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/highmem.h> 29 #include <linux/highuid.h> 30 #include <linux/personality.h> 31 #include <linux/ptrace.h> 32 #include <linux/init.h> 33 #include <linux/smp_lock.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 *); 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 (retval < 0) 141 return retval; 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 (retval < 0) 223 goto error; 224 225 retval = -ENOENT; 226 if (interpreter_name[phdr->p_filesz - 1] != '\0') 227 goto error; 228 229 kdebug("Using ELF interpreter %s", interpreter_name); 230 231 /* replace the program with the interpreter */ 232 interpreter = open_exec(interpreter_name); 233 retval = PTR_ERR(interpreter); 234 if (IS_ERR(interpreter)) { 235 interpreter = NULL; 236 goto error; 237 } 238 239 /* 240 * If the binary is not readable then enforce 241 * mm->dumpable = 0 regardless of the interpreter's 242 * permissions. 243 */ 244 if (file_permission(interpreter, MAY_READ) < 0) 245 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP; 246 247 retval = kernel_read(interpreter, 0, bprm->buf, 248 BINPRM_BUF_SIZE); 249 if (retval < 0) 250 goto error; 251 252 interp_params.hdr = *((struct elfhdr *) bprm->buf); 253 break; 254 255 case PT_LOAD: 256 #ifdef CONFIG_MMU 257 if (exec_params.load_addr == 0) 258 exec_params.load_addr = phdr->p_vaddr; 259 #endif 260 break; 261 } 262 263 } 264 265 if (elf_check_const_displacement(&exec_params.hdr)) 266 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 267 268 /* perform insanity checks on the interpreter */ 269 if (interpreter_name) { 270 retval = -ELIBBAD; 271 if (!is_elf_fdpic(&interp_params.hdr, interpreter)) 272 goto error; 273 274 interp_params.flags = ELF_FDPIC_FLAG_PRESENT; 275 276 /* read the interpreter's program header table */ 277 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter); 278 if (retval < 0) 279 goto error; 280 } 281 282 stack_size = exec_params.stack_size; 283 if (stack_size < interp_params.stack_size) 284 stack_size = interp_params.stack_size; 285 286 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 287 executable_stack = EXSTACK_ENABLE_X; 288 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 289 executable_stack = EXSTACK_DISABLE_X; 290 else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 291 executable_stack = EXSTACK_ENABLE_X; 292 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 293 executable_stack = EXSTACK_DISABLE_X; 294 else 295 executable_stack = EXSTACK_DEFAULT; 296 297 retval = -ENOEXEC; 298 if (stack_size == 0) 299 goto error; 300 301 if (elf_check_const_displacement(&interp_params.hdr)) 302 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 303 304 /* flush all traces of the currently running executable */ 305 retval = flush_old_exec(bprm); 306 if (retval) 307 goto error; 308 309 /* there's now no turning back... the old userspace image is dead, 310 * defunct, deceased, etc. after this point we have to exit via 311 * error_kill */ 312 set_personality(PER_LINUX_FDPIC); 313 set_binfmt(&elf_fdpic_format); 314 315 current->mm->start_code = 0; 316 current->mm->end_code = 0; 317 current->mm->start_stack = 0; 318 current->mm->start_data = 0; 319 current->mm->end_data = 0; 320 current->mm->context.exec_fdpic_loadmap = 0; 321 current->mm->context.interp_fdpic_loadmap = 0; 322 323 current->flags &= ~PF_FORKNOEXEC; 324 325 #ifdef CONFIG_MMU 326 elf_fdpic_arch_lay_out_mm(&exec_params, 327 &interp_params, 328 ¤t->mm->start_stack, 329 ¤t->mm->start_brk); 330 331 retval = setup_arg_pages(bprm, current->mm->start_stack, 332 executable_stack); 333 if (retval < 0) { 334 send_sig(SIGKILL, current, 0); 335 goto error_kill; 336 } 337 #endif 338 339 /* load the executable and interpreter into memory */ 340 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm, 341 "executable"); 342 if (retval < 0) 343 goto error_kill; 344 345 if (interpreter_name) { 346 retval = elf_fdpic_map_file(&interp_params, interpreter, 347 current->mm, "interpreter"); 348 if (retval < 0) { 349 printk(KERN_ERR "Unable to load interpreter\n"); 350 goto error_kill; 351 } 352 353 allow_write_access(interpreter); 354 fput(interpreter); 355 interpreter = NULL; 356 } 357 358 #ifdef CONFIG_MMU 359 if (!current->mm->start_brk) 360 current->mm->start_brk = current->mm->end_data; 361 362 current->mm->brk = current->mm->start_brk = 363 PAGE_ALIGN(current->mm->start_brk); 364 365 #else 366 /* create a stack and brk area big enough for everyone 367 * - the brk heap starts at the bottom and works up 368 * - the stack starts at the top and works down 369 */ 370 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK; 371 if (stack_size < PAGE_SIZE * 2) 372 stack_size = PAGE_SIZE * 2; 373 374 down_write(¤t->mm->mmap_sem); 375 current->mm->start_brk = do_mmap(NULL, 0, stack_size, 376 PROT_READ | PROT_WRITE | PROT_EXEC, 377 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN, 378 0); 379 380 if (IS_ERR_VALUE(current->mm->start_brk)) { 381 up_write(¤t->mm->mmap_sem); 382 retval = current->mm->start_brk; 383 current->mm->start_brk = 0; 384 goto error_kill; 385 } 386 387 /* expand the stack mapping to use up the entire allocation granule */ 388 fullsize = ksize((char *) current->mm->start_brk); 389 if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size, 390 fullsize, 0, 0))) 391 stack_size = fullsize; 392 up_write(¤t->mm->mmap_sem); 393 394 current->mm->brk = current->mm->start_brk; 395 current->mm->context.end_brk = current->mm->start_brk; 396 current->mm->context.end_brk += 397 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0; 398 current->mm->start_stack = current->mm->start_brk + stack_size; 399 #endif 400 401 compute_creds(bprm); 402 current->flags &= ~PF_FORKNOEXEC; 403 if (create_elf_fdpic_tables(bprm, current->mm, 404 &exec_params, &interp_params) < 0) 405 goto error_kill; 406 407 kdebug("- start_code %lx", current->mm->start_code); 408 kdebug("- end_code %lx", current->mm->end_code); 409 kdebug("- start_data %lx", current->mm->start_data); 410 kdebug("- end_data %lx", current->mm->end_data); 411 kdebug("- start_brk %lx", current->mm->start_brk); 412 kdebug("- brk %lx", current->mm->brk); 413 kdebug("- start_stack %lx", current->mm->start_stack); 414 415 #ifdef ELF_FDPIC_PLAT_INIT 416 /* 417 * The ABI may specify that certain registers be set up in special 418 * ways (on i386 %edx is the address of a DT_FINI function, for 419 * example. This macro performs whatever initialization to 420 * the regs structure is required. 421 */ 422 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr; 423 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr, 424 dynaddr); 425 #endif 426 427 /* everything is now ready... get the userspace context ready to roll */ 428 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr; 429 start_thread(regs, entryaddr, current->mm->start_stack); 430 431 if (unlikely(current->ptrace & PT_PTRACED)) { 432 if (current->ptrace & PT_TRACE_EXEC) 433 ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP); 434 else 435 send_sig(SIGTRAP, current, 0); 436 } 437 438 retval = 0; 439 440 error: 441 if (interpreter) { 442 allow_write_access(interpreter); 443 fput(interpreter); 444 } 445 kfree(interpreter_name); 446 kfree(exec_params.phdrs); 447 kfree(exec_params.loadmap); 448 kfree(interp_params.phdrs); 449 kfree(interp_params.loadmap); 450 return retval; 451 452 /* unrecoverable error - kill the process */ 453 error_kill: 454 send_sig(SIGSEGV, current, 0); 455 goto error; 456 457 } 458 459 /*****************************************************************************/ 460 /* 461 * present useful information to the program 462 */ 463 static int create_elf_fdpic_tables(struct linux_binprm *bprm, 464 struct mm_struct *mm, 465 struct elf_fdpic_params *exec_params, 466 struct elf_fdpic_params *interp_params) 467 { 468 unsigned long sp, csp, nitems; 469 elf_caddr_t __user *argv, *envp; 470 size_t platform_len = 0, len; 471 char *k_platform; 472 char __user *u_platform, *p; 473 long hwcap; 474 int loop; 475 476 /* we're going to shovel a whole load of stuff onto the stack */ 477 #ifdef CONFIG_MMU 478 sp = bprm->p; 479 #else 480 sp = mm->start_stack; 481 482 /* stack the program arguments and environment */ 483 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0) 484 return -EFAULT; 485 #endif 486 487 /* get hold of platform and hardware capabilities masks for the machine 488 * we are running on. In some cases (Sparc), this info is impossible 489 * to get, in others (i386) it is merely difficult. 490 */ 491 hwcap = ELF_HWCAP; 492 k_platform = ELF_PLATFORM; 493 u_platform = NULL; 494 495 if (k_platform) { 496 platform_len = strlen(k_platform) + 1; 497 sp -= platform_len; 498 u_platform = (char __user *) sp; 499 if (__copy_to_user(u_platform, k_platform, platform_len) != 0) 500 return -EFAULT; 501 } 502 503 #if defined(__i386__) && defined(CONFIG_SMP) 504 /* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions 505 * by the processes running on the same package. One thing we can do is 506 * to shuffle the initial stack for them. 507 * 508 * the conditionals here are unneeded, but kept in to make the code 509 * behaviour the same as pre change unless we have hyperthreaded 510 * processors. This keeps Mr Marcelo Person happier but should be 511 * removed for 2.5 512 */ 513 if (smp_num_siblings > 1) 514 sp = sp - ((current->pid % 64) << 7); 515 #endif 516 517 sp &= ~7UL; 518 519 /* stack the load map(s) */ 520 len = sizeof(struct elf32_fdpic_loadmap); 521 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs; 522 sp = (sp - len) & ~7UL; 523 exec_params->map_addr = sp; 524 525 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0) 526 return -EFAULT; 527 528 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp; 529 530 if (interp_params->loadmap) { 531 len = sizeof(struct elf32_fdpic_loadmap); 532 len += sizeof(struct elf32_fdpic_loadseg) * 533 interp_params->loadmap->nsegs; 534 sp = (sp - len) & ~7UL; 535 interp_params->map_addr = sp; 536 537 if (copy_to_user((void __user *) sp, interp_params->loadmap, 538 len) != 0) 539 return -EFAULT; 540 541 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp; 542 } 543 544 /* force 16 byte _final_ alignment here for generality */ 545 #define DLINFO_ITEMS 13 546 547 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0); 548 #ifdef DLINFO_ARCH_ITEMS 549 nitems += DLINFO_ARCH_ITEMS; 550 #endif 551 552 csp = sp; 553 sp -= nitems * 2 * sizeof(unsigned long); 554 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */ 555 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */ 556 sp -= 1 * sizeof(unsigned long); /* argc */ 557 558 csp -= sp & 15UL; 559 sp -= sp & 15UL; 560 561 /* put the ELF interpreter info on the stack */ 562 #define NEW_AUX_ENT(nr, id, val) \ 563 do { \ 564 struct { unsigned long _id, _val; } __user *ent; \ 565 \ 566 ent = (void __user *) csp; \ 567 __put_user((id), &ent[nr]._id); \ 568 __put_user((val), &ent[nr]._val); \ 569 } while (0) 570 571 csp -= 2 * sizeof(unsigned long); 572 NEW_AUX_ENT(0, AT_NULL, 0); 573 if (k_platform) { 574 csp -= 2 * sizeof(unsigned long); 575 NEW_AUX_ENT(0, AT_PLATFORM, 576 (elf_addr_t) (unsigned long) u_platform); 577 } 578 579 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long); 580 NEW_AUX_ENT( 0, AT_HWCAP, hwcap); 581 NEW_AUX_ENT( 1, AT_PAGESZ, PAGE_SIZE); 582 NEW_AUX_ENT( 2, AT_CLKTCK, CLOCKS_PER_SEC); 583 NEW_AUX_ENT( 3, AT_PHDR, exec_params->ph_addr); 584 NEW_AUX_ENT( 4, AT_PHENT, sizeof(struct elf_phdr)); 585 NEW_AUX_ENT( 5, AT_PHNUM, exec_params->hdr.e_phnum); 586 NEW_AUX_ENT( 6, AT_BASE, interp_params->elfhdr_addr); 587 NEW_AUX_ENT( 7, AT_FLAGS, 0); 588 NEW_AUX_ENT( 8, AT_ENTRY, exec_params->entry_addr); 589 NEW_AUX_ENT( 9, AT_UID, (elf_addr_t) current->uid); 590 NEW_AUX_ENT(10, AT_EUID, (elf_addr_t) current->euid); 591 NEW_AUX_ENT(11, AT_GID, (elf_addr_t) current->gid); 592 NEW_AUX_ENT(12, AT_EGID, (elf_addr_t) current->egid); 593 594 #ifdef ARCH_DLINFO 595 /* ARCH_DLINFO must come last so platform specific code can enforce 596 * special alignment requirements on the AUXV if necessary (eg. PPC). 597 */ 598 ARCH_DLINFO; 599 #endif 600 #undef NEW_AUX_ENT 601 602 /* allocate room for argv[] and envv[] */ 603 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t); 604 envp = (elf_caddr_t __user *) csp; 605 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t); 606 argv = (elf_caddr_t __user *) csp; 607 608 /* stack argc */ 609 csp -= sizeof(unsigned long); 610 __put_user(bprm->argc, (unsigned long __user *) csp); 611 612 BUG_ON(csp != sp); 613 614 /* fill in the argv[] array */ 615 #ifdef CONFIG_MMU 616 current->mm->arg_start = bprm->p; 617 #else 618 current->mm->arg_start = current->mm->start_stack - 619 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p); 620 #endif 621 622 p = (char __user *) current->mm->arg_start; 623 for (loop = bprm->argc; loop > 0; loop--) { 624 __put_user((elf_caddr_t) p, argv++); 625 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES); 626 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES) 627 return -EINVAL; 628 p += len; 629 } 630 __put_user(NULL, argv); 631 current->mm->arg_end = (unsigned long) p; 632 633 /* fill in the envv[] array */ 634 current->mm->env_start = (unsigned long) p; 635 for (loop = bprm->envc; loop > 0; loop--) { 636 __put_user((elf_caddr_t)(unsigned long) p, envp++); 637 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES); 638 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES) 639 return -EINVAL; 640 p += len; 641 } 642 __put_user(NULL, envp); 643 current->mm->env_end = (unsigned long) p; 644 645 mm->start_stack = (unsigned long) sp; 646 return 0; 647 } 648 649 /*****************************************************************************/ 650 /* 651 * transfer the program arguments and environment from the holding pages onto 652 * the stack 653 */ 654 #ifndef CONFIG_MMU 655 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm, 656 unsigned long *_sp) 657 { 658 unsigned long index, stop, sp; 659 char *src; 660 int ret = 0; 661 662 stop = bprm->p >> PAGE_SHIFT; 663 sp = *_sp; 664 665 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) { 666 src = kmap(bprm->page[index]); 667 sp -= PAGE_SIZE; 668 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0) 669 ret = -EFAULT; 670 kunmap(bprm->page[index]); 671 if (ret < 0) 672 goto out; 673 } 674 675 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15; 676 677 out: 678 return ret; 679 } 680 #endif 681 682 /*****************************************************************************/ 683 /* 684 * load the appropriate binary image (executable or interpreter) into memory 685 * - we assume no MMU is available 686 * - if no other PIC bits are set in params->hdr->e_flags 687 * - we assume that the LOADable segments in the binary are independently relocatable 688 * - we assume R/O executable segments are shareable 689 * - else 690 * - we assume the loadable parts of the image to require fixed displacement 691 * - the image is not shareable 692 */ 693 static int elf_fdpic_map_file(struct elf_fdpic_params *params, 694 struct file *file, 695 struct mm_struct *mm, 696 const char *what) 697 { 698 struct elf32_fdpic_loadmap *loadmap; 699 #ifdef CONFIG_MMU 700 struct elf32_fdpic_loadseg *mseg; 701 #endif 702 struct elf32_fdpic_loadseg *seg; 703 struct elf32_phdr *phdr; 704 unsigned long load_addr, stop; 705 unsigned nloads, tmp; 706 size_t size; 707 int loop, ret; 708 709 /* allocate a load map table */ 710 nloads = 0; 711 for (loop = 0; loop < params->hdr.e_phnum; loop++) 712 if (params->phdrs[loop].p_type == PT_LOAD) 713 nloads++; 714 715 if (nloads == 0) 716 return -ELIBBAD; 717 718 size = sizeof(*loadmap) + nloads * sizeof(*seg); 719 loadmap = kzalloc(size, GFP_KERNEL); 720 if (!loadmap) 721 return -ENOMEM; 722 723 params->loadmap = loadmap; 724 725 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION; 726 loadmap->nsegs = nloads; 727 728 load_addr = params->load_addr; 729 seg = loadmap->segs; 730 731 /* map the requested LOADs into the memory space */ 732 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 733 case ELF_FDPIC_FLAG_CONSTDISP: 734 case ELF_FDPIC_FLAG_CONTIGUOUS: 735 #ifndef CONFIG_MMU 736 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm); 737 if (ret < 0) 738 return ret; 739 break; 740 #endif 741 default: 742 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm); 743 if (ret < 0) 744 return ret; 745 break; 746 } 747 748 /* map the entry point */ 749 if (params->hdr.e_entry) { 750 seg = loadmap->segs; 751 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 752 if (params->hdr.e_entry >= seg->p_vaddr && 753 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) { 754 params->entry_addr = 755 (params->hdr.e_entry - seg->p_vaddr) + 756 seg->addr; 757 break; 758 } 759 } 760 } 761 762 /* determine where the program header table has wound up if mapped */ 763 stop = params->hdr.e_phoff; 764 stop += params->hdr.e_phnum * sizeof (struct elf_phdr); 765 phdr = params->phdrs; 766 767 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 768 if (phdr->p_type != PT_LOAD) 769 continue; 770 771 if (phdr->p_offset > params->hdr.e_phoff || 772 phdr->p_offset + phdr->p_filesz < stop) 773 continue; 774 775 seg = loadmap->segs; 776 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 777 if (phdr->p_vaddr >= seg->p_vaddr && 778 phdr->p_vaddr + phdr->p_filesz <= 779 seg->p_vaddr + seg->p_memsz) { 780 params->ph_addr = 781 (phdr->p_vaddr - seg->p_vaddr) + 782 seg->addr + 783 params->hdr.e_phoff - phdr->p_offset; 784 break; 785 } 786 } 787 break; 788 } 789 790 /* determine where the dynamic section has wound up if there is one */ 791 phdr = params->phdrs; 792 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 793 if (phdr->p_type != PT_DYNAMIC) 794 continue; 795 796 seg = loadmap->segs; 797 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 798 if (phdr->p_vaddr >= seg->p_vaddr && 799 phdr->p_vaddr + phdr->p_memsz <= 800 seg->p_vaddr + seg->p_memsz) { 801 params->dynamic_addr = 802 (phdr->p_vaddr - seg->p_vaddr) + 803 seg->addr; 804 805 /* check the dynamic section contains at least 806 * one item, and that the last item is a NULL 807 * entry */ 808 if (phdr->p_memsz == 0 || 809 phdr->p_memsz % sizeof(Elf32_Dyn) != 0) 810 goto dynamic_error; 811 812 tmp = phdr->p_memsz / sizeof(Elf32_Dyn); 813 if (((Elf32_Dyn *) 814 params->dynamic_addr)[tmp - 1].d_tag != 0) 815 goto dynamic_error; 816 break; 817 } 818 } 819 break; 820 } 821 822 /* now elide adjacent segments in the load map on MMU linux 823 * - on uClinux the holes between may actually be filled with system 824 * stuff or stuff from other processes 825 */ 826 #ifdef CONFIG_MMU 827 nloads = loadmap->nsegs; 828 mseg = loadmap->segs; 829 seg = mseg + 1; 830 for (loop = 1; loop < nloads; loop++) { 831 /* see if we have a candidate for merging */ 832 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) { 833 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz); 834 if (load_addr == (seg->addr & PAGE_MASK)) { 835 mseg->p_memsz += 836 load_addr - 837 (mseg->addr + mseg->p_memsz); 838 mseg->p_memsz += seg->addr & ~PAGE_MASK; 839 mseg->p_memsz += seg->p_memsz; 840 loadmap->nsegs--; 841 continue; 842 } 843 } 844 845 mseg++; 846 if (mseg != seg) 847 *mseg = *seg; 848 } 849 #endif 850 851 kdebug("Mapped Object [%s]:", what); 852 kdebug("- elfhdr : %lx", params->elfhdr_addr); 853 kdebug("- entry : %lx", params->entry_addr); 854 kdebug("- PHDR[] : %lx", params->ph_addr); 855 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr); 856 seg = loadmap->segs; 857 for (loop = 0; loop < loadmap->nsegs; loop++, seg++) 858 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]", 859 loop, 860 seg->addr, seg->addr + seg->p_memsz - 1, 861 seg->p_vaddr, seg->p_memsz); 862 863 return 0; 864 865 dynamic_error: 866 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n", 867 what, file->f_path.dentry->d_inode->i_ino); 868 return -ELIBBAD; 869 } 870 871 /*****************************************************************************/ 872 /* 873 * map a file with constant displacement under uClinux 874 */ 875 #ifndef CONFIG_MMU 876 static int elf_fdpic_map_file_constdisp_on_uclinux( 877 struct elf_fdpic_params *params, 878 struct file *file, 879 struct mm_struct *mm) 880 { 881 struct elf32_fdpic_loadseg *seg; 882 struct elf32_phdr *phdr; 883 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags; 884 loff_t fpos; 885 int loop, ret; 886 887 load_addr = params->load_addr; 888 seg = params->loadmap->segs; 889 890 /* determine the bounds of the contiguous overall allocation we must 891 * make */ 892 phdr = params->phdrs; 893 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 894 if (params->phdrs[loop].p_type != PT_LOAD) 895 continue; 896 897 if (base > phdr->p_vaddr) 898 base = phdr->p_vaddr; 899 if (top < phdr->p_vaddr + phdr->p_memsz) 900 top = phdr->p_vaddr + phdr->p_memsz; 901 } 902 903 /* allocate one big anon block for everything */ 904 mflags = MAP_PRIVATE; 905 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 906 mflags |= MAP_EXECUTABLE; 907 908 down_write(&mm->mmap_sem); 909 maddr = do_mmap(NULL, load_addr, top - base, 910 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0); 911 up_write(&mm->mmap_sem); 912 if (IS_ERR_VALUE(maddr)) 913 return (int) maddr; 914 915 if (load_addr != 0) 916 load_addr += PAGE_ALIGN(top - base); 917 918 /* and then load the file segments into it */ 919 phdr = params->phdrs; 920 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 921 if (params->phdrs[loop].p_type != PT_LOAD) 922 continue; 923 924 fpos = phdr->p_offset; 925 926 seg->addr = maddr + (phdr->p_vaddr - base); 927 seg->p_vaddr = phdr->p_vaddr; 928 seg->p_memsz = phdr->p_memsz; 929 930 ret = file->f_op->read(file, (void *) seg->addr, 931 phdr->p_filesz, &fpos); 932 if (ret < 0) 933 return ret; 934 935 /* map the ELF header address if in this segment */ 936 if (phdr->p_offset == 0) 937 params->elfhdr_addr = seg->addr; 938 939 /* clear any space allocated but not loaded */ 940 if (phdr->p_filesz < phdr->p_memsz) 941 clear_user((void *) (seg->addr + phdr->p_filesz), 942 phdr->p_memsz - phdr->p_filesz); 943 944 if (mm) { 945 if (phdr->p_flags & PF_X) { 946 if (!mm->start_code) { 947 mm->start_code = seg->addr; 948 mm->end_code = seg->addr + 949 phdr->p_memsz; 950 } 951 } else if (!mm->start_data) { 952 mm->start_data = seg->addr; 953 #ifndef CONFIG_MMU 954 mm->end_data = seg->addr + phdr->p_memsz; 955 #endif 956 } 957 958 #ifdef CONFIG_MMU 959 if (seg->addr + phdr->p_memsz > mm->end_data) 960 mm->end_data = seg->addr + phdr->p_memsz; 961 #endif 962 } 963 964 seg++; 965 } 966 967 return 0; 968 } 969 #endif 970 971 /*****************************************************************************/ 972 /* 973 * map a binary by direct mmap() of the individual PT_LOAD segments 974 */ 975 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params, 976 struct file *file, 977 struct mm_struct *mm) 978 { 979 struct elf32_fdpic_loadseg *seg; 980 struct elf32_phdr *phdr; 981 unsigned long load_addr, delta_vaddr; 982 int loop, dvset; 983 984 load_addr = params->load_addr; 985 delta_vaddr = 0; 986 dvset = 0; 987 988 seg = params->loadmap->segs; 989 990 /* deal with each load segment separately */ 991 phdr = params->phdrs; 992 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 993 unsigned long maddr, disp, excess, excess1; 994 int prot = 0, flags; 995 996 if (phdr->p_type != PT_LOAD) 997 continue; 998 999 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx", 1000 (unsigned long) phdr->p_vaddr, 1001 (unsigned long) phdr->p_offset, 1002 (unsigned long) phdr->p_filesz, 1003 (unsigned long) phdr->p_memsz); 1004 1005 /* determine the mapping parameters */ 1006 if (phdr->p_flags & PF_R) prot |= PROT_READ; 1007 if (phdr->p_flags & PF_W) prot |= PROT_WRITE; 1008 if (phdr->p_flags & PF_X) prot |= PROT_EXEC; 1009 1010 flags = MAP_PRIVATE | MAP_DENYWRITE; 1011 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 1012 flags |= MAP_EXECUTABLE; 1013 1014 maddr = 0; 1015 1016 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 1017 case ELF_FDPIC_FLAG_INDEPENDENT: 1018 /* PT_LOADs are independently locatable */ 1019 break; 1020 1021 case ELF_FDPIC_FLAG_HONOURVADDR: 1022 /* the specified virtual address must be honoured */ 1023 maddr = phdr->p_vaddr; 1024 flags |= MAP_FIXED; 1025 break; 1026 1027 case ELF_FDPIC_FLAG_CONSTDISP: 1028 /* constant displacement 1029 * - can be mapped anywhere, but must be mapped as a 1030 * unit 1031 */ 1032 if (!dvset) { 1033 maddr = load_addr; 1034 delta_vaddr = phdr->p_vaddr; 1035 dvset = 1; 1036 } else { 1037 maddr = load_addr + phdr->p_vaddr - delta_vaddr; 1038 flags |= MAP_FIXED; 1039 } 1040 break; 1041 1042 case ELF_FDPIC_FLAG_CONTIGUOUS: 1043 /* contiguity handled later */ 1044 break; 1045 1046 default: 1047 BUG(); 1048 } 1049 1050 maddr &= PAGE_MASK; 1051 1052 /* create the mapping */ 1053 disp = phdr->p_vaddr & ~PAGE_MASK; 1054 down_write(&mm->mmap_sem); 1055 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags, 1056 phdr->p_offset - disp); 1057 up_write(&mm->mmap_sem); 1058 1059 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx", 1060 loop, phdr->p_memsz + disp, prot, flags, 1061 phdr->p_offset - disp, maddr); 1062 1063 if (IS_ERR_VALUE(maddr)) 1064 return (int) maddr; 1065 1066 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) == 1067 ELF_FDPIC_FLAG_CONTIGUOUS) 1068 load_addr += PAGE_ALIGN(phdr->p_memsz + disp); 1069 1070 seg->addr = maddr + disp; 1071 seg->p_vaddr = phdr->p_vaddr; 1072 seg->p_memsz = phdr->p_memsz; 1073 1074 /* map the ELF header address if in this segment */ 1075 if (phdr->p_offset == 0) 1076 params->elfhdr_addr = seg->addr; 1077 1078 /* clear the bit between beginning of mapping and beginning of 1079 * PT_LOAD */ 1080 if (prot & PROT_WRITE && disp > 0) { 1081 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp); 1082 clear_user((void __user *) maddr, disp); 1083 maddr += disp; 1084 } 1085 1086 /* clear any space allocated but not loaded 1087 * - on uClinux we can just clear the lot 1088 * - on MMU linux we'll get a SIGBUS beyond the last page 1089 * extant in the file 1090 */ 1091 excess = phdr->p_memsz - phdr->p_filesz; 1092 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK); 1093 1094 #ifdef CONFIG_MMU 1095 if (excess > excess1) { 1096 unsigned long xaddr = maddr + phdr->p_filesz + excess1; 1097 unsigned long xmaddr; 1098 1099 flags |= MAP_FIXED | MAP_ANONYMOUS; 1100 down_write(&mm->mmap_sem); 1101 xmaddr = do_mmap(NULL, xaddr, excess - excess1, 1102 prot, flags, 0); 1103 up_write(&mm->mmap_sem); 1104 1105 kdebug("mmap[%d] <anon>" 1106 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx", 1107 loop, xaddr, excess - excess1, prot, flags, 1108 xmaddr); 1109 1110 if (xmaddr != xaddr) 1111 return -ENOMEM; 1112 } 1113 1114 if (prot & PROT_WRITE && excess1 > 0) { 1115 kdebug("clear[%d] ad=%lx sz=%lx", 1116 loop, maddr + phdr->p_filesz, excess1); 1117 clear_user((void __user *) maddr + phdr->p_filesz, 1118 excess1); 1119 } 1120 1121 #else 1122 if (excess > 0) { 1123 kdebug("clear[%d] ad=%lx sz=%lx", 1124 loop, maddr + phdr->p_filesz, excess); 1125 clear_user((void *) maddr + phdr->p_filesz, excess); 1126 } 1127 #endif 1128 1129 if (mm) { 1130 if (phdr->p_flags & PF_X) { 1131 if (!mm->start_code) { 1132 mm->start_code = maddr; 1133 mm->end_code = maddr + phdr->p_memsz; 1134 } 1135 } else if (!mm->start_data) { 1136 mm->start_data = maddr; 1137 mm->end_data = maddr + phdr->p_memsz; 1138 } 1139 } 1140 1141 seg++; 1142 } 1143 1144 return 0; 1145 } 1146 1147 /*****************************************************************************/ 1148 /* 1149 * ELF-FDPIC core dumper 1150 * 1151 * Modelled on fs/exec.c:aout_core_dump() 1152 * Jeremy Fitzhardinge <jeremy@sw.oz.au> 1153 * 1154 * Modelled on fs/binfmt_elf.c core dumper 1155 */ 1156 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE) 1157 1158 /* 1159 * These are the only things you should do on a core-file: use only these 1160 * functions to write out all the necessary info. 1161 */ 1162 static int dump_write(struct file *file, const void *addr, int nr) 1163 { 1164 return file->f_op->write(file, addr, nr, &file->f_pos) == nr; 1165 } 1166 1167 static int dump_seek(struct file *file, loff_t off) 1168 { 1169 if (file->f_op->llseek) { 1170 if (file->f_op->llseek(file, off, SEEK_SET) != off) 1171 return 0; 1172 } else { 1173 file->f_pos = off; 1174 } 1175 return 1; 1176 } 1177 1178 /* 1179 * Decide whether a segment is worth dumping; default is yes to be 1180 * sure (missing info is worse than too much; etc). 1181 * Personally I'd include everything, and use the coredump limit... 1182 * 1183 * I think we should skip something. But I am not sure how. H.J. 1184 */ 1185 static int maydump(struct vm_area_struct *vma) 1186 { 1187 /* Do not dump I/O mapped devices or special mappings */ 1188 if (vma->vm_flags & (VM_IO | VM_RESERVED)) { 1189 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags); 1190 return 0; 1191 } 1192 1193 /* If we may not read the contents, don't allow us to dump 1194 * them either. "dump_write()" can't handle it anyway. 1195 */ 1196 if (!(vma->vm_flags & VM_READ)) { 1197 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags); 1198 return 0; 1199 } 1200 1201 /* Dump shared memory only if mapped from an anonymous file. */ 1202 if (vma->vm_flags & VM_SHARED) { 1203 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) { 1204 kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags); 1205 return 1; 1206 } 1207 1208 kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags); 1209 return 0; 1210 } 1211 1212 #ifdef CONFIG_MMU 1213 /* If it hasn't been written to, don't write it out */ 1214 if (!vma->anon_vma) { 1215 kdcore("%08lx: %08lx: no (!anon)", vma->vm_start, vma->vm_flags); 1216 return 0; 1217 } 1218 #endif 1219 1220 kdcore("%08lx: %08lx: yes", vma->vm_start, vma->vm_flags); 1221 return 1; 1222 } 1223 1224 /* An ELF note in memory */ 1225 struct memelfnote 1226 { 1227 const char *name; 1228 int type; 1229 unsigned int datasz; 1230 void *data; 1231 }; 1232 1233 static int notesize(struct memelfnote *en) 1234 { 1235 int sz; 1236 1237 sz = sizeof(struct elf_note); 1238 sz += roundup(strlen(en->name) + 1, 4); 1239 sz += roundup(en->datasz, 4); 1240 1241 return sz; 1242 } 1243 1244 /* #define DEBUG */ 1245 1246 #define DUMP_WRITE(addr, nr) \ 1247 do { if (!dump_write(file, (addr), (nr))) return 0; } while(0) 1248 #define DUMP_SEEK(off) \ 1249 do { if (!dump_seek(file, (off))) return 0; } while(0) 1250 1251 static int writenote(struct memelfnote *men, struct file *file) 1252 { 1253 struct elf_note en; 1254 1255 en.n_namesz = strlen(men->name) + 1; 1256 en.n_descsz = men->datasz; 1257 en.n_type = men->type; 1258 1259 DUMP_WRITE(&en, sizeof(en)); 1260 DUMP_WRITE(men->name, en.n_namesz); 1261 /* XXX - cast from long long to long to avoid need for libgcc.a */ 1262 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */ 1263 DUMP_WRITE(men->data, men->datasz); 1264 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */ 1265 1266 return 1; 1267 } 1268 #undef DUMP_WRITE 1269 #undef DUMP_SEEK 1270 1271 #define DUMP_WRITE(addr, nr) \ 1272 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \ 1273 goto end_coredump; 1274 #define DUMP_SEEK(off) \ 1275 if (!dump_seek(file, (off))) \ 1276 goto end_coredump; 1277 1278 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs) 1279 { 1280 memcpy(elf->e_ident, ELFMAG, SELFMAG); 1281 elf->e_ident[EI_CLASS] = ELF_CLASS; 1282 elf->e_ident[EI_DATA] = ELF_DATA; 1283 elf->e_ident[EI_VERSION] = EV_CURRENT; 1284 elf->e_ident[EI_OSABI] = ELF_OSABI; 1285 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); 1286 1287 elf->e_type = ET_CORE; 1288 elf->e_machine = ELF_ARCH; 1289 elf->e_version = EV_CURRENT; 1290 elf->e_entry = 0; 1291 elf->e_phoff = sizeof(struct elfhdr); 1292 elf->e_shoff = 0; 1293 elf->e_flags = ELF_FDPIC_CORE_EFLAGS; 1294 elf->e_ehsize = sizeof(struct elfhdr); 1295 elf->e_phentsize = sizeof(struct elf_phdr); 1296 elf->e_phnum = segs; 1297 elf->e_shentsize = 0; 1298 elf->e_shnum = 0; 1299 elf->e_shstrndx = 0; 1300 return; 1301 } 1302 1303 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset) 1304 { 1305 phdr->p_type = PT_NOTE; 1306 phdr->p_offset = offset; 1307 phdr->p_vaddr = 0; 1308 phdr->p_paddr = 0; 1309 phdr->p_filesz = sz; 1310 phdr->p_memsz = 0; 1311 phdr->p_flags = 0; 1312 phdr->p_align = 0; 1313 return; 1314 } 1315 1316 static inline void fill_note(struct memelfnote *note, const char *name, int type, 1317 unsigned int sz, void *data) 1318 { 1319 note->name = name; 1320 note->type = type; 1321 note->datasz = sz; 1322 note->data = data; 1323 return; 1324 } 1325 1326 /* 1327 * fill up all the fields in prstatus from the given task struct, except 1328 * registers which need to be filled up seperately. 1329 */ 1330 static void fill_prstatus(struct elf_prstatus *prstatus, 1331 struct task_struct *p, long signr) 1332 { 1333 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 1334 prstatus->pr_sigpend = p->pending.signal.sig[0]; 1335 prstatus->pr_sighold = p->blocked.sig[0]; 1336 prstatus->pr_pid = p->pid; 1337 prstatus->pr_ppid = p->parent->pid; 1338 prstatus->pr_pgrp = process_group(p); 1339 prstatus->pr_sid = process_session(p); 1340 if (thread_group_leader(p)) { 1341 /* 1342 * This is the record for the group leader. Add in the 1343 * cumulative times of previous dead threads. This total 1344 * won't include the time of each live thread whose state 1345 * is included in the core dump. The final total reported 1346 * to our parent process when it calls wait4 will include 1347 * those sums as well as the little bit more time it takes 1348 * this and each other thread to finish dying after the 1349 * core dump synchronization phase. 1350 */ 1351 cputime_to_timeval(cputime_add(p->utime, p->signal->utime), 1352 &prstatus->pr_utime); 1353 cputime_to_timeval(cputime_add(p->stime, p->signal->stime), 1354 &prstatus->pr_stime); 1355 } else { 1356 cputime_to_timeval(p->utime, &prstatus->pr_utime); 1357 cputime_to_timeval(p->stime, &prstatus->pr_stime); 1358 } 1359 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime); 1360 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime); 1361 1362 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap; 1363 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap; 1364 } 1365 1366 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p, 1367 struct mm_struct *mm) 1368 { 1369 unsigned int i, len; 1370 1371 /* first copy the parameters from user space */ 1372 memset(psinfo, 0, sizeof(struct elf_prpsinfo)); 1373 1374 len = mm->arg_end - mm->arg_start; 1375 if (len >= ELF_PRARGSZ) 1376 len = ELF_PRARGSZ - 1; 1377 if (copy_from_user(&psinfo->pr_psargs, 1378 (const char __user *) mm->arg_start, len)) 1379 return -EFAULT; 1380 for (i = 0; i < len; i++) 1381 if (psinfo->pr_psargs[i] == 0) 1382 psinfo->pr_psargs[i] = ' '; 1383 psinfo->pr_psargs[len] = 0; 1384 1385 psinfo->pr_pid = p->pid; 1386 psinfo->pr_ppid = p->parent->pid; 1387 psinfo->pr_pgrp = process_group(p); 1388 psinfo->pr_sid = process_session(p); 1389 1390 i = p->state ? ffz(~p->state) + 1 : 0; 1391 psinfo->pr_state = i; 1392 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i]; 1393 psinfo->pr_zomb = psinfo->pr_sname == 'Z'; 1394 psinfo->pr_nice = task_nice(p); 1395 psinfo->pr_flag = p->flags; 1396 SET_UID(psinfo->pr_uid, p->uid); 1397 SET_GID(psinfo->pr_gid, p->gid); 1398 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname)); 1399 1400 return 0; 1401 } 1402 1403 /* Here is the structure in which status of each thread is captured. */ 1404 struct elf_thread_status 1405 { 1406 struct list_head list; 1407 struct elf_prstatus prstatus; /* NT_PRSTATUS */ 1408 elf_fpregset_t fpu; /* NT_PRFPREG */ 1409 struct task_struct *thread; 1410 #ifdef ELF_CORE_COPY_XFPREGS 1411 elf_fpxregset_t xfpu; /* NT_PRXFPREG */ 1412 #endif 1413 struct memelfnote notes[3]; 1414 int num_notes; 1415 }; 1416 1417 /* 1418 * In order to add the specific thread information for the elf file format, 1419 * we need to keep a linked list of every thread's pr_status and then create 1420 * a single section for them in the final core file. 1421 */ 1422 static int elf_dump_thread_status(long signr, struct elf_thread_status *t) 1423 { 1424 struct task_struct *p = t->thread; 1425 int sz = 0; 1426 1427 t->num_notes = 0; 1428 1429 fill_prstatus(&t->prstatus, p, signr); 1430 elf_core_copy_task_regs(p, &t->prstatus.pr_reg); 1431 1432 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus), 1433 &t->prstatus); 1434 t->num_notes++; 1435 sz += notesize(&t->notes[0]); 1436 1437 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu); 1438 if (t->prstatus.pr_fpvalid) { 1439 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu), 1440 &t->fpu); 1441 t->num_notes++; 1442 sz += notesize(&t->notes[1]); 1443 } 1444 1445 #ifdef ELF_CORE_COPY_XFPREGS 1446 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) { 1447 fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu), 1448 &t->xfpu); 1449 t->num_notes++; 1450 sz += notesize(&t->notes[2]); 1451 } 1452 #endif 1453 return sz; 1454 } 1455 1456 /* 1457 * dump the segments for an MMU process 1458 */ 1459 #ifdef CONFIG_MMU 1460 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm, 1461 size_t *size, unsigned long *limit) 1462 { 1463 struct vm_area_struct *vma; 1464 1465 for (vma = current->mm->mmap; vma; vma = vma->vm_next) { 1466 unsigned long addr; 1467 1468 if (!maydump(vma)) 1469 continue; 1470 1471 for (addr = vma->vm_start; 1472 addr < vma->vm_end; 1473 addr += PAGE_SIZE 1474 ) { 1475 struct vm_area_struct *vma; 1476 struct page *page; 1477 1478 if (get_user_pages(current, current->mm, addr, 1, 0, 1, 1479 &page, &vma) <= 0) { 1480 DUMP_SEEK(file->f_pos + PAGE_SIZE); 1481 } 1482 else if (page == ZERO_PAGE(addr)) { 1483 page_cache_release(page); 1484 DUMP_SEEK(file->f_pos + PAGE_SIZE); 1485 } 1486 else { 1487 void *kaddr; 1488 1489 flush_cache_page(vma, addr, page_to_pfn(page)); 1490 kaddr = kmap(page); 1491 if ((*size += PAGE_SIZE) > *limit || 1492 !dump_write(file, kaddr, PAGE_SIZE) 1493 ) { 1494 kunmap(page); 1495 page_cache_release(page); 1496 return -EIO; 1497 } 1498 kunmap(page); 1499 page_cache_release(page); 1500 } 1501 } 1502 } 1503 1504 return 0; 1505 1506 end_coredump: 1507 return -EFBIG; 1508 } 1509 #endif 1510 1511 /* 1512 * dump the segments for a NOMMU process 1513 */ 1514 #ifndef CONFIG_MMU 1515 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm, 1516 size_t *size, unsigned long *limit) 1517 { 1518 struct vm_list_struct *vml; 1519 1520 for (vml = current->mm->context.vmlist; vml; vml = vml->next) { 1521 struct vm_area_struct *vma = vml->vma; 1522 1523 if (!maydump(vma)) 1524 continue; 1525 1526 if ((*size += PAGE_SIZE) > *limit) 1527 return -EFBIG; 1528 1529 if (!dump_write(file, (void *) vma->vm_start, 1530 vma->vm_end - vma->vm_start)) 1531 return -EIO; 1532 } 1533 1534 return 0; 1535 } 1536 #endif 1537 1538 /* 1539 * Actual dumper 1540 * 1541 * This is a two-pass process; first we find the offsets of the bits, 1542 * and then they are actually written out. If we run out of core limit 1543 * we just truncate. 1544 */ 1545 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs, 1546 struct file *file) 1547 { 1548 #define NUM_NOTES 6 1549 int has_dumped = 0; 1550 mm_segment_t fs; 1551 int segs; 1552 size_t size = 0; 1553 int i; 1554 struct vm_area_struct *vma; 1555 struct elfhdr *elf = NULL; 1556 loff_t offset = 0, dataoff; 1557 unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur; 1558 int numnote; 1559 struct memelfnote *notes = NULL; 1560 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */ 1561 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */ 1562 struct task_struct *g, *p; 1563 LIST_HEAD(thread_list); 1564 struct list_head *t; 1565 elf_fpregset_t *fpu = NULL; 1566 #ifdef ELF_CORE_COPY_XFPREGS 1567 elf_fpxregset_t *xfpu = NULL; 1568 #endif 1569 int thread_status_size = 0; 1570 #ifndef CONFIG_MMU 1571 struct vm_list_struct *vml; 1572 #endif 1573 elf_addr_t *auxv; 1574 1575 /* 1576 * We no longer stop all VM operations. 1577 * 1578 * This is because those proceses that could possibly change map_count 1579 * or the mmap / vma pages are now blocked in do_exit on current 1580 * finishing this core dump. 1581 * 1582 * Only ptrace can touch these memory addresses, but it doesn't change 1583 * the map_count or the pages allocated. So no possibility of crashing 1584 * exists while dumping the mm->vm_next areas to the core file. 1585 */ 1586 1587 /* alloc memory for large data structures: too large to be on stack */ 1588 elf = kmalloc(sizeof(*elf), GFP_KERNEL); 1589 if (!elf) 1590 goto cleanup; 1591 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL); 1592 if (!prstatus) 1593 goto cleanup; 1594 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL); 1595 if (!psinfo) 1596 goto cleanup; 1597 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL); 1598 if (!notes) 1599 goto cleanup; 1600 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL); 1601 if (!fpu) 1602 goto cleanup; 1603 #ifdef ELF_CORE_COPY_XFPREGS 1604 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL); 1605 if (!xfpu) 1606 goto cleanup; 1607 #endif 1608 1609 if (signr) { 1610 struct elf_thread_status *tmp; 1611 rcu_read_lock(); 1612 do_each_thread(g,p) 1613 if (current->mm == p->mm && current != p) { 1614 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC); 1615 if (!tmp) { 1616 rcu_read_unlock(); 1617 goto cleanup; 1618 } 1619 tmp->thread = p; 1620 list_add(&tmp->list, &thread_list); 1621 } 1622 while_each_thread(g,p); 1623 rcu_read_unlock(); 1624 list_for_each(t, &thread_list) { 1625 struct elf_thread_status *tmp; 1626 int sz; 1627 1628 tmp = list_entry(t, struct elf_thread_status, list); 1629 sz = elf_dump_thread_status(signr, tmp); 1630 thread_status_size += sz; 1631 } 1632 } 1633 1634 /* now collect the dump for the current */ 1635 fill_prstatus(prstatus, current, signr); 1636 elf_core_copy_regs(&prstatus->pr_reg, regs); 1637 1638 #ifdef CONFIG_MMU 1639 segs = current->mm->map_count; 1640 #else 1641 segs = 0; 1642 for (vml = current->mm->context.vmlist; vml; vml = vml->next) 1643 segs++; 1644 #endif 1645 #ifdef ELF_CORE_EXTRA_PHDRS 1646 segs += ELF_CORE_EXTRA_PHDRS; 1647 #endif 1648 1649 /* Set up header */ 1650 fill_elf_fdpic_header(elf, segs + 1); /* including notes section */ 1651 1652 has_dumped = 1; 1653 current->flags |= PF_DUMPCORE; 1654 1655 /* 1656 * Set up the notes in similar form to SVR4 core dumps made 1657 * with info from their /proc. 1658 */ 1659 1660 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus); 1661 fill_psinfo(psinfo, current->group_leader, current->mm); 1662 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo); 1663 1664 numnote = 2; 1665 1666 auxv = (elf_addr_t *) current->mm->saved_auxv; 1667 1668 i = 0; 1669 do 1670 i += 2; 1671 while (auxv[i - 2] != AT_NULL); 1672 fill_note(¬es[numnote++], "CORE", NT_AUXV, 1673 i * sizeof(elf_addr_t), auxv); 1674 1675 /* Try to dump the FPU. */ 1676 if ((prstatus->pr_fpvalid = 1677 elf_core_copy_task_fpregs(current, regs, fpu))) 1678 fill_note(notes + numnote++, 1679 "CORE", NT_PRFPREG, sizeof(*fpu), fpu); 1680 #ifdef ELF_CORE_COPY_XFPREGS 1681 if (elf_core_copy_task_xfpregs(current, xfpu)) 1682 fill_note(notes + numnote++, 1683 "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu); 1684 #endif 1685 1686 fs = get_fs(); 1687 set_fs(KERNEL_DS); 1688 1689 DUMP_WRITE(elf, sizeof(*elf)); 1690 offset += sizeof(*elf); /* Elf header */ 1691 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers */ 1692 1693 /* Write notes phdr entry */ 1694 { 1695 struct elf_phdr phdr; 1696 int sz = 0; 1697 1698 for (i = 0; i < numnote; i++) 1699 sz += notesize(notes + i); 1700 1701 sz += thread_status_size; 1702 1703 fill_elf_note_phdr(&phdr, sz, offset); 1704 offset += sz; 1705 DUMP_WRITE(&phdr, sizeof(phdr)); 1706 } 1707 1708 /* Page-align dumped data */ 1709 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE); 1710 1711 /* write program headers for segments dump */ 1712 for ( 1713 #ifdef CONFIG_MMU 1714 vma = current->mm->mmap; vma; vma = vma->vm_next 1715 #else 1716 vml = current->mm->context.vmlist; vml; vml = vml->next 1717 #endif 1718 ) { 1719 struct elf_phdr phdr; 1720 size_t sz; 1721 1722 #ifndef CONFIG_MMU 1723 vma = vml->vma; 1724 #endif 1725 1726 sz = vma->vm_end - vma->vm_start; 1727 1728 phdr.p_type = PT_LOAD; 1729 phdr.p_offset = offset; 1730 phdr.p_vaddr = vma->vm_start; 1731 phdr.p_paddr = 0; 1732 phdr.p_filesz = maydump(vma) ? sz : 0; 1733 phdr.p_memsz = sz; 1734 offset += phdr.p_filesz; 1735 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0; 1736 if (vma->vm_flags & VM_WRITE) 1737 phdr.p_flags |= PF_W; 1738 if (vma->vm_flags & VM_EXEC) 1739 phdr.p_flags |= PF_X; 1740 phdr.p_align = ELF_EXEC_PAGESIZE; 1741 1742 DUMP_WRITE(&phdr, sizeof(phdr)); 1743 } 1744 1745 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS 1746 ELF_CORE_WRITE_EXTRA_PHDRS; 1747 #endif 1748 1749 /* write out the notes section */ 1750 for (i = 0; i < numnote; i++) 1751 if (!writenote(notes + i, file)) 1752 goto end_coredump; 1753 1754 /* write out the thread status notes section */ 1755 list_for_each(t, &thread_list) { 1756 struct elf_thread_status *tmp = 1757 list_entry(t, struct elf_thread_status, list); 1758 1759 for (i = 0; i < tmp->num_notes; i++) 1760 if (!writenote(&tmp->notes[i], file)) 1761 goto end_coredump; 1762 } 1763 1764 DUMP_SEEK(dataoff); 1765 1766 if (elf_fdpic_dump_segments(file, current->mm, &size, &limit) < 0) 1767 goto end_coredump; 1768 1769 #ifdef ELF_CORE_WRITE_EXTRA_DATA 1770 ELF_CORE_WRITE_EXTRA_DATA; 1771 #endif 1772 1773 if (file->f_pos != offset) { 1774 /* Sanity check */ 1775 printk(KERN_WARNING 1776 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n", 1777 file->f_pos, offset); 1778 } 1779 1780 end_coredump: 1781 set_fs(fs); 1782 1783 cleanup: 1784 while (!list_empty(&thread_list)) { 1785 struct list_head *tmp = thread_list.next; 1786 list_del(tmp); 1787 kfree(list_entry(tmp, struct elf_thread_status, list)); 1788 } 1789 1790 kfree(elf); 1791 kfree(prstatus); 1792 kfree(psinfo); 1793 kfree(notes); 1794 kfree(fpu); 1795 #ifdef ELF_CORE_COPY_XFPREGS 1796 kfree(xfpu); 1797 #endif 1798 return has_dumped; 1799 #undef NUM_NOTES 1800 } 1801 1802 #endif /* USE_ELF_CORE_DUMP */ 1803