fs/binfmt_elf.c

   1 /*
   2  * linux/fs/binfmt_elf.c
   3  *
   4  * These are the functions used to load ELF format executables as used
   5  * on SVr4 machines.  Information on the format may be found in the book
   6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
   7  * Tools".
   8  *
   9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/kernel.h>
  14 #include <linux/fs.h>
  15 #include <linux/stat.h>
  16 #include <linux/time.h>
  17 #include <linux/mm.h>
  18 #include <linux/mman.h>
  19 #include <linux/a.out.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/ptrace.h>
  27 #include <linux/slab.h>
  28 #include <linux/shm.h>
  29 #include <linux/personality.h>
  30 #include <linux/elfcore.h>
  31 #include <linux/init.h>
  32 #include <linux/highuid.h>
  33 #include <linux/smp.h>
  34 #include <linux/smp_lock.h>
  35 #include <linux/compiler.h>
  36 #include <linux/highmem.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/security.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/random.h>
  41 #include <linux/elf.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/param.h>
  44 #include <asm/page.h>
  45
  46 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
  47 static int load_elf_library(struct file *);
  48 static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int);
  49 extern int dump_fpu (struct pt_regs *, elf_fpregset_t *);
  50
  51 #ifndef elf_addr_t
  52 #define elf_addr_t unsigned long
  53 #endif
  54
  55 /*
  56  * If we don't support core dumping, then supply a NULL so we
  57  * don't even try.
  58  */
  59 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
  60 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file);
  61 #else
  62 #define elf_core_dump   NULL
  63 #endif
  64
  65 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  66 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  67 #else
  68 #define ELF_MIN_ALIGN   PAGE_SIZE
  69 #endif
  70
  71 #ifndef ELF_CORE_EFLAGS
  72 #define ELF_CORE_EFLAGS 0
  73 #endif
  74
  75 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  76 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  77 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  78
  79 static struct linux_binfmt elf_format = {
  80                 .module         = THIS_MODULE,
  81                 .load_binary    = load_elf_binary,
  82                 .load_shlib     = load_elf_library,
  83                 .core_dump      = elf_core_dump,
  84                 .min_coredump   = ELF_EXEC_PAGESIZE
  85 };
  86
  87 #define BAD_ADDR(x) ((unsigned long)(x) > TASK_SIZE)
  88
  89 static int set_brk(unsigned long start, unsigned long end)
  90 {
  91         start = ELF_PAGEALIGN(start);
  92         end = ELF_PAGEALIGN(end);
  93         if (end > start) {
  94                 unsigned long addr;
  95                 down_write(&current->mm->mmap_sem);
  96                 addr = do_brk(start, end - start);
  97                 up_write(&current->mm->mmap_sem);
  98                 if (BAD_ADDR(addr))
  99                         return addr;
 100         }
 101         current->mm->start_brk = current->mm->brk = end;
 102         return 0;
 103 }
 104
 105 /* We need to explicitly zero any fractional pages
 106    after the data section (i.e. bss).  This would
 107    contain the junk from the file that should not
 108    be in memory
 109  */
 110 static int padzero(unsigned long elf_bss)
 111 {
 112         unsigned long nbyte;
 113
 114         nbyte = ELF_PAGEOFFSET(elf_bss);
 115         if (nbyte) {
 116                 nbyte = ELF_MIN_ALIGN - nbyte;
 117                 if (clear_user((void __user *) elf_bss, nbyte))
 118                         return -EFAULT;
 119         }
 120         return 0;
 121 }
 122
 123 /* Let's use some macros to make this stack manipulation a litle clearer */
 124 #ifdef CONFIG_STACK_GROWSUP
 125 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
 126 #define STACK_ROUND(sp, items) \
 127         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
 128 #define STACK_ALLOC(sp, len) ({ \
 129         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
 130         old_sp; })
 131 #else
 132 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
 133 #define STACK_ROUND(sp, items) \
 134         (((unsigned long) (sp - items)) &~ 15UL)
 135 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
 136 #endif
 137
 138 static int
 139 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 140                 int interp_aout, unsigned long load_addr,
 141                 unsigned long interp_load_addr)
 142 {
 143         unsigned long p = bprm->p;
 144         int argc = bprm->argc;
 145         int envc = bprm->envc;
 146         elf_addr_t __user *argv;
 147         elf_addr_t __user *envp;
 148         elf_addr_t __user *sp;
 149         elf_addr_t __user *u_platform;
 150         const char *k_platform = ELF_PLATFORM;
 151         int items;
 152         elf_addr_t *elf_info;
 153         int ei_index = 0;
 154         struct task_struct *tsk = current;
 155
 156         /*
 157          * If this architecture has a platform capability string, copy it
 158          * to userspace.  In some cases (Sparc), this info is impossible
 159          * for userspace to get any other way, in others (i386) it is
 160          * merely difficult.
 161          */
 162         u_platform = NULL;
 163         if (k_platform) {
 164                 size_t len = strlen(k_platform) + 1;
 165
 166                 /*
 167                  * In some cases (e.g. Hyper-Threading), we want to avoid L1
 168                  * evictions by the processes running on the same package. One
 169                  * thing we can do is to shuffle the initial stack for them.
 170                  */
 171
 172                 p = arch_align_stack(p);
 173
 174                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 175                 if (__copy_to_user(u_platform, k_platform, len))
 176                         return -EFAULT;
 177         }
 178
 179         /* Create the ELF interpreter info */
 180         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 181 #define NEW_AUX_ENT(id, val) \
 182         do { \
 183                 elf_info[ei_index++] = id; \
 184                 elf_info[ei_index++] = val; \
 185         } while (0)
 186
 187 #ifdef ARCH_DLINFO
 188         /*
 189          * ARCH_DLINFO must come first so PPC can do its special alignment of
 190          * AUXV.
 191          */
 192         ARCH_DLINFO;
 193 #endif
 194         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 195         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 196         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 197         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 198         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 199         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 200         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 201         NEW_AUX_ENT(AT_FLAGS, 0);
 202         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 203         NEW_AUX_ENT(AT_UID, tsk->uid);
 204         NEW_AUX_ENT(AT_EUID, tsk->euid);
 205         NEW_AUX_ENT(AT_GID, tsk->gid);
 206         NEW_AUX_ENT(AT_EGID, tsk->egid);
 207         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 208         if (k_platform) {
 209                 NEW_AUX_ENT(AT_PLATFORM,
 210                             (elf_addr_t)(unsigned long)u_platform);
 211         }
 212         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 213                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 214         }
 215 #undef NEW_AUX_ENT
 216         /* AT_NULL is zero; clear the rest too */
 217         memset(&elf_info[ei_index], 0,
 218                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 219
 220         /* And advance past the AT_NULL entry.  */
 221         ei_index += 2;
 222
 223         sp = STACK_ADD(p, ei_index);
 224
 225         items = (argc + 1) + (envc + 1);
 226         if (interp_aout) {
 227                 items += 3; /* a.out interpreters require argv & envp too */
 228         } else {
 229                 items += 1; /* ELF interpreters only put argc on the stack */
 230         }
 231         bprm->p = STACK_ROUND(sp, items);
 232
 233         /* Point sp at the lowest address on the stack */
 234 #ifdef CONFIG_STACK_GROWSUP
 235         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 236         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 237 #else
 238         sp = (elf_addr_t __user *)bprm->p;
 239 #endif
 240
 241         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 242         if (__put_user(argc, sp++))
 243                 return -EFAULT;
 244         if (interp_aout) {
 245                 argv = sp + 2;
 246                 envp = argv + argc + 1;
 247                 __put_user((elf_addr_t)(unsigned long)argv, sp++);
 248                 __put_user((elf_addr_t)(unsigned long)envp, sp++);
 249         } else {
 250                 argv = sp;
 251                 envp = argv + argc + 1;
 252         }
 253
 254         /* Populate argv and envp */
 255         p = current->mm->arg_end = current->mm->arg_start;
 256         while (argc-- > 0) {
 257                 size_t len;
 258                 __put_user((elf_addr_t)p, argv++);
 259                 len = strnlen_user((void __user *)p, PAGE_SIZE*MAX_ARG_PAGES);
 260                 if (!len || len > PAGE_SIZE*MAX_ARG_PAGES)
 261                         return 0;
 262                 p += len;
 263         }
 264         if (__put_user(0, argv))
 265                 return -EFAULT;
 266         current->mm->arg_end = current->mm->env_start = p;
 267         while (envc-- > 0) {
 268                 size_t len;
 269                 __put_user((elf_addr_t)p, envp++);
 270                 len = strnlen_user((void __user *)p, PAGE_SIZE*MAX_ARG_PAGES);
 271                 if (!len || len > PAGE_SIZE*MAX_ARG_PAGES)
 272                         return 0;
 273                 p += len;
 274         }
 275         if (__put_user(0, envp))
 276                 return -EFAULT;
 277         current->mm->env_end = p;
 278
 279         /* Put the elf_info on the stack in the right place.  */
 280         sp = (elf_addr_t __user *)envp + 1;
 281         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 282                 return -EFAULT;
 283         return 0;
 284 }
 285
 286 #ifndef elf_map
 287
 288 static unsigned long elf_map(struct file *filep, unsigned long addr,
 289                 struct elf_phdr *eppnt, int prot, int type)
 290 {
 291         unsigned long map_addr;
 292         unsigned long pageoffset = ELF_PAGEOFFSET(eppnt->p_vaddr);
 293
 294         down_write(&current->mm->mmap_sem);
 295         /* mmap() will return -EINVAL if given a zero size, but a
 296          * segment with zero filesize is perfectly valid */
 297         if (eppnt->p_filesz + pageoffset)
 298                 map_addr = do_mmap(filep, ELF_PAGESTART(addr),
 299                                    eppnt->p_filesz + pageoffset, prot, type,
 300                                    eppnt->p_offset - pageoffset);
 301         else
 302                 map_addr = ELF_PAGESTART(addr);
 303         up_write(&current->mm->mmap_sem);
 304         return(map_addr);
 305 }
 306
 307 #endif /* !elf_map */
 308
 309 /* This is much more generalized than the library routine read function,
 310    so we keep this separate.  Technically the library read function
 311    is only provided so that we can read a.out libraries that have
 312    an ELF header */
 313
 314 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 315                 struct file *interpreter, unsigned long *interp_load_addr)
 316 {
 317         struct elf_phdr *elf_phdata;
 318         struct elf_phdr *eppnt;
 319         unsigned long load_addr = 0;
 320         int load_addr_set = 0;
 321         unsigned long last_bss = 0, elf_bss = 0;
 322         unsigned long error = ~0UL;
 323         int retval, i, size;
 324
 325         /* First of all, some simple consistency checks */
 326         if (interp_elf_ex->e_type != ET_EXEC &&
 327             interp_elf_ex->e_type != ET_DYN)
 328                 goto out;
 329         if (!elf_check_arch(interp_elf_ex))
 330                 goto out;
 331         if (!interpreter->f_op || !interpreter->f_op->mmap)
 332                 goto out;
 333
 334         /*
 335          * If the size of this structure has changed, then punt, since
 336          * we will be doing the wrong thing.
 337          */
 338         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 339                 goto out;
 340         if (interp_elf_ex->e_phnum < 1 ||
 341                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 342                 goto out;
 343
 344         /* Now read in all of the header information */
 345         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 346         if (size > ELF_MIN_ALIGN)
 347                 goto out;
 348         elf_phdata = kmalloc(size, GFP_KERNEL);
 349         if (!elf_phdata)
 350                 goto out;
 351
 352         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 353                              (char *)elf_phdata,size);
 354         error = -EIO;
 355         if (retval != size) {
 356                 if (retval < 0)
 357                         error = retval;
 358                 goto out_close;
 359         }
 360
 361         eppnt = elf_phdata;
 362         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 363                 if (eppnt->p_type == PT_LOAD) {
 364                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 365                         int elf_prot = 0;
 366                         unsigned long vaddr = 0;
 367                         unsigned long k, map_addr;
 368
 369                         if (eppnt->p_flags & PF_R)
 370                                 elf_prot = PROT_READ;
 371                         if (eppnt->p_flags & PF_W)
 372                                 elf_prot |= PROT_WRITE;
 373                         if (eppnt->p_flags & PF_X)
 374                                 elf_prot |= PROT_EXEC;
 375                         vaddr = eppnt->p_vaddr;
 376                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 377                                 elf_type |= MAP_FIXED;
 378
 379                         map_addr = elf_map(interpreter, load_addr + vaddr,
 380                                            eppnt, elf_prot, elf_type);
 381                         error = map_addr;
 382                         if (BAD_ADDR(map_addr))
 383                                 goto out_close;
 384
 385                         if (!load_addr_set &&
 386                             interp_elf_ex->e_type == ET_DYN) {
 387                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 388                                 load_addr_set = 1;
 389                         }
 390
 391                         /*
 392                          * Check to see if the section's size will overflow the
 393                          * allowed task size. Note that p_filesz must always be
 394                          * <= p_memsize so it's only necessary to check p_memsz.
 395                          */
 396                         k = load_addr + eppnt->p_vaddr;
 397                         if (k > TASK_SIZE ||
 398                             eppnt->p_filesz > eppnt->p_memsz ||
 399                             eppnt->p_memsz > TASK_SIZE ||
 400                             TASK_SIZE - eppnt->p_memsz < k) {
 401                                 error = -ENOMEM;
 402                                 goto out_close;
 403                         }
 404
 405                         /*
 406                          * Find the end of the file mapping for this phdr, and
 407                          * keep track of the largest address we see for this.
 408                          */
 409                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 410                         if (k > elf_bss)
 411                                 elf_bss = k;
 412
 413                         /*
 414                          * Do the same thing for the memory mapping - between
 415                          * elf_bss and last_bss is the bss section.
 416                          */
 417                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 418                         if (k > last_bss)
 419                                 last_bss = k;
 420                 }
 421         }
 422
 423         /*
 424          * Now fill out the bss section.  First pad the last page up
 425          * to the page boundary, and then perform a mmap to make sure
 426          * that there are zero-mapped pages up to and including the
 427          * last bss page.
 428          */
 429         if (padzero(elf_bss)) {
 430                 error = -EFAULT;
 431                 goto out_close;
 432         }
 433
 434         /* What we have mapped so far */
 435         elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 436
 437         /* Map the last of the bss segment */
 438         if (last_bss > elf_bss) {
 439                 down_write(&current->mm->mmap_sem);
 440                 error = do_brk(elf_bss, last_bss - elf_bss);
 441                 up_write(&current->mm->mmap_sem);
 442                 if (BAD_ADDR(error))
 443                         goto out_close;
 444         }
 445
 446         *interp_load_addr = load_addr;
 447         error = ((unsigned long)interp_elf_ex->e_entry) + load_addr;
 448
 449 out_close:
 450         kfree(elf_phdata);
 451 out:
 452         return error;
 453 }
 454
 455 static unsigned long load_aout_interp(struct exec *interp_ex,
 456                 struct file *interpreter)
 457 {
 458         unsigned long text_data, elf_entry = ~0UL;
 459         char __user * addr;
 460         loff_t offset;
 461
 462         current->mm->end_code = interp_ex->a_text;
 463         text_data = interp_ex->a_text + interp_ex->a_data;
 464         current->mm->end_data = text_data;
 465         current->mm->brk = interp_ex->a_bss + text_data;
 466
 467         switch (N_MAGIC(*interp_ex)) {
 468         case OMAGIC:
 469                 offset = 32;
 470                 addr = (char __user *)0;
 471                 break;
 472         case ZMAGIC:
 473         case QMAGIC:
 474                 offset = N_TXTOFF(*interp_ex);
 475                 addr = (char __user *)N_TXTADDR(*interp_ex);
 476                 break;
 477         default:
 478                 goto out;
 479         }
 480
 481         down_write(&current->mm->mmap_sem);
 482         do_brk(0, text_data);
 483         up_write(&current->mm->mmap_sem);
 484         if (!interpreter->f_op || !interpreter->f_op->read)
 485                 goto out;
 486         if (interpreter->f_op->read(interpreter, addr, text_data, &offset) < 0)
 487                 goto out;
 488         flush_icache_range((unsigned long)addr,
 489                            (unsigned long)addr + text_data);
 490
 491         down_write(&current->mm->mmap_sem);
 492         do_brk(ELF_PAGESTART(text_data + ELF_MIN_ALIGN - 1),
 493                 interp_ex->a_bss);
 494         up_write(&current->mm->mmap_sem);
 495         elf_entry = interp_ex->a_entry;
 496
 497 out:
 498         return elf_entry;
 499 }
 500
 501 /*
 502  * These are the functions used to load ELF style executables and shared
 503  * libraries.  There is no binary dependent code anywhere else.
 504  */
 505
 506 #define INTERPRETER_NONE 0
 507 #define INTERPRETER_AOUT 1
 508 #define INTERPRETER_ELF 2
 509
 510 #ifndef STACK_RND_MASK
 511 #define STACK_RND_MASK 0x7ff            /* with 4K pages 8MB of VA */
 512 #endif
 513
 514 static unsigned long randomize_stack_top(unsigned long stack_top)
 515 {
 516         unsigned int random_variable = 0;
 517
 518         if (current->flags & PF_RANDOMIZE) {
 519                 random_variable = get_random_int() & STACK_RND_MASK;
 520                 random_variable <<= PAGE_SHIFT;
 521         }
 522 #ifdef CONFIG_STACK_GROWSUP
 523         return PAGE_ALIGN(stack_top) + random_variable;
 524 #else
 525         return PAGE_ALIGN(stack_top) - random_variable;
 526 #endif
 527 }
 528
 529 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 530 {
 531         struct file *interpreter = NULL; /* to shut gcc up */
 532         unsigned long load_addr = 0, load_bias = 0;
 533         int load_addr_set = 0;
 534         char * elf_interpreter = NULL;
 535         unsigned int interpreter_type = INTERPRETER_NONE;
 536         unsigned char ibcs2_interpreter = 0;
 537         unsigned long error;
 538         struct elf_phdr *elf_ppnt, *elf_phdata;
 539         unsigned long elf_bss, elf_brk;
 540         int elf_exec_fileno;
 541         int retval, i;
 542         unsigned int size;
 543         unsigned long elf_entry, interp_load_addr = 0;
 544         unsigned long start_code, end_code, start_data, end_data;
 545         unsigned long reloc_func_desc = 0;
 546         char passed_fileno[6];
 547         struct files_struct *files;
 548         int have_pt_gnu_stack, executable_stack = EXSTACK_DEFAULT;
 549         unsigned long def_flags = 0;
 550         struct {
 551                 struct elfhdr elf_ex;
 552                 struct elfhdr interp_elf_ex;
 553                 struct exec interp_ex;
 554         } *loc;
 555
 556         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 557         if (!loc) {
 558                 retval = -ENOMEM;
 559                 goto out_ret;
 560         }
 561
 562         /* Get the exec-header */
 563         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 564
 565         retval = -ENOEXEC;
 566         /* First of all, some simple consistency checks */
 567         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 568                 goto out;
 569
 570         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 571                 goto out;
 572         if (!elf_check_arch(&loc->elf_ex))
 573                 goto out;
 574         if (!bprm->file->f_op||!bprm->file->f_op->mmap)
 575                 goto out;
 576
 577         /* Now read in all of the header information */
 578         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 579                 goto out;
 580         if (loc->elf_ex.e_phnum < 1 ||
 581                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 582                 goto out;
 583         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 584         retval = -ENOMEM;
 585         elf_phdata = kmalloc(size, GFP_KERNEL);
 586         if (!elf_phdata)
 587                 goto out;
 588
 589         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 590                              (char *)elf_phdata, size);
 591         if (retval != size) {
 592                 if (retval >= 0)
 593                         retval = -EIO;
 594                 goto out_free_ph;
 595         }
 596
 597         files = current->files; /* Refcounted so ok */
 598         retval = unshare_files();
 599         if (retval < 0)
 600                 goto out_free_ph;
 601         if (files == current->files) {
 602                 put_files_struct(files);
 603                 files = NULL;
 604         }
 605
 606         /* exec will make our files private anyway, but for the a.out
 607            loader stuff we need to do it earlier */
 608         retval = get_unused_fd();
 609         if (retval < 0)
 610                 goto out_free_fh;
 611         get_file(bprm->file);
 612         fd_install(elf_exec_fileno = retval, bprm->file);
 613
 614         elf_ppnt = elf_phdata;
 615         elf_bss = 0;
 616         elf_brk = 0;
 617
 618         start_code = ~0UL;
 619         end_code = 0;
 620         start_data = 0;
 621         end_data = 0;
 622
 623         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 624                 if (elf_ppnt->p_type == PT_INTERP) {
 625                         /* This is the program interpreter used for
 626                          * shared libraries - for now assume that this
 627                          * is an a.out format binary
 628                          */
 629                         retval = -ENOEXEC;
 630                         if (elf_ppnt->p_filesz > PATH_MAX ||
 631                             elf_ppnt->p_filesz < 2)
 632                                 goto out_free_file;
 633
 634                         retval = -ENOMEM;
 635                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 636                                                   GFP_KERNEL);
 637                         if (!elf_interpreter)
 638                                 goto out_free_file;
 639
 640                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 641                                              elf_interpreter,
 642                                              elf_ppnt->p_filesz);
 643                         if (retval != elf_ppnt->p_filesz) {
 644                                 if (retval >= 0)
 645                                         retval = -EIO;
 646                                 goto out_free_interp;
 647                         }
 648                         /* make sure path is NULL terminated */
 649                         retval = -ENOEXEC;
 650                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 651                                 goto out_free_interp;
 652
 653                         /* If the program interpreter is one of these two,
 654                          * then assume an iBCS2 image. Otherwise assume
 655                          * a native linux image.
 656                          */
 657                         if (strcmp(elf_interpreter,"/usr/lib/libc.so.1") == 0 ||
 658                             strcmp(elf_interpreter,"/usr/lib/ld.so.1") == 0)
 659                                 ibcs2_interpreter = 1;
 660
 661                         /*
 662                          * The early SET_PERSONALITY here is so that the lookup
 663                          * for the interpreter happens in the namespace of the
 664                          * to-be-execed image.  SET_PERSONALITY can select an
 665                          * alternate root.
 666                          *
 667                          * However, SET_PERSONALITY is NOT allowed to switch
 668                          * this task into the new images's memory mapping
 669                          * policy - that is, TASK_SIZE must still evaluate to
 670                          * that which is appropriate to the execing application.
 671                          * This is because exit_mmap() needs to have TASK_SIZE
 672                          * evaluate to the size of the old image.
 673                          *
 674                          * So if (say) a 64-bit application is execing a 32-bit
 675                          * application it is the architecture's responsibility
 676                          * to defer changing the value of TASK_SIZE until the
 677                          * switch really is going to happen - do this in
 678                          * flush_thread().      - akpm
 679                          */
 680                         SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
 681
 682                         interpreter = open_exec(elf_interpreter);
 683                         retval = PTR_ERR(interpreter);
 684                         if (IS_ERR(interpreter))
 685                                 goto out_free_interp;
 686                         retval = kernel_read(interpreter, 0, bprm->buf,
 687                                              BINPRM_BUF_SIZE);
 688                         if (retval != BINPRM_BUF_SIZE) {
 689                                 if (retval >= 0)
 690                                         retval = -EIO;
 691                                 goto out_free_dentry;
 692                         }
 693
 694                         /* Get the exec headers */
 695                         loc->interp_ex = *((struct exec *)bprm->buf);
 696                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
 697                         break;
 698                 }
 699                 elf_ppnt++;
 700         }
 701
 702         elf_ppnt = elf_phdata;
 703         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 704                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 705                         if (elf_ppnt->p_flags & PF_X)
 706                                 executable_stack = EXSTACK_ENABLE_X;
 707                         else
 708                                 executable_stack = EXSTACK_DISABLE_X;
 709                         break;
 710                 }
 711         have_pt_gnu_stack = (i < loc->elf_ex.e_phnum);
 712
 713         /* Some simple consistency checks for the interpreter */
 714         if (elf_interpreter) {
 715                 interpreter_type = INTERPRETER_ELF | INTERPRETER_AOUT;
 716
 717                 /* Now figure out which format our binary is */
 718                 if ((N_MAGIC(loc->interp_ex) != OMAGIC) &&
 719                     (N_MAGIC(loc->interp_ex) != ZMAGIC) &&
 720                     (N_MAGIC(loc->interp_ex) != QMAGIC))
 721                         interpreter_type = INTERPRETER_ELF;
 722
 723                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 724                         interpreter_type &= ~INTERPRETER_ELF;
 725
 726                 retval = -ELIBBAD;
 727                 if (!interpreter_type)
 728                         goto out_free_dentry;
 729
 730                 /* Make sure only one type was selected */
 731                 if ((interpreter_type & INTERPRETER_ELF) &&
 732                      interpreter_type != INTERPRETER_ELF) {
 733                         // FIXME - ratelimit this before re-enabling
 734                         // printk(KERN_WARNING "ELF: Ambiguous type, using ELF\n");
 735                         interpreter_type = INTERPRETER_ELF;
 736                 }
 737                 /* Verify the interpreter has a valid arch */
 738                 if ((interpreter_type == INTERPRETER_ELF) &&
 739                     !elf_check_arch(&loc->interp_elf_ex))
 740                         goto out_free_dentry;
 741         } else {
 742                 /* Executables without an interpreter also need a personality  */
 743                 SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
 744         }
 745
 746         /* OK, we are done with that, now set up the arg stuff,
 747            and then start this sucker up */
 748         if ((!bprm->sh_bang) && (interpreter_type == INTERPRETER_AOUT)) {
 749                 char *passed_p = passed_fileno;
 750                 sprintf(passed_fileno, "%d", elf_exec_fileno);
 751
 752                 if (elf_interpreter) {
 753                         retval = copy_strings_kernel(1, &passed_p, bprm);
 754                         if (retval)
 755                                 goto out_free_dentry;
 756                         bprm->argc++;
 757                 }
 758         }
 759
 760         /* Flush all traces of the currently running executable */
 761         retval = flush_old_exec(bprm);
 762         if (retval)
 763                 goto out_free_dentry;
 764
 765         /* Discard our unneeded old files struct */
 766         if (files) {
 767                 put_files_struct(files);
 768                 files = NULL;
 769         }
 770
 771         /* OK, This is the point of no return */
 772         current->mm->start_data = 0;
 773         current->mm->end_data = 0;
 774         current->mm->end_code = 0;
 775         current->mm->mmap = NULL;
 776         current->flags &= ~PF_FORKNOEXEC;
 777         current->mm->def_flags = def_flags;
 778
 779         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 780            may depend on the personality.  */
 781         SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
 782         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 783                 current->personality |= READ_IMPLIES_EXEC;
 784
 785         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 786                 current->flags |= PF_RANDOMIZE;
 787         arch_pick_mmap_layout(current->mm);
 788
 789         /* Do this so that we can load the interpreter, if need be.  We will
 790            change some of these later */
 791         current->mm->free_area_cache = current->mm->mmap_base;
 792         current->mm->cached_hole_size = 0;
 793         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 794                                  executable_stack);
 795         if (retval < 0) {
 796                 send_sig(SIGKILL, current, 0);
 797                 goto out_free_dentry;
 798         }
 799
 800         current->mm->start_stack = bprm->p;
 801
 802         /* Now we do a little grungy work by mmaping the ELF image into
 803            the correct location in memory.  At this point, we assume that
 804            the image should be loaded at fixed address, not at a variable
 805            address. */
 806         for(i = 0, elf_ppnt = elf_phdata;
 807             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 808                 int elf_prot = 0, elf_flags;
 809                 unsigned long k, vaddr;
 810
 811                 if (elf_ppnt->p_type != PT_LOAD)
 812                         continue;
 813
 814                 if (unlikely (elf_brk > elf_bss)) {
 815                         unsigned long nbyte;
 816
 817                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 818                            before this one. Map anonymous pages, if needed,
 819                            and clear the area.  */
 820                         retval = set_brk (elf_bss + load_bias,
 821                                           elf_brk + load_bias);
 822                         if (retval) {
 823                                 send_sig(SIGKILL, current, 0);
 824                                 goto out_free_dentry;
 825                         }
 826                         nbyte = ELF_PAGEOFFSET(elf_bss);
 827                         if (nbyte) {
 828                                 nbyte = ELF_MIN_ALIGN - nbyte;
 829                                 if (nbyte > elf_brk - elf_bss)
 830                                         nbyte = elf_brk - elf_bss;
 831                                 if (clear_user((void __user *)elf_bss +
 832                                                         load_bias, nbyte)) {
 833                                         /*
 834                                          * This bss-zeroing can fail if the ELF
 835                                          * file specifies odd protections. So
 836                                          * we don't check the return value
 837                                          */
 838                                 }
 839                         }
 840                 }
 841
 842                 if (elf_ppnt->p_flags & PF_R)
 843                         elf_prot |= PROT_READ;
 844                 if (elf_ppnt->p_flags & PF_W)
 845                         elf_prot |= PROT_WRITE;
 846                 if (elf_ppnt->p_flags & PF_X)
 847                         elf_prot |= PROT_EXEC;
 848
 849                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 850
 851                 vaddr = elf_ppnt->p_vaddr;
 852                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 853                         elf_flags |= MAP_FIXED;
 854                 } else if (loc->elf_ex.e_type == ET_DYN) {
 855                         /* Try and get dynamic programs out of the way of the
 856                          * default mmap base, as well as whatever program they
 857                          * might try to exec.  This is because the brk will
 858                          * follow the loader, and is not movable.  */
 859                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 860                 }
 861
 862                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 863                                 elf_prot, elf_flags);
 864                 if (BAD_ADDR(error)) {
 865                         send_sig(SIGKILL, current, 0);
 866                         goto out_free_dentry;
 867                 }
 868
 869                 if (!load_addr_set) {
 870                         load_addr_set = 1;
 871                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 872                         if (loc->elf_ex.e_type == ET_DYN) {
 873                                 load_bias += error -
 874                                              ELF_PAGESTART(load_bias + vaddr);
 875                                 load_addr += load_bias;
 876                                 reloc_func_desc = load_bias;
 877                         }
 878                 }
 879                 k = elf_ppnt->p_vaddr;
 880                 if (k < start_code)
 881                         start_code = k;
 882                 if (start_data < k)
 883                         start_data = k;
 884
 885                 /*
 886                  * Check to see if the section's size will overflow the
 887                  * allowed task size. Note that p_filesz must always be
 888                  * <= p_memsz so it is only necessary to check p_memsz.
 889                  */
 890                 if (k > TASK_SIZE || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 891                     elf_ppnt->p_memsz > TASK_SIZE ||
 892                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 893                         /* set_brk can never work. Avoid overflows. */
 894                         send_sig(SIGKILL, current, 0);
 895                         goto out_free_dentry;
 896                 }
 897
 898                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 899
 900                 if (k > elf_bss)
 901                         elf_bss = k;
 902                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 903                         end_code = k;
 904                 if (end_data < k)
 905                         end_data = k;
 906                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 907                 if (k > elf_brk)
 908                         elf_brk = k;
 909         }
 910
 911         loc->elf_ex.e_entry += load_bias;
 912         elf_bss += load_bias;
 913         elf_brk += load_bias;
 914         start_code += load_bias;
 915         end_code += load_bias;
 916         start_data += load_bias;
 917         end_data += load_bias;
 918
 919         /* Calling set_brk effectively mmaps the pages that we need
 920          * for the bss and break sections.  We must do this before
 921          * mapping in the interpreter, to make sure it doesn't wind
 922          * up getting placed where the bss needs to go.
 923          */
 924         retval = set_brk(elf_bss, elf_brk);
 925         if (retval) {
 926                 send_sig(SIGKILL, current, 0);
 927                 goto out_free_dentry;
 928         }
 929         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 930                 send_sig(SIGSEGV, current, 0);
 931                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 932                 goto out_free_dentry;
 933         }
 934
 935         if (elf_interpreter) {
 936                 if (interpreter_type == INTERPRETER_AOUT)
 937                         elf_entry = load_aout_interp(&loc->interp_ex,
 938                                                      interpreter);
 939                 else
 940                         elf_entry = load_elf_interp(&loc->interp_elf_ex,
 941                                                     interpreter,
 942                                                     &interp_load_addr);
 943                 if (BAD_ADDR(elf_entry)) {
 944                         printk(KERN_ERR "Unable to load interpreter %.128s\n",
 945                                 elf_interpreter);
 946                         force_sig(SIGSEGV, current);
 947                         retval = -ENOEXEC; /* Nobody gets to see this, but.. */
 948                         goto out_free_dentry;
 949                 }
 950                 reloc_func_desc = interp_load_addr;
 951
 952                 allow_write_access(interpreter);
 953                 fput(interpreter);
 954                 kfree(elf_interpreter);
 955         } else {
 956                 elf_entry = loc->elf_ex.e_entry;
 957                 if (BAD_ADDR(elf_entry)) {
 958                         send_sig(SIGSEGV, current, 0);
 959                         retval = -ENOEXEC; /* Nobody gets to see this, but.. */
 960                         goto out_free_dentry;
 961                 }
 962         }
 963
 964         kfree(elf_phdata);
 965
 966         if (interpreter_type != INTERPRETER_AOUT)
 967                 sys_close(elf_exec_fileno);
 968
 969         set_binfmt(&elf_format);
 970
 971 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 972         retval = arch_setup_additional_pages(bprm, executable_stack);
 973         if (retval < 0) {
 974                 send_sig(SIGKILL, current, 0);
 975                 goto out;
 976         }
 977 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 978
 979         compute_creds(bprm);
 980         current->flags &= ~PF_FORKNOEXEC;
 981         create_elf_tables(bprm, &loc->elf_ex,
 982                           (interpreter_type == INTERPRETER_AOUT),
 983                           load_addr, interp_load_addr);
 984         /* N.B. passed_fileno might not be initialized? */
 985         if (interpreter_type == INTERPRETER_AOUT)
 986                 current->mm->arg_start += strlen(passed_fileno) + 1;
 987         current->mm->end_code = end_code;
 988         current->mm->start_code = start_code;
 989         current->mm->start_data = start_data;
 990         current->mm->end_data = end_data;
 991         current->mm->start_stack = bprm->p;
 992
 993         if (current->personality & MMAP_PAGE_ZERO) {
 994                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 995                    and some applications "depend" upon this behavior.
 996                    Since we do not have the power to recompile these, we
 997                    emulate the SVr4 behavior. Sigh. */
 998                 down_write(&current->mm->mmap_sem);
 999                 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1000                                 MAP_FIXED | MAP_PRIVATE, 0);
1001                 up_write(&current->mm->mmap_sem);
1002         }
1003
1004 #ifdef ELF_PLAT_INIT
1005         /*
1006          * The ABI may specify that certain registers be set up in special
1007          * ways (on i386 %edx is the address of a DT_FINI function, for
1008          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1009          * that the e_entry field is the address of the function descriptor
1010          * for the startup routine, rather than the address of the startup
1011          * routine itself.  This macro performs whatever initialization to
1012          * the regs structure is required as well as any relocations to the
1013          * function descriptor entries when executing dynamically links apps.
1014          */
1015         ELF_PLAT_INIT(regs, reloc_func_desc);
1016 #endif
1017
1018         start_thread(regs, elf_entry, bprm->p);
1019         if (unlikely(current->ptrace & PT_PTRACED)) {
1020                 if (current->ptrace & PT_TRACE_EXEC)
1021                         ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
1022                 else
1023                         send_sig(SIGTRAP, current, 0);
1024         }
1025         retval = 0;
1026 out:
1027         kfree(loc);
1028 out_ret:
1029         return retval;
1030
1031         /* error cleanup */
1032 out_free_dentry:
1033         allow_write_access(interpreter);
1034         if (interpreter)
1035                 fput(interpreter);
1036 out_free_interp:
1037         kfree(elf_interpreter);
1038 out_free_file:
1039         sys_close(elf_exec_fileno);
1040 out_free_fh:
1041         if (files) {
1042                 put_files_struct(current->files);
1043                 current->files = files;
1044         }
1045 out_free_ph:
1046         kfree(elf_phdata);
1047         goto out;
1048 }
1049
1050 /* This is really simpleminded and specialized - we are loading an
1051    a.out library that is given an ELF header. */
1052 static int load_elf_library(struct file *file)
1053 {
1054         struct elf_phdr *elf_phdata;
1055         struct elf_phdr *eppnt;
1056         unsigned long elf_bss, bss, len;
1057         int retval, error, i, j;
1058         struct elfhdr elf_ex;
1059
1060         error = -ENOEXEC;
1061         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1062         if (retval != sizeof(elf_ex))
1063                 goto out;
1064
1065         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1066                 goto out;
1067
1068         /* First of all, some simple consistency checks */
1069         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1070             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1071                 goto out;
1072
1073         /* Now read in all of the header information */
1074
1075         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1076         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1077
1078         error = -ENOMEM;
1079         elf_phdata = kmalloc(j, GFP_KERNEL);
1080         if (!elf_phdata)
1081                 goto out;
1082
1083         eppnt = elf_phdata;
1084         error = -ENOEXEC;
1085         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1086         if (retval != j)
1087                 goto out_free_ph;
1088
1089         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1090                 if ((eppnt + i)->p_type == PT_LOAD)
1091                         j++;
1092         if (j != 1)
1093                 goto out_free_ph;
1094
1095         while (eppnt->p_type != PT_LOAD)
1096                 eppnt++;
1097
1098         /* Now use mmap to map the library into memory. */
1099         down_write(&current->mm->mmap_sem);
1100         error = do_mmap(file,
1101                         ELF_PAGESTART(eppnt->p_vaddr),
1102                         (eppnt->p_filesz +
1103                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1104                         PROT_READ | PROT_WRITE | PROT_EXEC,
1105                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1106                         (eppnt->p_offset -
1107                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1108         up_write(&current->mm->mmap_sem);
1109         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1110                 goto out_free_ph;
1111
1112         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1113         if (padzero(elf_bss)) {
1114                 error = -EFAULT;
1115                 goto out_free_ph;
1116         }
1117
1118         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1119                             ELF_MIN_ALIGN - 1);
1120         bss = eppnt->p_memsz + eppnt->p_vaddr;
1121         if (bss > len) {
1122                 down_write(&current->mm->mmap_sem);
1123                 do_brk(len, bss - len);
1124                 up_write(&current->mm->mmap_sem);
1125         }
1126         error = 0;
1127
1128 out_free_ph:
1129         kfree(elf_phdata);
1130 out:
1131         return error;
1132 }
1133
1134 /*
1135  * Note that some platforms still use traditional core dumps and not
1136  * the ELF core dump.  Each platform can select it as appropriate.
1137  */
1138 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1139
1140 /*
1141  * ELF core dumper
1142  *
1143  * Modelled on fs/exec.c:aout_core_dump()
1144  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1145  */
1146 /*
1147  * These are the only things you should do on a core-file: use only these
1148  * functions to write out all the necessary info.
1149  */
1150 static int dump_write(struct file *file, const void *addr, int nr)
1151 {
1152         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1153 }
1154
1155 static int dump_seek(struct file *file, loff_t off)
1156 {
1157         if (file->f_op->llseek) {
1158                 if (file->f_op->llseek(file, off, 0) != off)
1159                         return 0;
1160         } else
1161                 file->f_pos = off;
1162         return 1;
1163 }
1164
1165 /*
1166  * Decide whether a segment is worth dumping; default is yes to be
1167  * sure (missing info is worse than too much; etc).
1168  * Personally I'd include everything, and use the coredump limit...
1169  *
1170  * I think we should skip something. But I am not sure how. H.J.
1171  */
1172 static int maydump(struct vm_area_struct *vma)
1173 {
1174         /* Do not dump I/O mapped devices or special mappings */
1175         if (vma->vm_flags & (VM_IO | VM_RESERVED))
1176                 return 0;
1177
1178         /* Dump shared memory only if mapped from an anonymous file. */
1179         if (vma->vm_flags & VM_SHARED)
1180                 return vma->vm_file->f_dentry->d_inode->i_nlink == 0;
1181
1182         /* If it hasn't been written to, don't write it out */
1183         if (!vma->anon_vma)
1184                 return 0;
1185
1186         return 1;
1187 }
1188
1189 #define roundup(x, y) ((((x) + ((y) - 1)) / (y)) * (y))
1190
1191 /* An ELF note in memory */
1192 struct memelfnote
1193 {
1194         const char *name;
1195         int type;
1196         unsigned int datasz;
1197         void *data;
1198 };
1199
1200 static int notesize(struct memelfnote *en)
1201 {
1202         int sz;
1203
1204         sz = sizeof(struct elf_note);
1205         sz += roundup(strlen(en->name) + 1, 4);
1206         sz += roundup(en->datasz, 4);
1207
1208         return sz;
1209 }
1210
1211 #define DUMP_WRITE(addr, nr)    \
1212         do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1213 #define DUMP_SEEK(off)  \
1214         do { if (!dump_seek(file, (off))) return 0; } while(0)
1215
1216 static int writenote(struct memelfnote *men, struct file *file)
1217 {
1218         struct elf_note en;
1219
1220         en.n_namesz = strlen(men->name) + 1;
1221         en.n_descsz = men->datasz;
1222         en.n_type = men->type;
1223
1224         DUMP_WRITE(&en, sizeof(en));
1225         DUMP_WRITE(men->name, en.n_namesz);
1226         /* XXX - cast from long long to long to avoid need for libgcc.a */
1227         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1228         DUMP_WRITE(men->data, men->datasz);
1229         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1230
1231         return 1;
1232 }
1233 #undef DUMP_WRITE
1234 #undef DUMP_SEEK
1235
1236 #define DUMP_WRITE(addr, nr)    \
1237         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1238                 goto end_coredump;
1239 #define DUMP_SEEK(off)  \
1240         if (!dump_seek(file, (off))) \
1241                 goto end_coredump;
1242
1243 static void fill_elf_header(struct elfhdr *elf, int segs)
1244 {
1245         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1246         elf->e_ident[EI_CLASS] = ELF_CLASS;
1247         elf->e_ident[EI_DATA] = ELF_DATA;
1248         elf->e_ident[EI_VERSION] = EV_CURRENT;
1249         elf->e_ident[EI_OSABI] = ELF_OSABI;
1250         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1251
1252         elf->e_type = ET_CORE;
1253         elf->e_machine = ELF_ARCH;
1254         elf->e_version = EV_CURRENT;
1255         elf->e_entry = 0;
1256         elf->e_phoff = sizeof(struct elfhdr);
1257         elf->e_shoff = 0;
1258         elf->e_flags = ELF_CORE_EFLAGS;
1259         elf->e_ehsize = sizeof(struct elfhdr);
1260         elf->e_phentsize = sizeof(struct elf_phdr);
1261         elf->e_phnum = segs;
1262         elf->e_shentsize = 0;
1263         elf->e_shnum = 0;
1264         elf->e_shstrndx = 0;
1265         return;
1266 }
1267
1268 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, off_t offset)
1269 {
1270         phdr->p_type = PT_NOTE;
1271         phdr->p_offset = offset;
1272         phdr->p_vaddr = 0;
1273         phdr->p_paddr = 0;
1274         phdr->p_filesz = sz;
1275         phdr->p_memsz = 0;
1276         phdr->p_flags = 0;
1277         phdr->p_align = 0;
1278         return;
1279 }
1280
1281 static void fill_note(struct memelfnote *note, const char *name, int type,
1282                 unsigned int sz, void *data)
1283 {
1284         note->name = name;
1285         note->type = type;
1286         note->datasz = sz;
1287         note->data = data;
1288         return;
1289 }
1290
1291 /*
1292  * fill up all the fields in prstatus from the given task struct, except
1293  * registers which need to be filled up separately.
1294  */
1295 static void fill_prstatus(struct elf_prstatus *prstatus,
1296                 struct task_struct *p, long signr)
1297 {
1298         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1299         prstatus->pr_sigpend = p->pending.signal.sig[0];
1300         prstatus->pr_sighold = p->blocked.sig[0];
1301         prstatus->pr_pid = p->pid;
1302         prstatus->pr_ppid = p->parent->pid;
1303         prstatus->pr_pgrp = process_group(p);
1304         prstatus->pr_sid = p->signal->session;
1305         if (thread_group_leader(p)) {
1306                 /*
1307                  * This is the record for the group leader.  Add in the
1308                  * cumulative times of previous dead threads.  This total
1309                  * won't include the time of each live thread whose state
1310                  * is included in the core dump.  The final total reported
1311                  * to our parent process when it calls wait4 will include
1312                  * those sums as well as the little bit more time it takes
1313                  * this and each other thread to finish dying after the
1314                  * core dump synchronization phase.
1315                  */
1316                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1317                                    &prstatus->pr_utime);
1318                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1319                                    &prstatus->pr_stime);
1320         } else {
1321                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1322                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1323         }
1324         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1325         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1326 }
1327
1328 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1329                        struct mm_struct *mm)
1330 {
1331         unsigned int i, len;
1332
1333         /* first copy the parameters from user space */
1334         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1335
1336         len = mm->arg_end - mm->arg_start;
1337         if (len >= ELF_PRARGSZ)
1338                 len = ELF_PRARGSZ-1;
1339         if (copy_from_user(&psinfo->pr_psargs,
1340                            (const char __user *)mm->arg_start, len))
1341                 return -EFAULT;
1342         for(i = 0; i < len; i++)
1343                 if (psinfo->pr_psargs[i] == 0)
1344                         psinfo->pr_psargs[i] = ' ';
1345         psinfo->pr_psargs[len] = 0;
1346
1347         psinfo->pr_pid = p->pid;
1348         psinfo->pr_ppid = p->parent->pid;
1349         psinfo->pr_pgrp = process_group(p);
1350         psinfo->pr_sid = p->signal->session;
1351
1352         i = p->state ? ffz(~p->state) + 1 : 0;
1353         psinfo->pr_state = i;
1354         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1355         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1356         psinfo->pr_nice = task_nice(p);
1357         psinfo->pr_flag = p->flags;
1358         SET_UID(psinfo->pr_uid, p->uid);
1359         SET_GID(psinfo->pr_gid, p->gid);
1360         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1361
1362         return 0;
1363 }
1364
1365 /* Here is the structure in which status of each thread is captured. */
1366 struct elf_thread_status
1367 {
1368         struct list_head list;
1369         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1370         elf_fpregset_t fpu;             /* NT_PRFPREG */
1371         struct task_struct *thread;
1372 #ifdef ELF_CORE_COPY_XFPREGS
1373         elf_fpxregset_t xfpu;           /* NT_PRXFPREG */
1374 #endif
1375         struct memelfnote notes[3];
1376         int num_notes;
1377 };
1378
1379 /*
1380  * In order to add the specific thread information for the elf file format,
1381  * we need to keep a linked list of every threads pr_status and then create
1382  * a single section for them in the final core file.
1383  */
1384 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1385 {
1386         int sz = 0;
1387         struct task_struct *p = t->thread;
1388         t->num_notes = 0;
1389
1390         fill_prstatus(&t->prstatus, p, signr);
1391         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1392
1393         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1394                   &(t->prstatus));
1395         t->num_notes++;
1396         sz += notesize(&t->notes[0]);
1397
1398         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1399                                                                 &t->fpu))) {
1400                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1401                           &(t->fpu));
1402                 t->num_notes++;
1403                 sz += notesize(&t->notes[1]);
1404         }
1405
1406 #ifdef ELF_CORE_COPY_XFPREGS
1407         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1408                 fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu),
1409                           &t->xfpu);
1410                 t->num_notes++;
1411                 sz += notesize(&t->notes[2]);
1412         }
1413 #endif
1414         return sz;
1415 }
1416
1417 /*
1418  * Actual dumper
1419  *
1420  * This is a two-pass process; first we find the offsets of the bits,
1421  * and then they are actually written out.  If we run out of core limit
1422  * we just truncate.
1423  */
1424 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file)
1425 {
1426 #define NUM_NOTES       6
1427         int has_dumped = 0;
1428         mm_segment_t fs;
1429         int segs;
1430         size_t size = 0;
1431         int i;
1432         struct vm_area_struct *vma;
1433         struct elfhdr *elf = NULL;
1434         off_t offset = 0, dataoff;
1435         unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1436         int numnote;
1437         struct memelfnote *notes = NULL;
1438         struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1439         struct elf_prpsinfo *psinfo = NULL;     /* NT_PRPSINFO */
1440         struct task_struct *g, *p;
1441         LIST_HEAD(thread_list);
1442         struct list_head *t;
1443         elf_fpregset_t *fpu = NULL;
1444 #ifdef ELF_CORE_COPY_XFPREGS
1445         elf_fpxregset_t *xfpu = NULL;
1446 #endif
1447         int thread_status_size = 0;
1448         elf_addr_t *auxv;
1449
1450         /*
1451          * We no longer stop all VM operations.
1452          *
1453          * This is because those proceses that could possibly change map_count
1454          * or the mmap / vma pages are now blocked in do_exit on current
1455          * finishing this core dump.
1456          *
1457          * Only ptrace can touch these memory addresses, but it doesn't change
1458          * the map_count or the pages allocated. So no possibility of crashing
1459          * exists while dumping the mm->vm_next areas to the core file.
1460          */
1461
1462         /* alloc memory for large data structures: too large to be on stack */
1463         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1464         if (!elf)
1465                 goto cleanup;
1466         prstatus = kmalloc(sizeof(*prstatus), GFP_KERNEL);
1467         if (!prstatus)
1468                 goto cleanup;
1469         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1470         if (!psinfo)
1471                 goto cleanup;
1472         notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1473         if (!notes)
1474                 goto cleanup;
1475         fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1476         if (!fpu)
1477                 goto cleanup;
1478 #ifdef ELF_CORE_COPY_XFPREGS
1479         xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1480         if (!xfpu)
1481                 goto cleanup;
1482 #endif
1483
1484         if (signr) {
1485                 struct elf_thread_status *tmp;
1486                 read_lock(&tasklist_lock);
1487                 do_each_thread(g,p)
1488                         if (current->mm == p->mm && current != p) {
1489                                 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1490                                 if (!tmp) {
1491                                         read_unlock(&tasklist_lock);
1492                                         goto cleanup;
1493                                 }
1494                                 INIT_LIST_HEAD(&tmp->list);
1495                                 tmp->thread = p;
1496                                 list_add(&tmp->list, &thread_list);
1497                         }
1498                 while_each_thread(g,p);
1499                 read_unlock(&tasklist_lock);
1500                 list_for_each(t, &thread_list) {
1501                         struct elf_thread_status *tmp;
1502                         int sz;
1503
1504                         tmp = list_entry(t, struct elf_thread_status, list);
1505                         sz = elf_dump_thread_status(signr, tmp);
1506                         thread_status_size += sz;
1507                 }
1508         }
1509         /* now collect the dump for the current */
1510         memset(prstatus, 0, sizeof(*prstatus));
1511         fill_prstatus(prstatus, current, signr);
1512         elf_core_copy_regs(&prstatus->pr_reg, regs);
1513
1514         segs = current->mm->map_count;
1515 #ifdef ELF_CORE_EXTRA_PHDRS
1516         segs += ELF_CORE_EXTRA_PHDRS;
1517 #endif
1518
1519         /* Set up header */
1520         fill_elf_header(elf, segs + 1); /* including notes section */
1521
1522         has_dumped = 1;
1523         current->flags |= PF_DUMPCORE;
1524
1525         /*
1526          * Set up the notes in similar form to SVR4 core dumps made
1527          * with info from their /proc.
1528          */
1529
1530         fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1531         fill_psinfo(psinfo, current->group_leader, current->mm);
1532         fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1533
1534         numnote = 2;
1535
1536         auxv = (elf_addr_t *)current->mm->saved_auxv;
1537
1538         i = 0;
1539         do
1540                 i += 2;
1541         while (auxv[i - 2] != AT_NULL);
1542         fill_note(&notes[numnote++], "CORE", NT_AUXV,
1543                   i * sizeof(elf_addr_t), auxv);
1544
1545         /* Try to dump the FPU. */
1546         if ((prstatus->pr_fpvalid =
1547              elf_core_copy_task_fpregs(current, regs, fpu)))
1548                 fill_note(notes + numnote++,
1549                           "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1550 #ifdef ELF_CORE_COPY_XFPREGS
1551         if (elf_core_copy_task_xfpregs(current, xfpu))
1552                 fill_note(notes + numnote++,
1553                           "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu);
1554 #endif
1555
1556         fs = get_fs();
1557         set_fs(KERNEL_DS);
1558
1559         DUMP_WRITE(elf, sizeof(*elf));
1560         offset += sizeof(*elf);                         /* Elf header */
1561         offset += (segs+1) * sizeof(struct elf_phdr);   /* Program headers */
1562
1563         /* Write notes phdr entry */
1564         {
1565                 struct elf_phdr phdr;
1566                 int sz = 0;
1567
1568                 for (i = 0; i < numnote; i++)
1569                         sz += notesize(notes + i);
1570
1571                 sz += thread_status_size;
1572
1573                 fill_elf_note_phdr(&phdr, sz, offset);
1574                 offset += sz;
1575                 DUMP_WRITE(&phdr, sizeof(phdr));
1576         }
1577
1578         /* Page-align dumped data */
1579         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1580
1581         /* Write program headers for segments dump */
1582         for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1583                 struct elf_phdr phdr;
1584                 size_t sz;
1585
1586                 sz = vma->vm_end - vma->vm_start;
1587
1588                 phdr.p_type = PT_LOAD;
1589                 phdr.p_offset = offset;
1590                 phdr.p_vaddr = vma->vm_start;
1591                 phdr.p_paddr = 0;
1592                 phdr.p_filesz = maydump(vma) ? sz : 0;
1593                 phdr.p_memsz = sz;
1594                 offset += phdr.p_filesz;
1595                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1596                 if (vma->vm_flags & VM_WRITE)
1597                         phdr.p_flags |= PF_W;
1598                 if (vma->vm_flags & VM_EXEC)
1599                         phdr.p_flags |= PF_X;
1600                 phdr.p_align = ELF_EXEC_PAGESIZE;
1601
1602                 DUMP_WRITE(&phdr, sizeof(phdr));
1603         }
1604
1605 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1606         ELF_CORE_WRITE_EXTRA_PHDRS;
1607 #endif
1608
1609         /* write out the notes section */
1610         for (i = 0; i < numnote; i++)
1611                 if (!writenote(notes + i, file))
1612                         goto end_coredump;
1613
1614         /* write out the thread status notes section */
1615         list_for_each(t, &thread_list) {
1616                 struct elf_thread_status *tmp =
1617                                 list_entry(t, struct elf_thread_status, list);
1618
1619                 for (i = 0; i < tmp->num_notes; i++)
1620                         if (!writenote(&tmp->notes[i], file))
1621                                 goto end_coredump;
1622         }
1623
1624         DUMP_SEEK(dataoff);
1625
1626         for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1627                 unsigned long addr;
1628
1629                 if (!maydump(vma))
1630                         continue;
1631
1632                 for (addr = vma->vm_start;
1633                      addr < vma->vm_end;
1634                      addr += PAGE_SIZE) {
1635                         struct page *page;
1636                         struct vm_area_struct *vma;
1637
1638                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1639                                                 &page, &vma) <= 0) {
1640                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1641                         } else {
1642                                 if (page == ZERO_PAGE(addr)) {
1643                                         DUMP_SEEK(file->f_pos + PAGE_SIZE);
1644                                 } else {
1645                                         void *kaddr;
1646                                         flush_cache_page(vma, addr,
1647                                                          page_to_pfn(page));
1648                                         kaddr = kmap(page);
1649                                         if ((size += PAGE_SIZE) > limit ||
1650                                             !dump_write(file, kaddr,
1651                                             PAGE_SIZE)) {
1652                                                 kunmap(page);
1653                                                 page_cache_release(page);
1654                                                 goto end_coredump;
1655                                         }
1656                                         kunmap(page);
1657                                 }
1658                                 page_cache_release(page);
1659                         }
1660                 }
1661         }
1662
1663 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1664         ELF_CORE_WRITE_EXTRA_DATA;
1665 #endif
1666
1667         if ((off_t)file->f_pos != offset) {
1668                 /* Sanity check */
1669                 printk(KERN_WARNING
1670                        "elf_core_dump: file->f_pos (%ld) != offset (%ld)\n",
1671                        (off_t)file->f_pos, offset);
1672         }
1673
1674 end_coredump:
1675         set_fs(fs);
1676
1677 cleanup:
1678         while (!list_empty(&thread_list)) {
1679                 struct list_head *tmp = thread_list.next;
1680                 list_del(tmp);
1681                 kfree(list_entry(tmp, struct elf_thread_status, list));
1682         }
1683
1684         kfree(elf);
1685         kfree(prstatus);
1686         kfree(psinfo);
1687         kfree(notes);
1688         kfree(fpu);
1689 #ifdef ELF_CORE_COPY_XFPREGS
1690         kfree(xfpu);
1691 #endif
1692         return has_dumped;
1693 #undef NUM_NOTES
1694 }
1695
1696 #endif          /* USE_ELF_CORE_DUMP */
1697
1698 static int __init init_elf_binfmt(void)
1699 {
1700         return register_binfmt(&elf_format);
1701 }
1702
1703 static void __exit exit_elf_binfmt(void)
1704 {
1705         /* Remove the COFF and ELF loaders. */
1706         unregister_binfmt(&elf_format);
1707 }
1708
1709 core_initcall(init_elf_binfmt);
1710 module_exit(exit_elf_binfmt);
1711 MODULE_LICENSE("GPL");