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 memset(&exec_params, 0, sizeof(exec_params)); 183 memset(&interp_params, 0, sizeof(interp_params)); 184 185 exec_params.hdr = *(struct elfhdr *) bprm->buf; 186 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE; 187 188 /* check that this is a binary we know how to deal with */ 189 retval = -ENOEXEC; 190 if (!is_elf_fdpic(&exec_params.hdr, bprm->file)) 191 goto error; 192 193 /* read the program header table */ 194 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file); 195 if (retval < 0) 196 goto error; 197 198 /* scan for a program header that specifies an interpreter */ 199 phdr = exec_params.phdrs; 200 201 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) { 202 switch (phdr->p_type) { 203 case PT_INTERP: 204 retval = -ENOMEM; 205 if (phdr->p_filesz > PATH_MAX) 206 goto error; 207 retval = -ENOENT; 208 if (phdr->p_filesz < 2) 209 goto error; 210 211 /* read the name of the interpreter into memory */ 212 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL); 213 if (!interpreter_name) 214 goto error; 215 216 retval = kernel_read(bprm->file, 217 phdr->p_offset, 218 interpreter_name, 219 phdr->p_filesz); 220 if (retval < 0) 221 goto error; 222 223 retval = -ENOENT; 224 if (interpreter_name[phdr->p_filesz - 1] != '\0') 225 goto error; 226 227 kdebug("Using ELF interpreter %s", interpreter_name); 228 229 /* replace the program with the interpreter */ 230 interpreter = open_exec(interpreter_name); 231 retval = PTR_ERR(interpreter); 232 if (IS_ERR(interpreter)) { 233 interpreter = NULL; 234 goto error; 235 } 236 237 /* 238 * If the binary is not readable then enforce 239 * mm->dumpable = 0 regardless of the interpreter's 240 * permissions. 241 */ 242 if (file_permission(interpreter, MAY_READ) < 0) 243 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP; 244 245 retval = kernel_read(interpreter, 0, bprm->buf, 246 BINPRM_BUF_SIZE); 247 if (retval < 0) 248 goto error; 249 250 interp_params.hdr = *((struct elfhdr *) bprm->buf); 251 break; 252 253 case PT_LOAD: 254 #ifdef CONFIG_MMU 255 if (exec_params.load_addr == 0) 256 exec_params.load_addr = phdr->p_vaddr; 257 #endif 258 break; 259 } 260 261 } 262 263 if (elf_check_const_displacement(&exec_params.hdr)) 264 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 265 266 /* perform insanity checks on the interpreter */ 267 if (interpreter_name) { 268 retval = -ELIBBAD; 269 if (!is_elf_fdpic(&interp_params.hdr, interpreter)) 270 goto error; 271 272 interp_params.flags = ELF_FDPIC_FLAG_PRESENT; 273 274 /* read the interpreter's program header table */ 275 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter); 276 if (retval < 0) 277 goto error; 278 } 279 280 stack_size = exec_params.stack_size; 281 if (stack_size < interp_params.stack_size) 282 stack_size = interp_params.stack_size; 283 284 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 285 executable_stack = EXSTACK_ENABLE_X; 286 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 287 executable_stack = EXSTACK_DISABLE_X; 288 else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 289 executable_stack = EXSTACK_ENABLE_X; 290 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 291 executable_stack = EXSTACK_DISABLE_X; 292 else 293 executable_stack = EXSTACK_DEFAULT; 294 295 retval = -ENOEXEC; 296 if (stack_size == 0) 297 goto error; 298 299 if (elf_check_const_displacement(&interp_params.hdr)) 300 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 301 302 /* flush all traces of the currently running executable */ 303 retval = flush_old_exec(bprm); 304 if (retval) 305 goto error; 306 307 /* there's now no turning back... the old userspace image is dead, 308 * defunct, deceased, etc. after this point we have to exit via 309 * error_kill */ 310 set_personality(PER_LINUX_FDPIC); 311 set_binfmt(&elf_fdpic_format); 312 313 current->mm->start_code = 0; 314 current->mm->end_code = 0; 315 current->mm->start_stack = 0; 316 current->mm->start_data = 0; 317 current->mm->end_data = 0; 318 current->mm->context.exec_fdpic_loadmap = 0; 319 current->mm->context.interp_fdpic_loadmap = 0; 320 321 current->flags &= ~PF_FORKNOEXEC; 322 323 #ifdef CONFIG_MMU 324 elf_fdpic_arch_lay_out_mm(&exec_params, 325 &interp_params, 326 ¤t->mm->start_stack, 327 ¤t->mm->start_brk); 328 329 retval = setup_arg_pages(bprm, current->mm->start_stack, 330 executable_stack); 331 if (retval < 0) { 332 send_sig(SIGKILL, current, 0); 333 goto error_kill; 334 } 335 #endif 336 337 /* load the executable and interpreter into memory */ 338 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm, 339 "executable"); 340 if (retval < 0) 341 goto error_kill; 342 343 if (interpreter_name) { 344 retval = elf_fdpic_map_file(&interp_params, interpreter, 345 current->mm, "interpreter"); 346 if (retval < 0) { 347 printk(KERN_ERR "Unable to load interpreter\n"); 348 goto error_kill; 349 } 350 351 allow_write_access(interpreter); 352 fput(interpreter); 353 interpreter = NULL; 354 } 355 356 #ifdef CONFIG_MMU 357 if (!current->mm->start_brk) 358 current->mm->start_brk = current->mm->end_data; 359 360 current->mm->brk = current->mm->start_brk = 361 PAGE_ALIGN(current->mm->start_brk); 362 363 #else 364 /* create a stack and brk area big enough for everyone 365 * - the brk heap starts at the bottom and works up 366 * - the stack starts at the top and works down 367 */ 368 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK; 369 if (stack_size < PAGE_SIZE * 2) 370 stack_size = PAGE_SIZE * 2; 371 372 down_write(¤t->mm->mmap_sem); 373 current->mm->start_brk = do_mmap(NULL, 0, stack_size, 374 PROT_READ | PROT_WRITE | PROT_EXEC, 375 MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN, 376 0); 377 378 if (IS_ERR_VALUE(current->mm->start_brk)) { 379 up_write(¤t->mm->mmap_sem); 380 retval = current->mm->start_brk; 381 current->mm->start_brk = 0; 382 goto error_kill; 383 } 384 385 /* expand the stack mapping to use up the entire allocation granule */ 386 fullsize = ksize((char *) current->mm->start_brk); 387 if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size, 388 fullsize, 0, 0))) 389 stack_size = fullsize; 390 up_write(¤t->mm->mmap_sem); 391 392 current->mm->brk = current->mm->start_brk; 393 current->mm->context.end_brk = current->mm->start_brk; 394 current->mm->context.end_brk += 395 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0; 396 current->mm->start_stack = current->mm->start_brk + stack_size; 397 #endif 398 399 compute_creds(bprm); 400 current->flags &= ~PF_FORKNOEXEC; 401 if (create_elf_fdpic_tables(bprm, current->mm, 402 &exec_params, &interp_params) < 0) 403 goto error_kill; 404 405 kdebug("- start_code %lx", current->mm->start_code); 406 kdebug("- end_code %lx", current->mm->end_code); 407 kdebug("- start_data %lx", current->mm->start_data); 408 kdebug("- end_data %lx", current->mm->end_data); 409 kdebug("- start_brk %lx", current->mm->start_brk); 410 kdebug("- brk %lx", current->mm->brk); 411 kdebug("- start_stack %lx", current->mm->start_stack); 412 413 #ifdef ELF_FDPIC_PLAT_INIT 414 /* 415 * The ABI may specify that certain registers be set up in special 416 * ways (on i386 %edx is the address of a DT_FINI function, for 417 * example. This macro performs whatever initialization to 418 * the regs structure is required. 419 */ 420 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr; 421 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr, 422 dynaddr); 423 #endif 424 425 /* everything is now ready... get the userspace context ready to roll */ 426 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr; 427 start_thread(regs, entryaddr, current->mm->start_stack); 428 429 if (unlikely(current->ptrace & PT_PTRACED)) { 430 if (current->ptrace & PT_TRACE_EXEC) 431 ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP); 432 else 433 send_sig(SIGTRAP, current, 0); 434 } 435 436 retval = 0; 437 438 error: 439 if (interpreter) { 440 allow_write_access(interpreter); 441 fput(interpreter); 442 } 443 kfree(interpreter_name); 444 kfree(exec_params.phdrs); 445 kfree(exec_params.loadmap); 446 kfree(interp_params.phdrs); 447 kfree(interp_params.loadmap); 448 return retval; 449 450 /* unrecoverable error - kill the process */ 451 error_kill: 452 send_sig(SIGSEGV, current, 0); 453 goto error; 454 455 } 456 457 /*****************************************************************************/ 458 /* 459 * present useful information to the program 460 */ 461 static int create_elf_fdpic_tables(struct linux_binprm *bprm, 462 struct mm_struct *mm, 463 struct elf_fdpic_params *exec_params, 464 struct elf_fdpic_params *interp_params) 465 { 466 unsigned long sp, csp, nitems; 467 elf_caddr_t __user *argv, *envp; 468 size_t platform_len = 0, len; 469 char *k_platform; 470 char __user *u_platform, *p; 471 long hwcap; 472 int loop; 473 474 /* we're going to shovel a whole load of stuff onto the stack */ 475 #ifdef CONFIG_MMU 476 sp = bprm->p; 477 #else 478 sp = mm->start_stack; 479 480 /* stack the program arguments and environment */ 481 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0) 482 return -EFAULT; 483 #endif 484 485 /* get hold of platform and hardware capabilities masks for the machine 486 * we are running on. In some cases (Sparc), this info is impossible 487 * to get, in others (i386) it is merely difficult. 488 */ 489 hwcap = ELF_HWCAP; 490 k_platform = ELF_PLATFORM; 491 u_platform = NULL; 492 493 if (k_platform) { 494 platform_len = strlen(k_platform) + 1; 495 sp -= platform_len; 496 u_platform = (char __user *) sp; 497 if (__copy_to_user(u_platform, k_platform, platform_len) != 0) 498 return -EFAULT; 499 } 500 501 #if defined(__i386__) && defined(CONFIG_SMP) 502 /* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions 503 * by the processes running on the same package. One thing we can do is 504 * to shuffle the initial stack for them. 505 * 506 * the conditionals here are unneeded, but kept in to make the code 507 * behaviour the same as pre change unless we have hyperthreaded 508 * processors. This keeps Mr Marcelo Person happier but should be 509 * removed for 2.5 510 */ 511 if (smp_num_siblings > 1) 512 sp = sp - ((current->pid % 64) << 7); 513 #endif 514 515 sp &= ~7UL; 516 517 /* stack the load map(s) */ 518 len = sizeof(struct elf32_fdpic_loadmap); 519 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs; 520 sp = (sp - len) & ~7UL; 521 exec_params->map_addr = sp; 522 523 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0) 524 return -EFAULT; 525 526 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp; 527 528 if (interp_params->loadmap) { 529 len = sizeof(struct elf32_fdpic_loadmap); 530 len += sizeof(struct elf32_fdpic_loadseg) * 531 interp_params->loadmap->nsegs; 532 sp = (sp - len) & ~7UL; 533 interp_params->map_addr = sp; 534 535 if (copy_to_user((void __user *) sp, interp_params->loadmap, 536 len) != 0) 537 return -EFAULT; 538 539 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp; 540 } 541 542 /* force 16 byte _final_ alignment here for generality */ 543 #define DLINFO_ITEMS 13 544 545 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0); 546 #ifdef DLINFO_ARCH_ITEMS 547 nitems += DLINFO_ARCH_ITEMS; 548 #endif 549 550 csp = sp; 551 sp -= nitems * 2 * sizeof(unsigned long); 552 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */ 553 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */ 554 sp -= 1 * sizeof(unsigned long); /* argc */ 555 556 csp -= sp & 15UL; 557 sp -= sp & 15UL; 558 559 /* put the ELF interpreter info on the stack */ 560 #define NEW_AUX_ENT(nr, id, val) \ 561 do { \ 562 struct { unsigned long _id, _val; } __user *ent; \ 563 \ 564 ent = (void __user *) csp; \ 565 __put_user((id), &ent[nr]._id); \ 566 __put_user((val), &ent[nr]._val); \ 567 } while (0) 568 569 csp -= 2 * sizeof(unsigned long); 570 NEW_AUX_ENT(0, AT_NULL, 0); 571 if (k_platform) { 572 csp -= 2 * sizeof(unsigned long); 573 NEW_AUX_ENT(0, AT_PLATFORM, 574 (elf_addr_t) (unsigned long) u_platform); 575 } 576 577 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long); 578 NEW_AUX_ENT( 0, AT_HWCAP, hwcap); 579 NEW_AUX_ENT( 1, AT_PAGESZ, PAGE_SIZE); 580 NEW_AUX_ENT( 2, AT_CLKTCK, CLOCKS_PER_SEC); 581 NEW_AUX_ENT( 3, AT_PHDR, exec_params->ph_addr); 582 NEW_AUX_ENT( 4, AT_PHENT, sizeof(struct elf_phdr)); 583 NEW_AUX_ENT( 5, AT_PHNUM, exec_params->hdr.e_phnum); 584 NEW_AUX_ENT( 6, AT_BASE, interp_params->elfhdr_addr); 585 NEW_AUX_ENT( 7, AT_FLAGS, 0); 586 NEW_AUX_ENT( 8, AT_ENTRY, exec_params->entry_addr); 587 NEW_AUX_ENT( 9, AT_UID, (elf_addr_t) current->uid); 588 NEW_AUX_ENT(10, AT_EUID, (elf_addr_t) current->euid); 589 NEW_AUX_ENT(11, AT_GID, (elf_addr_t) current->gid); 590 NEW_AUX_ENT(12, AT_EGID, (elf_addr_t) current->egid); 591 592 #ifdef ARCH_DLINFO 593 /* ARCH_DLINFO must come last so platform specific code can enforce 594 * special alignment requirements on the AUXV if necessary (eg. PPC). 595 */ 596 ARCH_DLINFO; 597 #endif 598 #undef NEW_AUX_ENT 599 600 /* allocate room for argv[] and envv[] */ 601 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t); 602 envp = (elf_caddr_t __user *) csp; 603 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t); 604 argv = (elf_caddr_t __user *) csp; 605 606 /* stack argc */ 607 csp -= sizeof(unsigned long); 608 __put_user(bprm->argc, (unsigned long __user *) csp); 609 610 BUG_ON(csp != sp); 611 612 /* fill in the argv[] array */ 613 #ifdef CONFIG_MMU 614 current->mm->arg_start = bprm->p; 615 #else 616 current->mm->arg_start = current->mm->start_stack - 617 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p); 618 #endif 619 620 p = (char __user *) current->mm->arg_start; 621 for (loop = bprm->argc; loop > 0; loop--) { 622 __put_user((elf_caddr_t) p, argv++); 623 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES); 624 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES) 625 return -EINVAL; 626 p += len; 627 } 628 __put_user(NULL, argv); 629 current->mm->arg_end = (unsigned long) p; 630 631 /* fill in the envv[] array */ 632 current->mm->env_start = (unsigned long) p; 633 for (loop = bprm->envc; loop > 0; loop--) { 634 __put_user((elf_caddr_t)(unsigned long) p, envp++); 635 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES); 636 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES) 637 return -EINVAL; 638 p += len; 639 } 640 __put_user(NULL, envp); 641 current->mm->env_end = (unsigned long) p; 642 643 mm->start_stack = (unsigned long) sp; 644 return 0; 645 } 646 647 /*****************************************************************************/ 648 /* 649 * transfer the program arguments and environment from the holding pages onto 650 * the stack 651 */ 652 #ifndef CONFIG_MMU 653 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm, 654 unsigned long *_sp) 655 { 656 unsigned long index, stop, sp; 657 char *src; 658 int ret = 0; 659 660 stop = bprm->p >> PAGE_SHIFT; 661 sp = *_sp; 662 663 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) { 664 src = kmap(bprm->page[index]); 665 sp -= PAGE_SIZE; 666 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0) 667 ret = -EFAULT; 668 kunmap(bprm->page[index]); 669 if (ret < 0) 670 goto out; 671 } 672 673 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15; 674 675 out: 676 return ret; 677 } 678 #endif 679 680 /*****************************************************************************/ 681 /* 682 * load the appropriate binary image (executable or interpreter) into memory 683 * - we assume no MMU is available 684 * - if no other PIC bits are set in params->hdr->e_flags 685 * - we assume that the LOADable segments in the binary are independently relocatable 686 * - we assume R/O executable segments are shareable 687 * - else 688 * - we assume the loadable parts of the image to require fixed displacement 689 * - the image is not shareable 690 */ 691 static int elf_fdpic_map_file(struct elf_fdpic_params *params, 692 struct file *file, 693 struct mm_struct *mm, 694 const char *what) 695 { 696 struct elf32_fdpic_loadmap *loadmap; 697 #ifdef CONFIG_MMU 698 struct elf32_fdpic_loadseg *mseg; 699 #endif 700 struct elf32_fdpic_loadseg *seg; 701 struct elf32_phdr *phdr; 702 unsigned long load_addr, stop; 703 unsigned nloads, tmp; 704 size_t size; 705 int loop, ret; 706 707 /* allocate a load map table */ 708 nloads = 0; 709 for (loop = 0; loop < params->hdr.e_phnum; loop++) 710 if (params->phdrs[loop].p_type == PT_LOAD) 711 nloads++; 712 713 if (nloads == 0) 714 return -ELIBBAD; 715 716 size = sizeof(*loadmap) + nloads * sizeof(*seg); 717 loadmap = kzalloc(size, GFP_KERNEL); 718 if (!loadmap) 719 return -ENOMEM; 720 721 params->loadmap = loadmap; 722 723 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION; 724 loadmap->nsegs = nloads; 725 726 load_addr = params->load_addr; 727 seg = loadmap->segs; 728 729 /* map the requested LOADs into the memory space */ 730 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 731 case ELF_FDPIC_FLAG_CONSTDISP: 732 case ELF_FDPIC_FLAG_CONTIGUOUS: 733 #ifndef CONFIG_MMU 734 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm); 735 if (ret < 0) 736 return ret; 737 break; 738 #endif 739 default: 740 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm); 741 if (ret < 0) 742 return ret; 743 break; 744 } 745 746 /* map the entry point */ 747 if (params->hdr.e_entry) { 748 seg = loadmap->segs; 749 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 750 if (params->hdr.e_entry >= seg->p_vaddr && 751 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) { 752 params->entry_addr = 753 (params->hdr.e_entry - seg->p_vaddr) + 754 seg->addr; 755 break; 756 } 757 } 758 } 759 760 /* determine where the program header table has wound up if mapped */ 761 stop = params->hdr.e_phoff; 762 stop += params->hdr.e_phnum * sizeof (struct elf_phdr); 763 phdr = params->phdrs; 764 765 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 766 if (phdr->p_type != PT_LOAD) 767 continue; 768 769 if (phdr->p_offset > params->hdr.e_phoff || 770 phdr->p_offset + phdr->p_filesz < stop) 771 continue; 772 773 seg = loadmap->segs; 774 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 775 if (phdr->p_vaddr >= seg->p_vaddr && 776 phdr->p_vaddr + phdr->p_filesz <= 777 seg->p_vaddr + seg->p_memsz) { 778 params->ph_addr = 779 (phdr->p_vaddr - seg->p_vaddr) + 780 seg->addr + 781 params->hdr.e_phoff - phdr->p_offset; 782 break; 783 } 784 } 785 break; 786 } 787 788 /* determine where the dynamic section has wound up if there is one */ 789 phdr = params->phdrs; 790 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 791 if (phdr->p_type != PT_DYNAMIC) 792 continue; 793 794 seg = loadmap->segs; 795 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 796 if (phdr->p_vaddr >= seg->p_vaddr && 797 phdr->p_vaddr + phdr->p_memsz <= 798 seg->p_vaddr + seg->p_memsz) { 799 params->dynamic_addr = 800 (phdr->p_vaddr - seg->p_vaddr) + 801 seg->addr; 802 803 /* check the dynamic section contains at least 804 * one item, and that the last item is a NULL 805 * entry */ 806 if (phdr->p_memsz == 0 || 807 phdr->p_memsz % sizeof(Elf32_Dyn) != 0) 808 goto dynamic_error; 809 810 tmp = phdr->p_memsz / sizeof(Elf32_Dyn); 811 if (((Elf32_Dyn *) 812 params->dynamic_addr)[tmp - 1].d_tag != 0) 813 goto dynamic_error; 814 break; 815 } 816 } 817 break; 818 } 819 820 /* now elide adjacent segments in the load map on MMU linux 821 * - on uClinux the holes between may actually be filled with system 822 * stuff or stuff from other processes 823 */ 824 #ifdef CONFIG_MMU 825 nloads = loadmap->nsegs; 826 mseg = loadmap->segs; 827 seg = mseg + 1; 828 for (loop = 1; loop < nloads; loop++) { 829 /* see if we have a candidate for merging */ 830 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) { 831 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz); 832 if (load_addr == (seg->addr & PAGE_MASK)) { 833 mseg->p_memsz += 834 load_addr - 835 (mseg->addr + mseg->p_memsz); 836 mseg->p_memsz += seg->addr & ~PAGE_MASK; 837 mseg->p_memsz += seg->p_memsz; 838 loadmap->nsegs--; 839 continue; 840 } 841 } 842 843 mseg++; 844 if (mseg != seg) 845 *mseg = *seg; 846 } 847 #endif 848 849 kdebug("Mapped Object [%s]:", what); 850 kdebug("- elfhdr : %lx", params->elfhdr_addr); 851 kdebug("- entry : %lx", params->entry_addr); 852 kdebug("- PHDR[] : %lx", params->ph_addr); 853 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr); 854 seg = loadmap->segs; 855 for (loop = 0; loop < loadmap->nsegs; loop++, seg++) 856 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]", 857 loop, 858 seg->addr, seg->addr + seg->p_memsz - 1, 859 seg->p_vaddr, seg->p_memsz); 860 861 return 0; 862 863 dynamic_error: 864 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n", 865 what, file->f_path.dentry->d_inode->i_ino); 866 return -ELIBBAD; 867 } 868 869 /*****************************************************************************/ 870 /* 871 * map a file with constant displacement under uClinux 872 */ 873 #ifndef CONFIG_MMU 874 static int elf_fdpic_map_file_constdisp_on_uclinux( 875 struct elf_fdpic_params *params, 876 struct file *file, 877 struct mm_struct *mm) 878 { 879 struct elf32_fdpic_loadseg *seg; 880 struct elf32_phdr *phdr; 881 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags; 882 loff_t fpos; 883 int loop, ret; 884 885 load_addr = params->load_addr; 886 seg = params->loadmap->segs; 887 888 /* determine the bounds of the contiguous overall allocation we must 889 * make */ 890 phdr = params->phdrs; 891 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 892 if (params->phdrs[loop].p_type != PT_LOAD) 893 continue; 894 895 if (base > phdr->p_vaddr) 896 base = phdr->p_vaddr; 897 if (top < phdr->p_vaddr + phdr->p_memsz) 898 top = phdr->p_vaddr + phdr->p_memsz; 899 } 900 901 /* allocate one big anon block for everything */ 902 mflags = MAP_PRIVATE; 903 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 904 mflags |= MAP_EXECUTABLE; 905 906 down_write(&mm->mmap_sem); 907 maddr = do_mmap(NULL, load_addr, top - base, 908 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0); 909 up_write(&mm->mmap_sem); 910 if (IS_ERR_VALUE(maddr)) 911 return (int) maddr; 912 913 if (load_addr != 0) 914 load_addr += PAGE_ALIGN(top - base); 915 916 /* and then load the file segments into it */ 917 phdr = params->phdrs; 918 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 919 if (params->phdrs[loop].p_type != PT_LOAD) 920 continue; 921 922 fpos = phdr->p_offset; 923 924 seg->addr = maddr + (phdr->p_vaddr - base); 925 seg->p_vaddr = phdr->p_vaddr; 926 seg->p_memsz = phdr->p_memsz; 927 928 ret = file->f_op->read(file, (void *) seg->addr, 929 phdr->p_filesz, &fpos); 930 if (ret < 0) 931 return ret; 932 933 /* map the ELF header address if in this segment */ 934 if (phdr->p_offset == 0) 935 params->elfhdr_addr = seg->addr; 936 937 /* clear any space allocated but not loaded */ 938 if (phdr->p_filesz < phdr->p_memsz) 939 clear_user((void *) (seg->addr + phdr->p_filesz), 940 phdr->p_memsz - phdr->p_filesz); 941 942 if (mm) { 943 if (phdr->p_flags & PF_X) { 944 mm->start_code = seg->addr; 945 mm->end_code = seg->addr + phdr->p_memsz; 946 } else if (!mm->start_data) { 947 mm->start_data = seg->addr; 948 #ifndef CONFIG_MMU 949 mm->end_data = seg->addr + phdr->p_memsz; 950 #endif 951 } 952 953 #ifdef CONFIG_MMU 954 if (seg->addr + phdr->p_memsz > mm->end_data) 955 mm->end_data = seg->addr + phdr->p_memsz; 956 #endif 957 } 958 959 seg++; 960 } 961 962 return 0; 963 } 964 #endif 965 966 /*****************************************************************************/ 967 /* 968 * map a binary by direct mmap() of the individual PT_LOAD segments 969 */ 970 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params, 971 struct file *file, 972 struct mm_struct *mm) 973 { 974 struct elf32_fdpic_loadseg *seg; 975 struct elf32_phdr *phdr; 976 unsigned long load_addr, delta_vaddr; 977 int loop, dvset; 978 979 load_addr = params->load_addr; 980 delta_vaddr = 0; 981 dvset = 0; 982 983 seg = params->loadmap->segs; 984 985 /* deal with each load segment separately */ 986 phdr = params->phdrs; 987 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 988 unsigned long maddr, disp, excess, excess1; 989 int prot = 0, flags; 990 991 if (phdr->p_type != PT_LOAD) 992 continue; 993 994 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx", 995 (unsigned long) phdr->p_vaddr, 996 (unsigned long) phdr->p_offset, 997 (unsigned long) phdr->p_filesz, 998 (unsigned long) phdr->p_memsz); 999 1000 /* determine the mapping parameters */ 1001 if (phdr->p_flags & PF_R) prot |= PROT_READ; 1002 if (phdr->p_flags & PF_W) prot |= PROT_WRITE; 1003 if (phdr->p_flags & PF_X) prot |= PROT_EXEC; 1004 1005 flags = MAP_PRIVATE | MAP_DENYWRITE; 1006 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 1007 flags |= MAP_EXECUTABLE; 1008 1009 maddr = 0; 1010 1011 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 1012 case ELF_FDPIC_FLAG_INDEPENDENT: 1013 /* PT_LOADs are independently locatable */ 1014 break; 1015 1016 case ELF_FDPIC_FLAG_HONOURVADDR: 1017 /* the specified virtual address must be honoured */ 1018 maddr = phdr->p_vaddr; 1019 flags |= MAP_FIXED; 1020 break; 1021 1022 case ELF_FDPIC_FLAG_CONSTDISP: 1023 /* constant displacement 1024 * - can be mapped anywhere, but must be mapped as a 1025 * unit 1026 */ 1027 if (!dvset) { 1028 maddr = load_addr; 1029 delta_vaddr = phdr->p_vaddr; 1030 dvset = 1; 1031 } else { 1032 maddr = load_addr + phdr->p_vaddr - delta_vaddr; 1033 flags |= MAP_FIXED; 1034 } 1035 break; 1036 1037 case ELF_FDPIC_FLAG_CONTIGUOUS: 1038 /* contiguity handled later */ 1039 break; 1040 1041 default: 1042 BUG(); 1043 } 1044 1045 maddr &= PAGE_MASK; 1046 1047 /* create the mapping */ 1048 disp = phdr->p_vaddr & ~PAGE_MASK; 1049 down_write(&mm->mmap_sem); 1050 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags, 1051 phdr->p_offset - disp); 1052 up_write(&mm->mmap_sem); 1053 1054 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx", 1055 loop, phdr->p_memsz + disp, prot, flags, 1056 phdr->p_offset - disp, maddr); 1057 1058 if (IS_ERR_VALUE(maddr)) 1059 return (int) maddr; 1060 1061 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) == 1062 ELF_FDPIC_FLAG_CONTIGUOUS) 1063 load_addr += PAGE_ALIGN(phdr->p_memsz + disp); 1064 1065 seg->addr = maddr + disp; 1066 seg->p_vaddr = phdr->p_vaddr; 1067 seg->p_memsz = phdr->p_memsz; 1068 1069 /* map the ELF header address if in this segment */ 1070 if (phdr->p_offset == 0) 1071 params->elfhdr_addr = seg->addr; 1072 1073 /* clear the bit between beginning of mapping and beginning of 1074 * PT_LOAD */ 1075 if (prot & PROT_WRITE && disp > 0) { 1076 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp); 1077 clear_user((void __user *) maddr, disp); 1078 maddr += disp; 1079 } 1080 1081 /* clear any space allocated but not loaded 1082 * - on uClinux we can just clear the lot 1083 * - on MMU linux we'll get a SIGBUS beyond the last page 1084 * extant in the file 1085 */ 1086 excess = phdr->p_memsz - phdr->p_filesz; 1087 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK); 1088 1089 #ifdef CONFIG_MMU 1090 if (excess > excess1) { 1091 unsigned long xaddr = maddr + phdr->p_filesz + excess1; 1092 unsigned long xmaddr; 1093 1094 flags |= MAP_FIXED | MAP_ANONYMOUS; 1095 down_write(&mm->mmap_sem); 1096 xmaddr = do_mmap(NULL, xaddr, excess - excess1, 1097 prot, flags, 0); 1098 up_write(&mm->mmap_sem); 1099 1100 kdebug("mmap[%d] <anon>" 1101 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx", 1102 loop, xaddr, excess - excess1, prot, flags, 1103 xmaddr); 1104 1105 if (xmaddr != xaddr) 1106 return -ENOMEM; 1107 } 1108 1109 if (prot & PROT_WRITE && excess1 > 0) { 1110 kdebug("clear[%d] ad=%lx sz=%lx", 1111 loop, maddr + phdr->p_filesz, excess1); 1112 clear_user((void __user *) maddr + phdr->p_filesz, 1113 excess1); 1114 } 1115 1116 #else 1117 if (excess > 0) { 1118 kdebug("clear[%d] ad=%lx sz=%lx", 1119 loop, maddr + phdr->p_filesz, excess); 1120 clear_user((void *) maddr + phdr->p_filesz, excess); 1121 } 1122 #endif 1123 1124 if (mm) { 1125 if (phdr->p_flags & PF_X) { 1126 mm->start_code = maddr; 1127 mm->end_code = maddr + phdr->p_memsz; 1128 } else if (!mm->start_data) { 1129 mm->start_data = maddr; 1130 mm->end_data = maddr + phdr->p_memsz; 1131 } 1132 } 1133 1134 seg++; 1135 } 1136 1137 return 0; 1138 } 1139 1140 /*****************************************************************************/ 1141 /* 1142 * ELF-FDPIC core dumper 1143 * 1144 * Modelled on fs/exec.c:aout_core_dump() 1145 * Jeremy Fitzhardinge <jeremy@sw.oz.au> 1146 * 1147 * Modelled on fs/binfmt_elf.c core dumper 1148 */ 1149 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE) 1150 1151 /* 1152 * These are the only things you should do on a core-file: use only these 1153 * functions to write out all the necessary info. 1154 */ 1155 static int dump_write(struct file *file, const void *addr, int nr) 1156 { 1157 return file->f_op->write(file, addr, nr, &file->f_pos) == nr; 1158 } 1159 1160 static int dump_seek(struct file *file, loff_t off) 1161 { 1162 if (file->f_op->llseek) { 1163 if (file->f_op->llseek(file, off, SEEK_SET) != off) 1164 return 0; 1165 } else { 1166 file->f_pos = off; 1167 } 1168 return 1; 1169 } 1170 1171 /* 1172 * Decide whether a segment is worth dumping; default is yes to be 1173 * sure (missing info is worse than too much; etc). 1174 * Personally I'd include everything, and use the coredump limit... 1175 * 1176 * I think we should skip something. But I am not sure how. H.J. 1177 */ 1178 static int maydump(struct vm_area_struct *vma) 1179 { 1180 /* Do not dump I/O mapped devices or special mappings */ 1181 if (vma->vm_flags & (VM_IO | VM_RESERVED)) { 1182 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags); 1183 return 0; 1184 } 1185 1186 /* If we may not read the contents, don't allow us to dump 1187 * them either. "dump_write()" can't handle it anyway. 1188 */ 1189 if (!(vma->vm_flags & VM_READ)) { 1190 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags); 1191 return 0; 1192 } 1193 1194 /* Dump shared memory only if mapped from an anonymous file. */ 1195 if (vma->vm_flags & VM_SHARED) { 1196 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) { 1197 kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags); 1198 return 1; 1199 } 1200 1201 kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags); 1202 return 0; 1203 } 1204 1205 #ifdef CONFIG_MMU 1206 /* If it hasn't been written to, don't write it out */ 1207 if (!vma->anon_vma) { 1208 kdcore("%08lx: %08lx: no (!anon)", vma->vm_start, vma->vm_flags); 1209 return 0; 1210 } 1211 #endif 1212 1213 kdcore("%08lx: %08lx: yes", vma->vm_start, vma->vm_flags); 1214 return 1; 1215 } 1216 1217 /* An ELF note in memory */ 1218 struct memelfnote 1219 { 1220 const char *name; 1221 int type; 1222 unsigned int datasz; 1223 void *data; 1224 }; 1225 1226 static int notesize(struct memelfnote *en) 1227 { 1228 int sz; 1229 1230 sz = sizeof(struct elf_note); 1231 sz += roundup(strlen(en->name) + 1, 4); 1232 sz += roundup(en->datasz, 4); 1233 1234 return sz; 1235 } 1236 1237 /* #define DEBUG */ 1238 1239 #define DUMP_WRITE(addr, nr) \ 1240 do { if (!dump_write(file, (addr), (nr))) return 0; } while(0) 1241 #define DUMP_SEEK(off) \ 1242 do { if (!dump_seek(file, (off))) return 0; } while(0) 1243 1244 static int writenote(struct memelfnote *men, struct file *file) 1245 { 1246 struct elf_note en; 1247 1248 en.n_namesz = strlen(men->name) + 1; 1249 en.n_descsz = men->datasz; 1250 en.n_type = men->type; 1251 1252 DUMP_WRITE(&en, sizeof(en)); 1253 DUMP_WRITE(men->name, en.n_namesz); 1254 /* XXX - cast from long long to long to avoid need for libgcc.a */ 1255 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */ 1256 DUMP_WRITE(men->data, men->datasz); 1257 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */ 1258 1259 return 1; 1260 } 1261 #undef DUMP_WRITE 1262 #undef DUMP_SEEK 1263 1264 #define DUMP_WRITE(addr, nr) \ 1265 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \ 1266 goto end_coredump; 1267 #define DUMP_SEEK(off) \ 1268 if (!dump_seek(file, (off))) \ 1269 goto end_coredump; 1270 1271 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs) 1272 { 1273 memcpy(elf->e_ident, ELFMAG, SELFMAG); 1274 elf->e_ident[EI_CLASS] = ELF_CLASS; 1275 elf->e_ident[EI_DATA] = ELF_DATA; 1276 elf->e_ident[EI_VERSION] = EV_CURRENT; 1277 elf->e_ident[EI_OSABI] = ELF_OSABI; 1278 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); 1279 1280 elf->e_type = ET_CORE; 1281 elf->e_machine = ELF_ARCH; 1282 elf->e_version = EV_CURRENT; 1283 elf->e_entry = 0; 1284 elf->e_phoff = sizeof(struct elfhdr); 1285 elf->e_shoff = 0; 1286 elf->e_flags = ELF_FDPIC_CORE_EFLAGS; 1287 elf->e_ehsize = sizeof(struct elfhdr); 1288 elf->e_phentsize = sizeof(struct elf_phdr); 1289 elf->e_phnum = segs; 1290 elf->e_shentsize = 0; 1291 elf->e_shnum = 0; 1292 elf->e_shstrndx = 0; 1293 return; 1294 } 1295 1296 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset) 1297 { 1298 phdr->p_type = PT_NOTE; 1299 phdr->p_offset = offset; 1300 phdr->p_vaddr = 0; 1301 phdr->p_paddr = 0; 1302 phdr->p_filesz = sz; 1303 phdr->p_memsz = 0; 1304 phdr->p_flags = 0; 1305 phdr->p_align = 0; 1306 return; 1307 } 1308 1309 static inline void fill_note(struct memelfnote *note, const char *name, int type, 1310 unsigned int sz, void *data) 1311 { 1312 note->name = name; 1313 note->type = type; 1314 note->datasz = sz; 1315 note->data = data; 1316 return; 1317 } 1318 1319 /* 1320 * fill up all the fields in prstatus from the given task struct, except 1321 * registers which need to be filled up seperately. 1322 */ 1323 static void fill_prstatus(struct elf_prstatus *prstatus, 1324 struct task_struct *p, long signr) 1325 { 1326 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 1327 prstatus->pr_sigpend = p->pending.signal.sig[0]; 1328 prstatus->pr_sighold = p->blocked.sig[0]; 1329 prstatus->pr_pid = p->pid; 1330 prstatus->pr_ppid = p->parent->pid; 1331 prstatus->pr_pgrp = process_group(p); 1332 prstatus->pr_sid = process_session(p); 1333 if (thread_group_leader(p)) { 1334 /* 1335 * This is the record for the group leader. Add in the 1336 * cumulative times of previous dead threads. This total 1337 * won't include the time of each live thread whose state 1338 * is included in the core dump. The final total reported 1339 * to our parent process when it calls wait4 will include 1340 * those sums as well as the little bit more time it takes 1341 * this and each other thread to finish dying after the 1342 * core dump synchronization phase. 1343 */ 1344 cputime_to_timeval(cputime_add(p->utime, p->signal->utime), 1345 &prstatus->pr_utime); 1346 cputime_to_timeval(cputime_add(p->stime, p->signal->stime), 1347 &prstatus->pr_stime); 1348 } else { 1349 cputime_to_timeval(p->utime, &prstatus->pr_utime); 1350 cputime_to_timeval(p->stime, &prstatus->pr_stime); 1351 } 1352 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime); 1353 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime); 1354 1355 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap; 1356 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap; 1357 } 1358 1359 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p, 1360 struct mm_struct *mm) 1361 { 1362 unsigned int i, len; 1363 1364 /* first copy the parameters from user space */ 1365 memset(psinfo, 0, sizeof(struct elf_prpsinfo)); 1366 1367 len = mm->arg_end - mm->arg_start; 1368 if (len >= ELF_PRARGSZ) 1369 len = ELF_PRARGSZ - 1; 1370 if (copy_from_user(&psinfo->pr_psargs, 1371 (const char __user *) mm->arg_start, len)) 1372 return -EFAULT; 1373 for (i = 0; i < len; i++) 1374 if (psinfo->pr_psargs[i] == 0) 1375 psinfo->pr_psargs[i] = ' '; 1376 psinfo->pr_psargs[len] = 0; 1377 1378 psinfo->pr_pid = p->pid; 1379 psinfo->pr_ppid = p->parent->pid; 1380 psinfo->pr_pgrp = process_group(p); 1381 psinfo->pr_sid = process_session(p); 1382 1383 i = p->state ? ffz(~p->state) + 1 : 0; 1384 psinfo->pr_state = i; 1385 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i]; 1386 psinfo->pr_zomb = psinfo->pr_sname == 'Z'; 1387 psinfo->pr_nice = task_nice(p); 1388 psinfo->pr_flag = p->flags; 1389 SET_UID(psinfo->pr_uid, p->uid); 1390 SET_GID(psinfo->pr_gid, p->gid); 1391 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname)); 1392 1393 return 0; 1394 } 1395 1396 /* Here is the structure in which status of each thread is captured. */ 1397 struct elf_thread_status 1398 { 1399 struct list_head list; 1400 struct elf_prstatus prstatus; /* NT_PRSTATUS */ 1401 elf_fpregset_t fpu; /* NT_PRFPREG */ 1402 struct task_struct *thread; 1403 #ifdef ELF_CORE_COPY_XFPREGS 1404 elf_fpxregset_t xfpu; /* NT_PRXFPREG */ 1405 #endif 1406 struct memelfnote notes[3]; 1407 int num_notes; 1408 }; 1409 1410 /* 1411 * In order to add the specific thread information for the elf file format, 1412 * we need to keep a linked list of every thread's pr_status and then create 1413 * a single section for them in the final core file. 1414 */ 1415 static int elf_dump_thread_status(long signr, struct elf_thread_status *t) 1416 { 1417 struct task_struct *p = t->thread; 1418 int sz = 0; 1419 1420 t->num_notes = 0; 1421 1422 fill_prstatus(&t->prstatus, p, signr); 1423 elf_core_copy_task_regs(p, &t->prstatus.pr_reg); 1424 1425 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus), 1426 &t->prstatus); 1427 t->num_notes++; 1428 sz += notesize(&t->notes[0]); 1429 1430 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu); 1431 if (t->prstatus.pr_fpvalid) { 1432 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu), 1433 &t->fpu); 1434 t->num_notes++; 1435 sz += notesize(&t->notes[1]); 1436 } 1437 1438 #ifdef ELF_CORE_COPY_XFPREGS 1439 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) { 1440 fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu), 1441 &t->xfpu); 1442 t->num_notes++; 1443 sz += notesize(&t->notes[2]); 1444 } 1445 #endif 1446 return sz; 1447 } 1448 1449 /* 1450 * dump the segments for an MMU process 1451 */ 1452 #ifdef CONFIG_MMU 1453 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm, 1454 size_t *size, unsigned long *limit) 1455 { 1456 struct vm_area_struct *vma; 1457 1458 for (vma = current->mm->mmap; vma; vma = vma->vm_next) { 1459 unsigned long addr; 1460 1461 if (!maydump(vma)) 1462 continue; 1463 1464 for (addr = vma->vm_start; 1465 addr < vma->vm_end; 1466 addr += PAGE_SIZE 1467 ) { 1468 struct vm_area_struct *vma; 1469 struct page *page; 1470 1471 if (get_user_pages(current, current->mm, addr, 1, 0, 1, 1472 &page, &vma) <= 0) { 1473 DUMP_SEEK(file->f_pos + PAGE_SIZE); 1474 } 1475 else if (page == ZERO_PAGE(addr)) { 1476 DUMP_SEEK(file->f_pos + PAGE_SIZE); 1477 page_cache_release(page); 1478 } 1479 else { 1480 void *kaddr; 1481 1482 flush_cache_page(vma, addr, page_to_pfn(page)); 1483 kaddr = kmap(page); 1484 if ((*size += PAGE_SIZE) > *limit || 1485 !dump_write(file, kaddr, PAGE_SIZE) 1486 ) { 1487 kunmap(page); 1488 page_cache_release(page); 1489 return -EIO; 1490 } 1491 kunmap(page); 1492 page_cache_release(page); 1493 } 1494 } 1495 } 1496 1497 return 0; 1498 1499 end_coredump: 1500 return -EFBIG; 1501 } 1502 #endif 1503 1504 /* 1505 * dump the segments for a NOMMU process 1506 */ 1507 #ifndef CONFIG_MMU 1508 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm, 1509 size_t *size, unsigned long *limit) 1510 { 1511 struct vm_list_struct *vml; 1512 1513 for (vml = current->mm->context.vmlist; vml; vml = vml->next) { 1514 struct vm_area_struct *vma = vml->vma; 1515 1516 if (!maydump(vma)) 1517 continue; 1518 1519 if ((*size += PAGE_SIZE) > *limit) 1520 return -EFBIG; 1521 1522 if (!dump_write(file, (void *) vma->vm_start, 1523 vma->vm_end - vma->vm_start)) 1524 return -EIO; 1525 } 1526 1527 return 0; 1528 } 1529 #endif 1530 1531 /* 1532 * Actual dumper 1533 * 1534 * This is a two-pass process; first we find the offsets of the bits, 1535 * and then they are actually written out. If we run out of core limit 1536 * we just truncate. 1537 */ 1538 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs, 1539 struct file *file) 1540 { 1541 #define NUM_NOTES 6 1542 int has_dumped = 0; 1543 mm_segment_t fs; 1544 int segs; 1545 size_t size = 0; 1546 int i; 1547 struct vm_area_struct *vma; 1548 struct elfhdr *elf = NULL; 1549 loff_t offset = 0, dataoff; 1550 unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur; 1551 int numnote; 1552 struct memelfnote *notes = NULL; 1553 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */ 1554 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */ 1555 struct task_struct *g, *p; 1556 LIST_HEAD(thread_list); 1557 struct list_head *t; 1558 elf_fpregset_t *fpu = NULL; 1559 #ifdef ELF_CORE_COPY_XFPREGS 1560 elf_fpxregset_t *xfpu = NULL; 1561 #endif 1562 int thread_status_size = 0; 1563 #ifndef CONFIG_MMU 1564 struct vm_list_struct *vml; 1565 #endif 1566 elf_addr_t *auxv; 1567 1568 /* 1569 * We no longer stop all VM operations. 1570 * 1571 * This is because those proceses that could possibly change map_count 1572 * or the mmap / vma pages are now blocked in do_exit on current 1573 * finishing this core dump. 1574 * 1575 * Only ptrace can touch these memory addresses, but it doesn't change 1576 * the map_count or the pages allocated. So no possibility of crashing 1577 * exists while dumping the mm->vm_next areas to the core file. 1578 */ 1579 1580 /* alloc memory for large data structures: too large to be on stack */ 1581 elf = kmalloc(sizeof(*elf), GFP_KERNEL); 1582 if (!elf) 1583 goto cleanup; 1584 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL); 1585 if (!prstatus) 1586 goto cleanup; 1587 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL); 1588 if (!psinfo) 1589 goto cleanup; 1590 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL); 1591 if (!notes) 1592 goto cleanup; 1593 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL); 1594 if (!fpu) 1595 goto cleanup; 1596 #ifdef ELF_CORE_COPY_XFPREGS 1597 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL); 1598 if (!xfpu) 1599 goto cleanup; 1600 #endif 1601 1602 if (signr) { 1603 struct elf_thread_status *tmp; 1604 rcu_read_lock(); 1605 do_each_thread(g,p) 1606 if (current->mm == p->mm && current != p) { 1607 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC); 1608 if (!tmp) { 1609 rcu_read_unlock(); 1610 goto cleanup; 1611 } 1612 tmp->thread = p; 1613 list_add(&tmp->list, &thread_list); 1614 } 1615 while_each_thread(g,p); 1616 rcu_read_unlock(); 1617 list_for_each(t, &thread_list) { 1618 struct elf_thread_status *tmp; 1619 int sz; 1620 1621 tmp = list_entry(t, struct elf_thread_status, list); 1622 sz = elf_dump_thread_status(signr, tmp); 1623 thread_status_size += sz; 1624 } 1625 } 1626 1627 /* now collect the dump for the current */ 1628 fill_prstatus(prstatus, current, signr); 1629 elf_core_copy_regs(&prstatus->pr_reg, regs); 1630 1631 #ifdef CONFIG_MMU 1632 segs = current->mm->map_count; 1633 #else 1634 segs = 0; 1635 for (vml = current->mm->context.vmlist; vml; vml = vml->next) 1636 segs++; 1637 #endif 1638 #ifdef ELF_CORE_EXTRA_PHDRS 1639 segs += ELF_CORE_EXTRA_PHDRS; 1640 #endif 1641 1642 /* Set up header */ 1643 fill_elf_fdpic_header(elf, segs + 1); /* including notes section */ 1644 1645 has_dumped = 1; 1646 current->flags |= PF_DUMPCORE; 1647 1648 /* 1649 * Set up the notes in similar form to SVR4 core dumps made 1650 * with info from their /proc. 1651 */ 1652 1653 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus); 1654 fill_psinfo(psinfo, current->group_leader, current->mm); 1655 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo); 1656 1657 numnote = 2; 1658 1659 auxv = (elf_addr_t *) current->mm->saved_auxv; 1660 1661 i = 0; 1662 do 1663 i += 2; 1664 while (auxv[i - 2] != AT_NULL); 1665 fill_note(¬es[numnote++], "CORE", NT_AUXV, 1666 i * sizeof(elf_addr_t), auxv); 1667 1668 /* Try to dump the FPU. */ 1669 if ((prstatus->pr_fpvalid = 1670 elf_core_copy_task_fpregs(current, regs, fpu))) 1671 fill_note(notes + numnote++, 1672 "CORE", NT_PRFPREG, sizeof(*fpu), fpu); 1673 #ifdef ELF_CORE_COPY_XFPREGS 1674 if (elf_core_copy_task_xfpregs(current, xfpu)) 1675 fill_note(notes + numnote++, 1676 "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu); 1677 #endif 1678 1679 fs = get_fs(); 1680 set_fs(KERNEL_DS); 1681 1682 DUMP_WRITE(elf, sizeof(*elf)); 1683 offset += sizeof(*elf); /* Elf header */ 1684 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers */ 1685 1686 /* Write notes phdr entry */ 1687 { 1688 struct elf_phdr phdr; 1689 int sz = 0; 1690 1691 for (i = 0; i < numnote; i++) 1692 sz += notesize(notes + i); 1693 1694 sz += thread_status_size; 1695 1696 fill_elf_note_phdr(&phdr, sz, offset); 1697 offset += sz; 1698 DUMP_WRITE(&phdr, sizeof(phdr)); 1699 } 1700 1701 /* Page-align dumped data */ 1702 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE); 1703 1704 /* write program headers for segments dump */ 1705 for ( 1706 #ifdef CONFIG_MMU 1707 vma = current->mm->mmap; vma; vma = vma->vm_next 1708 #else 1709 vml = current->mm->context.vmlist; vml; vml = vml->next 1710 #endif 1711 ) { 1712 struct elf_phdr phdr; 1713 size_t sz; 1714 1715 #ifndef CONFIG_MMU 1716 vma = vml->vma; 1717 #endif 1718 1719 sz = vma->vm_end - vma->vm_start; 1720 1721 phdr.p_type = PT_LOAD; 1722 phdr.p_offset = offset; 1723 phdr.p_vaddr = vma->vm_start; 1724 phdr.p_paddr = 0; 1725 phdr.p_filesz = maydump(vma) ? sz : 0; 1726 phdr.p_memsz = sz; 1727 offset += phdr.p_filesz; 1728 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0; 1729 if (vma->vm_flags & VM_WRITE) 1730 phdr.p_flags |= PF_W; 1731 if (vma->vm_flags & VM_EXEC) 1732 phdr.p_flags |= PF_X; 1733 phdr.p_align = ELF_EXEC_PAGESIZE; 1734 1735 DUMP_WRITE(&phdr, sizeof(phdr)); 1736 } 1737 1738 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS 1739 ELF_CORE_WRITE_EXTRA_PHDRS; 1740 #endif 1741 1742 /* write out the notes section */ 1743 for (i = 0; i < numnote; i++) 1744 if (!writenote(notes + i, file)) 1745 goto end_coredump; 1746 1747 /* write out the thread status notes section */ 1748 list_for_each(t, &thread_list) { 1749 struct elf_thread_status *tmp = 1750 list_entry(t, struct elf_thread_status, list); 1751 1752 for (i = 0; i < tmp->num_notes; i++) 1753 if (!writenote(&tmp->notes[i], file)) 1754 goto end_coredump; 1755 } 1756 1757 DUMP_SEEK(dataoff); 1758 1759 if (elf_fdpic_dump_segments(file, current->mm, &size, &limit) < 0) 1760 goto end_coredump; 1761 1762 #ifdef ELF_CORE_WRITE_EXTRA_DATA 1763 ELF_CORE_WRITE_EXTRA_DATA; 1764 #endif 1765 1766 if (file->f_pos != offset) { 1767 /* Sanity check */ 1768 printk(KERN_WARNING 1769 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n", 1770 file->f_pos, offset); 1771 } 1772 1773 end_coredump: 1774 set_fs(fs); 1775 1776 cleanup: 1777 while (!list_empty(&thread_list)) { 1778 struct list_head *tmp = thread_list.next; 1779 list_del(tmp); 1780 kfree(list_entry(tmp, struct elf_thread_status, list)); 1781 } 1782 1783 kfree(elf); 1784 kfree(prstatus); 1785 kfree(psinfo); 1786 kfree(notes); 1787 kfree(fpu); 1788 #ifdef ELF_CORE_COPY_XFPREGS 1789 kfree(xfpu); 1790 #endif 1791 return has_dumped; 1792 #undef NUM_NOTES 1793 } 1794 1795 #endif /* USE_ELF_CORE_DUMP */ 1796