1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2017 Dell EMC 5 * Copyright (c) 2000-2001, 2003 David O'Brien 6 * Copyright (c) 1995-1996 Søren Schmidt 7 * Copyright (c) 1996 Peter Wemm 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer 15 * in this position and unchanged. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. The name of the author may not be used to endorse or promote products 20 * derived from this software without specific prior written permission 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "opt_capsicum.h" 35 36 #include <sys/param.h> 37 #include <sys/capsicum.h> 38 #include <sys/compressor.h> 39 #include <sys/exec.h> 40 #include <sys/fcntl.h> 41 #include <sys/imgact.h> 42 #include <sys/imgact_elf.h> 43 #include <sys/jail.h> 44 #include <sys/kernel.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mman.h> 49 #include <sys/namei.h> 50 #include <sys/proc.h> 51 #include <sys/procfs.h> 52 #include <sys/ptrace.h> 53 #include <sys/racct.h> 54 #include <sys/reg.h> 55 #include <sys/resourcevar.h> 56 #include <sys/rwlock.h> 57 #include <sys/sbuf.h> 58 #include <sys/sf_buf.h> 59 #include <sys/smp.h> 60 #include <sys/systm.h> 61 #include <sys/signalvar.h> 62 #include <sys/stat.h> 63 #include <sys/sx.h> 64 #include <sys/syscall.h> 65 #include <sys/sysctl.h> 66 #include <sys/sysent.h> 67 #include <sys/ucoredump.h> 68 #include <sys/vnode.h> 69 #include <sys/syslog.h> 70 #include <sys/eventhandler.h> 71 #include <sys/user.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_kern.h> 75 #include <vm/vm_param.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_object.h> 79 #include <vm/vm_extern.h> 80 81 #include <machine/elf.h> 82 #include <machine/md_var.h> 83 84 #define ELF_NOTE_ROUNDSIZE 4 85 #define OLD_EI_BRAND 8 86 87 /* 88 * ELF_ABI_NAME is a string name of the ELF ABI. ELF_ABI_ID is used 89 * to build variable names. 90 */ 91 #define ELF_ABI_NAME __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 92 #define ELF_ABI_ID __CONCAT(elf, __ELF_WORD_SIZE) 93 94 static int __elfN(check_header)(const Elf_Ehdr *hdr); 95 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 96 const char *interp, int32_t *osrel, uint32_t *fctl0); 97 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 98 u_long *entry); 99 static int __elfN(load_section)(const struct image_params *imgp, 100 vm_ooffset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 101 vm_prot_t prot); 102 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 103 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note, 104 int32_t *osrel); 105 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 106 static bool __elfN(check_note)(struct image_params *imgp, 107 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0, 108 uint32_t *fctl0); 109 static vm_prot_t __elfN(trans_prot)(Elf_Word); 110 static Elf_Word __elfN(untrans_prot)(vm_prot_t); 111 static size_t __elfN(prepare_register_notes)(struct thread *td, 112 struct note_info_list *list, struct thread *target_td); 113 114 SYSCTL_NODE(_kern, OID_AUTO, ELF_ABI_ID, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 115 ""); 116 117 #define ELF_NODE_OID __CONCAT(_kern_, ELF_ABI_ID) 118 119 int __elfN(fallback_brand) = -1; 120 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, 121 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, 122 ELF_ABI_NAME " brand of last resort"); 123 124 static int elf_legacy_coredump = 0; 125 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 126 &elf_legacy_coredump, 0, 127 "include all and only RW pages in core dumps"); 128 129 int __elfN(nxstack) = 130 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \ 131 defined(__arm__) || defined(__aarch64__) || \ 132 defined(__riscv) 133 1; 134 #else 135 0; 136 #endif 137 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, 138 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, 139 ELF_ABI_NAME ": support PT_GNU_STACK for non-executable stack control"); 140 141 #if defined(__amd64__) 142 static int __elfN(vdso) = 1; 143 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, 144 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0, 145 ELF_ABI_NAME ": enable vdso preloading"); 146 #else 147 static int __elfN(vdso) = 0; 148 #endif 149 150 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 151 int i386_read_exec = 0; 152 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 153 "enable execution from readable segments"); 154 #endif 155 156 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR; 157 static int 158 sysctl_pie_base(SYSCTL_HANDLER_ARGS) 159 { 160 u_long val; 161 int error; 162 163 val = __elfN(pie_base); 164 error = sysctl_handle_long(oidp, &val, 0, req); 165 if (error != 0 || req->newptr == NULL) 166 return (error); 167 if ((val & PAGE_MASK) != 0) 168 return (EINVAL); 169 __elfN(pie_base) = val; 170 return (0); 171 } 172 SYSCTL_PROC(ELF_NODE_OID, OID_AUTO, pie_base, 173 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 174 sysctl_pie_base, "LU", 175 "PIE load base without randomization"); 176 177 SYSCTL_NODE(ELF_NODE_OID, OID_AUTO, aslr, 178 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 179 ""); 180 #define ASLR_NODE_OID __CONCAT(ELF_NODE_OID, _aslr) 181 182 /* 183 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures 184 * have limited address space (which can cause issues for applications with 185 * high memory use) so we leave it off there. 186 */ 187 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64; 188 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN, 189 &__elfN(aslr_enabled), 0, 190 ELF_ABI_NAME ": enable address map randomization"); 191 192 /* 193 * Enable ASLR by default for 64-bit PIE binaries. 194 */ 195 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64; 196 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN, 197 &__elfN(pie_aslr_enabled), 0, 198 ELF_ABI_NAME ": enable address map randomization for PIE binaries"); 199 200 /* 201 * Sbrk is deprecated and it can be assumed that in most cases it will not be 202 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR 203 * to use the bss grow region. 204 */ 205 static int __elfN(aslr_honor_sbrk) = 0; 206 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW, 207 &__elfN(aslr_honor_sbrk), 0, 208 ELF_ABI_NAME ": assume sbrk is used"); 209 210 static int __elfN(aslr_stack) = __ELF_WORD_SIZE == 64; 211 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN, 212 &__elfN(aslr_stack), 0, 213 ELF_ABI_NAME ": enable stack address randomization"); 214 215 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64; 216 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN, 217 &__elfN(aslr_shared_page), 0, 218 ELF_ABI_NAME ": enable shared page address randomization"); 219 220 static int __elfN(sigfastblock) = 1; 221 SYSCTL_INT(ELF_NODE_OID, OID_AUTO, sigfastblock, 222 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0, 223 "enable sigfastblock for new processes"); 224 225 static bool __elfN(allow_wx) = true; 226 SYSCTL_BOOL(ELF_NODE_OID, OID_AUTO, allow_wx, 227 CTLFLAG_RWTUN, &__elfN(allow_wx), 0, 228 "Allow pages to be mapped simultaneously writable and executable"); 229 230 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 231 232 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a)) 233 234 Elf_Brandnote __elfN(freebsd_brandnote) = { 235 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), 236 .hdr.n_descsz = sizeof(int32_t), 237 .hdr.n_type = NT_FREEBSD_ABI_TAG, 238 .vendor = FREEBSD_ABI_VENDOR, 239 .flags = BN_TRANSLATE_OSREL, 240 .trans_osrel = __elfN(freebsd_trans_osrel) 241 }; 242 243 static bool 244 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) 245 { 246 uintptr_t p; 247 248 p = (uintptr_t)(note + 1); 249 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 250 *osrel = *(const int32_t *)(p); 251 252 return (true); 253 } 254 255 static int GNU_KFREEBSD_ABI_DESC = 3; 256 257 Elf_Brandnote __elfN(kfreebsd_brandnote) = { 258 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), 259 .hdr.n_descsz = 16, /* XXX at least 16 */ 260 .hdr.n_type = 1, 261 .vendor = GNU_ABI_VENDOR, 262 .flags = BN_TRANSLATE_OSREL, 263 .trans_osrel = kfreebsd_trans_osrel 264 }; 265 266 static bool 267 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) 268 { 269 const Elf32_Word *desc; 270 uintptr_t p; 271 272 p = (uintptr_t)(note + 1); 273 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 274 275 desc = (const Elf32_Word *)p; 276 if (desc[0] != GNU_KFREEBSD_ABI_DESC) 277 return (false); 278 279 /* 280 * Debian GNU/kFreeBSD embed the earliest compatible kernel version 281 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way. 282 */ 283 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; 284 285 return (true); 286 } 287 288 int 289 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 290 { 291 int i; 292 293 for (i = 0; i < MAX_BRANDS; i++) { 294 if (elf_brand_list[i] == NULL) { 295 elf_brand_list[i] = entry; 296 break; 297 } 298 } 299 if (i == MAX_BRANDS) { 300 printf("WARNING: %s: could not insert brandinfo entry: %p\n", 301 __func__, entry); 302 return (-1); 303 } 304 return (0); 305 } 306 307 int 308 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 309 { 310 int i; 311 312 for (i = 0; i < MAX_BRANDS; i++) { 313 if (elf_brand_list[i] == entry) { 314 elf_brand_list[i] = NULL; 315 break; 316 } 317 } 318 if (i == MAX_BRANDS) 319 return (-1); 320 return (0); 321 } 322 323 bool 324 __elfN(brand_inuse)(Elf_Brandinfo *entry) 325 { 326 struct proc *p; 327 bool rval = false; 328 329 sx_slock(&allproc_lock); 330 FOREACH_PROC_IN_SYSTEM(p) { 331 if (p->p_sysent == entry->sysvec) { 332 rval = true; 333 break; 334 } 335 } 336 sx_sunlock(&allproc_lock); 337 338 return (rval); 339 } 340 341 static Elf_Brandinfo * 342 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, 343 int32_t *osrel, uint32_t *fctl0) 344 { 345 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 346 Elf_Brandinfo *bi, *bi_m; 347 bool ret, has_fctl0; 348 int i, interp_name_len; 349 350 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0; 351 352 /* 353 * We support four types of branding -- (1) the ELF EI_OSABI field 354 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 355 * branding w/in the ELF header, (3) path of the `interp_path' 356 * field, and (4) the ".note.ABI-tag" ELF section. 357 */ 358 359 /* Look for an ".note.ABI-tag" ELF section */ 360 bi_m = NULL; 361 for (i = 0; i < MAX_BRANDS; i++) { 362 bi = elf_brand_list[i]; 363 if (bi == NULL) 364 continue; 365 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 366 continue; 367 if (hdr->e_machine == bi->machine && (bi->flags & 368 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 369 has_fctl0 = false; 370 *fctl0 = 0; 371 *osrel = 0; 372 ret = __elfN(check_note)(imgp, bi->brand_note, osrel, 373 &has_fctl0, fctl0); 374 /* Give brand a chance to veto check_note's guess */ 375 if (ret && bi->header_supported) { 376 ret = bi->header_supported(imgp, osrel, 377 has_fctl0 ? fctl0 : NULL); 378 } 379 /* 380 * If note checker claimed the binary, but the 381 * interpreter path in the image does not 382 * match default one for the brand, try to 383 * search for other brands with the same 384 * interpreter. Either there is better brand 385 * with the right interpreter, or, failing 386 * this, we return first brand which accepted 387 * our note and, optionally, header. 388 */ 389 if (ret && bi_m == NULL && interp != NULL && 390 (bi->interp_path == NULL || 391 (strlen(bi->interp_path) + 1 != interp_name_len || 392 strncmp(interp, bi->interp_path, interp_name_len) 393 != 0))) { 394 bi_m = bi; 395 ret = 0; 396 } 397 if (ret) 398 return (bi); 399 } 400 } 401 if (bi_m != NULL) 402 return (bi_m); 403 404 /* If the executable has a brand, search for it in the brand list. */ 405 for (i = 0; i < MAX_BRANDS; i++) { 406 bi = elf_brand_list[i]; 407 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 408 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 409 continue; 410 if (hdr->e_machine == bi->machine && 411 (hdr->e_ident[EI_OSABI] == bi->brand || 412 (bi->compat_3_brand != NULL && 413 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 414 bi->compat_3_brand) == 0))) { 415 /* Looks good, but give brand a chance to veto */ 416 if (bi->header_supported == NULL || 417 bi->header_supported(imgp, NULL, NULL)) { 418 /* 419 * Again, prefer strictly matching 420 * interpreter path. 421 */ 422 if (interp_name_len == 0 && 423 bi->interp_path == NULL) 424 return (bi); 425 if (bi->interp_path != NULL && 426 strlen(bi->interp_path) + 1 == 427 interp_name_len && strncmp(interp, 428 bi->interp_path, interp_name_len) == 0) 429 return (bi); 430 if (bi_m == NULL) 431 bi_m = bi; 432 } 433 } 434 } 435 if (bi_m != NULL) 436 return (bi_m); 437 438 /* No known brand, see if the header is recognized by any brand */ 439 for (i = 0; i < MAX_BRANDS; i++) { 440 bi = elf_brand_list[i]; 441 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 442 bi->header_supported == NULL) 443 continue; 444 if (hdr->e_machine == bi->machine) { 445 ret = bi->header_supported(imgp, NULL, NULL); 446 if (ret) 447 return (bi); 448 } 449 } 450 451 /* Lacking a known brand, search for a recognized interpreter. */ 452 if (interp != NULL) { 453 for (i = 0; i < MAX_BRANDS; i++) { 454 bi = elf_brand_list[i]; 455 if (bi == NULL || (bi->flags & 456 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 457 != 0) 458 continue; 459 if (hdr->e_machine == bi->machine && 460 bi->interp_path != NULL && 461 /* ELF image p_filesz includes terminating zero */ 462 strlen(bi->interp_path) + 1 == interp_name_len && 463 strncmp(interp, bi->interp_path, interp_name_len) 464 == 0 && (bi->header_supported == NULL || 465 bi->header_supported(imgp, NULL, NULL))) 466 return (bi); 467 } 468 } 469 470 /* Lacking a recognized interpreter, try the default brand */ 471 for (i = 0; i < MAX_BRANDS; i++) { 472 bi = elf_brand_list[i]; 473 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 474 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 475 continue; 476 if (hdr->e_machine == bi->machine && 477 __elfN(fallback_brand) == bi->brand && 478 (bi->header_supported == NULL || 479 bi->header_supported(imgp, NULL, NULL))) 480 return (bi); 481 } 482 return (NULL); 483 } 484 485 static bool 486 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr) 487 { 488 return (hdr->e_phoff <= PAGE_SIZE && 489 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff); 490 } 491 492 static int 493 __elfN(check_header)(const Elf_Ehdr *hdr) 494 { 495 Elf_Brandinfo *bi; 496 int i; 497 498 if (!IS_ELF(*hdr) || 499 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 500 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 501 hdr->e_ident[EI_VERSION] != EV_CURRENT || 502 hdr->e_phentsize != sizeof(Elf_Phdr) || 503 hdr->e_version != ELF_TARG_VER) 504 return (ENOEXEC); 505 506 /* 507 * Make sure we have at least one brand for this machine. 508 */ 509 510 for (i = 0; i < MAX_BRANDS; i++) { 511 bi = elf_brand_list[i]; 512 if (bi != NULL && bi->machine == hdr->e_machine) 513 break; 514 } 515 if (i == MAX_BRANDS) 516 return (ENOEXEC); 517 518 return (0); 519 } 520 521 static int 522 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 523 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 524 { 525 struct sf_buf *sf; 526 int error; 527 vm_offset_t off; 528 529 /* 530 * Create the page if it doesn't exist yet. Ignore errors. 531 */ 532 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 533 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 534 535 /* 536 * Find the page from the underlying object. 537 */ 538 if (object != NULL) { 539 sf = vm_imgact_map_page(object, offset); 540 if (sf == NULL) 541 return (KERN_FAILURE); 542 off = offset - trunc_page(offset); 543 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 544 end - start); 545 vm_imgact_unmap_page(sf); 546 if (error != 0) 547 return (KERN_FAILURE); 548 } 549 550 return (KERN_SUCCESS); 551 } 552 553 static int 554 __elfN(map_insert)(const struct image_params *imgp, vm_map_t map, 555 vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, 556 vm_prot_t prot, int cow) 557 { 558 struct sf_buf *sf; 559 vm_offset_t off; 560 vm_size_t sz; 561 int error, locked, rv; 562 563 if (start != trunc_page(start)) { 564 rv = __elfN(map_partial)(map, object, offset, start, 565 round_page(start), prot); 566 if (rv != KERN_SUCCESS) 567 return (rv); 568 offset += round_page(start) - start; 569 start = round_page(start); 570 } 571 if (end != round_page(end)) { 572 rv = __elfN(map_partial)(map, object, offset + 573 trunc_page(end) - start, trunc_page(end), end, prot); 574 if (rv != KERN_SUCCESS) 575 return (rv); 576 end = trunc_page(end); 577 } 578 if (start >= end) 579 return (KERN_SUCCESS); 580 if ((offset & PAGE_MASK) != 0) { 581 /* 582 * The mapping is not page aligned. This means that we have 583 * to copy the data. 584 */ 585 rv = vm_map_fixed(map, NULL, 0, start, end - start, 586 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 587 if (rv != KERN_SUCCESS) 588 return (rv); 589 if (object == NULL) 590 return (KERN_SUCCESS); 591 for (; start < end; start += sz) { 592 sf = vm_imgact_map_page(object, offset); 593 if (sf == NULL) 594 return (KERN_FAILURE); 595 off = offset - trunc_page(offset); 596 sz = end - start; 597 if (sz > PAGE_SIZE - off) 598 sz = PAGE_SIZE - off; 599 error = copyout((caddr_t)sf_buf_kva(sf) + off, 600 (caddr_t)start, sz); 601 vm_imgact_unmap_page(sf); 602 if (error != 0) 603 return (KERN_FAILURE); 604 offset += sz; 605 } 606 } else { 607 vm_object_reference(object); 608 rv = vm_map_fixed(map, object, offset, start, end - start, 609 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL | 610 (object != NULL ? MAP_VN_EXEC : 0)); 611 if (rv != KERN_SUCCESS) { 612 locked = VOP_ISLOCKED(imgp->vp); 613 VOP_UNLOCK(imgp->vp); 614 vm_object_deallocate(object); 615 vn_lock(imgp->vp, locked | LK_RETRY); 616 return (rv); 617 } else if (object != NULL) { 618 MPASS(imgp->vp->v_object == object); 619 VOP_SET_TEXT_CHECKED(imgp->vp); 620 } 621 } 622 return (KERN_SUCCESS); 623 } 624 625 static int 626 __elfN(load_section)(const struct image_params *imgp, vm_ooffset_t offset, 627 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot) 628 { 629 struct sf_buf *sf; 630 size_t map_len; 631 vm_map_t map; 632 vm_object_t object; 633 vm_offset_t map_addr; 634 int error, rv, cow; 635 size_t copy_len; 636 vm_ooffset_t file_addr; 637 638 /* 639 * It's necessary to fail if the filsz + offset taken from the 640 * header is greater than the actual file pager object's size. 641 * If we were to allow this, then the vm_map_find() below would 642 * walk right off the end of the file object and into the ether. 643 * 644 * While I'm here, might as well check for something else that 645 * is invalid: filsz cannot be greater than memsz. 646 */ 647 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 648 filsz > memsz) { 649 uprintf("elf_load_section: truncated ELF file\n"); 650 return (ENOEXEC); 651 } 652 653 object = imgp->object; 654 map = &imgp->proc->p_vmspace->vm_map; 655 map_addr = trunc_page((vm_offset_t)vmaddr); 656 file_addr = trunc_page(offset); 657 658 /* 659 * We have two choices. We can either clear the data in the last page 660 * of an oversized mapping, or we can start the anon mapping a page 661 * early and copy the initialized data into that first page. We 662 * choose the second. 663 */ 664 if (filsz == 0) 665 map_len = 0; 666 else if (memsz > filsz) 667 map_len = trunc_page(offset + filsz) - file_addr; 668 else 669 map_len = round_page(offset + filsz) - file_addr; 670 671 if (map_len != 0) { 672 /* cow flags: don't dump readonly sections in core */ 673 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 674 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 675 676 rv = __elfN(map_insert)(imgp, map, object, file_addr, 677 map_addr, map_addr + map_len, prot, cow); 678 if (rv != KERN_SUCCESS) 679 return (EINVAL); 680 681 /* we can stop now if we've covered it all */ 682 if (memsz == filsz) 683 return (0); 684 } 685 686 /* 687 * We have to get the remaining bit of the file into the first part 688 * of the oversized map segment. This is normally because the .data 689 * segment in the file is extended to provide bss. It's a neat idea 690 * to try and save a page, but it's a pain in the behind to implement. 691 */ 692 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset + 693 filsz); 694 map_addr = trunc_page((vm_offset_t)vmaddr + filsz); 695 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr; 696 697 /* This had damn well better be true! */ 698 if (map_len != 0) { 699 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 700 map_addr + map_len, prot, 0); 701 if (rv != KERN_SUCCESS) 702 return (EINVAL); 703 } 704 705 if (copy_len != 0) { 706 sf = vm_imgact_map_page(object, offset + filsz); 707 if (sf == NULL) 708 return (EIO); 709 710 /* send the page fragment to user space */ 711 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr, 712 copy_len); 713 vm_imgact_unmap_page(sf); 714 if (error != 0) 715 return (error); 716 } 717 718 /* 719 * Remove write access to the page if it was only granted by map_insert 720 * to allow copyout. 721 */ 722 if ((prot & VM_PROT_WRITE) == 0) 723 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 724 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT); 725 726 return (0); 727 } 728 729 static int 730 __elfN(load_sections)(const struct image_params *imgp, const Elf_Ehdr *hdr, 731 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp) 732 { 733 vm_prot_t prot; 734 u_long base_addr; 735 bool first; 736 int error, i; 737 738 ASSERT_VOP_LOCKED(imgp->vp, __func__); 739 740 base_addr = 0; 741 first = true; 742 743 for (i = 0; i < hdr->e_phnum; i++) { 744 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 745 continue; 746 747 /* Loadable segment */ 748 prot = __elfN(trans_prot)(phdr[i].p_flags); 749 error = __elfN(load_section)(imgp, phdr[i].p_offset, 750 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 751 phdr[i].p_memsz, phdr[i].p_filesz, prot); 752 if (error != 0) 753 return (error); 754 755 /* 756 * Establish the base address if this is the first segment. 757 */ 758 if (first) { 759 base_addr = trunc_page(phdr[i].p_vaddr + rbase); 760 first = false; 761 } 762 } 763 764 if (base_addrp != NULL) 765 *base_addrp = base_addr; 766 767 return (0); 768 } 769 770 /* 771 * Load the file "file" into memory. It may be either a shared object 772 * or an executable. 773 * 774 * The "addr" reference parameter is in/out. On entry, it specifies 775 * the address where a shared object should be loaded. If the file is 776 * an executable, this value is ignored. On exit, "addr" specifies 777 * where the file was actually loaded. 778 * 779 * The "entry" reference parameter is out only. On exit, it specifies 780 * the entry point for the loaded file. 781 */ 782 static int 783 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 784 u_long *entry) 785 { 786 struct { 787 struct nameidata nd; 788 struct vattr attr; 789 struct image_params image_params; 790 } *tempdata; 791 const Elf_Ehdr *hdr = NULL; 792 const Elf_Phdr *phdr = NULL; 793 struct nameidata *nd; 794 struct vattr *attr; 795 struct image_params *imgp; 796 u_long rbase; 797 u_long base_addr = 0; 798 int error; 799 800 #ifdef CAPABILITY_MODE 801 /* 802 * XXXJA: This check can go away once we are sufficiently confident 803 * that the checks in namei() are correct. 804 */ 805 if (IN_CAPABILITY_MODE(curthread)) 806 return (ECAPMODE); 807 #endif 808 809 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO); 810 nd = &tempdata->nd; 811 attr = &tempdata->attr; 812 imgp = &tempdata->image_params; 813 814 /* 815 * Initialize part of the common data 816 */ 817 imgp->proc = p; 818 imgp->attr = attr; 819 820 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF, 821 UIO_SYSSPACE, file); 822 if ((error = namei(nd)) != 0) { 823 nd->ni_vp = NULL; 824 goto fail; 825 } 826 NDFREE_PNBUF(nd); 827 imgp->vp = nd->ni_vp; 828 829 /* 830 * Check permissions, modes, uid, etc on the file, and "open" it. 831 */ 832 error = exec_check_permissions(imgp); 833 if (error) 834 goto fail; 835 836 error = exec_map_first_page(imgp); 837 if (error) 838 goto fail; 839 840 imgp->object = nd->ni_vp->v_object; 841 842 hdr = (const Elf_Ehdr *)imgp->image_header; 843 if ((error = __elfN(check_header)(hdr)) != 0) 844 goto fail; 845 if (hdr->e_type == ET_DYN) 846 rbase = *addr; 847 else if (hdr->e_type == ET_EXEC) 848 rbase = 0; 849 else { 850 error = ENOEXEC; 851 goto fail; 852 } 853 854 /* Only support headers that fit within first page for now */ 855 if (!__elfN(phdr_in_zero_page)(hdr)) { 856 error = ENOEXEC; 857 goto fail; 858 } 859 860 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 861 if (!aligned(phdr, Elf_Addr)) { 862 error = ENOEXEC; 863 goto fail; 864 } 865 866 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr); 867 if (error != 0) 868 goto fail; 869 870 if (p->p_sysent->sv_protect != NULL) 871 p->p_sysent->sv_protect(imgp, SVP_INTERP); 872 873 *addr = base_addr; 874 *entry = (unsigned long)hdr->e_entry + rbase; 875 876 fail: 877 if (imgp->firstpage) 878 exec_unmap_first_page(imgp); 879 880 if (nd->ni_vp) { 881 if (imgp->textset) 882 VOP_UNSET_TEXT_CHECKED(nd->ni_vp); 883 vput(nd->ni_vp); 884 } 885 free(tempdata, M_TEMP); 886 887 return (error); 888 } 889 890 /* 891 * Select randomized valid address in the map map, between minv and 892 * maxv, with specified alignment. The [minv, maxv) range must belong 893 * to the map. Note that function only allocates the address, it is 894 * up to caller to clamp maxv in a way that the final allocation 895 * length fit into the map. 896 * 897 * Result is returned in *resp, error code indicates that arguments 898 * did not pass sanity checks for overflow and range correctness. 899 */ 900 static int 901 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv, 902 u_int align, u_long *resp) 903 { 904 u_long rbase, res; 905 906 MPASS(vm_map_min(map) <= minv); 907 908 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) { 909 uprintf("Invalid ELF segments layout\n"); 910 return (ENOEXEC); 911 } 912 913 arc4rand(&rbase, sizeof(rbase), 0); 914 res = roundup(minv, (u_long)align) + rbase % (maxv - minv); 915 res &= ~((u_long)align - 1); 916 if (res >= maxv) 917 res -= align; 918 919 KASSERT(res >= minv, 920 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx", 921 res, minv, maxv, rbase)); 922 KASSERT(res < maxv, 923 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx", 924 res, maxv, minv, rbase)); 925 926 *resp = res; 927 return (0); 928 } 929 930 static int 931 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr, 932 const Elf_Phdr *phdr) 933 { 934 struct vmspace *vmspace; 935 const char *err_str; 936 u_long text_size, data_size, total_size, text_addr, data_addr; 937 u_long seg_size, seg_addr; 938 int i; 939 940 err_str = NULL; 941 text_size = data_size = total_size = text_addr = data_addr = 0; 942 943 for (i = 0; i < hdr->e_phnum; i++) { 944 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0) 945 continue; 946 947 seg_addr = trunc_page(phdr[i].p_vaddr + imgp->et_dyn_addr); 948 seg_size = round_page(phdr[i].p_memsz + 949 phdr[i].p_vaddr + imgp->et_dyn_addr - seg_addr); 950 951 /* 952 * Make the largest executable segment the official 953 * text segment and all others data. 954 * 955 * Note that obreak() assumes that data_addr + data_size == end 956 * of data load area, and the ELF file format expects segments 957 * to be sorted by address. If multiple data segments exist, 958 * the last one will be used. 959 */ 960 961 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) { 962 text_size = seg_size; 963 text_addr = seg_addr; 964 } else { 965 data_size = seg_size; 966 data_addr = seg_addr; 967 } 968 total_size += seg_size; 969 } 970 971 if (data_addr == 0 && data_size == 0) { 972 data_addr = text_addr; 973 data_size = text_size; 974 } 975 976 /* 977 * Check limits. It should be safe to check the 978 * limits after loading the segments since we do 979 * not actually fault in all the segments pages. 980 */ 981 PROC_LOCK(imgp->proc); 982 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 983 err_str = "Data segment size exceeds process limit"; 984 else if (text_size > maxtsiz) 985 err_str = "Text segment size exceeds system limit"; 986 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 987 err_str = "Total segment size exceeds process limit"; 988 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 989 err_str = "Data segment size exceeds resource limit"; 990 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 991 err_str = "Total segment size exceeds resource limit"; 992 PROC_UNLOCK(imgp->proc); 993 if (err_str != NULL) { 994 uprintf("%s\n", err_str); 995 return (ENOMEM); 996 } 997 998 vmspace = imgp->proc->p_vmspace; 999 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 1000 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 1001 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 1002 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 1003 1004 return (0); 1005 } 1006 1007 static int 1008 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr, 1009 char **interpp, bool *free_interpp) 1010 { 1011 struct thread *td; 1012 char *interp; 1013 int error, interp_name_len; 1014 1015 KASSERT(phdr->p_type == PT_INTERP, 1016 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type)); 1017 ASSERT_VOP_LOCKED(imgp->vp, __func__); 1018 1019 td = curthread; 1020 1021 /* Path to interpreter */ 1022 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) { 1023 uprintf("Invalid PT_INTERP\n"); 1024 return (ENOEXEC); 1025 } 1026 1027 interp_name_len = phdr->p_filesz; 1028 if (phdr->p_offset > PAGE_SIZE || 1029 interp_name_len > PAGE_SIZE - phdr->p_offset) { 1030 /* 1031 * The vnode lock might be needed by the pagedaemon to 1032 * clean pages owned by the vnode. Do not allow sleep 1033 * waiting for memory with the vnode locked, instead 1034 * try non-sleepable allocation first, and if it 1035 * fails, go to the slow path were we drop the lock 1036 * and do M_WAITOK. A text reference prevents 1037 * modifications to the vnode content. 1038 */ 1039 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT); 1040 if (interp == NULL) { 1041 VOP_UNLOCK(imgp->vp); 1042 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK); 1043 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1044 } 1045 1046 error = vn_rdwr(UIO_READ, imgp->vp, interp, 1047 interp_name_len, phdr->p_offset, 1048 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 1049 NOCRED, NULL, td); 1050 if (error != 0) { 1051 free(interp, M_TEMP); 1052 uprintf("i/o error PT_INTERP %d\n", error); 1053 return (error); 1054 } 1055 interp[interp_name_len] = '\0'; 1056 1057 *interpp = interp; 1058 *free_interpp = true; 1059 return (0); 1060 } 1061 1062 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset; 1063 if (interp[interp_name_len - 1] != '\0') { 1064 uprintf("Invalid PT_INTERP\n"); 1065 return (ENOEXEC); 1066 } 1067 1068 *interpp = interp; 1069 *free_interpp = false; 1070 return (0); 1071 } 1072 1073 static int 1074 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info, 1075 const char *interp, u_long *addr, u_long *entry) 1076 { 1077 int error; 1078 1079 if (brand_info->interp_newpath != NULL && 1080 (brand_info->interp_path == NULL || 1081 strcmp(interp, brand_info->interp_path) == 0)) { 1082 error = __elfN(load_file)(imgp->proc, 1083 brand_info->interp_newpath, addr, entry); 1084 if (error == 0) 1085 return (0); 1086 } 1087 1088 error = __elfN(load_file)(imgp->proc, interp, addr, entry); 1089 if (error == 0) 1090 return (0); 1091 1092 uprintf("ELF interpreter %s not found, error %d\n", interp, error); 1093 return (error); 1094 } 1095 1096 /* 1097 * Impossible et_dyn_addr initial value indicating that the real base 1098 * must be calculated later with some randomization applied. 1099 */ 1100 #define ET_DYN_ADDR_RAND 1 1101 1102 static int 1103 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 1104 { 1105 struct thread *td; 1106 const Elf_Ehdr *hdr; 1107 const Elf_Phdr *phdr; 1108 Elf_Auxargs *elf_auxargs; 1109 struct vmspace *vmspace; 1110 vm_map_t map; 1111 char *interp; 1112 Elf_Brandinfo *brand_info; 1113 struct sysentvec *sv; 1114 u_long addr, baddr, entry, proghdr; 1115 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc; 1116 uint32_t fctl0; 1117 int32_t osrel; 1118 bool free_interp; 1119 int error, i, n; 1120 1121 hdr = (const Elf_Ehdr *)imgp->image_header; 1122 1123 /* 1124 * Do we have a valid ELF header ? 1125 * 1126 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 1127 * if particular brand doesn't support it. 1128 */ 1129 if (__elfN(check_header)(hdr) != 0 || 1130 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 1131 return (-1); 1132 1133 /* 1134 * From here on down, we return an errno, not -1, as we've 1135 * detected an ELF file. 1136 */ 1137 1138 if (!__elfN(phdr_in_zero_page)(hdr)) { 1139 uprintf("Program headers not in the first page\n"); 1140 return (ENOEXEC); 1141 } 1142 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 1143 if (!aligned(phdr, Elf_Addr)) { 1144 uprintf("Unaligned program headers\n"); 1145 return (ENOEXEC); 1146 } 1147 1148 n = error = 0; 1149 baddr = 0; 1150 osrel = 0; 1151 fctl0 = 0; 1152 entry = proghdr = 0; 1153 interp = NULL; 1154 free_interp = false; 1155 td = curthread; 1156 1157 /* 1158 * Somewhat arbitrary, limit accepted max alignment for the 1159 * loadable segment to the max supported superpage size. Too 1160 * large alignment requests are not useful and are indicators 1161 * of corrupted or outright malicious binary. 1162 */ 1163 maxalign = PAGE_SIZE; 1164 maxsalign = PAGE_SIZE * 1024; 1165 for (i = MAXPAGESIZES - 1; i > 0; i--) { 1166 if (pagesizes[i] > maxsalign) { 1167 maxsalign = pagesizes[i]; 1168 break; 1169 } 1170 } 1171 1172 mapsz = 0; 1173 1174 for (i = 0; i < hdr->e_phnum; i++) { 1175 switch (phdr[i].p_type) { 1176 case PT_LOAD: 1177 if (n == 0) 1178 baddr = phdr[i].p_vaddr; 1179 if (!powerof2(phdr[i].p_align) || 1180 phdr[i].p_align > maxsalign) { 1181 uprintf("Invalid segment alignment\n"); 1182 error = ENOEXEC; 1183 goto ret; 1184 } 1185 if (phdr[i].p_align > maxalign) 1186 maxalign = phdr[i].p_align; 1187 if (mapsz + phdr[i].p_memsz < mapsz) { 1188 uprintf("Mapsize overflow\n"); 1189 error = ENOEXEC; 1190 goto ret; 1191 } 1192 mapsz += phdr[i].p_memsz; 1193 n++; 1194 1195 /* 1196 * If this segment contains the program headers, 1197 * remember their virtual address for the AT_PHDR 1198 * aux entry. Static binaries don't usually include 1199 * a PT_PHDR entry. 1200 */ 1201 if (phdr[i].p_offset == 0 && 1202 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= 1203 phdr[i].p_filesz) 1204 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 1205 break; 1206 case PT_INTERP: 1207 /* Path to interpreter */ 1208 if (interp != NULL) { 1209 uprintf("Multiple PT_INTERP headers\n"); 1210 error = ENOEXEC; 1211 goto ret; 1212 } 1213 error = __elfN(get_interp)(imgp, &phdr[i], &interp, 1214 &free_interp); 1215 if (error != 0) 1216 goto ret; 1217 break; 1218 case PT_GNU_STACK: 1219 if (__elfN(nxstack)) { 1220 imgp->stack_prot = 1221 __elfN(trans_prot)(phdr[i].p_flags); 1222 if ((imgp->stack_prot & VM_PROT_RW) != 1223 VM_PROT_RW) { 1224 uprintf("Invalid PT_GNU_STACK\n"); 1225 error = ENOEXEC; 1226 goto ret; 1227 } 1228 } 1229 imgp->stack_sz = phdr[i].p_memsz; 1230 break; 1231 case PT_PHDR: /* Program header table info */ 1232 proghdr = phdr[i].p_vaddr; 1233 break; 1234 } 1235 } 1236 1237 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0); 1238 if (brand_info == NULL) { 1239 uprintf("ELF binary type \"%u\" not known.\n", 1240 hdr->e_ident[EI_OSABI]); 1241 error = ENOEXEC; 1242 goto ret; 1243 } 1244 sv = brand_info->sysvec; 1245 if (hdr->e_type == ET_DYN) { 1246 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 1247 uprintf("Cannot execute shared object\n"); 1248 error = ENOEXEC; 1249 goto ret; 1250 } 1251 /* 1252 * Honour the base load address from the dso if it is 1253 * non-zero for some reason. 1254 */ 1255 if (baddr == 0) { 1256 if ((sv->sv_flags & SV_ASLR) == 0 || 1257 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) 1258 imgp->et_dyn_addr = __elfN(pie_base); 1259 else if ((__elfN(pie_aslr_enabled) && 1260 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) || 1261 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0) 1262 imgp->et_dyn_addr = ET_DYN_ADDR_RAND; 1263 else 1264 imgp->et_dyn_addr = __elfN(pie_base); 1265 } 1266 } 1267 1268 /* 1269 * Avoid a possible deadlock if the current address space is destroyed 1270 * and that address space maps the locked vnode. In the common case, 1271 * the locked vnode's v_usecount is decremented but remains greater 1272 * than zero. Consequently, the vnode lock is not needed by vrele(). 1273 * However, in cases where the vnode lock is external, such as nullfs, 1274 * v_usecount may become zero. 1275 * 1276 * The VV_TEXT flag prevents modifications to the executable while 1277 * the vnode is unlocked. 1278 */ 1279 VOP_UNLOCK(imgp->vp); 1280 1281 /* 1282 * Decide whether to enable randomization of user mappings. 1283 * First, reset user preferences for the setid binaries. 1284 * Then, account for the support of the randomization by the 1285 * ABI, by user preferences, and make special treatment for 1286 * PIE binaries. 1287 */ 1288 if (imgp->credential_setid) { 1289 PROC_LOCK(imgp->proc); 1290 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE | 1291 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC); 1292 PROC_UNLOCK(imgp->proc); 1293 } 1294 if ((sv->sv_flags & SV_ASLR) == 0 || 1295 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 || 1296 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) { 1297 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND, 1298 ("imgp->et_dyn_addr == RAND and !ASLR")); 1299 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 || 1300 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) || 1301 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1302 imgp->map_flags |= MAP_ASLR; 1303 /* 1304 * If user does not care about sbrk, utilize the bss 1305 * grow region for mappings as well. We can select 1306 * the base for the image anywere and still not suffer 1307 * from the fragmentation. 1308 */ 1309 if (!__elfN(aslr_honor_sbrk) || 1310 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0) 1311 imgp->map_flags |= MAP_ASLR_IGNSTART; 1312 if (__elfN(aslr_stack)) 1313 imgp->map_flags |= MAP_ASLR_STACK; 1314 if (__elfN(aslr_shared_page)) 1315 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE; 1316 } 1317 1318 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 && 1319 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) || 1320 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0) 1321 imgp->map_flags |= MAP_WXORX; 1322 1323 error = exec_new_vmspace(imgp, sv); 1324 1325 imgp->proc->p_sysent = sv; 1326 imgp->proc->p_elf_brandinfo = brand_info; 1327 1328 vmspace = imgp->proc->p_vmspace; 1329 map = &vmspace->vm_map; 1330 maxv = sv->sv_usrstack; 1331 if ((imgp->map_flags & MAP_ASLR_STACK) == 0) 1332 maxv -= lim_max(td, RLIMIT_STACK); 1333 if (error == 0 && mapsz >= maxv - vm_map_min(map)) { 1334 uprintf("Excessive mapping size\n"); 1335 error = ENOEXEC; 1336 } 1337 1338 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) { 1339 KASSERT((map->flags & MAP_ASLR) != 0, 1340 ("ET_DYN_ADDR_RAND but !MAP_ASLR")); 1341 error = __CONCAT(rnd_, __elfN(base))(map, 1342 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA), 1343 /* reserve half of the address space to interpreter */ 1344 maxv / 2, maxalign, &imgp->et_dyn_addr); 1345 } 1346 1347 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1348 if (error != 0) 1349 goto ret; 1350 1351 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL); 1352 if (error != 0) 1353 goto ret; 1354 1355 error = __elfN(enforce_limits)(imgp, hdr, phdr); 1356 if (error != 0) 1357 goto ret; 1358 1359 /* 1360 * We load the dynamic linker where a userland call 1361 * to mmap(0, ...) would put it. The rationale behind this 1362 * calculation is that it leaves room for the heap to grow to 1363 * its maximum allowed size. 1364 */ 1365 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1366 RLIMIT_DATA)); 1367 if ((map->flags & MAP_ASLR) != 0) { 1368 maxv1 = maxv / 2 + addr / 2; 1369 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1, 1370 #if VM_NRESERVLEVEL > 0 1371 pagesizes[VM_NRESERVLEVEL] != 0 ? 1372 /* Align anon_loc to the largest superpage size. */ 1373 pagesizes[VM_NRESERVLEVEL] : 1374 #endif 1375 pagesizes[0], &anon_loc); 1376 if (error != 0) 1377 goto ret; 1378 map->anon_loc = anon_loc; 1379 } else { 1380 map->anon_loc = addr; 1381 } 1382 1383 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr; 1384 imgp->entry_addr = entry; 1385 1386 if (sv->sv_protect != NULL) 1387 sv->sv_protect(imgp, SVP_IMAGE); 1388 1389 if (interp != NULL) { 1390 VOP_UNLOCK(imgp->vp); 1391 if ((map->flags & MAP_ASLR) != 0) { 1392 /* Assume that interpreter fits into 1/4 of AS */ 1393 maxv1 = maxv / 2 + addr / 2; 1394 error = __CONCAT(rnd_, __elfN(base))(map, addr, 1395 maxv1, PAGE_SIZE, &addr); 1396 } 1397 if (error == 0) { 1398 error = __elfN(load_interp)(imgp, brand_info, interp, 1399 &addr, &imgp->entry_addr); 1400 } 1401 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1402 if (error != 0) 1403 goto ret; 1404 } else 1405 addr = imgp->et_dyn_addr; 1406 1407 error = exec_map_stack(imgp); 1408 if (error != 0) 1409 goto ret; 1410 1411 /* 1412 * Construct auxargs table (used by the copyout_auxargs routine) 1413 */ 1414 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT); 1415 if (elf_auxargs == NULL) { 1416 VOP_UNLOCK(imgp->vp); 1417 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1418 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 1419 } 1420 elf_auxargs->execfd = -1; 1421 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr; 1422 elf_auxargs->phent = hdr->e_phentsize; 1423 elf_auxargs->phnum = hdr->e_phnum; 1424 elf_auxargs->pagesz = PAGE_SIZE; 1425 elf_auxargs->base = addr; 1426 elf_auxargs->flags = 0; 1427 elf_auxargs->entry = entry; 1428 elf_auxargs->hdr_eflags = hdr->e_flags; 1429 1430 imgp->auxargs = elf_auxargs; 1431 imgp->interpreted = 0; 1432 imgp->reloc_base = addr; 1433 imgp->proc->p_osrel = osrel; 1434 imgp->proc->p_fctl0 = fctl0; 1435 imgp->proc->p_elf_flags = hdr->e_flags; 1436 1437 ret: 1438 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock"); 1439 if (free_interp) 1440 free(interp, M_TEMP); 1441 return (error); 1442 } 1443 1444 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE) 1445 1446 int 1447 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base) 1448 { 1449 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1450 Elf_Auxinfo *argarray, *pos; 1451 struct vmspace *vmspace; 1452 rlim_t stacksz; 1453 int error, oc; 1454 uint32_t bsdflags; 1455 1456 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP, 1457 M_WAITOK | M_ZERO); 1458 1459 vmspace = imgp->proc->p_vmspace; 1460 1461 if (args->execfd != -1) 1462 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1463 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1464 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1465 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1466 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1467 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1468 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1469 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1470 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1471 if (imgp->execpathp != 0) 1472 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp); 1473 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1474 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1475 if (imgp->canary != 0) { 1476 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary); 1477 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1478 } 1479 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1480 if (imgp->pagesizes != 0) { 1481 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes); 1482 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1483 } 1484 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) { 1485 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1486 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset); 1487 } 1488 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1489 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1490 imgp->sysent->sv_stackprot); 1491 if (imgp->sysent->sv_hwcap != NULL) 1492 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap); 1493 if (imgp->sysent->sv_hwcap2 != NULL) 1494 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2); 1495 if (imgp->sysent->sv_hwcap3 != NULL) 1496 AUXARGS_ENTRY(pos, AT_HWCAP3, *imgp->sysent->sv_hwcap3); 1497 if (imgp->sysent->sv_hwcap4 != NULL) 1498 AUXARGS_ENTRY(pos, AT_HWCAP4, *imgp->sysent->sv_hwcap4); 1499 bsdflags = 0; 1500 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0; 1501 oc = atomic_load_int(&vm_overcommit); 1502 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) != 1503 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0; 1504 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags); 1505 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc); 1506 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv); 1507 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc); 1508 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv); 1509 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings); 1510 #ifdef RANDOM_FENESTRASX 1511 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) { 1512 AUXARGS_ENTRY(pos, AT_FXRNG, 1513 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset); 1514 } 1515 #endif 1516 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) { 1517 AUXARGS_ENTRY(pos, AT_KPRELOAD, 1518 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset); 1519 } 1520 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop)); 1521 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur; 1522 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz); 1523 AUXARGS_ENTRY(pos, AT_NULL, 0); 1524 1525 free(imgp->auxargs, M_TEMP); 1526 imgp->auxargs = NULL; 1527 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs")); 1528 1529 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT); 1530 free(argarray, M_TEMP); 1531 return (error); 1532 } 1533 1534 int 1535 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp) 1536 { 1537 Elf_Addr *base; 1538 1539 base = (Elf_Addr *)*stack_base; 1540 base--; 1541 if (elf_suword(base, imgp->args->argc) == -1) 1542 return (EFAULT); 1543 *stack_base = (uintptr_t)base; 1544 return (0); 1545 } 1546 1547 /* 1548 * Code for generating ELF core dumps. 1549 */ 1550 1551 typedef void (*segment_callback)(vm_map_entry_t, void *); 1552 1553 /* Closure for cb_put_phdr(). */ 1554 struct phdr_closure { 1555 Elf_Phdr *phdr; /* Program header to fill in */ 1556 Elf_Off offset; /* Offset of segment in core file */ 1557 }; 1558 1559 struct note_info { 1560 int type; /* Note type. */ 1561 struct regset *regset; /* Register set. */ 1562 outfunc_t outfunc; /* Output function. */ 1563 void *outarg; /* Argument for the output function. */ 1564 size_t outsize; /* Output size. */ 1565 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1566 }; 1567 1568 TAILQ_HEAD(note_info_list, note_info); 1569 1570 static void cb_put_phdr(vm_map_entry_t, void *); 1571 static void cb_size_segment(vm_map_entry_t, void *); 1572 static void each_dumpable_segment(struct thread *, segment_callback, void *, 1573 int); 1574 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1575 struct note_info_list *, size_t, int); 1576 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *); 1577 1578 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1579 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1580 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1581 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1582 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1583 static void __elfN(note_procstat_kqueues)(void *, struct sbuf *, size_t *); 1584 static void note_procstat_files(void *, struct sbuf *, size_t *); 1585 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1586 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1587 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1588 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1589 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1590 1591 static int 1592 core_compressed_write(void *base, size_t len, off_t offset, void *arg) 1593 { 1594 1595 return (core_write((struct coredump_params *)arg, base, len, offset, 1596 UIO_SYSSPACE, NULL)); 1597 } 1598 1599 int 1600 __elfN(coredump)(struct thread *td, struct coredump_writer *cdw, off_t limit, int flags) 1601 { 1602 struct ucred *cred = td->td_ucred; 1603 int compm, error = 0; 1604 struct sseg_closure seginfo; 1605 struct note_info_list notelst; 1606 struct coredump_params params; 1607 struct note_info *ninfo; 1608 void *hdr, *tmpbuf; 1609 size_t hdrsize, notesz, coresize; 1610 1611 hdr = NULL; 1612 tmpbuf = NULL; 1613 TAILQ_INIT(¬elst); 1614 1615 /* Size the program segments. */ 1616 __elfN(size_segments)(td, &seginfo, flags); 1617 1618 /* 1619 * Collect info about the core file header area. 1620 */ 1621 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1622 if (seginfo.count + 1 >= PN_XNUM) 1623 hdrsize += sizeof(Elf_Shdr); 1624 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz); 1625 coresize = round_page(hdrsize + notesz) + seginfo.size; 1626 1627 /* Set up core dump parameters. */ 1628 params.offset = 0; 1629 params.active_cred = cred; 1630 params.td = td; 1631 params.cdw = cdw; 1632 params.comp = NULL; 1633 1634 #ifdef RACCT 1635 if (racct_enable) { 1636 PROC_LOCK(td->td_proc); 1637 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1638 PROC_UNLOCK(td->td_proc); 1639 if (error != 0) { 1640 error = EFAULT; 1641 goto done; 1642 } 1643 } 1644 #endif 1645 if (coresize >= limit) { 1646 error = EFAULT; 1647 goto done; 1648 } 1649 1650 /* Create a compression stream if necessary. */ 1651 compm = compress_user_cores; 1652 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP && 1653 compm == 0) 1654 compm = COMPRESS_GZIP; 1655 if (compm != 0) { 1656 params.comp = compressor_init(core_compressed_write, 1657 compm, CORE_BUF_SIZE, 1658 compress_user_cores_level, ¶ms); 1659 if (params.comp == NULL) { 1660 error = EFAULT; 1661 goto done; 1662 } 1663 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1664 } 1665 1666 if (cdw->init_fn != NULL) { 1667 error = (*cdw->init_fn)(cdw, ¶ms); 1668 if (error != 0) 1669 goto done; 1670 } 1671 1672 /* 1673 * Allocate memory for building the header, fill it up, 1674 * and write it out following the notes. 1675 */ 1676 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1677 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1678 notesz, flags); 1679 1680 /* Write the contents of all of the writable segments. */ 1681 if (error == 0) { 1682 Elf_Phdr *php; 1683 off_t offset; 1684 int i; 1685 1686 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1687 offset = round_page(hdrsize + notesz); 1688 for (i = 0; i < seginfo.count; i++) { 1689 error = core_output((char *)(uintptr_t)php->p_vaddr, 1690 php->p_filesz, offset, ¶ms, tmpbuf); 1691 if (error != 0) 1692 break; 1693 offset += php->p_filesz; 1694 php++; 1695 } 1696 if (error == 0 && params.comp != NULL) 1697 error = compressor_flush(params.comp); 1698 } 1699 if (error) { 1700 log(LOG_WARNING, 1701 "Failed to write core file for process %s (error %d)\n", 1702 curproc->p_comm, error); 1703 } 1704 1705 done: 1706 free(tmpbuf, M_TEMP); 1707 if (params.comp != NULL) 1708 compressor_fini(params.comp); 1709 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1710 TAILQ_REMOVE(¬elst, ninfo, link); 1711 free(ninfo, M_TEMP); 1712 } 1713 if (hdr != NULL) 1714 free(hdr, M_TEMP); 1715 1716 return (error); 1717 } 1718 1719 /* 1720 * A callback for each_dumpable_segment() to write out the segment's 1721 * program header entry. 1722 */ 1723 static void 1724 cb_put_phdr(vm_map_entry_t entry, void *closure) 1725 { 1726 struct phdr_closure *phc = (struct phdr_closure *)closure; 1727 Elf_Phdr *phdr = phc->phdr; 1728 1729 phc->offset = round_page(phc->offset); 1730 1731 phdr->p_type = PT_LOAD; 1732 phdr->p_offset = phc->offset; 1733 phdr->p_vaddr = entry->start; 1734 phdr->p_paddr = 0; 1735 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1736 phdr->p_align = PAGE_SIZE; 1737 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1738 1739 phc->offset += phdr->p_filesz; 1740 phc->phdr++; 1741 } 1742 1743 /* 1744 * A callback for each_dumpable_segment() to gather information about 1745 * the number of segments and their total size. 1746 */ 1747 static void 1748 cb_size_segment(vm_map_entry_t entry, void *closure) 1749 { 1750 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1751 1752 ssc->count++; 1753 ssc->size += entry->end - entry->start; 1754 } 1755 1756 void 1757 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo, 1758 int flags) 1759 { 1760 seginfo->count = 0; 1761 seginfo->size = 0; 1762 1763 each_dumpable_segment(td, cb_size_segment, seginfo, flags); 1764 } 1765 1766 /* 1767 * For each writable segment in the process's memory map, call the given 1768 * function with a pointer to the map entry and some arbitrary 1769 * caller-supplied data. 1770 */ 1771 static void 1772 each_dumpable_segment(struct thread *td, segment_callback func, void *closure, 1773 int flags) 1774 { 1775 struct proc *p = td->td_proc; 1776 vm_map_t map = &p->p_vmspace->vm_map; 1777 vm_map_entry_t entry; 1778 vm_object_t backing_object, object; 1779 bool ignore_entry; 1780 1781 vm_map_lock_read(map); 1782 VM_MAP_ENTRY_FOREACH(entry, map) { 1783 /* 1784 * Don't dump inaccessible mappings, deal with legacy 1785 * coredump mode. 1786 * 1787 * Note that read-only segments related to the elf binary 1788 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1789 * need to arbitrarily ignore such segments. 1790 */ 1791 if ((flags & SVC_ALL) == 0) { 1792 if (elf_legacy_coredump) { 1793 if ((entry->protection & VM_PROT_RW) != 1794 VM_PROT_RW) 1795 continue; 1796 } else { 1797 if ((entry->protection & VM_PROT_ALL) == 0) 1798 continue; 1799 } 1800 } 1801 1802 /* 1803 * Dont include memory segment in the coredump if 1804 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1805 * madvise(2). Do not dump submaps (i.e. parts of the 1806 * kernel map). 1807 */ 1808 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) 1809 continue; 1810 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 && 1811 (flags & SVC_ALL) == 0) 1812 continue; 1813 if ((object = entry->object.vm_object) == NULL) 1814 continue; 1815 1816 /* Ignore memory-mapped devices and such things. */ 1817 VM_OBJECT_RLOCK(object); 1818 while ((backing_object = object->backing_object) != NULL) { 1819 VM_OBJECT_RLOCK(backing_object); 1820 VM_OBJECT_RUNLOCK(object); 1821 object = backing_object; 1822 } 1823 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0; 1824 VM_OBJECT_RUNLOCK(object); 1825 if (ignore_entry) 1826 continue; 1827 1828 (*func)(entry, closure); 1829 } 1830 vm_map_unlock_read(map); 1831 } 1832 1833 /* 1834 * Write the core file header to the file, including padding up to 1835 * the page boundary. 1836 */ 1837 static int 1838 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1839 size_t hdrsize, struct note_info_list *notelst, size_t notesz, 1840 int flags) 1841 { 1842 struct note_info *ninfo; 1843 struct sbuf *sb; 1844 int error; 1845 1846 /* Fill in the header. */ 1847 bzero(hdr, hdrsize); 1848 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags); 1849 1850 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1851 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1852 sbuf_start_section(sb, NULL); 1853 sbuf_bcat(sb, hdr, hdrsize); 1854 TAILQ_FOREACH(ninfo, notelst, link) 1855 __elfN(putnote)(p->td, ninfo, sb); 1856 /* Align up to a page boundary for the program segments. */ 1857 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1858 error = sbuf_finish(sb); 1859 sbuf_delete(sb); 1860 1861 return (error); 1862 } 1863 1864 void 1865 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1866 size_t *sizep) 1867 { 1868 struct proc *p; 1869 struct thread *thr; 1870 size_t size; 1871 1872 p = td->td_proc; 1873 size = 0; 1874 1875 size += __elfN(register_note)(td, list, NT_PRPSINFO, 1876 __elfN(note_prpsinfo), p); 1877 1878 /* 1879 * To have the debugger select the right thread (LWP) as the initial 1880 * thread, we dump the state of the thread passed to us in td first. 1881 * This is the thread that causes the core dump and thus likely to 1882 * be the right thread one wants to have selected in the debugger. 1883 */ 1884 thr = td; 1885 while (thr != NULL) { 1886 size += __elfN(prepare_register_notes)(td, list, thr); 1887 size += __elfN(register_note)(td, list, -1, 1888 __elfN(note_threadmd), thr); 1889 1890 thr = thr == td ? TAILQ_FIRST(&p->p_threads) : 1891 TAILQ_NEXT(thr, td_plist); 1892 if (thr == td) 1893 thr = TAILQ_NEXT(thr, td_plist); 1894 } 1895 1896 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC, 1897 __elfN(note_procstat_proc), p); 1898 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES, 1899 note_procstat_files, p); 1900 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP, 1901 note_procstat_vmmap, p); 1902 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS, 1903 note_procstat_groups, p); 1904 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK, 1905 note_procstat_umask, p); 1906 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT, 1907 note_procstat_rlimit, p); 1908 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL, 1909 note_procstat_osrel, p); 1910 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS, 1911 __elfN(note_procstat_psstrings), p); 1912 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV, 1913 __elfN(note_procstat_auxv), p); 1914 size += __elfN(register_note)(td, list, NT_PROCSTAT_KQUEUES, 1915 __elfN(note_procstat_kqueues), p); 1916 1917 *sizep = size; 1918 } 1919 1920 void 1921 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1922 size_t notesz, int flags) 1923 { 1924 Elf_Ehdr *ehdr; 1925 Elf_Phdr *phdr; 1926 Elf_Shdr *shdr; 1927 struct phdr_closure phc; 1928 Elf_Brandinfo *bi; 1929 1930 ehdr = (Elf_Ehdr *)hdr; 1931 bi = td->td_proc->p_elf_brandinfo; 1932 1933 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1934 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1935 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1936 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1937 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1938 ehdr->e_ident[EI_DATA] = ELF_DATA; 1939 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1940 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi; 1941 ehdr->e_ident[EI_ABIVERSION] = 0; 1942 ehdr->e_ident[EI_PAD] = 0; 1943 ehdr->e_type = ET_CORE; 1944 ehdr->e_machine = bi->machine; 1945 ehdr->e_version = EV_CURRENT; 1946 ehdr->e_entry = 0; 1947 ehdr->e_phoff = sizeof(Elf_Ehdr); 1948 ehdr->e_flags = td->td_proc->p_elf_flags; 1949 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1950 ehdr->e_phentsize = sizeof(Elf_Phdr); 1951 ehdr->e_shentsize = sizeof(Elf_Shdr); 1952 ehdr->e_shstrndx = SHN_UNDEF; 1953 if (numsegs + 1 < PN_XNUM) { 1954 ehdr->e_phnum = numsegs + 1; 1955 ehdr->e_shnum = 0; 1956 } else { 1957 ehdr->e_phnum = PN_XNUM; 1958 ehdr->e_shnum = 1; 1959 1960 ehdr->e_shoff = ehdr->e_phoff + 1961 (numsegs + 1) * ehdr->e_phentsize; 1962 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1963 ("e_shoff: %zu, hdrsize - shdr: %zu", 1964 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1965 1966 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1967 memset(shdr, 0, sizeof(*shdr)); 1968 /* 1969 * A special first section is used to hold large segment and 1970 * section counts. This was proposed by Sun Microsystems in 1971 * Solaris and has been adopted by Linux; the standard ELF 1972 * tools are already familiar with the technique. 1973 * 1974 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1975 * (or 12-7 depending on the version of the document) for more 1976 * details. 1977 */ 1978 shdr->sh_type = SHT_NULL; 1979 shdr->sh_size = ehdr->e_shnum; 1980 shdr->sh_link = ehdr->e_shstrndx; 1981 shdr->sh_info = numsegs + 1; 1982 } 1983 1984 /* 1985 * Fill in the program header entries. 1986 */ 1987 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1988 1989 /* The note segement. */ 1990 phdr->p_type = PT_NOTE; 1991 phdr->p_offset = hdrsize; 1992 phdr->p_vaddr = 0; 1993 phdr->p_paddr = 0; 1994 phdr->p_filesz = notesz; 1995 phdr->p_memsz = 0; 1996 phdr->p_flags = PF_R; 1997 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1998 phdr++; 1999 2000 /* All the writable segments from the program. */ 2001 phc.phdr = phdr; 2002 phc.offset = round_page(hdrsize + notesz); 2003 each_dumpable_segment(td, cb_put_phdr, &phc, flags); 2004 } 2005 2006 static size_t 2007 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list, 2008 struct regset *regset, struct thread *target_td) 2009 { 2010 const struct sysentvec *sv; 2011 struct note_info *ninfo; 2012 size_t size, notesize; 2013 2014 size = 0; 2015 if (!regset->get(regset, target_td, NULL, &size) || size == 0) 2016 return (0); 2017 2018 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2019 ninfo->type = regset->note; 2020 ninfo->regset = regset; 2021 ninfo->outarg = target_td; 2022 ninfo->outsize = size; 2023 TAILQ_INSERT_TAIL(list, ninfo, link); 2024 2025 sv = td->td_proc->p_sysent; 2026 notesize = sizeof(Elf_Note) + /* note header */ 2027 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2028 /* note name */ 2029 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2030 2031 return (notesize); 2032 } 2033 2034 size_t 2035 __elfN(register_note)(struct thread *td, struct note_info_list *list, 2036 int type, outfunc_t out, void *arg) 2037 { 2038 const struct sysentvec *sv; 2039 struct note_info *ninfo; 2040 size_t size, notesize; 2041 2042 sv = td->td_proc->p_sysent; 2043 size = 0; 2044 out(arg, NULL, &size); 2045 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 2046 ninfo->type = type; 2047 ninfo->outfunc = out; 2048 ninfo->outarg = arg; 2049 ninfo->outsize = size; 2050 TAILQ_INSERT_TAIL(list, ninfo, link); 2051 2052 if (type == -1) 2053 return (size); 2054 2055 notesize = sizeof(Elf_Note) + /* note header */ 2056 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) + 2057 /* note name */ 2058 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2059 2060 return (notesize); 2061 } 2062 2063 static size_t 2064 append_note_data(const void *src, void *dst, size_t len) 2065 { 2066 size_t padded_len; 2067 2068 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 2069 if (dst != NULL) { 2070 bcopy(src, dst, len); 2071 bzero((char *)dst + len, padded_len - len); 2072 } 2073 return (padded_len); 2074 } 2075 2076 size_t 2077 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 2078 { 2079 Elf_Note *note; 2080 char *buf; 2081 size_t notesize; 2082 2083 buf = dst; 2084 if (buf != NULL) { 2085 note = (Elf_Note *)buf; 2086 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 2087 note->n_descsz = size; 2088 note->n_type = type; 2089 buf += sizeof(*note); 2090 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 2091 sizeof(FREEBSD_ABI_VENDOR)); 2092 append_note_data(src, buf, size); 2093 if (descp != NULL) 2094 *descp = buf; 2095 } 2096 2097 notesize = sizeof(Elf_Note) + /* note header */ 2098 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 2099 /* note name */ 2100 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 2101 2102 return (notesize); 2103 } 2104 2105 static void 2106 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb) 2107 { 2108 Elf_Note note; 2109 const struct sysentvec *sv; 2110 ssize_t old_len, sect_len; 2111 size_t new_len, descsz, i; 2112 2113 if (ninfo->type == -1) { 2114 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2115 return; 2116 } 2117 2118 sv = td->td_proc->p_sysent; 2119 2120 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1; 2121 note.n_descsz = ninfo->outsize; 2122 note.n_type = ninfo->type; 2123 2124 sbuf_bcat(sb, ¬e, sizeof(note)); 2125 sbuf_start_section(sb, &old_len); 2126 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor, 2127 strlen(sv->sv_elf_core_abi_vendor) + 1); 2128 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2129 if (note.n_descsz == 0) 2130 return; 2131 sbuf_start_section(sb, &old_len); 2132 if (ninfo->regset != NULL) { 2133 struct regset *regset = ninfo->regset; 2134 void *buf; 2135 2136 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK); 2137 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize); 2138 sbuf_bcat(sb, buf, ninfo->outsize); 2139 free(buf, M_TEMP); 2140 } else 2141 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 2142 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 2143 if (sect_len < 0) 2144 return; 2145 2146 new_len = (size_t)sect_len; 2147 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 2148 if (new_len < descsz) { 2149 /* 2150 * It is expected that individual note emitters will correctly 2151 * predict their expected output size and fill up to that size 2152 * themselves, padding in a format-specific way if needed. 2153 * However, in case they don't, just do it here with zeros. 2154 */ 2155 for (i = 0; i < descsz - new_len; i++) 2156 sbuf_putc(sb, 0); 2157 } else if (new_len > descsz) { 2158 /* 2159 * We can't always truncate sb -- we may have drained some 2160 * of it already. 2161 */ 2162 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 2163 "read it (%zu > %zu). Since it is longer than " 2164 "expected, this coredump's notes are corrupt. THIS " 2165 "IS A BUG in the note_procstat routine for type %u.\n", 2166 __func__, (unsigned)note.n_type, new_len, descsz, 2167 (unsigned)note.n_type)); 2168 } 2169 } 2170 2171 /* 2172 * Miscellaneous note out functions. 2173 */ 2174 2175 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2176 #include <compat/freebsd32/freebsd32.h> 2177 #include <compat/freebsd32/freebsd32_signal.h> 2178 2179 typedef struct prstatus32 elf_prstatus_t; 2180 typedef struct prpsinfo32 elf_prpsinfo_t; 2181 typedef struct fpreg32 elf_prfpregset_t; 2182 typedef struct fpreg32 elf_fpregset_t; 2183 typedef struct reg32 elf_gregset_t; 2184 typedef struct thrmisc32 elf_thrmisc_t; 2185 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t; 2186 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 2187 typedef struct kinfo_proc32 elf_kinfo_proc_t; 2188 typedef uint32_t elf_ps_strings_t; 2189 #else 2190 typedef prstatus_t elf_prstatus_t; 2191 typedef prpsinfo_t elf_prpsinfo_t; 2192 typedef prfpregset_t elf_prfpregset_t; 2193 typedef prfpregset_t elf_fpregset_t; 2194 typedef gregset_t elf_gregset_t; 2195 typedef thrmisc_t elf_thrmisc_t; 2196 typedef struct ptrace_lwpinfo elf_lwpinfo_t; 2197 #define ELF_KERN_PROC_MASK 0 2198 typedef struct kinfo_proc elf_kinfo_proc_t; 2199 typedef vm_offset_t elf_ps_strings_t; 2200 #endif 2201 2202 static void 2203 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 2204 { 2205 struct sbuf sbarg; 2206 size_t len; 2207 char *cp, *end; 2208 struct proc *p; 2209 elf_prpsinfo_t *psinfo; 2210 int error; 2211 2212 p = arg; 2213 if (sb != NULL) { 2214 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 2215 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 2216 psinfo->pr_version = PRPSINFO_VERSION; 2217 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 2218 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 2219 PROC_LOCK(p); 2220 if (p->p_args != NULL) { 2221 len = sizeof(psinfo->pr_psargs) - 1; 2222 if (len > p->p_args->ar_length) 2223 len = p->p_args->ar_length; 2224 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 2225 PROC_UNLOCK(p); 2226 error = 0; 2227 } else { 2228 _PHOLD(p); 2229 PROC_UNLOCK(p); 2230 sbuf_new(&sbarg, psinfo->pr_psargs, 2231 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 2232 error = proc_getargv(curthread, p, &sbarg); 2233 PRELE(p); 2234 if (sbuf_finish(&sbarg) == 0) { 2235 len = sbuf_len(&sbarg); 2236 if (len > 0) 2237 len--; 2238 } else { 2239 len = sizeof(psinfo->pr_psargs) - 1; 2240 } 2241 sbuf_delete(&sbarg); 2242 } 2243 if (error != 0 || len == 0 || (ssize_t)len == -1) 2244 strlcpy(psinfo->pr_psargs, p->p_comm, 2245 sizeof(psinfo->pr_psargs)); 2246 else { 2247 KASSERT(len < sizeof(psinfo->pr_psargs), 2248 ("len is too long: %zu vs %zu", len, 2249 sizeof(psinfo->pr_psargs))); 2250 cp = psinfo->pr_psargs; 2251 end = cp + len - 1; 2252 for (;;) { 2253 cp = memchr(cp, '\0', end - cp); 2254 if (cp == NULL) 2255 break; 2256 *cp = ' '; 2257 } 2258 } 2259 psinfo->pr_pid = p->p_pid; 2260 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 2261 free(psinfo, M_TEMP); 2262 } 2263 *sizep = sizeof(*psinfo); 2264 } 2265 2266 static bool 2267 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf, 2268 size_t *sizep) 2269 { 2270 elf_prstatus_t *status; 2271 2272 if (buf != NULL) { 2273 KASSERT(*sizep == sizeof(*status), ("%s: invalid size", 2274 __func__)); 2275 status = buf; 2276 memset(status, 0, *sizep); 2277 status->pr_version = PRSTATUS_VERSION; 2278 status->pr_statussz = sizeof(elf_prstatus_t); 2279 status->pr_gregsetsz = sizeof(elf_gregset_t); 2280 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 2281 status->pr_osreldate = osreldate; 2282 status->pr_cursig = td->td_proc->p_sig; 2283 status->pr_pid = td->td_tid; 2284 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2285 fill_regs32(td, &status->pr_reg); 2286 #else 2287 fill_regs(td, &status->pr_reg); 2288 #endif 2289 } 2290 *sizep = sizeof(*status); 2291 return (true); 2292 } 2293 2294 static bool 2295 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf, 2296 size_t size) 2297 { 2298 elf_prstatus_t *status; 2299 2300 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__)); 2301 status = buf; 2302 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2303 set_regs32(td, &status->pr_reg); 2304 #else 2305 set_regs(td, &status->pr_reg); 2306 #endif 2307 return (true); 2308 } 2309 2310 static struct regset __elfN(regset_prstatus) = { 2311 .note = NT_PRSTATUS, 2312 .size = sizeof(elf_prstatus_t), 2313 .get = __elfN(get_prstatus), 2314 .set = __elfN(set_prstatus), 2315 }; 2316 ELF_REGSET(__elfN(regset_prstatus)); 2317 2318 static bool 2319 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf, 2320 size_t *sizep) 2321 { 2322 elf_prfpregset_t *fpregset; 2323 2324 if (buf != NULL) { 2325 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size", 2326 __func__)); 2327 fpregset = buf; 2328 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2329 fill_fpregs32(td, fpregset); 2330 #else 2331 fill_fpregs(td, fpregset); 2332 #endif 2333 } 2334 *sizep = sizeof(*fpregset); 2335 return (true); 2336 } 2337 2338 static bool 2339 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf, 2340 size_t size) 2341 { 2342 elf_prfpregset_t *fpregset; 2343 2344 fpregset = buf; 2345 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__)); 2346 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2347 set_fpregs32(td, fpregset); 2348 #else 2349 set_fpregs(td, fpregset); 2350 #endif 2351 return (true); 2352 } 2353 2354 static struct regset __elfN(regset_fpregset) = { 2355 .note = NT_FPREGSET, 2356 .size = sizeof(elf_prfpregset_t), 2357 .get = __elfN(get_fpregset), 2358 .set = __elfN(set_fpregset), 2359 }; 2360 ELF_REGSET(__elfN(regset_fpregset)); 2361 2362 static bool 2363 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf, 2364 size_t *sizep) 2365 { 2366 elf_thrmisc_t *thrmisc; 2367 2368 if (buf != NULL) { 2369 KASSERT(*sizep == sizeof(*thrmisc), 2370 ("%s: invalid size", __func__)); 2371 thrmisc = buf; 2372 bzero(thrmisc, sizeof(*thrmisc)); 2373 strcpy(thrmisc->pr_tname, td->td_name); 2374 } 2375 *sizep = sizeof(*thrmisc); 2376 return (true); 2377 } 2378 2379 static struct regset __elfN(regset_thrmisc) = { 2380 .note = NT_THRMISC, 2381 .size = sizeof(elf_thrmisc_t), 2382 .get = __elfN(get_thrmisc), 2383 }; 2384 ELF_REGSET(__elfN(regset_thrmisc)); 2385 2386 static bool 2387 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf, 2388 size_t *sizep) 2389 { 2390 elf_lwpinfo_t pl; 2391 size_t size; 2392 int structsize; 2393 2394 size = sizeof(structsize) + sizeof(pl); 2395 if (buf != NULL) { 2396 KASSERT(*sizep == size, ("%s: invalid size", __func__)); 2397 structsize = sizeof(pl); 2398 memcpy(buf, &structsize, sizeof(structsize)); 2399 bzero(&pl, sizeof(pl)); 2400 pl.pl_lwpid = td->td_tid; 2401 pl.pl_event = PL_EVENT_NONE; 2402 pl.pl_sigmask = td->td_sigmask; 2403 pl.pl_siglist = td->td_siglist; 2404 if (td->td_si.si_signo != 0) { 2405 pl.pl_event = PL_EVENT_SIGNAL; 2406 pl.pl_flags |= PL_FLAG_SI; 2407 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2408 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo); 2409 #else 2410 pl.pl_siginfo = td->td_si; 2411 #endif 2412 } 2413 strcpy(pl.pl_tdname, td->td_name); 2414 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/ 2415 memcpy((int *)buf + 1, &pl, sizeof(pl)); 2416 } 2417 *sizep = size; 2418 return (true); 2419 } 2420 2421 static struct regset __elfN(regset_lwpinfo) = { 2422 .note = NT_PTLWPINFO, 2423 .size = sizeof(int) + sizeof(elf_lwpinfo_t), 2424 .get = __elfN(get_lwpinfo), 2425 }; 2426 ELF_REGSET(__elfN(regset_lwpinfo)); 2427 2428 static size_t 2429 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list, 2430 struct thread *target_td) 2431 { 2432 struct sysentvec *sv = td->td_proc->p_sysent; 2433 struct regset **regsetp, **regset_end, *regset; 2434 size_t size; 2435 2436 size = 0; 2437 2438 if (target_td == td) 2439 cpu_update_pcb(target_td); 2440 2441 /* NT_PRSTATUS must be the first register set note. */ 2442 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus), 2443 target_td); 2444 2445 regsetp = sv->sv_regset_begin; 2446 if (regsetp == NULL) { 2447 /* XXX: This shouldn't be true for any FreeBSD ABIs. */ 2448 size += __elfN(register_regset_note)(td, list, 2449 &__elfN(regset_fpregset), target_td); 2450 return (size); 2451 } 2452 regset_end = sv->sv_regset_end; 2453 MPASS(regset_end != NULL); 2454 for (; regsetp < regset_end; regsetp++) { 2455 regset = *regsetp; 2456 if (regset->note == NT_PRSTATUS) 2457 continue; 2458 size += __elfN(register_regset_note)(td, list, regset, 2459 target_td); 2460 } 2461 return (size); 2462 } 2463 2464 /* 2465 * Allow for MD specific notes, as well as any MD 2466 * specific preparations for writing MI notes. 2467 */ 2468 static void 2469 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2470 { 2471 struct thread *td; 2472 void *buf; 2473 size_t size; 2474 2475 td = (struct thread *)arg; 2476 size = *sizep; 2477 if (size != 0 && sb != NULL) 2478 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2479 else 2480 buf = NULL; 2481 size = 0; 2482 __elfN(dump_thread)(td, buf, &size); 2483 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2484 if (size != 0 && sb != NULL) 2485 sbuf_bcat(sb, buf, size); 2486 free(buf, M_TEMP); 2487 *sizep = size; 2488 } 2489 2490 #ifdef KINFO_PROC_SIZE 2491 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2492 #endif 2493 2494 static void 2495 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2496 { 2497 struct proc *p; 2498 size_t size; 2499 int structsize; 2500 2501 p = arg; 2502 size = sizeof(structsize) + p->p_numthreads * 2503 sizeof(elf_kinfo_proc_t); 2504 2505 if (sb != NULL) { 2506 KASSERT(*sizep == size, ("invalid size")); 2507 structsize = sizeof(elf_kinfo_proc_t); 2508 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2509 sx_slock(&proctree_lock); 2510 PROC_LOCK(p); 2511 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2512 sx_sunlock(&proctree_lock); 2513 } 2514 *sizep = size; 2515 } 2516 2517 #ifdef KINFO_FILE_SIZE 2518 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2519 #endif 2520 2521 static void 2522 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2523 { 2524 struct proc *p; 2525 size_t size, sect_sz, i; 2526 ssize_t start_len, sect_len; 2527 int structsize, filedesc_flags; 2528 2529 if (coredump_pack_fileinfo) 2530 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2531 else 2532 filedesc_flags = 0; 2533 2534 p = arg; 2535 structsize = sizeof(struct kinfo_file); 2536 if (sb == NULL) { 2537 size = 0; 2538 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2539 sbuf_set_drain(sb, sbuf_count_drain, &size); 2540 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2541 PROC_LOCK(p); 2542 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2543 sbuf_finish(sb); 2544 sbuf_delete(sb); 2545 *sizep = size; 2546 } else { 2547 sbuf_start_section(sb, &start_len); 2548 2549 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2550 PROC_LOCK(p); 2551 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2552 filedesc_flags); 2553 2554 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2555 if (sect_len < 0) 2556 return; 2557 sect_sz = sect_len; 2558 2559 KASSERT(sect_sz <= *sizep, 2560 ("kern_proc_filedesc_out did not respect maxlen; " 2561 "requested %zu, got %zu", *sizep - sizeof(structsize), 2562 sect_sz - sizeof(structsize))); 2563 2564 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2565 sbuf_putc(sb, 0); 2566 } 2567 } 2568 2569 #ifdef KINFO_VMENTRY_SIZE 2570 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2571 #endif 2572 2573 static void 2574 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2575 { 2576 struct proc *p; 2577 size_t size; 2578 int structsize, vmmap_flags; 2579 2580 if (coredump_pack_vmmapinfo) 2581 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2582 else 2583 vmmap_flags = 0; 2584 2585 p = arg; 2586 structsize = sizeof(struct kinfo_vmentry); 2587 if (sb == NULL) { 2588 size = 0; 2589 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2590 sbuf_set_drain(sb, sbuf_count_drain, &size); 2591 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2592 PROC_LOCK(p); 2593 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2594 sbuf_finish(sb); 2595 sbuf_delete(sb); 2596 *sizep = size; 2597 } else { 2598 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2599 PROC_LOCK(p); 2600 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2601 vmmap_flags); 2602 } 2603 } 2604 2605 static void 2606 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2607 { 2608 struct proc *p; 2609 size_t size; 2610 int structsize; 2611 2612 p = arg; 2613 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2614 if (sb != NULL) { 2615 KASSERT(*sizep == size, ("invalid size")); 2616 structsize = sizeof(gid_t); 2617 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2618 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2619 sizeof(gid_t)); 2620 } 2621 *sizep = size; 2622 } 2623 2624 static void 2625 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2626 { 2627 struct proc *p; 2628 size_t size; 2629 int structsize; 2630 2631 p = arg; 2632 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask); 2633 if (sb != NULL) { 2634 KASSERT(*sizep == size, ("invalid size")); 2635 structsize = sizeof(p->p_pd->pd_cmask); 2636 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2637 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask)); 2638 } 2639 *sizep = size; 2640 } 2641 2642 static void 2643 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2644 { 2645 struct proc *p; 2646 struct rlimit rlim[RLIM_NLIMITS]; 2647 size_t size; 2648 int structsize, i; 2649 2650 p = arg; 2651 size = sizeof(structsize) + sizeof(rlim); 2652 if (sb != NULL) { 2653 KASSERT(*sizep == size, ("invalid size")); 2654 structsize = sizeof(rlim); 2655 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2656 PROC_LOCK(p); 2657 for (i = 0; i < RLIM_NLIMITS; i++) 2658 lim_rlimit_proc(p, i, &rlim[i]); 2659 PROC_UNLOCK(p); 2660 sbuf_bcat(sb, rlim, sizeof(rlim)); 2661 } 2662 *sizep = size; 2663 } 2664 2665 static void 2666 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2667 { 2668 struct proc *p; 2669 size_t size; 2670 int structsize; 2671 2672 p = arg; 2673 size = sizeof(structsize) + sizeof(p->p_osrel); 2674 if (sb != NULL) { 2675 KASSERT(*sizep == size, ("invalid size")); 2676 structsize = sizeof(p->p_osrel); 2677 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2678 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2679 } 2680 *sizep = size; 2681 } 2682 2683 static void 2684 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2685 { 2686 struct proc *p; 2687 elf_ps_strings_t ps_strings; 2688 size_t size; 2689 int structsize; 2690 2691 p = arg; 2692 size = sizeof(structsize) + sizeof(ps_strings); 2693 if (sb != NULL) { 2694 KASSERT(*sizep == size, ("invalid size")); 2695 structsize = sizeof(ps_strings); 2696 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2697 ps_strings = PTROUT(PROC_PS_STRINGS(p)); 2698 #else 2699 ps_strings = PROC_PS_STRINGS(p); 2700 #endif 2701 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2702 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2703 } 2704 *sizep = size; 2705 } 2706 2707 static void 2708 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2709 { 2710 struct proc *p; 2711 size_t size; 2712 int structsize; 2713 2714 p = arg; 2715 if (sb == NULL) { 2716 size = 0; 2717 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo), 2718 SBUF_FIXEDLEN); 2719 sbuf_set_drain(sb, sbuf_count_drain, &size); 2720 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2721 PHOLD(p); 2722 proc_getauxv(curthread, p, sb); 2723 PRELE(p); 2724 sbuf_finish(sb); 2725 sbuf_delete(sb); 2726 *sizep = size; 2727 } else { 2728 structsize = sizeof(Elf_Auxinfo); 2729 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2730 PHOLD(p); 2731 proc_getauxv(curthread, p, sb); 2732 PRELE(p); 2733 } 2734 } 2735 2736 static void 2737 __elfN(note_procstat_kqueues)(void *arg, struct sbuf *sb, size_t *sizep) 2738 { 2739 struct proc *p; 2740 size_t size, sect_sz, i; 2741 ssize_t start_len, sect_len; 2742 int structsize; 2743 bool compat32; 2744 2745 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2746 compat32 = true; 2747 structsize = sizeof(struct kinfo_knote32); 2748 #else 2749 compat32 = false; 2750 structsize = sizeof(struct kinfo_knote); 2751 #endif 2752 p = arg; 2753 if (sb == NULL) { 2754 size = 0; 2755 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2756 sbuf_set_drain(sb, sbuf_count_drain, &size); 2757 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2758 kern_proc_kqueues_out(p, sb, -1, compat32); 2759 sbuf_finish(sb); 2760 sbuf_delete(sb); 2761 *sizep = size; 2762 } else { 2763 sbuf_start_section(sb, &start_len); 2764 2765 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2766 kern_proc_kqueues_out(p, sb, *sizep - sizeof(structsize), 2767 compat32); 2768 2769 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2770 if (sect_len < 0) 2771 return; 2772 sect_sz = sect_len; 2773 2774 KASSERT(sect_sz <= *sizep, 2775 ("kern_proc_kqueue_out did not respect maxlen; " 2776 "requested %zu, got %zu", *sizep - sizeof(structsize), 2777 sect_sz - sizeof(structsize))); 2778 2779 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2780 sbuf_putc(sb, 0); 2781 } 2782 } 2783 2784 #define MAX_NOTES_LOOP 4096 2785 bool 2786 __elfN(parse_notes)(const struct image_params *imgp, const Elf_Note *checknote, 2787 const char *note_vendor, const Elf_Phdr *pnote, 2788 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg) 2789 { 2790 const Elf_Note *note, *note0, *note_end; 2791 const char *note_name; 2792 char *buf; 2793 int i, error; 2794 bool res; 2795 2796 /* We need some limit, might as well use PAGE_SIZE. */ 2797 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2798 return (false); 2799 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2800 if (pnote->p_offset > PAGE_SIZE || 2801 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2802 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT); 2803 if (buf == NULL) { 2804 VOP_UNLOCK(imgp->vp); 2805 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2806 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 2807 } 2808 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2809 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2810 curthread->td_ucred, NOCRED, NULL, curthread); 2811 if (error != 0) { 2812 uprintf("i/o error PT_NOTE\n"); 2813 goto retf; 2814 } 2815 note = note0 = (const Elf_Note *)buf; 2816 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2817 } else { 2818 note = note0 = (const Elf_Note *)(imgp->image_header + 2819 pnote->p_offset); 2820 note_end = (const Elf_Note *)(imgp->image_header + 2821 pnote->p_offset + pnote->p_filesz); 2822 buf = NULL; 2823 } 2824 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end; 2825 i++) { 2826 if (!aligned(note, Elf32_Addr)) { 2827 uprintf("Unaligned ELF note\n"); 2828 goto retf; 2829 } 2830 if ((const char *)note_end - (const char *)note < 2831 sizeof(Elf_Note)) { 2832 uprintf("ELF note to short\n"); 2833 goto retf; 2834 } 2835 if (note->n_namesz != checknote->n_namesz || 2836 note->n_descsz != checknote->n_descsz || 2837 note->n_type != checknote->n_type) 2838 goto nextnote; 2839 note_name = (const char *)(note + 1); 2840 if (note_name + checknote->n_namesz >= 2841 (const char *)note_end || strncmp(note_vendor, 2842 note_name, checknote->n_namesz) != 0) 2843 goto nextnote; 2844 2845 if (cb(note, cb_arg, &res)) 2846 goto ret; 2847 nextnote: 2848 note = (const Elf_Note *)((const char *)(note + 1) + 2849 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2850 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2851 } 2852 if (i >= MAX_NOTES_LOOP) 2853 uprintf("ELF note parser reached %d notes\n", i); 2854 retf: 2855 res = false; 2856 ret: 2857 free(buf, M_TEMP); 2858 return (res); 2859 } 2860 2861 struct brandnote_cb_arg { 2862 Elf_Brandnote *brandnote; 2863 int32_t *osrel; 2864 }; 2865 2866 static bool 2867 brandnote_cb(const Elf_Note *note, void *arg0, bool *res) 2868 { 2869 struct brandnote_cb_arg *arg; 2870 2871 arg = arg0; 2872 2873 /* 2874 * Fetch the osreldate for binary from the ELF OSABI-note if 2875 * necessary. 2876 */ 2877 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 && 2878 arg->brandnote->trans_osrel != NULL ? 2879 arg->brandnote->trans_osrel(note, arg->osrel) : true; 2880 2881 return (true); 2882 } 2883 2884 static Elf_Note fctl_note = { 2885 .n_namesz = sizeof(FREEBSD_ABI_VENDOR), 2886 .n_descsz = sizeof(uint32_t), 2887 .n_type = NT_FREEBSD_FEATURE_CTL, 2888 }; 2889 2890 struct fctl_cb_arg { 2891 bool *has_fctl0; 2892 uint32_t *fctl0; 2893 }; 2894 2895 static bool 2896 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res) 2897 { 2898 struct fctl_cb_arg *arg; 2899 const Elf32_Word *desc; 2900 uintptr_t p; 2901 2902 arg = arg0; 2903 p = (uintptr_t)(note + 1); 2904 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); 2905 desc = (const Elf32_Word *)p; 2906 *arg->has_fctl0 = true; 2907 *arg->fctl0 = desc[0]; 2908 *res = true; 2909 return (true); 2910 } 2911 2912 /* 2913 * Try to find the appropriate ABI-note section for checknote, fetch 2914 * the osreldate and feature control flags for binary from the ELF 2915 * OSABI-note. Only the first page of the image is searched, the same 2916 * as for headers. 2917 */ 2918 static bool 2919 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote, 2920 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0) 2921 { 2922 const Elf_Phdr *phdr; 2923 const Elf_Ehdr *hdr; 2924 struct brandnote_cb_arg b_arg; 2925 struct fctl_cb_arg f_arg; 2926 int i, j; 2927 2928 hdr = (const Elf_Ehdr *)imgp->image_header; 2929 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2930 b_arg.brandnote = brandnote; 2931 b_arg.osrel = osrel; 2932 f_arg.has_fctl0 = has_fctl0; 2933 f_arg.fctl0 = fctl0; 2934 2935 for (i = 0; i < hdr->e_phnum; i++) { 2936 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, 2937 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb, 2938 &b_arg)) { 2939 for (j = 0; j < hdr->e_phnum; j++) { 2940 if (phdr[j].p_type == PT_NOTE && 2941 __elfN(parse_notes)(imgp, &fctl_note, 2942 FREEBSD_ABI_VENDOR, &phdr[j], 2943 note_fctl_cb, &f_arg)) 2944 break; 2945 } 2946 return (true); 2947 } 2948 } 2949 return (false); 2950 2951 } 2952 2953 /* 2954 * Tell kern_execve.c about it, with a little help from the linker. 2955 */ 2956 static struct execsw __elfN(execsw) = { 2957 .ex_imgact = __CONCAT(exec_, __elfN(imgact)), 2958 .ex_name = ELF_ABI_NAME 2959 }; 2960 EXEC_SET(ELF_ABI_ID, __elfN(execsw)); 2961 2962 static vm_prot_t 2963 __elfN(trans_prot)(Elf_Word flags) 2964 { 2965 vm_prot_t prot; 2966 2967 prot = 0; 2968 if (flags & PF_X) 2969 prot |= VM_PROT_EXECUTE; 2970 if (flags & PF_W) 2971 prot |= VM_PROT_WRITE; 2972 if (flags & PF_R) 2973 prot |= VM_PROT_READ; 2974 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__)) 2975 if (i386_read_exec && (flags & PF_R)) 2976 prot |= VM_PROT_EXECUTE; 2977 #endif 2978 return (prot); 2979 } 2980 2981 static Elf_Word 2982 __elfN(untrans_prot)(vm_prot_t prot) 2983 { 2984 Elf_Word flags; 2985 2986 flags = 0; 2987 if (prot & VM_PROT_EXECUTE) 2988 flags |= PF_X; 2989 if (prot & VM_PROT_READ) 2990 flags |= PF_R; 2991 if (prot & VM_PROT_WRITE) 2992 flags |= PF_W; 2993 return (flags); 2994 } 2995