imgact_elf.c revision 316576
1/*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 S��ren 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: stable/11/sys/kern/imgact_elf.c 316576 2017-04-06 15:19:18Z kib $"); 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 87static int __elfN(check_header)(const Elf_Ehdr *hdr); 88static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, 89 const char *interp, int interp_name_len, int32_t *osrel); 90static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 91 u_long *entry, size_t pagesize); 92static 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); 95static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 96static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, 97 int32_t *osrel); 98static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); 99static boolean_t __elfN(check_note)(struct image_params *imgp, 100 Elf_Brandnote *checknote, int32_t *osrel); 101static vm_prot_t __elfN(trans_prot)(Elf_Word); 102static Elf_Word __elfN(untrans_prot)(vm_prot_t); 103 104SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 105 ""); 106 107#define CORE_BUF_SIZE (16 * 1024) 108 109int __elfN(fallback_brand) = -1; 110SYSCTL_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 114static int elf_legacy_coredump = 0; 115SYSCTL_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 119int __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 126SYSCTL_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__) 132int i386_read_exec = 0; 133SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, 134 "enable execution from readable segments"); 135#endif 136#endif 137 138static 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 144static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; 145 146Elf_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 155static 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 167static const char GNU_ABI_VENDOR[] = "GNU"; 168static int GNU_KFREEBSD_ABI_DESC = 3; 169 170Elf_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 179static boolean_t 180kfreebsd_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 201int 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 220int 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 236int 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 254static 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 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0) 277 continue; 278 if (hdr->e_machine == bi->machine && (bi->flags & 279 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { 280 ret = __elfN(check_note)(imgp, bi->brand_note, osrel); 281 /* Give brand a chance to veto check_note's guess */ 282 if (ret && bi->header_supported) 283 ret = bi->header_supported(imgp); 284 /* 285 * If note checker claimed the binary, but the 286 * interpreter path in the image does not 287 * match default one for the brand, try to 288 * search for other brands with the same 289 * interpreter. Either there is better brand 290 * with the right interpreter, or, failing 291 * this, we return first brand which accepted 292 * our note and, optionally, header. 293 */ 294 if (ret && bi_m == NULL && interp != NULL && 295 (bi->interp_path == NULL || 296 (strlen(bi->interp_path) + 1 != interp_name_len || 297 strncmp(interp, bi->interp_path, interp_name_len) 298 != 0))) { 299 bi_m = bi; 300 ret = 0; 301 } 302 if (ret) 303 return (bi); 304 } 305 } 306 if (bi_m != NULL) 307 return (bi_m); 308 309 /* If the executable has a brand, search for it in the brand list. */ 310 for (i = 0; i < MAX_BRANDS; i++) { 311 bi = elf_brand_list[i]; 312 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 313 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 314 continue; 315 if (hdr->e_machine == bi->machine && 316 (hdr->e_ident[EI_OSABI] == bi->brand || 317 (bi->compat_3_brand != NULL && 318 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 319 bi->compat_3_brand) == 0))) { 320 /* Looks good, but give brand a chance to veto */ 321 if (!bi->header_supported || 322 bi->header_supported(imgp)) { 323 /* 324 * Again, prefer strictly matching 325 * interpreter path. 326 */ 327 if (interp_name_len == 0 && 328 bi->interp_path == NULL) 329 return (bi); 330 if (bi->interp_path != NULL && 331 strlen(bi->interp_path) + 1 == 332 interp_name_len && strncmp(interp, 333 bi->interp_path, interp_name_len) == 0) 334 return (bi); 335 if (bi_m == NULL) 336 bi_m = bi; 337 } 338 } 339 } 340 if (bi_m != NULL) 341 return (bi_m); 342 343 /* No known brand, see if the header is recognized by any brand */ 344 for (i = 0; i < MAX_BRANDS; i++) { 345 bi = elf_brand_list[i]; 346 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 347 bi->header_supported == NULL) 348 continue; 349 if (hdr->e_machine == bi->machine) { 350 ret = bi->header_supported(imgp); 351 if (ret) 352 return (bi); 353 } 354 } 355 356 /* Lacking a known brand, search for a recognized interpreter. */ 357 if (interp != NULL) { 358 for (i = 0; i < MAX_BRANDS; i++) { 359 bi = elf_brand_list[i]; 360 if (bi == NULL || (bi->flags & 361 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC)) 362 != 0) 363 continue; 364 if (hdr->e_machine == bi->machine && 365 bi->interp_path != NULL && 366 /* ELF image p_filesz includes terminating zero */ 367 strlen(bi->interp_path) + 1 == interp_name_len && 368 strncmp(interp, bi->interp_path, interp_name_len) 369 == 0) 370 return (bi); 371 } 372 } 373 374 /* Lacking a recognized interpreter, try the default brand */ 375 for (i = 0; i < MAX_BRANDS; i++) { 376 bi = elf_brand_list[i]; 377 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 || 378 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)) 379 continue; 380 if (hdr->e_machine == bi->machine && 381 __elfN(fallback_brand) == bi->brand) 382 return (bi); 383 } 384 return (NULL); 385} 386 387static int 388__elfN(check_header)(const Elf_Ehdr *hdr) 389{ 390 Elf_Brandinfo *bi; 391 int i; 392 393 if (!IS_ELF(*hdr) || 394 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 395 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 396 hdr->e_ident[EI_VERSION] != EV_CURRENT || 397 hdr->e_phentsize != sizeof(Elf_Phdr) || 398 hdr->e_version != ELF_TARG_VER) 399 return (ENOEXEC); 400 401 /* 402 * Make sure we have at least one brand for this machine. 403 */ 404 405 for (i = 0; i < MAX_BRANDS; i++) { 406 bi = elf_brand_list[i]; 407 if (bi != NULL && bi->machine == hdr->e_machine) 408 break; 409 } 410 if (i == MAX_BRANDS) 411 return (ENOEXEC); 412 413 return (0); 414} 415 416static int 417__elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 418 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 419{ 420 struct sf_buf *sf; 421 int error; 422 vm_offset_t off; 423 424 /* 425 * Create the page if it doesn't exist yet. Ignore errors. 426 */ 427 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) - 428 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL); 429 430 /* 431 * Find the page from the underlying object. 432 */ 433 if (object != NULL) { 434 sf = vm_imgact_map_page(object, offset); 435 if (sf == NULL) 436 return (KERN_FAILURE); 437 off = offset - trunc_page(offset); 438 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 439 end - start); 440 vm_imgact_unmap_page(sf); 441 if (error != 0) 442 return (KERN_FAILURE); 443 } 444 445 return (KERN_SUCCESS); 446} 447 448static int 449__elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object, 450 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot, 451 int cow) 452{ 453 struct sf_buf *sf; 454 vm_offset_t off; 455 vm_size_t sz; 456 int error, locked, rv; 457 458 if (start != trunc_page(start)) { 459 rv = __elfN(map_partial)(map, object, offset, start, 460 round_page(start), prot); 461 if (rv != KERN_SUCCESS) 462 return (rv); 463 offset += round_page(start) - start; 464 start = round_page(start); 465 } 466 if (end != round_page(end)) { 467 rv = __elfN(map_partial)(map, object, offset + 468 trunc_page(end) - start, trunc_page(end), end, prot); 469 if (rv != KERN_SUCCESS) 470 return (rv); 471 end = trunc_page(end); 472 } 473 if (start >= end) 474 return (KERN_SUCCESS); 475 if ((offset & PAGE_MASK) != 0) { 476 /* 477 * The mapping is not page aligned. This means that we have 478 * to copy the data. 479 */ 480 rv = vm_map_fixed(map, NULL, 0, start, end - start, 481 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL); 482 if (rv != KERN_SUCCESS) 483 return (rv); 484 if (object == NULL) 485 return (KERN_SUCCESS); 486 for (; start < end; start += sz) { 487 sf = vm_imgact_map_page(object, offset); 488 if (sf == NULL) 489 return (KERN_FAILURE); 490 off = offset - trunc_page(offset); 491 sz = end - start; 492 if (sz > PAGE_SIZE - off) 493 sz = PAGE_SIZE - off; 494 error = copyout((caddr_t)sf_buf_kva(sf) + off, 495 (caddr_t)start, sz); 496 vm_imgact_unmap_page(sf); 497 if (error != 0) 498 return (KERN_FAILURE); 499 offset += sz; 500 } 501 } else { 502 vm_object_reference(object); 503 rv = vm_map_fixed(map, object, offset, start, end - start, 504 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL); 505 if (rv != KERN_SUCCESS) { 506 locked = VOP_ISLOCKED(imgp->vp); 507 VOP_UNLOCK(imgp->vp, 0); 508 vm_object_deallocate(object); 509 vn_lock(imgp->vp, locked | LK_RETRY); 510 return (rv); 511 } 512 } 513 return (KERN_SUCCESS); 514} 515 516static int 517__elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset, 518 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 519 size_t pagesize) 520{ 521 struct sf_buf *sf; 522 size_t map_len; 523 vm_map_t map; 524 vm_object_t object; 525 vm_offset_t off, map_addr; 526 int error, rv, cow; 527 size_t copy_len; 528 vm_ooffset_t file_addr; 529 530 /* 531 * It's necessary to fail if the filsz + offset taken from the 532 * header is greater than the actual file pager object's size. 533 * If we were to allow this, then the vm_map_find() below would 534 * walk right off the end of the file object and into the ether. 535 * 536 * While I'm here, might as well check for something else that 537 * is invalid: filsz cannot be greater than memsz. 538 */ 539 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) || 540 filsz > memsz) { 541 uprintf("elf_load_section: truncated ELF file\n"); 542 return (ENOEXEC); 543 } 544 545 object = imgp->object; 546 map = &imgp->proc->p_vmspace->vm_map; 547 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 548 file_addr = trunc_page_ps(offset, pagesize); 549 550 /* 551 * We have two choices. We can either clear the data in the last page 552 * of an oversized mapping, or we can start the anon mapping a page 553 * early and copy the initialized data into that first page. We 554 * choose the second. 555 */ 556 if (filsz == 0) 557 map_len = 0; 558 else if (memsz > filsz) 559 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 560 else 561 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 562 563 if (map_len != 0) { 564 /* cow flags: don't dump readonly sections in core */ 565 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 566 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 567 568 rv = __elfN(map_insert)(imgp, map, 569 object, 570 file_addr, /* file offset */ 571 map_addr, /* virtual start */ 572 map_addr + map_len,/* virtual end */ 573 prot, 574 cow); 575 if (rv != KERN_SUCCESS) 576 return (EINVAL); 577 578 /* we can stop now if we've covered it all */ 579 if (memsz == filsz) 580 return (0); 581 } 582 583 584 /* 585 * We have to get the remaining bit of the file into the first part 586 * of the oversized map segment. This is normally because the .data 587 * segment in the file is extended to provide bss. It's a neat idea 588 * to try and save a page, but it's a pain in the behind to implement. 589 */ 590 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset + 591 filsz, pagesize); 592 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 593 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 594 map_addr; 595 596 /* This had damn well better be true! */ 597 if (map_len != 0) { 598 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr, 599 map_addr + map_len, VM_PROT_ALL, 0); 600 if (rv != KERN_SUCCESS) 601 return (EINVAL); 602 } 603 604 if (copy_len != 0) { 605 sf = vm_imgact_map_page(object, offset + filsz); 606 if (sf == NULL) 607 return (EIO); 608 609 /* send the page fragment to user space */ 610 off = trunc_page_ps(offset + filsz, pagesize) - 611 trunc_page(offset + filsz); 612 error = copyout((caddr_t)sf_buf_kva(sf) + off, 613 (caddr_t)map_addr, copy_len); 614 vm_imgact_unmap_page(sf); 615 if (error != 0) 616 return (error); 617 } 618 619 /* 620 * set it to the specified protection. 621 */ 622 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 623 map_len), prot, FALSE); 624 625 return (0); 626} 627 628/* 629 * Load the file "file" into memory. It may be either a shared object 630 * or an executable. 631 * 632 * The "addr" reference parameter is in/out. On entry, it specifies 633 * the address where a shared object should be loaded. If the file is 634 * an executable, this value is ignored. On exit, "addr" specifies 635 * where the file was actually loaded. 636 * 637 * The "entry" reference parameter is out only. On exit, it specifies 638 * the entry point for the loaded file. 639 */ 640static int 641__elfN(load_file)(struct proc *p, const char *file, u_long *addr, 642 u_long *entry, size_t pagesize) 643{ 644 struct { 645 struct nameidata nd; 646 struct vattr attr; 647 struct image_params image_params; 648 } *tempdata; 649 const Elf_Ehdr *hdr = NULL; 650 const Elf_Phdr *phdr = NULL; 651 struct nameidata *nd; 652 struct vattr *attr; 653 struct image_params *imgp; 654 vm_prot_t prot; 655 u_long rbase; 656 u_long base_addr = 0; 657 int error, i, numsegs; 658 659#ifdef CAPABILITY_MODE 660 /* 661 * XXXJA: This check can go away once we are sufficiently confident 662 * that the checks in namei() are correct. 663 */ 664 if (IN_CAPABILITY_MODE(curthread)) 665 return (ECAPMODE); 666#endif 667 668 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 669 nd = &tempdata->nd; 670 attr = &tempdata->attr; 671 imgp = &tempdata->image_params; 672 673 /* 674 * Initialize part of the common data 675 */ 676 imgp->proc = p; 677 imgp->attr = attr; 678 imgp->firstpage = NULL; 679 imgp->image_header = NULL; 680 imgp->object = NULL; 681 imgp->execlabel = NULL; 682 683 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); 684 if ((error = namei(nd)) != 0) { 685 nd->ni_vp = NULL; 686 goto fail; 687 } 688 NDFREE(nd, NDF_ONLY_PNBUF); 689 imgp->vp = nd->ni_vp; 690 691 /* 692 * Check permissions, modes, uid, etc on the file, and "open" it. 693 */ 694 error = exec_check_permissions(imgp); 695 if (error) 696 goto fail; 697 698 error = exec_map_first_page(imgp); 699 if (error) 700 goto fail; 701 702 /* 703 * Also make certain that the interpreter stays the same, so set 704 * its VV_TEXT flag, too. 705 */ 706 VOP_SET_TEXT(nd->ni_vp); 707 708 imgp->object = nd->ni_vp->v_object; 709 710 hdr = (const Elf_Ehdr *)imgp->image_header; 711 if ((error = __elfN(check_header)(hdr)) != 0) 712 goto fail; 713 if (hdr->e_type == ET_DYN) 714 rbase = *addr; 715 else if (hdr->e_type == ET_EXEC) 716 rbase = 0; 717 else { 718 error = ENOEXEC; 719 goto fail; 720 } 721 722 /* Only support headers that fit within first page for now */ 723 if ((hdr->e_phoff > PAGE_SIZE) || 724 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 725 error = ENOEXEC; 726 goto fail; 727 } 728 729 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 730 if (!aligned(phdr, Elf_Addr)) { 731 error = ENOEXEC; 732 goto fail; 733 } 734 735 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 736 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 737 /* Loadable segment */ 738 prot = __elfN(trans_prot)(phdr[i].p_flags); 739 error = __elfN(load_section)(imgp, phdr[i].p_offset, 740 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 741 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); 742 if (error != 0) 743 goto fail; 744 /* 745 * Establish the base address if this is the 746 * first segment. 747 */ 748 if (numsegs == 0) 749 base_addr = trunc_page(phdr[i].p_vaddr + 750 rbase); 751 numsegs++; 752 } 753 } 754 *addr = base_addr; 755 *entry = (unsigned long)hdr->e_entry + rbase; 756 757fail: 758 if (imgp->firstpage) 759 exec_unmap_first_page(imgp); 760 761 if (nd->ni_vp) 762 vput(nd->ni_vp); 763 764 free(tempdata, M_TEMP); 765 766 return (error); 767} 768 769static int 770__CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 771{ 772 struct thread *td; 773 const Elf_Ehdr *hdr; 774 const Elf_Phdr *phdr; 775 Elf_Auxargs *elf_auxargs; 776 struct vmspace *vmspace; 777 const char *err_str, *newinterp; 778 char *interp, *interp_buf, *path; 779 Elf_Brandinfo *brand_info; 780 struct sysentvec *sv; 781 vm_prot_t prot; 782 u_long text_size, data_size, total_size, text_addr, data_addr; 783 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr; 784 int32_t osrel; 785 int error, i, n, interp_name_len, have_interp; 786 787 hdr = (const Elf_Ehdr *)imgp->image_header; 788 789 /* 790 * Do we have a valid ELF header ? 791 * 792 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 793 * if particular brand doesn't support it. 794 */ 795 if (__elfN(check_header)(hdr) != 0 || 796 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 797 return (-1); 798 799 /* 800 * From here on down, we return an errno, not -1, as we've 801 * detected an ELF file. 802 */ 803 804 if ((hdr->e_phoff > PAGE_SIZE) || 805 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 806 /* Only support headers in first page for now */ 807 uprintf("Program headers not in the first page\n"); 808 return (ENOEXEC); 809 } 810 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 811 if (!aligned(phdr, Elf_Addr)) { 812 uprintf("Unaligned program headers\n"); 813 return (ENOEXEC); 814 } 815 816 n = error = 0; 817 baddr = 0; 818 osrel = 0; 819 text_size = data_size = total_size = text_addr = data_addr = 0; 820 entry = proghdr = 0; 821 interp_name_len = 0; 822 err_str = newinterp = NULL; 823 interp = interp_buf = NULL; 824 td = curthread; 825 826 for (i = 0; i < hdr->e_phnum; i++) { 827 switch (phdr[i].p_type) { 828 case PT_LOAD: 829 if (n == 0) 830 baddr = phdr[i].p_vaddr; 831 n++; 832 break; 833 case PT_INTERP: 834 /* Path to interpreter */ 835 if (phdr[i].p_filesz > MAXPATHLEN) { 836 uprintf("Invalid PT_INTERP\n"); 837 error = ENOEXEC; 838 goto ret; 839 } 840 if (interp != NULL) { 841 uprintf("Multiple PT_INTERP headers\n"); 842 error = ENOEXEC; 843 goto ret; 844 } 845 interp_name_len = phdr[i].p_filesz; 846 if (phdr[i].p_offset > PAGE_SIZE || 847 interp_name_len > PAGE_SIZE - phdr[i].p_offset) { 848 VOP_UNLOCK(imgp->vp, 0); 849 interp_buf = malloc(interp_name_len + 1, M_TEMP, 850 M_WAITOK); 851 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 852 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf, 853 interp_name_len, phdr[i].p_offset, 854 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 855 NOCRED, NULL, td); 856 if (error != 0) { 857 uprintf("i/o error PT_INTERP\n"); 858 goto ret; 859 } 860 interp_buf[interp_name_len] = '\0'; 861 interp = interp_buf; 862 } else { 863 interp = __DECONST(char *, imgp->image_header) + 864 phdr[i].p_offset; 865 } 866 break; 867 case PT_GNU_STACK: 868 if (__elfN(nxstack)) 869 imgp->stack_prot = 870 __elfN(trans_prot)(phdr[i].p_flags); 871 imgp->stack_sz = phdr[i].p_memsz; 872 break; 873 } 874 } 875 876 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 877 &osrel); 878 if (brand_info == NULL) { 879 uprintf("ELF binary type \"%u\" not known.\n", 880 hdr->e_ident[EI_OSABI]); 881 error = ENOEXEC; 882 goto ret; 883 } 884 if (hdr->e_type == ET_DYN) { 885 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 886 uprintf("Cannot execute shared object\n"); 887 error = ENOEXEC; 888 goto ret; 889 } 890 /* 891 * Honour the base load address from the dso if it is 892 * non-zero for some reason. 893 */ 894 if (baddr == 0) 895 et_dyn_addr = ET_DYN_LOAD_ADDR; 896 else 897 et_dyn_addr = 0; 898 } else 899 et_dyn_addr = 0; 900 sv = brand_info->sysvec; 901 if (interp != NULL && brand_info->interp_newpath != NULL) 902 newinterp = brand_info->interp_newpath; 903 904 /* 905 * Avoid a possible deadlock if the current address space is destroyed 906 * and that address space maps the locked vnode. In the common case, 907 * the locked vnode's v_usecount is decremented but remains greater 908 * than zero. Consequently, the vnode lock is not needed by vrele(). 909 * However, in cases where the vnode lock is external, such as nullfs, 910 * v_usecount may become zero. 911 * 912 * The VV_TEXT flag prevents modifications to the executable while 913 * the vnode is unlocked. 914 */ 915 VOP_UNLOCK(imgp->vp, 0); 916 917 error = exec_new_vmspace(imgp, sv); 918 imgp->proc->p_sysent = sv; 919 920 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 921 if (error != 0) 922 goto ret; 923 924 for (i = 0; i < hdr->e_phnum; i++) { 925 switch (phdr[i].p_type) { 926 case PT_LOAD: /* Loadable segment */ 927 if (phdr[i].p_memsz == 0) 928 break; 929 prot = __elfN(trans_prot)(phdr[i].p_flags); 930 error = __elfN(load_section)(imgp, phdr[i].p_offset, 931 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 932 phdr[i].p_memsz, phdr[i].p_filesz, prot, 933 sv->sv_pagesize); 934 if (error != 0) 935 goto ret; 936 937 /* 938 * If this segment contains the program headers, 939 * remember their virtual address for the AT_PHDR 940 * aux entry. Static binaries don't usually include 941 * a PT_PHDR entry. 942 */ 943 if (phdr[i].p_offset == 0 && 944 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 945 <= phdr[i].p_filesz) 946 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 947 et_dyn_addr; 948 949 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 950 seg_size = round_page(phdr[i].p_memsz + 951 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 952 953 /* 954 * Make the largest executable segment the official 955 * text segment and all others data. 956 * 957 * Note that obreak() assumes that data_addr + 958 * data_size == end of data load area, and the ELF 959 * file format expects segments to be sorted by 960 * address. If multiple data segments exist, the 961 * last one will be used. 962 */ 963 964 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 965 text_size = seg_size; 966 text_addr = seg_addr; 967 } else { 968 data_size = seg_size; 969 data_addr = seg_addr; 970 } 971 total_size += seg_size; 972 break; 973 case PT_PHDR: /* Program header table info */ 974 proghdr = phdr[i].p_vaddr + et_dyn_addr; 975 break; 976 default: 977 break; 978 } 979 } 980 981 if (data_addr == 0 && data_size == 0) { 982 data_addr = text_addr; 983 data_size = text_size; 984 } 985 986 entry = (u_long)hdr->e_entry + et_dyn_addr; 987 988 /* 989 * Check limits. It should be safe to check the 990 * limits after loading the segments since we do 991 * not actually fault in all the segments pages. 992 */ 993 PROC_LOCK(imgp->proc); 994 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 995 err_str = "Data segment size exceeds process limit"; 996 else if (text_size > maxtsiz) 997 err_str = "Text segment size exceeds system limit"; 998 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 999 err_str = "Total segment size exceeds process limit"; 1000 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 1001 err_str = "Data segment size exceeds resource limit"; 1002 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 1003 err_str = "Total segment size exceeds resource limit"; 1004 if (err_str != NULL) { 1005 PROC_UNLOCK(imgp->proc); 1006 uprintf("%s\n", err_str); 1007 error = ENOMEM; 1008 goto ret; 1009 } 1010 1011 vmspace = imgp->proc->p_vmspace; 1012 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 1013 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 1014 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 1015 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 1016 1017 /* 1018 * We load the dynamic linker where a userland call 1019 * to mmap(0, ...) would put it. The rationale behind this 1020 * calculation is that it leaves room for the heap to grow to 1021 * its maximum allowed size. 1022 */ 1023 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1024 RLIMIT_DATA)); 1025 PROC_UNLOCK(imgp->proc); 1026 1027 imgp->entry_addr = entry; 1028 1029 if (interp != NULL) { 1030 have_interp = FALSE; 1031 VOP_UNLOCK(imgp->vp, 0); 1032 if (brand_info->emul_path != NULL && 1033 brand_info->emul_path[0] != '\0') { 1034 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1035 snprintf(path, MAXPATHLEN, "%s%s", 1036 brand_info->emul_path, interp); 1037 error = __elfN(load_file)(imgp->proc, path, &addr, 1038 &imgp->entry_addr, sv->sv_pagesize); 1039 free(path, M_TEMP); 1040 if (error == 0) 1041 have_interp = TRUE; 1042 } 1043 if (!have_interp && newinterp != NULL && 1044 (brand_info->interp_path == NULL || 1045 strcmp(interp, brand_info->interp_path) == 0)) { 1046 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 1047 &imgp->entry_addr, sv->sv_pagesize); 1048 if (error == 0) 1049 have_interp = TRUE; 1050 } 1051 if (!have_interp) { 1052 error = __elfN(load_file)(imgp->proc, interp, &addr, 1053 &imgp->entry_addr, sv->sv_pagesize); 1054 } 1055 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 1056 if (error != 0) { 1057 uprintf("ELF interpreter %s not found, error %d\n", 1058 interp, error); 1059 goto ret; 1060 } 1061 } else 1062 addr = et_dyn_addr; 1063 1064 /* 1065 * Construct auxargs table (used by the fixup routine) 1066 */ 1067 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1068 elf_auxargs->execfd = -1; 1069 elf_auxargs->phdr = proghdr; 1070 elf_auxargs->phent = hdr->e_phentsize; 1071 elf_auxargs->phnum = hdr->e_phnum; 1072 elf_auxargs->pagesz = PAGE_SIZE; 1073 elf_auxargs->base = addr; 1074 elf_auxargs->flags = 0; 1075 elf_auxargs->entry = entry; 1076 elf_auxargs->hdr_eflags = hdr->e_flags; 1077 1078 imgp->auxargs = elf_auxargs; 1079 imgp->interpreted = 0; 1080 imgp->reloc_base = addr; 1081 imgp->proc->p_osrel = osrel; 1082 1083 ret: 1084 free(interp_buf, M_TEMP); 1085 return (error); 1086} 1087 1088#define suword __CONCAT(suword, __ELF_WORD_SIZE) 1089 1090int 1091__elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 1092{ 1093 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1094 Elf_Addr *base; 1095 Elf_Addr *pos; 1096 1097 base = (Elf_Addr *)*stack_base; 1098 pos = base + (imgp->args->argc + imgp->args->envc + 2); 1099 1100 if (args->execfd != -1) 1101 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1102 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1103 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1104 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1105 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1106 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1107 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1108 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1109#ifdef AT_EHDRFLAGS 1110 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1111#endif 1112 if (imgp->execpathp != 0) 1113 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 1114 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1115 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1116 if (imgp->canary != 0) { 1117 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 1118 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1119 } 1120 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1121 if (imgp->pagesizes != 0) { 1122 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1123 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1124 } 1125 if (imgp->sysent->sv_timekeep_base != 0) { 1126 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1127 imgp->sysent->sv_timekeep_base); 1128 } 1129 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1130 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1131 imgp->sysent->sv_stackprot); 1132 AUXARGS_ENTRY(pos, AT_NULL, 0); 1133 1134 free(imgp->auxargs, M_TEMP); 1135 imgp->auxargs = NULL; 1136 1137 base--; 1138 suword(base, (long)imgp->args->argc); 1139 *stack_base = (register_t *)base; 1140 return (0); 1141} 1142 1143/* 1144 * Code for generating ELF core dumps. 1145 */ 1146 1147typedef void (*segment_callback)(vm_map_entry_t, void *); 1148 1149/* Closure for cb_put_phdr(). */ 1150struct phdr_closure { 1151 Elf_Phdr *phdr; /* Program header to fill in */ 1152 Elf_Off offset; /* Offset of segment in core file */ 1153}; 1154 1155/* Closure for cb_size_segment(). */ 1156struct sseg_closure { 1157 int count; /* Count of writable segments. */ 1158 size_t size; /* Total size of all writable segments. */ 1159}; 1160 1161typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); 1162 1163struct note_info { 1164 int type; /* Note type. */ 1165 outfunc_t outfunc; /* Output function. */ 1166 void *outarg; /* Argument for the output function. */ 1167 size_t outsize; /* Output size. */ 1168 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1169}; 1170 1171TAILQ_HEAD(note_info_list, note_info); 1172 1173/* Coredump output parameters. */ 1174struct coredump_params { 1175 off_t offset; 1176 struct ucred *active_cred; 1177 struct ucred *file_cred; 1178 struct thread *td; 1179 struct vnode *vp; 1180 struct gzio_stream *gzs; 1181}; 1182 1183static void cb_put_phdr(vm_map_entry_t, void *); 1184static void cb_size_segment(vm_map_entry_t, void *); 1185static int core_write(struct coredump_params *, const void *, size_t, off_t, 1186 enum uio_seg); 1187static void each_writable_segment(struct thread *, segment_callback, void *); 1188static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1189 struct note_info_list *, size_t); 1190static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, 1191 size_t *); 1192static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); 1193static void __elfN(putnote)(struct note_info *, struct sbuf *); 1194static size_t register_note(struct note_info_list *, int, outfunc_t, void *); 1195static int sbuf_drain_core_output(void *, const char *, int); 1196static int sbuf_drain_count(void *arg, const char *data, int len); 1197 1198static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); 1199static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1200static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); 1201static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1202static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); 1203static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1204static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1205static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1206static void note_procstat_files(void *, struct sbuf *, size_t *); 1207static void note_procstat_groups(void *, struct sbuf *, size_t *); 1208static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1209static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1210static void note_procstat_umask(void *, struct sbuf *, size_t *); 1211static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1212 1213#ifdef GZIO 1214extern int compress_user_cores_gzlevel; 1215 1216/* 1217 * Write out a core segment to the compression stream. 1218 */ 1219static int 1220compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len) 1221{ 1222 u_int chunk_len; 1223 int error; 1224 1225 while (len > 0) { 1226 chunk_len = MIN(len, CORE_BUF_SIZE); 1227 1228 /* 1229 * We can get EFAULT error here. 1230 * In that case zero out the current chunk of the segment. 1231 */ 1232 error = copyin(base, buf, chunk_len); 1233 if (error != 0) 1234 bzero(buf, chunk_len); 1235 error = gzio_write(p->gzs, buf, chunk_len); 1236 if (error != 0) 1237 break; 1238 base += chunk_len; 1239 len -= chunk_len; 1240 } 1241 return (error); 1242} 1243 1244static int 1245core_gz_write(void *base, size_t len, off_t offset, void *arg) 1246{ 1247 1248 return (core_write((struct coredump_params *)arg, base, len, offset, 1249 UIO_SYSSPACE)); 1250} 1251#endif /* GZIO */ 1252 1253static int 1254core_write(struct coredump_params *p, const void *base, size_t len, 1255 off_t offset, enum uio_seg seg) 1256{ 1257 1258 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base), 1259 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1260 p->active_cred, p->file_cred, NULL, p->td)); 1261} 1262 1263static int 1264core_output(void *base, size_t len, off_t offset, struct coredump_params *p, 1265 void *tmpbuf) 1266{ 1267 int error; 1268 1269#ifdef GZIO 1270 if (p->gzs != NULL) 1271 return (compress_chunk(p, base, tmpbuf, len)); 1272#endif 1273 /* 1274 * EFAULT is a non-fatal error that we can get, for example, 1275 * if the segment is backed by a file but extends beyond its 1276 * end. 1277 */ 1278 error = core_write(p, base, len, offset, UIO_USERSPACE); 1279 if (error == EFAULT) { 1280 log(LOG_WARNING, "Failed to fully fault in a core file segment " 1281 "at VA %p with size 0x%zx to be written at offset 0x%jx " 1282 "for process %s\n", base, len, offset, curproc->p_comm); 1283 1284 /* 1285 * Write a "real" zero byte at the end of the target region 1286 * in the case this is the last segment. 1287 * The intermediate space will be implicitly zero-filled. 1288 */ 1289 error = core_write(p, zero_region, 1, offset + len - 1, 1290 UIO_SYSSPACE); 1291 } 1292 return (error); 1293} 1294 1295/* 1296 * Drain into a core file. 1297 */ 1298static int 1299sbuf_drain_core_output(void *arg, const char *data, int len) 1300{ 1301 struct coredump_params *p; 1302 int error, locked; 1303 1304 p = (struct coredump_params *)arg; 1305 1306 /* 1307 * Some kern_proc out routines that print to this sbuf may 1308 * call us with the process lock held. Draining with the 1309 * non-sleepable lock held is unsafe. The lock is needed for 1310 * those routines when dumping a live process. In our case we 1311 * can safely release the lock before draining and acquire 1312 * again after. 1313 */ 1314 locked = PROC_LOCKED(p->td->td_proc); 1315 if (locked) 1316 PROC_UNLOCK(p->td->td_proc); 1317#ifdef GZIO 1318 if (p->gzs != NULL) 1319 error = gzio_write(p->gzs, __DECONST(char *, data), len); 1320 else 1321#endif 1322 error = core_write(p, __DECONST(void *, data), len, p->offset, 1323 UIO_SYSSPACE); 1324 if (locked) 1325 PROC_LOCK(p->td->td_proc); 1326 if (error != 0) 1327 return (-error); 1328 p->offset += len; 1329 return (len); 1330} 1331 1332/* 1333 * Drain into a counter. 1334 */ 1335static int 1336sbuf_drain_count(void *arg, const char *data __unused, int len) 1337{ 1338 size_t *sizep; 1339 1340 sizep = (size_t *)arg; 1341 *sizep += len; 1342 return (len); 1343} 1344 1345int 1346__elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1347{ 1348 struct ucred *cred = td->td_ucred; 1349 int error = 0; 1350 struct sseg_closure seginfo; 1351 struct note_info_list notelst; 1352 struct coredump_params params; 1353 struct note_info *ninfo; 1354 void *hdr, *tmpbuf; 1355 size_t hdrsize, notesz, coresize; 1356#ifdef GZIO 1357 boolean_t compress; 1358 1359 compress = (flags & IMGACT_CORE_COMPRESS) != 0; 1360#endif 1361 hdr = NULL; 1362 tmpbuf = NULL; 1363 TAILQ_INIT(¬elst); 1364 1365 /* Size the program segments. */ 1366 seginfo.count = 0; 1367 seginfo.size = 0; 1368 each_writable_segment(td, cb_size_segment, &seginfo); 1369 1370 /* 1371 * Collect info about the core file header area. 1372 */ 1373 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1374 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1375 coresize = round_page(hdrsize + notesz) + seginfo.size; 1376 1377 /* Set up core dump parameters. */ 1378 params.offset = 0; 1379 params.active_cred = cred; 1380 params.file_cred = NOCRED; 1381 params.td = td; 1382 params.vp = vp; 1383 params.gzs = NULL; 1384 1385#ifdef RACCT 1386 if (racct_enable) { 1387 PROC_LOCK(td->td_proc); 1388 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1389 PROC_UNLOCK(td->td_proc); 1390 if (error != 0) { 1391 error = EFAULT; 1392 goto done; 1393 } 1394 } 1395#endif 1396 if (coresize >= limit) { 1397 error = EFAULT; 1398 goto done; 1399 } 1400 1401#ifdef GZIO 1402 /* Create a compression stream if necessary. */ 1403 if (compress) { 1404 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE, 1405 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms); 1406 if (params.gzs == NULL) { 1407 error = EFAULT; 1408 goto done; 1409 } 1410 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1411 } 1412#endif 1413 1414 /* 1415 * Allocate memory for building the header, fill it up, 1416 * and write it out following the notes. 1417 */ 1418 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1419 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1420 notesz); 1421 1422 /* Write the contents of all of the writable segments. */ 1423 if (error == 0) { 1424 Elf_Phdr *php; 1425 off_t offset; 1426 int i; 1427 1428 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1429 offset = round_page(hdrsize + notesz); 1430 for (i = 0; i < seginfo.count; i++) { 1431 error = core_output((caddr_t)(uintptr_t)php->p_vaddr, 1432 php->p_filesz, offset, ¶ms, tmpbuf); 1433 if (error != 0) 1434 break; 1435 offset += php->p_filesz; 1436 php++; 1437 } 1438#ifdef GZIO 1439 if (error == 0 && compress) 1440 error = gzio_flush(params.gzs); 1441#endif 1442 } 1443 if (error) { 1444 log(LOG_WARNING, 1445 "Failed to write core file for process %s (error %d)\n", 1446 curproc->p_comm, error); 1447 } 1448 1449done: 1450#ifdef GZIO 1451 if (compress) { 1452 free(tmpbuf, M_TEMP); 1453 if (params.gzs != NULL) 1454 gzio_fini(params.gzs); 1455 } 1456#endif 1457 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1458 TAILQ_REMOVE(¬elst, ninfo, link); 1459 free(ninfo, M_TEMP); 1460 } 1461 if (hdr != NULL) 1462 free(hdr, M_TEMP); 1463 1464 return (error); 1465} 1466 1467/* 1468 * A callback for each_writable_segment() to write out the segment's 1469 * program header entry. 1470 */ 1471static void 1472cb_put_phdr(entry, closure) 1473 vm_map_entry_t entry; 1474 void *closure; 1475{ 1476 struct phdr_closure *phc = (struct phdr_closure *)closure; 1477 Elf_Phdr *phdr = phc->phdr; 1478 1479 phc->offset = round_page(phc->offset); 1480 1481 phdr->p_type = PT_LOAD; 1482 phdr->p_offset = phc->offset; 1483 phdr->p_vaddr = entry->start; 1484 phdr->p_paddr = 0; 1485 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1486 phdr->p_align = PAGE_SIZE; 1487 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1488 1489 phc->offset += phdr->p_filesz; 1490 phc->phdr++; 1491} 1492 1493/* 1494 * A callback for each_writable_segment() to gather information about 1495 * the number of segments and their total size. 1496 */ 1497static void 1498cb_size_segment(entry, closure) 1499 vm_map_entry_t entry; 1500 void *closure; 1501{ 1502 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1503 1504 ssc->count++; 1505 ssc->size += entry->end - entry->start; 1506} 1507 1508/* 1509 * For each writable segment in the process's memory map, call the given 1510 * function with a pointer to the map entry and some arbitrary 1511 * caller-supplied data. 1512 */ 1513static void 1514each_writable_segment(td, func, closure) 1515 struct thread *td; 1516 segment_callback func; 1517 void *closure; 1518{ 1519 struct proc *p = td->td_proc; 1520 vm_map_t map = &p->p_vmspace->vm_map; 1521 vm_map_entry_t entry; 1522 vm_object_t backing_object, object; 1523 boolean_t ignore_entry; 1524 1525 vm_map_lock_read(map); 1526 for (entry = map->header.next; entry != &map->header; 1527 entry = entry->next) { 1528 /* 1529 * Don't dump inaccessible mappings, deal with legacy 1530 * coredump mode. 1531 * 1532 * Note that read-only segments related to the elf binary 1533 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1534 * need to arbitrarily ignore such segments. 1535 */ 1536 if (elf_legacy_coredump) { 1537 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1538 continue; 1539 } else { 1540 if ((entry->protection & VM_PROT_ALL) == 0) 1541 continue; 1542 } 1543 1544 /* 1545 * Dont include memory segment in the coredump if 1546 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1547 * madvise(2). Do not dump submaps (i.e. parts of the 1548 * kernel map). 1549 */ 1550 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1551 continue; 1552 1553 if ((object = entry->object.vm_object) == NULL) 1554 continue; 1555 1556 /* Ignore memory-mapped devices and such things. */ 1557 VM_OBJECT_RLOCK(object); 1558 while ((backing_object = object->backing_object) != NULL) { 1559 VM_OBJECT_RLOCK(backing_object); 1560 VM_OBJECT_RUNLOCK(object); 1561 object = backing_object; 1562 } 1563 ignore_entry = object->type != OBJT_DEFAULT && 1564 object->type != OBJT_SWAP && object->type != OBJT_VNODE && 1565 object->type != OBJT_PHYS; 1566 VM_OBJECT_RUNLOCK(object); 1567 if (ignore_entry) 1568 continue; 1569 1570 (*func)(entry, closure); 1571 } 1572 vm_map_unlock_read(map); 1573} 1574 1575/* 1576 * Write the core file header to the file, including padding up to 1577 * the page boundary. 1578 */ 1579static int 1580__elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1581 size_t hdrsize, struct note_info_list *notelst, size_t notesz) 1582{ 1583 struct note_info *ninfo; 1584 struct sbuf *sb; 1585 int error; 1586 1587 /* Fill in the header. */ 1588 bzero(hdr, hdrsize); 1589 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz); 1590 1591 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1592 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1593 sbuf_start_section(sb, NULL); 1594 sbuf_bcat(sb, hdr, hdrsize); 1595 TAILQ_FOREACH(ninfo, notelst, link) 1596 __elfN(putnote)(ninfo, sb); 1597 /* Align up to a page boundary for the program segments. */ 1598 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1599 error = sbuf_finish(sb); 1600 sbuf_delete(sb); 1601 1602 return (error); 1603} 1604 1605static void 1606__elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1607 size_t *sizep) 1608{ 1609 struct proc *p; 1610 struct thread *thr; 1611 size_t size; 1612 1613 p = td->td_proc; 1614 size = 0; 1615 1616 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1617 1618 /* 1619 * To have the debugger select the right thread (LWP) as the initial 1620 * thread, we dump the state of the thread passed to us in td first. 1621 * This is the thread that causes the core dump and thus likely to 1622 * be the right thread one wants to have selected in the debugger. 1623 */ 1624 thr = td; 1625 while (thr != NULL) { 1626 size += register_note(list, NT_PRSTATUS, 1627 __elfN(note_prstatus), thr); 1628 size += register_note(list, NT_FPREGSET, 1629 __elfN(note_fpregset), thr); 1630 size += register_note(list, NT_THRMISC, 1631 __elfN(note_thrmisc), thr); 1632 size += register_note(list, -1, 1633 __elfN(note_threadmd), thr); 1634 1635 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1636 TAILQ_NEXT(thr, td_plist); 1637 if (thr == td) 1638 thr = TAILQ_NEXT(thr, td_plist); 1639 } 1640 1641 size += register_note(list, NT_PROCSTAT_PROC, 1642 __elfN(note_procstat_proc), p); 1643 size += register_note(list, NT_PROCSTAT_FILES, 1644 note_procstat_files, p); 1645 size += register_note(list, NT_PROCSTAT_VMMAP, 1646 note_procstat_vmmap, p); 1647 size += register_note(list, NT_PROCSTAT_GROUPS, 1648 note_procstat_groups, p); 1649 size += register_note(list, NT_PROCSTAT_UMASK, 1650 note_procstat_umask, p); 1651 size += register_note(list, NT_PROCSTAT_RLIMIT, 1652 note_procstat_rlimit, p); 1653 size += register_note(list, NT_PROCSTAT_OSREL, 1654 note_procstat_osrel, p); 1655 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1656 __elfN(note_procstat_psstrings), p); 1657 size += register_note(list, NT_PROCSTAT_AUXV, 1658 __elfN(note_procstat_auxv), p); 1659 1660 *sizep = size; 1661} 1662 1663static void 1664__elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1665 size_t notesz) 1666{ 1667 Elf_Ehdr *ehdr; 1668 Elf_Phdr *phdr; 1669 struct phdr_closure phc; 1670 1671 ehdr = (Elf_Ehdr *)hdr; 1672 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)); 1673 1674 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1675 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1676 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1677 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1678 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1679 ehdr->e_ident[EI_DATA] = ELF_DATA; 1680 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1681 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1682 ehdr->e_ident[EI_ABIVERSION] = 0; 1683 ehdr->e_ident[EI_PAD] = 0; 1684 ehdr->e_type = ET_CORE; 1685#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1686 ehdr->e_machine = ELF_ARCH32; 1687#else 1688 ehdr->e_machine = ELF_ARCH; 1689#endif 1690 ehdr->e_version = EV_CURRENT; 1691 ehdr->e_entry = 0; 1692 ehdr->e_phoff = sizeof(Elf_Ehdr); 1693 ehdr->e_flags = 0; 1694 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1695 ehdr->e_phentsize = sizeof(Elf_Phdr); 1696 ehdr->e_phnum = numsegs + 1; 1697 ehdr->e_shentsize = sizeof(Elf_Shdr); 1698 ehdr->e_shnum = 0; 1699 ehdr->e_shstrndx = SHN_UNDEF; 1700 1701 /* 1702 * Fill in the program header entries. 1703 */ 1704 1705 /* The note segement. */ 1706 phdr->p_type = PT_NOTE; 1707 phdr->p_offset = hdrsize; 1708 phdr->p_vaddr = 0; 1709 phdr->p_paddr = 0; 1710 phdr->p_filesz = notesz; 1711 phdr->p_memsz = 0; 1712 phdr->p_flags = PF_R; 1713 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1714 phdr++; 1715 1716 /* All the writable segments from the program. */ 1717 phc.phdr = phdr; 1718 phc.offset = round_page(hdrsize + notesz); 1719 each_writable_segment(td, cb_put_phdr, &phc); 1720} 1721 1722static size_t 1723register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) 1724{ 1725 struct note_info *ninfo; 1726 size_t size, notesize; 1727 1728 size = 0; 1729 out(arg, NULL, &size); 1730 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1731 ninfo->type = type; 1732 ninfo->outfunc = out; 1733 ninfo->outarg = arg; 1734 ninfo->outsize = size; 1735 TAILQ_INSERT_TAIL(list, ninfo, link); 1736 1737 if (type == -1) 1738 return (size); 1739 1740 notesize = sizeof(Elf_Note) + /* note header */ 1741 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1742 /* note name */ 1743 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1744 1745 return (notesize); 1746} 1747 1748static size_t 1749append_note_data(const void *src, void *dst, size_t len) 1750{ 1751 size_t padded_len; 1752 1753 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 1754 if (dst != NULL) { 1755 bcopy(src, dst, len); 1756 bzero((char *)dst + len, padded_len - len); 1757 } 1758 return (padded_len); 1759} 1760 1761size_t 1762__elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 1763{ 1764 Elf_Note *note; 1765 char *buf; 1766 size_t notesize; 1767 1768 buf = dst; 1769 if (buf != NULL) { 1770 note = (Elf_Note *)buf; 1771 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1772 note->n_descsz = size; 1773 note->n_type = type; 1774 buf += sizeof(*note); 1775 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 1776 sizeof(FREEBSD_ABI_VENDOR)); 1777 append_note_data(src, buf, size); 1778 if (descp != NULL) 1779 *descp = buf; 1780 } 1781 1782 notesize = sizeof(Elf_Note) + /* note header */ 1783 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1784 /* note name */ 1785 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1786 1787 return (notesize); 1788} 1789 1790static void 1791__elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) 1792{ 1793 Elf_Note note; 1794 ssize_t old_len, sect_len; 1795 size_t new_len, descsz, i; 1796 1797 if (ninfo->type == -1) { 1798 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1799 return; 1800 } 1801 1802 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1803 note.n_descsz = ninfo->outsize; 1804 note.n_type = ninfo->type; 1805 1806 sbuf_bcat(sb, ¬e, sizeof(note)); 1807 sbuf_start_section(sb, &old_len); 1808 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR)); 1809 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1810 if (note.n_descsz == 0) 1811 return; 1812 sbuf_start_section(sb, &old_len); 1813 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1814 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1815 if (sect_len < 0) 1816 return; 1817 1818 new_len = (size_t)sect_len; 1819 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 1820 if (new_len < descsz) { 1821 /* 1822 * It is expected that individual note emitters will correctly 1823 * predict their expected output size and fill up to that size 1824 * themselves, padding in a format-specific way if needed. 1825 * However, in case they don't, just do it here with zeros. 1826 */ 1827 for (i = 0; i < descsz - new_len; i++) 1828 sbuf_putc(sb, 0); 1829 } else if (new_len > descsz) { 1830 /* 1831 * We can't always truncate sb -- we may have drained some 1832 * of it already. 1833 */ 1834 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 1835 "read it (%zu > %zu). Since it is longer than " 1836 "expected, this coredump's notes are corrupt. THIS " 1837 "IS A BUG in the note_procstat routine for type %u.\n", 1838 __func__, (unsigned)note.n_type, new_len, descsz, 1839 (unsigned)note.n_type)); 1840 } 1841} 1842 1843/* 1844 * Miscellaneous note out functions. 1845 */ 1846 1847#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1848#include <compat/freebsd32/freebsd32.h> 1849 1850typedef struct prstatus32 elf_prstatus_t; 1851typedef struct prpsinfo32 elf_prpsinfo_t; 1852typedef struct fpreg32 elf_prfpregset_t; 1853typedef struct fpreg32 elf_fpregset_t; 1854typedef struct reg32 elf_gregset_t; 1855typedef struct thrmisc32 elf_thrmisc_t; 1856#define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1857typedef struct kinfo_proc32 elf_kinfo_proc_t; 1858typedef uint32_t elf_ps_strings_t; 1859#else 1860typedef prstatus_t elf_prstatus_t; 1861typedef prpsinfo_t elf_prpsinfo_t; 1862typedef prfpregset_t elf_prfpregset_t; 1863typedef prfpregset_t elf_fpregset_t; 1864typedef gregset_t elf_gregset_t; 1865typedef thrmisc_t elf_thrmisc_t; 1866#define ELF_KERN_PROC_MASK 0 1867typedef struct kinfo_proc elf_kinfo_proc_t; 1868typedef vm_offset_t elf_ps_strings_t; 1869#endif 1870 1871static void 1872__elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 1873{ 1874 struct sbuf sbarg; 1875 size_t len; 1876 char *cp, *end; 1877 struct proc *p; 1878 elf_prpsinfo_t *psinfo; 1879 int error; 1880 1881 p = (struct proc *)arg; 1882 if (sb != NULL) { 1883 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 1884 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 1885 psinfo->pr_version = PRPSINFO_VERSION; 1886 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1887 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1888 PROC_LOCK(p); 1889 if (p->p_args != NULL) { 1890 len = sizeof(psinfo->pr_psargs) - 1; 1891 if (len > p->p_args->ar_length) 1892 len = p->p_args->ar_length; 1893 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 1894 PROC_UNLOCK(p); 1895 error = 0; 1896 } else { 1897 _PHOLD(p); 1898 PROC_UNLOCK(p); 1899 sbuf_new(&sbarg, psinfo->pr_psargs, 1900 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 1901 error = proc_getargv(curthread, p, &sbarg); 1902 PRELE(p); 1903 if (sbuf_finish(&sbarg) == 0) 1904 len = sbuf_len(&sbarg) - 1; 1905 else 1906 len = sizeof(psinfo->pr_psargs) - 1; 1907 sbuf_delete(&sbarg); 1908 } 1909 if (error || len == 0) 1910 strlcpy(psinfo->pr_psargs, p->p_comm, 1911 sizeof(psinfo->pr_psargs)); 1912 else { 1913 KASSERT(len < sizeof(psinfo->pr_psargs), 1914 ("len is too long: %zu vs %zu", len, 1915 sizeof(psinfo->pr_psargs))); 1916 cp = psinfo->pr_psargs; 1917 end = cp + len - 1; 1918 for (;;) { 1919 cp = memchr(cp, '\0', end - cp); 1920 if (cp == NULL) 1921 break; 1922 *cp = ' '; 1923 } 1924 } 1925 psinfo->pr_pid = p->p_pid; 1926 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 1927 free(psinfo, M_TEMP); 1928 } 1929 *sizep = sizeof(*psinfo); 1930} 1931 1932static void 1933__elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) 1934{ 1935 struct thread *td; 1936 elf_prstatus_t *status; 1937 1938 td = (struct thread *)arg; 1939 if (sb != NULL) { 1940 KASSERT(*sizep == sizeof(*status), ("invalid size")); 1941 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); 1942 status->pr_version = PRSTATUS_VERSION; 1943 status->pr_statussz = sizeof(elf_prstatus_t); 1944 status->pr_gregsetsz = sizeof(elf_gregset_t); 1945 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1946 status->pr_osreldate = osreldate; 1947 status->pr_cursig = td->td_proc->p_sig; 1948 status->pr_pid = td->td_tid; 1949#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1950 fill_regs32(td, &status->pr_reg); 1951#else 1952 fill_regs(td, &status->pr_reg); 1953#endif 1954 sbuf_bcat(sb, status, sizeof(*status)); 1955 free(status, M_TEMP); 1956 } 1957 *sizep = sizeof(*status); 1958} 1959 1960static void 1961__elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) 1962{ 1963 struct thread *td; 1964 elf_prfpregset_t *fpregset; 1965 1966 td = (struct thread *)arg; 1967 if (sb != NULL) { 1968 KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); 1969 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); 1970#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1971 fill_fpregs32(td, fpregset); 1972#else 1973 fill_fpregs(td, fpregset); 1974#endif 1975 sbuf_bcat(sb, fpregset, sizeof(*fpregset)); 1976 free(fpregset, M_TEMP); 1977 } 1978 *sizep = sizeof(*fpregset); 1979} 1980 1981static void 1982__elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) 1983{ 1984 struct thread *td; 1985 elf_thrmisc_t thrmisc; 1986 1987 td = (struct thread *)arg; 1988 if (sb != NULL) { 1989 KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); 1990 bzero(&thrmisc._pad, sizeof(thrmisc._pad)); 1991 strcpy(thrmisc.pr_tname, td->td_name); 1992 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); 1993 } 1994 *sizep = sizeof(thrmisc); 1995} 1996 1997/* 1998 * Allow for MD specific notes, as well as any MD 1999 * specific preparations for writing MI notes. 2000 */ 2001static void 2002__elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 2003{ 2004 struct thread *td; 2005 void *buf; 2006 size_t size; 2007 2008 td = (struct thread *)arg; 2009 size = *sizep; 2010 if (size != 0 && sb != NULL) 2011 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 2012 else 2013 buf = NULL; 2014 size = 0; 2015 __elfN(dump_thread)(td, buf, &size); 2016 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 2017 if (size != 0 && sb != NULL) 2018 sbuf_bcat(sb, buf, size); 2019 free(buf, M_TEMP); 2020 *sizep = size; 2021} 2022 2023#ifdef KINFO_PROC_SIZE 2024CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2025#endif 2026 2027static void 2028__elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2029{ 2030 struct proc *p; 2031 size_t size; 2032 int structsize; 2033 2034 p = (struct proc *)arg; 2035 size = sizeof(structsize) + p->p_numthreads * 2036 sizeof(elf_kinfo_proc_t); 2037 2038 if (sb != NULL) { 2039 KASSERT(*sizep == size, ("invalid size")); 2040 structsize = sizeof(elf_kinfo_proc_t); 2041 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2042 sx_slock(&proctree_lock); 2043 PROC_LOCK(p); 2044 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2045 sx_sunlock(&proctree_lock); 2046 } 2047 *sizep = size; 2048} 2049 2050#ifdef KINFO_FILE_SIZE 2051CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2052#endif 2053 2054static void 2055note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2056{ 2057 struct proc *p; 2058 size_t size, sect_sz, i; 2059 ssize_t start_len, sect_len; 2060 int structsize, filedesc_flags; 2061 2062 if (coredump_pack_fileinfo) 2063 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2064 else 2065 filedesc_flags = 0; 2066 2067 p = (struct proc *)arg; 2068 structsize = sizeof(struct kinfo_file); 2069 if (sb == NULL) { 2070 size = 0; 2071 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2072 sbuf_set_drain(sb, sbuf_drain_count, &size); 2073 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2074 PROC_LOCK(p); 2075 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2076 sbuf_finish(sb); 2077 sbuf_delete(sb); 2078 *sizep = size; 2079 } else { 2080 sbuf_start_section(sb, &start_len); 2081 2082 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2083 PROC_LOCK(p); 2084 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2085 filedesc_flags); 2086 2087 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2088 if (sect_len < 0) 2089 return; 2090 sect_sz = sect_len; 2091 2092 KASSERT(sect_sz <= *sizep, 2093 ("kern_proc_filedesc_out did not respect maxlen; " 2094 "requested %zu, got %zu", *sizep - sizeof(structsize), 2095 sect_sz - sizeof(structsize))); 2096 2097 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2098 sbuf_putc(sb, 0); 2099 } 2100} 2101 2102#ifdef KINFO_VMENTRY_SIZE 2103CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2104#endif 2105 2106static void 2107note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2108{ 2109 struct proc *p; 2110 size_t size; 2111 int structsize, vmmap_flags; 2112 2113 if (coredump_pack_vmmapinfo) 2114 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2115 else 2116 vmmap_flags = 0; 2117 2118 p = (struct proc *)arg; 2119 structsize = sizeof(struct kinfo_vmentry); 2120 if (sb == NULL) { 2121 size = 0; 2122 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2123 sbuf_set_drain(sb, sbuf_drain_count, &size); 2124 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2125 PROC_LOCK(p); 2126 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2127 sbuf_finish(sb); 2128 sbuf_delete(sb); 2129 *sizep = size; 2130 } else { 2131 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2132 PROC_LOCK(p); 2133 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2134 vmmap_flags); 2135 } 2136} 2137 2138static void 2139note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2140{ 2141 struct proc *p; 2142 size_t size; 2143 int structsize; 2144 2145 p = (struct proc *)arg; 2146 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2147 if (sb != NULL) { 2148 KASSERT(*sizep == size, ("invalid size")); 2149 structsize = sizeof(gid_t); 2150 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2151 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2152 sizeof(gid_t)); 2153 } 2154 *sizep = size; 2155} 2156 2157static void 2158note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2159{ 2160 struct proc *p; 2161 size_t size; 2162 int structsize; 2163 2164 p = (struct proc *)arg; 2165 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); 2166 if (sb != NULL) { 2167 KASSERT(*sizep == size, ("invalid size")); 2168 structsize = sizeof(p->p_fd->fd_cmask); 2169 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2170 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); 2171 } 2172 *sizep = size; 2173} 2174 2175static void 2176note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2177{ 2178 struct proc *p; 2179 struct rlimit rlim[RLIM_NLIMITS]; 2180 size_t size; 2181 int structsize, i; 2182 2183 p = (struct proc *)arg; 2184 size = sizeof(structsize) + sizeof(rlim); 2185 if (sb != NULL) { 2186 KASSERT(*sizep == size, ("invalid size")); 2187 structsize = sizeof(rlim); 2188 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2189 PROC_LOCK(p); 2190 for (i = 0; i < RLIM_NLIMITS; i++) 2191 lim_rlimit_proc(p, i, &rlim[i]); 2192 PROC_UNLOCK(p); 2193 sbuf_bcat(sb, rlim, sizeof(rlim)); 2194 } 2195 *sizep = size; 2196} 2197 2198static void 2199note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2200{ 2201 struct proc *p; 2202 size_t size; 2203 int structsize; 2204 2205 p = (struct proc *)arg; 2206 size = sizeof(structsize) + sizeof(p->p_osrel); 2207 if (sb != NULL) { 2208 KASSERT(*sizep == size, ("invalid size")); 2209 structsize = sizeof(p->p_osrel); 2210 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2211 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2212 } 2213 *sizep = size; 2214} 2215 2216static void 2217__elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2218{ 2219 struct proc *p; 2220 elf_ps_strings_t ps_strings; 2221 size_t size; 2222 int structsize; 2223 2224 p = (struct proc *)arg; 2225 size = sizeof(structsize) + sizeof(ps_strings); 2226 if (sb != NULL) { 2227 KASSERT(*sizep == size, ("invalid size")); 2228 structsize = sizeof(ps_strings); 2229#if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2230 ps_strings = PTROUT(p->p_sysent->sv_psstrings); 2231#else 2232 ps_strings = p->p_sysent->sv_psstrings; 2233#endif 2234 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2235 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2236 } 2237 *sizep = size; 2238} 2239 2240static void 2241__elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2242{ 2243 struct proc *p; 2244 size_t size; 2245 int structsize; 2246 2247 p = (struct proc *)arg; 2248 if (sb == NULL) { 2249 size = 0; 2250 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2251 sbuf_set_drain(sb, sbuf_drain_count, &size); 2252 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2253 PHOLD(p); 2254 proc_getauxv(curthread, p, sb); 2255 PRELE(p); 2256 sbuf_finish(sb); 2257 sbuf_delete(sb); 2258 *sizep = size; 2259 } else { 2260 structsize = sizeof(Elf_Auxinfo); 2261 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2262 PHOLD(p); 2263 proc_getauxv(curthread, p, sb); 2264 PRELE(p); 2265 } 2266} 2267 2268static boolean_t 2269__elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 2270 int32_t *osrel, const Elf_Phdr *pnote) 2271{ 2272 const Elf_Note *note, *note0, *note_end; 2273 const char *note_name; 2274 char *buf; 2275 int i, error; 2276 boolean_t res; 2277 2278 /* We need some limit, might as well use PAGE_SIZE. */ 2279 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2280 return (FALSE); 2281 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2282 if (pnote->p_offset > PAGE_SIZE || 2283 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2284 VOP_UNLOCK(imgp->vp, 0); 2285 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2286 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 2287 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2288 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2289 curthread->td_ucred, NOCRED, NULL, curthread); 2290 if (error != 0) { 2291 uprintf("i/o error PT_NOTE\n"); 2292 res = FALSE; 2293 goto ret; 2294 } 2295 note = note0 = (const Elf_Note *)buf; 2296 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2297 } else { 2298 note = note0 = (const Elf_Note *)(imgp->image_header + 2299 pnote->p_offset); 2300 note_end = (const Elf_Note *)(imgp->image_header + 2301 pnote->p_offset + pnote->p_filesz); 2302 buf = NULL; 2303 } 2304 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 2305 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 2306 (const char *)note < sizeof(Elf_Note)) { 2307 res = FALSE; 2308 goto ret; 2309 } 2310 if (note->n_namesz != checknote->hdr.n_namesz || 2311 note->n_descsz != checknote->hdr.n_descsz || 2312 note->n_type != checknote->hdr.n_type) 2313 goto nextnote; 2314 note_name = (const char *)(note + 1); 2315 if (note_name + checknote->hdr.n_namesz >= 2316 (const char *)note_end || strncmp(checknote->vendor, 2317 note_name, checknote->hdr.n_namesz) != 0) 2318 goto nextnote; 2319 2320 /* 2321 * Fetch the osreldate for binary 2322 * from the ELF OSABI-note if necessary. 2323 */ 2324 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 2325 checknote->trans_osrel != NULL) { 2326 res = checknote->trans_osrel(note, osrel); 2327 goto ret; 2328 } 2329 res = TRUE; 2330 goto ret; 2331nextnote: 2332 note = (const Elf_Note *)((const char *)(note + 1) + 2333 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2334 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2335 } 2336 res = FALSE; 2337ret: 2338 free(buf, M_TEMP); 2339 return (res); 2340} 2341 2342/* 2343 * Try to find the appropriate ABI-note section for checknote, 2344 * fetch the osreldate for binary from the ELF OSABI-note. Only the 2345 * first page of the image is searched, the same as for headers. 2346 */ 2347static boolean_t 2348__elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 2349 int32_t *osrel) 2350{ 2351 const Elf_Phdr *phdr; 2352 const Elf_Ehdr *hdr; 2353 int i; 2354 2355 hdr = (const Elf_Ehdr *)imgp->image_header; 2356 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2357 2358 for (i = 0; i < hdr->e_phnum; i++) { 2359 if (phdr[i].p_type == PT_NOTE && 2360 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 2361 return (TRUE); 2362 } 2363 return (FALSE); 2364 2365} 2366 2367/* 2368 * Tell kern_execve.c about it, with a little help from the linker. 2369 */ 2370static struct execsw __elfN(execsw) = { 2371 __CONCAT(exec_, __elfN(imgact)), 2372 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2373}; 2374EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2375 2376static vm_prot_t 2377__elfN(trans_prot)(Elf_Word flags) 2378{ 2379 vm_prot_t prot; 2380 2381 prot = 0; 2382 if (flags & PF_X) 2383 prot |= VM_PROT_EXECUTE; 2384 if (flags & PF_W) 2385 prot |= VM_PROT_WRITE; 2386 if (flags & PF_R) 2387 prot |= VM_PROT_READ; 2388#if __ELF_WORD_SIZE == 32 2389#if defined(__amd64__) 2390 if (i386_read_exec && (flags & PF_R)) 2391 prot |= VM_PROT_EXECUTE; 2392#endif 2393#endif 2394 return (prot); 2395} 2396 2397static Elf_Word 2398__elfN(untrans_prot)(vm_prot_t prot) 2399{ 2400 Elf_Word flags; 2401 2402 flags = 0; 2403 if (prot & VM_PROT_EXECUTE) 2404 flags |= PF_X; 2405 if (prot & VM_PROT_READ) 2406 flags |= PF_R; 2407 if (prot & VM_PROT_WRITE) 2408 flags |= PF_W; 2409 return (flags); 2410} 2411