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