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 if (hdr->e_type == ET_DYN) { 863 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { 864 uprintf("Cannot execute shared object\n"); 865 error = ENOEXEC; 866 goto ret; 867 } 868 /* 869 * Honour the base load address from the dso if it is 870 * non-zero for some reason. 871 */ 872 if (baddr == 0) 873 et_dyn_addr = ET_DYN_LOAD_ADDR; 874 else 875 et_dyn_addr = 0; 876 } else 877 et_dyn_addr = 0; 878 sv = brand_info->sysvec; 879 if (interp != NULL && brand_info->interp_newpath != NULL) 880 newinterp = brand_info->interp_newpath; 881 882 /* 883 * Avoid a possible deadlock if the current address space is destroyed 884 * and that address space maps the locked vnode. In the common case, 885 * the locked vnode's v_usecount is decremented but remains greater 886 * than zero. Consequently, the vnode lock is not needed by vrele(). 887 * However, in cases where the vnode lock is external, such as nullfs, 888 * v_usecount may become zero. 889 * 890 * The VV_TEXT flag prevents modifications to the executable while 891 * the vnode is unlocked. 892 */ 893 VOP_UNLOCK(imgp->vp, 0); 894 895 error = exec_new_vmspace(imgp, sv); 896 imgp->proc->p_sysent = sv; 897 898 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 899 if (error != 0) 900 goto ret; 901 902 for (i = 0; i < hdr->e_phnum; i++) { 903 switch (phdr[i].p_type) { 904 case PT_LOAD: /* Loadable segment */ 905 if (phdr[i].p_memsz == 0) 906 break; 907 prot = __elfN(trans_prot)(phdr[i].p_flags); 908 error = __elfN(load_section)(imgp, phdr[i].p_offset, 909 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, 910 phdr[i].p_memsz, phdr[i].p_filesz, prot, 911 sv->sv_pagesize); 912 if (error != 0) 913 goto ret; 914 915 /* 916 * If this segment contains the program headers, 917 * remember their virtual address for the AT_PHDR 918 * aux entry. Static binaries don't usually include 919 * a PT_PHDR entry. 920 */ 921 if (phdr[i].p_offset == 0 && 922 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 923 <= phdr[i].p_filesz) 924 proghdr = phdr[i].p_vaddr + hdr->e_phoff + 925 et_dyn_addr; 926 927 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); 928 seg_size = round_page(phdr[i].p_memsz + 929 phdr[i].p_vaddr + et_dyn_addr - seg_addr); 930 931 /* 932 * Make the largest executable segment the official 933 * text segment and all others data. 934 * 935 * Note that obreak() assumes that data_addr + 936 * data_size == end of data load area, and the ELF 937 * file format expects segments to be sorted by 938 * address. If multiple data segments exist, the 939 * last one will be used. 940 */ 941 942 if (phdr[i].p_flags & PF_X && text_size < seg_size) { 943 text_size = seg_size; 944 text_addr = seg_addr; 945 } else { 946 data_size = seg_size; 947 data_addr = seg_addr; 948 } 949 total_size += seg_size; 950 break; 951 case PT_PHDR: /* Program header table info */ 952 proghdr = phdr[i].p_vaddr + et_dyn_addr; 953 break; 954 default: 955 break; 956 } 957 } 958 959 if (data_addr == 0 && data_size == 0) { 960 data_addr = text_addr; 961 data_size = text_size; 962 } 963 964 entry = (u_long)hdr->e_entry + et_dyn_addr; 965 966 /* 967 * Check limits. It should be safe to check the 968 * limits after loading the segments since we do 969 * not actually fault in all the segments pages. 970 */ 971 PROC_LOCK(imgp->proc); 972 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) 973 err_str = "Data segment size exceeds process limit"; 974 else if (text_size > maxtsiz) 975 err_str = "Text segment size exceeds system limit"; 976 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) 977 err_str = "Total segment size exceeds process limit"; 978 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) 979 err_str = "Data segment size exceeds resource limit"; 980 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) 981 err_str = "Total segment size exceeds resource limit"; 982 if (err_str != NULL) { 983 PROC_UNLOCK(imgp->proc); 984 uprintf("%s\n", err_str); 985 error = ENOMEM; 986 goto ret; 987 } 988 989 vmspace = imgp->proc->p_vmspace; 990 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 991 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 992 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 993 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 994 995 /* 996 * We load the dynamic linker where a userland call 997 * to mmap(0, ...) would put it. The rationale behind this 998 * calculation is that it leaves room for the heap to grow to 999 * its maximum allowed size. 1000 */ 1001 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td, 1002 RLIMIT_DATA)); 1003 PROC_UNLOCK(imgp->proc); 1004 1005 imgp->entry_addr = entry; 1006 1007 if (interp != NULL) { 1008 have_interp = FALSE; 1009 VOP_UNLOCK(imgp->vp, 0); 1010 if (brand_info->emul_path != NULL && 1011 brand_info->emul_path[0] != '\0') { 1012 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1013 snprintf(path, MAXPATHLEN, "%s%s", 1014 brand_info->emul_path, interp); 1015 error = __elfN(load_file)(imgp->proc, path, &addr, 1016 &imgp->entry_addr, sv->sv_pagesize); 1017 free(path, M_TEMP); 1018 if (error == 0) 1019 have_interp = TRUE; 1020 } 1021 if (!have_interp && newinterp != NULL && 1022 (brand_info->interp_path == NULL || 1023 strcmp(interp, brand_info->interp_path) == 0)) { 1024 error = __elfN(load_file)(imgp->proc, newinterp, &addr, 1025 &imgp->entry_addr, sv->sv_pagesize); 1026 if (error == 0) 1027 have_interp = TRUE; 1028 } 1029 if (!have_interp) { 1030 error = __elfN(load_file)(imgp->proc, interp, &addr, 1031 &imgp->entry_addr, sv->sv_pagesize); 1032 } 1033 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 1034 if (error != 0) { 1035 uprintf("ELF interpreter %s not found, error %d\n", 1036 interp, error); 1037 goto ret; 1038 } 1039 } else 1040 addr = et_dyn_addr; 1041 1042 /* 1043 * Construct auxargs table (used by the fixup routine) 1044 */ 1045 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 1046 elf_auxargs->execfd = -1; 1047 elf_auxargs->phdr = proghdr; 1048 elf_auxargs->phent = hdr->e_phentsize; 1049 elf_auxargs->phnum = hdr->e_phnum; 1050 elf_auxargs->pagesz = PAGE_SIZE; 1051 elf_auxargs->base = addr; 1052 elf_auxargs->flags = 0; 1053 elf_auxargs->entry = entry; 1054 elf_auxargs->hdr_eflags = hdr->e_flags; 1055 1056 imgp->auxargs = elf_auxargs; 1057 imgp->interpreted = 0; 1058 imgp->reloc_base = addr; 1059 imgp->proc->p_osrel = osrel; 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 *, 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 copyin(base, buf, chunk_len); 1206 error = gzio_write(p->gzs, buf, chunk_len); 1207 if (error != 0) 1208 break; 1209 base += chunk_len; 1210 len -= chunk_len; 1211 } 1212 return (error); 1213 } 1214 1215 static int 1216 core_gz_write(void *base, size_t len, off_t offset, void *arg) 1217 { 1218 1219 return (core_write((struct coredump_params *)arg, base, len, offset, 1220 UIO_SYSSPACE)); 1221 } 1222 #endif /* GZIO */ 1223 1224 static int 1225 core_write(struct coredump_params *p, void *base, size_t len, off_t offset, 1226 enum uio_seg seg) 1227 { 1228 1229 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset, 1230 seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1231 p->active_cred, p->file_cred, NULL, p->td)); 1232 } 1233 1234 static int 1235 core_output(void *base, size_t len, off_t offset, struct coredump_params *p, 1236 void *tmpbuf) 1237 { 1238 1239 #ifdef GZIO 1240 if (p->gzs != NULL) 1241 return (compress_chunk(p, base, tmpbuf, len)); 1242 #endif 1243 return (core_write(p, base, len, offset, UIO_USERSPACE)); 1244 } 1245 1246 /* 1247 * Drain into a core file. 1248 */ 1249 static int 1250 sbuf_drain_core_output(void *arg, const char *data, int len) 1251 { 1252 struct coredump_params *p; 1253 int error, locked; 1254 1255 p = (struct coredump_params *)arg; 1256 1257 /* 1258 * Some kern_proc out routines that print to this sbuf may 1259 * call us with the process lock held. Draining with the 1260 * non-sleepable lock held is unsafe. The lock is needed for 1261 * those routines when dumping a live process. In our case we 1262 * can safely release the lock before draining and acquire 1263 * again after. 1264 */ 1265 locked = PROC_LOCKED(p->td->td_proc); 1266 if (locked) 1267 PROC_UNLOCK(p->td->td_proc); 1268 #ifdef GZIO 1269 if (p->gzs != NULL) 1270 error = gzio_write(p->gzs, __DECONST(char *, data), len); 1271 else 1272 #endif 1273 error = core_write(p, __DECONST(void *, data), len, p->offset, 1274 UIO_SYSSPACE); 1275 if (locked) 1276 PROC_LOCK(p->td->td_proc); 1277 if (error != 0) 1278 return (-error); 1279 p->offset += len; 1280 return (len); 1281 } 1282 1283 /* 1284 * Drain into a counter. 1285 */ 1286 static int 1287 sbuf_drain_count(void *arg, const char *data __unused, int len) 1288 { 1289 size_t *sizep; 1290 1291 sizep = (size_t *)arg; 1292 *sizep += len; 1293 return (len); 1294 } 1295 1296 int 1297 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) 1298 { 1299 struct ucred *cred = td->td_ucred; 1300 int error = 0; 1301 struct sseg_closure seginfo; 1302 struct note_info_list notelst; 1303 struct coredump_params params; 1304 struct note_info *ninfo; 1305 void *hdr, *tmpbuf; 1306 size_t hdrsize, notesz, coresize; 1307 #ifdef GZIO 1308 boolean_t compress; 1309 1310 compress = (flags & IMGACT_CORE_COMPRESS) != 0; 1311 #endif 1312 hdr = NULL; 1313 tmpbuf = NULL; 1314 TAILQ_INIT(¬elst); 1315 1316 /* Size the program segments. */ 1317 seginfo.count = 0; 1318 seginfo.size = 0; 1319 each_dumpable_segment(td, cb_size_segment, &seginfo); 1320 1321 /* 1322 * Collect info about the core file header area. 1323 */ 1324 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); 1325 if (seginfo.count + 1 >= PN_XNUM) 1326 hdrsize += sizeof(Elf_Shdr); 1327 __elfN(prepare_notes)(td, ¬elst, ¬esz); 1328 coresize = round_page(hdrsize + notesz) + seginfo.size; 1329 1330 /* Set up core dump parameters. */ 1331 params.offset = 0; 1332 params.active_cred = cred; 1333 params.file_cred = NOCRED; 1334 params.td = td; 1335 params.vp = vp; 1336 params.gzs = NULL; 1337 1338 #ifdef RACCT 1339 if (racct_enable) { 1340 PROC_LOCK(td->td_proc); 1341 error = racct_add(td->td_proc, RACCT_CORE, coresize); 1342 PROC_UNLOCK(td->td_proc); 1343 if (error != 0) { 1344 error = EFAULT; 1345 goto done; 1346 } 1347 } 1348 #endif 1349 if (coresize >= limit) { 1350 error = EFAULT; 1351 goto done; 1352 } 1353 1354 #ifdef GZIO 1355 /* Create a compression stream if necessary. */ 1356 if (compress) { 1357 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE, 1358 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms); 1359 if (params.gzs == NULL) { 1360 error = EFAULT; 1361 goto done; 1362 } 1363 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); 1364 } 1365 #endif 1366 1367 /* 1368 * Allocate memory for building the header, fill it up, 1369 * and write it out following the notes. 1370 */ 1371 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 1372 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, 1373 notesz); 1374 1375 /* Write the contents of all of the writable segments. */ 1376 if (error == 0) { 1377 Elf_Phdr *php; 1378 off_t offset; 1379 int i; 1380 1381 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 1382 offset = round_page(hdrsize + notesz); 1383 for (i = 0; i < seginfo.count; i++) { 1384 error = core_output((caddr_t)(uintptr_t)php->p_vaddr, 1385 php->p_filesz, offset, ¶ms, tmpbuf); 1386 if (error != 0) 1387 break; 1388 offset += php->p_filesz; 1389 php++; 1390 } 1391 #ifdef GZIO 1392 if (error == 0 && compress) 1393 error = gzio_flush(params.gzs); 1394 #endif 1395 } 1396 if (error) { 1397 log(LOG_WARNING, 1398 "Failed to write core file for process %s (error %d)\n", 1399 curproc->p_comm, error); 1400 } 1401 1402 done: 1403 #ifdef GZIO 1404 if (compress) { 1405 free(tmpbuf, M_TEMP); 1406 if (params.gzs != NULL) 1407 gzio_fini(params.gzs); 1408 } 1409 #endif 1410 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { 1411 TAILQ_REMOVE(¬elst, ninfo, link); 1412 free(ninfo, M_TEMP); 1413 } 1414 if (hdr != NULL) 1415 free(hdr, M_TEMP); 1416 1417 return (error); 1418 } 1419 1420 /* 1421 * A callback for each_dumpable_segment() to write out the segment's 1422 * program header entry. 1423 */ 1424 static void 1425 cb_put_phdr(entry, closure) 1426 vm_map_entry_t entry; 1427 void *closure; 1428 { 1429 struct phdr_closure *phc = (struct phdr_closure *)closure; 1430 Elf_Phdr *phdr = phc->phdr; 1431 1432 phc->offset = round_page(phc->offset); 1433 1434 phdr->p_type = PT_LOAD; 1435 phdr->p_offset = phc->offset; 1436 phdr->p_vaddr = entry->start; 1437 phdr->p_paddr = 0; 1438 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 1439 phdr->p_align = PAGE_SIZE; 1440 phdr->p_flags = __elfN(untrans_prot)(entry->protection); 1441 1442 phc->offset += phdr->p_filesz; 1443 phc->phdr++; 1444 } 1445 1446 /* 1447 * A callback for each_dumpable_segment() to gather information about 1448 * the number of segments and their total size. 1449 */ 1450 static void 1451 cb_size_segment(vm_map_entry_t entry, void *closure) 1452 { 1453 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1454 1455 ssc->count++; 1456 ssc->size += entry->end - entry->start; 1457 } 1458 1459 /* 1460 * For each writable segment in the process's memory map, call the given 1461 * function with a pointer to the map entry and some arbitrary 1462 * caller-supplied data. 1463 */ 1464 static void 1465 each_dumpable_segment(struct thread *td, segment_callback func, void *closure) 1466 { 1467 struct proc *p = td->td_proc; 1468 vm_map_t map = &p->p_vmspace->vm_map; 1469 vm_map_entry_t entry; 1470 vm_object_t backing_object, object; 1471 boolean_t ignore_entry; 1472 1473 vm_map_lock_read(map); 1474 for (entry = map->header.next; entry != &map->header; 1475 entry = entry->next) { 1476 /* 1477 * Don't dump inaccessible mappings, deal with legacy 1478 * coredump mode. 1479 * 1480 * Note that read-only segments related to the elf binary 1481 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1482 * need to arbitrarily ignore such segments. 1483 */ 1484 if (elf_legacy_coredump) { 1485 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1486 continue; 1487 } else { 1488 if ((entry->protection & VM_PROT_ALL) == 0) 1489 continue; 1490 } 1491 1492 /* 1493 * Dont include memory segment in the coredump if 1494 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1495 * madvise(2). Do not dump submaps (i.e. parts of the 1496 * kernel map). 1497 */ 1498 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1499 continue; 1500 1501 if ((object = entry->object.vm_object) == NULL) 1502 continue; 1503 1504 /* Ignore memory-mapped devices and such things. */ 1505 VM_OBJECT_RLOCK(object); 1506 while ((backing_object = object->backing_object) != NULL) { 1507 VM_OBJECT_RLOCK(backing_object); 1508 VM_OBJECT_RUNLOCK(object); 1509 object = backing_object; 1510 } 1511 ignore_entry = object->type != OBJT_DEFAULT && 1512 object->type != OBJT_SWAP && object->type != OBJT_VNODE && 1513 object->type != OBJT_PHYS; 1514 VM_OBJECT_RUNLOCK(object); 1515 if (ignore_entry) 1516 continue; 1517 1518 (*func)(entry, closure); 1519 } 1520 vm_map_unlock_read(map); 1521 } 1522 1523 /* 1524 * Write the core file header to the file, including padding up to 1525 * the page boundary. 1526 */ 1527 static int 1528 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, 1529 size_t hdrsize, struct note_info_list *notelst, size_t notesz) 1530 { 1531 struct note_info *ninfo; 1532 struct sbuf *sb; 1533 int error; 1534 1535 /* Fill in the header. */ 1536 bzero(hdr, hdrsize); 1537 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz); 1538 1539 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); 1540 sbuf_set_drain(sb, sbuf_drain_core_output, p); 1541 sbuf_start_section(sb, NULL); 1542 sbuf_bcat(sb, hdr, hdrsize); 1543 TAILQ_FOREACH(ninfo, notelst, link) 1544 __elfN(putnote)(ninfo, sb); 1545 /* Align up to a page boundary for the program segments. */ 1546 sbuf_end_section(sb, -1, PAGE_SIZE, 0); 1547 error = sbuf_finish(sb); 1548 sbuf_delete(sb); 1549 1550 return (error); 1551 } 1552 1553 static void 1554 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, 1555 size_t *sizep) 1556 { 1557 struct proc *p; 1558 struct thread *thr; 1559 size_t size; 1560 1561 p = td->td_proc; 1562 size = 0; 1563 1564 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); 1565 1566 /* 1567 * To have the debugger select the right thread (LWP) as the initial 1568 * thread, we dump the state of the thread passed to us in td first. 1569 * This is the thread that causes the core dump and thus likely to 1570 * be the right thread one wants to have selected in the debugger. 1571 */ 1572 thr = td; 1573 while (thr != NULL) { 1574 size += register_note(list, NT_PRSTATUS, 1575 __elfN(note_prstatus), thr); 1576 size += register_note(list, NT_FPREGSET, 1577 __elfN(note_fpregset), thr); 1578 size += register_note(list, NT_THRMISC, 1579 __elfN(note_thrmisc), thr); 1580 size += register_note(list, -1, 1581 __elfN(note_threadmd), thr); 1582 1583 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1584 TAILQ_NEXT(thr, td_plist); 1585 if (thr == td) 1586 thr = TAILQ_NEXT(thr, td_plist); 1587 } 1588 1589 size += register_note(list, NT_PROCSTAT_PROC, 1590 __elfN(note_procstat_proc), p); 1591 size += register_note(list, NT_PROCSTAT_FILES, 1592 note_procstat_files, p); 1593 size += register_note(list, NT_PROCSTAT_VMMAP, 1594 note_procstat_vmmap, p); 1595 size += register_note(list, NT_PROCSTAT_GROUPS, 1596 note_procstat_groups, p); 1597 size += register_note(list, NT_PROCSTAT_UMASK, 1598 note_procstat_umask, p); 1599 size += register_note(list, NT_PROCSTAT_RLIMIT, 1600 note_procstat_rlimit, p); 1601 size += register_note(list, NT_PROCSTAT_OSREL, 1602 note_procstat_osrel, p); 1603 size += register_note(list, NT_PROCSTAT_PSSTRINGS, 1604 __elfN(note_procstat_psstrings), p); 1605 size += register_note(list, NT_PROCSTAT_AUXV, 1606 __elfN(note_procstat_auxv), p); 1607 1608 *sizep = size; 1609 } 1610 1611 static void 1612 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, 1613 size_t notesz) 1614 { 1615 Elf_Ehdr *ehdr; 1616 Elf_Phdr *phdr; 1617 Elf_Shdr *shdr; 1618 struct phdr_closure phc; 1619 1620 ehdr = (Elf_Ehdr *)hdr; 1621 1622 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1623 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1624 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1625 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1626 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1627 ehdr->e_ident[EI_DATA] = ELF_DATA; 1628 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1629 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1630 ehdr->e_ident[EI_ABIVERSION] = 0; 1631 ehdr->e_ident[EI_PAD] = 0; 1632 ehdr->e_type = ET_CORE; 1633 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1634 ehdr->e_machine = ELF_ARCH32; 1635 #else 1636 ehdr->e_machine = ELF_ARCH; 1637 #endif 1638 ehdr->e_version = EV_CURRENT; 1639 ehdr->e_entry = 0; 1640 ehdr->e_phoff = sizeof(Elf_Ehdr); 1641 ehdr->e_flags = 0; 1642 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1643 ehdr->e_phentsize = sizeof(Elf_Phdr); 1644 ehdr->e_shentsize = sizeof(Elf_Shdr); 1645 ehdr->e_shstrndx = SHN_UNDEF; 1646 if (numsegs + 1 < PN_XNUM) { 1647 ehdr->e_phnum = numsegs + 1; 1648 ehdr->e_shnum = 0; 1649 } else { 1650 ehdr->e_phnum = PN_XNUM; 1651 ehdr->e_shnum = 1; 1652 1653 ehdr->e_shoff = ehdr->e_phoff + 1654 (numsegs + 1) * ehdr->e_phentsize; 1655 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr), 1656 ("e_shoff: %zu, hdrsize - shdr: %zu", 1657 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr))); 1658 1659 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff); 1660 memset(shdr, 0, sizeof(*shdr)); 1661 /* 1662 * A special first section is used to hold large segment and 1663 * section counts. This was proposed by Sun Microsystems in 1664 * Solaris and has been adopted by Linux; the standard ELF 1665 * tools are already familiar with the technique. 1666 * 1667 * See table 7-7 of the Solaris "Linker and Libraries Guide" 1668 * (or 12-7 depending on the version of the document) for more 1669 * details. 1670 */ 1671 shdr->sh_type = SHT_NULL; 1672 shdr->sh_size = ehdr->e_shnum; 1673 shdr->sh_link = ehdr->e_shstrndx; 1674 shdr->sh_info = numsegs + 1; 1675 } 1676 1677 /* 1678 * Fill in the program header entries. 1679 */ 1680 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff); 1681 1682 /* The note segement. */ 1683 phdr->p_type = PT_NOTE; 1684 phdr->p_offset = hdrsize; 1685 phdr->p_vaddr = 0; 1686 phdr->p_paddr = 0; 1687 phdr->p_filesz = notesz; 1688 phdr->p_memsz = 0; 1689 phdr->p_flags = PF_R; 1690 phdr->p_align = ELF_NOTE_ROUNDSIZE; 1691 phdr++; 1692 1693 /* All the writable segments from the program. */ 1694 phc.phdr = phdr; 1695 phc.offset = round_page(hdrsize + notesz); 1696 each_dumpable_segment(td, cb_put_phdr, &phc); 1697 } 1698 1699 static size_t 1700 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) 1701 { 1702 struct note_info *ninfo; 1703 size_t size, notesize; 1704 1705 size = 0; 1706 out(arg, NULL, &size); 1707 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); 1708 ninfo->type = type; 1709 ninfo->outfunc = out; 1710 ninfo->outarg = arg; 1711 ninfo->outsize = size; 1712 TAILQ_INSERT_TAIL(list, ninfo, link); 1713 1714 if (type == -1) 1715 return (size); 1716 1717 notesize = sizeof(Elf_Note) + /* note header */ 1718 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1719 /* note name */ 1720 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1721 1722 return (notesize); 1723 } 1724 1725 static size_t 1726 append_note_data(const void *src, void *dst, size_t len) 1727 { 1728 size_t padded_len; 1729 1730 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); 1731 if (dst != NULL) { 1732 bcopy(src, dst, len); 1733 bzero((char *)dst + len, padded_len - len); 1734 } 1735 return (padded_len); 1736 } 1737 1738 size_t 1739 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) 1740 { 1741 Elf_Note *note; 1742 char *buf; 1743 size_t notesize; 1744 1745 buf = dst; 1746 if (buf != NULL) { 1747 note = (Elf_Note *)buf; 1748 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1749 note->n_descsz = size; 1750 note->n_type = type; 1751 buf += sizeof(*note); 1752 buf += append_note_data(FREEBSD_ABI_VENDOR, buf, 1753 sizeof(FREEBSD_ABI_VENDOR)); 1754 append_note_data(src, buf, size); 1755 if (descp != NULL) 1756 *descp = buf; 1757 } 1758 1759 notesize = sizeof(Elf_Note) + /* note header */ 1760 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + 1761 /* note name */ 1762 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ 1763 1764 return (notesize); 1765 } 1766 1767 static void 1768 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) 1769 { 1770 Elf_Note note; 1771 ssize_t old_len, sect_len; 1772 size_t new_len, descsz, i; 1773 1774 if (ninfo->type == -1) { 1775 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1776 return; 1777 } 1778 1779 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR); 1780 note.n_descsz = ninfo->outsize; 1781 note.n_type = ninfo->type; 1782 1783 sbuf_bcat(sb, ¬e, sizeof(note)); 1784 sbuf_start_section(sb, &old_len); 1785 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR)); 1786 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1787 if (note.n_descsz == 0) 1788 return; 1789 sbuf_start_section(sb, &old_len); 1790 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); 1791 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); 1792 if (sect_len < 0) 1793 return; 1794 1795 new_len = (size_t)sect_len; 1796 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE); 1797 if (new_len < descsz) { 1798 /* 1799 * It is expected that individual note emitters will correctly 1800 * predict their expected output size and fill up to that size 1801 * themselves, padding in a format-specific way if needed. 1802 * However, in case they don't, just do it here with zeros. 1803 */ 1804 for (i = 0; i < descsz - new_len; i++) 1805 sbuf_putc(sb, 0); 1806 } else if (new_len > descsz) { 1807 /* 1808 * We can't always truncate sb -- we may have drained some 1809 * of it already. 1810 */ 1811 KASSERT(new_len == descsz, ("%s: Note type %u changed as we " 1812 "read it (%zu > %zu). Since it is longer than " 1813 "expected, this coredump's notes are corrupt. THIS " 1814 "IS A BUG in the note_procstat routine for type %u.\n", 1815 __func__, (unsigned)note.n_type, new_len, descsz, 1816 (unsigned)note.n_type)); 1817 } 1818 } 1819 1820 /* 1821 * Miscellaneous note out functions. 1822 */ 1823 1824 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1825 #include <compat/freebsd32/freebsd32.h> 1826 1827 typedef struct prstatus32 elf_prstatus_t; 1828 typedef struct prpsinfo32 elf_prpsinfo_t; 1829 typedef struct fpreg32 elf_prfpregset_t; 1830 typedef struct fpreg32 elf_fpregset_t; 1831 typedef struct reg32 elf_gregset_t; 1832 typedef struct thrmisc32 elf_thrmisc_t; 1833 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 1834 typedef struct kinfo_proc32 elf_kinfo_proc_t; 1835 typedef uint32_t elf_ps_strings_t; 1836 #else 1837 typedef prstatus_t elf_prstatus_t; 1838 typedef prpsinfo_t elf_prpsinfo_t; 1839 typedef prfpregset_t elf_prfpregset_t; 1840 typedef prfpregset_t elf_fpregset_t; 1841 typedef gregset_t elf_gregset_t; 1842 typedef thrmisc_t elf_thrmisc_t; 1843 #define ELF_KERN_PROC_MASK 0 1844 typedef struct kinfo_proc elf_kinfo_proc_t; 1845 typedef vm_offset_t elf_ps_strings_t; 1846 #endif 1847 1848 static void 1849 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) 1850 { 1851 struct sbuf sbarg; 1852 size_t len; 1853 char *cp, *end; 1854 struct proc *p; 1855 elf_prpsinfo_t *psinfo; 1856 int error; 1857 1858 p = (struct proc *)arg; 1859 if (sb != NULL) { 1860 KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); 1861 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); 1862 psinfo->pr_version = PRPSINFO_VERSION; 1863 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1864 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1865 PROC_LOCK(p); 1866 if (p->p_args != NULL) { 1867 len = sizeof(psinfo->pr_psargs) - 1; 1868 if (len > p->p_args->ar_length) 1869 len = p->p_args->ar_length; 1870 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len); 1871 PROC_UNLOCK(p); 1872 error = 0; 1873 } else { 1874 _PHOLD(p); 1875 PROC_UNLOCK(p); 1876 sbuf_new(&sbarg, psinfo->pr_psargs, 1877 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN); 1878 error = proc_getargv(curthread, p, &sbarg); 1879 PRELE(p); 1880 if (sbuf_finish(&sbarg) == 0) 1881 len = sbuf_len(&sbarg) - 1; 1882 else 1883 len = sizeof(psinfo->pr_psargs) - 1; 1884 sbuf_delete(&sbarg); 1885 } 1886 if (error || len == 0) 1887 strlcpy(psinfo->pr_psargs, p->p_comm, 1888 sizeof(psinfo->pr_psargs)); 1889 else { 1890 KASSERT(len < sizeof(psinfo->pr_psargs), 1891 ("len is too long: %zu vs %zu", len, 1892 sizeof(psinfo->pr_psargs))); 1893 cp = psinfo->pr_psargs; 1894 end = cp + len - 1; 1895 for (;;) { 1896 cp = memchr(cp, '\0', end - cp); 1897 if (cp == NULL) 1898 break; 1899 *cp = ' '; 1900 } 1901 } 1902 psinfo->pr_pid = p->p_pid; 1903 sbuf_bcat(sb, psinfo, sizeof(*psinfo)); 1904 free(psinfo, M_TEMP); 1905 } 1906 *sizep = sizeof(*psinfo); 1907 } 1908 1909 static void 1910 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) 1911 { 1912 struct thread *td; 1913 elf_prstatus_t *status; 1914 1915 td = (struct thread *)arg; 1916 if (sb != NULL) { 1917 KASSERT(*sizep == sizeof(*status), ("invalid size")); 1918 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); 1919 status->pr_version = PRSTATUS_VERSION; 1920 status->pr_statussz = sizeof(elf_prstatus_t); 1921 status->pr_gregsetsz = sizeof(elf_gregset_t); 1922 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1923 status->pr_osreldate = osreldate; 1924 status->pr_cursig = td->td_proc->p_sig; 1925 status->pr_pid = td->td_tid; 1926 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1927 fill_regs32(td, &status->pr_reg); 1928 #else 1929 fill_regs(td, &status->pr_reg); 1930 #endif 1931 sbuf_bcat(sb, status, sizeof(*status)); 1932 free(status, M_TEMP); 1933 } 1934 *sizep = sizeof(*status); 1935 } 1936 1937 static void 1938 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) 1939 { 1940 struct thread *td; 1941 elf_prfpregset_t *fpregset; 1942 1943 td = (struct thread *)arg; 1944 if (sb != NULL) { 1945 KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); 1946 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); 1947 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 1948 fill_fpregs32(td, fpregset); 1949 #else 1950 fill_fpregs(td, fpregset); 1951 #endif 1952 sbuf_bcat(sb, fpregset, sizeof(*fpregset)); 1953 free(fpregset, M_TEMP); 1954 } 1955 *sizep = sizeof(*fpregset); 1956 } 1957 1958 static void 1959 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) 1960 { 1961 struct thread *td; 1962 elf_thrmisc_t thrmisc; 1963 1964 td = (struct thread *)arg; 1965 if (sb != NULL) { 1966 KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); 1967 bzero(&thrmisc._pad, sizeof(thrmisc._pad)); 1968 strcpy(thrmisc.pr_tname, td->td_name); 1969 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); 1970 } 1971 *sizep = sizeof(thrmisc); 1972 } 1973 1974 /* 1975 * Allow for MD specific notes, as well as any MD 1976 * specific preparations for writing MI notes. 1977 */ 1978 static void 1979 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) 1980 { 1981 struct thread *td; 1982 void *buf; 1983 size_t size; 1984 1985 td = (struct thread *)arg; 1986 size = *sizep; 1987 if (size != 0 && sb != NULL) 1988 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); 1989 else 1990 buf = NULL; 1991 size = 0; 1992 __elfN(dump_thread)(td, buf, &size); 1993 KASSERT(sb == NULL || *sizep == size, ("invalid size")); 1994 if (size != 0 && sb != NULL) 1995 sbuf_bcat(sb, buf, size); 1996 free(buf, M_TEMP); 1997 *sizep = size; 1998 } 1999 2000 #ifdef KINFO_PROC_SIZE 2001 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); 2002 #endif 2003 2004 static void 2005 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) 2006 { 2007 struct proc *p; 2008 size_t size; 2009 int structsize; 2010 2011 p = (struct proc *)arg; 2012 size = sizeof(structsize) + p->p_numthreads * 2013 sizeof(elf_kinfo_proc_t); 2014 2015 if (sb != NULL) { 2016 KASSERT(*sizep == size, ("invalid size")); 2017 structsize = sizeof(elf_kinfo_proc_t); 2018 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2019 sx_slock(&proctree_lock); 2020 PROC_LOCK(p); 2021 kern_proc_out(p, sb, ELF_KERN_PROC_MASK); 2022 sx_sunlock(&proctree_lock); 2023 } 2024 *sizep = size; 2025 } 2026 2027 #ifdef KINFO_FILE_SIZE 2028 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 2029 #endif 2030 2031 static void 2032 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) 2033 { 2034 struct proc *p; 2035 size_t size, sect_sz, i; 2036 ssize_t start_len, sect_len; 2037 int structsize, filedesc_flags; 2038 2039 if (coredump_pack_fileinfo) 2040 filedesc_flags = KERN_FILEDESC_PACK_KINFO; 2041 else 2042 filedesc_flags = 0; 2043 2044 p = (struct proc *)arg; 2045 structsize = sizeof(struct kinfo_file); 2046 if (sb == NULL) { 2047 size = 0; 2048 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2049 sbuf_set_drain(sb, sbuf_drain_count, &size); 2050 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2051 PROC_LOCK(p); 2052 kern_proc_filedesc_out(p, sb, -1, filedesc_flags); 2053 sbuf_finish(sb); 2054 sbuf_delete(sb); 2055 *sizep = size; 2056 } else { 2057 sbuf_start_section(sb, &start_len); 2058 2059 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2060 PROC_LOCK(p); 2061 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize), 2062 filedesc_flags); 2063 2064 sect_len = sbuf_end_section(sb, start_len, 0, 0); 2065 if (sect_len < 0) 2066 return; 2067 sect_sz = sect_len; 2068 2069 KASSERT(sect_sz <= *sizep, 2070 ("kern_proc_filedesc_out did not respect maxlen; " 2071 "requested %zu, got %zu", *sizep - sizeof(structsize), 2072 sect_sz - sizeof(structsize))); 2073 2074 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++) 2075 sbuf_putc(sb, 0); 2076 } 2077 } 2078 2079 #ifdef KINFO_VMENTRY_SIZE 2080 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); 2081 #endif 2082 2083 static void 2084 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) 2085 { 2086 struct proc *p; 2087 size_t size; 2088 int structsize, vmmap_flags; 2089 2090 if (coredump_pack_vmmapinfo) 2091 vmmap_flags = KERN_VMMAP_PACK_KINFO; 2092 else 2093 vmmap_flags = 0; 2094 2095 p = (struct proc *)arg; 2096 structsize = sizeof(struct kinfo_vmentry); 2097 if (sb == NULL) { 2098 size = 0; 2099 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2100 sbuf_set_drain(sb, sbuf_drain_count, &size); 2101 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2102 PROC_LOCK(p); 2103 kern_proc_vmmap_out(p, sb, -1, vmmap_flags); 2104 sbuf_finish(sb); 2105 sbuf_delete(sb); 2106 *sizep = size; 2107 } else { 2108 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2109 PROC_LOCK(p); 2110 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize), 2111 vmmap_flags); 2112 } 2113 } 2114 2115 static void 2116 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) 2117 { 2118 struct proc *p; 2119 size_t size; 2120 int structsize; 2121 2122 p = (struct proc *)arg; 2123 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); 2124 if (sb != NULL) { 2125 KASSERT(*sizep == size, ("invalid size")); 2126 structsize = sizeof(gid_t); 2127 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2128 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * 2129 sizeof(gid_t)); 2130 } 2131 *sizep = size; 2132 } 2133 2134 static void 2135 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) 2136 { 2137 struct proc *p; 2138 size_t size; 2139 int structsize; 2140 2141 p = (struct proc *)arg; 2142 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); 2143 if (sb != NULL) { 2144 KASSERT(*sizep == size, ("invalid size")); 2145 structsize = sizeof(p->p_fd->fd_cmask); 2146 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2147 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); 2148 } 2149 *sizep = size; 2150 } 2151 2152 static void 2153 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) 2154 { 2155 struct proc *p; 2156 struct rlimit rlim[RLIM_NLIMITS]; 2157 size_t size; 2158 int structsize, i; 2159 2160 p = (struct proc *)arg; 2161 size = sizeof(structsize) + sizeof(rlim); 2162 if (sb != NULL) { 2163 KASSERT(*sizep == size, ("invalid size")); 2164 structsize = sizeof(rlim); 2165 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2166 PROC_LOCK(p); 2167 for (i = 0; i < RLIM_NLIMITS; i++) 2168 lim_rlimit_proc(p, i, &rlim[i]); 2169 PROC_UNLOCK(p); 2170 sbuf_bcat(sb, rlim, sizeof(rlim)); 2171 } 2172 *sizep = size; 2173 } 2174 2175 static void 2176 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) 2177 { 2178 struct proc *p; 2179 size_t size; 2180 int structsize; 2181 2182 p = (struct proc *)arg; 2183 size = sizeof(structsize) + sizeof(p->p_osrel); 2184 if (sb != NULL) { 2185 KASSERT(*sizep == size, ("invalid size")); 2186 structsize = sizeof(p->p_osrel); 2187 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2188 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); 2189 } 2190 *sizep = size; 2191 } 2192 2193 static void 2194 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) 2195 { 2196 struct proc *p; 2197 elf_ps_strings_t ps_strings; 2198 size_t size; 2199 int structsize; 2200 2201 p = (struct proc *)arg; 2202 size = sizeof(structsize) + sizeof(ps_strings); 2203 if (sb != NULL) { 2204 KASSERT(*sizep == size, ("invalid size")); 2205 structsize = sizeof(ps_strings); 2206 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 2207 ps_strings = PTROUT(p->p_sysent->sv_psstrings); 2208 #else 2209 ps_strings = p->p_sysent->sv_psstrings; 2210 #endif 2211 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2212 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); 2213 } 2214 *sizep = size; 2215 } 2216 2217 static void 2218 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) 2219 { 2220 struct proc *p; 2221 size_t size; 2222 int structsize; 2223 2224 p = (struct proc *)arg; 2225 if (sb == NULL) { 2226 size = 0; 2227 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); 2228 sbuf_set_drain(sb, sbuf_drain_count, &size); 2229 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2230 PHOLD(p); 2231 proc_getauxv(curthread, p, sb); 2232 PRELE(p); 2233 sbuf_finish(sb); 2234 sbuf_delete(sb); 2235 *sizep = size; 2236 } else { 2237 structsize = sizeof(Elf_Auxinfo); 2238 sbuf_bcat(sb, &structsize, sizeof(structsize)); 2239 PHOLD(p); 2240 proc_getauxv(curthread, p, sb); 2241 PRELE(p); 2242 } 2243 } 2244 2245 static boolean_t 2246 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, 2247 int32_t *osrel, const Elf_Phdr *pnote) 2248 { 2249 const Elf_Note *note, *note0, *note_end; 2250 const char *note_name; 2251 char *buf; 2252 int i, error; 2253 boolean_t res; 2254 2255 /* We need some limit, might as well use PAGE_SIZE. */ 2256 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE) 2257 return (FALSE); 2258 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes"); 2259 if (pnote->p_offset > PAGE_SIZE || 2260 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) { 2261 VOP_UNLOCK(imgp->vp, 0); 2262 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK); 2263 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 2264 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz, 2265 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED, 2266 curthread->td_ucred, NOCRED, NULL, curthread); 2267 if (error != 0) { 2268 uprintf("i/o error PT_NOTE\n"); 2269 res = FALSE; 2270 goto ret; 2271 } 2272 note = note0 = (const Elf_Note *)buf; 2273 note_end = (const Elf_Note *)(buf + pnote->p_filesz); 2274 } else { 2275 note = note0 = (const Elf_Note *)(imgp->image_header + 2276 pnote->p_offset); 2277 note_end = (const Elf_Note *)(imgp->image_header + 2278 pnote->p_offset + pnote->p_filesz); 2279 buf = NULL; 2280 } 2281 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { 2282 if (!aligned(note, Elf32_Addr) || (const char *)note_end - 2283 (const char *)note < sizeof(Elf_Note)) { 2284 res = FALSE; 2285 goto ret; 2286 } 2287 if (note->n_namesz != checknote->hdr.n_namesz || 2288 note->n_descsz != checknote->hdr.n_descsz || 2289 note->n_type != checknote->hdr.n_type) 2290 goto nextnote; 2291 note_name = (const char *)(note + 1); 2292 if (note_name + checknote->hdr.n_namesz >= 2293 (const char *)note_end || strncmp(checknote->vendor, 2294 note_name, checknote->hdr.n_namesz) != 0) 2295 goto nextnote; 2296 2297 /* 2298 * Fetch the osreldate for binary 2299 * from the ELF OSABI-note if necessary. 2300 */ 2301 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && 2302 checknote->trans_osrel != NULL) { 2303 res = checknote->trans_osrel(note, osrel); 2304 goto ret; 2305 } 2306 res = TRUE; 2307 goto ret; 2308 nextnote: 2309 note = (const Elf_Note *)((const char *)(note + 1) + 2310 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + 2311 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); 2312 } 2313 res = FALSE; 2314 ret: 2315 free(buf, M_TEMP); 2316 return (res); 2317 } 2318 2319 /* 2320 * Try to find the appropriate ABI-note section for checknote, 2321 * fetch the osreldate for binary from the ELF OSABI-note. Only the 2322 * first page of the image is searched, the same as for headers. 2323 */ 2324 static boolean_t 2325 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, 2326 int32_t *osrel) 2327 { 2328 const Elf_Phdr *phdr; 2329 const Elf_Ehdr *hdr; 2330 int i; 2331 2332 hdr = (const Elf_Ehdr *)imgp->image_header; 2333 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 2334 2335 for (i = 0; i < hdr->e_phnum; i++) { 2336 if (phdr[i].p_type == PT_NOTE && 2337 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) 2338 return (TRUE); 2339 } 2340 return (FALSE); 2341 2342 } 2343 2344 /* 2345 * Tell kern_execve.c about it, with a little help from the linker. 2346 */ 2347 static struct execsw __elfN(execsw) = { 2348 __CONCAT(exec_, __elfN(imgact)), 2349 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 2350 }; 2351 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 2352 2353 static vm_prot_t 2354 __elfN(trans_prot)(Elf_Word flags) 2355 { 2356 vm_prot_t prot; 2357 2358 prot = 0; 2359 if (flags & PF_X) 2360 prot |= VM_PROT_EXECUTE; 2361 if (flags & PF_W) 2362 prot |= VM_PROT_WRITE; 2363 if (flags & PF_R) 2364 prot |= VM_PROT_READ; 2365 #if __ELF_WORD_SIZE == 32 2366 #if defined(__amd64__) 2367 if (i386_read_exec && (flags & PF_R)) 2368 prot |= VM_PROT_EXECUTE; 2369 #endif 2370 #endif 2371 return (prot); 2372 } 2373 2374 static Elf_Word 2375 __elfN(untrans_prot)(vm_prot_t prot) 2376 { 2377 Elf_Word flags; 2378 2379 flags = 0; 2380 if (prot & VM_PROT_EXECUTE) 2381 flags |= PF_X; 2382 if (prot & VM_PROT_READ) 2383 flags |= PF_R; 2384 if (prot & VM_PROT_WRITE) 2385 flags |= PF_W; 2386 return (flags); 2387 } 2388