL; 787 struct nameidata *nd; 788 struct vattr *attr; 789 struct image_params *imgp; 790 u_long rbase; 791 u_long base_addr = 0; 792 int error; 793 794 #ifdef CAPABILITY_MODE 795 /* 796 * XXXJA: This check can go away once we are sufficiently confident 797 * that the checks in namei() are correct. 798 */ 799 if (IN_CAPABILITY_MODE(curthread)) 800 return (ECAPMODE); 801 #endif 802 803 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO); 804 nd = &tempdata->nd; 805 attr = &tempdata->attr; 806 imgp = &tempdata->image_params; 807 808 /* 809 * Initialize part of the common data 810 */ 811 imgp->proc = p; 812 imgp->attr = attr; 813 814 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF, 815 UIO_SYSSPACE, file); 816 if ((error = namei(nd)) != 0) { 817 nd->ni_vp = NULL; 818 goto fail; 819 } 820 NDFREE_PNBUF(nd); 821 imgp->vp = nd->ni_vp; 822 823 /* 824 * Check permissions, modes, uid, etc on the file, and "open" it. 825 */ 826 error = exec_check_permissions(imgp); 827 if (error) 828 goto fail; 829 830 error = exec_map_first_page(imgp); 831 if (error) 832 goto fail; 833 834 imgp->object = nd->ni_vp->v_object; 835 836 hdr = (const Elf_Ehdr *)imgp->image_header; 837 if ((error = __elfN(check_header)(hdr)) != 0) 838 goto fail; 839 if (hdr->e_type == ET_DYN) 840 rbase = *addr; 841 else if (hdr->e_type == ET_EXEC) 842 rbase = 0; 843 else { 844 error = ENOEXEC; 845 goto fail; 846 } 847 848 /* Only support headers that fit within first page for now */ 849 if (!__elfN(phdr_in_zero_page)(hdr)) { 850 error = ENOEXEC; 851 goto fail; 852 } 853 854 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 855 if (!aligned(phdr, Elf_Addr)) { 856 error = ENOEXEC; 857 goto fail; 858 } 859 860 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr); 861 if (error != 0) 862 goto fail; 863 864 if (p->p_sysent->sv_protect != NULL) 865 p->p_sysent->sv_protect(imgp, SVP_INTERP); 866 867 *addr = base_addr; 868 *entry = (unsigned long)hdr->e_entry + rbase; 869 870 fail: 871 if (imgp->firstpage) 872 exec_unmap_first_page(imgp); 873 874 if (nd->ni_vp) { 875 if (imgp->textset) 876 VOP_UNSET_TEXT_CHECKED(nd->ni_vp); 877 vput(nd->ni_vp); 878 } 879 free(tempdata, M_TEMP); 880 881 return (error); 882 } 883 884 /* 885 * Select randomized valid address in the map map, between minv and 886 * maxv, with specified alignment. The [minv, maxv) range must belong 887 * to the map. Note that function only allocates the address, it is 888 * up to caller to clamp maxv in a way that the final allocation 889 * length fit into the map. 890 * 891 * Result is returned in *resp, error code indicates that arguments 892 * did not pass sanity checks for overflow and range correctness. 893 */ 894 static int 895 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv, 896 u_int align, u_long *resp) 897 { 898 u_long rbase, res; 899 900 MPASS(vm_map_min(map) <= minv); 901 902 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) { 903 uprintf("Invalid ELF segments layout\n"); 904 return (ENOEXEC); 905 } 906 907 arc4rand(&rbase, sizeof(rbase), 0); 908 res = roundup(minv, (u_long)align) + rbase % (maxv - minv); 909 res &= ~((u_long)align - 1); 910 if (res >= maxv) 911 res -= align; 912 913 KASSERT(res >= minv, 914 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx", 915 res, minv, maxv, rbase)); 916 KASSERT(res < maxv, 917 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx", 918 res, maxv, minv, rbase)); 919 920 *resp = res; 921 return (0); 922 } 923 924 static int 925 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr, 926 const Elf_Phdr *phdr) 927 { 928 struct vmspace *vmspace; 929 const char *err_str; 930 u_long text_size, data_size, total_size, text_addr, data_addr; 931 u_long seg_size, seg_addr; 932 int i; 933 934 err_str = NULL; 935 text_size = data_size = total_size = text_addr = data_addr = 0; 936 937 for (i = 0; i < hdr->e_phnum; i++) { 938 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 939 continue; 940 941 seg_addr = trunc_page(phdr[i].p_vaddr + imgp->et_dyn_addr); 942 seg_size = round_page(phdr[i].p_memsz + 943 phdr[i].p_vaddr + imgp->et_dyn_addr - seg_addr); 944 945 /* 946 * Make the largest executable segment the official 947 * text segment and all others data. 948 * 949 * Note that obreak() assumes that data_addr + data_size == end 950 * of data load area, and the ELF file format expects segments 951 * to be sorted by address. If multiple data segments exist, 952 * the last one will be used. 953 */ 954 955 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) { 956 text_size = seg_size; 957 text_addr = seg_addr; 958 } else { 959 data_size = seg_size; 960 data_addr = seg_addr; 961 } 962 total_size += seg_size; 963 } 964 965 if (data_addr == 0 && data_size == 0) { 966 data_addr = text_addr; 967 data_size = text_size; 968 } 969 970 /* 971 * Check limits. It should be safe to check the 972 * limits after loading the segments since we do 973 * not actually fault in all the segments pages. 974 */ 975 PROC_LOCK(imgp->proc); 976 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 977 err_str = "Data segment size exceeds process limit"; 978 else if (text_size > maxtsiz) 979 err_str = "Text segment size exceeds system limit"; 980 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 981 err_str = "Total segment size exceeds process limit"; 982 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 983 err_str = "Data segment size exceeds resource limit"; 984 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 985 err_str = "Total segment size exceeds resource limit"; 986 PROC_UNLOCK(imgp->proc); 987 if (err_str != NULL) { 988 uprintf("%s\n", err_str); 989 return (ENOMEM); 990 } 991 992 vmspace = imgp->proc->p_vmspace; 993 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 994 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 995 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 996 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 997 998 return (0); 999 } 1000 1001 static int 1002 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr, 1003 char **interpp, bool *free_interpp) 1004 { 1005 struct thread *td; 1006 char *interp; 1007 int error, interp_name_len; 1008 1009 KASSERT(phdr->p_type == PT_INTERP, 1010 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type)); 1011 ASSERT_VOP_LOCKED(imgp->vp, __func__); 1012 1013 td = curthread; 1014 1015 /* Path to interpreter */ 1016 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) { 1017 uprintf("Invalid PT_INTERP\n"); 1018 return (ENOEXEC); 1019 } 1020 1021 interp_name_len = phdr->p_filesz; 1022 if (phdr->p_offset > PAGE_SIZE || 1023 interp_name_len > PAGE_SIZE - phdr->p_offset) { 1024 /* 1025 * The vnode lock might be needed by the pagedaemon to 1026 * clean pages owned by the vnode. Do not allow sleep 1027 * waiting for memory with the vnode locked, instead 1028 * try non-sleepable allocation first, and if it 1029 * fails, go to the slow path were we drop the lock 1030 * and do M_WAITOK. A text reference prevents 1031 * modifications to the vnode content. 1032 */ 1033 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT); 1034 if (interp == NULL) { 1035 VOP_UNLOCK(imgp->vp); 1036 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK); 1037 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1038 } 1039 1040 error = vn_rdwr(UIO_READ, imgp->vp, interp, 1041 interp_name_len, phdr->p_offset, 1042 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 1043 NOCRED, NULL, td); 1044 if (error != 0) { 1045 free(interp, M_TEMP); 1046 uprintf("i/o error PT_INTERP %d\n", error); 1047 return (error); 1048 } 1049 interp[interp_name_len] = '\0'; 1050 1051 *interpp = interp; 1052 *free_interpp = true; 1053 return (0); 1054 } 1055 1056 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset; 1057 if (interp[interp_name_len - 1] != '\0') { 1058 uprintf("Invalid PT_INTERP\n"); 1059 return (ENOEXEC); 1060 } 1061 1062 *interpp = interp; 1063 *free_interpp = false; 1064 return (0); 1065 } 1066 1067 static int 1068 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info, 1069 const char *interp, u_long *addr, u_long *entry) 1070 { 1071 int error; 1072 1073 if (brand_info->interp_newpath != NULL && 1074 (brand_info->interp_path == NULL || 1075 strcmp(interp, brand_info->interp_path) == 0)) { 1076 error = __elfN(load_file)(imgp->proc, 1077 brand_info->interp_newpath, addr, entry); 1078 if (error == 0) 1079 return (0); 1080 } 1081 1082 error = __elfN(load_file)(imgp->proc, interp, addr, entry); 1083 if (error == 0) 1084 return (0); 1085 1086 uprintf("ELF interpreter %s not found, error %d\n", interp, error); 1087 return (error); 1088 } 1089 1090 /* 1091 * Impossible et_dyn_addr initial value indicating that the real base 1092 * must be calculated later with some randomization applied. 1093 */ 1094 #define ET_DYN_ADDR_RAND 1 1095 1096 static int 1097 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 1098 { 1099 struct thread *td; 1100 const Elf_Ehdr *hdr; 1101 const Elf_Phdr *phdr; 1102 Elf_Auxargs *elf_auxargs; 1103 struct vmspace *vmspace; 1104 vm_map_t map; 1105 char *interp; 1106 Elf_Brandinfo *brand_info; 1107 struct sysentvec *sv; 1108 u_long addr, baddr, entry, proghdr; 1109 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc; 1110 uint32_t fctl0; 1111 int32_t osrel; 1112 bool free_interp; 1113 int error, i, n; 1114 1115 hdr = (const Elf_Ehdr *)imgp->image_header; 1116 1117 /* 1118 * Do we have a valid ELF header ? 1119 * 1120 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 1121 * if particular brand doesn't support it. 1122 */ 1123 if (__elfN(check_header)(hdr) != 0 || 1124 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 1125 return (-1); 1126 1127 /* 1128 * From here on down, we return an errno, not -1, as we've 1129 * detected an ELF file. 1130 */ 1131 1132 if (!__elfN(phdr_in_zero_page)(hdr)) { 1133 uprintf("Program headers not in the first page\n"); 1134 return (ENOEXEC); 1135 } 1136 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1137 if (!aligned(phdr, Elf_Addr)) { 1138 uprintf("Unaligned program headers\n"); 1139 return (ENOEXEC); 1140 } 1141 1142 n = error = 0; 1143 baddr = 0; 1144 osrel = 0; 1145 fctl0 = 0; 1146 entry = proghdr = 0; 1147 interp = NULL; 1148 free_interp = false; 1149 td = curthread; 1150 1151 /* 1152 * Somewhat arbitrary, limit accepted max alignment for the 1153 * loadable segment to the max supported superpage size. Too 1154 * large alignment requests are not useful and are indicators 1155 * of corrupted or outright malicious binary. 1156 */ 1157 maxalign = PAGE_SIZE; 1158 maxsalign = PAGE_SIZE * 1024; 1159 for (i = MAXPAGESIZES - 1; i > 0; i--) { 1160 if (pagesizes[i] > maxsalign) 1161 maxsalign = pagesizes[i]; 1162 } 1163 1164 mapsz = 0; 1165 1166 for (i = 0; i < hdr->e_phnum; i++) { 1167 switch (phdr[i].p_type) { 1168 case PT_LOAD: 1169 if (n == 0) 1170 baddr = phdr[i].p_vaddr; 1171 if (!powerof2(phdr[i].p_align) || 1172 phdr[i].p_align > maxsalign) { 1173 uprintf("Invalid segment alignment\n"); 1174 error = ENOEXEC; 1175 goto ret; 1176 } 1177 if (phdr[i].p_align > maxalign) 1178 maxalign = phdr[i].p_align; 1179 if (mapsz + phdr[i].p_memsz < mapsz) { 1180 uprintf("Mapsize overflow\n"); 1181 error = ENOEXEC; 1182 goto ret; 1183 } 1184 mapsz += phdr[i].p_memsz; 1185 n++; 1186 1187 /* 1188 * If this segment contains the program headers, 1189 * remember their virtual address for the AT_PHDR 1190 * aux entry. Static binaries don't usually include 1191 * a PT_PHDR entry. 1192 */ 1193 if (phdr[i].p_offset == 0 && 1194 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= 1195 phdr[i].p_filesz) 1196 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 1197 break; 1198 case PT_INTERP: 1199 /* Path to interpreter */ 1200 if (interp != NULL) { 1201 uprintf("Multiple PT_INTERP headers\n"); 1202 error = ENOEXEC; 1203 goto ret; 1204 } 1205 error = __elfN(get_interp)(imgp, &phdr[i], &interp, 1206 &free_interp); 1207 if (error != 0) 1208 goto ret; 1209 break; 1210 case PT_GNU_STACK: 1211 if (__elfN(nxstack)) { 1212 imgp->stack_prot = 1213 __elfN(trans_prot)(phdr[i].p_flags); 1214 if ((imgp->stack_prot & VM_PROT_RW) != 1215 VM_PROT_RW) { 1216 uprintf("Invalid PT_GNU_STACK\n"); 1217 error = ENOEXEC; 1218 goto ret; 1219 } 1220 } 1221 imgp->stack_sz = phdr[i].p_memsz; 1222 break; 1223 case PT_PHDR: /* Program header table info */ 1224 proghdr = phdr[i].p_vaddr; 1225 break; 1226 } 1227 } 1228 1229 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0); 1230 if (brand_info == NULL) { 1231 uprintf("ELF binary type \"%u\" not known.\n", 1232 hdr->e_ident[EI_OSABI]); 1233 error = ENOEXEC; 1234 goto ret; 1235 } 1236 sv = brand_info->sysvec; 1237 if (hdr->e_type == ET_DYN) { 1238 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 1239 uprintf("Cannot execute shared object\n"); 1240 error = ENOEXEC; 1241 goto ret; 1242 } 1243 /* 1244 * Honour the base load address from the dso if it is 1245 * non-zero for some reason. 1246 */ 1247 if (baddr == 0) { 1248 if ((sv->sv_flags & SV_ASLR) == 0 || 1249 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) 1250 imgp->et_dyn_addr = __elfN(pie_base); 1251 else if ((__elfN(pie_aslr_enabled) && 1252 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) || 1253 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0) 1254 imgp->et_dyn_addr = ET_DYN_ADDR_RAND; 1255 else 1256 imgp->et_dyn_addr = __elfN(pie_base); 1257 } 1258 } 1259 1260 /* 1261 * Avoid a possible deadlock if the current address space is destroyed 1262 * and that address space maps the locked vnode. In the common case, 1263 * the locked vnode's v_usecount is decremented but remains greater 1264 * than zero. Consequently, the vnode lock is not needed by vrele(). 1265 * However, in cases where the vnode lock is external, such as nullfs, 1266 * v_usecount may become zero. 1267 * 1268 * The VV_TEXT flag prevents modifications to the executable while 1269 * the vnode is unlocked. 1270 */ 1271 VOP_UNLOCK(imgp->vp); 1272 1273 /* 1274 * Decide whether to enable randomization of user mappings. 1275 * First, reset user preferences for the setid binaries. 1276 * Then, account for the support of the randomization by the 1277 * ABI, by user preferences, and make special treatment for 1278 * PIE binaries. 1279 */ 1280 if (imgp->credential_setid) { 1281 PROC_LOCK(imgp->proc); 1282 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE | 1283 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 1284 PROC_UNLOCK(imgp->proc); 1285 } 1286 if ((sv->sv_flags & SV_ASLR) == 0 || 1287 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 || 1288 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) { 1289 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND, 1290 ("imgp->et_dyn_addr == RAND and !ASLR")); 1291 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 || 1292 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) || 1293 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1294 imgp->map_flags |= MAP_ASLR; 1295 /* 1296 * If user does not care about sbrk, utilize the bss 1297 * grow region for mappings as well. We can select 1298 * the base for the image anywere and still not suffer 1299 * from the fragmentation. 1300 */ 1301 if (!__elfN(aslr_honor_sbrk) || 1302 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0) 1303 imgp->map_flags |= MAP_ASLR_IGNSTART; 1304 if (__elfN(aslr_stack)) 1305 imgp->map_flags |= MAP_ASLR_STACK; 1306 if (__elfN(aslr_shared_page)) 1307 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE; 1308 } 1309 1310 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 && 1311 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) || 1312 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0) 1313 imgp->map_flags |= MAP_WXORX; 1314 1315 error = exec_new_vmspace(imgp, sv); 1316 1317 imgp->proc->p_sysent = sv; 1318 imgp->proc->p_elf_brandinfo = brand_info; 1319 1320 vmspace = imgp->proc->p_vmspace; 1321 map = &vmspace->vm_map; 1322 maxv = sv->sv_usrstack; 1323 if ((imgp->map_flags & MAP_ASLR_STACK) == 0) 1324 maxv -= lim_max(td, RLIMIT_STACK); 1325 if (error == 0 && mapsz >= maxv - vm_map_min(map)) { 1326 uprintf("Excessive mapping size\n"); 1327 error = ENOEXEC; 1328 } 1329 1330 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1331 KASSERT((map->flags & MAP_ASLR) != 0, 1332 ("ET_DYN_ADDR_RAND but !MAP_ASLR")); 1333 error = __CONCAT(rnd_, __elfN(base))(map, 1334 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA), 1335 /* reserve half of the address space to interpreter */ 1336 maxv / 2, maxalign, &imgp->et_dyn_addr); 1337 } 1338 1339 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1340 if (error != 0) 1341 goto ret; 1342 1343 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL); 1344 if (error != 0) 1345 goto ret; 1346 1347 error = __elfN(enforce_limits)(imgp, hdr, phdr); 1348 if (error != 0) 1349 goto ret; 1350 1351 /* 1352 * We load the dynamic linker where a userland call 1353 * to mmap(0, ...) would put it. The rationale behind this 1354 * calculation is that it leaves room for the heap to grow to 1355 * its maximum allowed size. 1356 */ 1357 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1358 RLIMIT_DATA)); 1359 if ((map->flags & MAP_ASLR) != 0) { 1360 maxv1 = maxv / 2 + addr / 2; 1361 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1, 1362 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ? 1363 pagesizes[1] : pagesizes[0], &anon_loc); 1364 if (error != 0) 1365 goto ret; 1366 map->anon_loc = anon_loc; 1367 } else { 1368 map->anon_loc = addr; 1369 } 1370 1371 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr; 1372 imgp->entry_addr = entry; 1373 1374 if (sv->sv_protect != NULL) 1375 sv->sv_protect(imgp, SVP_IMAGE); 1376 1377 if (interp != NULL) { 1378 VOP_UNLOCK(imgp->vp); 1379 if ((map->flags & MAP_ASLR) != 0) { 1380 /* Assume that interpreter fits into 1/4 of AS */ 1381 maxv1 = maxv / 2 + addr / 2; 1382 error = __CONCAT(rnd_, __elfN(base))(map, addr, 1383 maxv1, PAGE_SIZE, &addr); 1384 } 1385 if (error == 0) { 1386 error = __elfN(load_interp)(imgp, brand_info, interp, 1387 &addr, &imgp->entry_addr); 1388 } 1389 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1390 if (error != 0) 1391 goto ret; 1392 } else 1393 addr = imgp->et_dyn_addr; 1394 1395 error = exec_map_stack(imgp); 1396 if (error != 0) 1397 goto ret; 1398 1399 /* 1400 * Construct auxargs table (used by the copyout_auxargs routine) 1401 */ 1402 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT); 1403 if (elf_auxargs == NULL) { 1404 VOP_UNLOCK(imgp->vp); 1405 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1406 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1407 } 1408 elf_auxargs->execfd = -1; 1409 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr; 1410 elf_auxargs->phent = hdr->e_phentsize; 1411 elf_auxargs->phnum = hdr->e_phnum; 1412 elf_auxargs->pagesz = PAGE_SIZE; 1413 elf_auxargs->base = addr; 1414 elf_auxargs->flags = 0; 1415 elf_auxargs->entry = entry; 1416 elf_auxargs->hdr_eflags = hdr->e_flags; 1417 1418 imgp->auxargs = elf_auxargs; 1419 imgp->interpreted = 0; 1420 imgp->reloc_base = addr; 1421 imgp->proc->p_osrel = osrel; 1422 imgp->proc->p_fctl0 = fctl0; 1423 imgp->proc->p_elf_flags = hdr->e_flags; 1424 1425 ret: 1426 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock"); 1427 if (free_interp) 1428 free(interp, M_TEMP); 1429 return (error); 1430 } 1431 1432 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE) 1433 1434 int 1435 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base) 1436 { 1437 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1438 Elf_Auxinfo *argarray, *pos; 1439 struct vmspace *vmspace; 1440 rlim_t stacksz; 1441 int error, oc; 1442 uint32_t bsdflags; 1443 1444 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1445 M_WAITOK | M_ZERO); 1446 1447 vmspace = imgp->proc->p_vmspace; 1448 1449 if (args->execfd != -1) 1450 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1451 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1452 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1453 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1454 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1455 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1456 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1457 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1458 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1459 if (imgp->execpathp != 0) 1460 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp); 1461 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1462 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1463 if (imgp->canary != 0) { 1464 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary); 1465 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1466 } 1467 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1468 if (imgp->pagesizes != 0) { 1469 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes); 1470 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1471 } 1472 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) { 1473 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1474 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset); 1475 } 1476 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1477 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1478 imgp->sysent->sv_stackprot); 1479 if (imgp->sysent->sv_hwcap != NULL) 1480 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1481 if (imgp->sysent->sv_hwcap2 != NULL) 1482 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1483 bsdflags = 0; 1484 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0; 1485 oc = atomic_load_int(&vm_overcommit); 1486 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) != 1487 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0; 1488 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags); 1489 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc); 1490 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv); 1491 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc); 1492 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv); 1493 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings); 1494 #ifdef RANDOM_FENESTRASX 1495 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) { 1496 AUXARGS_ENTRY(pos, AT_FXRNG, 1497 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset); 1498 } 1499 #endif 1500 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) { 1501 AUXARGS_ENTRY(pos, AT_KPRELOAD, 1502 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset); 1503 } 1504 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop)); 1505 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur; 1506 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz); 1507 AUXARGS_ENTRY(pos, AT_NULL, 0); 1508 1509 free(imgp->auxargs, M_TEMP); 1510 imgp->auxargs = NULL; 1511 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1512 1513 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT); 1514 free(argarray, M_TEMP); 1515 return (error); 1516 } 1517 1518 int 1519 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp) 1520 { 1521 Elf_Addr *base; 1522 1523 base = (Elf_Addr *)*stack_base; 1524 base--; 1525 if (elf_suword(base, imgp->args->argc) == -1) 1526 return (EFAULT); 1527 *stack_base = (uintptr_t)base; 1528 return (0); 1529 } 1530 1531 /* 1532 * Code for generating ELF core dumps. 1533 */ 1534 1535 typedef void (*segment_callback)(vm_map_entry_t, void *); 1536 1537 /* Closure for cb_put_phdr(). */ 1538 struct phdr_closure { 1539 Elf_Phdr *phdr; /* Program header to fill in */ 1540 Elf_Off offset; /* Offset of segment in core file */ 1541 }; 1542 1543 struct note_info { 1544 int type; /* Note type. */ 1545 struct regset *regset; /* Register set. */ 1546 outfunc_t outfunc; /* Output function. */ 1547 void *outarg; /* Argument for the output function. */ 1548 size_t outsize; /* Output size. */ 1549 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1550 }; 1551 1552 TAILQ_HEAD(note_info_list, note_info); 1553 1554 extern int compress_user_cores; 1555 extern int compress_user_cores_level; 1556 1557 static void cb_put_phdr(vm_map_entry_t, void *); 1558 static void cb_size_segment(vm_map_entry_t, void *); 1559 static void each_dumpable_segment(struct thread *, segment_callback, void *, 1560 int); 1561 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1562 struct note_info_list *, size_t, int); 1563 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *); 1564 1565 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1566 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1567 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1568 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1569 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1570 static void note_procstat_files(void *, struct sbuf *, size_t *); 1571 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1572 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1573 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1574 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1575 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1576 1577 static int 1578 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1579 { 1580 1581 return (core_write((struct coredump_params *)arg, base, len, offset, 1582 UIO_SYSSPACE, NULL)); 1583 } 1584 1585 int 1586 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1587 { 1588 struct ucred *cred = td->td_ucred; 1589 int compm, error = 0; 1590 struct sseg_closure seginfo; 1591 struct note_info_list notelst; 1592 struct coredump_params params; 1593 struct note_info *ninfo; 1594 void *hdr, *tmpbuf; 1595 size_t hdrsize, notesz, coresize; 1596 1597 hdr = NULL; 1598 tmpbuf = NULL; 1599 TAILQ_INIT(¬elst); 1600 1601 /* Size the program segments. */ 1602 __elfN(size_segments)(td, &seginfo, flags); 1603 1604 /* 1605 * Collect info about the core file header area. 1606 */ 1607 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1608 if (seginfo.count + 1 >= PN_XNUM) 1609 hdrsize += sizeof(Elf_Shdr); 1610 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz); 1611 coresize = round_page(hdrsize + notesz) + seginfo.size; 1612 1613 /* Set up core dump parameters. */ 1614 params.offset = 0; 1615 params.active_cred = cred; 1616 params.file_cred = NOCRED; 1617 params.td = td; 1618 params.vp = vp; 1619 params.comp = NULL; 1620 1621 #ifdef RACCT 1622 if (racct_enable) { 1623 PROC_LOCK(td->td_proc); 1624 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1625 PROC_UNLOCK(td->td_proc); 1626 if (error != 0) { 1627 error = EFAULT; 1628 goto done; 1629 } 1630 } 1631 #endif 1632 if (coresize >= limit) { 1633 error = EFAULT; 1634 goto done; 1635 } 1636 1637 /* Create a compression stream if necessary. */ 1638 compm = compress_user_cores; 1639 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP && 1640 compm == 0) 1641 compm = COMPRESS_GZIP; 1642 if (compm != 0) { 1643 params.comp = compressor_init(core_compressed_write, 1644 compm, CORE_BUF_SIZE, 1645 compress_user_cores_level, ¶ms); 1646 if (params.comp == NULL) { 1647 error = EFAULT; 1648 goto done; 1649 } 1650 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1651 } 1652 1653 /* 1654 * Allocate memory for building the header, fill it up, 1655 * and write it out following the notes. 1656 */ 1657 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1658 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1659 notesz, flags); 1660 1661 /* Write the contents of all of the writable segments. */ 1662 if (error == 0) { 1663 Elf_Phdr *php; 1664 off_t offset; 1665 int i; 1666 1667 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1668 offset = round_page(hdrsize + notesz); 1669 for (i = 0; i < seginfo.count; i++) { 1670 error = core_output((char *)(uintptr_t)php->p_vaddr, 1671 php->p_filesz, offset, ¶ms, tmpbuf); 1672 if (error != 0) 1673 break; 1674 offset += php->p_filesz; 1675 php++; 1676 } 1677 if (error == 0 && params.comp != NULL) 1678 error = compressor_flush(params.comp); 1679 } 1680 if (error) { 1681 log(LOG_WARNING, 1682 "Failed to write core file for process %s (error %d)\n", 1683 curproc->p_comm, error); 1684 } 1685 1686 done: 1687 free(tmpbuf, M_TEMP); 1688 if (params.comp != NULL) 1689 compressor_fini(params.comp); 1690 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1691 TAILQ_REMOVE(¬elst, ninfo, link); 1692 free(ninfo, M_TEMP); 1693 } 1694 if (hdr != NULL) 1695 free(hdr, M_TEMP); 1696 1697 return (error); 1698 } 1699 1700 /* 1701 * A callback for each_dumpable_segment() to write out the segment's 1702 * program header entry. 1703 */ 1704 static void 1705 cb_put_phdr(vm_map_entry_t entry, void *closure) 1706 { 1707 struct phdr_closure *phc = (struct phdr_closure *)closure; 1708 Elf_Phdr *phdr = phc->phdr; 1709 1710 phc->offset = round_page(phc->offset); 1711 1712 phdr->p_type = PT_LOAD; 1713 phdr->p_offset = phc->offset; 1714 phdr->p_vaddr = entry->start; 1715 phdr->p_paddr = 0; 1716 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1717 phdr->p_align = PAGE_SIZE; 1718 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1719 1720 phc->offset += phdr->p_filesz; 1721 phc->phdr++; 1722 } 1723 1724 /* 1725 * A callback for each_dumpable_segment() to gather information about 1726 * the number of segments and their total size. 1727 */ 1728 static void 1729 cb_size_segment(vm_map_entry_t entry, void *closure) 1730 { 1731 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1732 1733 ssc->count++; 1734 ssc->size += entry->end - entry->start; 1735 } 1736 1737 void 1738 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo, 1739 int flags) 1740 { 1741 seginfo->count = 0; 1742 seginfo->size = 0; 1743 1744 each_dumpable_segment(td, cb_size_segment, seginfo, flags); 1745 } 1746 1747 /* 1748 * For each writable segment in the process's memory map, call the given 1749 * function with a pointer to the map entry and some arbitrary 1750 * caller-supplied data. 1751 */ 1752 static void 1753 each_dumpable_segment(struct thread *td, segment_callback func, void *closure, 1754 int flags) 1755 { 1756 struct proc *p = td->td_proc; 1757 vm_map_t map = &p->p_vmspace->vm_map; 1758 vm_map_entry_t entry; 1759 vm_object_t backing_object, object; 1760 bool ignore_entry; 1761 1762 vm_map_lock_read(map); 1763 VM_MAP_ENTRY_FOREACH(entry, map) { 1764 /* 1765 * Don't dump inaccessible mappings, deal with legacy 1766 * coredump mode. 1767 * 1768 * Note that read-only segments related to the elf binary 1769 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1770 * need to arbitrarily ignore such segments. 1771 */ 1772 if ((flags & SVC_ALL) == 0) { 1773 if (elf_legacy_coredump) { 1774 if ((entry->protection & VM_PROT_RW) != 1775 VM_PROT_RW) 1776 continue; 1777 } else { 1778 if ((entry->protection & VM_PROT_ALL) == 0) 1779 continue; 1780 } 1781 } 1782 1783 /* 1784 * Dont include memory segment in the coredump if 1785 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1786 * madvise(2). Do not dump submaps (i.e. parts of the 1787 * kernel map). 1788 */ 1789 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 1790 continue; 1791 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 && 1792 (flags & SVC_ALL) == 0) 1793 continue; 1794 if ((object = entry->object.vm_object) == NULL) 1795 continue; 1796 1797 /* Ignore memory-mapped devices and such things. */ 1798 VM_OBJECT_RLOCK(object); 1799 while ((backing_object = object->backing_object) != NULL) { 1800 VM_OBJECT_RLOCK(backing_object); 1801 VM_OBJECT_RUNLOCK(object); 1802 object = backing_object; 1803 } 1804 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0; 1805 VM_OBJECT_RUNLOCK(object); 1806 if (ignore_entry) 1807 continue; 1808 1809 (*func)(entry, closure); 1810 } 1811 vm_map_unlock_read(map); 1812 } 1813 1814 /* 1815 * Write the core file header to the file, including padding up to 1816 * the page boundary. 1817 */ 1818 static int 1819 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1820 size_t hdrsize, struct note_info_list *notelst, size_t notesz, 1821 int flags) 1822 { 1823 struct note_info *ninfo; 1824 struct sbuf *sb; 1825 int error; 1826 1827 /* Fill in the header. */ 1828 bzero(hdr, hdrsize); 1829 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags); 1830 1831 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1832 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1833 sbuf_start_section(sb, NULL); 1834 sbuf_bcat(sb, hdr, hdrsize); 1835 TAILQ_FOREACH(ninfo, notelst, link) 1836 __elfN(putnote)(p->td, ninfo, sb); 1837 /* Align up to a page boundary for the program segments. */ 1838 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1839 error = sbuf_finish(sb); 1840 sbuf_delete(sb); 1841 1842 return (error); 1843 } 1844 1845 void 1846 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1847 size_t *sizep) 1848 { 1849 struct proc *p; 1850 struct thread *thr; 1851 size_t size; 1852 1853 p = td->td_proc; 1854 size = 0; 1855 1856 size += __elfN(register_note)(td, list, NT_PRPSINFO, 1857 __elfN(note_prpsinfo), p); 1858 1859 /* 1860 * To have the debugger select the right thread (LWP) as the initial 1861 * thread, we dump the state of the thread passed to us in td first. 1862 * This is the thread that causes the core dump and thus likely to 1863 * be the right thread one wants to have selected in the debugger. 1864 */ 1865 thr = td; 1866 while (thr != NULL) { 1867 size += __elfN(prepare_register_notes)(td, list, thr); 1868 size += __elfN(register_note)(td, list, -1, 1869 __elfN(note_threadmd), thr); 1870 1871 thr = thr == td ? TAILQ_FIRST(&p->p_threads) : 1872 TAILQ_NEXT(thr, td_plist); 1873 if (thr == td) 1874 thr = TAILQ_NEXT(thr, td_plist); 1875 } 1876 1877 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC, 1878 __elfN(note_procstat_proc), p); 1879 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES, 1880 note_procstat_files, p); 1881 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP, 1882 note_procstat_vmmap, p); 1883 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS, 1884 note_procstat_groups, p); 1885 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK, 1886 note_procstat_umask, p); 1887 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT, 1888 note_procstat_rlimit, p); 1889 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL, 1890 note_procstat_osrel, p); 1891 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS, 1892 __elfN(note_procstat_psstrings), p); 1893 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV, 1894 __elfN(note_procstat_auxv), p); 1895 1896 *sizep = size; 1897 } 1898 1899 void 1900 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1901 size_t notesz, int flags) 1902 { 1903 Elf_Ehdr *ehdr; 1904 Elf_Phdr *phdr; 1905 Elf_Shdr *shdr; 1906 struct phdr_closure phc; 1907 Elf_Brandinfo *bi; 1908 1909 ehdr = (Elf_Ehdr *)hdr; 1910 bi = td->td_proc->p_elf_brandinfo; 1911 1912 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1913 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1914 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1915 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1916 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1917 ehdr->e_ident[EI_DATA] = ELF_DATA; 1918 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1919 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi; 1920 ehdr->e_ident[EI_ABIVERSION] = 0; 1921 ehdr->e_ident[EI_PAD] = 0; 1922 ehdr->e_type = ET_CORE; 1923 ehdr->e_machine = bi->machine; 1924 ehdr->e_version = EV_CURRENT; 1925 ehdr->e_entry = 0; 1926 ehdr->e_phoff = sizeof(Elf_Ehdr); 1927 ehdr->e_flags = td->td_proc->p_elf_flags; 1928 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1929 ehdr->e_phentsize = sizeof(Elf_Phdr); 1930 ehdr->e_shentsize = sizeof(Elf_Shdr); 1931 ehdr->e_shstrndx = SHN_UNDEF; 1932 if (numsegs + 1 < PN_XNUM) { 1933 ehdr->e_phnum = numsegs + 1; 1934 ehdr->e_shnum = 0; 1935 } else { 1936 ehdr->e_phnum = PN_XNUM; 1937 ehdr->e_shnum = 1; 1938 1939 ehdr->e_shoff = ehdr->e_phoff + 1940 (numsegs + 1) * ehdr->e_phentsize; 1941 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1942 ("e_shoff: %zu, hdrsize - shdr: %zu", 1943 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1944 1945 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1946 memset(shdr, 0, sizeof(*shdr)); 1947 /* 1948 * A special first section is used to hold large segment and 1949 * section counts. This was proposed by Sun Microsystems in 1950 * Solaris and has been adopted by Linux; the standard ELF 1951 * tools are already familiar with the technique. 1952 * 1953 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1954 * (or 12-7 depending on the version of the document) for more 1955 * details. 1956 */ 1957 shdr->sh_type = SHT_NULL; 1958 shdr->sh_size = ehdr->e_shnum; 1959 shdr->sh_link = ehdr->e_shstrndx; 1960 shdr->sh_info = numsegs + 1; 1961 } 1962 1963 /* 1964 * Fill in the program header entries. 1965 */ 1966 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1967 1968 /* The note segement. */ 1969 phdr->p_type = PT_NOTE; 1970 phdr->p_offset = hdrsize; 1971 phdr->p_vaddr = 0; 1972 phdr->p_paddr = 0; 1973 phdr->p_filesz = notesz; 1974 phdr->p_memsz = 0; 1975 phdr->p_flags = PF_R; 1976 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1977 phdr++; 1978 1979 /* All the writable segments from the program. */ 1980 phc.phdr = phdr; 1981 phc.offset = round_page(hdrsize + notesz); 1982 each_dumpable_segment(td, cb_put_phdr, &phc, flags); 1983 } 1984 1985 static size_t 1986 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list, 1987 struct regset *regset, struct thread *target_td) 1988 { 1989 const struct sysentvec *sv; 1990 struct note_info *ninfo; 1991 size_t size, notesize; 1992 1993 size = 0; 1994 if (!regset->get(regset, target_td, NULL, &size) || size == 0) 1995 return (0); 1996 1997 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1998 ninfo->type = regset->note; 1999 ninfo->regset = regset; 2000 ninfo->outarg = target_td; 2001 ninfo->outsize = size; 2002 TAILQ_INSERT_TAIL(list, ninfo, link); 2003 2004 sv = td->td_proc->p_sysent; 2005 notesize = sizeof(Elf_Note) + /* note header */ 2006 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2007 /* note name */ 2008 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2009 2010 return (notesize); 2011 } 2012 2013 size_t 2014 __elfN(register_note)(struct thread *td, struct note_info_list *list, 2015 int type, outfunc_t out, void *arg) 2016 { 2017 const struct sysentvec *sv; 2018 struct note_info *ninfo; 2019 size_t size, notesize; 2020 2021 sv = td->td_proc->p_sysent; 2022 size = 0; 2023 out(arg, NULL, &size); 2024 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2025 ninfo->type = type; 2026 ninfo->outfunc = out; 2027 ninfo->outarg = arg; 2028 ninfo->outsize = size; 2029 TAILQ_INSERT_TAIL(list, ninfo, link); 2030 2031 if (type == -1) 2032 return (size); 2033 2034 notesize = sizeof(Elf_Note) + /* note header */ 2035 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2036 /* note name */ 2037 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2038 2039 return (notesize); 2040 } 2041 2042 static size_t 2043 append_note_data(const void *src, void *dst, size_t len) 2044 { 2045 size_t padded_len; 2046 2047 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 2048 if (dst != NULL) { 2049 bcopy(src, dst, len); 2050 bzero((char *)dst + len, padded_len - len); 2051 } 2052 return (padded_len); 2053 } 2054 2055 size_t 2056 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 2057 { 2058 Elf_Note *note; 2059 char *buf; 2060 size_t notesize; 2061 2062 buf = dst; 2063 if (buf != NULL) { 2064 note = (Elf_Note *)buf; 2065 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 2066 note->n_descsz = size; 2067 note->n_type = type; 2068 buf += sizeof(*note); 2069 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 2070 sizeof(FREEBSD_ABI_VENDOR)); 2071 append_note_data(src, buf, size); 2072 if (descp != NULL) 2073 *descp = buf; 2074 } 2075 2076 notesize = sizeof(Elf_Note) + /* note header */ 2077 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 2078 /* note name */ 2079 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2080 2081 return (notesize); 2082 } 2083 2084 static void 2085 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb) 2086 { 2087 Elf_Note note; 2088 const struct sysentvec *sv; 2089 ssize_t old_len, sect_len; 2090 size_t new_len, descsz, i; 2091 2092 if (ninfo->type == -1) { 2093 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2094 return; 2095 } 2096 2097 sv = td->td_proc->p_sysent; 2098 2099 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1; 2100 note.n_descsz = ninfo->outsize; 2101 note.n_type = ninfo->type; 2102 2103 sbuf_bcat(sb, ¬e, sizeof(note)); 2104 sbuf_start_section(sb, &old_len); 2105 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor, 2106 strlen(sv->sv_elf_core_abi_vendor) + 1); 2107 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2108 if (note.n_descsz == 0) 2109 return; 2110 sbuf_start_section(sb, &old_len); 2111 if (ninfo->regset != NULL) { 2112 struct regset *regset = ninfo->regset; 2113 void *buf; 2114 2115 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK); 2116 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize); 2117 sbuf_bcat(sb, buf, ninfo->outsize); 2118 free(buf, M_TEMP); 2119 } else 2120 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2121 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2122 if (sect_len < 0) 2123 return; 2124 2125 new_len = (size_t)sect_len; 2126 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 2127 if (new_len < descsz) { 2128 /* 2129 * It is expected that individual note emitters will correctly 2130 * predict their expected output size and fill up to that size 2131 * themselves, padding in a format-specific way if needed. 2132 * However, in case they don't, just do it here with zeros. 2133 */ 2134 for (i = 0; i < descsz - new_len; i++) 2135 sbuf_putc(sb, 0); 2136 } else if (new_len > descsz) { 2137 /* 2138 * We can't always truncate sb -- we may have drained some 2139 * of it already. 2140 */ 2141 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 2142 "read it (%zu > %zu). Since it is longer than " 2143 "expected, this coredump's notes are corrupt. THIS " 2144 "IS A BUG in the note_procstat routine for type %u.\n", 2145 __func__, (unsigned)note.n_type, new_len, descsz, 2146 (unsigned)note.n_type)); 2147 } 2148 } 2149 2150 /* 2151 * Miscellaneous note out functions. 2152 */ 2153 2154 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2155 #include <compat/freebsd32/freebsd32.h> 2156 #include <compat/freebsd32/freebsd32_signal.h> 2157 2158 typedef struct prstatus32 elf_prstatus_t; 2159 typedef struct prpsinfo32 elf_prpsinfo_t; 2160 typedef struct fpreg32 elf_prfpregset_t; 2161 typedef struct fpreg32 elf_fpregset_t; 2162 typedef struct reg32 elf_gregset_t; 2163 typedef struct thrmisc32 elf_thrmisc_t; 2164 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t; 2165 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 2166 typedef struct kinfo_proc32 elf_kinfo_proc_t; 2167 typedef uint32_t elf_ps_strings_t; 2168 #else 2169 typedef prstatus_t elf_prstatus_t; 2170 typedef prpsinfo_t elf_prpsinfo_t; 2171 typedef prfpregset_t elf_prfpregset_t; 2172 typedef prfpregset_t elf_fpregset_t; 2173 typedef gregset_t elf_gregset_t; 2174 typedef thrmisc_t elf_thrmisc_t; 2175 typedef struct ptrace_lwpinfo elf_lwpinfo_t; 2176 #define ELF_KERN_PROC_MASK 0 2177 typedef struct kinfo_proc elf_kinfo_proc_t; 2178 typedef vm_offset_t elf_ps_strings_t; 2179 #endif 2180 2181 static void 2182 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2183 { 2184 struct sbuf sbarg; 2185 size_t len; 2186 char *cp, *end; 2187 struct proc *p; 2188 elf_prpsinfo_t *psinfo; 2189 int error; 2190 2191 p = arg; 2192 if (sb != NULL) { 2193 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 2194 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 2195 psinfo->pr_version = PRPSINFO_VERSION; 2196 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 2197 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 2198 PROC_LOCK(p); 2199 if (p->p_args != NULL) { 2200 len = sizeof(psinfo->pr_psargs) - 1; 2201 if (len > p->p_args->ar_length) 2202 len = p->p_args->ar_length; 2203 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 2204 PROC_UNLOCK(p); 2205 error = 0; 2206 } else { 2207 _PHOLD(p); 2208 PROC_UNLOCK(p); 2209 sbuf_new(&sbarg, psinfo->pr_psargs, 2210 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 2211 error = proc_getargv(curthread, p, &sbarg); 2212 PRELE(p); 2213 if (sbuf_finish(&sbarg) == 0) { 2214 len = sbuf_len(&sbarg); 2215 if (len > 0) 2216 len--; 2217 } else { 2218 len = sizeof(psinfo->pr_psargs) - 1; 2219 } 2220 sbuf_delete(&sbarg); 2221 } 2222 if (error != 0 || len == 0 || (ssize_t)len == -1) 2223 strlcpy(psinfo->pr_psargs, p->p_comm, 2224 sizeof(psinfo->pr_psargs)); 2225 else { 2226 KASSERT(len < sizeof(psinfo->pr_psargs), 2227 ("len is too long: %zu vs %zu", len, 2228 sizeof(psinfo->pr_psargs))); 2229 cp = psinfo->pr_psargs; 2230 end = cp + len - 1; 2231 for (;;) { 2232 cp = memchr(cp, '\0', end - cp); 2233 if (cp == NULL) 2234 break; 2235 *cp = ' '; 2236 } 2237 } 2238 psinfo->pr_pid = p->p_pid; 2239 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 2240 free(psinfo, M_TEMP); 2241 } 2242 *sizep = sizeof(*psinfo); 2243 } 2244 2245 static bool 2246 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf, 2247 size_t *sizep) 2248 { 2249 elf_prstatus_t *status; 2250 2251 if (buf != NULL) { 2252 KASSERT(*sizep == sizeof(*status), ("%s: invalid size", 2253 __func__)); 2254 status = buf; 2255 memset(status, 0, *sizep); 2256 status->pr_version = PRSTATUS_VERSION; 2257 status->pr_statussz = sizeof(elf_prstatus_t); 2258 status->pr_gregsetsz = sizeof(elf_gregset_t); 2259 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 2260 status->pr_osreldate = osreldate; 2261 status->pr_cursig = td->td_proc->p_sig; 2262 status->pr_pid = td->td_tid; 2263 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2264 fill_regs32(td, &status->pr_reg); 2265 #else 2266 fill_regs(td, &status->pr_reg); 2267 #endif 2268 } 2269 *sizep = sizeof(*status); 2270 return (true); 2271 } 2272 2273 static bool 2274 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf, 2275 size_t size) 2276 { 2277 elf_prstatus_t *status; 2278 2279 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__)); 2280 status = buf; 2281 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2282 set_regs32(td, &status->pr_reg); 2283 #else 2284 set_regs(td, &status->pr_reg); 2285 #endif 2286 return (true); 2287 } 2288 2289 static struct regset __elfN(regset_prstatus) = { 2290 .note = NT_PRSTATUS, 2291 .size = sizeof(elf_prstatus_t), 2292 .get = __elfN(get_prstatus), 2293 .set = __elfN(set_prstatus), 2294 }; 2295 ELF_REGSET(__elfN(regset_prstatus)); 2296 2297 static bool 2298 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf, 2299 size_t *sizep) 2300 { 2301 elf_prfpregset_t *fpregset; 2302 2303 if (buf != NULL) { 2304 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size", 2305 __func__)); 2306 fpregset = buf; 2307 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2308 fill_fpregs32(td, fpregset); 2309 #else 2310 fill_fpregs(td, fpregset); 2311 #endif 2312 } 2313 *sizep = sizeof(*fpregset); 2314 return (true); 2315 } 2316 2317 static bool 2318 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf, 2319 size_t size) 2320 { 2321 elf_prfpregset_t *fpregset; 2322 2323 fpregset = buf; 2324 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__)); 2325 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2326 set_fpregs32(td, fpregset); 2327 #else 2328 set_fpregs(td, fpregset); 2329 #endif 2330 return (true); 2331 } 2332 2333 static struct regset __elfN(regset_fpregset) = { 2334 .note = NT_FPREGSET, 2335 .size = sizeof(elf_prfpregset_t), 2336 .get = __elfN(get_fpregset), 2337 .set = __elfN(set_fpregset), 2338 }; 2339 ELF_REGSET(__elfN(regset_fpregset)); 2340 2341 static bool 2342 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf, 2343 size_t *sizep) 2344 { 2345 elf_thrmisc_t *thrmisc; 2346 2347 if (buf != NULL) { 2348 KASSERT(*sizep == sizeof(*thrmisc), 2349 ("%s: invalid size", __func__)); 2350 thrmisc = buf; 2351 bzero(thrmisc, sizeof(*thrmisc)); 2352 strcpy(thrmisc->pr_tname, td->td_name); 2353 } 2354 *sizep = sizeof(*thrmisc); 2355 return (true); 2356 } 2357 2358 static struct regset __elfN(regset_thrmisc) = { 2359 .note = NT_THRMISC, 2360 .size = sizeof(elf_thrmisc_t), 2361 .get = __elfN(get_thrmisc), 2362 }; 2363 ELF_REGSET(__elfN(regset_thrmisc)); 2364 2365 static bool 2366 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf, 2367 size_t *sizep) 2368 { 2369 elf_lwpinfo_t pl; 2370 size_t size; 2371 int structsize; 2372 2373 size = sizeof(structsize) + sizeof(pl); 2374 if (buf != NULL) { 2375 KASSERT(*sizep == size, ("%s: invalid size", __func__)); 2376 structsize = sizeof(pl); 2377 memcpy(buf, &structsize, sizeof(structsize)); 2378 bzero(&pl, sizeof(pl)); 2379 pl.pl_lwpid = td->td_tid; 2380 pl.pl_event = PL_EVENT_NONE; 2381 pl.pl_sigmask = td->td_sigmask; 2382 pl.pl_siglist = td->td_siglist; 2383 if (td->td_si.si_signo != 0) { 2384 pl.pl_event = PL_EVENT_SIGNAL; 2385 pl.pl_flags |= PL_FLAG_SI; 2386 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2387 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2388 #else 2389 pl.pl_siginfo = td->td_si; 2390 #endif 2391 } 2392 strcpy(pl.pl_tdname, td->td_name); 2393 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2394 memcpy((int *)buf + 1, &pl, sizeof(pl)); 2395 } 2396 *sizep = size; 2397 return (true); 2398 } 2399 2400 static struct regset __elfN(regset_lwpinfo) = { 2401 .note = NT_PTLWPINFO, 2402 .size = sizeof(int) + sizeof(elf_lwpinfo_t), 2403 .get = __elfN(get_lwpinfo), 2404 }; 2405 ELF_REGSET(__elfN(regset_lwpinfo)); 2406 2407 static size_t 2408 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list, 2409 struct thread *target_td) 2410 { 2411 struct sysentvec *sv = td->td_proc->p_sysent; 2412 struct regset **regsetp, **regset_end, *regset; 2413 size_t size; 2414 2415 size = 0; 2416 2417 /* NT_PRSTATUS must be the first register set note. */ 2418 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus), 2419 target_td); 2420 2421 regsetp = sv->sv_regset_begin; 2422 if (regsetp == NULL) { 2423 /* XXX: This shouldn't be true for any FreeBSD ABIs. */ 2424 size += __elfN(register_regset_note)(td, list, 2425 &__elfN(regset_fpregset), target_td); 2426 return (size); 2427 } 2428 regset_end = sv->sv_regset_end; 2429 MPASS(regset_end != NULL); 2430 for (; regsetp < regset_end; regsetp++) { 2431 regset = *regsetp; 2432 if (regset->note == NT_PRSTATUS) 2433 continue; 2434 size += __elfN(register_regset_note)(td, list, regset, 2435 target_td); 2436 } 2437 return (size); 2438 } 2439 2440 /* 2441 * Allow for MD specific notes, as well as any MD 2442 * specific preparations for writing MI notes. 2443 */ 2444 static void 2445 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2446 { 2447 struct thread *td; 2448 void *buf; 2449 size_t size; 2450 2451 td = (struct thread *)arg; 2452 size = *sizep; 2453 if (size != 0 && sb != NULL) 2454 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2455 else 2456 buf = NULL; 2457 size = 0; 2458 __elfN(dump_thread)(td, buf, &size); 2459 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2460 if (size != 0 && sb != NULL) 2461 sbuf_bcat(sb, buf, size); 2462 free(buf, M_TEMP); 2463 *sizep = size; 2464 } 2465 2466 #ifdef KINFO_PROC_SIZE 2467 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2468 #endif 2469 2470 static void 2471 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2472 { 2473 struct proc *p; 2474 size_t size; 2475 int structsize; 2476 2477 p = arg; 2478 size = sizeof(structsize) + p->p_numthreads * 2479 sizeof(elf_kinfo_proc_t); 2480 2481 if (sb != NULL) { 2482 KASSERT(*sizep == size, ("invalid size")); 2483 structsize = sizeof(elf_kinfo_proc_t); 2484 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2485 sx_slock(&proctree_lock); 2486 PROC_LOCK(p); 2487 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2488 sx_sunlock(&proctree_lock); 2489 } 2490 *sizep = size; 2491 } 2492 2493 #ifdef KINFO_FILE_SIZE 2494 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2495 #endif 2496 2497 static void 2498 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2499 { 2500 struct proc *p; 2501 size_t size, sect_sz, i; 2502 ssize_t start_len, sect_len; 2503 int structsize, filedesc_flags; 2504 2505 if (coredump_pack_fileinfo) 2506 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2507 else 2508 filedesc_flags = 0; 2509 2510 p = arg; 2511 structsize = sizeof(struct kinfo_file); 2512 if (sb == NULL) { 2513 size = 0; 2514 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2515 sbuf_set_drain(sb, sbuf_count_drain, &size); 2516 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2517 PROC_LOCK(p); 2518 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2519 sbuf_finish(sb); 2520 sbuf_delete(sb); 2521 *sizep = size; 2522 } else { 2523 sbuf_start_section(sb, &start_len); 2524 2525 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2526 PROC_LOCK(p); 2527 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2528 filedesc_flags); 2529 2530 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2531 if (sect_len < 0) 2532 return; 2533 sect_sz = sect_len; 2534 2535 KASSERT(sect_sz <= *sizep, 2536 ("kern_proc_filedesc_out did not respect maxlen; " 2537 "requested %zu, got %zu", *sizep - sizeof(structsize), 2538 sect_sz - sizeof(structsize))); 2539 2540 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2541 sbuf_putc(sb, 0); 2542 } 2543 } 2544 2545 #ifdef KINFO_VMENTRY_SIZE 2546 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2547 #endif 2548 2549 static void 2550 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2551 { 2552 struct proc *p; 2553 size_t size; 2554 int structsize, vmmap_flags; 2555 2556 if (coredump_pack_vmmapinfo) 2557 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2558 else 2559 vmmap_flags = 0; 2560 2561 p = arg; 2562 structsize = sizeof(struct kinfo_vmentry); 2563 if (sb == NULL) { 2564 size = 0; 2565 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2566 sbuf_set_drain(sb, sbuf_count_drain, &size); 2567 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2568 PROC_LOCK(p); 2569 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2570 sbuf_finish(sb); 2571 sbuf_delete(sb); 2572 *sizep = size; 2573 } else { 2574 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2575 PROC_LOCK(p); 2576 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2577 vmmap_flags); 2578 } 2579 } 2580 2581 static void 2582 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2583 { 2584 struct proc *p; 2585 size_t size; 2586 int structsize; 2587 2588 p = arg; 2589 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2590 if (sb != NULL) { 2591 KASSERT(*sizep == size, ("invalid size")); 2592 structsize = sizeof(gid_t); 2593 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2594 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2595 sizeof(gid_t)); 2596 } 2597 *sizep = size; 2598 } 2599 2600 static void 2601 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2602 { 2603 struct proc *p; 2604 size_t size; 2605 int structsize; 2606 2607 p = arg; 2608 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask); 2609 if (sb != NULL) { 2610 KASSERT(*sizep == size, ("invalid size")); 2611 structsize = sizeof(p->p_pd->pd_cmask); 2612 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2613 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask)); 2614 } 2615 *sizep = size; 2616 } 2617 2618 static void 2619 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2620 { 2621 struct proc *p; 2622 struct rlimit rlim[RLIM_NLIMITS]; 2623 size_t size; 2624 int structsize, i; 2625 2626 p = arg; 2627 size = sizeof(structsize) + sizeof(rlim); 2628 if (sb != NULL) { 2629 KASSERT(*sizep == size, ("invalid size")); 2630 structsize = sizeof(rlim); 2631 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2632 PROC_LOCK(p); 2633 for (i = 0; i < RLIM_NLIMITS; i++) 2634 lim_rlimit_proc(p, i, &rlim[i]); 2635 PROC_UNLOCK(p); 2636 sbuf_bcat(sb, rlim, sizeof(rlim)); 2637 } 2638 *sizep = size; 2639 } 2640 2641 static void 2642 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2643 { 2644 struct proc *p; 2645 size_t size; 2646 int structsize; 2647 2648 p = arg; 2649 size = sizeof(structsize) + sizeof(p->p_osrel); 2650 if (sb != NULL) { 2651 KASSERT(*sizep == size, ("invalid size")); 2652 structsize = sizeof(p->p_osrel); 2653 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2654 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2655 } 2656 *sizep = size; 2657 } 2658 2659 static void 2660 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2661 { 2662 struct proc *p; 2663 elf_ps_strings_t ps_strings; 2664 size_t size; 2665 int structsize; 2666 2667 p = arg; 2668 size = sizeof(structsize) + sizeof(ps_strings); 2669 if (sb != NULL) { 2670 KASSERT(*sizep == size, ("invalid size")); 2671 structsize = sizeof(ps_strings); 2672 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2673 ps_strings = PTROUT(PROC_PS_STRINGS(p)); 2674 #else 2675 ps_strings = PROC_PS_STRINGS(p); 2676 #endif 2677 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2678 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2679 } 2680 *sizep = size; 2681 } 2682 2683 static void 2684 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2685 { 2686 struct proc *p; 2687 size_t size; 2688 int structsize; 2689 2690 p = arg; 2691 if (sb == NULL) { 2692 size = 0; 2693 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo), 2694 SBUF_FIXEDLEN); 2695 sbuf_set_drain(sb, sbuf_count_drain, &size); 2696 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2697 PHOLD(p); 2698 proc_getauxv(curthread, p, sb); 2699 PRELE(p); 2700 sbuf_finish(sb); 2701 sbuf_delete(sb); 2702 *sizep = size; 2703 } else { 2704 structsize = sizeof(Elf_Auxinfo); 2705 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2706 PHOLD(p); 2707 proc_getauxv(curthread, p, sb); 2708 PRELE(p); 2709 } 2710 } 2711 2712 #define MAX_NOTES_LOOP 4096 2713 bool 2714 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote, 2715 const char *note_vendor, const Elf_Phdr *pnote, 2716 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg) 2717 { 2718 const Elf_Note *note, *note0, *note_end; 2719 const char *note_name; 2720 char *buf; 2721 int i, error; 2722 bool res; 2723 2724 /* We need some limit, might as well use PAGE_SIZE. */ 2725 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2726 return (false); 2727 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2728 if (pnote->p_offset > PAGE_SIZE || 2729 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2730 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT); 2731 if (buf == NULL) { 2732 VOP_UNLOCK(imgp->vp); 2733 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2734 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 2735 } 2736 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2737 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2738 curthread->td_ucred, NOCRED, NULL, curthread); 2739 if (error != 0) { 2740 uprintf("i/o error PT_NOTE\n"); 2741 goto retf; 2742 } 2743 note = note0 = (const Elf_Note *)buf; 2744 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2745 } else { 2746 note = note0 = (const Elf_Note *)(imgp->image_header + 2747 pnote->p_offset); 2748 note_end = (const Elf_Note *)(imgp->image_header + 2749 pnote->p_offset + pnote->p_filesz); 2750 buf = NULL; 2751 } 2752 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end; 2753 i++) { 2754 if (!aligned(note, Elf32_Addr)) { 2755 uprintf("Unaligned ELF note\n"); 2756 goto retf; 2757 } 2758 if ((const char *)note_end - (const char *)note < 2759 sizeof(Elf_Note)) { 2760 uprintf("ELF note to short\n"); 2761 goto retf; 2762 } 2763 if (note->n_namesz != checknote->n_namesz || 2764 note->n_descsz != checknote->n_descsz || 2765 note->n_type != checknote->n_type) 2766 goto nextnote; 2767 note_name = (const char *)(note + 1); 2768 if (note_name + checknote->n_namesz >= 2769 (const char *)note_end || strncmp(note_vendor, 2770 note_name, checknote->n_namesz) != 0) 2771 goto nextnote; 2772 2773 if (cb(note, cb_arg, &res)) 2774 goto ret; 2775 nextnote: 2776 note = (const Elf_Note *)((const char *)(note + 1) + 2777 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2778 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2779 } 2780 if (i >= MAX_NOTES_LOOP) 2781 uprintf("ELF note parser reached %d notes\n", i); 2782 retf: 2783 res = false; 2784 ret: 2785 free(buf, M_TEMP); 2786 return (res); 2787 } 2788 2789 struct brandnote_cb_arg { 2790 Elf_Brandnote *brandnote; 2791 int32_t *osrel; 2792 }; 2793 2794 static bool 2795 brandnote_cb(const Elf_Note *note, void *arg0, bool *res) 2796 { 2797 struct brandnote_cb_arg *arg; 2798 2799 arg = arg0; 2800 2801 /* 2802 * Fetch the osreldate for binary from the ELF OSABI-note if 2803 * necessary. 2804 */ 2805 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 && 2806 arg->brandnote->trans_osrel != NULL ? 2807 arg->brandnote->trans_osrel(note, arg->osrel) : true; 2808 2809 return (true); 2810 } 2811 2812 static Elf_Note fctl_note = { 2813 .n_namesz = sizeof(FREEBSD_ABI_VENDOR), 2814 .n_descsz = sizeof(uint32_t), 2815 .n_type = NT_FREEBSD_FEATURE_CTL, 2816 }; 2817 2818 struct fctl_cb_arg { 2819 bool *has_fctl0; 2820 uint32_t *fctl0; 2821 }; 2822 2823 static bool 2824 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res) 2825 { 2826 struct fctl_cb_arg *arg; 2827 const Elf32_Word *desc; 2828 uintptr_t p; 2829 2830 arg = arg0; 2831 p = (uintptr_t)(note + 1); 2832 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 2833 desc = (const Elf32_Word *)p; 2834 *arg->has_fctl0 = true; 2835 *arg->fctl0 = desc[0]; 2836 *res = true; 2837 return (true); 2838 } 2839 2840 /* 2841 * Try to find the appropriate ABI-note section for checknote, fetch 2842 * the osreldate and feature control flags for binary from the ELF 2843 * OSABI-note. Only the first page of the image is searched, the same 2844 * as for headers. 2845 */ 2846 static bool 2847 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote, 2848 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0) 2849 { 2850 const Elf_Phdr *phdr; 2851 const Elf_Ehdr *hdr; 2852 struct brandnote_cb_arg b_arg; 2853 struct fctl_cb_arg f_arg; 2854 int i, j; 2855 2856 hdr = (const Elf_Ehdr *)imgp->image_header; 2857 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2858 b_arg.brandnote = brandnote; 2859 b_arg.osrel = osrel; 2860 f_arg.has_fctl0 = has_fctl0; 2861 f_arg.fctl0 = fctl0; 2862 2863 for (i = 0; i < hdr->e_phnum; i++) { 2864 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, 2865 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb, 2866 &b_arg)) { 2867 for (j = 0; j < hdr->e_phnum; j++) { 2868 if (phdr[j].p_type == PT_NOTE && 2869 __elfN(parse_notes)(imgp, &fctl_note, 2870 FREEBSD_ABI_VENDOR, &phdr[j], 2871 note_fctl_cb, &f_arg)) 2872 break; 2873 } 2874 return (true); 2875 } 2876 } 2877 return (false); 2878 2879 } 2880 2881 /* 2882 * Tell kern_execve.c about it, with a little help from the linker. 2883 */ 2884 static struct execsw __elfN(execsw) = { 2885 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 2886 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2887 }; 2888 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2889 2890 static vm_prot_t 2891 __elfN(trans_prot)(Elf_Word flags) 2892 { 2893 vm_prot_t prot; 2894 2895 prot = 0; 2896 if (flags & PF_X) 2897 prot |= VM_PROT_EXECUTE; 2898 if (flags & PF_W) 2899 prot |= VM_PROT_WRITE; 2900 if (flags & PF_R) 2901 prot |= VM_PROT_READ; 2902 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 2903 if (i386_read_exec && (flags & PF_R)) 2904 prot |= VM_PROT_EXECUTE; 2905 #endif 2906 return (prot); 2907 } 2908 2909 static Elf_Word 2910 __elfN(untrans_prot)(vm_prot_t prot) 2911 { 2912 Elf_Word flags; 2913 2914 flags = 0; 2915 if (prot & VM_PROT_EXECUTE) 2916 flags |= PF_X; 2917 if (prot & VM_PROT_READ) 2918 flags |= PF_R; 2919 if (prot & VM_PROT_WRITE) 2920 flags |= PF_W; 2921 return (flags); 2922 } 2923