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