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