imgact_elf.c revision 153585
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: head/sys/kern/imgact_elf.c 153585 2005-12-20 23:42:18Z alc $"); 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 80static int __elfN(check_header)(const Elf_Ehdr *hdr); 81static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr, 82 const char *interp); 83static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, 84 u_long *entry, size_t pagesize); 85static int __elfN(load_section)(struct proc *p, 86 struct vmspace *vmspace, struct vnode *vp, vm_object_t object, 87 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, 88 vm_prot_t prot, size_t pagesize); 89static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); 90 91SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, 92 ""); 93 94int __elfN(fallback_brand) = -1; 95SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 96 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0, 97 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); 98TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand", 99 &__elfN(fallback_brand)); 100 101int __elfN(can_exec_dyn) = 0; 102SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, 103 can_exec_dyn, CTLFLAG_RW, &__elfN(can_exec_dyn), 0, 104 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " can exec shared libraries"); 105 106static int elf_trace = 0; 107SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, ""); 108 109static int elf_legacy_coredump = 0; 110SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 111 &elf_legacy_coredump, 0, ""); 112 113static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; 114 115int 116__elfN(insert_brand_entry)(Elf_Brandinfo *entry) 117{ 118 int i; 119 120 for (i = 0; i < MAX_BRANDS; i++) { 121 if (elf_brand_list[i] == NULL) { 122 elf_brand_list[i] = entry; 123 break; 124 } 125 } 126 if (i == MAX_BRANDS) 127 return (-1); 128 return (0); 129} 130 131int 132__elfN(remove_brand_entry)(Elf_Brandinfo *entry) 133{ 134 int i; 135 136 for (i = 0; i < MAX_BRANDS; i++) { 137 if (elf_brand_list[i] == entry) { 138 elf_brand_list[i] = NULL; 139 break; 140 } 141 } 142 if (i == MAX_BRANDS) 143 return (-1); 144 return (0); 145} 146 147int 148__elfN(brand_inuse)(Elf_Brandinfo *entry) 149{ 150 struct proc *p; 151 int rval = FALSE; 152 153 sx_slock(&allproc_lock); 154 LIST_FOREACH(p, &allproc, p_list) { 155 if (p->p_sysent == entry->sysvec) { 156 rval = TRUE; 157 break; 158 } 159 } 160 sx_sunlock(&allproc_lock); 161 162 return (rval); 163} 164 165static Elf_Brandinfo * 166__elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp) 167{ 168 Elf_Brandinfo *bi; 169 int i; 170 171 /* 172 * We support three types of branding -- (1) the ELF EI_OSABI field 173 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string 174 * branding w/in the ELF header, and (3) path of the `interp_path' 175 * field. We should also look for an ".note.ABI-tag" ELF section now 176 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones. 177 */ 178 179 /* If the executable has a brand, search for it in the brand list. */ 180 for (i = 0; i < MAX_BRANDS; i++) { 181 bi = elf_brand_list[i]; 182 if (bi != NULL && hdr->e_machine == bi->machine && 183 (hdr->e_ident[EI_OSABI] == bi->brand || 184 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], 185 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) 186 return (bi); 187 } 188 189 /* Lacking a known brand, search for a recognized interpreter. */ 190 if (interp != NULL) { 191 for (i = 0; i < MAX_BRANDS; i++) { 192 bi = elf_brand_list[i]; 193 if (bi != NULL && hdr->e_machine == bi->machine && 194 strcmp(interp, bi->interp_path) == 0) 195 return (bi); 196 } 197 } 198 199 /* Lacking a recognized interpreter, try the default brand */ 200 for (i = 0; i < MAX_BRANDS; i++) { 201 bi = elf_brand_list[i]; 202 if (bi != NULL && hdr->e_machine == bi->machine && 203 __elfN(fallback_brand) == bi->brand) 204 return (bi); 205 } 206 return (NULL); 207} 208 209static int 210__elfN(check_header)(const Elf_Ehdr *hdr) 211{ 212 Elf_Brandinfo *bi; 213 int i; 214 215 if (!IS_ELF(*hdr) || 216 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 217 hdr->e_ident[EI_DATA] != ELF_TARG_DATA || 218 hdr->e_ident[EI_VERSION] != EV_CURRENT || 219 hdr->e_phentsize != sizeof(Elf_Phdr) || 220 hdr->e_version != ELF_TARG_VER) 221 return (ENOEXEC); 222 223 /* 224 * Make sure we have at least one brand for this machine. 225 */ 226 227 for (i = 0; i < MAX_BRANDS; i++) { 228 bi = elf_brand_list[i]; 229 if (bi != NULL && bi->machine == hdr->e_machine) 230 break; 231 } 232 if (i == MAX_BRANDS) 233 return (ENOEXEC); 234 235 return (0); 236} 237 238static int 239__elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 240 vm_offset_t start, vm_offset_t end, vm_prot_t prot) 241{ 242 struct sf_buf *sf; 243 int error; 244 vm_offset_t off; 245 246 /* 247 * Create the page if it doesn't exist yet. Ignore errors. 248 */ 249 vm_map_lock(map); 250 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), 251 VM_PROT_ALL, VM_PROT_ALL, 0); 252 vm_map_unlock(map); 253 254 /* 255 * Find the page from the underlying object. 256 */ 257 if (object) { 258 sf = vm_imgact_map_page(object, offset); 259 if (sf == NULL) 260 return (KERN_FAILURE); 261 off = offset - trunc_page(offset); 262 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, 263 end - start); 264 vm_imgact_unmap_page(sf); 265 if (error) { 266 return (KERN_FAILURE); 267 } 268 } 269 270 return (KERN_SUCCESS); 271} 272 273static int 274__elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, 275 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) 276{ 277 struct sf_buf *sf; 278 vm_offset_t off; 279 vm_size_t sz; 280 int error, rv; 281 282 if (start != trunc_page(start)) { 283 rv = __elfN(map_partial)(map, object, offset, start, 284 round_page(start), prot); 285 if (rv) 286 return (rv); 287 offset += round_page(start) - start; 288 start = round_page(start); 289 } 290 if (end != round_page(end)) { 291 rv = __elfN(map_partial)(map, object, offset + 292 trunc_page(end) - start, trunc_page(end), end, prot); 293 if (rv) 294 return (rv); 295 end = trunc_page(end); 296 } 297 if (end > start) { 298 if (offset & PAGE_MASK) { 299 /* 300 * The mapping is not page aligned. This means we have 301 * to copy the data. Sigh. 302 */ 303 rv = vm_map_find(map, NULL, 0, &start, end - start, 304 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); 305 if (rv) 306 return (rv); 307 if (object == NULL) 308 return (KERN_SUCCESS); 309 for (; start < end; start += sz) { 310 sf = vm_imgact_map_page(object, offset); 311 if (sf == NULL) 312 return (KERN_FAILURE); 313 off = offset - trunc_page(offset); 314 sz = end - start; 315 if (sz > PAGE_SIZE - off) 316 sz = PAGE_SIZE - off; 317 error = copyout((caddr_t)sf_buf_kva(sf) + off, 318 (caddr_t)start, sz); 319 vm_imgact_unmap_page(sf); 320 if (error) { 321 return (KERN_FAILURE); 322 } 323 offset += sz; 324 } 325 rv = KERN_SUCCESS; 326 } else { 327 vm_map_lock(map); 328 rv = vm_map_insert(map, object, offset, start, end, 329 prot, VM_PROT_ALL, cow); 330 vm_map_unlock(map); 331 } 332 return (rv); 333 } else { 334 return (KERN_SUCCESS); 335 } 336} 337 338static int 339__elfN(load_section)(struct proc *p, struct vmspace *vmspace, 340 struct vnode *vp, vm_object_t object, vm_offset_t offset, 341 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, 342 size_t pagesize) 343{ 344 struct sf_buf *sf; 345 size_t map_len; 346 vm_offset_t map_addr; 347 int error, rv, cow; 348 size_t copy_len; 349 vm_offset_t file_addr; 350 351 /* 352 * It's necessary to fail if the filsz + offset taken from the 353 * header is greater than the actual file pager object's size. 354 * If we were to allow this, then the vm_map_find() below would 355 * walk right off the end of the file object and into the ether. 356 * 357 * While I'm here, might as well check for something else that 358 * is invalid: filsz cannot be greater than memsz. 359 */ 360 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size || 361 filsz > memsz) { 362 uprintf("elf_load_section: truncated ELF file\n"); 363 return (ENOEXEC); 364 } 365 366#define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) 367#define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) 368 369 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); 370 file_addr = trunc_page_ps(offset, pagesize); 371 372 /* 373 * We have two choices. We can either clear the data in the last page 374 * of an oversized mapping, or we can start the anon mapping a page 375 * early and copy the initialized data into that first page. We 376 * choose the second.. 377 */ 378 if (memsz > filsz) 379 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; 380 else 381 map_len = round_page_ps(offset + filsz, pagesize) - file_addr; 382 383 if (map_len != 0) { 384 vm_object_reference(object); 385 386 /* cow flags: don't dump readonly sections in core */ 387 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | 388 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); 389 390 rv = __elfN(map_insert)(&vmspace->vm_map, 391 object, 392 file_addr, /* file offset */ 393 map_addr, /* virtual start */ 394 map_addr + map_len,/* virtual end */ 395 prot, 396 cow); 397 if (rv != KERN_SUCCESS) { 398 vm_object_deallocate(object); 399 return (EINVAL); 400 } 401 402 /* we can stop now if we've covered it all */ 403 if (memsz == filsz) { 404 return (0); 405 } 406 } 407 408 409 /* 410 * We have to get the remaining bit of the file into the first part 411 * of the oversized map segment. This is normally because the .data 412 * segment in the file is extended to provide bss. It's a neat idea 413 * to try and save a page, but it's a pain in the behind to implement. 414 */ 415 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); 416 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); 417 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - 418 map_addr; 419 420 /* This had damn well better be true! */ 421 if (map_len != 0) { 422 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr, 423 map_addr + map_len, VM_PROT_ALL, 0); 424 if (rv != KERN_SUCCESS) { 425 return (EINVAL); 426 } 427 } 428 429 if (copy_len != 0) { 430 vm_offset_t off; 431 432 sf = vm_imgact_map_page(object, offset + filsz); 433 if (sf == NULL) 434 return (EIO); 435 436 /* send the page fragment to user space */ 437 off = trunc_page_ps(offset + filsz, pagesize) - 438 trunc_page(offset + filsz); 439 error = copyout((caddr_t)sf_buf_kva(sf) + off, 440 (caddr_t)map_addr, copy_len); 441 vm_imgact_unmap_page(sf); 442 if (error) { 443 return (error); 444 } 445 } 446 447 /* 448 * set it to the specified protection. 449 * XXX had better undo the damage from pasting over the cracks here! 450 */ 451 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr), 452 round_page(map_addr + map_len), prot, FALSE); 453 454 return (0); 455} 456 457/* 458 * Load the file "file" into memory. It may be either a shared object 459 * or an executable. 460 * 461 * The "addr" reference parameter is in/out. On entry, it specifies 462 * the address where a shared object should be loaded. If the file is 463 * an executable, this value is ignored. On exit, "addr" specifies 464 * where the file was actually loaded. 465 * 466 * The "entry" reference parameter is out only. On exit, it specifies 467 * the entry point for the loaded file. 468 */ 469static int 470__elfN(load_file)(struct proc *p, const char *file, u_long *addr, 471 u_long *entry, size_t pagesize) 472{ 473 struct { 474 struct nameidata nd; 475 struct vattr attr; 476 struct image_params image_params; 477 } *tempdata; 478 const Elf_Ehdr *hdr = NULL; 479 const Elf_Phdr *phdr = NULL; 480 struct nameidata *nd; 481 struct vmspace *vmspace = p->p_vmspace; 482 struct vattr *attr; 483 struct image_params *imgp; 484 vm_prot_t prot; 485 u_long rbase; 486 u_long base_addr = 0; 487 int vfslocked, error, i, numsegs; 488 489 if (curthread->td_proc != p) 490 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */ 491 492 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); 493 nd = &tempdata->nd; 494 attr = &tempdata->attr; 495 imgp = &tempdata->image_params; 496 497 /* 498 * Initialize part of the common data 499 */ 500 imgp->proc = p; 501 imgp->attr = attr; 502 imgp->firstpage = NULL; 503 imgp->image_header = NULL; 504 imgp->object = NULL; 505 imgp->execlabel = NULL; 506 507 /* XXXKSE */ 508 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, 509 curthread); 510 vfslocked = 0; 511 if ((error = namei(nd)) != 0) { 512 nd->ni_vp = NULL; 513 goto fail; 514 } 515 vfslocked = NDHASGIANT(nd); 516 NDFREE(nd, NDF_ONLY_PNBUF); 517 imgp->vp = nd->ni_vp; 518 519 /* 520 * Check permissions, modes, uid, etc on the file, and "open" it. 521 */ 522 error = exec_check_permissions(imgp); 523 if (error) { 524 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 525 goto fail; 526 } 527 528 error = exec_map_first_page(imgp); 529 /* 530 * Also make certain that the interpreter stays the same, so set 531 * its VV_TEXT flag, too. 532 */ 533 if (error == 0) 534 nd->ni_vp->v_vflag |= VV_TEXT; 535 536 imgp->object = nd->ni_vp->v_object; 537 vm_object_reference(imgp->object); 538 539 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */ 540 if (error) 541 goto fail; 542 543 hdr = (const Elf_Ehdr *)imgp->image_header; 544 if ((error = __elfN(check_header)(hdr)) != 0) 545 goto fail; 546 if (hdr->e_type == ET_DYN) 547 rbase = *addr; 548 else if (hdr->e_type == ET_EXEC) 549 rbase = 0; 550 else { 551 error = ENOEXEC; 552 goto fail; 553 } 554 555 /* Only support headers that fit within first page for now */ 556 /* (multiplication of two Elf_Half fields will not overflow) */ 557 if ((hdr->e_phoff > PAGE_SIZE) || 558 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) { 559 error = ENOEXEC; 560 goto fail; 561 } 562 563 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 564 565 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { 566 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */ 567 prot = 0; 568 if (phdr[i].p_flags & PF_X) 569 prot |= VM_PROT_EXECUTE; 570 if (phdr[i].p_flags & PF_W) 571 prot |= VM_PROT_WRITE; 572 if (phdr[i].p_flags & PF_R) 573 prot |= VM_PROT_READ; 574 575 if ((error = __elfN(load_section)(p, vmspace, 576 nd->ni_vp, imgp->object, phdr[i].p_offset, 577 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, 578 phdr[i].p_memsz, phdr[i].p_filesz, prot, 579 pagesize)) != 0) 580 goto fail; 581 /* 582 * Establish the base address if this is the 583 * first segment. 584 */ 585 if (numsegs == 0) 586 base_addr = trunc_page(phdr[i].p_vaddr + 587 rbase); 588 numsegs++; 589 } 590 } 591 *addr = base_addr; 592 *entry = (unsigned long)hdr->e_entry + rbase; 593 594fail: 595 if (imgp->firstpage) 596 exec_unmap_first_page(imgp); 597 if (imgp->object) 598 vm_object_deallocate(imgp->object); 599 600 if (nd->ni_vp) 601 vrele(nd->ni_vp); 602 603 VFS_UNLOCK_GIANT(vfslocked); 604 free(tempdata, M_TEMP); 605 606 return (error); 607} 608 609static int 610__CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) 611{ 612 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; 613 const Elf_Phdr *phdr; 614 Elf_Auxargs *elf_auxargs = NULL; 615 struct vmspace *vmspace; 616 vm_prot_t prot; 617 u_long text_size = 0, data_size = 0, total_size = 0; 618 u_long text_addr = 0, data_addr = 0; 619 u_long seg_size, seg_addr; 620 u_long addr, entry = 0, proghdr = 0; 621 int error = 0, i; 622 const char *interp = NULL; 623 Elf_Brandinfo *brand_info; 624 char *path; 625 struct thread *td = curthread; 626 struct sysentvec *sv; 627 628 /* 629 * Do we have a valid ELF header ? 630 */ 631 if (__elfN(check_header)(hdr) != 0 || (hdr->e_type != ET_EXEC 632 && (!__elfN(can_exec_dyn) || hdr->e_type != ET_DYN))) 633 return (-1); 634 635 /* 636 * From here on down, we return an errno, not -1, as we've 637 * detected an ELF file. 638 */ 639 640 if ((hdr->e_phoff > PAGE_SIZE) || 641 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) { 642 /* Only support headers in first page for now */ 643 return (ENOEXEC); 644 } 645 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); 646 647 /* 648 * From this point on, we may have resources that need to be freed. 649 */ 650 651 VOP_UNLOCK(imgp->vp, 0, td); 652 653 for (i = 0; i < hdr->e_phnum; i++) { 654 switch (phdr[i].p_type) { 655 case PT_INTERP: /* Path to interpreter */ 656 if (phdr[i].p_filesz > MAXPATHLEN || 657 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) { 658 error = ENOEXEC; 659 goto fail; 660 } 661 interp = imgp->image_header + phdr[i].p_offset; 662 break; 663 default: 664 break; 665 } 666 } 667 668 brand_info = __elfN(get_brandinfo)(hdr, interp); 669 if (brand_info == NULL) { 670 uprintf("ELF binary type \"%u\" not known.\n", 671 hdr->e_ident[EI_OSABI]); 672 error = ENOEXEC; 673 goto fail; 674 } 675 sv = brand_info->sysvec; 676 if (interp != NULL && brand_info->interp_newpath != NULL) 677 interp = brand_info->interp_newpath; 678 679 exec_new_vmspace(imgp, sv); 680 681 vmspace = imgp->proc->p_vmspace; 682 683 for (i = 0; i < hdr->e_phnum; i++) { 684 switch (phdr[i].p_type) { 685 case PT_LOAD: /* Loadable segment */ 686 prot = 0; 687 if (phdr[i].p_flags & PF_X) 688 prot |= VM_PROT_EXECUTE; 689 if (phdr[i].p_flags & PF_W) 690 prot |= VM_PROT_WRITE; 691 if (phdr[i].p_flags & PF_R) 692 prot |= VM_PROT_READ; 693 694#if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER) 695 /* 696 * Some x86 binaries assume read == executable, 697 * notably the M3 runtime and therefore cvsup 698 */ 699 if (prot & VM_PROT_READ) 700 prot |= VM_PROT_EXECUTE; 701#endif 702 703 if ((error = __elfN(load_section)(imgp->proc, vmspace, 704 imgp->vp, imgp->object, phdr[i].p_offset, 705 (caddr_t)(uintptr_t)phdr[i].p_vaddr, 706 phdr[i].p_memsz, phdr[i].p_filesz, prot, 707 sv->sv_pagesize)) != 0) 708 goto fail; 709 710 /* 711 * If this segment contains the program headers, 712 * remember their virtual address for the AT_PHDR 713 * aux entry. Static binaries don't usually include 714 * a PT_PHDR entry. 715 */ 716 if (phdr[i].p_offset == 0 && 717 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize 718 <= phdr[i].p_filesz) 719 proghdr = phdr[i].p_vaddr + hdr->e_phoff; 720 721 seg_addr = trunc_page(phdr[i].p_vaddr); 722 seg_size = round_page(phdr[i].p_memsz + 723 phdr[i].p_vaddr - seg_addr); 724 725 /* 726 * Is this .text or .data? We can't use 727 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the 728 * alpha terribly and possibly does other bad 729 * things so we stick to the old way of figuring 730 * it out: If the segment contains the program 731 * entry point, it's a text segment, otherwise it 732 * is a data segment. 733 * 734 * Note that obreak() assumes that data_addr + 735 * data_size == end of data load area, and the ELF 736 * file format expects segments to be sorted by 737 * address. If multiple data segments exist, the 738 * last one will be used. 739 */ 740 if (hdr->e_entry >= phdr[i].p_vaddr && 741 hdr->e_entry < (phdr[i].p_vaddr + 742 phdr[i].p_memsz)) { 743 text_size = seg_size; 744 text_addr = seg_addr; 745 entry = (u_long)hdr->e_entry; 746 } else { 747 data_size = seg_size; 748 data_addr = seg_addr; 749 } 750 total_size += seg_size; 751 break; 752 case PT_PHDR: /* Program header table info */ 753 proghdr = phdr[i].p_vaddr; 754 break; 755 default: 756 break; 757 } 758 } 759 760 if (data_addr == 0 && data_size == 0) { 761 data_addr = text_addr; 762 data_size = text_size; 763 } 764 765 /* 766 * Check limits. It should be safe to check the 767 * limits after loading the segments since we do 768 * not actually fault in all the segments pages. 769 */ 770 PROC_LOCK(imgp->proc); 771 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) || 772 text_size > maxtsiz || 773 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) { 774 PROC_UNLOCK(imgp->proc); 775 error = ENOMEM; 776 goto fail; 777 } 778 779 vmspace->vm_tsize = text_size >> PAGE_SHIFT; 780 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; 781 vmspace->vm_dsize = data_size >> PAGE_SHIFT; 782 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; 783 784 /* 785 * We load the dynamic linker where a userland call 786 * to mmap(0, ...) would put it. The rationale behind this 787 * calculation is that it leaves room for the heap to grow to 788 * its maximum allowed size. 789 */ 790 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr + 791 lim_max(imgp->proc, RLIMIT_DATA)); 792 PROC_UNLOCK(imgp->proc); 793 794 imgp->entry_addr = entry; 795 796 imgp->proc->p_sysent = sv; 797 if (interp != NULL && brand_info->emul_path != NULL && 798 brand_info->emul_path[0] != '\0') { 799 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 800 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path, 801 interp); 802 error = __elfN(load_file)(imgp->proc, path, &addr, 803 &imgp->entry_addr, sv->sv_pagesize); 804 free(path, M_TEMP); 805 if (error == 0) 806 interp = NULL; 807 } 808 if (interp != NULL) { 809 error = __elfN(load_file)(imgp->proc, interp, &addr, 810 &imgp->entry_addr, sv->sv_pagesize); 811 if (error != 0) { 812 uprintf("ELF interpreter %s not found\n", interp); 813 goto fail; 814 } 815 } 816 817 /* 818 * Construct auxargs table (used by the fixup routine) 819 */ 820 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); 821 elf_auxargs->execfd = -1; 822 elf_auxargs->phdr = proghdr; 823 elf_auxargs->phent = hdr->e_phentsize; 824 elf_auxargs->phnum = hdr->e_phnum; 825 elf_auxargs->pagesz = PAGE_SIZE; 826 elf_auxargs->base = addr; 827 elf_auxargs->flags = 0; 828 elf_auxargs->entry = entry; 829 elf_auxargs->trace = elf_trace; 830 831 imgp->auxargs = elf_auxargs; 832 imgp->interpreted = 0; 833 834fail: 835 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td); 836 return (error); 837} 838 839#define suword __CONCAT(suword, __ELF_WORD_SIZE) 840 841int 842__elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) 843{ 844 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; 845 Elf_Addr *base; 846 Elf_Addr *pos; 847 848 base = (Elf_Addr *)*stack_base; 849 pos = base + (imgp->args->argc + imgp->args->envc + 2); 850 851 if (args->trace) { 852 AUXARGS_ENTRY(pos, AT_DEBUG, 1); 853 } 854 if (args->execfd != -1) { 855 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); 856 } 857 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); 858 AUXARGS_ENTRY(pos, AT_PHENT, args->phent); 859 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); 860 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); 861 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); 862 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); 863 AUXARGS_ENTRY(pos, AT_BASE, args->base); 864 AUXARGS_ENTRY(pos, AT_NULL, 0); 865 866 free(imgp->auxargs, M_TEMP); 867 imgp->auxargs = NULL; 868 869 base--; 870 suword(base, (long)imgp->args->argc); 871 *stack_base = (register_t *)base; 872 return (0); 873} 874 875/* 876 * Code for generating ELF core dumps. 877 */ 878 879typedef void (*segment_callback)(vm_map_entry_t, void *); 880 881/* Closure for cb_put_phdr(). */ 882struct phdr_closure { 883 Elf_Phdr *phdr; /* Program header to fill in */ 884 Elf_Off offset; /* Offset of segment in core file */ 885}; 886 887/* Closure for cb_size_segment(). */ 888struct sseg_closure { 889 int count; /* Count of writable segments. */ 890 size_t size; /* Total size of all writable segments. */ 891}; 892 893static void cb_put_phdr(vm_map_entry_t, void *); 894static void cb_size_segment(vm_map_entry_t, void *); 895static void each_writable_segment(struct thread *, segment_callback, void *); 896static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *, 897 int, void *, size_t); 898static void __elfN(puthdr)(struct thread *, void *, size_t *, int); 899static void __elfN(putnote)(void *, size_t *, const char *, int, 900 const void *, size_t); 901 902extern int osreldate; 903 904int 905__elfN(coredump)(td, vp, limit) 906 struct thread *td; 907 struct vnode *vp; 908 off_t limit; 909{ 910 struct ucred *cred = td->td_ucred; 911 int error = 0; 912 struct sseg_closure seginfo; 913 void *hdr; 914 size_t hdrsize; 915 916 /* Size the program segments. */ 917 seginfo.count = 0; 918 seginfo.size = 0; 919 each_writable_segment(td, cb_size_segment, &seginfo); 920 921 /* 922 * Calculate the size of the core file header area by making 923 * a dry run of generating it. Nothing is written, but the 924 * size is calculated. 925 */ 926 hdrsize = 0; 927 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count); 928 929 if (hdrsize + seginfo.size >= limit) 930 return (EFAULT); 931 932 /* 933 * Allocate memory for building the header, fill it up, 934 * and write it out. 935 */ 936 hdr = malloc(hdrsize, M_TEMP, M_WAITOK); 937 if (hdr == NULL) { 938 return (EINVAL); 939 } 940 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize); 941 942 /* Write the contents of all of the writable segments. */ 943 if (error == 0) { 944 Elf_Phdr *php; 945 off_t offset; 946 int i; 947 948 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 949 offset = hdrsize; 950 for (i = 0; i < seginfo.count; i++) { 951 error = vn_rdwr_inchunks(UIO_WRITE, vp, 952 (caddr_t)(uintptr_t)php->p_vaddr, 953 php->p_filesz, offset, UIO_USERSPACE, 954 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 955 curthread); /* XXXKSE */ 956 if (error != 0) 957 break; 958 offset += php->p_filesz; 959 php++; 960 } 961 } 962 free(hdr, M_TEMP); 963 964 return (error); 965} 966 967/* 968 * A callback for each_writable_segment() to write out the segment's 969 * program header entry. 970 */ 971static void 972cb_put_phdr(entry, closure) 973 vm_map_entry_t entry; 974 void *closure; 975{ 976 struct phdr_closure *phc = (struct phdr_closure *)closure; 977 Elf_Phdr *phdr = phc->phdr; 978 979 phc->offset = round_page(phc->offset); 980 981 phdr->p_type = PT_LOAD; 982 phdr->p_offset = phc->offset; 983 phdr->p_vaddr = entry->start; 984 phdr->p_paddr = 0; 985 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 986 phdr->p_align = PAGE_SIZE; 987 phdr->p_flags = 0; 988 if (entry->protection & VM_PROT_READ) 989 phdr->p_flags |= PF_R; 990 if (entry->protection & VM_PROT_WRITE) 991 phdr->p_flags |= PF_W; 992 if (entry->protection & VM_PROT_EXECUTE) 993 phdr->p_flags |= PF_X; 994 995 phc->offset += phdr->p_filesz; 996 phc->phdr++; 997} 998 999/* 1000 * A callback for each_writable_segment() to gather information about 1001 * the number of segments and their total size. 1002 */ 1003static void 1004cb_size_segment(entry, closure) 1005 vm_map_entry_t entry; 1006 void *closure; 1007{ 1008 struct sseg_closure *ssc = (struct sseg_closure *)closure; 1009 1010 ssc->count++; 1011 ssc->size += entry->end - entry->start; 1012} 1013 1014/* 1015 * For each writable segment in the process's memory map, call the given 1016 * function with a pointer to the map entry and some arbitrary 1017 * caller-supplied data. 1018 */ 1019static void 1020each_writable_segment(td, func, closure) 1021 struct thread *td; 1022 segment_callback func; 1023 void *closure; 1024{ 1025 struct proc *p = td->td_proc; 1026 vm_map_t map = &p->p_vmspace->vm_map; 1027 vm_map_entry_t entry; 1028 1029 for (entry = map->header.next; entry != &map->header; 1030 entry = entry->next) { 1031 vm_object_t obj; 1032 1033 /* 1034 * Don't dump inaccessible mappings, deal with legacy 1035 * coredump mode. 1036 * 1037 * Note that read-only segments related to the elf binary 1038 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer 1039 * need to arbitrarily ignore such segments. 1040 */ 1041 if (elf_legacy_coredump) { 1042 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) 1043 continue; 1044 } else { 1045 if ((entry->protection & VM_PROT_ALL) == 0) 1046 continue; 1047 } 1048 1049 /* 1050 * Dont include memory segment in the coredump if 1051 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in 1052 * madvise(2). Do not dump submaps (i.e. parts of the 1053 * kernel map). 1054 */ 1055 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) 1056 continue; 1057 1058 if ((obj = entry->object.vm_object) == NULL) 1059 continue; 1060 1061 /* Find the deepest backing object. */ 1062 while (obj->backing_object != NULL) 1063 obj = obj->backing_object; 1064 1065 /* Ignore memory-mapped devices and such things. */ 1066 if (obj->type != OBJT_DEFAULT && 1067 obj->type != OBJT_SWAP && 1068 obj->type != OBJT_VNODE) 1069 continue; 1070 1071 (*func)(entry, closure); 1072 } 1073} 1074 1075/* 1076 * Write the core file header to the file, including padding up to 1077 * the page boundary. 1078 */ 1079static int 1080__elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize) 1081 struct thread *td; 1082 struct vnode *vp; 1083 struct ucred *cred; 1084 int numsegs; 1085 size_t hdrsize; 1086 void *hdr; 1087{ 1088 size_t off; 1089 1090 /* Fill in the header. */ 1091 bzero(hdr, hdrsize); 1092 off = 0; 1093 __elfN(puthdr)(td, hdr, &off, numsegs); 1094 1095 /* Write it to the core file. */ 1096 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0, 1097 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL, 1098 td)); /* XXXKSE */ 1099} 1100 1101#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1102typedef struct prstatus32 elf_prstatus_t; 1103typedef struct prpsinfo32 elf_prpsinfo_t; 1104typedef struct fpreg32 elf_prfpregset_t; 1105typedef struct fpreg32 elf_fpregset_t; 1106typedef struct reg32 elf_gregset_t; 1107#else 1108typedef prstatus_t elf_prstatus_t; 1109typedef prpsinfo_t elf_prpsinfo_t; 1110typedef prfpregset_t elf_prfpregset_t; 1111typedef prfpregset_t elf_fpregset_t; 1112typedef gregset_t elf_gregset_t; 1113#endif 1114 1115static void 1116__elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs) 1117{ 1118 struct { 1119 elf_prstatus_t status; 1120 elf_prfpregset_t fpregset; 1121 elf_prpsinfo_t psinfo; 1122 } *tempdata; 1123 elf_prstatus_t *status; 1124 elf_prfpregset_t *fpregset; 1125 elf_prpsinfo_t *psinfo; 1126 struct proc *p; 1127 struct thread *thr; 1128 size_t ehoff, noteoff, notesz, phoff; 1129 1130 p = td->td_proc; 1131 1132 ehoff = *off; 1133 *off += sizeof(Elf_Ehdr); 1134 1135 phoff = *off; 1136 *off += (numsegs + 1) * sizeof(Elf_Phdr); 1137 1138 noteoff = *off; 1139 /* 1140 * Don't allocate space for the notes if we're just calculating 1141 * the size of the header. We also don't collect the data. 1142 */ 1143 if (dst != NULL) { 1144 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK); 1145 status = &tempdata->status; 1146 fpregset = &tempdata->fpregset; 1147 psinfo = &tempdata->psinfo; 1148 } else { 1149 tempdata = NULL; 1150 status = NULL; 1151 fpregset = NULL; 1152 psinfo = NULL; 1153 } 1154 1155 if (dst != NULL) { 1156 psinfo->pr_version = PRPSINFO_VERSION; 1157 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); 1158 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); 1159 /* 1160 * XXX - We don't fill in the command line arguments properly 1161 * yet. 1162 */ 1163 strlcpy(psinfo->pr_psargs, p->p_comm, 1164 sizeof(psinfo->pr_psargs)); 1165 } 1166 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 1167 sizeof *psinfo); 1168 1169 /* 1170 * To have the debugger select the right thread (LWP) as the initial 1171 * thread, we dump the state of the thread passed to us in td first. 1172 * This is the thread that causes the core dump and thus likely to 1173 * be the right thread one wants to have selected in the debugger. 1174 */ 1175 thr = td; 1176 while (thr != NULL) { 1177 if (dst != NULL) { 1178 status->pr_version = PRSTATUS_VERSION; 1179 status->pr_statussz = sizeof(elf_prstatus_t); 1180 status->pr_gregsetsz = sizeof(elf_gregset_t); 1181 status->pr_fpregsetsz = sizeof(elf_fpregset_t); 1182 status->pr_osreldate = osreldate; 1183 status->pr_cursig = p->p_sig; 1184 status->pr_pid = thr->td_tid; 1185#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1186 fill_regs32(thr, &status->pr_reg); 1187 fill_fpregs32(thr, fpregset); 1188#else 1189 fill_regs(thr, &status->pr_reg); 1190 fill_fpregs(thr, fpregset); 1191#endif 1192 } 1193 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status, 1194 sizeof *status); 1195 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 1196 sizeof *fpregset); 1197 /* 1198 * Allow for MD specific notes, as well as any MD 1199 * specific preparations for writing MI notes. 1200 */ 1201 __elfN(dump_thread)(thr, dst, off); 1202 1203 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : 1204 TAILQ_NEXT(thr, td_plist); 1205 if (thr == td) 1206 thr = TAILQ_NEXT(thr, td_plist); 1207 } 1208 1209 notesz = *off - noteoff; 1210 1211 if (dst != NULL) 1212 free(tempdata, M_TEMP); 1213 1214 /* Align up to a page boundary for the program segments. */ 1215 *off = round_page(*off); 1216 1217 if (dst != NULL) { 1218 Elf_Ehdr *ehdr; 1219 Elf_Phdr *phdr; 1220 struct phdr_closure phc; 1221 1222 /* 1223 * Fill in the ELF header. 1224 */ 1225 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 1226 ehdr->e_ident[EI_MAG0] = ELFMAG0; 1227 ehdr->e_ident[EI_MAG1] = ELFMAG1; 1228 ehdr->e_ident[EI_MAG2] = ELFMAG2; 1229 ehdr->e_ident[EI_MAG3] = ELFMAG3; 1230 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 1231 ehdr->e_ident[EI_DATA] = ELF_DATA; 1232 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 1233 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 1234 ehdr->e_ident[EI_ABIVERSION] = 0; 1235 ehdr->e_ident[EI_PAD] = 0; 1236 ehdr->e_type = ET_CORE; 1237#if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32 1238 ehdr->e_machine = EM_386; 1239#else 1240 ehdr->e_machine = ELF_ARCH; 1241#endif 1242 ehdr->e_version = EV_CURRENT; 1243 ehdr->e_entry = 0; 1244 ehdr->e_phoff = phoff; 1245 ehdr->e_flags = 0; 1246 ehdr->e_ehsize = sizeof(Elf_Ehdr); 1247 ehdr->e_phentsize = sizeof(Elf_Phdr); 1248 ehdr->e_phnum = numsegs + 1; 1249 ehdr->e_shentsize = sizeof(Elf_Shdr); 1250 ehdr->e_shnum = 0; 1251 ehdr->e_shstrndx = SHN_UNDEF; 1252 1253 /* 1254 * Fill in the program header entries. 1255 */ 1256 phdr = (Elf_Phdr *)((char *)dst + phoff); 1257 1258 /* The note segement. */ 1259 phdr->p_type = PT_NOTE; 1260 phdr->p_offset = noteoff; 1261 phdr->p_vaddr = 0; 1262 phdr->p_paddr = 0; 1263 phdr->p_filesz = notesz; 1264 phdr->p_memsz = 0; 1265 phdr->p_flags = 0; 1266 phdr->p_align = 0; 1267 phdr++; 1268 1269 /* All the writable segments from the program. */ 1270 phc.phdr = phdr; 1271 phc.offset = *off; 1272 each_writable_segment(td, cb_put_phdr, &phc); 1273 } 1274} 1275 1276static void 1277__elfN(putnote)(void *dst, size_t *off, const char *name, int type, 1278 const void *desc, size_t descsz) 1279{ 1280 Elf_Note note; 1281 1282 note.n_namesz = strlen(name) + 1; 1283 note.n_descsz = descsz; 1284 note.n_type = type; 1285 if (dst != NULL) 1286 bcopy(¬e, (char *)dst + *off, sizeof note); 1287 *off += sizeof note; 1288 if (dst != NULL) 1289 bcopy(name, (char *)dst + *off, note.n_namesz); 1290 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 1291 if (dst != NULL) 1292 bcopy(desc, (char *)dst + *off, note.n_descsz); 1293 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 1294} 1295 1296/* 1297 * Tell kern_execve.c about it, with a little help from the linker. 1298 */ 1299static struct execsw __elfN(execsw) = { 1300 __CONCAT(exec_, __elfN(imgact)), 1301 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) 1302}; 1303EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); 1304