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_offset_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 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 313 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) { 314 /* Looks good, but give brand a chance to veto */ 315 if (!bi->header_supported || bi->header_supported(imgp)) 316 return (bi); 317 } 318 } 319 320 /* No known brand, see if the header is recognized by any brand */ 321 for (i = 0; i < MAX_BRANDS; i++) { 322 bi = elf_brand_list[i]; 323 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || 324 bi->header_supported == NULL) 325 continue; 326 if (hdr->e_machine == bi->machine) { 327 ret = bi->header_supported(imgp); 328 if (ret) 329 return (bi); 330 } 331 } 332 333 /* Lacking a known brand, search for a recognized interpreter. */ 334 if (interp != NULL) { 335 for (i = 0; i < MAX_BRANDS; i++) { 336 bi = elf_brand_list[i]; 337 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 338 continue; 339 if (hdr->e_machine == bi->machine && 340 /* ELF image p_filesz includes terminating zero */ 341 strlen(bi->interp_path) + 1 == interp_name_len && 342 strncmp(interp, bi->interp_path, interp_name_len) 343 == 0) 344 return (bi); 345 } 346 } 347 348 /* Lacking a recognized interpreter, try the default brand */ 349 for (i = 0; i < MAX_BRANDS; i++) { 350 bi = elf_brand_list[i]; 351 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) 352 continue; 353 if (hdr->e_machine == bi->machine && 354 __elfN(fallback_brand) == bi->brand) 355 return (bi); 356 } 357 return (NULL); 358 } 359 360 static int 361 __elfN(check_header)(const Elf_Ehdr *hdr) 362 { 363 Elf_Brandinfo *bi; 364 int i; 365 366 if (!IS_ELF(*hdr) || 367 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 368 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 369 hdr->e_ident[EI_VERSION] != EV_CURRENT || 370 hdr->e_phentsize != sizeof(Elf_Phdr) || 371 hdr->e_version != ELF_TARG_VER) 372 return (ENOEXEC); 373 374 /* 375 * Make sure we have at least one brand for this machine. 376 */ 377 378 for (i = 0; i < MAX_BRANDS; i++) { 379 bi = elf_brand_list[i]; 380 if (bi != NULL && bi->machine == hdr->e_machine) 381 break; 382 } 383 if (i == MAX_BRANDS) 384 return (ENOEXEC); 385 386 return (0); 387 } 388 389 static int 390 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 391 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 392 { 393 struct sf_buf *sf; 394 int error; 395 vm_offset_t off; 396 397 /* 398 * Create the page if it doesn't exist yet. Ignore errors. 399 */ 400 vm_map_lock(map); 401 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 402 VM_PROT_ALL, VM_PROT_ALL, 0); 403 vm_map_unlock(map); 404 405 /* 406 * Find the page from the underlying object. 407 */ 408 if (object) { 409 sf = vm_imgact_map_page(object, offset); 410 if (sf == NULL) 411 return (KERN_FAILURE); 412 off = offset - trunc_page(offset); 413 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 414 end - start); 415 vm_imgact_unmap_page(sf); 416 if (error) { 417 return (KERN_FAILURE); 418 } 419 } 420 421 return (KERN_SUCCESS); 422 } 423 424 static int 425 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 426 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 427 { 428 struct sf_buf *sf; 429 vm_offset_t off; 430 vm_size_t sz; 431 int error, rv; 432 433 if (start != trunc_page(start)) { 434 rv = __elfN(map_partial)(map, object, offset, start, 435 round_page(start), prot); 436 if (rv) 437 return (rv); 438 offset += round_page(start) - start; 439 start = round_page(start); 440 } 441 if (end != round_page(end)) { 442 rv = __elfN(map_partial)(map, object, offset + 443 trunc_page(end) - start, trunc_page(end), end, prot); 444 if (rv) 445 return (rv); 446 end = trunc_page(end); 447 } 448 if (end > start) { 449 if (offset & PAGE_MASK) { 450 /* 451 * The mapping is not page aligned. This means we have 452 * to copy the data. Sigh. 453 */ 454 rv = vm_map_find(map, NULL, 0, &start, end - start, 0, 455 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL, 456 0); 457 if (rv != KERN_SUCCESS) 458 return (rv); 459 if (object == NULL) 460 return (KERN_SUCCESS); 461 for (; start < end; start += sz) { 462 sf = vm_imgact_map_page(object, offset); 463 if (sf == NULL) 464 return (KERN_FAILURE); 465 off = offset - trunc_page(offset); 466 sz = end - start; 467 if (sz > PAGE_SIZE - off) 468 sz = PAGE_SIZE - off; 469 error = copyout((caddr_t)sf_buf_kva(sf) + off, 470 (caddr_t)start, sz); 471 vm_imgact_unmap_page(sf); 472 if (error != 0) 473 return (KERN_FAILURE); 474 offset += sz; 475 } 476 rv = KERN_SUCCESS; 477 } else { 478 vm_object_reference(object); 479 vm_map_lock(map); 480 rv = vm_map_insert(map, object, offset, start, end, 481 prot, VM_PROT_ALL, cow); 482 vm_map_unlock(map); 483 if (rv != KERN_SUCCESS) 484 vm_object_deallocate(object); 485 } 486 return (rv); 487 } else { 488 return (KERN_SUCCESS); 489 } 490 } 491 492 static int 493 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, 494 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 495 size_t pagesize) 496 { 497 struct sf_buf *sf; 498 size_t map_len; 499 vm_map_t map; 500 vm_object_t object; 501 vm_offset_t map_addr; 502 int error, rv, cow; 503 size_t copy_len; 504 vm_offset_t file_addr; 505 506 /* 507 * It's necessary to fail if the filsz + offset taken from the 508 * header is greater than the actual file pager object's size. 509 * If we were to allow this, then the vm_map_find() below would 510 * walk right off the end of the file object and into the ether. 511 * 512 * While I'm here, might as well check for something else that 513 * is invalid: filsz cannot be greater than memsz. 514 */ 515 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) { 516 uprintf("elf_load_section: truncated ELF file\n"); 517 return (ENOEXEC); 518 } 519 520 object = imgp->object; 521 map = &imgp->proc->p_vmspace->vm_map; 522 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 523 file_addr = trunc_page_ps(offset, pagesize); 524 525 /* 526 * We have two choices. We can either clear the data in the last page 527 * of an oversized mapping, or we can start the anon mapping a page 528 * early and copy the initialized data into that first page. We 529 * choose the second.. 530 */ 531 if (memsz > filsz) 532 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 533 else 534 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 535 536 if (map_len != 0) { 537 /* cow flags: don't dump readonly sections in core */ 538 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 539 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 540 541 rv = __elfN(map_insert)(map, 542 object, 543 file_addr, /* file offset */ 544 map_addr, /* virtual start */ 545 map_addr + map_len,/* virtual end */ 546 prot, 547 cow); 548 if (rv != KERN_SUCCESS) 549 return (EINVAL); 550 551 /* we can stop now if we've covered it all */ 552 if (memsz == filsz) { 553 return (0); 554 } 555 } 556 557 558 /* 559 * We have to get the remaining bit of the file into the first part 560 * of the oversized map segment. This is normally because the .data 561 * segment in the file is extended to provide bss. It's a neat idea 562 * to try and save a page, but it's a pain in the behind to implement. 563 */ 564 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 565 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 566 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 567 map_addr; 568 569 /* This had damn well better be true! */ 570 if (map_len != 0) { 571 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr + 572 map_len, VM_PROT_ALL, 0); 573 if (rv != KERN_SUCCESS) { 574 return (EINVAL); 575 } 576 } 577 578 if (copy_len != 0) { 579 vm_offset_t off; 580 581 sf = vm_imgact_map_page(object, offset + filsz); 582 if (sf == NULL) 583 return (EIO); 584 585 /* send the page fragment to user space */ 586 off = trunc_page_ps(offset + filsz, pagesize) - 587 trunc_page(offset + filsz); 588 error = copyout((caddr_t)sf_buf_kva(sf) + off, 589 (caddr_t)map_addr, copy_len); 590 vm_imgact_unmap_page(sf); 591 if (error) { 592 return (error); 593 } 594 } 595 596 /* 597 * set it to the specified protection. 598 * XXX had better undo the damage from pasting over the cracks here! 599 */ 600 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + 601 map_len), prot, FALSE); 602 603 return (0); 604 } 605 606 /* 607 * Load the file "file" into memory. It may be either a shared object 608 * or an executable. 609 * 610 * The "addr" reference parameter is in/out. On entry, it specifies 611 * the address where a shared object should be loaded. If the file is 612 * an executable, this value is ignored. On exit, "addr" specifies 613 * where the file was actually loaded. 614 * 615 * The "entry" reference parameter is out only. On exit, it specifies 616 * the entry point for the loaded file. 617 */ 618 static int 619 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 620 u_long *entry, size_t pagesize) 621 { 622 struct { 623 struct nameidata nd; 624 struct vattr attr; 625 struct image_params image_params; 626 } *tempdata; 627 const Elf_Ehdr *hdr = NULL; 628 const Elf_Phdr *phdr = NULL; 629 struct nameidata *nd; 630 struct vattr *attr; 631 struct image_params *imgp; 632 vm_prot_t prot; 633 u_long rbase; 634 u_long base_addr = 0; 635 int error, i, numsegs; 636 637 #ifdef CAPABILITY_MODE 638 /* 639 * XXXJA: This check can go away once we are sufficiently confident 640 * that the checks in namei() are correct. 641 */ 642 if (IN_CAPABILITY_MODE(curthread)) 643 return (ECAPMODE); 644 #endif 645 646 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 647 nd = &tempdata->nd; 648 attr = &tempdata->attr; 649 imgp = &tempdata->image_params; 650 651 /* 652 * Initialize part of the common data 653 */ 654 imgp->proc = p; 655 imgp->attr = attr; 656 imgp->firstpage = NULL; 657 imgp->image_header = NULL; 658 imgp->object = NULL; 659 imgp->execlabel = NULL; 660 661 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); 662 if ((error = namei(nd)) != 0) { 663 nd->ni_vp = NULL; 664 goto fail; 665 } 666 NDFREE(nd, NDF_ONLY_PNBUF); 667 imgp->vp = nd->ni_vp; 668 669 /* 670 * Check permissions, modes, uid, etc on the file, and "open" it. 671 */ 672 error = exec_check_permissions(imgp); 673 if (error) 674 goto fail; 675 676 error = exec_map_first_page(imgp); 677 if (error) 678 goto fail; 679 680 /* 681 * Also make certain that the interpreter stays the same, so set 682 * its VV_TEXT flag, too. 683 */ 684 VOP_SET_TEXT(nd->ni_vp); 685 686 imgp->object = nd->ni_vp->v_object; 687 688 hdr = (const Elf_Ehdr *)imgp->image_header; 689 if ((error = __elfN(check_header)(hdr)) != 0) 690 goto fail; 691 if (hdr->e_type == ET_DYN) 692 rbase = *addr; 693 else if (hdr->e_type == ET_EXEC) 694 rbase = 0; 695 else { 696 error = ENOEXEC; 697 goto fail; 698 } 699 700 /* Only support headers that fit within first page for now */ 701 if ((hdr->e_phoff > PAGE_SIZE) || 702 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 703 error = ENOEXEC; 704 goto fail; 705 } 706 707 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 708 if (!aligned(phdr, Elf_Addr)) { 709 error = ENOEXEC; 710 goto fail; 711 } 712 713 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 714 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { 715 /* Loadable segment */ 716 prot = __elfN(trans_prot)(phdr[i].p_flags); 717 error = __elfN(load_section)(imgp, phdr[i].p_offset, 718 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 719 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); 720 if (error != 0) 721 goto fail; 722 /* 723 * Establish the base address if this is the 724 * first segment. 725 */ 726 if (numsegs == 0) 727 base_addr = trunc_page(phdr[i].p_vaddr + 728 rbase); 729 numsegs++; 730 } 731 } 732 *addr = base_addr; 733 *entry = (unsigned long)hdr->e_entry + rbase; 734 735 fail: 736 if (imgp->firstpage) 737 exec_unmap_first_page(imgp); 738 739 if (nd->ni_vp) 740 vput(nd->ni_vp); 741 742 free(tempdata, M_TEMP); 743 744 return (error); 745 } 746 747 static int 748 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 749 { 750 struct thread *td; 751 const Elf_Ehdr *hdr; 752 const Elf_Phdr *phdr; 753 Elf_Auxargs *elf_auxargs; 754 struct vmspace *vmspace; 755 const char *err_str, *newinterp; 756 char *interp, *interp_buf, *path; 757 Elf_Brandinfo *brand_info; 758 struct sysentvec *sv; 759 vm_prot_t prot; 760 u_long text_size, data_size, total_size, text_addr, data_addr; 761 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr; 762 int32_t osrel; 763 int error, i, n, interp_name_len, have_interp; 764 765 hdr = (const Elf_Ehdr *)imgp->image_header; 766 767 /* 768 * Do we have a valid ELF header ? 769 * 770 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 771 * if particular brand doesn't support it. 772 */ 773 if (__elfN(check_header)(hdr) != 0 || 774 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 775 return (-1); 776 777 /* 778 * From here on down, we return an errno, not -1, as we've 779 * detected an ELF file. 780 */ 781 782 if ((hdr->e_phoff > PAGE_SIZE) || 783 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { 784 /* Only support headers in first page for now */ 785 uprintf("Program headers not in the first page\n"); 786 return (ENOEXEC); 787 } 788 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 789 if (!aligned(phdr, Elf_Addr)) { 790 uprintf("Unaligned program headers\n"); 791 return (ENOEXEC); 792 } 793 794 n = error = 0; 795 baddr = 0; 796 osrel = 0; 797 text_size = data_size = total_size = text_addr = data_addr = 0; 798 entry = proghdr = 0; 799 interp_name_len = 0; 800 err_str = newinterp = NULL; 801 interp = interp_buf = NULL; 802 td = curthread; 803 804 for (i = 0; i < hdr->e_phnum; i++) { 805 switch (phdr[i].p_type) { 806 case PT_LOAD: 807 if (n == 0) 808 baddr = phdr[i].p_vaddr; 809 n++; 810 break; 811 case PT_INTERP: 812 /* Path to interpreter */ 813 if (phdr[i].p_filesz > MAXPATHLEN) { 814 uprintf("Invalid PT_INTERP\n"); 815 error = ENOEXEC; 816 goto ret; 817 } 818 if (interp != NULL) { 819 uprintf("Multiple PT_INTERP headers\n"); 820 error = ENOEXEC; 821 goto ret; 822 } 823 interp_name_len = phdr[i].p_filesz; 824 if (phdr[i].p_offset > PAGE_SIZE || 825 interp_name_len > PAGE_SIZE - phdr[i].p_offset) { 826 VOP_UNLOCK(imgp->vp, 0); 827 interp_buf = malloc(interp_name_len + 1, M_TEMP, 828 M_WAITOK); 829 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 830 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf, 831 interp_name_len, phdr[i].p_offset, 832 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 833 NOCRED, NULL, td); 834 if (error != 0) { 835 uprintf("i/o error PT_INTERP\n"); 836 goto ret; 837 } 838 interp_buf[interp_name_len] = '\0'; 839 interp = interp_buf; 840 } else { 841 interp = __DECONST(char *, imgp->image_header) + 842 phdr[i].p_offset; 843 } 844 break; 845 case PT_GNU_STACK: 846 if (__elfN(nxstack)) 847 imgp->stack_prot = 848 __elfN(trans_prot)(phdr[i].p_flags); 849 imgp->stack_sz = phdr[i].p_memsz; 850 break; 851 } 852 } 853 854 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, 855 &osrel); 856 if (brand_info == NULL) { 857 uprintf("ELF binary type \"%u\" not known.\n", 858 hdr->e_ident[EI_OSABI]); 859 error = ENOEXEC; 860 goto ret; 861 } 862 et_dyn_addr = 0; 863 if (hdr->e_type == ET_DYN) { 864 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 865 uprintf("Cannot execute shared object\n"); 866 error = ENOEXEC; 867 goto ret; 868 } 869 /* 870 * Honour the base load address from the dso if it is 871 * non-zero for some reason. 872 */ 873 if (baddr == 0) 874 et_dyn_addr = ET_DYN_LOAD_ADDR; 875 } 876 sv = brand_info->sysvec; 877 if (interp != NULL && brand_info->interp_newpath != NULL) 878 newinterp = brand_info->interp_newpath; 879 880 /* 881 * Avoid a possible deadlock if the current address space is destroyed 882 * and that address space maps the locked vnode. In the common case, 883 * the locked vnode's v_usecount is decremented but remains greater 884 * than zero. Consequently, the vnode lock is not needed by vrele(). 885 * However, in cases where the vnode lock is external, such as nullfs, 886 * v_usecount may become zero. 887 * 888 * The VV_TEXT flag prevents modifications to the executable while 889 * the vnode is unlocked. 890 */ 891 VOP_UNLOCK(imgp->vp, 0); 892 893 error = exec_new_vmspace(imgp, sv); 894 imgp->proc->p_sysent = sv; 895 896 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 897 if (error != 0) 898 goto ret; 899 900 for (i = 0; i < hdr->e_phnum; i++) { 901 switch (phdr[i].p_type) { 902 case PT_LOAD: /* Loadable segment */ 903 if (phdr[i].p_memsz == 0) 904 break; 905 prot = __elfN(trans_prot)(phdr[i].p_flags); 906 error = __elfN(load_section)(imgp, phdr[i].p_offset, 907 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 908 phdr[i].p_memsz, phdr[i].p_filesz, prot, 909 sv->sv_pagesize); 910 if (error != 0) 911 goto ret; 912 913 /* 914 * If this segment contains the program headers, 915 * remember their virtual address for the AT_PHDR 916 * aux entry. Static binaries don't usually include 917 * a PT_PHDR entry. 918 */ 919 if (phdr[i].p_offset == 0 && 920 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 921 <= phdr[i].p_filesz) 922 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 923 et_dyn_addr; 924 925 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 926 seg_size = round_page(phdr[i].p_memsz + 927 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 928 929 /* 930 * Make the largest executable segment the official 931 * text segment and all others data. 932 * 933 * Note that obreak() assumes that data_addr + 934 * data_size == end of data load area, and the ELF 935 * file format expects segments to be sorted by 936 * address. If multiple data segments exist, the 937 * last one will be used. 938 */ 939 940 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 941 text_size = seg_size; 942 text_addr = seg_addr; 943 } else { 944 data_size = seg_size; 945 data_addr = seg_addr; 946 } 947 total_size += seg_size; 948 break; 949 case PT_PHDR: /* Program header table info */ 950 proghdr = phdr[i].p_vaddr + et_dyn_addr; 951 break; 952 default: 953 break; 954 } 955 } 956 957 if (data_addr == 0 && data_size == 0) { 958 data_addr = text_addr; 959 data_size = text_size; 960 } 961 962 entry = (u_long)hdr->e_entry + et_dyn_addr; 963 964 /* 965 * Check limits. It should be safe to check the 966 * limits after loading the segments since we do 967 * not actually fault in all the segments pages. 968 */ 969 PROC_LOCK(imgp->proc); 970 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 971 err_str = "Data segment size exceeds process limit"; 972 else if (text_size > maxtsiz) 973 err_str = "Text segment size exceeds system limit"; 974 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 975 err_str = "Total segment size exceeds process limit"; 976 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 977 err_str = "Data segment size exceeds resource limit"; 978 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 979 err_str = "Total segment size exceeds resource limit"; 980 if (err_str != NULL) { 981 PROC_UNLOCK(imgp->proc); 982 uprintf("%s\n", err_str); 983 error = ENOMEM; 984 goto ret; 985 } 986 987 vmspace = imgp->proc->p_vmspace; 988 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 989 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 990 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 991 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 992 993 /* 994 * We load the dynamic linker where a userland call 995 * to mmap(0, ...) would put it. The rationale behind this 996 * calculation is that it leaves room for the heap to grow to 997 * its maximum allowed size. 998 */ 999 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1000 RLIMIT_DATA)); 1001 PROC_UNLOCK(imgp->proc); 1002 1003 imgp->entry_addr = entry; 1004 1005 if (interp != NULL) { 1006 have_interp = FALSE; 1007 VOP_UNLOCK(imgp->vp, 0); 1008 if (brand_info->emul_path != NULL && 1009 brand_info->emul_path[0] != '\0') { 1010 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1011 snprintf(path, MAXPATHLEN, "%s%s", 1012 brand_info->emul_path, interp); 1013 error = __elfN(load_file)(imgp->proc, path, &addr, 1014 &imgp->entry_addr, sv->sv_pagesize); 1015 free(path, M_TEMP); 1016 if (error == 0) 1017 have_interp = TRUE; 1018 } 1019 if (!have_interp && newinterp != NULL && 1020 (brand_info->interp_path == NULL || 1021 strcmp(interp, brand_info->interp_path) == 0)) { 1022 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 1023 &imgp->entry_addr, sv->sv_pagesize); 1024 if (error == 0) 1025 have_interp = TRUE; 1026 } 1027 if (!have_interp) { 1028 error = __elfN(load_file)(imgp->proc, interp, &addr, 1029 &imgp->entry_addr, sv->sv_pagesize); 1030 } 1031 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 1032 if (error != 0) { 1033 uprintf("ELF interpreter %s not found, error %d\n", 1034 interp, error); 1035 goto ret; 1036 } 1037 } else 1038 addr = et_dyn_addr; 1039 1040 /* 1041 * Construct auxargs table (used by the fixup routine) 1042 */ 1043 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1044 elf_auxargs->execfd = -1; 1045 elf_auxargs->phdr = proghdr; 1046 elf_auxargs->phent = hdr->e_phentsize; 1047 elf_auxargs->phnum = hdr->e_phnum; 1048 elf_auxargs->pagesz = PAGE_SIZE; 1049 elf_auxargs->base = addr; 1050 elf_auxargs->flags = 0; 1051 elf_auxargs->entry = entry; 1052 elf_auxargs->hdr_eflags = hdr->e_flags; 1053 1054 imgp->auxargs = elf_auxargs; 1055 imgp->interpreted = 0; 1056 imgp->reloc_base = addr; 1057 imgp->proc->p_osrel = osrel; 1058 imgp->proc->p_elf_machine = hdr->e_machine; 1059 imgp->proc->p_elf_flags = hdr->e_flags; 1060 1061 ret: 1062 free(interp_buf, M_TEMP); 1063 return (error); 1064 } 1065 1066 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 1067 1068 int 1069 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 1070 { 1071 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 1072 Elf_Addr *base; 1073 Elf_Addr *pos; 1074 1075 base = (Elf_Addr *)*stack_base; 1076 pos = base + (imgp->args->argc + imgp->args->envc + 2); 1077 1078 if (args->execfd != -1) 1079 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 1080 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 1081 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 1082 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 1083 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 1084 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 1085 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 1086 AUXARGS_ENTRY(pos, AT_BASE, args->base); 1087 #ifdef AT_EHDRFLAGS 1088 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); 1089 #endif 1090 if (imgp->execpathp != 0) 1091 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); 1092 AUXARGS_ENTRY(pos, AT_OSRELDATE, 1093 imgp->proc->p_ucred->cr_prison->pr_osreldate); 1094 if (imgp->canary != 0) { 1095 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); 1096 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); 1097 } 1098 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); 1099 if (imgp->pagesizes != 0) { 1100 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); 1101 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); 1102 } 1103 if (imgp->sysent->sv_timekeep_base != 0) { 1104 AUXARGS_ENTRY(pos, AT_TIMEKEEP, 1105 imgp->sysent->sv_timekeep_base); 1106 } 1107 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj 1108 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1109 imgp->sysent->sv_stackprot); 1110 AUXARGS_ENTRY(pos, AT_NULL, 0); 1111 1112 free(imgp->auxargs, M_TEMP); 1113 imgp->auxargs = NULL; 1114 1115 base--; 1116 suword(base, (long)imgp->args->argc); 1117 *stack_base = (register_t *)base; 1118 return (0); 1119 } 1120 1121 /* 1122 * Code for generating ELF core dumps. 1123 */ 1124 1125 typedef void (*segment_callback)(vm_map_entry_t, void *); 1126 1127 /* Closure for cb_put_phdr(). */ 1128 struct phdr_closure { 1129 Elf_Phdr *phdr; /* Program header to fill in */ 1130 Elf_Off offset; /* Offset of segment in core file */ 1131 }; 1132 1133 /* Closure for cb_size_segment(). */ 1134 struct sseg_closure { 1135 int count; /* Count of writable segments. */ 1136 size_t size; /* Total size of all writable segments. */ 1137 }; 1138 1139 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); 1140 1141 struct note_info { 1142 int type; /* Note type. */ 1143 outfunc_t outfunc; /* Output function. */ 1144 void *outarg; /* Argument for the output function. */ 1145 size_t outsize; /* Output size. */ 1146 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ 1147 }; 1148 1149 TAILQ_HEAD(note_info_list, note_info); 1150 1151 /* Coredump output parameters. */ 1152 struct coredump_params { 1153 off_t offset; 1154 struct ucred *active_cred; 1155 struct ucred *file_cred; 1156 struct thread *td; 1157 struct vnode *vp; 1158 struct gzio_stream *gzs; 1159 }; 1160 1161 static void cb_put_phdr(vm_map_entry_t, void *); 1162 static void cb_size_segment(vm_map_entry_t, void *); 1163 static int core_write(struct coredump_params *, const void *, size_t, off_t, 1164 enum uio_seg); 1165 static void each_dumpable_segment(struct thread *, segment_callback, void *); 1166 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, 1167 struct note_info_list *, size_t); 1168 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, 1169 size_t *); 1170 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); 1171 static void __elfN(putnote)(struct note_info *, struct sbuf *); 1172 static size_t register_note(struct note_info_list *, int, outfunc_t, void *); 1173 static int sbuf_drain_core_output(void *, const char *, int); 1174 static int sbuf_drain_count(void *arg, const char *data, int len); 1175 1176 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); 1177 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); 1178 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); 1179 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); 1180 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); 1181 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); 1182 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); 1183 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); 1184 static void note_procstat_files(void *, struct sbuf *, size_t *); 1185 static void note_procstat_groups(void *, struct sbuf *, size_t *); 1186 static void note_procstat_osrel(void *, struct sbuf *, size_t *); 1187 static void note_procstat_rlimit(void *, struct sbuf *, size_t *); 1188 static void note_procstat_umask(void *, struct sbuf *, size_t *); 1189 static void note_procstat_vmmap(void *, struct sbuf *, size_t *); 1190 1191 #ifdef GZIO 1192 extern int compress_user_cores_gzlevel; 1193 1194 /* 1195 * Write out a core segment to the compression stream. 1196 */ 1197 static int 1198 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len) 1199 { 1200 u_int chunk_len; 1201 int error; 1202 1203 while (len > 0) { 1204 chunk_len = MIN(len, CORE_BUF_SIZE); 1205 1206 /* 1207 * We can get EFAULT error here. 1208 * In that case zero out the current chunk of the segment. 1209 */ 1210 error = copyin(base, buf, chunk_len); 1211 if (error != 0) 1212 bzero(buf, chunk_len); 1213 error = gzio_write(p->gzs, buf, chunk_len); 1214 if (error != 0) 1215 break; 1216 base += chunk_len; 1217 len -= chunk_len; 1218 } 1219 return (error); 1220 } 1221 1222 static int 1223 core_gz_write(void *base, size_t len, off_t offset, void *arg) 1224 { 1225 1226 return (core_write((struct coredump_params *)arg, base, len, offset, 1227 UIO_SYSSPACE)); 1228 } 1229 #endif /* GZIO */ 1230 1231 static int 1232 core_write(struct coredump_params *p, const void *base, size_t len, 1233 off_t offset, enum uio_seg seg) 1234 { 1235 1236 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base), 1237 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1238 p->active_cred, p->file_cred, NULL, p->td)); 1239 } 1240 1241 static int 1242 core_output(void *base, size_t len, off_t offset, struct coredump_params *p, 1243 void *tmpbuf) 1244 { 1245 int error; 1246 1247 #ifdef GZIO 1248 if (p->gzs != NULL) 1249 return (compress_chunk(p, base, tmpbuf, len)); 1250 #endif 1251 /* 1252 * EFAULT is a non-fatal error that we can get, for example, 1253 * if the segment is backed by a file but extends beyond its 1254 * end. 1255 */ 1256 error = core_write(p, base, len, offset, UIO_USERSPACE); 1257 if (error == EFAULT) { 1258 log(LOG_WARNING, "Failed to fully fault in a core file segment " 1259 "at VA %p with size 0x%zx to be written at offset 0x%jx " 1260 "for process %s\n", base, len, offset, curproc->p_comm); 1261 1262 /* 1263 * Write a "real" zero byte at the end of the target region 1264 * in the case this is the last segment. 1265 * The intermediate space will be implicitly zero-filled. 1266 */ 1267 error = core_write(p, zero_region, 1, offset + len - 1, 1268 UIO_SYSSPACE); 1269 } 1270 return (error); 1271 } 1272 1273 /* 1274 * Drain into a core file. 1275 */ 1276 static int 1277 sbuf_drain_core_output(void *arg, const char *data, int len) 1278 { 1279 struct coredump_params *p; 1280 int error, locked; 1281 1282 p = (struct coredump_params *)arg; 1283 1284 /* 1285 * Some kern_proc out routines that print to this sbuf may 1286 * call us with the process lock held. Draining with the 1287 * non-sleepable lock held is unsafe. The lock is needed for 1288 * those routines when dumping a live process. In our case we 1289 * can safely release the lock before draining and acquire 1290 * again after. 1291 */ 1292 locked = PROC_LOCKED(p->td->td_proc); 1293 if (locked) 1294 PROC_UNLOCK(p->td->td_proc); 1295 #ifdef GZIO 1296 if (p->gzs != NULL) 1297 error = gzio_write(p->gzs, __DECONST(char *, data), len); 1298 else 1299 #endif 1300 error = core_write(p, __DECONST(void *, data), len, p->offset, 1301 UIO_SYSSPACE); 1302 if (locked) 1303 PROC_LOCK(p->td->td_proc); 1304 if (error != 0) 1305 return (-error); 1306 p->offset += len; 1307 return (len); 1308 } 1309 1310 /* 1311 * Drain into a counter. 1312 */ 1313 static int 1314 sbuf_drain_count(void *arg, const char *data __unused, int len) 1315 { 1316 size_t *sizep; 1317 1318 sizep = (size_t *)arg; 1319 *sizep += len; 1320 return (len); 1321 } 1322 1323 int 1324 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1325 { 1326 struct ucred *cred = td->td_ucred; 1327 int error = 0; 1328 struct sseg_closure seginfo; 1329 struct note_info_list notelst; 1330 struct coredump_params params; 1331 struct note_info *ninfo; 1332 void *hdr, *tmpbuf; 1333 size_t hdrsize, notesz, coresize; 1334 #ifdef GZIO 1335 boolean_t compress; 1336 1337 compress = (flags & IMGACT_CORE_COMPRESS) != 0; 1338 #endif 1339 hdr = NULL; 1340 tmpbuf = NULL; 1341 TAILQ_INIT(¬elst); 1342 1343 /* Size the program segments. */ 1344 seginfo.count = 0; 1345 seginfo.size = 0; 1346 each_dumpable_segment(td, cb_size_segment, &seginfo); 1347 1348 /* 1349 * Collect info about the core file header area. 1350 */ 1351 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1352 if (seginfo.count + 1 >= PN_XNUM) 1353 hdrsize += sizeof(Elf_Shdr); 1354 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1355 coresize = round_page(hdrsize + notesz) + seginfo.size; 1356 1357 /* Set up core dump parameters. */ 1358 params.offset = 0; 1359 params.active_cred = cred; 1360 params.file_cred = NOCRED; 1361 params.td = td; 1362 params.vp = vp; 1363 params.gzs = NULL; 1364 1365 #ifdef RACCT 1366 if (racct_enable) { 1367 PROC_LOCK(td->td_proc); 1368 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1369 PROC_UNLOCK(td->td_proc); 1370 if (error != 0) { 1371 error = EFAULT; 1372 goto done; 1373 } 1374 } 1375 #endif 1376 if (coresize >= limit) { 1377 error = EFAULT; 1378 goto done; 1379 } 1380 1381 #ifdef GZIO 1382 /* Create a compression stream if necessary. */ 1383 if (compress) { 1384 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE, 1385 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms); 1386 if (params.gzs == NULL) { 1387 error = EFAULT; 1388 goto done; 1389 } 1390 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1391 } 1392 #endif 1393 1394 /* 1395 * Allocate memory for building the header, fill it up, 1396 * and write it out following the notes. 1397 */ 1398 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1399 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1400 notesz); 1401 1402 /* Write the contents of all of the writable segments. */ 1403 if (error == 0) { 1404 Elf_Phdr *php; 1405 off_t offset; 1406 int i; 1407 1408 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1409 offset = round_page(hdrsize + notesz); 1410 for (i = 0; i < seginfo.count; i++) { 1411 error = core_output((caddr_t)(uintptr_t)php->p_vaddr, 1412 php->p_filesz, offset, ¶ms, tmpbuf); 1413 if (error != 0) 1414 break; 1415 offset += php->p_filesz; 1416 php++; 1417 } 1418 #ifdef GZIO 1419 if (error == 0 && compress) 1420 error = gzio_flush(params.gzs); 1421 #endif 1422 } 1423 if (error) { 1424 log(LOG_WARNING, 1425 "Failed to write core file for process %s (error %d)\n", 1426 curproc->p_comm, error); 1427 } 1428 1429 done: 1430 #ifdef GZIO 1431 if (compress) { 1432 free(tmpbuf, M_TEMP); 1433 if (params.gzs != NULL) 1434 gzio_fini(params.gzs); 1435 } 1436 #endif 1437 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1438 TAILQ_REMOVE(¬elst, ninfo, link); 1439 free(ninfo, M_TEMP); 1440 } 1441 if (hdr != NULL) 1442 free(hdr, M_TEMP); 1443 1444 return (error); 1445 } 1446 1447 /* 1448 * A callback for each_dumpable_segment() to write out the segment's 1449 * program header entry. 1450 */ 1451 static void 1452 cb_put_phdr(entry, closure) 1453 vm_map_entry_t entry; 1454 void *closure; 1455 { 1456 struct phdr_closure *phc = (struct phdr_closure *)closure; 1457 Elf_Phdr *phdr = phc->phdr; 1458 1459 phc->offset = round_page(phc->offset); 1460 1461 phdr->p_type = PT_LOAD; 1462 phdr->p_offset = phc->offset; 1463 phdr->p_vaddr = entry->start; 1464 phdr->p_paddr = 0; 1465 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1466 phdr->p_align = PAGE_SIZE; 1467 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1468 1469 phc->offset += phdr->p_filesz; 1470 phc->phdr++; 1471 } 1472 1473 /* 1474 * A callback for each_dumpable_segment() to gather information about 1475 * the number of segments and their total size. 1476 */ 1477 static void 1478 cb_size_segment(vm_map_entry_t entry, void *closure) 1479 { 1480 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1481 1482 ssc->count++; 1483 ssc->size += entry->end - entry->start; 1484 } 1485 1486 /* 1487 * For each writable segment in the process's memory map, call the given 1488 * function with a pointer to the map entry and some arbitrary 1489 * caller-supplied data. 1490 */ 1491 static void 1492 each_dumpable_segment(struct thread *td, segment_callback func, void *closure) 1493 { 1494 struct proc *p = td->td_proc; 1495 vm_map_t map = &p->p_vmspace->vm_map; 1496 vm_map_entry_t entry; 1497 vm_object_t backing_object, object; 1498 boolean_t ignore_entry; 1499 1500 vm_map_lock_read(map); 1501 for (entry = map->header.next; entry != &map->header; 1502 entry = entry->next) { 1503 /* 1504 * Don't dump inaccessible mappings, deal with legacy 1505 * coredump mode. 1506 * 1507 * Note that read-only segments related to the elf binary 1508 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1509 * need to arbitrarily ignore such segments. 1510 */ 1511 if (elf_legacy_coredump) { 1512 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1513 continue; 1514 } else { 1515 if ((entry->protection & VM_PROT_ALL) == 0) 1516 continue; 1517 } 1518 1519 /* 1520 * Dont include memory segment in the coredump if 1521 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1522 * madvise(2). Do not dump submaps (i.e. parts of the 1523 * kernel map). 1524 */ 1525 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1526 continue; 1527 1528 if ((object = entry->object.vm_object) == NULL) 1529 continue; 1530 1531 /* Ignore memory-mapped devices and such things. */ 1532 VM_OBJECT_RLOCK(object); 1533 while ((backing_object = object->backing_object) != NULL) { 1534 VM_OBJECT_RLOCK(backing_object); 1535 VM_OBJECT_RUNLOCK(object); 1536 object = backing_object; 1537 } 1538 ignore_entry = object->type != OBJT_DEFAULT && 1539 object->type != OBJT_SWAP && object->type != OBJT_VNODE && 1540 object->type != OBJT_PHYS; 1541 VM_OBJECT_RUNLOCK(object); 1542 if (ignore_entry) 1543 continue; 1544 1545 (*func)(entry, closure); 1546 } 1547 vm_map_unlock_read(map); 1548 } 1549 1550 /* 1551 * Write the core file header to the file, including padding up to 1552 * the page boundary. 1553 */ 1554 static int 1555 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1556 size_t hdrsize, struct note_info_list *notelst, size_t notesz) 1557 { 1558 struct note_info *ninfo; 1559 struct sbuf *sb; 1560 int error; 1561 1562 /* Fill in the header. */ 1563 bzero(hdr, hdrsize); 1564 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz); 1565 1566 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1567 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1568 sbuf_start_section(sb, NULL); 1569 sbuf_bcat(sb, hdr, hdrsize); 1570 TAILQ_FOREACH(ninfo, notelst, link) 1571 __elfN(putnote)(ninfo, sb); 1572 /* Align up to a page boundary for the program segments. */ 1573 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1574 error = sbuf_finish(sb); 1575 sbuf_delete(sb); 1576 1577 return (error); 1578 } 1579 1580 static void 1581 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1582 size_t *sizep) 1583 { 1584 struct proc *p; 1585 struct thread *thr; 1586 size_t size; 1587 1588 p = td->td_proc; 1589 size = 0; 1590 1591 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1592 1593 /* 1594 * To have the debugger select the right thread (LWP) as the initial 1595 * thread, we dump the state of the thread passed to us in td first. 1596 * This is the thread that causes the core dump and thus likely to 1597 * be the right thread one wants to have selected in the debugger. 1598 */ 1599 thr = td; 1600 while (thr != NULL) { 1601 size += register_note(list, NT_PRSTATUS, 1602 __elfN(note_prstatus), thr); 1603 size += register_note(list, NT_FPREGSET, 1604 __elfN(note_fpregset), thr); 1605 size += register_note(list, NT_THRMISC, 1606 __elfN(note_thrmisc), thr); 1607 size += register_note(list, -1, 1608 __elfN(note_threadmd), thr); 1609 1610 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1611 TAILQ_NEXT(thr, td_plist); 1612 if (thr == td) 1613 thr = TAILQ_NEXT(thr, td_plist); 1614 } 1615 1616 size += register_note(list, NT_PROCSTAT_PROC, 1617 __elfN(note_procstat_proc), p); 1618 size += register_note(list, NT_PROCSTAT_FILES, 1619 note_procstat_files, p); 1620 size += register_note(list, NT_PROCSTAT_VMMAP, 1621 note_procstat_vmmap, p); 1622 size += register_note(list, NT_PROCSTAT_GROUPS, 1623 note_procstat_groups, p); 1624 size += register_note(list, NT_PROCSTAT_UMASK, 1625 note_procstat_umask, p); 1626 size += register_note(list, NT_PROCSTAT_RLIMIT, 1627 note_procstat_rlimit, p); 1628 size += register_note(list, NT_PROCSTAT_OSREL, 1629 note_procstat_osrel, p); 1630 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1631 __elfN(note_procstat_psstrings), p); 1632 size += register_note(list, NT_PROCSTAT_AUXV, 1633 __elfN(note_procstat_auxv), p); 1634 1635 *sizep = size; 1636 } 1637 1638 static void 1639 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1640 size_t notesz) 1641 { 1642 Elf_Ehdr *ehdr; 1643 Elf_Phdr *phdr; 1644 Elf_Shdr *shdr; 1645 struct phdr_closure phc; 1646 1647 ehdr = (Elf_Ehdr *)hdr; 1648 1649 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1650 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1651 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1652 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1653 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1654 ehdr->e_ident[EI_DATA] = ELF_DATA; 1655 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1656 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1657 ehdr->e_ident[EI_ABIVERSION] = 0; 1658 ehdr->e_ident[EI_PAD] = 0; 1659 ehdr->e_type = ET_CORE; 1660 ehdr->e_machine = td->td_proc->p_elf_machine; 1661 ehdr->e_version = EV_CURRENT; 1662 ehdr->e_entry = 0; 1663 ehdr->e_phoff = sizeof(Elf_Ehdr); 1664 ehdr->e_flags = td->td_proc->p_elf_flags; 1665 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1666 ehdr->e_phentsize = sizeof(Elf_Phdr); 1667 ehdr->e_shentsize = sizeof(Elf_Shdr); 1668 ehdr->e_shstrndx = SHN_UNDEF; 1669 if (numsegs + 1 < PN_XNUM) { 1670 ehdr->e_phnum = numsegs + 1; 1671 ehdr->e_shnum = 0; 1672 } else { 1673 ehdr->e_phnum = PN_XNUM; 1674 ehdr->e_shnum = 1; 1675 1676 ehdr->e_shoff = ehdr->e_phoff + 1677 (numsegs + 1) * ehdr->e_phentsize; 1678 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1679 ("e_shoff: %zu, hdrsize - shdr: %zu", 1680 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1681 1682 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1683 memset(shdr, 0, sizeof(*shdr)); 1684 /* 1685 * A special first section is used to hold large segment and 1686 * section counts. This was proposed by Sun Microsystems in 1687 * Solaris and has been adopted by Linux; the standard ELF 1688 * tools are already familiar with the technique. 1689 * 1690 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1691 * (or 12-7 depending on the version of the document) for more 1692 * details. 1693 */ 1694 shdr->sh_type = SHT_NULL; 1695 shdr->sh_size = ehdr->e_shnum; 1696 shdr->sh_link = ehdr->e_shstrndx; 1697 shdr->sh_info = numsegs + 1; 1698 } 1699 1700 /* 1701 * Fill in the program header entries. 1702 */ 1703 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 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_dumpable_segment(td, cb_put_phdr, &phc); 1720 } 1721 1722 static size_t 1723 register_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 1748 static size_t 1749 append_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 1761 size_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 1790 static 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 1850 typedef struct prstatus32 elf_prstatus_t; 1851 typedef struct prpsinfo32 elf_prpsinfo_t; 1852 typedef struct fpreg32 elf_prfpregset_t; 1853 typedef struct fpreg32 elf_fpregset_t; 1854 typedef struct reg32 elf_gregset_t; 1855 typedef struct thrmisc32 elf_thrmisc_t; 1856 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1857 typedef struct kinfo_proc32 elf_kinfo_proc_t; 1858 typedef uint32_t elf_ps_strings_t; 1859 #else 1860 typedef prstatus_t elf_prstatus_t; 1861 typedef prpsinfo_t elf_prpsinfo_t; 1862 typedef prfpregset_t elf_prfpregset_t; 1863 typedef prfpregset_t elf_fpregset_t; 1864 typedef gregset_t elf_gregset_t; 1865 typedef thrmisc_t elf_thrmisc_t; 1866 #define ELF_KERN_PROC_MASK 0 1867 typedef struct kinfo_proc elf_kinfo_proc_t; 1868 typedef vm_offset_t elf_ps_strings_t; 1869 #endif 1870 1871 static 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 1932 static 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 1960 static 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 1981 static 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 */ 2001 static 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 2024 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2025 #endif 2026 2027 static 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 2051 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2052 #endif 2053 2054 static void 2055 note_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 2103 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2104 #endif 2105 2106 static void 2107 note_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 2138 static void 2139 note_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 2157 static void 2158 note_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 2175 static void 2176 note_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 2198 static void 2199 note_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 2216 static 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 2240 static 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 2268 static 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; 2331 nextnote: 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; 2337 ret: 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 */ 2347 static 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 */ 2370 static struct execsw __elfN(execsw) = { 2371 __CONCAT(exec_, __elfN(imgact)), 2372 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2373 }; 2374 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2375 2376 static 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 2397 static 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