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