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 #ifndef elf_addr_t 44 #define elf_addr_t unsigned long 45 #endif 46 47 #if 0 48 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ ) 49 #else 50 #define kdebug(fmt, ...) do {} while(0) 51 #endif 52 53 #if 0 54 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ ) 55 #else 56 #define kdcore(fmt, ...) do {} while(0) 57 #endif 58 59 MODULE_LICENSE("GPL"); 60 61 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *); 62 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *); 63 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *, 64 struct mm_struct *, const char *); 65 66 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *, 67 struct elf_fdpic_params *, 68 struct elf_fdpic_params *); 69 70 #ifndef CONFIG_MMU 71 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *, 72 unsigned long *); 73 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *, 74 struct file *, 75 struct mm_struct *); 76 #endif 77 78 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *, 79 struct file *, struct mm_struct *); 80 81 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE) 82 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *); 83 #endif 84 85 static struct linux_binfmt elf_fdpic_format = { 86 .module = THIS_MODULE, 87 .load_binary = load_elf_fdpic_binary, 88 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE) 89 .core_dump = elf_fdpic_core_dump, 90 #endif 91 .min_coredump = ELF_EXEC_PAGESIZE, 92 }; 93 94 static int __init init_elf_fdpic_binfmt(void) 95 { 96 return register_binfmt(&elf_fdpic_format); 97 } 98 99 static void __exit exit_elf_fdpic_binfmt(void) 100 { 101 unregister_binfmt(&elf_fdpic_format); 102 } 103 104 core_initcall(init_elf_fdpic_binfmt); 105 module_exit(exit_elf_fdpic_binfmt); 106 107 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file) 108 { 109 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0) 110 return 0; 111 if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN) 112 return 0; 113 if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr)) 114 return 0; 115 if (!file->f_op || !file->f_op->mmap) 116 return 0; 117 return 1; 118 } 119 120 /*****************************************************************************/ 121 /* 122 * read the program headers table into memory 123 */ 124 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params, 125 struct file *file) 126 { 127 struct elf32_phdr *phdr; 128 unsigned long size; 129 int retval, loop; 130 131 if (params->hdr.e_phentsize != sizeof(struct elf_phdr)) 132 return -ENOMEM; 133 if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr)) 134 return -ENOMEM; 135 136 size = params->hdr.e_phnum * sizeof(struct elf_phdr); 137 params->phdrs = kmalloc(size, GFP_KERNEL); 138 if (!params->phdrs) 139 return -ENOMEM; 140 141 retval = kernel_read(file, params->hdr.e_phoff, 142 (char *) params->phdrs, size); 143 if (retval < 0) 144 return retval; 145 146 /* determine stack size for this binary */ 147 phdr = params->phdrs; 148 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 149 if (phdr->p_type != PT_GNU_STACK) 150 continue; 151 152 if (phdr->p_flags & PF_X) 153 params->flags |= ELF_FDPIC_FLAG_EXEC_STACK; 154 else 155 params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK; 156 157 params->stack_size = phdr->p_memsz; 158 break; 159 } 160 161 return 0; 162 } 163 164 /*****************************************************************************/ 165 /* 166 * load an fdpic binary into various bits of memory 167 */ 168 static int load_elf_fdpic_binary(struct linux_binprm *bprm, 169 struct pt_regs *regs) 170 { 171 struct elf_fdpic_params exec_params, interp_params; 172 struct elf_phdr *phdr; 173 unsigned long stack_size, entryaddr; 174 #ifndef CONFIG_MMU 175 unsigned long fullsize; 176 #endif 177 #ifdef ELF_FDPIC_PLAT_INIT 178 unsigned long dynaddr; 179 #endif 180 struct file *interpreter = NULL; /* to shut gcc up */ 181 char *interpreter_name = NULL; 182 int executable_stack; 183 int retval, i; 184 185 memset(&exec_params, 0, sizeof(exec_params)); 186 memset(&interp_params, 0, sizeof(interp_params)); 187 188 exec_params.hdr = *(struct elfhdr *) bprm->buf; 189 exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE; 190 191 /* check that this is a binary we know how to deal with */ 192 retval = -ENOEXEC; 193 if (!is_elf_fdpic(&exec_params.hdr, bprm->file)) 194 goto error; 195 196 /* read the program header table */ 197 retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file); 198 if (retval < 0) 199 goto error; 200 201 /* scan for a program header that specifies an interpreter */ 202 phdr = exec_params.phdrs; 203 204 for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) { 205 switch (phdr->p_type) { 206 case PT_INTERP: 207 retval = -ENOMEM; 208 if (phdr->p_filesz > PATH_MAX) 209 goto error; 210 retval = -ENOENT; 211 if (phdr->p_filesz < 2) 212 goto error; 213 214 /* read the name of the interpreter into memory */ 215 interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL); 216 if (!interpreter_name) 217 goto error; 218 219 retval = kernel_read(bprm->file, 220 phdr->p_offset, 221 interpreter_name, 222 phdr->p_filesz); 223 if (retval < 0) 224 goto error; 225 226 retval = -ENOENT; 227 if (interpreter_name[phdr->p_filesz - 1] != '\0') 228 goto error; 229 230 kdebug("Using ELF interpreter %s", interpreter_name); 231 232 /* replace the program with the interpreter */ 233 interpreter = open_exec(interpreter_name); 234 retval = PTR_ERR(interpreter); 235 if (IS_ERR(interpreter)) { 236 interpreter = NULL; 237 goto error; 238 } 239 240 retval = kernel_read(interpreter, 0, bprm->buf, 241 BINPRM_BUF_SIZE); 242 if (retval < 0) 243 goto error; 244 245 interp_params.hdr = *((struct elfhdr *) bprm->buf); 246 break; 247 248 case PT_LOAD: 249 #ifdef CONFIG_MMU 250 if (exec_params.load_addr == 0) 251 exec_params.load_addr = phdr->p_vaddr; 252 #endif 253 break; 254 } 255 256 } 257 258 if (elf_check_const_displacement(&exec_params.hdr)) 259 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 260 261 /* perform insanity checks on the interpreter */ 262 if (interpreter_name) { 263 retval = -ELIBBAD; 264 if (!is_elf_fdpic(&interp_params.hdr, interpreter)) 265 goto error; 266 267 interp_params.flags = ELF_FDPIC_FLAG_PRESENT; 268 269 /* read the interpreter's program header table */ 270 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter); 271 if (retval < 0) 272 goto error; 273 } 274 275 stack_size = exec_params.stack_size; 276 if (stack_size < interp_params.stack_size) 277 stack_size = interp_params.stack_size; 278 279 if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 280 executable_stack = EXSTACK_ENABLE_X; 281 else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 282 executable_stack = EXSTACK_DISABLE_X; 283 else if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK) 284 executable_stack = EXSTACK_ENABLE_X; 285 else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK) 286 executable_stack = EXSTACK_DISABLE_X; 287 else 288 executable_stack = EXSTACK_DEFAULT; 289 290 retval = -ENOEXEC; 291 if (stack_size == 0) 292 goto error; 293 294 if (elf_check_const_displacement(&interp_params.hdr)) 295 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP; 296 297 /* flush all traces of the currently running executable */ 298 retval = flush_old_exec(bprm); 299 if (retval) 300 goto error; 301 302 /* there's now no turning back... the old userspace image is dead, 303 * defunct, deceased, etc. after this point we have to exit via 304 * error_kill */ 305 set_personality(PER_LINUX_FDPIC); 306 set_binfmt(&elf_fdpic_format); 307 308 current->mm->start_code = 0; 309 current->mm->end_code = 0; 310 current->mm->start_stack = 0; 311 current->mm->start_data = 0; 312 current->mm->end_data = 0; 313 current->mm->context.exec_fdpic_loadmap = 0; 314 current->mm->context.interp_fdpic_loadmap = 0; 315 316 current->flags &= ~PF_FORKNOEXEC; 317 318 #ifdef CONFIG_MMU 319 elf_fdpic_arch_lay_out_mm(&exec_params, 320 &interp_params, 321 ¤t->mm->start_stack, 322 ¤t->mm->start_brk); 323 324 retval = setup_arg_pages(bprm, current->mm->start_stack, 325 executable_stack); 326 if (retval < 0) { 327 send_sig(SIGKILL, current, 0); 328 goto error_kill; 329 } 330 #endif 331 332 /* load the executable and interpreter into memory */ 333 retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm, 334 "executable"); 335 if (retval < 0) 336 goto error_kill; 337 338 if (interpreter_name) { 339 retval = elf_fdpic_map_file(&interp_params, interpreter, 340 current->mm, "interpreter"); 341 if (retval < 0) { 342 printk(KERN_ERR "Unable to load interpreter\n"); 343 goto error_kill; 344 } 345 346 allow_write_access(interpreter); 347 fput(interpreter); 348 interpreter = NULL; 349 } 350 351 #ifdef CONFIG_MMU 352 if (!current->mm->start_brk) 353 current->mm->start_brk = current->mm->end_data; 354 355 current->mm->brk = current->mm->start_brk = 356 PAGE_ALIGN(current->mm->start_brk); 357 358 #else 359 /* create a stack and brk area big enough for everyone 360 * - the brk heap starts at the bottom and works up 361 * - the stack starts at the top and works down 362 */ 363 stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK; 364 if (stack_size < PAGE_SIZE * 2) 365 stack_size = PAGE_SIZE * 2; 366 367 down_write(¤t->mm->mmap_sem); 368 current->mm->start_brk = do_mmap(NULL, 0, stack_size, 369 PROT_READ | PROT_WRITE | PROT_EXEC, 370 MAP_PRIVATE | MAP_ANON | MAP_GROWSDOWN, 371 0); 372 373 if (IS_ERR_VALUE(current->mm->start_brk)) { 374 up_write(¤t->mm->mmap_sem); 375 retval = current->mm->start_brk; 376 current->mm->start_brk = 0; 377 goto error_kill; 378 } 379 380 /* expand the stack mapping to use up the entire allocation granule */ 381 fullsize = ksize((char *) current->mm->start_brk); 382 if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size, 383 fullsize, 0, 0))) 384 stack_size = fullsize; 385 up_write(¤t->mm->mmap_sem); 386 387 current->mm->brk = current->mm->start_brk; 388 current->mm->context.end_brk = current->mm->start_brk; 389 current->mm->context.end_brk += 390 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0; 391 current->mm->start_stack = current->mm->start_brk + stack_size; 392 #endif 393 394 compute_creds(bprm); 395 current->flags &= ~PF_FORKNOEXEC; 396 if (create_elf_fdpic_tables(bprm, current->mm, 397 &exec_params, &interp_params) < 0) 398 goto error_kill; 399 400 kdebug("- start_code %lx", current->mm->start_code); 401 kdebug("- end_code %lx", current->mm->end_code); 402 kdebug("- start_data %lx", current->mm->start_data); 403 kdebug("- end_data %lx", current->mm->end_data); 404 kdebug("- start_brk %lx", current->mm->start_brk); 405 kdebug("- brk %lx", current->mm->brk); 406 kdebug("- start_stack %lx", current->mm->start_stack); 407 408 #ifdef ELF_FDPIC_PLAT_INIT 409 /* 410 * The ABI may specify that certain registers be set up in special 411 * ways (on i386 %edx is the address of a DT_FINI function, for 412 * example. This macro performs whatever initialization to 413 * the regs structure is required. 414 */ 415 dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr; 416 ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr, 417 dynaddr); 418 #endif 419 420 /* everything is now ready... get the userspace context ready to roll */ 421 entryaddr = interp_params.entry_addr ?: exec_params.entry_addr; 422 start_thread(regs, entryaddr, current->mm->start_stack); 423 424 if (unlikely(current->ptrace & PT_PTRACED)) { 425 if (current->ptrace & PT_TRACE_EXEC) 426 ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP); 427 else 428 send_sig(SIGTRAP, current, 0); 429 } 430 431 retval = 0; 432 433 error: 434 if (interpreter) { 435 allow_write_access(interpreter); 436 fput(interpreter); 437 } 438 kfree(interpreter_name); 439 kfree(exec_params.phdrs); 440 kfree(exec_params.loadmap); 441 kfree(interp_params.phdrs); 442 kfree(interp_params.loadmap); 443 return retval; 444 445 /* unrecoverable error - kill the process */ 446 error_kill: 447 send_sig(SIGSEGV, current, 0); 448 goto error; 449 450 } 451 452 /*****************************************************************************/ 453 /* 454 * present useful information to the program 455 */ 456 static int create_elf_fdpic_tables(struct linux_binprm *bprm, 457 struct mm_struct *mm, 458 struct elf_fdpic_params *exec_params, 459 struct elf_fdpic_params *interp_params) 460 { 461 unsigned long sp, csp, nitems; 462 elf_caddr_t __user *argv, *envp; 463 size_t platform_len = 0, len; 464 char *k_platform; 465 char __user *u_platform, *p; 466 long hwcap; 467 int loop; 468 469 /* we're going to shovel a whole load of stuff onto the stack */ 470 #ifdef CONFIG_MMU 471 sp = bprm->p; 472 #else 473 sp = mm->start_stack; 474 475 /* stack the program arguments and environment */ 476 if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0) 477 return -EFAULT; 478 #endif 479 480 /* get hold of platform and hardware capabilities masks for the machine 481 * we are running on. In some cases (Sparc), this info is impossible 482 * to get, in others (i386) it is merely difficult. 483 */ 484 hwcap = ELF_HWCAP; 485 k_platform = ELF_PLATFORM; 486 u_platform = NULL; 487 488 if (k_platform) { 489 platform_len = strlen(k_platform) + 1; 490 sp -= platform_len; 491 u_platform = (char __user *) sp; 492 if (__copy_to_user(u_platform, k_platform, platform_len) != 0) 493 return -EFAULT; 494 } 495 496 #if defined(__i386__) && defined(CONFIG_SMP) 497 /* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions 498 * by the processes running on the same package. One thing we can do is 499 * to shuffle the initial stack for them. 500 * 501 * the conditionals here are unneeded, but kept in to make the code 502 * behaviour the same as pre change unless we have hyperthreaded 503 * processors. This keeps Mr Marcelo Person happier but should be 504 * removed for 2.5 505 */ 506 if (smp_num_siblings > 1) 507 sp = sp - ((current->pid % 64) << 7); 508 #endif 509 510 sp &= ~7UL; 511 512 /* stack the load map(s) */ 513 len = sizeof(struct elf32_fdpic_loadmap); 514 len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs; 515 sp = (sp - len) & ~7UL; 516 exec_params->map_addr = sp; 517 518 if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0) 519 return -EFAULT; 520 521 current->mm->context.exec_fdpic_loadmap = (unsigned long) sp; 522 523 if (interp_params->loadmap) { 524 len = sizeof(struct elf32_fdpic_loadmap); 525 len += sizeof(struct elf32_fdpic_loadseg) * 526 interp_params->loadmap->nsegs; 527 sp = (sp - len) & ~7UL; 528 interp_params->map_addr = sp; 529 530 if (copy_to_user((void __user *) sp, interp_params->loadmap, 531 len) != 0) 532 return -EFAULT; 533 534 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp; 535 } 536 537 /* force 16 byte _final_ alignment here for generality */ 538 #define DLINFO_ITEMS 13 539 540 nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0); 541 #ifdef DLINFO_ARCH_ITEMS 542 nitems += DLINFO_ARCH_ITEMS; 543 #endif 544 545 csp = sp; 546 sp -= nitems * 2 * sizeof(unsigned long); 547 sp -= (bprm->envc + 1) * sizeof(char *); /* envv[] */ 548 sp -= (bprm->argc + 1) * sizeof(char *); /* argv[] */ 549 sp -= 1 * sizeof(unsigned long); /* argc */ 550 551 csp -= sp & 15UL; 552 sp -= sp & 15UL; 553 554 /* put the ELF interpreter info on the stack */ 555 #define NEW_AUX_ENT(nr, id, val) \ 556 do { \ 557 struct { unsigned long _id, _val; } __user *ent; \ 558 \ 559 ent = (void __user *) csp; \ 560 __put_user((id), &ent[nr]._id); \ 561 __put_user((val), &ent[nr]._val); \ 562 } while (0) 563 564 csp -= 2 * sizeof(unsigned long); 565 NEW_AUX_ENT(0, AT_NULL, 0); 566 if (k_platform) { 567 csp -= 2 * sizeof(unsigned long); 568 NEW_AUX_ENT(0, AT_PLATFORM, 569 (elf_addr_t) (unsigned long) u_platform); 570 } 571 572 csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long); 573 NEW_AUX_ENT( 0, AT_HWCAP, hwcap); 574 NEW_AUX_ENT( 1, AT_PAGESZ, PAGE_SIZE); 575 NEW_AUX_ENT( 2, AT_CLKTCK, CLOCKS_PER_SEC); 576 NEW_AUX_ENT( 3, AT_PHDR, exec_params->ph_addr); 577 NEW_AUX_ENT( 4, AT_PHENT, sizeof(struct elf_phdr)); 578 NEW_AUX_ENT( 5, AT_PHNUM, exec_params->hdr.e_phnum); 579 NEW_AUX_ENT( 6, AT_BASE, interp_params->elfhdr_addr); 580 NEW_AUX_ENT( 7, AT_FLAGS, 0); 581 NEW_AUX_ENT( 8, AT_ENTRY, exec_params->entry_addr); 582 NEW_AUX_ENT( 9, AT_UID, (elf_addr_t) current->uid); 583 NEW_AUX_ENT(10, AT_EUID, (elf_addr_t) current->euid); 584 NEW_AUX_ENT(11, AT_GID, (elf_addr_t) current->gid); 585 NEW_AUX_ENT(12, AT_EGID, (elf_addr_t) current->egid); 586 587 #ifdef ARCH_DLINFO 588 /* ARCH_DLINFO must come last so platform specific code can enforce 589 * special alignment requirements on the AUXV if necessary (eg. PPC). 590 */ 591 ARCH_DLINFO; 592 #endif 593 #undef NEW_AUX_ENT 594 595 /* allocate room for argv[] and envv[] */ 596 csp -= (bprm->envc + 1) * sizeof(elf_caddr_t); 597 envp = (elf_caddr_t __user *) csp; 598 csp -= (bprm->argc + 1) * sizeof(elf_caddr_t); 599 argv = (elf_caddr_t __user *) csp; 600 601 /* stack argc */ 602 csp -= sizeof(unsigned long); 603 __put_user(bprm->argc, (unsigned long __user *) csp); 604 605 BUG_ON(csp != sp); 606 607 /* fill in the argv[] array */ 608 #ifdef CONFIG_MMU 609 current->mm->arg_start = bprm->p; 610 #else 611 current->mm->arg_start = current->mm->start_stack - 612 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p); 613 #endif 614 615 p = (char __user *) current->mm->arg_start; 616 for (loop = bprm->argc; loop > 0; loop--) { 617 __put_user((elf_caddr_t) p, argv++); 618 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES); 619 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES) 620 return -EINVAL; 621 p += len; 622 } 623 __put_user(NULL, argv); 624 current->mm->arg_end = (unsigned long) p; 625 626 /* fill in the envv[] array */ 627 current->mm->env_start = (unsigned long) p; 628 for (loop = bprm->envc; loop > 0; loop--) { 629 __put_user((elf_caddr_t)(unsigned long) p, envp++); 630 len = strnlen_user(p, PAGE_SIZE * MAX_ARG_PAGES); 631 if (!len || len > PAGE_SIZE * MAX_ARG_PAGES) 632 return -EINVAL; 633 p += len; 634 } 635 __put_user(NULL, envp); 636 current->mm->env_end = (unsigned long) p; 637 638 mm->start_stack = (unsigned long) sp; 639 return 0; 640 } 641 642 /*****************************************************************************/ 643 /* 644 * transfer the program arguments and environment from the holding pages onto 645 * the stack 646 */ 647 #ifndef CONFIG_MMU 648 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm, 649 unsigned long *_sp) 650 { 651 unsigned long index, stop, sp; 652 char *src; 653 int ret = 0; 654 655 stop = bprm->p >> PAGE_SHIFT; 656 sp = *_sp; 657 658 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) { 659 src = kmap(bprm->page[index]); 660 sp -= PAGE_SIZE; 661 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0) 662 ret = -EFAULT; 663 kunmap(bprm->page[index]); 664 if (ret < 0) 665 goto out; 666 } 667 668 *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15; 669 670 out: 671 return ret; 672 } 673 #endif 674 675 /*****************************************************************************/ 676 /* 677 * load the appropriate binary image (executable or interpreter) into memory 678 * - we assume no MMU is available 679 * - if no other PIC bits are set in params->hdr->e_flags 680 * - we assume that the LOADable segments in the binary are independently relocatable 681 * - we assume R/O executable segments are shareable 682 * - else 683 * - we assume the loadable parts of the image to require fixed displacement 684 * - the image is not shareable 685 */ 686 static int elf_fdpic_map_file(struct elf_fdpic_params *params, 687 struct file *file, 688 struct mm_struct *mm, 689 const char *what) 690 { 691 struct elf32_fdpic_loadmap *loadmap; 692 #ifdef CONFIG_MMU 693 struct elf32_fdpic_loadseg *mseg; 694 #endif 695 struct elf32_fdpic_loadseg *seg; 696 struct elf32_phdr *phdr; 697 unsigned long load_addr, stop; 698 unsigned nloads, tmp; 699 size_t size; 700 int loop, ret; 701 702 /* allocate a load map table */ 703 nloads = 0; 704 for (loop = 0; loop < params->hdr.e_phnum; loop++) 705 if (params->phdrs[loop].p_type == PT_LOAD) 706 nloads++; 707 708 if (nloads == 0) 709 return -ELIBBAD; 710 711 size = sizeof(*loadmap) + nloads * sizeof(*seg); 712 loadmap = kmalloc(size, GFP_KERNEL); 713 if (!loadmap) 714 return -ENOMEM; 715 716 params->loadmap = loadmap; 717 memset(loadmap, 0, size); 718 719 loadmap->version = ELF32_FDPIC_LOADMAP_VERSION; 720 loadmap->nsegs = nloads; 721 722 load_addr = params->load_addr; 723 seg = loadmap->segs; 724 725 /* map the requested LOADs into the memory space */ 726 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 727 case ELF_FDPIC_FLAG_CONSTDISP: 728 case ELF_FDPIC_FLAG_CONTIGUOUS: 729 #ifndef CONFIG_MMU 730 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm); 731 if (ret < 0) 732 return ret; 733 break; 734 #endif 735 default: 736 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm); 737 if (ret < 0) 738 return ret; 739 break; 740 } 741 742 /* map the entry point */ 743 if (params->hdr.e_entry) { 744 seg = loadmap->segs; 745 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 746 if (params->hdr.e_entry >= seg->p_vaddr && 747 params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) { 748 params->entry_addr = 749 (params->hdr.e_entry - seg->p_vaddr) + 750 seg->addr; 751 break; 752 } 753 } 754 } 755 756 /* determine where the program header table has wound up if mapped */ 757 stop = params->hdr.e_phoff; 758 stop += params->hdr.e_phnum * sizeof (struct elf_phdr); 759 phdr = params->phdrs; 760 761 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 762 if (phdr->p_type != PT_LOAD) 763 continue; 764 765 if (phdr->p_offset > params->hdr.e_phoff || 766 phdr->p_offset + phdr->p_filesz < stop) 767 continue; 768 769 seg = loadmap->segs; 770 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 771 if (phdr->p_vaddr >= seg->p_vaddr && 772 phdr->p_vaddr + phdr->p_filesz <= 773 seg->p_vaddr + seg->p_memsz) { 774 params->ph_addr = 775 (phdr->p_vaddr - seg->p_vaddr) + 776 seg->addr + 777 params->hdr.e_phoff - phdr->p_offset; 778 break; 779 } 780 } 781 break; 782 } 783 784 /* determine where the dynamic section has wound up if there is one */ 785 phdr = params->phdrs; 786 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 787 if (phdr->p_type != PT_DYNAMIC) 788 continue; 789 790 seg = loadmap->segs; 791 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) { 792 if (phdr->p_vaddr >= seg->p_vaddr && 793 phdr->p_vaddr + phdr->p_memsz <= 794 seg->p_vaddr + seg->p_memsz) { 795 params->dynamic_addr = 796 (phdr->p_vaddr - seg->p_vaddr) + 797 seg->addr; 798 799 /* check the dynamic section contains at least 800 * one item, and that the last item is a NULL 801 * entry */ 802 if (phdr->p_memsz == 0 || 803 phdr->p_memsz % sizeof(Elf32_Dyn) != 0) 804 goto dynamic_error; 805 806 tmp = phdr->p_memsz / sizeof(Elf32_Dyn); 807 if (((Elf32_Dyn *) 808 params->dynamic_addr)[tmp - 1].d_tag != 0) 809 goto dynamic_error; 810 break; 811 } 812 } 813 break; 814 } 815 816 /* now elide adjacent segments in the load map on MMU linux 817 * - on uClinux the holes between may actually be filled with system 818 * stuff or stuff from other processes 819 */ 820 #ifdef CONFIG_MMU 821 nloads = loadmap->nsegs; 822 mseg = loadmap->segs; 823 seg = mseg + 1; 824 for (loop = 1; loop < nloads; loop++) { 825 /* see if we have a candidate for merging */ 826 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) { 827 load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz); 828 if (load_addr == (seg->addr & PAGE_MASK)) { 829 mseg->p_memsz += 830 load_addr - 831 (mseg->addr + mseg->p_memsz); 832 mseg->p_memsz += seg->addr & ~PAGE_MASK; 833 mseg->p_memsz += seg->p_memsz; 834 loadmap->nsegs--; 835 continue; 836 } 837 } 838 839 mseg++; 840 if (mseg != seg) 841 *mseg = *seg; 842 } 843 #endif 844 845 kdebug("Mapped Object [%s]:", what); 846 kdebug("- elfhdr : %lx", params->elfhdr_addr); 847 kdebug("- entry : %lx", params->entry_addr); 848 kdebug("- PHDR[] : %lx", params->ph_addr); 849 kdebug("- DYNAMIC[]: %lx", params->dynamic_addr); 850 seg = loadmap->segs; 851 for (loop = 0; loop < loadmap->nsegs; loop++, seg++) 852 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]", 853 loop, 854 seg->addr, seg->addr + seg->p_memsz - 1, 855 seg->p_vaddr, seg->p_memsz); 856 857 return 0; 858 859 dynamic_error: 860 printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n", 861 what, file->f_dentry->d_inode->i_ino); 862 return -ELIBBAD; 863 } 864 865 /*****************************************************************************/ 866 /* 867 * map a file with constant displacement under uClinux 868 */ 869 #ifndef CONFIG_MMU 870 static int elf_fdpic_map_file_constdisp_on_uclinux( 871 struct elf_fdpic_params *params, 872 struct file *file, 873 struct mm_struct *mm) 874 { 875 struct elf32_fdpic_loadseg *seg; 876 struct elf32_phdr *phdr; 877 unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags; 878 loff_t fpos; 879 int loop, ret; 880 881 load_addr = params->load_addr; 882 seg = params->loadmap->segs; 883 884 /* determine the bounds of the contiguous overall allocation we must 885 * make */ 886 phdr = params->phdrs; 887 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 888 if (params->phdrs[loop].p_type != PT_LOAD) 889 continue; 890 891 if (base > phdr->p_vaddr) 892 base = phdr->p_vaddr; 893 if (top < phdr->p_vaddr + phdr->p_memsz) 894 top = phdr->p_vaddr + phdr->p_memsz; 895 } 896 897 /* allocate one big anon block for everything */ 898 mflags = MAP_PRIVATE; 899 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 900 mflags |= MAP_EXECUTABLE; 901 902 down_write(&mm->mmap_sem); 903 maddr = do_mmap(NULL, load_addr, top - base, 904 PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0); 905 up_write(&mm->mmap_sem); 906 if (IS_ERR_VALUE(maddr)) 907 return (int) maddr; 908 909 if (load_addr != 0) 910 load_addr += PAGE_ALIGN(top - base); 911 912 /* and then load the file segments into it */ 913 phdr = params->phdrs; 914 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 915 if (params->phdrs[loop].p_type != PT_LOAD) 916 continue; 917 918 fpos = phdr->p_offset; 919 920 seg->addr = maddr + (phdr->p_vaddr - base); 921 seg->p_vaddr = phdr->p_vaddr; 922 seg->p_memsz = phdr->p_memsz; 923 924 ret = file->f_op->read(file, (void *) seg->addr, 925 phdr->p_filesz, &fpos); 926 if (ret < 0) 927 return ret; 928 929 /* map the ELF header address if in this segment */ 930 if (phdr->p_offset == 0) 931 params->elfhdr_addr = seg->addr; 932 933 /* clear any space allocated but not loaded */ 934 if (phdr->p_filesz < phdr->p_memsz) 935 clear_user((void *) (seg->addr + phdr->p_filesz), 936 phdr->p_memsz - phdr->p_filesz); 937 938 if (mm) { 939 if (phdr->p_flags & PF_X) { 940 mm->start_code = seg->addr; 941 mm->end_code = seg->addr + phdr->p_memsz; 942 } else if (!mm->start_data) { 943 mm->start_data = seg->addr; 944 #ifndef CONFIG_MMU 945 mm->end_data = seg->addr + phdr->p_memsz; 946 #endif 947 } 948 949 #ifdef CONFIG_MMU 950 if (seg->addr + phdr->p_memsz > mm->end_data) 951 mm->end_data = seg->addr + phdr->p_memsz; 952 #endif 953 } 954 955 seg++; 956 } 957 958 return 0; 959 } 960 #endif 961 962 /*****************************************************************************/ 963 /* 964 * map a binary by direct mmap() of the individual PT_LOAD segments 965 */ 966 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params, 967 struct file *file, 968 struct mm_struct *mm) 969 { 970 struct elf32_fdpic_loadseg *seg; 971 struct elf32_phdr *phdr; 972 unsigned long load_addr, delta_vaddr; 973 int loop, dvset; 974 975 load_addr = params->load_addr; 976 delta_vaddr = 0; 977 dvset = 0; 978 979 seg = params->loadmap->segs; 980 981 /* deal with each load segment separately */ 982 phdr = params->phdrs; 983 for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) { 984 unsigned long maddr, disp, excess, excess1; 985 int prot = 0, flags; 986 987 if (phdr->p_type != PT_LOAD) 988 continue; 989 990 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx", 991 (unsigned long) phdr->p_vaddr, 992 (unsigned long) phdr->p_offset, 993 (unsigned long) phdr->p_filesz, 994 (unsigned long) phdr->p_memsz); 995 996 /* determine the mapping parameters */ 997 if (phdr->p_flags & PF_R) prot |= PROT_READ; 998 if (phdr->p_flags & PF_W) prot |= PROT_WRITE; 999 if (phdr->p_flags & PF_X) prot |= PROT_EXEC; 1000 1001 flags = MAP_PRIVATE | MAP_DENYWRITE; 1002 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE) 1003 flags |= MAP_EXECUTABLE; 1004 1005 maddr = 0; 1006 1007 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) { 1008 case ELF_FDPIC_FLAG_INDEPENDENT: 1009 /* PT_LOADs are independently locatable */ 1010 break; 1011 1012 case ELF_FDPIC_FLAG_HONOURVADDR: 1013 /* the specified virtual address must be honoured */ 1014 maddr = phdr->p_vaddr; 1015 flags |= MAP_FIXED; 1016 break; 1017 1018 case ELF_FDPIC_FLAG_CONSTDISP: 1019 /* constant displacement 1020 * - can be mapped anywhere, but must be mapped as a 1021 * unit 1022 */ 1023 if (!dvset) { 1024 maddr = load_addr; 1025 delta_vaddr = phdr->p_vaddr; 1026 dvset = 1; 1027 } else { 1028 maddr = load_addr + phdr->p_vaddr - delta_vaddr; 1029 flags |= MAP_FIXED; 1030 } 1031 break; 1032 1033 case ELF_FDPIC_FLAG_CONTIGUOUS: 1034 /* contiguity handled later */ 1035 break; 1036 1037 default: 1038 BUG(); 1039 } 1040 1041 maddr &= PAGE_MASK; 1042 1043 /* create the mapping */ 1044 disp = phdr->p_vaddr & ~PAGE_MASK; 1045 down_write(&mm->mmap_sem); 1046 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags, 1047 phdr->p_offset - disp); 1048 up_write(&mm->mmap_sem); 1049 1050 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx", 1051 loop, phdr->p_memsz + disp, prot, flags, 1052 phdr->p_offset - disp, maddr); 1053 1054 if (IS_ERR_VALUE(maddr)) 1055 return (int) maddr; 1056 1057 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) == 1058 ELF_FDPIC_FLAG_CONTIGUOUS) 1059 load_addr += PAGE_ALIGN(phdr->p_memsz + disp); 1060 1061 seg->addr = maddr + disp; 1062 seg->p_vaddr = phdr->p_vaddr; 1063 seg->p_memsz = phdr->p_memsz; 1064 1065 /* map the ELF header address if in this segment */ 1066 if (phdr->p_offset == 0) 1067 params->elfhdr_addr = seg->addr; 1068 1069 /* clear the bit between beginning of mapping and beginning of 1070 * PT_LOAD */ 1071 if (prot & PROT_WRITE && disp > 0) { 1072 kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp); 1073 clear_user((void __user *) maddr, disp); 1074 maddr += disp; 1075 } 1076 1077 /* clear any space allocated but not loaded 1078 * - on uClinux we can just clear the lot 1079 * - on MMU linux we'll get a SIGBUS beyond the last page 1080 * extant in the file 1081 */ 1082 excess = phdr->p_memsz - phdr->p_filesz; 1083 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK); 1084 1085 #ifdef CONFIG_MMU 1086 if (excess > excess1) { 1087 unsigned long xaddr = maddr + phdr->p_filesz + excess1; 1088 unsigned long xmaddr; 1089 1090 flags |= MAP_FIXED | MAP_ANONYMOUS; 1091 down_write(&mm->mmap_sem); 1092 xmaddr = do_mmap(NULL, xaddr, excess - excess1, 1093 prot, flags, 0); 1094 up_write(&mm->mmap_sem); 1095 1096 kdebug("mmap[%d] <anon>" 1097 " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx", 1098 loop, xaddr, excess - excess1, prot, flags, 1099 xmaddr); 1100 1101 if (xmaddr != xaddr) 1102 return -ENOMEM; 1103 } 1104 1105 if (prot & PROT_WRITE && excess1 > 0) { 1106 kdebug("clear[%d] ad=%lx sz=%lx", 1107 loop, maddr + phdr->p_filesz, excess1); 1108 clear_user((void __user *) maddr + phdr->p_filesz, 1109 excess1); 1110 } 1111 1112 #else 1113 if (excess > 0) { 1114 kdebug("clear[%d] ad=%lx sz=%lx", 1115 loop, maddr + phdr->p_filesz, excess); 1116 clear_user((void *) maddr + phdr->p_filesz, excess); 1117 } 1118 #endif 1119 1120 if (mm) { 1121 if (phdr->p_flags & PF_X) { 1122 mm->start_code = maddr; 1123 mm->end_code = maddr + phdr->p_memsz; 1124 } else if (!mm->start_data) { 1125 mm->start_data = maddr; 1126 mm->end_data = maddr + phdr->p_memsz; 1127 } 1128 } 1129 1130 seg++; 1131 } 1132 1133 return 0; 1134 } 1135 1136 /*****************************************************************************/ 1137 /* 1138 * ELF-FDPIC core dumper 1139 * 1140 * Modelled on fs/exec.c:aout_core_dump() 1141 * Jeremy Fitzhardinge <jeremy@sw.oz.au> 1142 * 1143 * Modelled on fs/binfmt_elf.c core dumper 1144 */ 1145 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE) 1146 1147 /* 1148 * These are the only things you should do on a core-file: use only these 1149 * functions to write out all the necessary info. 1150 */ 1151 static int dump_write(struct file *file, const void *addr, int nr) 1152 { 1153 return file->f_op->write(file, addr, nr, &file->f_pos) == nr; 1154 } 1155 1156 static int dump_seek(struct file *file, loff_t off) 1157 { 1158 if (file->f_op->llseek) { 1159 if (file->f_op->llseek(file, off, SEEK_SET) != off) 1160 return 0; 1161 } else { 1162 file->f_pos = off; 1163 } 1164 return 1; 1165 } 1166 1167 /* 1168 * Decide whether a segment is worth dumping; default is yes to be 1169 * sure (missing info is worse than too much; etc). 1170 * Personally I'd include everything, and use the coredump limit... 1171 * 1172 * I think we should skip something. But I am not sure how. H.J. 1173 */ 1174 static int maydump(struct vm_area_struct *vma) 1175 { 1176 /* Do not dump I/O mapped devices or special mappings */ 1177 if (vma->vm_flags & (VM_IO | VM_RESERVED)) { 1178 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags); 1179 return 0; 1180 } 1181 1182 /* If we may not read the contents, don't allow us to dump 1183 * them either. "dump_write()" can't handle it anyway. 1184 */ 1185 if (!(vma->vm_flags & VM_READ)) { 1186 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags); 1187 return 0; 1188 } 1189 1190 /* Dump shared memory only if mapped from an anonymous file. */ 1191 if (vma->vm_flags & VM_SHARED) { 1192 if (vma->vm_file->f_dentry->d_inode->i_nlink == 0) { 1193 kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags); 1194 return 1; 1195 } 1196 1197 kdcore("%08lx: %08lx: no (share)", vma->vm_start, vma->vm_flags); 1198 return 0; 1199 } 1200 1201 #ifdef CONFIG_MMU 1202 /* If it hasn't been written to, don't write it out */ 1203 if (!vma->anon_vma) { 1204 kdcore("%08lx: %08lx: no (!anon)", vma->vm_start, vma->vm_flags); 1205 return 0; 1206 } 1207 #endif 1208 1209 kdcore("%08lx: %08lx: yes", vma->vm_start, vma->vm_flags); 1210 return 1; 1211 } 1212 1213 /* An ELF note in memory */ 1214 struct memelfnote 1215 { 1216 const char *name; 1217 int type; 1218 unsigned int datasz; 1219 void *data; 1220 }; 1221 1222 static int notesize(struct memelfnote *en) 1223 { 1224 int sz; 1225 1226 sz = sizeof(struct elf_note); 1227 sz += roundup(strlen(en->name) + 1, 4); 1228 sz += roundup(en->datasz, 4); 1229 1230 return sz; 1231 } 1232 1233 /* #define DEBUG */ 1234 1235 #define DUMP_WRITE(addr, nr) \ 1236 do { if (!dump_write(file, (addr), (nr))) return 0; } while(0) 1237 #define DUMP_SEEK(off) \ 1238 do { if (!dump_seek(file, (off))) return 0; } while(0) 1239 1240 static int writenote(struct memelfnote *men, struct file *file) 1241 { 1242 struct elf_note en; 1243 1244 en.n_namesz = strlen(men->name) + 1; 1245 en.n_descsz = men->datasz; 1246 en.n_type = men->type; 1247 1248 DUMP_WRITE(&en, sizeof(en)); 1249 DUMP_WRITE(men->name, en.n_namesz); 1250 /* XXX - cast from long long to long to avoid need for libgcc.a */ 1251 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */ 1252 DUMP_WRITE(men->data, men->datasz); 1253 DUMP_SEEK(roundup((unsigned long)file->f_pos, 4)); /* XXX */ 1254 1255 return 1; 1256 } 1257 #undef DUMP_WRITE 1258 #undef DUMP_SEEK 1259 1260 #define DUMP_WRITE(addr, nr) \ 1261 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \ 1262 goto end_coredump; 1263 #define DUMP_SEEK(off) \ 1264 if (!dump_seek(file, (off))) \ 1265 goto end_coredump; 1266 1267 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs) 1268 { 1269 memcpy(elf->e_ident, ELFMAG, SELFMAG); 1270 elf->e_ident[EI_CLASS] = ELF_CLASS; 1271 elf->e_ident[EI_DATA] = ELF_DATA; 1272 elf->e_ident[EI_VERSION] = EV_CURRENT; 1273 elf->e_ident[EI_OSABI] = ELF_OSABI; 1274 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); 1275 1276 elf->e_type = ET_CORE; 1277 elf->e_machine = ELF_ARCH; 1278 elf->e_version = EV_CURRENT; 1279 elf->e_entry = 0; 1280 elf->e_phoff = sizeof(struct elfhdr); 1281 elf->e_shoff = 0; 1282 elf->e_flags = ELF_FDPIC_CORE_EFLAGS; 1283 elf->e_ehsize = sizeof(struct elfhdr); 1284 elf->e_phentsize = sizeof(struct elf_phdr); 1285 elf->e_phnum = segs; 1286 elf->e_shentsize = 0; 1287 elf->e_shnum = 0; 1288 elf->e_shstrndx = 0; 1289 return; 1290 } 1291 1292 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset) 1293 { 1294 phdr->p_type = PT_NOTE; 1295 phdr->p_offset = offset; 1296 phdr->p_vaddr = 0; 1297 phdr->p_paddr = 0; 1298 phdr->p_filesz = sz; 1299 phdr->p_memsz = 0; 1300 phdr->p_flags = 0; 1301 phdr->p_align = 0; 1302 return; 1303 } 1304 1305 static inline void fill_note(struct memelfnote *note, const char *name, int type, 1306 unsigned int sz, void *data) 1307 { 1308 note->name = name; 1309 note->type = type; 1310 note->datasz = sz; 1311 note->data = data; 1312 return; 1313 } 1314 1315 /* 1316 * fill up all the fields in prstatus from the given task struct, except 1317 * registers which need to be filled up seperately. 1318 */ 1319 static void fill_prstatus(struct elf_prstatus *prstatus, 1320 struct task_struct *p, long signr) 1321 { 1322 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr; 1323 prstatus->pr_sigpend = p->pending.signal.sig[0]; 1324 prstatus->pr_sighold = p->blocked.sig[0]; 1325 prstatus->pr_pid = p->pid; 1326 prstatus->pr_ppid = p->parent->pid; 1327 prstatus->pr_pgrp = process_group(p); 1328 prstatus->pr_sid = p->signal->session; 1329 if (thread_group_leader(p)) { 1330 /* 1331 * This is the record for the group leader. Add in the 1332 * cumulative times of previous dead threads. This total 1333 * won't include the time of each live thread whose state 1334 * is included in the core dump. The final total reported 1335 * to our parent process when it calls wait4 will include 1336 * those sums as well as the little bit more time it takes 1337 * this and each other thread to finish dying after the 1338 * core dump synchronization phase. 1339 */ 1340 cputime_to_timeval(cputime_add(p->utime, p->signal->utime), 1341 &prstatus->pr_utime); 1342 cputime_to_timeval(cputime_add(p->stime, p->signal->stime), 1343 &prstatus->pr_stime); 1344 } else { 1345 cputime_to_timeval(p->utime, &prstatus->pr_utime); 1346 cputime_to_timeval(p->stime, &prstatus->pr_stime); 1347 } 1348 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime); 1349 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime); 1350 1351 prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap; 1352 prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap; 1353 } 1354 1355 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p, 1356 struct mm_struct *mm) 1357 { 1358 unsigned int i, len; 1359 1360 /* first copy the parameters from user space */ 1361 memset(psinfo, 0, sizeof(struct elf_prpsinfo)); 1362 1363 len = mm->arg_end - mm->arg_start; 1364 if (len >= ELF_PRARGSZ) 1365 len = ELF_PRARGSZ - 1; 1366 if (copy_from_user(&psinfo->pr_psargs, 1367 (const char __user *) mm->arg_start, len)) 1368 return -EFAULT; 1369 for (i = 0; i < len; i++) 1370 if (psinfo->pr_psargs[i] == 0) 1371 psinfo->pr_psargs[i] = ' '; 1372 psinfo->pr_psargs[len] = 0; 1373 1374 psinfo->pr_pid = p->pid; 1375 psinfo->pr_ppid = p->parent->pid; 1376 psinfo->pr_pgrp = process_group(p); 1377 psinfo->pr_sid = p->signal->session; 1378 1379 i = p->state ? ffz(~p->state) + 1 : 0; 1380 psinfo->pr_state = i; 1381 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i]; 1382 psinfo->pr_zomb = psinfo->pr_sname == 'Z'; 1383 psinfo->pr_nice = task_nice(p); 1384 psinfo->pr_flag = p->flags; 1385 SET_UID(psinfo->pr_uid, p->uid); 1386 SET_GID(psinfo->pr_gid, p->gid); 1387 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname)); 1388 1389 return 0; 1390 } 1391 1392 /* Here is the structure in which status of each thread is captured. */ 1393 struct elf_thread_status 1394 { 1395 struct list_head list; 1396 struct elf_prstatus prstatus; /* NT_PRSTATUS */ 1397 elf_fpregset_t fpu; /* NT_PRFPREG */ 1398 struct task_struct *thread; 1399 #ifdef ELF_CORE_COPY_XFPREGS 1400 elf_fpxregset_t xfpu; /* NT_PRXFPREG */ 1401 #endif 1402 struct memelfnote notes[3]; 1403 int num_notes; 1404 }; 1405 1406 /* 1407 * In order to add the specific thread information for the elf file format, 1408 * we need to keep a linked list of every thread's pr_status and then create 1409 * a single section for them in the final core file. 1410 */ 1411 static int elf_dump_thread_status(long signr, struct elf_thread_status *t) 1412 { 1413 struct task_struct *p = t->thread; 1414 int sz = 0; 1415 1416 t->num_notes = 0; 1417 1418 fill_prstatus(&t->prstatus, p, signr); 1419 elf_core_copy_task_regs(p, &t->prstatus.pr_reg); 1420 1421 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus), 1422 &t->prstatus); 1423 t->num_notes++; 1424 sz += notesize(&t->notes[0]); 1425 1426 t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu); 1427 if (t->prstatus.pr_fpvalid) { 1428 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu), 1429 &t->fpu); 1430 t->num_notes++; 1431 sz += notesize(&t->notes[1]); 1432 } 1433 1434 #ifdef ELF_CORE_COPY_XFPREGS 1435 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) { 1436 fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu), 1437 &t->xfpu); 1438 t->num_notes++; 1439 sz += notesize(&t->notes[2]); 1440 } 1441 #endif 1442 return sz; 1443 } 1444 1445 /* 1446 * dump the segments for an MMU process 1447 */ 1448 #ifdef CONFIG_MMU 1449 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm, 1450 size_t *size, unsigned long *limit) 1451 { 1452 struct vm_area_struct *vma; 1453 1454 for (vma = current->mm->mmap; vma; vma = vma->vm_next) { 1455 unsigned long addr; 1456 1457 if (!maydump(vma)) 1458 continue; 1459 1460 for (addr = vma->vm_start; 1461 addr < vma->vm_end; 1462 addr += PAGE_SIZE 1463 ) { 1464 struct vm_area_struct *vma; 1465 struct page *page; 1466 1467 if (get_user_pages(current, current->mm, addr, 1, 0, 1, 1468 &page, &vma) <= 0) { 1469 DUMP_SEEK(file->f_pos + PAGE_SIZE); 1470 } 1471 else if (page == ZERO_PAGE(addr)) { 1472 DUMP_SEEK(file->f_pos + PAGE_SIZE); 1473 page_cache_release(page); 1474 } 1475 else { 1476 void *kaddr; 1477 1478 flush_cache_page(vma, addr, page_to_pfn(page)); 1479 kaddr = kmap(page); 1480 if ((*size += PAGE_SIZE) > *limit || 1481 !dump_write(file, kaddr, PAGE_SIZE) 1482 ) { 1483 kunmap(page); 1484 page_cache_release(page); 1485 return -EIO; 1486 } 1487 kunmap(page); 1488 page_cache_release(page); 1489 } 1490 } 1491 } 1492 1493 return 0; 1494 1495 end_coredump: 1496 return -EFBIG; 1497 } 1498 #endif 1499 1500 /* 1501 * dump the segments for a NOMMU process 1502 */ 1503 #ifndef CONFIG_MMU 1504 static int elf_fdpic_dump_segments(struct file *file, struct mm_struct *mm, 1505 size_t *size, unsigned long *limit) 1506 { 1507 struct vm_list_struct *vml; 1508 1509 for (vml = current->mm->context.vmlist; vml; vml = vml->next) { 1510 struct vm_area_struct *vma = vml->vma; 1511 1512 if (!maydump(vma)) 1513 continue; 1514 1515 if ((*size += PAGE_SIZE) > *limit) 1516 return -EFBIG; 1517 1518 if (!dump_write(file, (void *) vma->vm_start, 1519 vma->vm_end - vma->vm_start)) 1520 return -EIO; 1521 } 1522 1523 return 0; 1524 } 1525 #endif 1526 1527 /* 1528 * Actual dumper 1529 * 1530 * This is a two-pass process; first we find the offsets of the bits, 1531 * and then they are actually written out. If we run out of core limit 1532 * we just truncate. 1533 */ 1534 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs, 1535 struct file *file) 1536 { 1537 #define NUM_NOTES 6 1538 int has_dumped = 0; 1539 mm_segment_t fs; 1540 int segs; 1541 size_t size = 0; 1542 int i; 1543 struct vm_area_struct *vma; 1544 struct elfhdr *elf = NULL; 1545 loff_t offset = 0, dataoff; 1546 unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur; 1547 int numnote; 1548 struct memelfnote *notes = NULL; 1549 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */ 1550 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */ 1551 struct task_struct *g, *p; 1552 LIST_HEAD(thread_list); 1553 struct list_head *t; 1554 elf_fpregset_t *fpu = NULL; 1555 #ifdef ELF_CORE_COPY_XFPREGS 1556 elf_fpxregset_t *xfpu = NULL; 1557 #endif 1558 int thread_status_size = 0; 1559 #ifndef CONFIG_MMU 1560 struct vm_list_struct *vml; 1561 #endif 1562 elf_addr_t *auxv; 1563 1564 /* 1565 * We no longer stop all VM operations. 1566 * 1567 * This is because those proceses that could possibly change map_count 1568 * or the mmap / vma pages are now blocked in do_exit on current 1569 * finishing this core dump. 1570 * 1571 * Only ptrace can touch these memory addresses, but it doesn't change 1572 * the map_count or the pages allocated. So no possibility of crashing 1573 * exists while dumping the mm->vm_next areas to the core file. 1574 */ 1575 1576 /* alloc memory for large data structures: too large to be on stack */ 1577 elf = kmalloc(sizeof(*elf), GFP_KERNEL); 1578 if (!elf) 1579 goto cleanup; 1580 prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL); 1581 if (!prstatus) 1582 goto cleanup; 1583 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL); 1584 if (!psinfo) 1585 goto cleanup; 1586 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL); 1587 if (!notes) 1588 goto cleanup; 1589 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL); 1590 if (!fpu) 1591 goto cleanup; 1592 #ifdef ELF_CORE_COPY_XFPREGS 1593 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL); 1594 if (!xfpu) 1595 goto cleanup; 1596 #endif 1597 1598 if (signr) { 1599 struct elf_thread_status *tmp; 1600 read_lock(&tasklist_lock); 1601 do_each_thread(g,p) 1602 if (current->mm == p->mm && current != p) { 1603 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC); 1604 if (!tmp) { 1605 read_unlock(&tasklist_lock); 1606 goto cleanup; 1607 } 1608 INIT_LIST_HEAD(&tmp->list); 1609 tmp->thread = p; 1610 list_add(&tmp->list, &thread_list); 1611 } 1612 while_each_thread(g,p); 1613 read_unlock(&tasklist_lock); 1614 list_for_each(t, &thread_list) { 1615 struct elf_thread_status *tmp; 1616 int sz; 1617 1618 tmp = list_entry(t, struct elf_thread_status, list); 1619 sz = elf_dump_thread_status(signr, tmp); 1620 thread_status_size += sz; 1621 } 1622 } 1623 1624 /* now collect the dump for the current */ 1625 fill_prstatus(prstatus, current, signr); 1626 elf_core_copy_regs(&prstatus->pr_reg, regs); 1627 1628 #ifdef CONFIG_MMU 1629 segs = current->mm->map_count; 1630 #else 1631 segs = 0; 1632 for (vml = current->mm->context.vmlist; vml; vml = vml->next) 1633 segs++; 1634 #endif 1635 #ifdef ELF_CORE_EXTRA_PHDRS 1636 segs += ELF_CORE_EXTRA_PHDRS; 1637 #endif 1638 1639 /* Set up header */ 1640 fill_elf_fdpic_header(elf, segs + 1); /* including notes section */ 1641 1642 has_dumped = 1; 1643 current->flags |= PF_DUMPCORE; 1644 1645 /* 1646 * Set up the notes in similar form to SVR4 core dumps made 1647 * with info from their /proc. 1648 */ 1649 1650 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus); 1651 fill_psinfo(psinfo, current->group_leader, current->mm); 1652 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo); 1653 1654 numnote = 2; 1655 1656 auxv = (elf_addr_t *) current->mm->saved_auxv; 1657 1658 i = 0; 1659 do 1660 i += 2; 1661 while (auxv[i - 2] != AT_NULL); 1662 fill_note(¬es[numnote++], "CORE", NT_AUXV, 1663 i * sizeof(elf_addr_t), auxv); 1664 1665 /* Try to dump the FPU. */ 1666 if ((prstatus->pr_fpvalid = 1667 elf_core_copy_task_fpregs(current, regs, fpu))) 1668 fill_note(notes + numnote++, 1669 "CORE", NT_PRFPREG, sizeof(*fpu), fpu); 1670 #ifdef ELF_CORE_COPY_XFPREGS 1671 if (elf_core_copy_task_xfpregs(current, xfpu)) 1672 fill_note(notes + numnote++, 1673 "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu); 1674 #endif 1675 1676 fs = get_fs(); 1677 set_fs(KERNEL_DS); 1678 1679 DUMP_WRITE(elf, sizeof(*elf)); 1680 offset += sizeof(*elf); /* Elf header */ 1681 offset += (segs+1) * sizeof(struct elf_phdr); /* Program headers */ 1682 1683 /* Write notes phdr entry */ 1684 { 1685 struct elf_phdr phdr; 1686 int sz = 0; 1687 1688 for (i = 0; i < numnote; i++) 1689 sz += notesize(notes + i); 1690 1691 sz += thread_status_size; 1692 1693 fill_elf_note_phdr(&phdr, sz, offset); 1694 offset += sz; 1695 DUMP_WRITE(&phdr, sizeof(phdr)); 1696 } 1697 1698 /* Page-align dumped data */ 1699 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE); 1700 1701 /* write program headers for segments dump */ 1702 for ( 1703 #ifdef CONFIG_MMU 1704 vma = current->mm->mmap; vma; vma = vma->vm_next 1705 #else 1706 vml = current->mm->context.vmlist; vml; vml = vml->next 1707 #endif 1708 ) { 1709 struct elf_phdr phdr; 1710 size_t sz; 1711 1712 #ifndef CONFIG_MMU 1713 vma = vml->vma; 1714 #endif 1715 1716 sz = vma->vm_end - vma->vm_start; 1717 1718 phdr.p_type = PT_LOAD; 1719 phdr.p_offset = offset; 1720 phdr.p_vaddr = vma->vm_start; 1721 phdr.p_paddr = 0; 1722 phdr.p_filesz = maydump(vma) ? sz : 0; 1723 phdr.p_memsz = sz; 1724 offset += phdr.p_filesz; 1725 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0; 1726 if (vma->vm_flags & VM_WRITE) 1727 phdr.p_flags |= PF_W; 1728 if (vma->vm_flags & VM_EXEC) 1729 phdr.p_flags |= PF_X; 1730 phdr.p_align = ELF_EXEC_PAGESIZE; 1731 1732 DUMP_WRITE(&phdr, sizeof(phdr)); 1733 } 1734 1735 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS 1736 ELF_CORE_WRITE_EXTRA_PHDRS; 1737 #endif 1738 1739 /* write out the notes section */ 1740 for (i = 0; i < numnote; i++) 1741 if (!writenote(notes + i, file)) 1742 goto end_coredump; 1743 1744 /* write out the thread status notes section */ 1745 list_for_each(t, &thread_list) { 1746 struct elf_thread_status *tmp = 1747 list_entry(t, struct elf_thread_status, list); 1748 1749 for (i = 0; i < tmp->num_notes; i++) 1750 if (!writenote(&tmp->notes[i], file)) 1751 goto end_coredump; 1752 } 1753 1754 DUMP_SEEK(dataoff); 1755 1756 if (elf_fdpic_dump_segments(file, current->mm, &size, &limit) < 0) 1757 goto end_coredump; 1758 1759 #ifdef ELF_CORE_WRITE_EXTRA_DATA 1760 ELF_CORE_WRITE_EXTRA_DATA; 1761 #endif 1762 1763 if (file->f_pos != offset) { 1764 /* Sanity check */ 1765 printk(KERN_WARNING 1766 "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n", 1767 file->f_pos, offset); 1768 } 1769 1770 end_coredump: 1771 set_fs(fs); 1772 1773 cleanup: 1774 while (!list_empty(&thread_list)) { 1775 struct list_head *tmp = thread_list.next; 1776 list_del(tmp); 1777 kfree(list_entry(tmp, struct elf_thread_status, list)); 1778 } 1779 1780 kfree(elf); 1781 kfree(prstatus); 1782 kfree(psinfo); 1783 kfree(notes); 1784 kfree(fpu); 1785 #ifdef ELF_CORE_COPY_XFPREGS 1786 kfree(xfpu); 1787 #endif 1788 return has_dumped; 1789 #undef NUM_NOTES 1790 } 1791 1792 #endif /* USE_ELF_CORE_DUMP */ 1793