1 /*- 2 * Copyright (c) 2000 David O'Brien 3 * Copyright (c) 1995-1996 S�ren Schmidt 4 * Copyright (c) 1996 Peter Wemm 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer 12 * in this position and unchanged. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. The name of the author may not be used to endorse or promote products 17 * derived from this software without specific prior written permission 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_compat.h" 35 36 #include <sys/param.h> 37 #include <sys/exec.h> 38 #include <sys/fcntl.h> 39 #include <sys/imgact.h> 40 #include <sys/imgact_elf.h> 41 #include <sys/kernel.h> 42 #include <sys/lock.h> 43 #include <sys/malloc.h> 44 #include <sys/mount.h> 45 #include <sys/mutex.h> 46 #include <sys/mman.h> 47 #include <sys/namei.h> 48 #include <sys/pioctl.h> 49 #include <sys/proc.h> 50 #include <sys/procfs.h> 51 #include <sys/resourcevar.h> 52 #include <sys/sf_buf.h> 53 #include <sys/systm.h> 54 #include <sys/signalvar.h> 55 #include <sys/stat.h> 56 #include <sys/sx.h> 57 #include <sys/syscall.h> 58 #include <sys/sysctl.h> 59 #include <sys/sysent.h> 60 #include <sys/vnode.h> 61 62 #include <vm/vm.h> 63 #include <vm/vm_kern.h> 64 #include <vm/vm_param.h> 65 #include <vm/pmap.h> 66 #include <vm/vm_map.h> 67 #include <vm/vm_object.h> 68 #include <vm/vm_extern.h> 69 70 #include <machine/elf.h> 71 #include <machine/md_var.h> 72 73 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 74 #include <machine/fpu.h> 75 #include <compat/ia32/ia32_reg.h> 76 #endif 77 78 #define OLD_EI_BRAND 8 79 80 static int __elfN(check_header)(const Elf_Ehdr *hdr); 81 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr, 82 const char *interp); 83 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 84 u_long *entry, size_t pagesize); 85 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object, 86 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 87 vm_prot_t prot, size_t pagesize); 88 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 89 90 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 91 ""); 92 93 int __elfN(fallback_brand) = -1; 94 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 95 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 96 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 97 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 98 &__elfN(fallback_brand)); 99 100 static int elf_trace = 0; 101 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, ""); 102 103 static int elf_legacy_coredump = 0; 104 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 105 &elf_legacy_coredump, 0, ""); 106 107 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 108 109 int 110 __elfN(insert_brand_entry)(Elf_Brandinfo *entry) 111 { 112 int i; 113 114 for (i = 0; i < MAX_BRANDS; i++) { 115 if (elf_brand_list[i] == NULL) { 116 elf_brand_list[i] = entry; 117 break; 118 } 119 } 120 if (i == MAX_BRANDS) 121 return (-1); 122 return (0); 123 } 124 125 int 126 __elfN(remove_brand_entry)(Elf_Brandinfo *entry) 127 { 128 int i; 129 130 for (i = 0; i < MAX_BRANDS; i++) { 131 if (elf_brand_list[i] == entry) { 132 elf_brand_list[i] = NULL; 133 break; 134 } 135 } 136 if (i == MAX_BRANDS) 137 return (-1); 138 return (0); 139 } 140 141 int 142 __elfN(brand_inuse)(Elf_Brandinfo *entry) 143 { 144 struct proc *p; 145 int rval = FALSE; 146 147 sx_slock(&allproc_lock); 148 LIST_FOREACH(p, &allproc, p_list) { 149 if (p->p_sysent == entry->sysvec) { 150 rval = TRUE; 151 break; 152 } 153 } 154 sx_sunlock(&allproc_lock); 155 156 return (rval); 157 } 158 159 static Elf_Brandinfo * 160 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp) 161 { 162 Elf_Brandinfo *bi; 163 int i; 164 165 /* 166 * We support three types of branding -- (1) the ELF EI_OSABI field 167 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 168 * branding w/in the ELF header, and (3) path of the `interp_path' 169 * field. We should also look for an ".note.ABI-tag" ELF section now 170 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 171 */ 172 173 /* If the executable has a brand, search for it in the brand list. */ 174 for (i = 0; i < MAX_BRANDS; i++) { 175 bi = elf_brand_list[i]; 176 if (bi != NULL && hdr->e_machine == bi->machine && 177 (hdr->e_ident[EI_OSABI] == bi->brand || 178 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 179 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 180 return (bi); 181 } 182 183 /* Lacking a known brand, search for a recognized interpreter. */ 184 if (interp != NULL) { 185 for (i = 0; i < MAX_BRANDS; i++) { 186 bi = elf_brand_list[i]; 187 if (bi != NULL && hdr->e_machine == bi->machine && 188 strcmp(interp, bi->interp_path) == 0) 189 return (bi); 190 } 191 } 192 193 /* Lacking a recognized interpreter, try the default brand */ 194 for (i = 0; i < MAX_BRANDS; i++) { 195 bi = elf_brand_list[i]; 196 if (bi != NULL && hdr->e_machine == bi->machine && 197 __elfN(fallback_brand) == bi->brand) 198 return (bi); 199 } 200 return (NULL); 201 } 202 203 static int 204 __elfN(check_header)(const Elf_Ehdr *hdr) 205 { 206 Elf_Brandinfo *bi; 207 int i; 208 209 if (!IS_ELF(*hdr) || 210 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 211 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 212 hdr->e_ident[EI_VERSION] != EV_CURRENT || 213 hdr->e_phentsize != sizeof(Elf_Phdr) || 214 hdr->e_version != ELF_TARG_VER) 215 return (ENOEXEC); 216 217 /* 218 * Make sure we have at least one brand for this machine. 219 */ 220 221 for (i = 0; i < MAX_BRANDS; i++) { 222 bi = elf_brand_list[i]; 223 if (bi != NULL && bi->machine == hdr->e_machine) 224 break; 225 } 226 if (i == MAX_BRANDS) 227 return (ENOEXEC); 228 229 return (0); 230 } 231 232 static int 233 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 234 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 235 { 236 struct sf_buf *sf; 237 int error; 238 vm_offset_t off; 239 240 /* 241 * Create the page if it doesn't exist yet. Ignore errors. 242 */ 243 vm_map_lock(map); 244 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 245 VM_PROT_ALL, VM_PROT_ALL, 0); 246 vm_map_unlock(map); 247 248 /* 249 * Find the page from the underlying object. 250 */ 251 if (object) { 252 sf = vm_imgact_map_page(object, offset); 253 if (sf == NULL) 254 return (KERN_FAILURE); 255 off = offset - trunc_page(offset); 256 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 257 end - start); 258 vm_imgact_unmap_page(sf); 259 if (error) { 260 return (KERN_FAILURE); 261 } 262 } 263 264 return (KERN_SUCCESS); 265 } 266 267 static int 268 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 269 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 270 { 271 struct sf_buf *sf; 272 vm_offset_t off; 273 vm_size_t sz; 274 int error, rv; 275 276 if (start != trunc_page(start)) { 277 rv = __elfN(map_partial)(map, object, offset, start, 278 round_page(start), prot); 279 if (rv) 280 return (rv); 281 offset += round_page(start) - start; 282 start = round_page(start); 283 } 284 if (end != round_page(end)) { 285 rv = __elfN(map_partial)(map, object, offset + 286 trunc_page(end) - start, trunc_page(end), end, prot); 287 if (rv) 288 return (rv); 289 end = trunc_page(end); 290 } 291 if (end > start) { 292 if (offset & PAGE_MASK) { 293 /* 294 * The mapping is not page aligned. This means we have 295 * to copy the data. Sigh. 296 */ 297 rv = vm_map_find(map, NULL, 0, &start, end - start, 298 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 299 if (rv) 300 return (rv); 301 if (object == NULL) 302 return (KERN_SUCCESS); 303 for (; start < end; start += sz) { 304 sf = vm_imgact_map_page(object, offset); 305 if (sf == NULL) 306 return (KERN_FAILURE); 307 off = offset - trunc_page(offset); 308 sz = end - start; 309 if (sz > PAGE_SIZE - off) 310 sz = PAGE_SIZE - off; 311 error = copyout((caddr_t)sf_buf_kva(sf) + off, 312 (caddr_t)start, sz); 313 vm_imgact_unmap_page(sf); 314 if (error) { 315 return (KERN_FAILURE); 316 } 317 offset += sz; 318 } 319 rv = KERN_SUCCESS; 320 } else { 321 vm_map_lock(map); 322 rv = vm_map_insert(map, object, offset, start, end, 323 prot, VM_PROT_ALL, cow); 324 vm_map_unlock(map); 325 } 326 return (rv); 327 } else { 328 return (KERN_SUCCESS); 329 } 330 } 331 332 static int 333 __elfN(load_section)(struct vmspace *vmspace, 334 vm_object_t object, vm_offset_t offset, 335 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 336 size_t pagesize) 337 { 338 struct sf_buf *sf; 339 size_t map_len; 340 vm_offset_t map_addr; 341 int error, rv, cow; 342 size_t copy_len; 343 vm_offset_t file_addr; 344 345 /* 346 * It's necessary to fail if the filsz + offset taken from the 347 * header is greater than the actual file pager object's size. 348 * If we were to allow this, then the vm_map_find() below would 349 * walk right off the end of the file object and into the ether. 350 * 351 * While I'm here, might as well check for something else that 352 * is invalid: filsz cannot be greater than memsz. 353 */ 354 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 355 filsz > memsz) { 356 uprintf("elf_load_section: truncated ELF file\n"); 357 return (ENOEXEC); 358 } 359 360 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 361 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 362 363 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 364 file_addr = trunc_page_ps(offset, pagesize); 365 366 /* 367 * We have two choices. We can either clear the data in the last page 368 * of an oversized mapping, or we can start the anon mapping a page 369 * early and copy the initialized data into that first page. We 370 * choose the second.. 371 */ 372 if (memsz > filsz) 373 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 374 else 375 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 376 377 if (map_len != 0) { 378 vm_object_reference(object); 379 380 /* cow flags: don't dump readonly sections in core */ 381 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 382 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 383 384 rv = __elfN(map_insert)(&vmspace->vm_map, 385 object, 386 file_addr, /* file offset */ 387 map_addr, /* virtual start */ 388 map_addr + map_len,/* virtual end */ 389 prot, 390 cow); 391 if (rv != KERN_SUCCESS) { 392 vm_object_deallocate(object); 393 return (EINVAL); 394 } 395 396 /* we can stop now if we've covered it all */ 397 if (memsz == filsz) { 398 return (0); 399 } 400 } 401 402 403 /* 404 * We have to get the remaining bit of the file into the first part 405 * of the oversized map segment. This is normally because the .data 406 * segment in the file is extended to provide bss. It's a neat idea 407 * to try and save a page, but it's a pain in the behind to implement. 408 */ 409 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 410 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 411 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 412 map_addr; 413 414 /* This had damn well better be true! */ 415 if (map_len != 0) { 416 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 417 map_addr + map_len, VM_PROT_ALL, 0); 418 if (rv != KERN_SUCCESS) { 419 return (EINVAL); 420 } 421 } 422 423 if (copy_len != 0) { 424 vm_offset_t off; 425 426 sf = vm_imgact_map_page(object, offset + filsz); 427 if (sf == NULL) 428 return (EIO); 429 430 /* send the page fragment to user space */ 431 off = trunc_page_ps(offset + filsz, pagesize) - 432 trunc_page(offset + filsz); 433 error = copyout((caddr_t)sf_buf_kva(sf) + off, 434 (caddr_t)map_addr, copy_len); 435 vm_imgact_unmap_page(sf); 436 if (error) { 437 return (error); 438 } 439 } 440 441 /* 442 * set it to the specified protection. 443 * XXX had better undo the damage from pasting over the cracks here! 444 */ 445 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 446 round_page(map_addr + map_len), prot, FALSE); 447 448 return (0); 449 } 450 451 /* 452 * Load the file "file" into memory. It may be either a shared object 453 * or an executable. 454 * 455 * The "addr" reference parameter is in/out. On entry, it specifies 456 * the address where a shared object should be loaded. If the file is 457 * an executable, this value is ignored. On exit, "addr" specifies 458 * where the file was actually loaded. 459 * 460 * The "entry" reference parameter is out only. On exit, it specifies 461 * the entry point for the loaded file. 462 */ 463 static int 464 __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 465 u_long *entry, size_t pagesize) 466 { 467 struct { 468 struct nameidata nd; 469 struct vattr attr; 470 struct image_params image_params; 471 } *tempdata; 472 const Elf_Ehdr *hdr = NULL; 473 const Elf_Phdr *phdr = NULL; 474 struct nameidata *nd; 475 struct vmspace *vmspace = p->p_vmspace; 476 struct vattr *attr; 477 struct image_params *imgp; 478 vm_prot_t prot; 479 u_long rbase; 480 u_long base_addr = 0; 481 int vfslocked, error, i, numsegs; 482 483 if (curthread->td_proc != p) 484 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */ 485 486 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 487 nd = &tempdata->nd; 488 attr = &tempdata->attr; 489 imgp = &tempdata->image_params; 490 491 /* 492 * Initialize part of the common data 493 */ 494 imgp->proc = p; 495 imgp->attr = attr; 496 imgp->firstpage = NULL; 497 imgp->image_header = NULL; 498 imgp->object = NULL; 499 imgp->execlabel = NULL; 500 501 /* XXXKSE */ 502 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, 503 curthread); 504 vfslocked = 0; 505 if ((error = namei(nd)) != 0) { 506 nd->ni_vp = NULL; 507 goto fail; 508 } 509 vfslocked = NDHASGIANT(nd); 510 NDFREE(nd, NDF_ONLY_PNBUF); 511 imgp->vp = nd->ni_vp; 512 513 /* 514 * Check permissions, modes, uid, etc on the file, and "open" it. 515 */ 516 error = exec_check_permissions(imgp); 517 if (error) 518 goto fail; 519 520 error = exec_map_first_page(imgp); 521 if (error) 522 goto fail; 523 524 /* 525 * Also make certain that the interpreter stays the same, so set 526 * its VV_TEXT flag, too. 527 */ 528 nd->ni_vp->v_vflag |= VV_TEXT; 529 530 imgp->object = nd->ni_vp->v_object; 531 532 hdr = (const Elf_Ehdr *)imgp->image_header; 533 if ((error = __elfN(check_header)(hdr)) != 0) 534 goto fail; 535 if (hdr->e_type == ET_DYN) 536 rbase = *addr; 537 else if (hdr->e_type == ET_EXEC) 538 rbase = 0; 539 else { 540 error = ENOEXEC; 541 goto fail; 542 } 543 544 /* Only support headers that fit within first page for now */ 545 /* (multiplication of two Elf_Half fields will not overflow) */ 546 if ((hdr->e_phoff > PAGE_SIZE) || 547 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 548 error = ENOEXEC; 549 goto fail; 550 } 551 552 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 553 554 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 555 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 556 prot = 0; 557 if (phdr[i].p_flags & PF_X) 558 prot |= VM_PROT_EXECUTE; 559 if (phdr[i].p_flags & PF_W) 560 prot |= VM_PROT_WRITE; 561 if (phdr[i].p_flags & PF_R) 562 prot |= VM_PROT_READ; 563 564 if ((error = __elfN(load_section)(vmspace, 565 imgp->object, phdr[i].p_offset, 566 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 567 phdr[i].p_memsz, phdr[i].p_filesz, prot, 568 pagesize)) != 0) 569 goto fail; 570 /* 571 * Establish the base address if this is the 572 * first segment. 573 */ 574 if (numsegs == 0) 575 base_addr = trunc_page(phdr[i].p_vaddr + 576 rbase); 577 numsegs++; 578 } 579 } 580 *addr = base_addr; 581 *entry = (unsigned long)hdr->e_entry + rbase; 582 583 fail: 584 if (imgp->firstpage) 585 exec_unmap_first_page(imgp); 586 587 if (nd->ni_vp) 588 vput(nd->ni_vp); 589 590 VFS_UNLOCK_GIANT(vfslocked); 591 free(tempdata, M_TEMP); 592 593 return (error); 594 } 595 596 static int 597 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 598 { 599 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 600 const Elf_Phdr *phdr; 601 Elf_Auxargs *elf_auxargs = NULL; 602 struct vmspace *vmspace; 603 vm_prot_t prot; 604 u_long text_size = 0, data_size = 0, total_size = 0; 605 u_long text_addr = 0, data_addr = 0; 606 u_long seg_size, seg_addr; 607 u_long addr, entry = 0, proghdr = 0; 608 int error = 0, i; 609 const char *interp = NULL; 610 Elf_Brandinfo *brand_info; 611 char *path; 612 struct thread *td = curthread; 613 struct sysentvec *sv; 614 615 /* 616 * Do we have a valid ELF header ? 617 * 618 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later 619 * if particular brand doesn't support it. 620 */ 621 if (__elfN(check_header)(hdr) != 0 || 622 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) 623 return (-1); 624 625 /* 626 * From here on down, we return an errno, not -1, as we've 627 * detected an ELF file. 628 */ 629 630 if ((hdr->e_phoff > PAGE_SIZE) || 631 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 632 /* Only support headers in first page for now */ 633 return (ENOEXEC); 634 } 635 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 636 for (i = 0; i < hdr->e_phnum; i++) { 637 switch (phdr[i].p_type) { 638 case PT_INTERP: /* Path to interpreter */ 639 if (phdr[i].p_filesz > MAXPATHLEN || 640 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) 641 return (ENOEXEC); 642 interp = imgp->image_header + phdr[i].p_offset; 643 break; 644 default: 645 break; 646 } 647 } 648 649 brand_info = __elfN(get_brandinfo)(hdr, interp); 650 if (brand_info == NULL) { 651 uprintf("ELF binary type \"%u\" not known.\n", 652 hdr->e_ident[EI_OSABI]); 653 return (ENOEXEC); 654 } 655 if (hdr->e_type == ET_DYN && 656 (brand_info->flags & BI_CAN_EXEC_DYN) == 0) 657 return (ENOEXEC); 658 sv = brand_info->sysvec; 659 if (interp != NULL && brand_info->interp_newpath != NULL) 660 interp = brand_info->interp_newpath; 661 662 /* 663 * Avoid a possible deadlock if the current address space is destroyed 664 * and that address space maps the locked vnode. In the common case, 665 * the locked vnode's v_usecount is decremented but remains greater 666 * than zero. Consequently, the vnode lock is not needed by vrele(). 667 * However, in cases where the vnode lock is external, such as nullfs, 668 * v_usecount may become zero. 669 */ 670 VOP_UNLOCK(imgp->vp, 0, td); 671 672 exec_new_vmspace(imgp, sv); 673 674 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td); 675 676 vmspace = imgp->proc->p_vmspace; 677 678 for (i = 0; i < hdr->e_phnum; i++) { 679 switch (phdr[i].p_type) { 680 case PT_LOAD: /* Loadable segment */ 681 prot = 0; 682 if (phdr[i].p_flags & PF_X) 683 prot |= VM_PROT_EXECUTE; 684 if (phdr[i].p_flags & PF_W) 685 prot |= VM_PROT_WRITE; 686 if (phdr[i].p_flags & PF_R) 687 prot |= VM_PROT_READ; 688 689 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 690 /* 691 * Some x86 binaries assume read == executable, 692 * notably the M3 runtime and therefore cvsup 693 */ 694 if (prot & VM_PROT_READ) 695 prot |= VM_PROT_EXECUTE; 696 #endif 697 698 if ((error = __elfN(load_section)(vmspace, 699 imgp->object, phdr[i].p_offset, 700 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 701 phdr[i].p_memsz, phdr[i].p_filesz, prot, 702 sv->sv_pagesize)) != 0) 703 return (error); 704 705 /* 706 * If this segment contains the program headers, 707 * remember their virtual address for the AT_PHDR 708 * aux entry. Static binaries don't usually include 709 * a PT_PHDR entry. 710 */ 711 if (phdr[i].p_offset == 0 && 712 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 713 <= phdr[i].p_filesz) 714 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 715 716 seg_addr = trunc_page(phdr[i].p_vaddr); 717 seg_size = round_page(phdr[i].p_memsz + 718 phdr[i].p_vaddr - seg_addr); 719 720 /* 721 * Is this .text or .data? We can't use 722 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 723 * alpha terribly and possibly does other bad 724 * things so we stick to the old way of figuring 725 * it out: If the segment contains the program 726 * entry point, it's a text segment, otherwise it 727 * is a data segment. 728 * 729 * Note that obreak() assumes that data_addr + 730 * data_size == end of data load area, and the ELF 731 * file format expects segments to be sorted by 732 * address. If multiple data segments exist, the 733 * last one will be used. 734 */ 735 if (hdr->e_entry >= phdr[i].p_vaddr && 736 hdr->e_entry < (phdr[i].p_vaddr + 737 phdr[i].p_memsz)) { 738 text_size = seg_size; 739 text_addr = seg_addr; 740 entry = (u_long)hdr->e_entry; 741 } else { 742 data_size = seg_size; 743 data_addr = seg_addr; 744 } 745 total_size += seg_size; 746 break; 747 case PT_PHDR: /* Program header table info */ 748 proghdr = phdr[i].p_vaddr; 749 break; 750 default: 751 break; 752 } 753 } 754 755 if (data_addr == 0 && data_size == 0) { 756 data_addr = text_addr; 757 data_size = text_size; 758 } 759 760 /* 761 * Check limits. It should be safe to check the 762 * limits after loading the segments since we do 763 * not actually fault in all the segments pages. 764 */ 765 PROC_LOCK(imgp->proc); 766 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 767 text_size > maxtsiz || 768 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) { 769 PROC_UNLOCK(imgp->proc); 770 return (ENOMEM); 771 } 772 773 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 774 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 775 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 776 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 777 778 /* 779 * We load the dynamic linker where a userland call 780 * to mmap(0, ...) would put it. The rationale behind this 781 * calculation is that it leaves room for the heap to grow to 782 * its maximum allowed size. 783 */ 784 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 785 lim_max(imgp->proc, RLIMIT_DATA)); 786 PROC_UNLOCK(imgp->proc); 787 788 imgp->entry_addr = entry; 789 790 imgp->proc->p_sysent = sv; 791 if (interp != NULL) { 792 VOP_UNLOCK(imgp->vp, 0, td); 793 if (brand_info->emul_path != NULL && 794 brand_info->emul_path[0] != '\0') { 795 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 796 snprintf(path, MAXPATHLEN, "%s%s", 797 brand_info->emul_path, interp); 798 error = __elfN(load_file)(imgp->proc, path, &addr, 799 &imgp->entry_addr, sv->sv_pagesize); 800 free(path, M_TEMP); 801 if (error == 0) 802 interp = NULL; 803 } 804 if (interp != NULL) { 805 error = __elfN(load_file)(imgp->proc, interp, &addr, 806 &imgp->entry_addr, sv->sv_pagesize); 807 } 808 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td); 809 if (error != 0) { 810 uprintf("ELF interpreter %s not found\n", interp); 811 return (error); 812 } 813 } 814 815 /* 816 * Construct auxargs table (used by the fixup routine) 817 */ 818 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 819 elf_auxargs->execfd = -1; 820 elf_auxargs->phdr = proghdr; 821 elf_auxargs->phent = hdr->e_phentsize; 822 elf_auxargs->phnum = hdr->e_phnum; 823 elf_auxargs->pagesz = PAGE_SIZE; 824 elf_auxargs->base = addr; 825 elf_auxargs->flags = 0; 826 elf_auxargs->entry = entry; 827 elf_auxargs->trace = elf_trace; 828 829 imgp->auxargs = elf_auxargs; 830 imgp->interpreted = 0; 831 832 return (error); 833 } 834 835 #define suword __CONCAT(suword, __ELF_WORD_SIZE) 836 837 int 838 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 839 { 840 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 841 Elf_Addr *base; 842 Elf_Addr *pos; 843 844 base = (Elf_Addr *)*stack_base; 845 pos = base + (imgp->args->argc + imgp->args->envc + 2); 846 847 if (args->trace) { 848 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 849 } 850 if (args->execfd != -1) { 851 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 852 } 853 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 854 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 855 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 856 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 857 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 858 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 859 AUXARGS_ENTRY(pos, AT_BASE, args->base); 860 AUXARGS_ENTRY(pos, AT_NULL, 0); 861 862 free(imgp->auxargs, M_TEMP); 863 imgp->auxargs = NULL; 864 865 base--; 866 suword(base, (long)imgp->args->argc); 867 *stack_base = (register_t *)base; 868 return (0); 869 } 870 871 /* 872 * Code for generating ELF core dumps. 873 */ 874 875 typedef void (*segment_callback)(vm_map_entry_t, void *); 876 877 /* Closure for cb_put_phdr(). */ 878 struct phdr_closure { 879 Elf_Phdr *phdr; /* Program header to fill in */ 880 Elf_Off offset; /* Offset of segment in core file */ 881 }; 882 883 /* Closure for cb_size_segment(). */ 884 struct sseg_closure { 885 int count; /* Count of writable segments. */ 886 size_t size; /* Total size of all writable segments. */ 887 }; 888 889 static void cb_put_phdr(vm_map_entry_t, void *); 890 static void cb_size_segment(vm_map_entry_t, void *); 891 static void each_writable_segment(struct thread *, segment_callback, void *); 892 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 893 int, void *, size_t); 894 static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 895 static void __elfN(putnote)(void *, size_t *, const char *, int, 896 const void *, size_t); 897 898 extern int osreldate; 899 900 int 901 __elfN(coredump)(td, vp, limit) 902 struct thread *td; 903 struct vnode *vp; 904 off_t limit; 905 { 906 struct ucred *cred = td->td_ucred; 907 int error = 0; 908 struct sseg_closure seginfo; 909 void *hdr; 910 size_t hdrsize; 911 912 /* Size the program segments. */ 913 seginfo.count = 0; 914 seginfo.size = 0; 915 each_writable_segment(td, cb_size_segment, &seginfo); 916 917 /* 918 * Calculate the size of the core file header area by making 919 * a dry run of generating it. Nothing is written, but the 920 * size is calculated. 921 */ 922 hdrsize = 0; 923 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 924 925 if (hdrsize + seginfo.size >= limit) 926 return (EFAULT); 927 928 /* 929 * Allocate memory for building the header, fill it up, 930 * and write it out. 931 */ 932 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 933 if (hdr == NULL) { 934 return (EINVAL); 935 } 936 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 937 938 /* Write the contents of all of the writable segments. */ 939 if (error == 0) { 940 Elf_Phdr *php; 941 off_t offset; 942 int i; 943 944 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 945 offset = hdrsize; 946 for (i = 0; i < seginfo.count; i++) { 947 error = vn_rdwr_inchunks(UIO_WRITE, vp, 948 (caddr_t)(uintptr_t)php->p_vaddr, 949 php->p_filesz, offset, UIO_USERSPACE, 950 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 951 curthread); /* XXXKSE */ 952 if (error != 0) 953 break; 954 offset += php->p_filesz; 955 php++; 956 } 957 } 958 free(hdr, M_TEMP); 959 960 return (error); 961 } 962 963 /* 964 * A callback for each_writable_segment() to write out the segment's 965 * program header entry. 966 */ 967 static void 968 cb_put_phdr(entry, closure) 969 vm_map_entry_t entry; 970 void *closure; 971 { 972 struct phdr_closure *phc = (struct phdr_closure *)closure; 973 Elf_Phdr *phdr = phc->phdr; 974 975 phc->offset = round_page(phc->offset); 976 977 phdr->p_type = PT_LOAD; 978 phdr->p_offset = phc->offset; 979 phdr->p_vaddr = entry->start; 980 phdr->p_paddr = 0; 981 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 982 phdr->p_align = PAGE_SIZE; 983 phdr->p_flags = 0; 984 if (entry->protection & VM_PROT_READ) 985 phdr->p_flags |= PF_R; 986 if (entry->protection & VM_PROT_WRITE) 987 phdr->p_flags |= PF_W; 988 if (entry->protection & VM_PROT_EXECUTE) 989 phdr->p_flags |= PF_X; 990 991 phc->offset += phdr->p_filesz; 992 phc->phdr++; 993 } 994 995 /* 996 * A callback for each_writable_segment() to gather information about 997 * the number of segments and their total size. 998 */ 999 static void 1000 cb_size_segment(entry, closure) 1001 vm_map_entry_t entry; 1002 void *closure; 1003 { 1004 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1005 1006 ssc->count++; 1007 ssc->size += entry->end - entry->start; 1008 } 1009 1010 /* 1011 * For each writable segment in the process's memory map, call the given 1012 * function with a pointer to the map entry and some arbitrary 1013 * caller-supplied data. 1014 */ 1015 static void 1016 each_writable_segment(td, func, closure) 1017 struct thread *td; 1018 segment_callback func; 1019 void *closure; 1020 { 1021 struct proc *p = td->td_proc; 1022 vm_map_t map = &p->p_vmspace->vm_map; 1023 vm_map_entry_t entry; 1024 1025 for (entry = map->header.next; entry != &map->header; 1026 entry = entry->next) { 1027 vm_object_t obj; 1028 1029 /* 1030 * Don't dump inaccessible mappings, deal with legacy 1031 * coredump mode. 1032 * 1033 * Note that read-only segments related to the elf binary 1034 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1035 * need to arbitrarily ignore such segments. 1036 */ 1037 if (elf_legacy_coredump) { 1038 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1039 continue; 1040 } else { 1041 if ((entry->protection & VM_PROT_ALL) == 0) 1042 continue; 1043 } 1044 1045 /* 1046 * Dont include memory segment in the coredump if 1047 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1048 * madvise(2). Do not dump submaps (i.e. parts of the 1049 * kernel map). 1050 */ 1051 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1052 continue; 1053 1054 if ((obj = entry->object.vm_object) == NULL) 1055 continue; 1056 1057 /* Find the deepest backing object. */ 1058 while (obj->backing_object != NULL) 1059 obj = obj->backing_object; 1060 1061 /* Ignore memory-mapped devices and such things. */ 1062 if (obj->type != OBJT_DEFAULT && 1063 obj->type != OBJT_SWAP && 1064 obj->type != OBJT_VNODE) 1065 continue; 1066 1067 (*func)(entry, closure); 1068 } 1069 } 1070 1071 /* 1072 * Write the core file header to the file, including padding up to 1073 * the page boundary. 1074 */ 1075 static int 1076 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1077 struct thread *td; 1078 struct vnode *vp; 1079 struct ucred *cred; 1080 int numsegs; 1081 size_t hdrsize; 1082 void *hdr; 1083 { 1084 size_t off; 1085 1086 /* Fill in the header. */ 1087 bzero(hdr, hdrsize); 1088 off = 0; 1089 __elfN(puthdr)(td, hdr, &off, numsegs); 1090 1091 /* Write it to the core file. */ 1092 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1093 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1094 td)); /* XXXKSE */ 1095 } 1096 1097 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1098 typedef struct prstatus32 elf_prstatus_t; 1099 typedef struct prpsinfo32 elf_prpsinfo_t; 1100 typedef struct fpreg32 elf_prfpregset_t; 1101 typedef struct fpreg32 elf_fpregset_t; 1102 typedef struct reg32 elf_gregset_t; 1103 #else 1104 typedef prstatus_t elf_prstatus_t; 1105 typedef prpsinfo_t elf_prpsinfo_t; 1106 typedef prfpregset_t elf_prfpregset_t; 1107 typedef prfpregset_t elf_fpregset_t; 1108 typedef gregset_t elf_gregset_t; 1109 #endif 1110 1111 static void 1112 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1113 { 1114 struct { 1115 elf_prstatus_t status; 1116 elf_prfpregset_t fpregset; 1117 elf_prpsinfo_t psinfo; 1118 } *tempdata; 1119 elf_prstatus_t *status; 1120 elf_prfpregset_t *fpregset; 1121 elf_prpsinfo_t *psinfo; 1122 struct proc *p; 1123 struct thread *thr; 1124 size_t ehoff, noteoff, notesz, phoff; 1125 1126 p = td->td_proc; 1127 1128 ehoff = *off; 1129 *off += sizeof(Elf_Ehdr); 1130 1131 phoff = *off; 1132 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1133 1134 noteoff = *off; 1135 /* 1136 * Don't allocate space for the notes if we're just calculating 1137 * the size of the header. We also don't collect the data. 1138 */ 1139 if (dst != NULL) { 1140 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1141 status = &tempdata->status; 1142 fpregset = &tempdata->fpregset; 1143 psinfo = &tempdata->psinfo; 1144 } else { 1145 tempdata = NULL; 1146 status = NULL; 1147 fpregset = NULL; 1148 psinfo = NULL; 1149 } 1150 1151 if (dst != NULL) { 1152 psinfo->pr_version = PRPSINFO_VERSION; 1153 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1154 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1155 /* 1156 * XXX - We don't fill in the command line arguments properly 1157 * yet. 1158 */ 1159 strlcpy(psinfo->pr_psargs, p->p_comm, 1160 sizeof(psinfo->pr_psargs)); 1161 } 1162 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1163 sizeof *psinfo); 1164 1165 /* 1166 * To have the debugger select the right thread (LWP) as the initial 1167 * thread, we dump the state of the thread passed to us in td first. 1168 * This is the thread that causes the core dump and thus likely to 1169 * be the right thread one wants to have selected in the debugger. 1170 */ 1171 thr = td; 1172 while (thr != NULL) { 1173 if (dst != NULL) { 1174 status->pr_version = PRSTATUS_VERSION; 1175 status->pr_statussz = sizeof(elf_prstatus_t); 1176 status->pr_gregsetsz = sizeof(elf_gregset_t); 1177 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1178 status->pr_osreldate = osreldate; 1179 status->pr_cursig = p->p_sig; 1180 status->pr_pid = thr->td_tid; 1181 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1182 fill_regs32(thr, &status->pr_reg); 1183 fill_fpregs32(thr, fpregset); 1184 #else 1185 fill_regs(thr, &status->pr_reg); 1186 fill_fpregs(thr, fpregset); 1187 #endif 1188 } 1189 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1190 sizeof *status); 1191 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1192 sizeof *fpregset); 1193 /* 1194 * Allow for MD specific notes, as well as any MD 1195 * specific preparations for writing MI notes. 1196 */ 1197 __elfN(dump_thread)(thr, dst, off); 1198 1199 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1200 TAILQ_NEXT(thr, td_plist); 1201 if (thr == td) 1202 thr = TAILQ_NEXT(thr, td_plist); 1203 } 1204 1205 notesz = *off - noteoff; 1206 1207 if (dst != NULL) 1208 free(tempdata, M_TEMP); 1209 1210 /* Align up to a page boundary for the program segments. */ 1211 *off = round_page(*off); 1212 1213 if (dst != NULL) { 1214 Elf_Ehdr *ehdr; 1215 Elf_Phdr *phdr; 1216 struct phdr_closure phc; 1217 1218 /* 1219 * Fill in the ELF header. 1220 */ 1221 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1222 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1223 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1224 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1225 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1226 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1227 ehdr->e_ident[EI_DATA] = ELF_DATA; 1228 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1229 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1230 ehdr->e_ident[EI_ABIVERSION] = 0; 1231 ehdr->e_ident[EI_PAD] = 0; 1232 ehdr->e_type = ET_CORE; 1233 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1234 ehdr->e_machine = EM_386; 1235 #else 1236 ehdr->e_machine = ELF_ARCH; 1237 #endif 1238 ehdr->e_version = EV_CURRENT; 1239 ehdr->e_entry = 0; 1240 ehdr->e_phoff = phoff; 1241 ehdr->e_flags = 0; 1242 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1243 ehdr->e_phentsize = sizeof(Elf_Phdr); 1244 ehdr->e_phnum = numsegs + 1; 1245 ehdr->e_shentsize = sizeof(Elf_Shdr); 1246 ehdr->e_shnum = 0; 1247 ehdr->e_shstrndx = SHN_UNDEF; 1248 1249 /* 1250 * Fill in the program header entries. 1251 */ 1252 phdr = (Elf_Phdr *)((char *)dst + phoff); 1253 1254 /* The note segement. */ 1255 phdr->p_type = PT_NOTE; 1256 phdr->p_offset = noteoff; 1257 phdr->p_vaddr = 0; 1258 phdr->p_paddr = 0; 1259 phdr->p_filesz = notesz; 1260 phdr->p_memsz = 0; 1261 phdr->p_flags = 0; 1262 phdr->p_align = 0; 1263 phdr++; 1264 1265 /* All the writable segments from the program. */ 1266 phc.phdr = phdr; 1267 phc.offset = *off; 1268 each_writable_segment(td, cb_put_phdr, &phc); 1269 } 1270 } 1271 1272 static void 1273 __elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1274 const void *desc, size_t descsz) 1275 { 1276 Elf_Note note; 1277 1278 note.n_namesz = strlen(name) + 1; 1279 note.n_descsz = descsz; 1280 note.n_type = type; 1281 if (dst != NULL) 1282 bcopy(¬e, (char *)dst + *off, sizeof note); 1283 *off += sizeof note; 1284 if (dst != NULL) 1285 bcopy(name, (char *)dst + *off, note.n_namesz); 1286 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1287 if (dst != NULL) 1288 bcopy(desc, (char *)dst + *off, note.n_descsz); 1289 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1290 } 1291 1292 /* 1293 * Tell kern_execve.c about it, with a little help from the linker. 1294 */ 1295 static struct execsw __elfN(execsw) = { 1296 __CONCAT(exec_, __elfN(imgact)), 1297 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1298 }; 1299 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1300