1 /* $OpenBSD: exec_elf.c,v 1.190 2024/08/21 03:16:25 deraadt Exp $ */ 2 3 /* 4 * Copyright (c) 1996 Per Fogelstrom 5 * All rights reserved. 6 * 7 * Copyright (c) 1994 Christos Zoulas 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. The name of the author may not be used to endorse or promote products 19 * derived from this software without specific prior written permission 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 22 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 23 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 24 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 26 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 30 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 31 * 32 */ 33 34 /* 35 * Copyright (c) 2001 Wasabi Systems, Inc. 36 * All rights reserved. 37 * 38 * Written by Jason R. Thorpe for Wasabi Systems, Inc. 39 * 40 * Redistribution and use in source and binary forms, with or without 41 * modification, are permitted provided that the following conditions 42 * are met: 43 * 1. Redistributions of source code must retain the above copyright 44 * notice, this list of conditions and the following disclaimer. 45 * 2. Redistributions in binary form must reproduce the above copyright 46 * notice, this list of conditions and the following disclaimer in the 47 * documentation and/or other materials provided with the distribution. 48 * 3. All advertising materials mentioning features or use of this software 49 * must display the following acknowledgement: 50 * This product includes software developed for the NetBSD Project by 51 * Wasabi Systems, Inc. 52 * 4. The name of Wasabi Systems, Inc. may not be used to endorse 53 * or promote products derived from this software without specific prior 54 * written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND 57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 58 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 59 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC 60 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 61 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 62 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 63 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 64 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 65 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 66 * POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/proc.h> 72 #include <sys/malloc.h> 73 #include <sys/pool.h> 74 #include <sys/mount.h> 75 #include <sys/namei.h> 76 #include <sys/vnode.h> 77 #include <sys/core.h> 78 #include <sys/exec.h> 79 #include <sys/exec_elf.h> 80 #include <sys/fcntl.h> 81 #include <sys/ptrace.h> 82 #include <sys/signalvar.h> 83 #include <sys/pledge.h> 84 #include <sys/syscall.h> 85 86 #include <sys/mman.h> 87 88 #include <uvm/uvm_extern.h> 89 90 #include <machine/reg.h> 91 #include <machine/exec.h> 92 #include <machine/elf.h> 93 94 int elf_load_file(struct proc *, char *, struct exec_package *, 95 struct elf_args *); 96 int elf_check_header(Elf_Ehdr *); 97 int elf_read_from(struct proc *, struct vnode *, u_long, void *, int); 98 void elf_load_psection(struct exec_vmcmd_set *, struct vnode *, 99 Elf_Phdr *, Elf_Addr *, Elf_Addr *, int *, int); 100 int elf_os_pt_note_name(Elf_Note *); 101 int elf_os_pt_note(struct proc *, struct exec_package *, Elf_Ehdr *, int *); 102 int elf_read_pintable(struct proc *p, struct vnode *vp, Elf_Phdr *pp, 103 u_int **pinp, int is_ldso, size_t len); 104 105 /* round up and down to page boundaries. */ 106 #define ELF_ROUND(a, b) (((a) + (b) - 1) & ~((b) - 1)) 107 #define ELF_TRUNC(a, b) ((a) & ~((b) - 1)) 108 109 /* 110 * We limit the number of program headers to 32, this should 111 * be a reasonable limit for ELF, the most we have seen so far is 12 112 */ 113 #define ELF_MAX_VALID_PHDR 32 114 115 #define ELF_NOTE_NAME_OPENBSD 0x01 116 117 struct elf_note_name { 118 char *name; 119 int id; 120 } elf_note_names[] = { 121 { "OpenBSD", ELF_NOTE_NAME_OPENBSD }, 122 }; 123 124 #define ELFROUNDSIZE sizeof(Elf_Word) 125 #define elfround(x) roundup((x), ELFROUNDSIZE) 126 127 128 /* 129 * Check header for validity; return 0 for ok, ENOEXEC if error 130 */ 131 int 132 elf_check_header(Elf_Ehdr *ehdr) 133 { 134 /* 135 * We need to check magic, class size, endianness, and version before 136 * we look at the rest of the Elf_Ehdr structure. These few elements 137 * are represented in a machine independent fashion. 138 */ 139 if (!IS_ELF(*ehdr) || 140 ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 141 ehdr->e_ident[EI_DATA] != ELF_TARG_DATA || 142 ehdr->e_ident[EI_VERSION] != ELF_TARG_VER) 143 return (ENOEXEC); 144 145 /* Now check the machine dependent header */ 146 if (ehdr->e_machine != ELF_TARG_MACH || 147 ehdr->e_version != ELF_TARG_VER) 148 return (ENOEXEC); 149 150 /* Don't allow an insane amount of sections. */ 151 if (ehdr->e_phnum > ELF_MAX_VALID_PHDR) 152 return (ENOEXEC); 153 154 return (0); 155 } 156 157 /* 158 * Load a psection at the appropriate address 159 */ 160 void 161 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp, 162 Elf_Phdr *ph, Elf_Addr *addr, Elf_Addr *size, int *prot, int flags) 163 { 164 u_long msize, lsize, psize, rm, rf; 165 long diff, offset, bdiff; 166 Elf_Addr base; 167 168 /* 169 * If the user specified an address, then we load there. 170 */ 171 if (*addr != ELF_NO_ADDR) { 172 if (ph->p_align > 1) { 173 *addr = ELF_TRUNC(*addr, ph->p_align); 174 diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align); 175 /* page align vaddr */ 176 base = *addr + trunc_page(ph->p_vaddr) 177 - ELF_TRUNC(ph->p_vaddr, ph->p_align); 178 } else { 179 diff = 0; 180 base = *addr + trunc_page(ph->p_vaddr) - ph->p_vaddr; 181 } 182 } else { 183 *addr = ph->p_vaddr; 184 if (ph->p_align > 1) 185 *addr = ELF_TRUNC(*addr, ph->p_align); 186 base = trunc_page(ph->p_vaddr); 187 diff = ph->p_vaddr - *addr; 188 } 189 bdiff = ph->p_vaddr - trunc_page(ph->p_vaddr); 190 191 /* 192 * Enforce W^X and map W|X segments without X permission 193 * initially. The dynamic linker will make these read-only 194 * and add back X permission after relocation processing. 195 * Static executables with W|X segments will probably crash. 196 */ 197 *prot |= (ph->p_flags & PF_R) ? PROT_READ : 0; 198 *prot |= (ph->p_flags & PF_W) ? PROT_WRITE : 0; 199 if ((ph->p_flags & PF_W) == 0) 200 *prot |= (ph->p_flags & PF_X) ? PROT_EXEC : 0; 201 202 /* 203 * Apply immutability as much as possible, but not text/rodata 204 * segments of textrel binaries, or RELRO or PT_OPENBSD_MUTABLE 205 * sections, or LOADS marked PF_OPENBSD_MUTABLE, or LOADS which 206 * violate W^X. 207 * Userland (meaning crt0 or ld.so) will repair those regions. 208 */ 209 if ((ph->p_flags & (PF_X | PF_W)) != (PF_X | PF_W) && 210 ((ph->p_flags & PF_OPENBSD_MUTABLE) == 0)) 211 flags |= VMCMD_IMMUTABLE; 212 if ((flags & VMCMD_TEXTREL) && (ph->p_flags & PF_W) == 0) 213 flags &= ~VMCMD_IMMUTABLE; 214 215 msize = ph->p_memsz + diff; 216 offset = ph->p_offset - bdiff; 217 lsize = ph->p_filesz + bdiff; 218 psize = round_page(lsize); 219 220 /* 221 * Because the pagedvn pager can't handle zero fill of the last 222 * data page if it's not page aligned we map the last page readvn. 223 */ 224 if (ph->p_flags & PF_W) { 225 psize = trunc_page(lsize); 226 if (psize > 0) 227 NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp, 228 offset, *prot, flags); 229 if (psize != lsize) { 230 NEW_VMCMD2(vcset, vmcmd_map_readvn, lsize - psize, 231 base + psize, vp, offset + psize, *prot, flags); 232 } 233 } else { 234 NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp, offset, 235 *prot, flags); 236 } 237 238 /* 239 * Check if we need to extend the size of the segment 240 */ 241 rm = round_page(*addr + ph->p_memsz + diff); 242 rf = round_page(*addr + ph->p_filesz + diff); 243 244 if (rm != rf) { 245 NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP, 0, 246 *prot, flags); 247 } 248 *size = msize; 249 } 250 251 /* 252 * Read from vnode into buffer at offset. 253 */ 254 int 255 elf_read_from(struct proc *p, struct vnode *vp, u_long off, void *buf, 256 int size) 257 { 258 int error; 259 size_t resid; 260 261 if ((error = vn_rdwr(UIO_READ, vp, buf, size, off, UIO_SYSSPACE, 262 0, p->p_ucred, &resid, p)) != 0) 263 return error; 264 /* 265 * See if we got all of it 266 */ 267 if (resid != 0) 268 return (ENOEXEC); 269 return (0); 270 } 271 272 /* 273 * rebase the pin offsets inside a base,len window for the text segment only. 274 */ 275 void 276 elf_adjustpins(vaddr_t *basep, size_t *lenp, u_int *pins, int npins, u_int offset) 277 { 278 int i; 279 280 /* Adjust offsets, base, len */ 281 for (i = 0; i < npins; i++) { 282 if (pins[i] == -1 || pins[i] == 0) 283 continue; 284 pins[i] -= offset; 285 } 286 *basep += offset; 287 *lenp -= offset; 288 } 289 290 int 291 elf_read_pintable(struct proc *p, struct vnode *vp, Elf_Phdr *pp, 292 u_int **pinp, int is_ldso, size_t len) 293 { 294 struct pinsyscalls { 295 u_int offset; 296 u_int sysno; 297 } *syscalls = NULL; 298 int i, nsyscalls = 0, npins = 0; 299 u_int *pins = NULL; 300 301 if (pp->p_filesz > SYS_MAXSYSCALL * 2 * sizeof(*syscalls) || 302 pp->p_filesz % sizeof(*syscalls) != 0) 303 goto bad; 304 nsyscalls = pp->p_filesz / sizeof(*syscalls); 305 syscalls = malloc(pp->p_filesz, M_PINSYSCALL, M_WAITOK); 306 if (elf_read_from(p, vp, pp->p_offset, syscalls, 307 pp->p_filesz) != 0) 308 goto bad; 309 310 /* Validate, and calculate pintable size */ 311 for (i = 0; i < nsyscalls; i++) { 312 if (syscalls[i].sysno <= 0 || 313 syscalls[i].sysno >= SYS_MAXSYSCALL || 314 syscalls[i].offset > len) 315 goto bad; 316 npins = MAX(npins, syscalls[i].sysno); 317 } 318 if (is_ldso) 319 npins = MAX(npins, SYS_kbind); /* XXX see ld.so/loader.c */ 320 npins++; 321 322 /* Fill pintable: 0 = invalid, -1 = allowed, else offset from base */ 323 pins = mallocarray(npins, sizeof(u_int), M_PINSYSCALL, M_WAITOK|M_ZERO); 324 for (i = 0; i < nsyscalls; i++) { 325 if (pins[syscalls[i].sysno]) 326 pins[syscalls[i].sysno] = -1; /* duplicated */ 327 else 328 pins[syscalls[i].sysno] = syscalls[i].offset; 329 } 330 if (is_ldso) 331 pins[SYS_kbind] = -1; /* XXX see ld.so/loader.c */ 332 *pinp = pins; 333 pins = NULL; 334 bad: 335 free(syscalls, M_PINSYSCALL, nsyscalls * sizeof(*syscalls)); 336 free(pins, M_PINSYSCALL, npins * sizeof(u_int)); 337 return npins; 338 } 339 340 /* 341 * Load a file (interpreter/library) pointed to by path [stolen from 342 * coff_load_shlib()]. Made slightly generic so it might be used externally. 343 */ 344 int 345 elf_load_file(struct proc *p, char *path, struct exec_package *epp, 346 struct elf_args *ap) 347 { 348 int error, i; 349 struct nameidata nd; 350 Elf_Ehdr eh; 351 Elf_Phdr *ph = NULL, *syscall_ph = NULL; 352 u_long phsize = 0; 353 Elf_Addr addr; 354 struct vnode *vp; 355 Elf_Phdr *base_ph = NULL; 356 struct interp_ld_sec { 357 Elf_Addr vaddr; 358 u_long memsz; 359 } loadmap[ELF_MAX_VALID_PHDR]; 360 int nload, idx = 0; 361 Elf_Addr pos; 362 int file_align; 363 int loop; 364 size_t randomizequota = ELF_RANDOMIZE_LIMIT; 365 vaddr_t text_start = -1, text_end = 0; 366 367 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path, p); 368 nd.ni_pledge = PLEDGE_RPATH; 369 nd.ni_unveil = UNVEIL_READ; 370 if ((error = namei(&nd)) != 0) { 371 return (error); 372 } 373 vp = nd.ni_vp; 374 if (vp->v_type != VREG) { 375 error = EACCES; 376 goto bad; 377 } 378 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 379 goto bad; 380 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 381 error = EACCES; 382 goto bad; 383 } 384 if ((error = VOP_ACCESS(vp, VREAD, p->p_ucred, p)) != 0) 385 goto bad1; 386 if ((error = elf_read_from(p, nd.ni_vp, 0, &eh, sizeof(eh))) != 0) 387 goto bad1; 388 389 if (elf_check_header(&eh) || eh.e_type != ET_DYN) { 390 error = ENOEXEC; 391 goto bad1; 392 } 393 394 ph = mallocarray(eh.e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK); 395 phsize = eh.e_phnum * sizeof(Elf_Phdr); 396 397 if ((error = elf_read_from(p, nd.ni_vp, eh.e_phoff, ph, phsize)) != 0) 398 goto bad1; 399 400 for (i = 0; i < eh.e_phnum; i++) { 401 if ((ph[i].p_align > 1) && !powerof2(ph[i].p_align)) { 402 error = EINVAL; 403 goto bad1; 404 } 405 406 if (ph[i].p_type == PT_LOAD) { 407 if (ph[i].p_filesz > ph[i].p_memsz || 408 ph[i].p_memsz == 0) { 409 error = EINVAL; 410 goto bad1; 411 } 412 loadmap[idx].vaddr = trunc_page(ph[i].p_vaddr); 413 loadmap[idx].memsz = round_page (ph[i].p_vaddr + 414 ph[i].p_memsz - loadmap[idx].vaddr); 415 file_align = ph[i].p_align; 416 idx++; 417 } 418 } 419 nload = idx; 420 421 /* 422 * Load the interpreter where a non-fixed mmap(NULL, ...) 423 * would (i.e. something safely out of the way). 424 */ 425 pos = uvm_map_hint(p->p_vmspace, PROT_EXEC, VM_MIN_ADDRESS, 426 VM_MAXUSER_ADDRESS); 427 pos = ELF_ROUND(pos, file_align); 428 429 loop = 0; 430 for (i = 0; i < nload;/**/) { 431 vaddr_t addr; 432 struct uvm_object *uobj; 433 off_t uoff; 434 size_t size; 435 436 #ifdef this_needs_fixing 437 if (i == 0) { 438 uobj = &vp->v_uvm.u_obj; 439 /* need to fix uoff */ 440 } else { 441 #endif 442 uobj = NULL; 443 uoff = 0; 444 #ifdef this_needs_fixing 445 } 446 #endif 447 448 addr = trunc_page(pos + loadmap[i].vaddr); 449 size = round_page(addr + loadmap[i].memsz) - addr; 450 451 /* CRAP - map_findspace does not avoid daddr+BRKSIZ */ 452 if ((addr + size > (vaddr_t)p->p_vmspace->vm_daddr) && 453 (addr < (vaddr_t)p->p_vmspace->vm_daddr + BRKSIZ)) 454 addr = round_page((vaddr_t)p->p_vmspace->vm_daddr + 455 BRKSIZ); 456 457 if (uvm_map_mquery(&p->p_vmspace->vm_map, &addr, size, 458 (i == 0 ? uoff : UVM_UNKNOWN_OFFSET), 0) != 0) { 459 if (loop == 0) { 460 loop = 1; 461 i = 0; 462 pos = 0; 463 continue; 464 } 465 error = ENOMEM; 466 goto bad1; 467 } 468 if (addr != pos + loadmap[i].vaddr) { 469 /* base changed. */ 470 pos = addr - trunc_page(loadmap[i].vaddr); 471 pos = ELF_ROUND(pos,file_align); 472 i = 0; 473 continue; 474 } 475 476 i++; 477 } 478 479 /* 480 * Load all the necessary sections 481 */ 482 for (i = 0; i < eh.e_phnum; i++) { 483 Elf_Addr size = 0; 484 int prot = 0; 485 int flags; 486 487 switch (ph[i].p_type) { 488 case PT_LOAD: 489 if (base_ph == NULL) { 490 flags = VMCMD_BASE; 491 addr = pos; 492 base_ph = &ph[i]; 493 } else { 494 flags = VMCMD_RELATIVE; 495 addr = ph[i].p_vaddr - base_ph->p_vaddr; 496 } 497 elf_load_psection(&epp->ep_vmcmds, nd.ni_vp, 498 &ph[i], &addr, &size, &prot, flags); 499 /* If entry is within this section it must be text */ 500 if (eh.e_entry >= ph[i].p_vaddr && 501 eh.e_entry < (ph[i].p_vaddr + size)) { 502 /* LOAD containing e_entry may not be writable */ 503 if (prot & PROT_WRITE) { 504 error = ENOEXEC; 505 goto bad1; 506 } 507 epp->ep_entry = addr + eh.e_entry - 508 ELF_TRUNC(ph[i].p_vaddr,ph[i].p_align); 509 if (flags == VMCMD_RELATIVE) 510 epp->ep_entry += pos; 511 ap->arg_interp = pos; 512 } 513 if (prot & PROT_EXEC) { 514 if (addr < text_start) 515 text_start = addr; 516 if (addr+size >= text_end) 517 text_end = addr + size; 518 } 519 addr += size; 520 break; 521 522 case PT_PHDR: 523 case PT_NOTE: 524 break; 525 526 case PT_OPENBSD_RANDOMIZE: 527 if (ph[i].p_memsz > randomizequota) { 528 error = ENOMEM; 529 goto bad1; 530 } 531 randomizequota -= ph[i].p_memsz; 532 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize, 533 ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0); 534 break; 535 536 case PT_DYNAMIC: 537 #if defined (__mips__) 538 /* DT_DEBUG is not ready on mips */ 539 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 540 ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0); 541 #endif 542 break; 543 case PT_GNU_RELRO: 544 case PT_OPENBSD_MUTABLE: 545 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 546 ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0); 547 break; 548 case PT_OPENBSD_SYSCALLS: 549 syscall_ph = &ph[i]; 550 break; 551 default: 552 break; 553 } 554 } 555 556 if (syscall_ph) { 557 struct process *pr = p->p_p; 558 vaddr_t base = pos; 559 size_t len = text_end; 560 u_int *pins; 561 int npins; 562 563 npins = elf_read_pintable(p, nd.ni_vp, syscall_ph, 564 &pins, 1, len); 565 if (npins) { 566 elf_adjustpins(&base, &len, pins, npins, 567 text_start); 568 pr->ps_pin.pn_start = base; 569 pr->ps_pin.pn_end = base + len; 570 pr->ps_pin.pn_pins = pins; 571 pr->ps_pin.pn_npins = npins; 572 } 573 } else { 574 error = EINVAL; /* no pin table */ 575 goto bad1; 576 } 577 578 vn_marktext(nd.ni_vp); 579 580 bad1: 581 VOP_CLOSE(nd.ni_vp, FREAD, p->p_ucred, p); 582 bad: 583 free(ph, M_TEMP, phsize); 584 585 vput(nd.ni_vp); 586 return (error); 587 } 588 589 /* 590 * Prepare an Elf binary's exec package 591 * 592 * First, set of the various offsets/lengths in the exec package. 593 * 594 * Then, mark the text image busy (so it can be demand paged) or error out if 595 * this is not possible. Finally, set up vmcmds for the text, data, bss, and 596 * stack segments. 597 */ 598 int 599 exec_elf_makecmds(struct proc *p, struct exec_package *epp) 600 { 601 Elf_Ehdr *eh = epp->ep_hdr; 602 Elf_Phdr *ph, *pp, *base_ph = NULL, *syscall_ph = NULL; 603 Elf_Addr phdr = 0, exe_base = 0, exe_end = 0; 604 int error, i, has_phdr = 0, names = 0, textrel = 0; 605 char *interp = NULL; 606 u_long phsize; 607 size_t randomizequota = ELF_RANDOMIZE_LIMIT; 608 609 if (epp->ep_hdrvalid < sizeof(Elf_Ehdr)) 610 return (ENOEXEC); 611 612 if (elf_check_header(eh) || 613 (eh->e_type != ET_EXEC && eh->e_type != ET_DYN)) 614 return (ENOEXEC); 615 616 /* 617 * check if vnode is in open for writing, because we want to demand- 618 * page out of it. if it is, don't do it, for various reasons. 619 */ 620 if (epp->ep_vp->v_writecount != 0) { 621 #ifdef DIAGNOSTIC 622 if (epp->ep_vp->v_flag & VTEXT) 623 panic("exec: a VTEXT vnode has writecount != 0"); 624 #endif 625 return (ETXTBSY); 626 } 627 /* 628 * Allocate space to hold all the program headers, and read them 629 * from the file 630 */ 631 ph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK); 632 phsize = eh->e_phnum * sizeof(Elf_Phdr); 633 634 if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff, ph, 635 phsize)) != 0) 636 goto bad; 637 638 epp->ep_tsize = ELF_NO_ADDR; 639 epp->ep_dsize = ELF_NO_ADDR; 640 641 for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) { 642 if ((pp->p_align > 1) && !powerof2(pp->p_align)) { 643 error = EINVAL; 644 goto bad; 645 } 646 647 if (pp->p_type == PT_INTERP && !interp) { 648 if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) 649 goto bad; 650 interp = pool_get(&namei_pool, PR_WAITOK); 651 if ((error = elf_read_from(p, epp->ep_vp, 652 pp->p_offset, interp, pp->p_filesz)) != 0) { 653 goto bad; 654 } 655 if (interp[pp->p_filesz - 1] != '\0') 656 goto bad; 657 } else if (pp->p_type == PT_LOAD) { 658 if (pp->p_filesz > pp->p_memsz || 659 pp->p_memsz == 0) { 660 error = EINVAL; 661 goto bad; 662 } 663 if (base_ph == NULL) 664 base_ph = pp; 665 } else if (pp->p_type == PT_PHDR) { 666 has_phdr = 1; 667 } 668 } 669 670 /* 671 * Verify this is an OpenBSD executable. If it's marked that way 672 * via a PT_NOTE then also check for a PT_OPENBSD_WXNEEDED segment. 673 */ 674 if ((error = elf_os_pt_note(p, epp, epp->ep_hdr, &names)) != 0) 675 goto bad; 676 if (eh->e_ident[EI_OSABI] == ELFOSABI_OPENBSD) 677 names |= ELF_NOTE_NAME_OPENBSD; 678 679 if (eh->e_type == ET_DYN) { 680 /* need phdr and load sections for PIE */ 681 if (!has_phdr || base_ph == NULL || base_ph->p_vaddr != 0) { 682 error = EINVAL; 683 goto bad; 684 } 685 /* randomize exe_base for PIE */ 686 exe_base = uvm_map_pie(base_ph->p_align); 687 688 /* 689 * Check if DYNAMIC contains DT_TEXTREL 690 */ 691 for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) { 692 Elf_Dyn *dt; 693 int j; 694 695 switch (pp->p_type) { 696 case PT_DYNAMIC: 697 if (pp->p_filesz > 64*1024) 698 break; 699 dt = malloc(pp->p_filesz, M_TEMP, M_WAITOK); 700 error = vn_rdwr(UIO_READ, epp->ep_vp, 701 (caddr_t)dt, pp->p_filesz, pp->p_offset, 702 UIO_SYSSPACE, IO_UNIT, p->p_ucred, NULL, p); 703 if (error) { 704 free(dt, M_TEMP, pp->p_filesz); 705 break; 706 } 707 for (j = 0; j < pp->p_filesz / sizeof(*dt); j++) { 708 if (dt[j].d_tag == DT_TEXTREL) { 709 textrel = VMCMD_TEXTREL; 710 break; 711 } 712 } 713 free(dt, M_TEMP, pp->p_filesz); 714 break; 715 default: 716 break; 717 } 718 } 719 } 720 721 /* 722 * Load all the necessary sections 723 */ 724 for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) { 725 Elf_Addr addr, size = 0; 726 int prot = 0; 727 int flags = 0; 728 729 switch (pp->p_type) { 730 case PT_LOAD: 731 if (exe_base != 0) { 732 if (pp == base_ph) { 733 flags = VMCMD_BASE; 734 addr = exe_base; 735 } else { 736 flags = VMCMD_RELATIVE; 737 addr = pp->p_vaddr - base_ph->p_vaddr; 738 } 739 } else 740 addr = ELF_NO_ADDR; 741 742 /* Static binaries may not call pinsyscalls() */ 743 if (interp == NULL) 744 p->p_vmspace->vm_map.flags |= VM_MAP_PINSYSCALL_ONCE; 745 746 /* 747 * Calculates size of text and data segments 748 * by starting at first and going to end of last. 749 * 'rwx' sections are treated as data. 750 * this is correct for BSS_PLT, but may not be 751 * for DATA_PLT, is fine for TEXT_PLT. 752 */ 753 elf_load_psection(&epp->ep_vmcmds, epp->ep_vp, 754 pp, &addr, &size, &prot, flags | textrel); 755 756 /* 757 * Update exe_base in case alignment was off. 758 * For PIE, addr is relative to exe_base so 759 * adjust it (non PIE exe_base is 0 so no change). 760 */ 761 if (flags == VMCMD_BASE) 762 exe_base = addr; 763 else 764 addr += exe_base; 765 766 /* 767 * Decide whether it's text or data by looking 768 * at the protection of the section 769 */ 770 if (prot & PROT_WRITE) { 771 /* data section */ 772 if (epp->ep_dsize == ELF_NO_ADDR) { 773 epp->ep_daddr = addr; 774 epp->ep_dsize = size; 775 } else { 776 if (addr < epp->ep_daddr) { 777 epp->ep_dsize = 778 epp->ep_dsize + 779 epp->ep_daddr - 780 addr; 781 epp->ep_daddr = addr; 782 } else 783 epp->ep_dsize = addr+size - 784 epp->ep_daddr; 785 } 786 } else if (prot & PROT_EXEC) { 787 /* text section */ 788 if (epp->ep_tsize == ELF_NO_ADDR) { 789 epp->ep_taddr = addr; 790 epp->ep_tsize = size; 791 } else { 792 if (addr < epp->ep_taddr) { 793 epp->ep_tsize = 794 epp->ep_tsize + 795 epp->ep_taddr - 796 addr; 797 epp->ep_taddr = addr; 798 } else 799 epp->ep_tsize = addr+size - 800 epp->ep_taddr; 801 } 802 if (interp == NULL) 803 exe_end = epp->ep_taddr + 804 epp->ep_tsize; /* end of TEXT */ 805 } 806 break; 807 808 case PT_SHLIB: 809 error = ENOEXEC; 810 goto bad; 811 812 case PT_INTERP: 813 /* Already did this one */ 814 case PT_NOTE: 815 break; 816 817 case PT_PHDR: 818 /* Note address of program headers (in text segment) */ 819 phdr = pp->p_vaddr; 820 break; 821 822 case PT_OPENBSD_RANDOMIZE: 823 if (ph[i].p_memsz > randomizequota) { 824 error = ENOMEM; 825 goto bad; 826 } 827 randomizequota -= ph[i].p_memsz; 828 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize, 829 ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0); 830 break; 831 832 case PT_DYNAMIC: 833 #if defined (__mips__) 834 /* DT_DEBUG is not ready on mips */ 835 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 836 ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0); 837 #endif 838 break; 839 case PT_GNU_RELRO: 840 case PT_OPENBSD_MUTABLE: 841 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_mutable, 842 ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0); 843 break; 844 case PT_OPENBSD_SYSCALLS: 845 if (interp == NULL) 846 syscall_ph = &ph[i]; 847 break; 848 default: 849 /* 850 * Not fatal, we don't need to understand everything 851 * :-) 852 */ 853 break; 854 } 855 } 856 857 if (syscall_ph) { 858 vaddr_t base = exe_base; 859 size_t len = exe_end - exe_base; 860 u_int *pins; 861 int npins; 862 863 npins = elf_read_pintable(p, epp->ep_vp, syscall_ph, 864 &pins, 0, len); 865 if (npins) { 866 elf_adjustpins(&base, &len, pins, npins, 867 epp->ep_taddr - exe_base); 868 epp->ep_pinstart = base; 869 epp->ep_pinend = base + len; 870 epp->ep_pins = pins; 871 epp->ep_npins = npins; 872 } 873 } 874 875 phdr += exe_base; 876 877 /* 878 * Strangely some linux programs may have all load sections marked 879 * writeable, in this case, textsize is not -1, but rather 0; 880 */ 881 if (epp->ep_tsize == ELF_NO_ADDR) 882 epp->ep_tsize = 0; 883 /* 884 * Another possibility is that it has all load sections marked 885 * read-only. Fake a zero-sized data segment right after the 886 * text segment. 887 */ 888 if (epp->ep_dsize == ELF_NO_ADDR) { 889 epp->ep_daddr = round_page(epp->ep_taddr + epp->ep_tsize); 890 epp->ep_dsize = 0; 891 } 892 893 epp->ep_interp = interp; 894 epp->ep_entry = eh->e_entry + exe_base; 895 896 /* 897 * Check if we found a dynamically linked binary and arrange to load 898 * its interpreter when the exec file is released. 899 */ 900 if (interp || eh->e_type == ET_DYN) { 901 struct elf_args *ap; 902 903 ap = malloc(sizeof(*ap), M_TEMP, M_WAITOK); 904 905 ap->arg_phaddr = phdr; 906 ap->arg_phentsize = eh->e_phentsize; 907 ap->arg_phnum = eh->e_phnum; 908 ap->arg_entry = eh->e_entry + exe_base; 909 ap->arg_interp = exe_base; 910 911 epp->ep_args = ap; 912 } 913 914 free(ph, M_TEMP, phsize); 915 vn_marktext(epp->ep_vp); 916 return (exec_setup_stack(p, epp)); 917 918 bad: 919 if (interp) 920 pool_put(&namei_pool, interp); 921 free(ph, M_TEMP, phsize); 922 kill_vmcmds(&epp->ep_vmcmds); 923 if (error == 0) 924 return (ENOEXEC); 925 return (error); 926 } 927 928 #ifdef __HAVE_CPU_HWCAP 929 unsigned long hwcap; 930 #endif /* __HAVE_CPU_HWCAP */ 931 932 #ifdef __HAVE_CPU_HWCAP2 933 unsigned long hwcap2; 934 #endif /* __HAVE_CPU_HWCAP2 */ 935 936 /* 937 * Phase II of load. It is now safe to load the interpreter. Info collected 938 * when loading the program is available for setup of the interpreter. 939 */ 940 int 941 exec_elf_fixup(struct proc *p, struct exec_package *epp) 942 { 943 char *interp; 944 int error = 0; 945 struct elf_args *ap; 946 AuxInfo ai[ELF_AUX_ENTRIES], *a; 947 948 ap = epp->ep_args; 949 if (ap == NULL) { 950 return (0); 951 } 952 953 interp = epp->ep_interp; 954 955 /* disable kbind in programs that don't use ld.so */ 956 if (interp == NULL) 957 p->p_p->ps_kbind_addr = BOGO_PC; 958 959 if (interp && 960 (error = elf_load_file(p, interp, epp, ap)) != 0) { 961 uprintf("execve: cannot load %s\n", interp); 962 free(ap, M_TEMP, sizeof *ap); 963 pool_put(&namei_pool, interp); 964 kill_vmcmds(&epp->ep_vmcmds); 965 return (error); 966 } 967 /* 968 * We have to do this ourselves... 969 */ 970 error = exec_process_vmcmds(p, epp); 971 972 /* 973 * Push extra arguments on the stack needed by dynamically 974 * linked binaries 975 */ 976 if (error == 0) { 977 memset(&ai, 0, sizeof ai); 978 a = ai; 979 980 a->au_id = AUX_phdr; 981 a->au_v = ap->arg_phaddr; 982 a++; 983 984 a->au_id = AUX_phent; 985 a->au_v = ap->arg_phentsize; 986 a++; 987 988 a->au_id = AUX_phnum; 989 a->au_v = ap->arg_phnum; 990 a++; 991 992 a->au_id = AUX_pagesz; 993 a->au_v = PAGE_SIZE; 994 a++; 995 996 a->au_id = AUX_base; 997 a->au_v = ap->arg_interp; 998 a++; 999 1000 a->au_id = AUX_flags; 1001 a->au_v = 0; 1002 a++; 1003 1004 a->au_id = AUX_entry; 1005 a->au_v = ap->arg_entry; 1006 a++; 1007 1008 #ifdef __HAVE_CPU_HWCAP 1009 a->au_id = AUX_hwcap; 1010 a->au_v = hwcap; 1011 a++; 1012 #endif /* __HAVE_CPU_HWCAP */ 1013 1014 #ifdef __HAVE_CPU_HWCAP2 1015 a->au_id = AUX_hwcap2; 1016 a->au_v = hwcap2; 1017 a++; 1018 #endif /* __HAVE_CPU_HWCAP2 */ 1019 1020 a->au_id = AUX_openbsd_timekeep; 1021 a->au_v = p->p_p->ps_timekeep; 1022 a++; 1023 1024 a->au_id = AUX_null; 1025 a->au_v = 0; 1026 a++; 1027 1028 error = copyout(ai, epp->ep_auxinfo, sizeof ai); 1029 } 1030 free(ap, M_TEMP, sizeof *ap); 1031 if (interp) 1032 pool_put(&namei_pool, interp); 1033 return (error); 1034 } 1035 1036 int 1037 elf_os_pt_note_name(Elf_Note *np) 1038 { 1039 int i, j; 1040 1041 for (i = 0; i < nitems(elf_note_names); i++) { 1042 size_t namlen = strlen(elf_note_names[i].name); 1043 if (np->namesz < namlen) 1044 continue; 1045 /* verify name padding (after the NUL) is NUL */ 1046 for (j = namlen + 1; j < elfround(np->namesz); j++) 1047 if (((char *)(np + 1))[j] != '\0') 1048 continue; 1049 /* verify desc padding is NUL */ 1050 for (j = np->descsz; j < elfround(np->descsz); j++) 1051 if (((char *)(np + 1))[j] != '\0') 1052 continue; 1053 if (strcmp((char *)(np + 1), elf_note_names[i].name) == 0) 1054 return elf_note_names[i].id; 1055 } 1056 return (0); 1057 } 1058 1059 int 1060 elf_os_pt_note(struct proc *p, struct exec_package *epp, Elf_Ehdr *eh, int *namesp) 1061 { 1062 Elf_Phdr *hph, *ph; 1063 Elf_Note *np = NULL; 1064 size_t phsize, offset, pfilesz = 0, total; 1065 int error, names = 0; 1066 1067 hph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK); 1068 phsize = eh->e_phnum * sizeof(Elf_Phdr); 1069 if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff, 1070 hph, phsize)) != 0) 1071 goto out1; 1072 1073 for (ph = hph; ph < &hph[eh->e_phnum]; ph++) { 1074 if (ph->p_type == PT_OPENBSD_WXNEEDED) { 1075 epp->ep_flags |= EXEC_WXNEEDED; 1076 continue; 1077 } 1078 if (ph->p_type == PT_OPENBSD_NOBTCFI) { 1079 epp->ep_flags |= EXEC_NOBTCFI; 1080 continue; 1081 } 1082 1083 if (ph->p_type != PT_NOTE || ph->p_filesz > 1024) 1084 continue; 1085 1086 if (np && ph->p_filesz != pfilesz) { 1087 free(np, M_TEMP, pfilesz); 1088 np = NULL; 1089 } 1090 if (!np) 1091 np = malloc(ph->p_filesz, M_TEMP, M_WAITOK); 1092 pfilesz = ph->p_filesz; 1093 if ((error = elf_read_from(p, epp->ep_vp, ph->p_offset, 1094 np, ph->p_filesz)) != 0) 1095 goto out2; 1096 1097 for (offset = 0; offset < ph->p_filesz; offset += total) { 1098 Elf_Note *np2 = (Elf_Note *)((char *)np + offset); 1099 1100 if (offset + sizeof(Elf_Note) > ph->p_filesz) 1101 break; 1102 total = sizeof(Elf_Note) + elfround(np2->namesz) + 1103 elfround(np2->descsz); 1104 if (offset + total > ph->p_filesz) 1105 break; 1106 names |= elf_os_pt_note_name(np2); 1107 } 1108 } 1109 1110 out2: 1111 free(np, M_TEMP, pfilesz); 1112 out1: 1113 free(hph, M_TEMP, phsize); 1114 *namesp = names; 1115 return ((names & ELF_NOTE_NAME_OPENBSD) ? 0 : ENOEXEC); 1116 } 1117 1118 /* 1119 * Start of routines related to dumping core 1120 */ 1121 1122 #ifdef SMALL_KERNEL 1123 int 1124 coredump_elf(struct proc *p, void *cookie) 1125 { 1126 return EPERM; 1127 } 1128 #else /* !SMALL_KERNEL */ 1129 1130 struct writesegs_state { 1131 off_t notestart; 1132 off_t secstart; 1133 off_t secoff; 1134 struct proc *p; 1135 void *iocookie; 1136 Elf_Phdr *psections; 1137 size_t psectionslen; 1138 size_t notesize; 1139 int npsections; 1140 }; 1141 1142 uvm_coredump_setup_cb coredump_setup_elf; 1143 uvm_coredump_walk_cb coredump_walk_elf; 1144 1145 int coredump_notes_elf(struct proc *, void *, size_t *); 1146 int coredump_note_elf(struct proc *, void *, size_t *); 1147 int coredump_writenote_elf(struct proc *, void *, Elf_Note *, 1148 const char *, void *); 1149 1150 extern vaddr_t sigcode_va; 1151 extern vsize_t sigcode_sz; 1152 1153 int 1154 coredump_elf(struct proc *p, void *cookie) 1155 { 1156 #ifdef DIAGNOSTIC 1157 off_t offset; 1158 #endif 1159 struct writesegs_state ws; 1160 size_t notesize; 1161 int error, i; 1162 1163 ws.p = p; 1164 ws.iocookie = cookie; 1165 ws.psections = NULL; 1166 1167 /* 1168 * Walk the map to get all the segment offsets and lengths, 1169 * write out the ELF header. 1170 */ 1171 error = uvm_coredump_walkmap(p, coredump_setup_elf, 1172 coredump_walk_elf, &ws); 1173 if (error) 1174 goto out; 1175 1176 error = coredump_write(cookie, UIO_SYSSPACE, ws.psections, 1177 ws.psectionslen, 0); 1178 if (error) 1179 goto out; 1180 1181 /* Write out the notes. */ 1182 error = coredump_notes_elf(p, cookie, ¬esize); 1183 if (error) 1184 goto out; 1185 1186 #ifdef DIAGNOSTIC 1187 if (notesize != ws.notesize) 1188 panic("coredump: notesize changed: %zu != %zu", 1189 ws.notesize, notesize); 1190 offset = ws.notestart + notesize; 1191 if (offset != ws.secstart) 1192 panic("coredump: offset %lld != secstart %lld", 1193 (long long) offset, (long long) ws.secstart); 1194 #endif 1195 1196 /* Pass 3: finally, write the sections themselves. */ 1197 for (i = 0; i < ws.npsections - 1; i++) { 1198 Elf_Phdr *pent = &ws.psections[i]; 1199 if (pent->p_filesz == 0) 1200 continue; 1201 1202 #ifdef DIAGNOSTIC 1203 if (offset != pent->p_offset) 1204 panic("coredump: offset %lld != p_offset[%d] %lld", 1205 (long long) offset, i, 1206 (long long) pent->p_filesz); 1207 #endif 1208 1209 /* 1210 * Since the sigcode is mapped execute-only, we can't 1211 * read it. So use the kernel mapping for it instead. 1212 */ 1213 if (pent->p_vaddr == p->p_p->ps_sigcode && 1214 pent->p_filesz == sigcode_sz) { 1215 error = coredump_write(cookie, UIO_SYSSPACE, 1216 (void *)sigcode_va, sigcode_sz, 0); 1217 } else { 1218 error = coredump_write(cookie, UIO_USERSPACE, 1219 (void *)(vaddr_t)pent->p_vaddr, pent->p_filesz, 1220 (pent->p_flags & PF_ISVNODE)); 1221 } 1222 if (error) 1223 goto out; 1224 1225 coredump_unmap(cookie, (vaddr_t)pent->p_vaddr, 1226 (vaddr_t)pent->p_vaddr + pent->p_filesz); 1227 1228 #ifdef DIAGNOSTIC 1229 offset += ws.psections[i].p_filesz; 1230 #endif 1231 } 1232 1233 out: 1234 free(ws.psections, M_TEMP, ws.psectionslen); 1235 return (error); 1236 } 1237 1238 1239 /* 1240 * Normally we lay out core files like this: 1241 * [ELF Header] [Program headers] [Notes] [data for PT_LOAD segments] 1242 * 1243 * However, if there's >= 65535 segments then it overflows the field 1244 * in the ELF header, so the standard specifies putting a magic 1245 * number there and saving the real count in the .sh_info field of 1246 * the first *section* header...which requires generating a section 1247 * header. To avoid confusing tools, we include an .shstrtab section 1248 * as well so all the indexes look valid. So in this case we lay 1249 * out the core file like this: 1250 * [ELF Header] [Section Headers] [.shstrtab] [Program headers] \ 1251 * [Notes] [data for PT_LOAD segments] 1252 * 1253 * The 'shstrtab' structure below is data for the second of the two 1254 * section headers, plus the .shstrtab itself, in one const buffer. 1255 */ 1256 static const struct { 1257 Elf_Shdr shdr; 1258 char shstrtab[sizeof(ELF_SHSTRTAB) + 1]; 1259 } shstrtab = { 1260 .shdr = { 1261 .sh_name = 1, /* offset in .shstrtab below */ 1262 .sh_type = SHT_STRTAB, 1263 .sh_offset = sizeof(Elf_Ehdr) + 2*sizeof(Elf_Shdr), 1264 .sh_size = sizeof(ELF_SHSTRTAB) + 1, 1265 .sh_addralign = 1, 1266 }, 1267 .shstrtab = "\0" ELF_SHSTRTAB, 1268 }; 1269 1270 int 1271 coredump_setup_elf(int segment_count, void *cookie) 1272 { 1273 Elf_Ehdr ehdr; 1274 struct writesegs_state *ws = cookie; 1275 Elf_Phdr *note; 1276 int error; 1277 1278 /* Get the count of segments, plus one for the PT_NOTE */ 1279 ws->npsections = segment_count + 1; 1280 1281 /* Get the size of the notes. */ 1282 error = coredump_notes_elf(ws->p, NULL, &ws->notesize); 1283 if (error) 1284 return error; 1285 1286 /* Setup the ELF header */ 1287 memset(&ehdr, 0, sizeof(ehdr)); 1288 memcpy(ehdr.e_ident, ELFMAG, SELFMAG); 1289 ehdr.e_ident[EI_CLASS] = ELF_TARG_CLASS; 1290 ehdr.e_ident[EI_DATA] = ELF_TARG_DATA; 1291 ehdr.e_ident[EI_VERSION] = EV_CURRENT; 1292 /* XXX Should be the OSABI/ABI version of the executable. */ 1293 ehdr.e_ident[EI_OSABI] = ELFOSABI_SYSV; 1294 ehdr.e_ident[EI_ABIVERSION] = 0; 1295 ehdr.e_type = ET_CORE; 1296 /* XXX This should be the e_machine of the executable. */ 1297 ehdr.e_machine = ELF_TARG_MACH; 1298 ehdr.e_version = EV_CURRENT; 1299 ehdr.e_entry = 0; 1300 ehdr.e_flags = 0; 1301 ehdr.e_ehsize = sizeof(ehdr); 1302 ehdr.e_phentsize = sizeof(Elf_Phdr); 1303 1304 if (ws->npsections < PN_XNUM) { 1305 ehdr.e_phoff = sizeof(ehdr); 1306 ehdr.e_shoff = 0; 1307 ehdr.e_phnum = ws->npsections; 1308 ehdr.e_shentsize = 0; 1309 ehdr.e_shnum = 0; 1310 ehdr.e_shstrndx = 0; 1311 } else { 1312 /* too many segments, use extension setup */ 1313 ehdr.e_shoff = sizeof(ehdr); 1314 ehdr.e_phnum = PN_XNUM; 1315 ehdr.e_shentsize = sizeof(Elf_Shdr); 1316 ehdr.e_shnum = 2; 1317 ehdr.e_shstrndx = 1; 1318 ehdr.e_phoff = shstrtab.shdr.sh_offset + shstrtab.shdr.sh_size; 1319 } 1320 1321 /* Write out the ELF header. */ 1322 error = coredump_write(ws->iocookie, UIO_SYSSPACE, &ehdr, sizeof(ehdr), 0); 1323 if (error) 1324 return error; 1325 1326 /* 1327 * If an section header is needed to store extension info, write 1328 * it out after the ELF header and before the program header. 1329 */ 1330 if (ehdr.e_shnum != 0) { 1331 Elf_Shdr shdr = { .sh_info = ws->npsections }; 1332 error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shdr, 1333 sizeof shdr, 0); 1334 if (error) 1335 return error; 1336 error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shstrtab, 1337 sizeof(shstrtab.shdr) + sizeof(shstrtab.shstrtab), 0); 1338 if (error) 1339 return error; 1340 } 1341 1342 /* 1343 * Allocate the segment header array and setup to collect 1344 * the section sizes and offsets 1345 */ 1346 ws->psections = mallocarray(ws->npsections, sizeof(Elf_Phdr), 1347 M_TEMP, M_WAITOK|M_CANFAIL|M_ZERO); 1348 if (ws->psections == NULL) 1349 return ENOMEM; 1350 ws->psectionslen = ws->npsections * sizeof(Elf_Phdr); 1351 1352 ws->notestart = ehdr.e_phoff + ws->psectionslen; 1353 ws->secstart = ws->notestart + ws->notesize; 1354 ws->secoff = ws->secstart; 1355 1356 /* Fill in the PT_NOTE segment header in the last slot */ 1357 note = &ws->psections[ws->npsections - 1]; 1358 note->p_type = PT_NOTE; 1359 note->p_offset = ws->notestart; 1360 note->p_vaddr = 0; 1361 note->p_paddr = 0; 1362 note->p_filesz = ws->notesize; 1363 note->p_memsz = 0; 1364 note->p_flags = PF_R; 1365 note->p_align = ELFROUNDSIZE; 1366 1367 return (0); 1368 } 1369 1370 int 1371 coredump_walk_elf(vaddr_t start, vaddr_t realend, vaddr_t end, vm_prot_t prot, 1372 int isvnode, int nsegment, void *cookie) 1373 { 1374 struct writesegs_state *ws = cookie; 1375 Elf_Phdr phdr; 1376 vsize_t size, realsize; 1377 1378 size = end - start; 1379 realsize = realend - start; 1380 1381 phdr.p_type = PT_LOAD; 1382 phdr.p_offset = ws->secoff; 1383 phdr.p_vaddr = start; 1384 phdr.p_paddr = 0; 1385 phdr.p_filesz = realsize; 1386 phdr.p_memsz = size; 1387 phdr.p_flags = 0; 1388 if (prot & PROT_READ) 1389 phdr.p_flags |= PF_R; 1390 if (prot & PROT_WRITE) 1391 phdr.p_flags |= PF_W; 1392 if (prot & PROT_EXEC) 1393 phdr.p_flags |= PF_X; 1394 if (isvnode) 1395 phdr.p_flags |= PF_ISVNODE; 1396 phdr.p_align = PAGE_SIZE; 1397 1398 ws->secoff += phdr.p_filesz; 1399 ws->psections[nsegment] = phdr; 1400 1401 return (0); 1402 } 1403 1404 int 1405 coredump_notes_elf(struct proc *p, void *iocookie, size_t *sizep) 1406 { 1407 struct elfcore_procinfo cpi; 1408 Elf_Note nhdr; 1409 struct process *pr = p->p_p; 1410 struct proc *q; 1411 size_t size, notesize; 1412 int error; 1413 1414 KASSERT(!P_HASSIBLING(p) || pr->ps_single != NULL); 1415 size = 0; 1416 1417 /* First, write an elfcore_procinfo. */ 1418 notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) + 1419 elfround(sizeof(cpi)); 1420 if (iocookie) { 1421 memset(&cpi, 0, sizeof(cpi)); 1422 1423 cpi.cpi_version = ELFCORE_PROCINFO_VERSION; 1424 cpi.cpi_cpisize = sizeof(cpi); 1425 cpi.cpi_signo = p->p_sisig; 1426 cpi.cpi_sigcode = p->p_sicode; 1427 1428 cpi.cpi_sigpend = p->p_siglist | pr->ps_siglist; 1429 cpi.cpi_sigmask = p->p_sigmask; 1430 cpi.cpi_sigignore = pr->ps_sigacts->ps_sigignore; 1431 cpi.cpi_sigcatch = pr->ps_sigacts->ps_sigcatch; 1432 1433 cpi.cpi_pid = pr->ps_pid; 1434 cpi.cpi_ppid = pr->ps_ppid; 1435 cpi.cpi_pgrp = pr->ps_pgid; 1436 if (pr->ps_session->s_leader) 1437 cpi.cpi_sid = pr->ps_session->s_leader->ps_pid; 1438 else 1439 cpi.cpi_sid = 0; 1440 1441 cpi.cpi_ruid = p->p_ucred->cr_ruid; 1442 cpi.cpi_euid = p->p_ucred->cr_uid; 1443 cpi.cpi_svuid = p->p_ucred->cr_svuid; 1444 1445 cpi.cpi_rgid = p->p_ucred->cr_rgid; 1446 cpi.cpi_egid = p->p_ucred->cr_gid; 1447 cpi.cpi_svgid = p->p_ucred->cr_svgid; 1448 1449 (void)strlcpy(cpi.cpi_name, pr->ps_comm, sizeof(cpi.cpi_name)); 1450 1451 nhdr.namesz = sizeof("OpenBSD"); 1452 nhdr.descsz = sizeof(cpi); 1453 nhdr.type = NT_OPENBSD_PROCINFO; 1454 1455 error = coredump_writenote_elf(p, iocookie, &nhdr, 1456 "OpenBSD", &cpi); 1457 if (error) 1458 return (error); 1459 } 1460 size += notesize; 1461 1462 /* Second, write an NT_OPENBSD_AUXV note. */ 1463 notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) + 1464 elfround(ELF_AUX_WORDS * sizeof(char *)); 1465 if (iocookie && pr->ps_auxinfo) { 1466 1467 nhdr.namesz = sizeof("OpenBSD"); 1468 nhdr.descsz = ELF_AUX_WORDS * sizeof(char *); 1469 nhdr.type = NT_OPENBSD_AUXV; 1470 1471 error = coredump_write(iocookie, UIO_SYSSPACE, 1472 &nhdr, sizeof(nhdr), 0); 1473 if (error) 1474 return (error); 1475 1476 error = coredump_write(iocookie, UIO_SYSSPACE, 1477 "OpenBSD", elfround(nhdr.namesz), 0); 1478 if (error) 1479 return (error); 1480 1481 error = coredump_write(iocookie, UIO_USERSPACE, 1482 (caddr_t)pr->ps_auxinfo, nhdr.descsz, 0); 1483 if (error) 1484 return (error); 1485 } 1486 size += notesize; 1487 1488 #ifdef PT_WCOOKIE 1489 notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) + 1490 elfround(sizeof(register_t)); 1491 if (iocookie) { 1492 register_t wcookie; 1493 1494 nhdr.namesz = sizeof("OpenBSD"); 1495 nhdr.descsz = sizeof(register_t); 1496 nhdr.type = NT_OPENBSD_WCOOKIE; 1497 1498 wcookie = process_get_wcookie(p); 1499 error = coredump_writenote_elf(p, iocookie, &nhdr, 1500 "OpenBSD", &wcookie); 1501 if (error) 1502 return (error); 1503 } 1504 size += notesize; 1505 #endif 1506 1507 /* 1508 * Now write the register info for the thread that caused the 1509 * coredump. 1510 */ 1511 error = coredump_note_elf(p, iocookie, ¬esize); 1512 if (error) 1513 return (error); 1514 size += notesize; 1515 1516 /* 1517 * Now, for each thread, write the register info and any other 1518 * per-thread notes. Since we're dumping core, all the other 1519 * threads in the process have been stopped and the list can't 1520 * change. 1521 */ 1522 TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) { 1523 if (q == p) /* we've taken care of this thread */ 1524 continue; 1525 error = coredump_note_elf(q, iocookie, ¬esize); 1526 if (error) 1527 return (error); 1528 size += notesize; 1529 } 1530 1531 *sizep = size; 1532 return (0); 1533 } 1534 1535 int 1536 coredump_note_elf(struct proc *p, void *iocookie, size_t *sizep) 1537 { 1538 Elf_Note nhdr; 1539 int size, notesize, error; 1540 int namesize; 1541 char name[64+ELFROUNDSIZE]; 1542 struct reg intreg; 1543 #ifdef PT_GETFPREGS 1544 struct fpreg freg; 1545 #endif 1546 #ifdef PT_PACMASK 1547 register_t pacmask[2]; 1548 #endif 1549 1550 size = 0; 1551 1552 snprintf(name, sizeof(name)-ELFROUNDSIZE, "%s@%d", 1553 "OpenBSD", p->p_tid + THREAD_PID_OFFSET); 1554 namesize = strlen(name) + 1; 1555 memset(name + namesize, 0, elfround(namesize) - namesize); 1556 1557 notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(intreg)); 1558 if (iocookie) { 1559 error = process_read_regs(p, &intreg); 1560 if (error) 1561 return (error); 1562 1563 nhdr.namesz = namesize; 1564 nhdr.descsz = sizeof(intreg); 1565 nhdr.type = NT_OPENBSD_REGS; 1566 1567 error = coredump_writenote_elf(p, iocookie, &nhdr, 1568 name, &intreg); 1569 if (error) 1570 return (error); 1571 1572 } 1573 size += notesize; 1574 1575 #ifdef PT_GETFPREGS 1576 notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(freg)); 1577 if (iocookie) { 1578 error = process_read_fpregs(p, &freg); 1579 if (error) 1580 return (error); 1581 1582 nhdr.namesz = namesize; 1583 nhdr.descsz = sizeof(freg); 1584 nhdr.type = NT_OPENBSD_FPREGS; 1585 1586 error = coredump_writenote_elf(p, iocookie, &nhdr, name, &freg); 1587 if (error) 1588 return (error); 1589 } 1590 size += notesize; 1591 #endif 1592 1593 #ifdef PT_PACMASK 1594 notesize = sizeof(nhdr) + elfround(namesize) + 1595 elfround(sizeof(pacmask)); 1596 if (iocookie) { 1597 pacmask[0] = pacmask[1] = process_get_pacmask(p); 1598 1599 nhdr.namesz = namesize; 1600 nhdr.descsz = sizeof(pacmask); 1601 nhdr.type = NT_OPENBSD_PACMASK; 1602 1603 error = coredump_writenote_elf(p, iocookie, &nhdr, 1604 name, &pacmask); 1605 if (error) 1606 return (error); 1607 } 1608 size += notesize; 1609 #endif 1610 1611 *sizep = size; 1612 /* XXX Add hook for machdep per-LWP notes. */ 1613 return (0); 1614 } 1615 1616 int 1617 coredump_writenote_elf(struct proc *p, void *cookie, Elf_Note *nhdr, 1618 const char *name, void *data) 1619 { 1620 int error; 1621 1622 error = coredump_write(cookie, UIO_SYSSPACE, nhdr, sizeof(*nhdr), 0); 1623 if (error) 1624 return error; 1625 1626 error = coredump_write(cookie, UIO_SYSSPACE, name, 1627 elfround(nhdr->namesz), 0); 1628 if (error) 1629 return error; 1630 1631 return coredump_write(cookie, UIO_SYSSPACE, data, nhdr->descsz, 0); 1632 } 1633 #endif /* !SMALL_KERNEL */ 1634