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