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, prot, 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 * Remove write access to the page if it was only granted by map_insert 607 * to allow copyout. 608 */ 609 if ((prot & VM_PROT_WRITE) == 0) 610 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 611 map_len), prot, FALSE); 612 613 return (0); 614 } 615 616 /* 617 * Load the file "file" into memory. It may be either a shared object 618 * or an executable. 619 * 620 * The "addr" reference parameter is in/out. On entry, it specifies 621 * the address where a shared object should be loaded. If the file is 622 * an executable, this value is ignored. On exit, "addr" specifies 623 * where the file was actually loaded. 624 * 625 * The "entry" reference parameter is out only. On exit, it specifies 626 * the entry point for the loaded file. 627 */ 628 static int 629 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 630 u_long *entry, size_t pagesize) 631 { 632 struct { 633 struct nameidata nd; 634 struct vattr attr; 635 struct image_params image_params; 636 } *tempdata; 637 const Elf_Ehdr *hdr = NULL; 638 const Elf_Phdr *phdr = NULL; 639 struct nameidata *nd; 640 struct vattr *attr; 641 struct image_params *imgp; 642 vm_prot_t prot; 643 u_long rbase; 644 u_long base_addr = 0; 645 int error, i, numsegs; 646 647 #ifdef CAPABILITY_MODE 648 /* 649 * XXXJA: This check can go away once we are sufficiently confident 650 * that the checks in namei() are correct. 651 */ 652 if (IN_CAPABILITY_MODE(curthread)) 653 return (ECAPMODE); 654 #endif 655 656 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 657 nd = &tempdata->nd; 658 attr = &tempdata->attr; 659 imgp = &tempdata->image_params; 660 661 /* 662 * Initialize part of the common data 663 */ 664 imgp->proc = p; 665 imgp->attr = attr; 666 imgp->firstpage = NULL; 667 imgp->image_header = NULL; 668 imgp->object = NULL; 669 imgp->execlabel = NULL; 670 671 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); 672 if ((error = namei(nd)) != 0) { 673 nd->ni_vp = NULL; 674 goto fail; 675 } 676 NDFREE(nd, NDF_ONLY_PNBUF); 677 imgp->vp = nd->ni_vp; 678 679 /* 680 * Check permissions, modes, uid, etc on the file, and "open" it. 681 */ 682 error = exec_check_permissions(imgp); 683 if (error) 684 goto fail; 685 686 error = exec_map_first_page(imgp); 687 if (error) 688 goto fail; 689 690 /* 691 * Also make certain that the interpreter stays the same, so set 692 * its VV_TEXT flag, too. 693 */ 694 VOP_SET_TEXT(nd->ni_vp); 695 696 imgp->object = nd->ni_vp->v_object; 697 698 hdr = (const Elf_Ehdr *)imgp->image_header; 699 if ((error = __elfN(check_header)(hdr)) != 0) 700 goto fail; 701 if (hdr->e_type == ET_DYN) 702 rbase = *addr; 703 else if (hdr->e_type == ET_EXEC) 704 rbase = 0; 705 else { 706 error = ENOEXEC; 707 goto fail; 708 } 709 710 /* Only support headers that fit within first page for now */ 711 if ((hdr->e_phoff > PAGE_SIZE) || 712 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 713 error = ENOEXEC; 714 goto fail; 715 } 716 717 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 718 if (!aligned(phdr, Elf_Addr)) { 719 error = ENOEXEC; 720 goto fail; 721 } 722 723 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 724 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 725 /* Loadable segment */ 726 prot = __elfN(trans_prot)(phdr[i].p_flags); 727 error = __elfN(load_section)(imgp, phdr[i].p_offset, 728 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 729 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); 730 if (error != 0) 731 goto fail; 732 /* 733 * Establish the base address if this is the 734 * first segment. 735 */ 736 if (numsegs == 0) 737 base_addr = trunc_page(phdr[i].p_vaddr + 738 rbase); 739 numsegs++; 740 } 741 } 742 *addr = base_addr; 743 *entry = (unsigned long)hdr->e_entry + rbase; 744 745 fail: 746 if (imgp->firstpage) 747 exec_unmap_first_page(imgp); 748 749 if (nd->ni_vp) 750 vput(nd->ni_vp); 751 752 free(tempdata, M_TEMP); 753 754 return (error); 755 } 756 757 static int 758 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 759 { 760 struct thread *td; 761 const Elf_Ehdr *hdr; 762 const Elf_Phdr *phdr; 763 Elf_Auxargs *elf_auxargs; 764 struct vmspace *vmspace; 765 const char *err_str, *newinterp; 766 char *interp, *interp_buf, *path; 767 Elf_Brandinfo *brand_info; 768 struct sysentvec *sv; 769 vm_prot_t prot; 770 u_long text_size, data_size, total_size, text_addr, data_addr; 771 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr; 772 int32_t osrel; 773 int error, i, n, interp_name_len, have_interp; 774 775 hdr = (const Elf_Ehdr *)imgp->image_header; 776 777 /* 778 * Do we have a valid ELF header ? 779 * 780 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 781 * if particular brand doesn't support it. 782 */ 783 if (__elfN(check_header)(hdr) != 0 || 784 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 785 return (-1); 786 787 /* 788 * From here on down, we return an errno, not -1, as we've 789 * detected an ELF file. 790 */ 791 792 if ((hdr->e_phoff > PAGE_SIZE) || 793 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 794 /* Only support headers in first page for now */ 795 uprintf("Program headers not in the first page\n"); 796 return (ENOEXEC); 797 } 798 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 799 if (!aligned(phdr, Elf_Addr)) { 800 uprintf("Unaligned program headers\n"); 801 return (ENOEXEC); 802 } 803 804 n = error = 0; 805 baddr = 0; 806 osrel = 0; 807 text_size = data_size = total_size = text_addr = data_addr = 0; 808 entry = proghdr = 0; 809 interp_name_len = 0; 810 err_str = newinterp = NULL; 811 interp = interp_buf = NULL; 812 td = curthread; 813 814 for (i = 0; i < hdr->e_phnum; i++) { 815 switch (phdr[i].p_type) { 816 case PT_LOAD: 817 if (n == 0) 818 baddr = phdr[i].p_vaddr; 819 n++; 820 break; 821 case PT_INTERP: 822 /* Path to interpreter */ 823 if (phdr[i].p_filesz > MAXPATHLEN) { 824 uprintf("Invalid PT_INTERP\n"); 825 error = ENOEXEC; 826 goto ret; 827 } 828 if (interp != NULL) { 829 uprintf("Multiple PT_INTERP headers\n"); 830 error = ENOEXEC; 831 goto ret; 832 } 833 interp_name_len = phdr[i].p_filesz; 834 if (phdr[i].p_offset > PAGE_SIZE || 835 interp_name_len > PAGE_SIZE - phdr[i].p_offset) { 836 VOP_UNLOCK(imgp->vp, 0); 837 interp_buf = malloc(interp_name_len + 1, M_TEMP, 838 M_WAITOK); 839 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 840 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf, 841 interp_name_len, phdr[i].p_offset, 842 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 843 NOCRED, NULL, td); 844 if (error != 0) { 845 uprintf("i/o error PT_INTERP\n"); 846 goto ret; 847 } 848 interp_buf[interp_name_len] = '\0'; 849 interp = interp_buf; 850 } else { 851 interp = __DECONST(char *, imgp->image_header) + 852 phdr[i].p_offset; 853 } 854 break; 855 case PT_GNU_STACK: 856 if (__elfN(nxstack)) 857 imgp->stack_prot = 858 __elfN(trans_prot)(phdr[i].p_flags); 859 imgp->stack_sz = phdr[i].p_memsz; 860 break; 861 } 862 } 863 864 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 865 &osrel); 866 if (brand_info == NULL) { 867 uprintf("ELF binary type \"%u\" not known.\n", 868 hdr->e_ident[EI_OSABI]); 869 error = ENOEXEC; 870 goto ret; 871 } 872 et_dyn_addr = 0; 873 if (hdr->e_type == ET_DYN) { 874 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 875 uprintf("Cannot execute shared object\n"); 876 error = ENOEXEC; 877 goto ret; 878 } 879 /* 880 * Honour the base load address from the dso if it is 881 * non-zero for some reason. 882 */ 883 if (baddr == 0) 884 et_dyn_addr = ET_DYN_LOAD_ADDR; 885 } 886 sv = brand_info->sysvec; 887 if (interp != NULL && brand_info->interp_newpath != NULL) 888 newinterp = brand_info->interp_newpath; 889 890 /* 891 * Avoid a possible deadlock if the current address space is destroyed 892 * and that address space maps the locked vnode. In the common case, 893 * the locked vnode's v_usecount is decremented but remains greater 894 * than zero. Consequently, the vnode lock is not needed by vrele(). 895 * However, in cases where the vnode lock is external, such as nullfs, 896 * v_usecount may become zero. 897 * 898 * The VV_TEXT flag prevents modifications to the executable while 899 * the vnode is unlocked. 900 */ 901 VOP_UNLOCK(imgp->vp, 0); 902 903 error = exec_new_vmspace(imgp, sv); 904 imgp->proc->p_sysent = sv; 905 906 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 907 if (error != 0) 908 goto ret; 909 910 for (i = 0; i < hdr->e_phnum; i++) { 911 switch (phdr[i].p_type) { 912 case PT_LOAD: /* Loadable segment */ 913 if (phdr[i].p_memsz == 0) 914 break; 915 prot = __elfN(trans_prot)(phdr[i].p_flags); 916 error = __elfN(load_section)(imgp, phdr[i].p_offset, 917 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 918 phdr[i].p_memsz, phdr[i].p_filesz, prot, 919 sv->sv_pagesize); 920 if (error != 0) 921 goto ret; 922 923 /* 924 * If this segment contains the program headers, 925 * remember their virtual address for the AT_PHDR 926 * aux entry. Static binaries don't usually include 927 * a PT_PHDR entry. 928 */ 929 if (phdr[i].p_offset == 0 && 930 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 931 <= phdr[i].p_filesz) 932 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 933 et_dyn_addr; 934 935 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 936 seg_size = round_page(phdr[i].p_memsz + 937 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 938 939 /* 940 * Make the largest executable segment the official 941 * text segment and all others data. 942 * 943 * Note that obreak() assumes that data_addr + 944 * data_size == end of data load area, and the ELF 945 * file format expects segments to be sorted by 946 * address. If multiple data segments exist, the 947 * last one will be used. 948 */ 949 950 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 951 text_size = seg_size; 952 text_addr = seg_addr; 953 } else { 954 data_size = seg_size; 955 data_addr = seg_addr; 956 } 957 total_size += seg_size; 958 break; 959 case PT_PHDR: /* Program header table info */ 960 proghdr = phdr[i].p_vaddr + et_dyn_addr; 961 break; 962 default: 963 break; 964 } 965 } 966 967 if (data_addr == 0 && data_size == 0) { 968 data_addr = text_addr; 969 data_size = text_size; 970 } 971 972 entry = (u_long)hdr->e_entry + et_dyn_addr; 973 974 /* 975 * Check limits. It should be safe to check the 976 * limits after loading the segments since we do 977 * not actually fault in all the segments pages. 978 */ 979 PROC_LOCK(imgp->proc); 980 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 981 err_str = "Data segment size exceeds process limit"; 982 else if (text_size > maxtsiz) 983 err_str = "Text segment size exceeds system limit"; 984 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 985 err_str = "Total segment size exceeds process limit"; 986 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 987 err_str = "Data segment size exceeds resource limit"; 988 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 989 err_str = "Total segment size exceeds resource limit"; 990 if (err_str != NULL) { 991 PROC_UNLOCK(imgp->proc); 992 uprintf("%s\n", err_str); 993 error = ENOMEM; 994 goto ret; 995 } 996 997 vmspace = imgp->proc->p_vmspace; 998 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 999 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 1000 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 1001 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 1002 1003 /* 1004 * We load the dynamic linker where a userland call 1005 * to mmap(0, ...) would put it. The rationale behind this 1006 * calculation is that it leaves room for the heap to grow to 1007 * its maximum allowed size. 1008 */ 1009 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1010 RLIMIT_DATA)); 1011 PROC_UNLOCK(imgp->proc); 1012 1013 imgp->entry_addr = entry; 1014 1015 if (interp != NULL) { 1016 have_interp = FALSE; 1017 VOP_UNLOCK(imgp->vp, 0); 1018 if (brand_info->emul_path != NULL && 1019 brand_info->emul_path[0] != '\0') { 1020 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1021 snprintf(path, MAXPATHLEN, "%s%s", 1022 brand_info->emul_path, interp); 1023 error = __elfN(load_file)(imgp->proc, path, &addr, 1024 &imgp->entry_addr, sv->sv_pagesize); 1025 free(path, M_TEMP); 1026 if (error == 0) 1027 have_interp = TRUE; 1028 } 1029 if (!have_interp && newinterp != NULL && 1030 (brand_info->interp_path == NULL || 1031 strcmp(interp, brand_info->interp_path) == 0)) { 1032 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 1033 &imgp->entry_addr, sv->sv_pagesize); 1034 if (error == 0) 1035 have_interp = TRUE; 1036 } 1037 if (!have_interp) { 1038 error = __elfN(load_file)(imgp->proc, interp, &addr, 1039 &imgp->entry_addr, sv->sv_pagesize); 1040 } 1041 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 1042 if (error != 0) { 1043 uprintf("ELF interpreter %s not found, error %d\n", 1044 interp, error); 1045 goto ret; 1046 } 1047 } else 1048 addr = et_dyn_addr; 1049 1050 /* 1051 * Construct auxargs table (used by the fixup routine) 1052 */ 1053 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1054 elf_auxargs->execfd = -1; 1055 elf_auxargs->phdr = proghdr; 1056 elf_auxargs->phent = hdr->e_phentsize; 1057 elf_auxargs->phnum = hdr->e_phnum; 1058 elf_auxargs->pagesz = PAGE_SIZE; 1059 elf_auxargs->base = addr; 1060 elf_auxargs->flags = 0; 1061 elf_auxargs->entry = entry; 1062 elf_auxargs->hdr_eflags = hdr->e_flags; 1063 1064 imgp->auxargs = elf_auxargs; 1065 imgp->interpreted = 0; 1066 imgp->reloc_base = addr; 1067 imgp->proc->p_osrel = osrel; 1068 imgp->proc->p_elf_machine = hdr->e_machine; 1069 imgp->proc->p_elf_flags = hdr->e_flags; 1070 1071 ret: 1072 free(interp_buf, M_TEMP); 1073 return (error); 1074 } 1075 1076 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 1077 1078 int 1079 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 1080 { 1081 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1082 Elf_Addr *base; 1083 Elf_Addr *pos; 1084 1085 base = (Elf_Addr *)*stack_base; 1086 pos = base + (imgp->args->argc + imgp->args->envc + 2); 1087 1088 if (args->execfd != -1) 1089 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1090 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1091 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1092 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1093 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1094 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1095 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1096 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1097 #ifdef AT_EHDRFLAGS 1098 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1099 #endif 1100 if (imgp->execpathp != 0) 1101 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 1102 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1103 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1104 if (imgp->canary != 0) { 1105 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 1106 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1107 } 1108 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1109 if (imgp->pagesizes != 0) { 1110 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1111 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1112 } 1113 if (imgp->sysent->sv_timekeep_base != 0) { 1114 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1115 imgp->sysent->sv_timekeep_base); 1116 } 1117 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1118 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1119 imgp->sysent->sv_stackprot); 1120 AUXARGS_ENTRY(pos, AT_NULL, 0); 1121 1122 free(imgp->auxargs, M_TEMP); 1123 imgp->auxargs = NULL; 1124 1125 base--; 1126 suword(base, (long)imgp->args->argc); 1127 *stack_base = (register_t *)base; 1128 return (0); 1129 } 1130 1131 /* 1132 * Code for generating ELF core dumps. 1133 */ 1134 1135 typedef void (*segment_callback)(vm_map_entry_t, void *); 1136 1137 /* Closure for cb_put_phdr(). */ 1138 struct phdr_closure { 1139 Elf_Phdr *phdr; /* Program header to fill in */ 1140 Elf_Off offset; /* Offset of segment in core file */ 1141 }; 1142 1143 /* Closure for cb_size_segment(). */ 1144 struct sseg_closure { 1145 int count; /* Count of writable segments. */ 1146 size_t size; /* Total size of all writable segments. */ 1147 }; 1148 1149 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); 1150 1151 struct note_info { 1152 int type; /* Note type. */ 1153 outfunc_t outfunc; /* Output function. */ 1154 void *outarg; /* Argument for the output function. */ 1155 size_t outsize; /* Output size. */ 1156 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1157 }; 1158 1159 TAILQ_HEAD(note_info_list, note_info); 1160 1161 /* Coredump output parameters. */ 1162 struct coredump_params { 1163 off_t offset; 1164 struct ucred *active_cred; 1165 struct ucred *file_cred; 1166 struct thread *td; 1167 struct vnode *vp; 1168 struct gzio_stream *gzs; 1169 }; 1170 1171 static void cb_put_phdr(vm_map_entry_t, void *); 1172 static void cb_size_segment(vm_map_entry_t, void *); 1173 static int core_write(struct coredump_params *, const void *, size_t, off_t, 1174 enum uio_seg); 1175 static void each_dumpable_segment(struct thread *, segment_callback, void *); 1176 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1177 struct note_info_list *, size_t); 1178 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, 1179 size_t *); 1180 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); 1181 static void __elfN(putnote)(struct note_info *, struct sbuf *); 1182 static size_t register_note(struct note_info_list *, int, outfunc_t, void *); 1183 static int sbuf_drain_core_output(void *, const char *, int); 1184 static int sbuf_drain_count(void *arg, const char *data, int len); 1185 1186 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); 1187 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1188 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); 1189 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1190 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); 1191 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1192 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1193 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1194 static void note_procstat_files(void *, struct sbuf *, size_t *); 1195 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1196 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1197 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1198 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1199 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1200 1201 #ifdef GZIO 1202 extern int compress_user_cores_gzlevel; 1203 1204 /* 1205 * Write out a core segment to the compression stream. 1206 */ 1207 static int 1208 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len) 1209 { 1210 u_int chunk_len; 1211 int error; 1212 1213 while (len > 0) { 1214 chunk_len = MIN(len, CORE_BUF_SIZE); 1215 1216 /* 1217 * We can get EFAULT error here. 1218 * In that case zero out the current chunk of the segment. 1219 */ 1220 error = copyin(base, buf, chunk_len); 1221 if (error != 0) 1222 bzero(buf, chunk_len); 1223 error = gzio_write(p->gzs, buf, chunk_len); 1224 if (error != 0) 1225 break; 1226 base += chunk_len; 1227 len -= chunk_len; 1228 } 1229 return (error); 1230 } 1231 1232 static int 1233 core_gz_write(void *base, size_t len, off_t offset, void *arg) 1234 { 1235 1236 return (core_write((struct coredump_params *)arg, base, len, offset, 1237 UIO_SYSSPACE)); 1238 } 1239 #endif /* GZIO */ 1240 1241 static int 1242 core_write(struct coredump_params *p, const void *base, size_t len, 1243 off_t offset, enum uio_seg seg) 1244 { 1245 1246 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base), 1247 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1248 p->active_cred, p->file_cred, NULL, p->td)); 1249 } 1250 1251 static int 1252 core_output(void *base, size_t len, off_t offset, struct coredump_params *p, 1253 void *tmpbuf) 1254 { 1255 int error; 1256 1257 #ifdef GZIO 1258 if (p->gzs != NULL) 1259 return (compress_chunk(p, base, tmpbuf, len)); 1260 #endif 1261 /* 1262 * EFAULT is a non-fatal error that we can get, for example, 1263 * if the segment is backed by a file but extends beyond its 1264 * end. 1265 */ 1266 error = core_write(p, base, len, offset, UIO_USERSPACE); 1267 if (error == EFAULT) { 1268 log(LOG_WARNING, "Failed to fully fault in a core file segment " 1269 "at VA %p with size 0x%zx to be written at offset 0x%jx " 1270 "for process %s\n", base, len, offset, curproc->p_comm); 1271 1272 /* 1273 * Write a "real" zero byte at the end of the target region 1274 * in the case this is the last segment. 1275 * The intermediate space will be implicitly zero-filled. 1276 */ 1277 error = core_write(p, zero_region, 1, offset + len - 1, 1278 UIO_SYSSPACE); 1279 } 1280 return (error); 1281 } 1282 1283 /* 1284 * Drain into a core file. 1285 */ 1286 static int 1287 sbuf_drain_core_output(void *arg, const char *data, int len) 1288 { 1289 struct coredump_params *p; 1290 int error, locked; 1291 1292 p = (struct coredump_params *)arg; 1293 1294 /* 1295 * Some kern_proc out routines that print to this sbuf may 1296 * call us with the process lock held. Draining with the 1297 * non-sleepable lock held is unsafe. The lock is needed for 1298 * those routines when dumping a live process. In our case we 1299 * can safely release the lock before draining and acquire 1300 * again after. 1301 */ 1302 locked = PROC_LOCKED(p->td->td_proc); 1303 if (locked) 1304 PROC_UNLOCK(p->td->td_proc); 1305 #ifdef GZIO 1306 if (p->gzs != NULL) 1307 error = gzio_write(p->gzs, __DECONST(char *, data), len); 1308 else 1309 #endif 1310 error = core_write(p, __DECONST(void *, data), len, p->offset, 1311 UIO_SYSSPACE); 1312 if (locked) 1313 PROC_LOCK(p->td->td_proc); 1314 if (error != 0) 1315 return (-error); 1316 p->offset += len; 1317 return (len); 1318 } 1319 1320 /* 1321 * Drain into a counter. 1322 */ 1323 static int 1324 sbuf_drain_count(void *arg, const char *data __unused, int len) 1325 { 1326 size_t *sizep; 1327 1328 sizep = (size_t *)arg; 1329 *sizep += len; 1330 return (len); 1331 } 1332 1333 int 1334 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1335 { 1336 struct ucred *cred = td->td_ucred; 1337 int error = 0; 1338 struct sseg_closure seginfo; 1339 struct note_info_list notelst; 1340 struct coredump_params params; 1341 struct note_info *ninfo; 1342 void *hdr, *tmpbuf; 1343 size_t hdrsize, notesz, coresize; 1344 #ifdef GZIO 1345 boolean_t compress; 1346 1347 compress = (flags & IMGACT_CORE_COMPRESS) != 0; 1348 #endif 1349 hdr = NULL; 1350 tmpbuf = NULL; 1351 TAILQ_INIT(¬elst); 1352 1353 /* Size the program segments. */ 1354 seginfo.count = 0; 1355 seginfo.size = 0; 1356 each_dumpable_segment(td, cb_size_segment, &seginfo); 1357 1358 /* 1359 * Collect info about the core file header area. 1360 */ 1361 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1362 if (seginfo.count + 1 >= PN_XNUM) 1363 hdrsize += sizeof(Elf_Shdr); 1364 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1365 coresize = round_page(hdrsize + notesz) + seginfo.size; 1366 1367 /* Set up core dump parameters. */ 1368 params.offset = 0; 1369 params.active_cred = cred; 1370 params.file_cred = NOCRED; 1371 params.td = td; 1372 params.vp = vp; 1373 params.gzs = NULL; 1374 1375 #ifdef RACCT 1376 if (racct_enable) { 1377 PROC_LOCK(td->td_proc); 1378 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1379 PROC_UNLOCK(td->td_proc); 1380 if (error != 0) { 1381 error = EFAULT; 1382 goto done; 1383 } 1384 } 1385 #endif 1386 if (coresize >= limit) { 1387 error = EFAULT; 1388 goto done; 1389 } 1390 1391 #ifdef GZIO 1392 /* Create a compression stream if necessary. */ 1393 if (compress) { 1394 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE, 1395 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms); 1396 if (params.gzs == NULL) { 1397 error = EFAULT; 1398 goto done; 1399 } 1400 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1401 } 1402 #endif 1403 1404 /* 1405 * Allocate memory for building the header, fill it up, 1406 * and write it out following the notes. 1407 */ 1408 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1409 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1410 notesz); 1411 1412 /* Write the contents of all of the writable segments. */ 1413 if (error == 0) { 1414 Elf_Phdr *php; 1415 off_t offset; 1416 int i; 1417 1418 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1419 offset = round_page(hdrsize + notesz); 1420 for (i = 0; i < seginfo.count; i++) { 1421 error = core_output((caddr_t)(uintptr_t)php->p_vaddr, 1422 php->p_filesz, offset, ¶ms, tmpbuf); 1423 if (error != 0) 1424 break; 1425 offset += php->p_filesz; 1426 php++; 1427 } 1428 #ifdef GZIO 1429 if (error == 0 && compress) 1430 error = gzio_flush(params.gzs); 1431 #endif 1432 } 1433 if (error) { 1434 log(LOG_WARNING, 1435 "Failed to write core file for process %s (error %d)\n", 1436 curproc->p_comm, error); 1437 } 1438 1439 done: 1440 #ifdef GZIO 1441 if (compress) { 1442 free(tmpbuf, M_TEMP); 1443 if (params.gzs != NULL) 1444 gzio_fini(params.gzs); 1445 } 1446 #endif 1447 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1448 TAILQ_REMOVE(¬elst, ninfo, link); 1449 free(ninfo, M_TEMP); 1450 } 1451 if (hdr != NULL) 1452 free(hdr, M_TEMP); 1453 1454 return (error); 1455 } 1456 1457 /* 1458 * A callback for each_dumpable_segment() to write out the segment's 1459 * program header entry. 1460 */ 1461 static void 1462 cb_put_phdr(entry, closure) 1463 vm_map_entry_t entry; 1464 void *closure; 1465 { 1466 struct phdr_closure *phc = (struct phdr_closure *)closure; 1467 Elf_Phdr *phdr = phc->phdr; 1468 1469 phc->offset = round_page(phc->offset); 1470 1471 phdr->p_type = PT_LOAD; 1472 phdr->p_offset = phc->offset; 1473 phdr->p_vaddr = entry->start; 1474 phdr->p_paddr = 0; 1475 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1476 phdr->p_align = PAGE_SIZE; 1477 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1478 1479 phc->offset += phdr->p_filesz; 1480 phc->phdr++; 1481 } 1482 1483 /* 1484 * A callback for each_dumpable_segment() to gather information about 1485 * the number of segments and their total size. 1486 */ 1487 static void 1488 cb_size_segment(vm_map_entry_t entry, void *closure) 1489 { 1490 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1491 1492 ssc->count++; 1493 ssc->size += entry->end - entry->start; 1494 } 1495 1496 /* 1497 * For each writable segment in the process's memory map, call the given 1498 * function with a pointer to the map entry and some arbitrary 1499 * caller-supplied data. 1500 */ 1501 static void 1502 each_dumpable_segment(struct thread *td, segment_callback func, void *closure) 1503 { 1504 struct proc *p = td->td_proc; 1505 vm_map_t map = &p->p_vmspace->vm_map; 1506 vm_map_entry_t entry; 1507 vm_object_t backing_object, object; 1508 boolean_t ignore_entry; 1509 1510 vm_map_lock_read(map); 1511 for (entry = map->header.next; entry != &map->header; 1512 entry = entry->next) { 1513 /* 1514 * Don't dump inaccessible mappings, deal with legacy 1515 * coredump mode. 1516 * 1517 * Note that read-only segments related to the elf binary 1518 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1519 * need to arbitrarily ignore such segments. 1520 */ 1521 if (elf_legacy_coredump) { 1522 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1523 continue; 1524 } else { 1525 if ((entry->protection & VM_PROT_ALL) == 0) 1526 continue; 1527 } 1528 1529 /* 1530 * Dont include memory segment in the coredump if 1531 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1532 * madvise(2). Do not dump submaps (i.e. parts of the 1533 * kernel map). 1534 */ 1535 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1536 continue; 1537 1538 if ((object = entry->object.vm_object) == NULL) 1539 continue; 1540 1541 /* Ignore memory-mapped devices and such things. */ 1542 VM_OBJECT_RLOCK(object); 1543 while ((backing_object = object->backing_object) != NULL) { 1544 VM_OBJECT_RLOCK(backing_object); 1545 VM_OBJECT_RUNLOCK(object); 1546 object = backing_object; 1547 } 1548 ignore_entry = object->type != OBJT_DEFAULT && 1549 object->type != OBJT_SWAP && object->type != OBJT_VNODE && 1550 object->type != OBJT_PHYS; 1551 VM_OBJECT_RUNLOCK(object); 1552 if (ignore_entry) 1553 continue; 1554 1555 (*func)(entry, closure); 1556 } 1557 vm_map_unlock_read(map); 1558 } 1559 1560 /* 1561 * Write the core file header to the file, including padding up to 1562 * the page boundary. 1563 */ 1564 static int 1565 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1566 size_t hdrsize, struct note_info_list *notelst, size_t notesz) 1567 { 1568 struct note_info *ninfo; 1569 struct sbuf *sb; 1570 int error; 1571 1572 /* Fill in the header. */ 1573 bzero(hdr, hdrsize); 1574 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz); 1575 1576 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1577 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1578 sbuf_start_section(sb, NULL); 1579 sbuf_bcat(sb, hdr, hdrsize); 1580 TAILQ_FOREACH(ninfo, notelst, link) 1581 __elfN(putnote)(ninfo, sb); 1582 /* Align up to a page boundary for the program segments. */ 1583 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1584 error = sbuf_finish(sb); 1585 sbuf_delete(sb); 1586 1587 return (error); 1588 } 1589 1590 static void 1591 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1592 size_t *sizep) 1593 { 1594 struct proc *p; 1595 struct thread *thr; 1596 size_t size; 1597 1598 p = td->td_proc; 1599 size = 0; 1600 1601 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1602 1603 /* 1604 * To have the debugger select the right thread (LWP) as the initial 1605 * thread, we dump the state of the thread passed to us in td first. 1606 * This is the thread that causes the core dump and thus likely to 1607 * be the right thread one wants to have selected in the debugger. 1608 */ 1609 thr = td; 1610 while (thr != NULL) { 1611 size += register_note(list, NT_PRSTATUS, 1612 __elfN(note_prstatus), thr); 1613 size += register_note(list, NT_FPREGSET, 1614 __elfN(note_fpregset), thr); 1615 size += register_note(list, NT_THRMISC, 1616 __elfN(note_thrmisc), thr); 1617 size += register_note(list, -1, 1618 __elfN(note_threadmd), thr); 1619 1620 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1621 TAILQ_NEXT(thr, td_plist); 1622 if (thr == td) 1623 thr = TAILQ_NEXT(thr, td_plist); 1624 } 1625 1626 size += register_note(list, NT_PROCSTAT_PROC, 1627 __elfN(note_procstat_proc), p); 1628 size += register_note(list, NT_PROCSTAT_FILES, 1629 note_procstat_files, p); 1630 size += register_note(list, NT_PROCSTAT_VMMAP, 1631 note_procstat_vmmap, p); 1632 size += register_note(list, NT_PROCSTAT_GROUPS, 1633 note_procstat_groups, p); 1634 size += register_note(list, NT_PROCSTAT_UMASK, 1635 note_procstat_umask, p); 1636 size += register_note(list, NT_PROCSTAT_RLIMIT, 1637 note_procstat_rlimit, p); 1638 size += register_note(list, NT_PROCSTAT_OSREL, 1639 note_procstat_osrel, p); 1640 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1641 __elfN(note_procstat_psstrings), p); 1642 size += register_note(list, NT_PROCSTAT_AUXV, 1643 __elfN(note_procstat_auxv), p); 1644 1645 *sizep = size; 1646 } 1647 1648 static void 1649 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1650 size_t notesz) 1651 { 1652 Elf_Ehdr *ehdr; 1653 Elf_Phdr *phdr; 1654 Elf_Shdr *shdr; 1655 struct phdr_closure phc; 1656 1657 ehdr = (Elf_Ehdr *)hdr; 1658 1659 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1660 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1661 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1662 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1663 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1664 ehdr->e_ident[EI_DATA] = ELF_DATA; 1665 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1666 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1667 ehdr->e_ident[EI_ABIVERSION] = 0; 1668 ehdr->e_ident[EI_PAD] = 0; 1669 ehdr->e_type = ET_CORE; 1670 ehdr->e_machine = td->td_proc->p_elf_machine; 1671 ehdr->e_version = EV_CURRENT; 1672 ehdr->e_entry = 0; 1673 ehdr->e_phoff = sizeof(Elf_Ehdr); 1674 ehdr->e_flags = td->td_proc->p_elf_flags; 1675 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1676 ehdr->e_phentsize = sizeof(Elf_Phdr); 1677 ehdr->e_shentsize = sizeof(Elf_Shdr); 1678 ehdr->e_shstrndx = SHN_UNDEF; 1679 if (numsegs + 1 < PN_XNUM) { 1680 ehdr->e_phnum = numsegs + 1; 1681 ehdr->e_shnum = 0; 1682 } else { 1683 ehdr->e_phnum = PN_XNUM; 1684 ehdr->e_shnum = 1; 1685 1686 ehdr->e_shoff = ehdr->e_phoff + 1687 (numsegs + 1) * ehdr->e_phentsize; 1688 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1689 ("e_shoff: %zu, hdrsize - shdr: %zu", 1690 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1691 1692 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1693 memset(shdr, 0, sizeof(*shdr)); 1694 /* 1695 * A special first section is used to hold large segment and 1696 * section counts. This was proposed by Sun Microsystems in 1697 * Solaris and has been adopted by Linux; the standard ELF 1698 * tools are already familiar with the technique. 1699 * 1700 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1701 * (or 12-7 depending on the version of the document) for more 1702 * details. 1703 */ 1704 shdr->sh_type = SHT_NULL; 1705 shdr->sh_size = ehdr->e_shnum; 1706 shdr->sh_link = ehdr->e_shstrndx; 1707 shdr->sh_info = numsegs + 1; 1708 } 1709 1710 /* 1711 * Fill in the program header entries. 1712 */ 1713 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1714 1715 /* The note segement. */ 1716 phdr->p_type = PT_NOTE; 1717 phdr->p_offset = hdrsize; 1718 phdr->p_vaddr = 0; 1719 phdr->p_paddr = 0; 1720 phdr->p_filesz = notesz; 1721 phdr->p_memsz = 0; 1722 phdr->p_flags = PF_R; 1723 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1724 phdr++; 1725 1726 /* All the writable segments from the program. */ 1727 phc.phdr = phdr; 1728 phc.offset = round_page(hdrsize + notesz); 1729 each_dumpable_segment(td, cb_put_phdr, &phc); 1730 } 1731 1732 static size_t 1733 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) 1734 { 1735 struct note_info *ninfo; 1736 size_t size, notesize; 1737 1738 size = 0; 1739 out(arg, NULL, &size); 1740 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1741 ninfo->type = type; 1742 ninfo->outfunc = out; 1743 ninfo->outarg = arg; 1744 ninfo->outsize = size; 1745 TAILQ_INSERT_TAIL(list, ninfo, link); 1746 1747 if (type == -1) 1748 return (size); 1749 1750 notesize = sizeof(Elf_Note) + /* note header */ 1751 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1752 /* note name */ 1753 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1754 1755 return (notesize); 1756 } 1757 1758 static size_t 1759 append_note_data(const void *src, void *dst, size_t len) 1760 { 1761 size_t padded_len; 1762 1763 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 1764 if (dst != NULL) { 1765 bcopy(src, dst, len); 1766 bzero((char *)dst + len, padded_len - len); 1767 } 1768 return (padded_len); 1769 } 1770 1771 size_t 1772 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 1773 { 1774 Elf_Note *note; 1775 char *buf; 1776 size_t notesize; 1777 1778 buf = dst; 1779 if (buf != NULL) { 1780 note = (Elf_Note *)buf; 1781 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1782 note->n_descsz = size; 1783 note->n_type = type; 1784 buf += sizeof(*note); 1785 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 1786 sizeof(FREEBSD_ABI_VENDOR)); 1787 append_note_data(src, buf, size); 1788 if (descp != NULL) 1789 *descp = buf; 1790 } 1791 1792 notesize = sizeof(Elf_Note) + /* note header */ 1793 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1794 /* note name */ 1795 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1796 1797 return (notesize); 1798 } 1799 1800 static void 1801 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) 1802 { 1803 Elf_Note note; 1804 ssize_t old_len, sect_len; 1805 size_t new_len, descsz, i; 1806 1807 if (ninfo->type == -1) { 1808 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1809 return; 1810 } 1811 1812 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1813 note.n_descsz = ninfo->outsize; 1814 note.n_type = ninfo->type; 1815 1816 sbuf_bcat(sb, ¬e, sizeof(note)); 1817 sbuf_start_section(sb, &old_len); 1818 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR)); 1819 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1820 if (note.n_descsz == 0) 1821 return; 1822 sbuf_start_section(sb, &old_len); 1823 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1824 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1825 if (sect_len < 0) 1826 return; 1827 1828 new_len = (size_t)sect_len; 1829 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 1830 if (new_len < descsz) { 1831 /* 1832 * It is expected that individual note emitters will correctly 1833 * predict their expected output size and fill up to that size 1834 * themselves, padding in a format-specific way if needed. 1835 * However, in case they don't, just do it here with zeros. 1836 */ 1837 for (i = 0; i < descsz - new_len; i++) 1838 sbuf_putc(sb, 0); 1839 } else if (new_len > descsz) { 1840 /* 1841 * We can't always truncate sb -- we may have drained some 1842 * of it already. 1843 */ 1844 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 1845 "read it (%zu > %zu). Since it is longer than " 1846 "expected, this coredump's notes are corrupt. THIS " 1847 "IS A BUG in the note_procstat routine for type %u.\n", 1848 __func__, (unsigned)note.n_type, new_len, descsz, 1849 (unsigned)note.n_type)); 1850 } 1851 } 1852 1853 /* 1854 * Miscellaneous note out functions. 1855 */ 1856 1857 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1858 #include <compat/freebsd32/freebsd32.h> 1859 1860 typedef struct prstatus32 elf_prstatus_t; 1861 typedef struct prpsinfo32 elf_prpsinfo_t; 1862 typedef struct fpreg32 elf_prfpregset_t; 1863 typedef struct fpreg32 elf_fpregset_t; 1864 typedef struct reg32 elf_gregset_t; 1865 typedef struct thrmisc32 elf_thrmisc_t; 1866 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1867 typedef struct kinfo_proc32 elf_kinfo_proc_t; 1868 typedef uint32_t elf_ps_strings_t; 1869 #else 1870 typedef prstatus_t elf_prstatus_t; 1871 typedef prpsinfo_t elf_prpsinfo_t; 1872 typedef prfpregset_t elf_prfpregset_t; 1873 typedef prfpregset_t elf_fpregset_t; 1874 typedef gregset_t elf_gregset_t; 1875 typedef thrmisc_t elf_thrmisc_t; 1876 #define ELF_KERN_PROC_MASK 0 1877 typedef struct kinfo_proc elf_kinfo_proc_t; 1878 typedef vm_offset_t elf_ps_strings_t; 1879 #endif 1880 1881 static void 1882 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 1883 { 1884 struct sbuf sbarg; 1885 size_t len; 1886 char *cp, *end; 1887 struct proc *p; 1888 elf_prpsinfo_t *psinfo; 1889 int error; 1890 1891 p = (struct proc *)arg; 1892 if (sb != NULL) { 1893 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 1894 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 1895 psinfo->pr_version = PRPSINFO_VERSION; 1896 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1897 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1898 PROC_LOCK(p); 1899 if (p->p_args != NULL) { 1900 len = sizeof(psinfo->pr_psargs) - 1; 1901 if (len > p->p_args->ar_length) 1902 len = p->p_args->ar_length; 1903 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 1904 PROC_UNLOCK(p); 1905 error = 0; 1906 } else { 1907 _PHOLD(p); 1908 PROC_UNLOCK(p); 1909 sbuf_new(&sbarg, psinfo->pr_psargs, 1910 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 1911 error = proc_getargv(curthread, p, &sbarg); 1912 PRELE(p); 1913 if (sbuf_finish(&sbarg) == 0) 1914 len = sbuf_len(&sbarg) - 1; 1915 else 1916 len = sizeof(psinfo->pr_psargs) - 1; 1917 sbuf_delete(&sbarg); 1918 } 1919 if (error || len == 0) 1920 strlcpy(psinfo->pr_psargs, p->p_comm, 1921 sizeof(psinfo->pr_psargs)); 1922 else { 1923 KASSERT(len < sizeof(psinfo->pr_psargs), 1924 ("len is too long: %zu vs %zu", len, 1925 sizeof(psinfo->pr_psargs))); 1926 cp = psinfo->pr_psargs; 1927 end = cp + len - 1; 1928 for (;;) { 1929 cp = memchr(cp, '\0', end - cp); 1930 if (cp == NULL) 1931 break; 1932 *cp = ' '; 1933 } 1934 } 1935 psinfo->pr_pid = p->p_pid; 1936 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 1937 free(psinfo, M_TEMP); 1938 } 1939 *sizep = sizeof(*psinfo); 1940 } 1941 1942 static void 1943 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) 1944 { 1945 struct thread *td; 1946 elf_prstatus_t *status; 1947 1948 td = (struct thread *)arg; 1949 if (sb != NULL) { 1950 KASSERT(*sizep == sizeof(*status), ("invalid size")); 1951 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); 1952 status->pr_version = PRSTATUS_VERSION; 1953 status->pr_statussz = sizeof(elf_prstatus_t); 1954 status->pr_gregsetsz = sizeof(elf_gregset_t); 1955 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1956 status->pr_osreldate = osreldate; 1957 status->pr_cursig = td->td_proc->p_sig; 1958 status->pr_pid = td->td_tid; 1959 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1960 fill_regs32(td, &status->pr_reg); 1961 #else 1962 fill_regs(td, &status->pr_reg); 1963 #endif 1964 sbuf_bcat(sb, status, sizeof(*status)); 1965 free(status, M_TEMP); 1966 } 1967 *sizep = sizeof(*status); 1968 } 1969 1970 static void 1971 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) 1972 { 1973 struct thread *td; 1974 elf_prfpregset_t *fpregset; 1975 1976 td = (struct thread *)arg; 1977 if (sb != NULL) { 1978 KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); 1979 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); 1980 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1981 fill_fpregs32(td, fpregset); 1982 #else 1983 fill_fpregs(td, fpregset); 1984 #endif 1985 sbuf_bcat(sb, fpregset, sizeof(*fpregset)); 1986 free(fpregset, M_TEMP); 1987 } 1988 *sizep = sizeof(*fpregset); 1989 } 1990 1991 static void 1992 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) 1993 { 1994 struct thread *td; 1995 elf_thrmisc_t thrmisc; 1996 1997 td = (struct thread *)arg; 1998 if (sb != NULL) { 1999 KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); 2000 bzero(&thrmisc._pad, sizeof(thrmisc._pad)); 2001 strcpy(thrmisc.pr_tname, td->td_name); 2002 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); 2003 } 2004 *sizep = sizeof(thrmisc); 2005 } 2006 2007 /* 2008 * Allow for MD specific notes, as well as any MD 2009 * specific preparations for writing MI notes. 2010 */ 2011 static void 2012 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2013 { 2014 struct thread *td; 2015 void *buf; 2016 size_t size; 2017 2018 td = (struct thread *)arg; 2019 size = *sizep; 2020 if (size != 0 && sb != NULL) 2021 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2022 else 2023 buf = NULL; 2024 size = 0; 2025 __elfN(dump_thread)(td, buf, &size); 2026 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2027 if (size != 0 && sb != NULL) 2028 sbuf_bcat(sb, buf, size); 2029 free(buf, M_TEMP); 2030 *sizep = size; 2031 } 2032 2033 #ifdef KINFO_PROC_SIZE 2034 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2035 #endif 2036 2037 static void 2038 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2039 { 2040 struct proc *p; 2041 size_t size; 2042 int structsize; 2043 2044 p = (struct proc *)arg; 2045 size = sizeof(structsize) + p->p_numthreads * 2046 sizeof(elf_kinfo_proc_t); 2047 2048 if (sb != NULL) { 2049 KASSERT(*sizep == size, ("invalid size")); 2050 structsize = sizeof(elf_kinfo_proc_t); 2051 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2052 sx_slock(&proctree_lock); 2053 PROC_LOCK(p); 2054 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2055 sx_sunlock(&proctree_lock); 2056 } 2057 *sizep = size; 2058 } 2059 2060 #ifdef KINFO_FILE_SIZE 2061 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2062 #endif 2063 2064 static void 2065 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2066 { 2067 struct proc *p; 2068 size_t size, sect_sz, i; 2069 ssize_t start_len, sect_len; 2070 int structsize, filedesc_flags; 2071 2072 if (coredump_pack_fileinfo) 2073 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2074 else 2075 filedesc_flags = 0; 2076 2077 p = (struct proc *)arg; 2078 structsize = sizeof(struct kinfo_file); 2079 if (sb == NULL) { 2080 size = 0; 2081 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2082 sbuf_set_drain(sb, sbuf_drain_count, &size); 2083 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2084 PROC_LOCK(p); 2085 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2086 sbuf_finish(sb); 2087 sbuf_delete(sb); 2088 *sizep = size; 2089 } else { 2090 sbuf_start_section(sb, &start_len); 2091 2092 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2093 PROC_LOCK(p); 2094 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2095 filedesc_flags); 2096 2097 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2098 if (sect_len < 0) 2099 return; 2100 sect_sz = sect_len; 2101 2102 KASSERT(sect_sz <= *sizep, 2103 ("kern_proc_filedesc_out did not respect maxlen; " 2104 "requested %zu, got %zu", *sizep - sizeof(structsize), 2105 sect_sz - sizeof(structsize))); 2106 2107 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2108 sbuf_putc(sb, 0); 2109 } 2110 } 2111 2112 #ifdef KINFO_VMENTRY_SIZE 2113 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2114 #endif 2115 2116 static void 2117 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2118 { 2119 struct proc *p; 2120 size_t size; 2121 int structsize, vmmap_flags; 2122 2123 if (coredump_pack_vmmapinfo) 2124 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2125 else 2126 vmmap_flags = 0; 2127 2128 p = (struct proc *)arg; 2129 structsize = sizeof(struct kinfo_vmentry); 2130 if (sb == NULL) { 2131 size = 0; 2132 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2133 sbuf_set_drain(sb, sbuf_drain_count, &size); 2134 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2135 PROC_LOCK(p); 2136 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2137 sbuf_finish(sb); 2138 sbuf_delete(sb); 2139 *sizep = size; 2140 } else { 2141 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2142 PROC_LOCK(p); 2143 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2144 vmmap_flags); 2145 } 2146 } 2147 2148 static void 2149 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2150 { 2151 struct proc *p; 2152 size_t size; 2153 int structsize; 2154 2155 p = (struct proc *)arg; 2156 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2157 if (sb != NULL) { 2158 KASSERT(*sizep == size, ("invalid size")); 2159 structsize = sizeof(gid_t); 2160 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2161 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2162 sizeof(gid_t)); 2163 } 2164 *sizep = size; 2165 } 2166 2167 static void 2168 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2169 { 2170 struct proc *p; 2171 size_t size; 2172 int structsize; 2173 2174 p = (struct proc *)arg; 2175 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); 2176 if (sb != NULL) { 2177 KASSERT(*sizep == size, ("invalid size")); 2178 structsize = sizeof(p->p_fd->fd_cmask); 2179 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2180 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); 2181 } 2182 *sizep = size; 2183 } 2184 2185 static void 2186 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2187 { 2188 struct proc *p; 2189 struct rlimit rlim[RLIM_NLIMITS]; 2190 size_t size; 2191 int structsize, i; 2192 2193 p = (struct proc *)arg; 2194 size = sizeof(structsize) + sizeof(rlim); 2195 if (sb != NULL) { 2196 KASSERT(*sizep == size, ("invalid size")); 2197 structsize = sizeof(rlim); 2198 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2199 PROC_LOCK(p); 2200 for (i = 0; i < RLIM_NLIMITS; i++) 2201 lim_rlimit_proc(p, i, &rlim[i]); 2202 PROC_UNLOCK(p); 2203 sbuf_bcat(sb, rlim, sizeof(rlim)); 2204 } 2205 *sizep = size; 2206 } 2207 2208 static void 2209 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2210 { 2211 struct proc *p; 2212 size_t size; 2213 int structsize; 2214 2215 p = (struct proc *)arg; 2216 size = sizeof(structsize) + sizeof(p->p_osrel); 2217 if (sb != NULL) { 2218 KASSERT(*sizep == size, ("invalid size")); 2219 structsize = sizeof(p->p_osrel); 2220 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2221 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2222 } 2223 *sizep = size; 2224 } 2225 2226 static void 2227 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2228 { 2229 struct proc *p; 2230 elf_ps_strings_t ps_strings; 2231 size_t size; 2232 int structsize; 2233 2234 p = (struct proc *)arg; 2235 size = sizeof(structsize) + sizeof(ps_strings); 2236 if (sb != NULL) { 2237 KASSERT(*sizep == size, ("invalid size")); 2238 structsize = sizeof(ps_strings); 2239 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2240 ps_strings = PTROUT(p->p_sysent->sv_psstrings); 2241 #else 2242 ps_strings = p->p_sysent->sv_psstrings; 2243 #endif 2244 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2245 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2246 } 2247 *sizep = size; 2248 } 2249 2250 static void 2251 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2252 { 2253 struct proc *p; 2254 size_t size; 2255 int structsize; 2256 2257 p = (struct proc *)arg; 2258 if (sb == NULL) { 2259 size = 0; 2260 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2261 sbuf_set_drain(sb, sbuf_drain_count, &size); 2262 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2263 PHOLD(p); 2264 proc_getauxv(curthread, p, sb); 2265 PRELE(p); 2266 sbuf_finish(sb); 2267 sbuf_delete(sb); 2268 *sizep = size; 2269 } else { 2270 structsize = sizeof(Elf_Auxinfo); 2271 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2272 PHOLD(p); 2273 proc_getauxv(curthread, p, sb); 2274 PRELE(p); 2275 } 2276 } 2277 2278 static boolean_t 2279 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 2280 int32_t *osrel, const Elf_Phdr *pnote) 2281 { 2282 const Elf_Note *note, *note0, *note_end; 2283 const char *note_name; 2284 char *buf; 2285 int i, error; 2286 boolean_t res; 2287 2288 /* We need some limit, might as well use PAGE_SIZE. */ 2289 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2290 return (FALSE); 2291 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2292 if (pnote->p_offset > PAGE_SIZE || 2293 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2294 VOP_UNLOCK(imgp->vp, 0); 2295 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2296 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 2297 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2298 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2299 curthread->td_ucred, NOCRED, NULL, curthread); 2300 if (error != 0) { 2301 uprintf("i/o error PT_NOTE\n"); 2302 res = FALSE; 2303 goto ret; 2304 } 2305 note = note0 = (const Elf_Note *)buf; 2306 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2307 } else { 2308 note = note0 = (const Elf_Note *)(imgp->image_header + 2309 pnote->p_offset); 2310 note_end = (const Elf_Note *)(imgp->image_header + 2311 pnote->p_offset + pnote->p_filesz); 2312 buf = NULL; 2313 } 2314 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 2315 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 2316 (const char *)note < sizeof(Elf_Note)) { 2317 res = FALSE; 2318 goto ret; 2319 } 2320 if (note->n_namesz != checknote->hdr.n_namesz || 2321 note->n_descsz != checknote->hdr.n_descsz || 2322 note->n_type != checknote->hdr.n_type) 2323 goto nextnote; 2324 note_name = (const char *)(note + 1); 2325 if (note_name + checknote->hdr.n_namesz >= 2326 (const char *)note_end || strncmp(checknote->vendor, 2327 note_name, checknote->hdr.n_namesz) != 0) 2328 goto nextnote; 2329 2330 /* 2331 * Fetch the osreldate for binary 2332 * from the ELF OSABI-note if necessary. 2333 */ 2334 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 2335 checknote->trans_osrel != NULL) { 2336 res = checknote->trans_osrel(note, osrel); 2337 goto ret; 2338 } 2339 res = TRUE; 2340 goto ret; 2341 nextnote: 2342 note = (const Elf_Note *)((const char *)(note + 1) + 2343 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2344 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2345 } 2346 res = FALSE; 2347 ret: 2348 free(buf, M_TEMP); 2349 return (res); 2350 } 2351 2352 /* 2353 * Try to find the appropriate ABI-note section for checknote, 2354 * fetch the osreldate for binary from the ELF OSABI-note. Only the 2355 * first page of the image is searched, the same as for headers. 2356 */ 2357 static boolean_t 2358 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 2359 int32_t *osrel) 2360 { 2361 const Elf_Phdr *phdr; 2362 const Elf_Ehdr *hdr; 2363 int i; 2364 2365 hdr = (const Elf_Ehdr *)imgp->image_header; 2366 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2367 2368 for (i = 0; i < hdr->e_phnum; i++) { 2369 if (phdr[i].p_type == PT_NOTE && 2370 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 2371 return (TRUE); 2372 } 2373 return (FALSE); 2374 2375 } 2376 2377 /* 2378 * Tell kern_execve.c about it, with a little help from the linker. 2379 */ 2380 static struct execsw __elfN(execsw) = { 2381 __CONCAT(exec_, __elfN(imgact)), 2382 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2383 }; 2384 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2385 2386 static vm_prot_t 2387 __elfN(trans_prot)(Elf_Word flags) 2388 { 2389 vm_prot_t prot; 2390 2391 prot = 0; 2392 if (flags & PF_X) 2393 prot |= VM_PROT_EXECUTE; 2394 if (flags & PF_W) 2395 prot |= VM_PROT_WRITE; 2396 if (flags & PF_R) 2397 prot |= VM_PROT_READ; 2398 #if __ELF_WORD_SIZE == 32 2399 #if defined(__amd64__) 2400 if (i386_read_exec && (flags & PF_R)) 2401 prot |= VM_PROT_EXECUTE; 2402 #endif 2403 #endif 2404 return (prot); 2405 } 2406 2407 static Elf_Word 2408 __elfN(untrans_prot)(vm_prot_t prot) 2409 { 2410 Elf_Word flags; 2411 2412 flags = 0; 2413 if (prot & VM_PROT_EXECUTE) 2414 flags |= PF_X; 2415 if (prot & VM_PROT_READ) 2416 flags |= PF_R; 2417 if (prot & VM_PROT_WRITE) 2418 flags |= PF_W; 2419 return (flags); 2420 } 2421