1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format 2 * 3 * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved. 4 * Written by David Howells (dhowells@redhat.com) 5 * Derived from binfmt_elf.c 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License 9 * as published by the Free Software Foundation; either version 10 * 2 of the License, or (at your option) any later version. 11 */ 12 13 #include <linux/module.h> 14 15 #include <linux/fs.h> 16 #include <linux/stat.h> 17 #include <linux/sched.h> 18 #include <linux/mm.h> 19 #include <linux/mman.h> 20 #include <linux/errno.h> 21 #include <linux/signal.h> 22 #include <linux/binfmts.h> 23 #include <linux/string.h> 24 #include <linux/file.h> 25 #include <linux/fcntl.h> 26 #include <linux/slab.h> 27 #include <linux/pagemap.h> 28 #include <linux/security.h> 29 #include <linux/highmem.h> 30 #include <linux/highuid.h> 31 #include <linux/personality.h> 32 #include <linux/ptrace.h> 33 #include <linux/init.h> 34 #include <linux/elf.h> 35 #include <linux/elf-fdpic.h> 36 #include <linux/elfcore.h> 37 #include <linux/coredump.h> 38 39 #include <asm/uaccess.h> 40 #include <asm/param.h> 41 #include <asm/pgalloc.h> 42 43 typedef char *elf_caddr_t; 44 45 #if 0 46 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ ) 47 #else 48 #define kdebug(fmt, ...) do {} while(0) 49 #endif 50 51 #if 0 52 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ ) 53 #else 54 #define kdcore(fmt, ...) do {} while(0) 55 #endif 56 57 MODULE_LICENSE("GPL"); 58 59 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *); 60 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *); 61 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *, 62 struct mm_struct *, const char *); 63 64 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *, 65 struct elf_fdpic_params *, 66 struct elf_fdpic_params *); 67 68 #ifndef CONFIG_MMU 69 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *, 70 unsigned long *); 71 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *, 72 struct file *, 73 struct mm_struct *); 74 #endif 75 76 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *, 77 struct file *, struct mm_struct *); 78 79 #ifdef CONFIG_ELF_CORE 80 static int elf_fdpic_core_dump(struct coredump_params *cprm); 81 #endif 82 83 static struct linux_binfmt elf_fdpic_format = { 84 .module = THIS_MODULE, 85 .load_binary = load_elf_fdpic_binary, 86 #ifdef CONFIG_ELF_CORE 87 .core_dump = elf_fdpic_core_dump, 88 #endif 89 .min_coredump = ELF_EXEC_PAGESIZE, 90 }; 91 92 static int __init init_elf_fdpic_binfmt(void) 93 { 94 register_binfmt(&elf_fdpic_format); 95 return 0; 96 } 97 98 static void __exit exit_elf_fdpic_binfmt(void) 99 { 100 unregister_binfmt(&elf_fdpic_format); 101 } 102 103 core_initcall(init_elf_fdpic_binfmt); 104 module_exit(exit_elf_fdpic_binfmt); 105 106 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file) 107 { 108 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0) 109 return 0; 110 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN) 111 return 0; 112 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr)) 113 return 0; 114 if (!file->f_op || !file->f_op->mmap) 115 return 0; 116 return 1; 117 } 118 119 /*****************************************************************************/ 120 /* 121 * read the program headers table into memory 122 */ 123 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params, 124 struct file *file) 125 { 126 struct elf32_phdr *phdr; 127 unsigned long size; 128 int retval, loop; 129 130 if (params->hdr.e_phentsize != sizeof(struct elf_phdr)) 131 return -ENOMEM; 132 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr)) 133 return -ENOMEM; 134 135 size = params->hdr.e_phnum * sizeof(struct elf_phdr); 136 params->phdrs = kmalloc(size, GFP_KERNEL); 137 if (!params->phdrs) 138 return -ENOMEM; 139 140 retval = kernel_read(file, params->hdr.e_phoff, 141 (char *) params->phdrs, size); 142 if (unlikely(retval != size)) 143 return retval < 0 ? retval : -ENOEXEC; 144 145 /* determine stack size for this binary */ 146 phdr = params->phdrs; 147 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 148 if (phdr->p_type != PT_GNU_STACK) 149 continue; 150 151 if (phdr->p_flags & PF_X) 152 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK; 153 else 154 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK; 155 156 params->stack_size = phdr->p_memsz; 157 break; 158 } 159 160 return 0; 161 } 162 163 /*****************************************************************************/ 164 /* 165 * load an fdpic binary into various bits of memory 166 */ 167 static int load_elf_fdpic_binary(struct linux_binprm *bprm, 168 struct pt_regs *regs) 169 { 170 struct elf_fdpic_params exec_params, interp_params; 171 struct elf_phdr *phdr; 172 unsigned long stack_size, entryaddr; 173 #ifdef ELF_FDPIC_PLAT_INIT 174 unsigned long dynaddr; 175 #endif 176 #ifndef CONFIG_MMU 177 unsigned long stack_prot; 178 #endif 179 struct file *interpreter = NULL; /* to shut gcc up */ 180 char *interpreter_name = NULL; 181 int executable_stack; 182 int retval, i; 183 184 kdebug("____ LOAD %d ____", current->pid); 185 186 memset(&exec_params, 0, sizeof(exec_params)); 187 memset(&interp_params, 0, sizeof(interp_params)); 188 189 exec_params.hdr = *(struct elfhdr *) bprm->buf; 190 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE; 191 192 /* check that this is a binary we know how to deal with */ 193 retval = -ENOEXEC; 194 if (!is_elf_fdpic(&exec_params.hdr, bprm->file)) 195 goto error; 196 197 /* read the program header table */ 198 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file); 199 if (retval < 0) 200 goto error; 201 202 /* scan for a program header that specifies an interpreter */ 203 phdr = exec_params.phdrs; 204 205 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) { 206 switch (phdr->p_type) { 207 case PT_INTERP: 208 retval = -ENOMEM; 209 if (phdr->p_filesz > PATH_MAX) 210 goto error; 211 retval = -ENOENT; 212 if (phdr->p_filesz < 2) 213 goto error; 214 215 /* read the name of the interpreter into memory */ 216 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL); 217 if (!interpreter_name) 218 goto error; 219 220 retval = kernel_read(bprm->file, 221 phdr->p_offset, 222 interpreter_name, 223 phdr->p_filesz); 224 if (unlikely(retval != phdr->p_filesz)) { 225 if (retval >= 0) 226 retval = -ENOEXEC; 227 goto error; 228 } 229 230 retval = -ENOENT; 231 if (interpreter_name[phdr->p_filesz - 1] != '\0') 232 goto error; 233 234 kdebug("Using ELF interpreter %s", interpreter_name); 235 236 /* replace the program with the interpreter */ 237 interpreter = open_exec(interpreter_name); 238 retval = PTR_ERR(interpreter); 239 if (IS_ERR(interpreter)) { 240 interpreter = NULL; 241 goto error; 242 } 243 244 /* 245 * If the binary is not readable then enforce 246 * mm->dumpable = 0 regardless of the interpreter's 247 * permissions. 248 */ 249 would_dump(bprm, interpreter); 250 251 retval = kernel_read(interpreter, 0, bprm->buf, 252 BINPRM_BUF_SIZE); 253 if (unlikely(retval != BINPRM_BUF_SIZE)) { 254 if (retval >= 0) 255 retval = -ENOEXEC; 256 goto error; 257 } 258 259 interp_params.hdr = *((struct elfhdr *) bprm->buf); 260 break; 261 262 case PT_LOAD: 263 #ifdef CONFIG_MMU 264 if (exec_params.load_addr == 0) 265 exec_params.load_addr = phdr->p_vaddr; 266 #endif 267 break; 268 } 269 270 } 271 272 if (elf_check_const_displacement(&exec_params.hdr)) 273 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 274 275 /* perform insanity checks on the interpreter */ 276 if (interpreter_name) { 277 retval = -ELIBBAD; 278 if (!is_elf_fdpic(&interp_params.hdr, interpreter)) 279 goto error; 280 281 interp_params.flags = ELF_FDPIC_FLAG_PRESENT; 282 283 /* read the interpreter's program header table */ 284 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter); 285 if (retval < 0) 286 goto error; 287 } 288 289 stack_size = exec_params.stack_size; 290 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 291 executable_stack = EXSTACK_ENABLE_X; 292 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 293 executable_stack = EXSTACK_DISABLE_X; 294 else 295 executable_stack = EXSTACK_DEFAULT; 296 297 if (stack_size == 0) { 298 stack_size = interp_params.stack_size; 299 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 300 executable_stack = EXSTACK_ENABLE_X; 301 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 302 executable_stack = EXSTACK_DISABLE_X; 303 else 304 executable_stack = EXSTACK_DEFAULT; 305 } 306 307 retval = -ENOEXEC; 308 if (stack_size == 0) 309 goto error; 310 311 if (elf_check_const_displacement(&interp_params.hdr)) 312 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 313 314 /* flush all traces of the currently running executable */ 315 retval = flush_old_exec(bprm); 316 if (retval) 317 goto error; 318 319 /* there's now no turning back... the old userspace image is dead, 320 * defunct, deceased, etc. after this point we have to exit via 321 * error_kill */ 322 set_personality(PER_LINUX_FDPIC); 323 if (elf_read_implies_exec(&exec_params.hdr, executable_stack)) 324 current->personality |= READ_IMPLIES_EXEC; 325 326 setup_new_exec(bprm); 327 328 set_binfmt(&elf_fdpic_format); 329 330 current->mm->start_code = 0; 331 current->mm->end_code = 0; 332 current->mm->start_stack = 0; 333 current->mm->start_data = 0; 334 current->mm->end_data = 0; 335 current->mm->context.exec_fdpic_loadmap = 0; 336 current->mm->context.interp_fdpic_loadmap = 0; 337 338 #ifdef CONFIG_MMU 339 elf_fdpic_arch_lay_out_mm(&exec_params, 340 &interp_params, 341 ¤t->mm->start_stack, 342 ¤t->mm->start_brk); 343 344 retval = setup_arg_pages(bprm, current->mm->start_stack, 345 executable_stack); 346 if (retval < 0) { 347 send_sig(SIGKILL, current, 0); 348 goto error_kill; 349 } 350 #endif 351 352 /* load the executable and interpreter into memory */ 353 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm, 354 "executable"); 355 if (retval < 0) 356 goto error_kill; 357 358 if (interpreter_name) { 359 retval = elf_fdpic_map_file(&interp_params, interpreter, 360 current->mm, "interpreter"); 361 if (retval < 0) { 362 printk(KERN_ERR "Unable to load interpreter\n"); 363 goto error_kill; 364 } 365 366 allow_write_access(interpreter); 367 fput(interpreter); 368 interpreter = NULL; 369 } 370 371 #ifdef CONFIG_MMU 372 if (!current->mm->start_brk) 373 current->mm->start_brk = current->mm->end_data; 374 375 current->mm->brk = current->mm->start_brk = 376 PAGE_ALIGN(current->mm->start_brk); 377 378 #else 379 /* create a stack and brk area big enough for everyone 380 * - the brk heap starts at the bottom and works up 381 * - the stack starts at the top and works down 382 */ 383 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK; 384 if (stack_size < PAGE_SIZE * 2) 385 stack_size = PAGE_SIZE * 2; 386 387 stack_prot = PROT_READ | PROT_WRITE; 388 if (executable_stack == EXSTACK_ENABLE_X || 389 (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC)) 390 stack_prot |= PROT_EXEC; 391 392 down_write(¤t->mm->mmap_sem); 393 current->mm->start_brk = do_mmap(NULL, 0, stack_size, stack_prot, 394 MAP_PRIVATE | MAP_ANONYMOUS | 395 MAP_UNINITIALIZED | MAP_GROWSDOWN, 396 0); 397 398 if (IS_ERR_VALUE(current->mm->start_brk)) { 399 up_write(¤t->mm->mmap_sem); 400 retval = current->mm->start_brk; 401 current->mm->start_brk = 0; 402 goto error_kill; 403 } 404 405 up_write(¤t->mm->mmap_sem); 406 407 current->mm->brk = current->mm->start_brk; 408 current->mm->context.end_brk = current->mm->start_brk; 409 current->mm->context.end_brk += 410 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0; 411 current->mm->start_stack = current->mm->start_brk + stack_size; 412 #endif 413 414 install_exec_creds(bprm); 415 if (create_elf_fdpic_tables(bprm, current->mm, 416 &exec_params, &interp_params) < 0) 417 goto error_kill; 418 419 kdebug("- start_code %lx", current->mm->start_code); 420 kdebug("- end_code %lx", current->mm->end_code); 421 kdebug("- start_data %lx", current->mm->start_data); 422 kdebug("- end_data %lx", current->mm->end_data); 423 kdebug("- start_brk %lx", current->mm->start_brk); 424 kdebug("- brk %lx", current->mm->brk); 425 kdebug("- start_stack %lx", current->mm->start_stack); 426 427 #ifdef ELF_FDPIC_PLAT_INIT 428 /* 429 * The ABI may specify that certain registers be set up in special 430 * ways (on i386 %edx is the address of a DT_FINI function, for 431 * example. This macro performs whatever initialization to 432 * the regs structure is required. 433 */ 434 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr; 435 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr, 436 dynaddr); 437 #endif 438 439 /* everything is now ready... get the userspace context ready to roll */ 440 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr; 441 start_thread(regs, entryaddr, current->mm->start_stack); 442 443 retval = 0; 444 445 error: 446 if (interpreter) { 447 allow_write_access(interpreter); 448 fput(interpreter); 449 } 450 kfree(interpreter_name); 451 kfree(exec_params.phdrs); 452 kfree(exec_params.loadmap); 453 kfree(interp_params.phdrs); 454 kfree(interp_params.loadmap); 455 return retval; 456 457 /* unrecoverable error - kill the process */ 458 error_kill: 459 send_sig(SIGSEGV, current, 0); 460 goto error; 461 462 } 463 464 /*****************************************************************************/ 465 466 #ifndef ELF_BASE_PLATFORM 467 /* 468 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture. 469 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value 470 * will be copied to the user stack in the same manner as AT_PLATFORM. 471 */ 472 #define ELF_BASE_PLATFORM NULL 473 #endif 474 475 /* 476 * present useful information to the program by shovelling it onto the new 477 * process's stack 478 */ 479 static int create_elf_fdpic_tables(struct linux_binprm *bprm, 480 struct mm_struct *mm, 481 struct elf_fdpic_params *exec_params, 482 struct elf_fdpic_params *interp_params) 483 { 484 const struct cred *cred = current_cred(); 485 unsigned long sp, csp, nitems; 486 elf_caddr_t __user *argv, *envp; 487 size_t platform_len = 0, len; 488 char *k_platform, *k_base_platform; 489 char __user *u_platform, *u_base_platform, *p; 490 long hwcap; 491 int loop; 492 int nr; /* reset for each csp adjustment */ 493 494 #ifdef CONFIG_MMU 495 /* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions 496 * by the processes running on the same package. One thing we can do is 497 * to shuffle the initial stack for them, so we give the architecture 498 * an opportunity to do so here. 499 */ 500 sp = arch_align_stack(bprm->p); 501 #else 502 sp = mm->start_stack; 503 504 /* stack the program arguments and environment */ 505 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0) 506 return -EFAULT; 507 #endif 508 509 hwcap = ELF_HWCAP; 510 511 /* 512 * If this architecture has a platform capability string, copy it 513 * to userspace. In some cases (Sparc), this info is impossible 514 * for userspace to get any other way, in others (i386) it is 515 * merely difficult. 516 */ 517 k_platform = ELF_PLATFORM; 518 u_platform = NULL; 519 520 if (k_platform) { 521 platform_len = strlen(k_platform) + 1; 522 sp -= platform_len; 523 u_platform = (char __user *) sp; 524 if (__copy_to_user(u_platform, k_platform, platform_len) != 0) 525 return -EFAULT; 526 } 527 528 /* 529 * If this architecture has a "base" platform capability 530 * string, copy it to userspace. 531 */ 532 k_base_platform = ELF_BASE_PLATFORM; 533 u_base_platform = NULL; 534 535 if (k_base_platform) { 536 platform_len = strlen(k_base_platform) + 1; 537 sp -= platform_len; 538 u_base_platform = (char __user *) sp; 539 if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0) 540 return -EFAULT; 541 } 542 543 sp &= ~7UL; 544 545 /* stack the load map(s) */ 546 len = sizeof(struct elf32_fdpic_loadmap); 547 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs; 548 sp = (sp - len) & ~7UL; 549 exec_params->map_addr = sp; 550 551 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0) 552 return -EFAULT; 553 554 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp; 555 556 if (interp_params->loadmap) { 557 len = sizeof(struct elf32_fdpic_loadmap); 558 len += sizeof(struct elf32_fdpic_loadseg) * 559 interp_params->loadmap->nsegs; 560 sp = (sp - len) & ~7UL; 561 interp_params->map_addr = sp; 562 563 if (copy_to_user((void __user *) sp, interp_params->loadmap, 564 len) != 0) 565 return -EFAULT; 566 567 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp; 568 } 569 570 /* force 16 byte _final_ alignment here for generality */ 571 #define DLINFO_ITEMS 15 572 573 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) + 574 (k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH; 575 576 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) 577 nitems++; 578 579 csp = sp; 580 sp -= nitems * 2 * sizeof(unsigned long); 581 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */ 582 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */ 583 sp -= 1 * sizeof(unsigned long); /* argc */ 584 585 csp -= sp & 15UL; 586 sp -= sp & 15UL; 587 588 /* put the ELF interpreter info on the stack */ 589 #define NEW_AUX_ENT(id, val) \ 590 do { \ 591 struct { unsigned long _id, _val; } __user *ent; \ 592 \ 593 ent = (void __user *) csp; \ 594 __put_user((id), &ent[nr]._id); \ 595 __put_user((val), &ent[nr]._val); \ 596 nr++; \ 597 } while (0) 598 599 nr = 0; 600 csp -= 2 * sizeof(unsigned long); 601 NEW_AUX_ENT(AT_NULL, 0); 602 if (k_platform) { 603 nr = 0; 604 csp -= 2 * sizeof(unsigned long); 605 NEW_AUX_ENT(AT_PLATFORM, 606 (elf_addr_t) (unsigned long) u_platform); 607 } 608 609 if (k_base_platform) { 610 nr = 0; 611 csp -= 2 * sizeof(unsigned long); 612 NEW_AUX_ENT(AT_BASE_PLATFORM, 613 (elf_addr_t) (unsigned long) u_base_platform); 614 } 615 616 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) { 617 nr = 0; 618 csp -= 2 * sizeof(unsigned long); 619 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data); 620 } 621 622 nr = 0; 623 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long); 624 NEW_AUX_ENT(AT_HWCAP, hwcap); 625 NEW_AUX_ENT(AT_PAGESZ, PAGE_SIZE); 626 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC); 627 NEW_AUX_ENT(AT_PHDR, exec_params->ph_addr); 628 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr)); 629 NEW_AUX_ENT(AT_PHNUM, exec_params->hdr.e_phnum); 630 NEW_AUX_ENT(AT_BASE, interp_params->elfhdr_addr); 631 NEW_AUX_ENT(AT_FLAGS, 0); 632 NEW_AUX_ENT(AT_ENTRY, exec_params->entry_addr); 633 NEW_AUX_ENT(AT_UID, (elf_addr_t) cred->uid); 634 NEW_AUX_ENT(AT_EUID, (elf_addr_t) cred->euid); 635 NEW_AUX_ENT(AT_GID, (elf_addr_t) cred->gid); 636 NEW_AUX_ENT(AT_EGID, (elf_addr_t) cred->egid); 637 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm)); 638 NEW_AUX_ENT(AT_EXECFN, bprm->exec); 639 640 #ifdef ARCH_DLINFO 641 nr = 0; 642 csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long); 643 644 /* ARCH_DLINFO must come last so platform specific code can enforce 645 * special alignment requirements on the AUXV if necessary (eg. PPC). 646 */ 647 ARCH_DLINFO; 648 #endif 649 #undef NEW_AUX_ENT 650 651 /* allocate room for argv[] and envv[] */ 652 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t); 653 envp = (elf_caddr_t __user *) csp; 654 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t); 655 argv = (elf_caddr_t __user *) csp; 656 657 /* stack argc */ 658 csp -= sizeof(unsigned long); 659 __put_user(bprm->argc, (unsigned long __user *) csp); 660 661 BUG_ON(csp != sp); 662 663 /* fill in the argv[] array */ 664 #ifdef CONFIG_MMU 665 current->mm->arg_start = bprm->p; 666 #else 667 current->mm->arg_start = current->mm->start_stack - 668 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p); 669 #endif 670 671 p = (char __user *) current->mm->arg_start; 672 for (loop = bprm->argc; loop > 0; loop--) { 673 __put_user((elf_caddr_t) p, argv++); 674 len = strnlen_user(p, MAX_ARG_STRLEN); 675 if (!len || len > MAX_ARG_STRLEN) 676 return -EINVAL; 677 p += len; 678 } 679 __put_user(NULL, argv); 680 current->mm->arg_end = (unsigned long) p; 681 682 /* fill in the envv[] array */ 683 current->mm->env_start = (unsigned long) p; 684 for (loop = bprm->envc; loop > 0; loop--) { 685 __put_user((elf_caddr_t)(unsigned long) p, envp++); 686 len = strnlen_user(p, MAX_ARG_STRLEN); 687 if (!len || len > MAX_ARG_STRLEN) 688 return -EINVAL; 689 p += len; 690 } 691 __put_user(NULL, envp); 692 current->mm->env_end = (unsigned long) p; 693 694 mm->start_stack = (unsigned long) sp; 695 return 0; 696 } 697 698 /*****************************************************************************/ 699 /* 700 * transfer the program arguments and environment from the holding pages onto 701 * the stack 702 */ 703 #ifndef CONFIG_MMU 704 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm, 705 unsigned long *_sp) 706 { 707 unsigned long index, stop, sp; 708 char *src; 709 int ret = 0; 710 711 stop = bprm->p >> PAGE_SHIFT; 712 sp = *_sp; 713 714 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) { 715 src = kmap(bprm->page[index]); 716 sp -= PAGE_SIZE; 717 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0) 718 ret = -EFAULT; 719 kunmap(bprm->page[index]); 720 if (ret < 0) 721 goto out; 722 } 723 724 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15; 725 726 out: 727 return ret; 728 } 729 #endif 730 731 /*****************************************************************************/ 732 /* 733 * load the appropriate binary image (executable or interpreter) into memory 734 * - we assume no MMU is available 735 * - if no other PIC bits are set in params->hdr->e_flags 736 * - we assume that the LOADable segments in the binary are independently relocatable 737 * - we assume R/O executable segments are shareable 738 * - else 739 * - we assume the loadable parts of the image to require fixed displacement 740 * - the image is not shareable 741 */ 742 static int elf_fdpic_map_file(struct elf_fdpic_params *params, 743 struct file *file, 744 struct mm_struct *mm, 745 const char *what) 746 { 747 struct elf32_fdpic_loadmap *loadmap; 748 #ifdef CONFIG_MMU 749 struct elf32_fdpic_loadseg *mseg; 750 #endif 751 struct elf32_fdpic_loadseg *seg; 752 struct elf32_phdr *phdr; 753 unsigned long load_addr, stop; 754 unsigned nloads, tmp; 755 size_t size; 756 int loop, ret; 757 758 /* allocate a load map table */ 759 nloads = 0; 760 for (loop = 0; loop < params->hdr.e_phnum; loop++) 761 if (params->phdrs[loop].p_type == PT_LOAD) 762 nloads++; 763 764 if (nloads == 0) 765 return -ELIBBAD; 766 767 size = sizeof(*loadmap) + nloads * sizeof(*seg); 768 loadmap = kzalloc(size, GFP_KERNEL); 769 if (!loadmap) 770 return -ENOMEM; 771 772 params->loadmap = loadmap; 773 774 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION; 775 loadmap->nsegs = nloads; 776 777 load_addr = params->load_addr; 778 seg = loadmap->segs; 779 780 /* map the requested LOADs into the memory space */ 781 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 782 case ELF_FDPIC_FLAG_CONSTDISP: 783 case ELF_FDPIC_FLAG_CONTIGUOUS: 784 #ifndef CONFIG_MMU 785 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm); 786 if (ret < 0) 787 return ret; 788 break; 789 #endif 790 default: 791 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm); 792 if (ret < 0) 793 return ret; 794 break; 795 } 796 797 /* map the entry point */ 798 if (params->hdr.e_entry) { 799 seg = loadmap->segs; 800 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 801 if (params->hdr.e_entry >= seg->p_vaddr && 802 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) { 803 params->entry_addr = 804 (params->hdr.e_entry - seg->p_vaddr) + 805 seg->addr; 806 break; 807 } 808 } 809 } 810 811 /* determine where the program header table has wound up if mapped */ 812 stop = params->hdr.e_phoff; 813 stop += params->hdr.e_phnum * sizeof (struct elf_phdr); 814 phdr = params->phdrs; 815 816 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 817 if (phdr->p_type != PT_LOAD) 818 continue; 819 820 if (phdr->p_offset > params->hdr.e_phoff || 821 phdr->p_offset + phdr->p_filesz < stop) 822 continue; 823 824 seg = loadmap->segs; 825 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 826 if (phdr->p_vaddr >= seg->p_vaddr && 827 phdr->p_vaddr + phdr->p_filesz <= 828 seg->p_vaddr + seg->p_memsz) { 829 params->ph_addr = 830 (phdr->p_vaddr - seg->p_vaddr) + 831 seg->addr + 832 params->hdr.e_phoff - phdr->p_offset; 833 break; 834 } 835 } 836 break; 837 } 838 839 /* determine where the dynamic section has wound up if there is one */ 840 phdr = params->phdrs; 841 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 842 if (phdr->p_type != PT_DYNAMIC) 843 continue; 844 845 seg = loadmap->segs; 846 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 847 if (phdr->p_vaddr >= seg->p_vaddr && 848 phdr->p_vaddr + phdr->p_memsz <= 849 seg->p_vaddr + seg->p_memsz) { 850 params->dynamic_addr = 851 (phdr->p_vaddr - seg->p_vaddr) + 852 seg->addr; 853 854 /* check the dynamic section contains at least 855 * one item, and that the last item is a NULL 856 * entry */ 857 if (phdr->p_memsz == 0 || 858 phdr->p_memsz % sizeof(Elf32_Dyn) != 0) 859 goto dynamic_error; 860 861 tmp = phdr->p_memsz / sizeof(Elf32_Dyn); 862 if (((Elf32_Dyn *) 863 params->dynamic_addr)[tmp - 1].d_tag != 0) 864 goto dynamic_error; 865 break; 866 } 867 } 868 break; 869 } 870 871 /* now elide adjacent segments in the load map on MMU linux 872 * - on uClinux the holes between may actually be filled with system 873 * stuff or stuff from other processes 874 */ 875 #ifdef CONFIG_MMU 876 nloads = loadmap->nsegs; 877 mseg = loadmap->segs; 878 seg = mseg + 1; 879 for (loop = 1; loop < nloads; loop++) { 880 /* see if we have a candidate for merging */ 881 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) { 882 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz); 883 if (load_addr == (seg->addr & PAGE_MASK)) { 884 mseg->p_memsz += 885 load_addr - 886 (mseg->addr + mseg->p_memsz); 887 mseg->p_memsz += seg->addr & ~PAGE_MASK; 888 mseg->p_memsz += seg->p_memsz; 889 loadmap->nsegs--; 890 continue; 891 } 892 } 893 894 mseg++; 895 if (mseg != seg) 896 *mseg = *seg; 897 } 898 #endif 899 900 kdebug("Mapped Object [%s]:", what); 901 kdebug("- elfhdr : %lx", params->elfhdr_addr); 902 kdebug("- entry : %lx", params->entry_addr); 903 kdebug("- PHDR[] : %lx", params->ph_addr); 904 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr); 905 seg = loadmap->segs; 906 for (loop = 0; loop < loadmap->nsegs; loop++, seg++) 907 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]", 908 loop, 909 seg->addr, seg->addr + seg->p_memsz - 1, 910 seg->p_vaddr, seg->p_memsz); 911 912 return 0; 913 914 dynamic_error: 915 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n", 916 what, file->f_path.dentry->d_inode->i_ino); 917 return -ELIBBAD; 918 } 919 920 /*****************************************************************************/ 921 /* 922 * map a file with constant displacement under uClinux 923 */ 924 #ifndef CONFIG_MMU 925 static int elf_fdpic_map_file_constdisp_on_uclinux( 926 struct elf_fdpic_params *params, 927 struct file *file, 928 struct mm_struct *mm) 929 { 930 struct elf32_fdpic_loadseg *seg; 931 struct elf32_phdr *phdr; 932 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags; 933 loff_t fpos; 934 int loop, ret; 935 936 load_addr = params->load_addr; 937 seg = params->loadmap->segs; 938 939 /* determine the bounds of the contiguous overall allocation we must 940 * make */ 941 phdr = params->phdrs; 942 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 943 if (params->phdrs[loop].p_type != PT_LOAD) 944 continue; 945 946 if (base > phdr->p_vaddr) 947 base = phdr->p_vaddr; 948 if (top < phdr->p_vaddr + phdr->p_memsz) 949 top = phdr->p_vaddr + phdr->p_memsz; 950 } 951 952 /* allocate one big anon block for everything */ 953 mflags = MAP_PRIVATE; 954 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 955 mflags |= MAP_EXECUTABLE; 956 957 down_write(&mm->mmap_sem); 958 maddr = do_mmap(NULL, load_addr, top - base, 959 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0); 960 up_write(&mm->mmap_sem); 961 if (IS_ERR_VALUE(maddr)) 962 return (int) maddr; 963 964 if (load_addr != 0) 965 load_addr += PAGE_ALIGN(top - base); 966 967 /* and then load the file segments into it */ 968 phdr = params->phdrs; 969 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 970 if (params->phdrs[loop].p_type != PT_LOAD) 971 continue; 972 973 fpos = phdr->p_offset; 974 975 seg->addr = maddr + (phdr->p_vaddr - base); 976 seg->p_vaddr = phdr->p_vaddr; 977 seg->p_memsz = phdr->p_memsz; 978 979 ret = file->f_op->read(file, (void *) seg->addr, 980 phdr->p_filesz, &fpos); 981 if (ret < 0) 982 return ret; 983 984 /* map the ELF header address if in this segment */ 985 if (phdr->p_offset == 0) 986 params->elfhdr_addr = seg->addr; 987 988 /* clear any space allocated but not loaded */ 989 if (phdr->p_filesz < phdr->p_memsz) { 990 if (clear_user((void *) (seg->addr + phdr->p_filesz), 991 phdr->p_memsz - phdr->p_filesz)) 992 return -EFAULT; 993 } 994 995 if (mm) { 996 if (phdr->p_flags & PF_X) { 997 if (!mm->start_code) { 998 mm->start_code = seg->addr; 999 mm->end_code = seg->addr + 1000 phdr->p_memsz; 1001 } 1002 } else if (!mm->start_data) { 1003 mm->start_data = seg->addr; 1004 mm->end_data = seg->addr + phdr->p_memsz; 1005 } 1006 } 1007 1008 seg++; 1009 } 1010 1011 return 0; 1012 } 1013 #endif 1014 1015 /*****************************************************************************/ 1016 /* 1017 * map a binary by direct mmap() of the individual PT_LOAD segments 1018 */ 1019 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params, 1020 struct file *file, 1021 struct mm_struct *mm) 1022 { 1023 struct elf32_fdpic_loadseg *seg; 1024 struct elf32_phdr *phdr; 1025 unsigned long load_addr, delta_vaddr; 1026 int loop, dvset; 1027 1028 load_addr = params->load_addr; 1029 delta_vaddr = 0; 1030 dvset = 0; 1031 1032 seg = params->loadmap->segs; 1033 1034 /* deal with each load segment separately */ 1035 phdr = params->phdrs; 1036 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 1037 unsigned long maddr, disp, excess, excess1; 1038 int prot = 0, flags; 1039 1040 if (phdr->p_type != PT_LOAD) 1041 continue; 1042 1043 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx", 1044 (unsigned long) phdr->p_vaddr, 1045 (unsigned long) phdr->p_offset, 1046 (unsigned long) phdr->p_filesz, 1047 (unsigned long) phdr->p_memsz); 1048 1049 /* determine the mapping parameters */ 1050 if (phdr->p_flags & PF_R) prot |= PROT_READ; 1051 if (phdr->p_flags & PF_W) prot |= PROT_WRITE; 1052 if (phdr->p_flags & PF_X) prot |= PROT_EXEC; 1053 1054 flags = MAP_PRIVATE | MAP_DENYWRITE; 1055 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 1056 flags |= MAP_EXECUTABLE; 1057 1058 maddr = 0; 1059 1060 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 1061 case ELF_FDPIC_FLAG_INDEPENDENT: 1062 /* PT_LOADs are independently locatable */ 1063 break; 1064 1065 case ELF_FDPIC_FLAG_HONOURVADDR: 1066 /* the specified virtual address must be honoured */ 1067 maddr = phdr->p_vaddr; 1068 flags |= MAP_FIXED; 1069 break; 1070 1071 case ELF_FDPIC_FLAG_CONSTDISP: 1072 /* constant displacement 1073 * - can be mapped anywhere, but must be mapped as a 1074 * unit 1075 */ 1076 if (!dvset) { 1077 maddr = load_addr; 1078 delta_vaddr = phdr->p_vaddr; 1079 dvset = 1; 1080 } else { 1081 maddr = load_addr + phdr->p_vaddr - delta_vaddr; 1082 flags |= MAP_FIXED; 1083 } 1084 break; 1085 1086 case ELF_FDPIC_FLAG_CONTIGUOUS: 1087 /* contiguity handled later */ 1088 break; 1089 1090 default: 1091 BUG(); 1092 } 1093 1094 maddr &= PAGE_MASK; 1095 1096 /* create the mapping */ 1097 disp = phdr->p_vaddr & ~PAGE_MASK; 1098 down_write(&mm->mmap_sem); 1099 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags, 1100 phdr->p_offset - disp); 1101 up_write(&mm->mmap_sem); 1102 1103 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx", 1104 loop, phdr->p_memsz + disp, prot, flags, 1105 phdr->p_offset - disp, maddr); 1106 1107 if (IS_ERR_VALUE(maddr)) 1108 return (int) maddr; 1109 1110 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) == 1111 ELF_FDPIC_FLAG_CONTIGUOUS) 1112 load_addr += PAGE_ALIGN(phdr->p_memsz + disp); 1113 1114 seg->addr = maddr + disp; 1115 seg->p_vaddr = phdr->p_vaddr; 1116 seg->p_memsz = phdr->p_memsz; 1117 1118 /* map the ELF header address if in this segment */ 1119 if (phdr->p_offset == 0) 1120 params->elfhdr_addr = seg->addr; 1121 1122 /* clear the bit between beginning of mapping and beginning of 1123 * PT_LOAD */ 1124 if (prot & PROT_WRITE && disp > 0) { 1125 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp); 1126 if (clear_user((void __user *) maddr, disp)) 1127 return -EFAULT; 1128 maddr += disp; 1129 } 1130 1131 /* clear any space allocated but not loaded 1132 * - on uClinux we can just clear the lot 1133 * - on MMU linux we'll get a SIGBUS beyond the last page 1134 * extant in the file 1135 */ 1136 excess = phdr->p_memsz - phdr->p_filesz; 1137 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK); 1138 1139 #ifdef CONFIG_MMU 1140 if (excess > excess1) { 1141 unsigned long xaddr = maddr + phdr->p_filesz + excess1; 1142 unsigned long xmaddr; 1143 1144 flags |= MAP_FIXED | MAP_ANONYMOUS; 1145 down_write(&mm->mmap_sem); 1146 xmaddr = do_mmap(NULL, xaddr, excess - excess1, 1147 prot, flags, 0); 1148 up_write(&mm->mmap_sem); 1149 1150 kdebug("mmap[%d] <anon>" 1151 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx", 1152 loop, xaddr, excess - excess1, prot, flags, 1153 xmaddr); 1154 1155 if (xmaddr != xaddr) 1156 return -ENOMEM; 1157 } 1158 1159 if (prot & PROT_WRITE && excess1 > 0) { 1160 kdebug("clear[%d] ad=%lx sz=%lx", 1161 loop, maddr + phdr->p_filesz, excess1); 1162 if (clear_user((void __user *) maddr + phdr->p_filesz, 1163 excess1)) 1164 return -EFAULT; 1165 } 1166 1167 #else 1168 if (excess > 0) { 1169 kdebug("clear[%d] ad=%lx sz=%lx", 1170 loop, maddr + phdr->p_filesz, excess); 1171 if (clear_user((void *) maddr + phdr->p_filesz, excess)) 1172 return -EFAULT; 1173 } 1174 #endif 1175 1176 if (mm) { 1177 if (phdr->p_flags & PF_X) { 1178 if (!mm->start_code) { 1179 mm->start_code = maddr; 1180 mm->end_code = maddr + phdr->p_memsz; 1181 } 1182 } else if (!mm->start_data) { 1183 mm->start_data = maddr; 1184 mm->end_data = maddr + phdr->p_memsz; 1185 } 1186 } 1187 1188 seg++; 1189 } 1190 1191 return 0; 1192 } 1193 1194 /*****************************************************************************/ 1195 /* 1196 * ELF-FDPIC core dumper 1197 * 1198 * Modelled on fs/exec.c:aout_core_dump() 1199 * Jeremy Fitzhardinge <jeremy@sw.oz.au> 1200 * 1201 * Modelled on fs/binfmt_elf.c core dumper 1202 */ 1203 #ifdef CONFIG_ELF_CORE 1204 1205 /* 1206 * Decide whether a segment is worth dumping; default is yes to be 1207 * sure (missing info is worse than too much; etc). 1208 * Personally I'd include everything, and use the coredump limit... 1209 * 1210 * I think we should skip something. But I am not sure how. H.J. 1211 */ 1212 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags) 1213 { 1214 int dump_ok; 1215 1216 /* Do not dump I/O mapped devices or special mappings */ 1217 if (vma->vm_flags & (VM_IO | VM_RESERVED)) { 1218 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags); 1219 return 0; 1220 } 1221 1222 /* If we may not read the contents, don't allow us to dump 1223 * them either. "dump_write()" can't handle it anyway. 1224 */ 1225 if (!(vma->vm_flags & VM_READ)) { 1226 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags); 1227 return 0; 1228 } 1229 1230 /* By default, dump shared memory if mapped from an anonymous file. */ 1231 if (vma->vm_flags & VM_SHARED) { 1232 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) { 1233 dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags); 1234 kdcore("%08lx: %08lx: %s (share)", vma->vm_start, 1235 vma->vm_flags, dump_ok ? "yes" : "no"); 1236 return dump_ok; 1237 } 1238 1239 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags); 1240 kdcore("%08lx: %08lx: %s (share)", vma->vm_start, 1241 vma->vm_flags, dump_ok ? "yes" : "no"); 1242 return dump_ok; 1243 } 1244 1245 #ifdef CONFIG_MMU 1246 /* By default, if it hasn't been written to, don't write it out */ 1247 if (!vma->anon_vma) { 1248 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags); 1249 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start, 1250 vma->vm_flags, dump_ok ? "yes" : "no"); 1251 return dump_ok; 1252 } 1253 #endif 1254 1255 dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags); 1256 kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags, 1257 dump_ok ? "yes" : "no"); 1258 return dump_ok; 1259 } 1260 1261 /* An ELF note in memory */ 1262 struct memelfnote 1263 { 1264 const char *name; 1265 int type; 1266 unsigned int datasz; 1267 void *data; 1268 }; 1269 1270 static int notesize(struct memelfnote *en) 1271 { 1272 int sz; 1273 1274 sz = sizeof(struct elf_note); 1275 sz += roundup(strlen(en->name) + 1, 4); 1276 sz += roundup(en->datasz, 4); 1277 1278 return sz; 1279 } 1280 1281 /* #define DEBUG */ 1282 1283 #define DUMP_WRITE(addr, nr, foffset) \ 1284 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0) 1285 1286 static int alignfile(struct file *file, loff_t *foffset) 1287 { 1288 static const char buf[4] = { 0, }; 1289 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset); 1290 return 1; 1291 } 1292 1293 static int writenote(struct memelfnote *men, struct file *file, 1294 loff_t *foffset) 1295 { 1296 struct elf_note en; 1297 en.n_namesz = strlen(men->name) + 1; 1298 en.n_descsz = men->datasz; 1299 en.n_type = men->type; 1300 1301 DUMP_WRITE(&en, sizeof(en), foffset); 1302 DUMP_WRITE(men->name, en.n_namesz, foffset); 1303 if (!alignfile(file, foffset)) 1304 return 0; 1305 DUMP_WRITE(men->data, men->datasz, foffset); 1306 if (!alignfile(file, foffset)) 1307 return 0; 1308 1309 return 1; 1310 } 1311 #undef DUMP_WRITE 1312 1313 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs) 1314 { 1315 memcpy(elf->e_ident, ELFMAG, SELFMAG); 1316 elf->e_ident[EI_CLASS] = ELF_CLASS; 1317 elf->e_ident[EI_DATA] = ELF_DATA; 1318 elf->e_ident[EI_VERSION] = EV_CURRENT; 1319 elf->e_ident[EI_OSABI] = ELF_OSABI; 1320 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); 1321 1322 elf->e_type = ET_CORE; 1323 elf->e_machine = ELF_ARCH; 1324 elf->e_version = EV_CURRENT; 1325 elf->e_entry = 0; 1326 elf->e_phoff = sizeof(struct elfhdr); 1327 elf->e_shoff = 0; 1328 elf->e_flags = ELF_FDPIC_CORE_EFLAGS; 1329 elf->e_ehsize = sizeof(struct elfhdr); 1330 elf->e_phentsize = sizeof(struct elf_phdr); 1331 elf->e_phnum = segs; 1332 elf->e_shentsize = 0; 1333 elf->e_shnum = 0; 1334 elf->e_shstrndx = 0; 1335 return; 1336 } 1337 1338 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset) 1339 { 1340 phdr->p_type = PT_NOTE; 1341 phdr->p_offset = offset; 1342 phdr->p_vaddr = 0; 1343 phdr->p_paddr = 0; 1344 phdr->p_filesz = sz; 1345 phdr->p_memsz = 0; 1346 phdr->p_flags = 0; 1347 phdr->p_align = 0; 1348 return; 1349 } 1350 1351 static inline void fill_note(struct memelfnote *note, const char *name, int type, 1352 unsigned int sz, void *data) 1353 { 1354 note->name = name; 1355 note->type = type; 1356 note->datasz = sz; 1357 note->data = data; 1358 return; 1359 } 1360 1361 /* 1362 * fill up all the fields in prstatus from the given task struct, except 1363 * registers which need to be filled up separately. 1364 */ 1365 static void fill_prstatus(struct elf_prstatus *prstatus, 1366 struct task_struct *p, long signr) 1367 { 1368 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 1369 prstatus->pr_sigpend = p->pending.signal.sig[0]; 1370 prstatus->pr_sighold = p->blocked.sig[0]; 1371 rcu_read_lock(); 1372 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent)); 1373 rcu_read_unlock(); 1374 prstatus->pr_pid = task_pid_vnr(p); 1375 prstatus->pr_pgrp = task_pgrp_vnr(p); 1376 prstatus->pr_sid = task_session_vnr(p); 1377 if (thread_group_leader(p)) { 1378 struct task_cputime cputime; 1379 1380 /* 1381 * This is the record for the group leader. It shows the 1382 * group-wide total, not its individual thread total. 1383 */ 1384 thread_group_cputime(p, &cputime); 1385 cputime_to_timeval(cputime.utime, &prstatus->pr_utime); 1386 cputime_to_timeval(cputime.stime, &prstatus->pr_stime); 1387 } else { 1388 cputime_to_timeval(p->utime, &prstatus->pr_utime); 1389 cputime_to_timeval(p->stime, &prstatus->pr_stime); 1390 } 1391 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime); 1392 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime); 1393 1394 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap; 1395 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap; 1396 } 1397 1398 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p, 1399 struct mm_struct *mm) 1400 { 1401 const struct cred *cred; 1402 unsigned int i, len; 1403 1404 /* first copy the parameters from user space */ 1405 memset(psinfo, 0, sizeof(struct elf_prpsinfo)); 1406 1407 len = mm->arg_end - mm->arg_start; 1408 if (len >= ELF_PRARGSZ) 1409 len = ELF_PRARGSZ - 1; 1410 if (copy_from_user(&psinfo->pr_psargs, 1411 (const char __user *) mm->arg_start, len)) 1412 return -EFAULT; 1413 for (i = 0; i < len; i++) 1414 if (psinfo->pr_psargs[i] == 0) 1415 psinfo->pr_psargs[i] = ' '; 1416 psinfo->pr_psargs[len] = 0; 1417 1418 rcu_read_lock(); 1419 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent)); 1420 rcu_read_unlock(); 1421 psinfo->pr_pid = task_pid_vnr(p); 1422 psinfo->pr_pgrp = task_pgrp_vnr(p); 1423 psinfo->pr_sid = task_session_vnr(p); 1424 1425 i = p->state ? ffz(~p->state) + 1 : 0; 1426 psinfo->pr_state = i; 1427 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i]; 1428 psinfo->pr_zomb = psinfo->pr_sname == 'Z'; 1429 psinfo->pr_nice = task_nice(p); 1430 psinfo->pr_flag = p->flags; 1431 rcu_read_lock(); 1432 cred = __task_cred(p); 1433 SET_UID(psinfo->pr_uid, cred->uid); 1434 SET_GID(psinfo->pr_gid, cred->gid); 1435 rcu_read_unlock(); 1436 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname)); 1437 1438 return 0; 1439 } 1440 1441 /* Here is the structure in which status of each thread is captured. */ 1442 struct elf_thread_status 1443 { 1444 struct list_head list; 1445 struct elf_prstatus prstatus; /* NT_PRSTATUS */ 1446 elf_fpregset_t fpu; /* NT_PRFPREG */ 1447 struct task_struct *thread; 1448 #ifdef ELF_CORE_COPY_XFPREGS 1449 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */ 1450 #endif 1451 struct memelfnote notes[3]; 1452 int num_notes; 1453 }; 1454 1455 /* 1456 * In order to add the specific thread information for the elf file format, 1457 * we need to keep a linked list of every thread's pr_status and then create 1458 * a single section for them in the final core file. 1459 */ 1460 static int elf_dump_thread_status(long signr, struct elf_thread_status *t) 1461 { 1462 struct task_struct *p = t->thread; 1463 int sz = 0; 1464 1465 t->num_notes = 0; 1466 1467 fill_prstatus(&t->prstatus, p, signr); 1468 elf_core_copy_task_regs(p, &t->prstatus.pr_reg); 1469 1470 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus), 1471 &t->prstatus); 1472 t->num_notes++; 1473 sz += notesize(&t->notes[0]); 1474 1475 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu); 1476 if (t->prstatus.pr_fpvalid) { 1477 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu), 1478 &t->fpu); 1479 t->num_notes++; 1480 sz += notesize(&t->notes[1]); 1481 } 1482 1483 #ifdef ELF_CORE_COPY_XFPREGS 1484 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) { 1485 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE, 1486 sizeof(t->xfpu), &t->xfpu); 1487 t->num_notes++; 1488 sz += notesize(&t->notes[2]); 1489 } 1490 #endif 1491 return sz; 1492 } 1493 1494 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum, 1495 elf_addr_t e_shoff, int segs) 1496 { 1497 elf->e_shoff = e_shoff; 1498 elf->e_shentsize = sizeof(*shdr4extnum); 1499 elf->e_shnum = 1; 1500 elf->e_shstrndx = SHN_UNDEF; 1501 1502 memset(shdr4extnum, 0, sizeof(*shdr4extnum)); 1503 1504 shdr4extnum->sh_type = SHT_NULL; 1505 shdr4extnum->sh_size = elf->e_shnum; 1506 shdr4extnum->sh_link = elf->e_shstrndx; 1507 shdr4extnum->sh_info = segs; 1508 } 1509 1510 /* 1511 * dump the segments for an MMU process 1512 */ 1513 #ifdef CONFIG_MMU 1514 static int elf_fdpic_dump_segments(struct file *file, size_t *size, 1515 unsigned long *limit, unsigned long mm_flags) 1516 { 1517 struct vm_area_struct *vma; 1518 int err = 0; 1519 1520 for (vma = current->mm->mmap; vma; vma = vma->vm_next) { 1521 unsigned long addr; 1522 1523 if (!maydump(vma, mm_flags)) 1524 continue; 1525 1526 for (addr = vma->vm_start; addr < vma->vm_end; 1527 addr += PAGE_SIZE) { 1528 struct page *page = get_dump_page(addr); 1529 if (page) { 1530 void *kaddr = kmap(page); 1531 *size += PAGE_SIZE; 1532 if (*size > *limit) 1533 err = -EFBIG; 1534 else if (!dump_write(file, kaddr, PAGE_SIZE)) 1535 err = -EIO; 1536 kunmap(page); 1537 page_cache_release(page); 1538 } else if (!dump_seek(file, PAGE_SIZE)) 1539 err = -EFBIG; 1540 if (err) 1541 goto out; 1542 } 1543 } 1544 out: 1545 return err; 1546 } 1547 #endif 1548 1549 /* 1550 * dump the segments for a NOMMU process 1551 */ 1552 #ifndef CONFIG_MMU 1553 static int elf_fdpic_dump_segments(struct file *file, size_t *size, 1554 unsigned long *limit, unsigned long mm_flags) 1555 { 1556 struct vm_area_struct *vma; 1557 1558 for (vma = current->mm->mmap; vma; vma = vma->vm_next) { 1559 if (!maydump(vma, mm_flags)) 1560 continue; 1561 1562 if ((*size += PAGE_SIZE) > *limit) 1563 return -EFBIG; 1564 1565 if (!dump_write(file, (void *) vma->vm_start, 1566 vma->vm_end - vma->vm_start)) 1567 return -EIO; 1568 } 1569 1570 return 0; 1571 } 1572 #endif 1573 1574 static size_t elf_core_vma_data_size(unsigned long mm_flags) 1575 { 1576 struct vm_area_struct *vma; 1577 size_t size = 0; 1578 1579 for (vma = current->mm->mmap; vma; vma = vma->vm_next) 1580 if (maydump(vma, mm_flags)) 1581 size += vma->vm_end - vma->vm_start; 1582 return size; 1583 } 1584 1585 /* 1586 * Actual dumper 1587 * 1588 * This is a two-pass process; first we find the offsets of the bits, 1589 * and then they are actually written out. If we run out of core limit 1590 * we just truncate. 1591 */ 1592 static int elf_fdpic_core_dump(struct coredump_params *cprm) 1593 { 1594 #define NUM_NOTES 6 1595 int has_dumped = 0; 1596 mm_segment_t fs; 1597 int segs; 1598 size_t size = 0; 1599 int i; 1600 struct vm_area_struct *vma; 1601 struct elfhdr *elf = NULL; 1602 loff_t offset = 0, dataoff, foffset; 1603 int numnote; 1604 struct memelfnote *notes = NULL; 1605 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */ 1606 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */ 1607 LIST_HEAD(thread_list); 1608 struct list_head *t; 1609 elf_fpregset_t *fpu = NULL; 1610 #ifdef ELF_CORE_COPY_XFPREGS 1611 elf_fpxregset_t *xfpu = NULL; 1612 #endif 1613 int thread_status_size = 0; 1614 elf_addr_t *auxv; 1615 struct elf_phdr *phdr4note = NULL; 1616 struct elf_shdr *shdr4extnum = NULL; 1617 Elf_Half e_phnum; 1618 elf_addr_t e_shoff; 1619 1620 /* 1621 * We no longer stop all VM operations. 1622 * 1623 * This is because those proceses that could possibly change map_count 1624 * or the mmap / vma pages are now blocked in do_exit on current 1625 * finishing this core dump. 1626 * 1627 * Only ptrace can touch these memory addresses, but it doesn't change 1628 * the map_count or the pages allocated. So no possibility of crashing 1629 * exists while dumping the mm->vm_next areas to the core file. 1630 */ 1631 1632 /* alloc memory for large data structures: too large to be on stack */ 1633 elf = kmalloc(sizeof(*elf), GFP_KERNEL); 1634 if (!elf) 1635 goto cleanup; 1636 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL); 1637 if (!prstatus) 1638 goto cleanup; 1639 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL); 1640 if (!psinfo) 1641 goto cleanup; 1642 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL); 1643 if (!notes) 1644 goto cleanup; 1645 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL); 1646 if (!fpu) 1647 goto cleanup; 1648 #ifdef ELF_CORE_COPY_XFPREGS 1649 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL); 1650 if (!xfpu) 1651 goto cleanup; 1652 #endif 1653 1654 if (cprm->signr) { 1655 struct core_thread *ct; 1656 struct elf_thread_status *tmp; 1657 1658 for (ct = current->mm->core_state->dumper.next; 1659 ct; ct = ct->next) { 1660 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL); 1661 if (!tmp) 1662 goto cleanup; 1663 1664 tmp->thread = ct->task; 1665 list_add(&tmp->list, &thread_list); 1666 } 1667 1668 list_for_each(t, &thread_list) { 1669 struct elf_thread_status *tmp; 1670 int sz; 1671 1672 tmp = list_entry(t, struct elf_thread_status, list); 1673 sz = elf_dump_thread_status(cprm->signr, tmp); 1674 thread_status_size += sz; 1675 } 1676 } 1677 1678 /* now collect the dump for the current */ 1679 fill_prstatus(prstatus, current, cprm->signr); 1680 elf_core_copy_regs(&prstatus->pr_reg, cprm->regs); 1681 1682 segs = current->mm->map_count; 1683 segs += elf_core_extra_phdrs(); 1684 1685 /* for notes section */ 1686 segs++; 1687 1688 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid 1689 * this, kernel supports extended numbering. Have a look at 1690 * include/linux/elf.h for further information. */ 1691 e_phnum = segs > PN_XNUM ? PN_XNUM : segs; 1692 1693 /* Set up header */ 1694 fill_elf_fdpic_header(elf, e_phnum); 1695 1696 has_dumped = 1; 1697 current->flags |= PF_DUMPCORE; 1698 1699 /* 1700 * Set up the notes in similar form to SVR4 core dumps made 1701 * with info from their /proc. 1702 */ 1703 1704 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus); 1705 fill_psinfo(psinfo, current->group_leader, current->mm); 1706 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo); 1707 1708 numnote = 2; 1709 1710 auxv = (elf_addr_t *) current->mm->saved_auxv; 1711 1712 i = 0; 1713 do 1714 i += 2; 1715 while (auxv[i - 2] != AT_NULL); 1716 fill_note(¬es[numnote++], "CORE", NT_AUXV, 1717 i * sizeof(elf_addr_t), auxv); 1718 1719 /* Try to dump the FPU. */ 1720 if ((prstatus->pr_fpvalid = 1721 elf_core_copy_task_fpregs(current, cprm->regs, fpu))) 1722 fill_note(notes + numnote++, 1723 "CORE", NT_PRFPREG, sizeof(*fpu), fpu); 1724 #ifdef ELF_CORE_COPY_XFPREGS 1725 if (elf_core_copy_task_xfpregs(current, xfpu)) 1726 fill_note(notes + numnote++, 1727 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu); 1728 #endif 1729 1730 fs = get_fs(); 1731 set_fs(KERNEL_DS); 1732 1733 offset += sizeof(*elf); /* Elf header */ 1734 offset += segs * sizeof(struct elf_phdr); /* Program headers */ 1735 foffset = offset; 1736 1737 /* Write notes phdr entry */ 1738 { 1739 int sz = 0; 1740 1741 for (i = 0; i < numnote; i++) 1742 sz += notesize(notes + i); 1743 1744 sz += thread_status_size; 1745 1746 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL); 1747 if (!phdr4note) 1748 goto end_coredump; 1749 1750 fill_elf_note_phdr(phdr4note, sz, offset); 1751 offset += sz; 1752 } 1753 1754 /* Page-align dumped data */ 1755 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE); 1756 1757 offset += elf_core_vma_data_size(cprm->mm_flags); 1758 offset += elf_core_extra_data_size(); 1759 e_shoff = offset; 1760 1761 if (e_phnum == PN_XNUM) { 1762 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL); 1763 if (!shdr4extnum) 1764 goto end_coredump; 1765 fill_extnum_info(elf, shdr4extnum, e_shoff, segs); 1766 } 1767 1768 offset = dataoff; 1769 1770 size += sizeof(*elf); 1771 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf))) 1772 goto end_coredump; 1773 1774 size += sizeof(*phdr4note); 1775 if (size > cprm->limit 1776 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note))) 1777 goto end_coredump; 1778 1779 /* write program headers for segments dump */ 1780 for (vma = current->mm->mmap; vma; vma = vma->vm_next) { 1781 struct elf_phdr phdr; 1782 size_t sz; 1783 1784 sz = vma->vm_end - vma->vm_start; 1785 1786 phdr.p_type = PT_LOAD; 1787 phdr.p_offset = offset; 1788 phdr.p_vaddr = vma->vm_start; 1789 phdr.p_paddr = 0; 1790 phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0; 1791 phdr.p_memsz = sz; 1792 offset += phdr.p_filesz; 1793 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0; 1794 if (vma->vm_flags & VM_WRITE) 1795 phdr.p_flags |= PF_W; 1796 if (vma->vm_flags & VM_EXEC) 1797 phdr.p_flags |= PF_X; 1798 phdr.p_align = ELF_EXEC_PAGESIZE; 1799 1800 size += sizeof(phdr); 1801 if (size > cprm->limit 1802 || !dump_write(cprm->file, &phdr, sizeof(phdr))) 1803 goto end_coredump; 1804 } 1805 1806 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit)) 1807 goto end_coredump; 1808 1809 /* write out the notes section */ 1810 for (i = 0; i < numnote; i++) 1811 if (!writenote(notes + i, cprm->file, &foffset)) 1812 goto end_coredump; 1813 1814 /* write out the thread status notes section */ 1815 list_for_each(t, &thread_list) { 1816 struct elf_thread_status *tmp = 1817 list_entry(t, struct elf_thread_status, list); 1818 1819 for (i = 0; i < tmp->num_notes; i++) 1820 if (!writenote(&tmp->notes[i], cprm->file, &foffset)) 1821 goto end_coredump; 1822 } 1823 1824 if (!dump_seek(cprm->file, dataoff - foffset)) 1825 goto end_coredump; 1826 1827 if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit, 1828 cprm->mm_flags) < 0) 1829 goto end_coredump; 1830 1831 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit)) 1832 goto end_coredump; 1833 1834 if (e_phnum == PN_XNUM) { 1835 size += sizeof(*shdr4extnum); 1836 if (size > cprm->limit 1837 || !dump_write(cprm->file, shdr4extnum, 1838 sizeof(*shdr4extnum))) 1839 goto end_coredump; 1840 } 1841 1842 if (cprm->file->f_pos != offset) { 1843 /* Sanity check */ 1844 printk(KERN_WARNING 1845 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n", 1846 cprm->file->f_pos, offset); 1847 } 1848 1849 end_coredump: 1850 set_fs(fs); 1851 1852 cleanup: 1853 while (!list_empty(&thread_list)) { 1854 struct list_head *tmp = thread_list.next; 1855 list_del(tmp); 1856 kfree(list_entry(tmp, struct elf_thread_status, list)); 1857 } 1858 kfree(phdr4note); 1859 kfree(elf); 1860 kfree(prstatus); 1861 kfree(psinfo); 1862 kfree(notes); 1863 kfree(fpu); 1864 kfree(shdr4extnum); 1865 #ifdef ELF_CORE_COPY_XFPREGS 1866 kfree(xfpu); 1867 #endif 1868 return has_dumped; 1869 #undef NUM_NOTES 1870 } 1871 1872 #endif /* CONFIG_ELF_CORE */ 1873