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/mm.h>
  16 #include <linux/mman.h>
  17 #include <linux/errno.h>
  18 #include <linux/signal.h>
  19 #include <linux/binfmts.h>
  20 #include <linux/string.h>
  21 #include <linux/file.h>
  22 #include <linux/slab.h>
  23 #include <linux/personality.h>
  24 #include <linux/elfcore.h>
  25 #include <linux/init.h>
  26 #include <linux/highuid.h>
  27 #include <linux/compiler.h>
  28 #include <linux/highmem.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/security.h>
  31 #include <linux/random.h>
  32 #include <linux/elf.h>
  33 #include <linux/utsname.h>
  34 #include <linux/coredump.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/param.h>
  37 #include <asm/page.h>
  38
  39 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
  40 static int load_elf_library(struct file *);
  41 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
  42                                 int, int, unsigned long);
  43
  44 /*
  45  * If we don't support core dumping, then supply a NULL so we
  46  * don't even try.
  47  */
  48 #ifdef CONFIG_ELF_CORE
  49 static int elf_core_dump(struct coredump_params *cprm);
  50 #else
  51 #define elf_core_dump   NULL
  52 #endif
  53
  54 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  55 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  56 #else
  57 #define ELF_MIN_ALIGN   PAGE_SIZE
  58 #endif
  59
  60 #ifndef ELF_CORE_EFLAGS
  61 #define ELF_CORE_EFLAGS 0
  62 #endif
  63
  64 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  65 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  66 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  67
  68 static struct linux_binfmt elf_format = {
  69         .module         = THIS_MODULE,
  70         .load_binary    = load_elf_binary,
  71         .load_shlib     = load_elf_library,
  72         .core_dump      = elf_core_dump,
  73         .min_coredump   = ELF_EXEC_PAGESIZE,
  74 };
  75
  76 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
  77
  78 static int set_brk(unsigned long start, unsigned long end)
  79 {
  80         start = ELF_PAGEALIGN(start);
  81         end = ELF_PAGEALIGN(end);
  82         if (end > start) {
  83                 unsigned long addr;
  84                 down_write(&current->mm->mmap_sem);
  85                 addr = do_brk(start, end - start);
  86                 up_write(&current->mm->mmap_sem);
  87                 if (BAD_ADDR(addr))
  88                         return addr;
  89         }
  90         current->mm->start_brk = current->mm->brk = end;
  91         return 0;
  92 }
  93
  94 /* We need to explicitly zero any fractional pages
  95    after the data section (i.e. bss).  This would
  96    contain the junk from the file that should not
  97    be in memory
  98  */
  99 static int padzero(unsigned long elf_bss)
 100 {
 101         unsigned long nbyte;
 102
 103         nbyte = ELF_PAGEOFFSET(elf_bss);
 104         if (nbyte) {
 105                 nbyte = ELF_MIN_ALIGN - nbyte;
 106                 if (clear_user((void __user *) elf_bss, nbyte))
 107                         return -EFAULT;
 108         }
 109         return 0;
 110 }
 111
 112 /* Let's use some macros to make this stack manipulation a little clearer */
 113 #ifdef CONFIG_STACK_GROWSUP
 114 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
 115 #define STACK_ROUND(sp, items) \
 116         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
 117 #define STACK_ALLOC(sp, len) ({ \
 118         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
 119         old_sp; })
 120 #else
 121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
 122 #define STACK_ROUND(sp, items) \
 123         (((unsigned long) (sp - items)) &~ 15UL)
 124 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
 125 #endif
 126
 127 #ifndef ELF_BASE_PLATFORM
 128 /*
 129  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
 130  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
 131  * will be copied to the user stack in the same manner as AT_PLATFORM.
 132  */
 133 #define ELF_BASE_PLATFORM NULL
 134 #endif
 135
 136 static int
 137 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 138                 unsigned long load_addr, unsigned long interp_load_addr)
 139 {
 140         unsigned long p = bprm->p;
 141         int argc = bprm->argc;
 142         int envc = bprm->envc;
 143         elf_addr_t __user *argv;
 144         elf_addr_t __user *envp;
 145         elf_addr_t __user *sp;
 146         elf_addr_t __user *u_platform;
 147         elf_addr_t __user *u_base_platform;
 148         elf_addr_t __user *u_rand_bytes;
 149         const char *k_platform = ELF_PLATFORM;
 150         const char *k_base_platform = ELF_BASE_PLATFORM;
 151         unsigned char k_rand_bytes[16];
 152         int items;
 153         elf_addr_t *elf_info;
 154         int ei_index = 0;
 155         const struct cred *cred = current_cred();
 156         struct vm_area_struct *vma;
 157
 158         /*
 159          * In some cases (e.g. Hyper-Threading), we want to avoid L1
 160          * evictions by the processes running on the same package. One
 161          * thing we can do is to shuffle the initial stack for them.
 162          */
 163
 164         p = arch_align_stack(p);
 165
 166         /*
 167          * If this architecture has a platform capability string, copy it
 168          * to userspace.  In some cases (Sparc), this info is impossible
 169          * for userspace to get any other way, in others (i386) it is
 170          * merely difficult.
 171          */
 172         u_platform = NULL;
 173         if (k_platform) {
 174                 size_t len = strlen(k_platform) + 1;
 175
 176                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 177                 if (__copy_to_user(u_platform, k_platform, len))
 178                         return -EFAULT;
 179         }
 180
 181         /*
 182          * If this architecture has a "base" platform capability
 183          * string, copy it to userspace.
 184          */
 185         u_base_platform = NULL;
 186         if (k_base_platform) {
 187                 size_t len = strlen(k_base_platform) + 1;
 188
 189                 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 190                 if (__copy_to_user(u_base_platform, k_base_platform, len))
 191                         return -EFAULT;
 192         }
 193
 194         /*
 195          * Generate 16 random bytes for userspace PRNG seeding.
 196          */
 197         get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
 198         u_rand_bytes = (elf_addr_t __user *)
 199                        STACK_ALLOC(p, sizeof(k_rand_bytes));
 200         if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
 201                 return -EFAULT;
 202
 203         /* Create the ELF interpreter info */
 204         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 205         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
 206 #define NEW_AUX_ENT(id, val) \
 207         do { \
 208                 elf_info[ei_index++] = id; \
 209                 elf_info[ei_index++] = val; \
 210         } while (0)
 211
 212 #ifdef ARCH_DLINFO
 213         /*
 214          * ARCH_DLINFO must come first so PPC can do its special alignment of
 215          * AUXV.
 216          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
 217          * ARCH_DLINFO changes
 218          */
 219         ARCH_DLINFO;
 220 #endif
 221         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 222         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 223         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 224         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 225         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 226         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 227         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 228         NEW_AUX_ENT(AT_FLAGS, 0);
 229         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 230         NEW_AUX_ENT(AT_UID, cred->uid);
 231         NEW_AUX_ENT(AT_EUID, cred->euid);
 232         NEW_AUX_ENT(AT_GID, cred->gid);
 233         NEW_AUX_ENT(AT_EGID, cred->egid);
 234         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 235         NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
 236         NEW_AUX_ENT(AT_EXECFN, bprm->exec);
 237         if (k_platform) {
 238                 NEW_AUX_ENT(AT_PLATFORM,
 239                             (elf_addr_t)(unsigned long)u_platform);
 240         }
 241         if (k_base_platform) {
 242                 NEW_AUX_ENT(AT_BASE_PLATFORM,
 243                             (elf_addr_t)(unsigned long)u_base_platform);
 244         }
 245         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 246                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 247         }
 248 #undef NEW_AUX_ENT
 249         /* AT_NULL is zero; clear the rest too */
 250         memset(&elf_info[ei_index], 0,
 251                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 252
 253         /* And advance past the AT_NULL entry.  */
 254         ei_index += 2;
 255
 256         sp = STACK_ADD(p, ei_index);
 257
 258         items = (argc + 1) + (envc + 1) + 1;
 259         bprm->p = STACK_ROUND(sp, items);
 260
 261         /* Point sp at the lowest address on the stack */
 262 #ifdef CONFIG_STACK_GROWSUP
 263         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 264         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 265 #else
 266         sp = (elf_addr_t __user *)bprm->p;
 267 #endif
 268
 269
 270         /*
 271          * Grow the stack manually; some architectures have a limit on how
 272          * far ahead a user-space access may be in order to grow the stack.
 273          */
 274         vma = find_extend_vma(current->mm, bprm->p);
 275         if (!vma)
 276                 return -EFAULT;
 277
 278         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 279         if (__put_user(argc, sp++))
 280                 return -EFAULT;
 281         argv = sp;
 282         envp = argv + argc + 1;
 283
 284         /* Populate argv and envp */
 285         p = current->mm->arg_end = current->mm->arg_start;
 286         while (argc-- > 0) {
 287                 size_t len;
 288                 if (__put_user((elf_addr_t)p, argv++))
 289                         return -EFAULT;
 290                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 291                 if (!len || len > MAX_ARG_STRLEN)
 292                         return -EINVAL;
 293                 p += len;
 294         }
 295         if (__put_user(0, argv))
 296                 return -EFAULT;
 297         current->mm->arg_end = current->mm->env_start = p;
 298         while (envc-- > 0) {
 299                 size_t len;
 300                 if (__put_user((elf_addr_t)p, envp++))
 301                         return -EFAULT;
 302                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 303                 if (!len || len > MAX_ARG_STRLEN)
 304                         return -EINVAL;
 305                 p += len;
 306         }
 307         if (__put_user(0, envp))
 308                 return -EFAULT;
 309         current->mm->env_end = p;
 310
 311         /* Put the elf_info on the stack in the right place.  */
 312         sp = (elf_addr_t __user *)envp + 1;
 313         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 314                 return -EFAULT;
 315         return 0;
 316 }
 317
 318 static unsigned long elf_map(struct file *filep, unsigned long addr,
 319                 struct elf_phdr *eppnt, int prot, int type,
 320                 unsigned long total_size)
 321 {
 322         unsigned long map_addr;
 323         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
 324         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
 325         addr = ELF_PAGESTART(addr);
 326         size = ELF_PAGEALIGN(size);
 327
 328         /* mmap() will return -EINVAL if given a zero size, but a
 329          * segment with zero filesize is perfectly valid */
 330         if (!size)
 331                 return addr;
 332
 333         down_write(&current->mm->mmap_sem);
 334         /*
 335         * total_size is the size of the ELF (interpreter) image.
 336         * The _first_ mmap needs to know the full size, otherwise
 337         * randomization might put this image into an overlapping
 338         * position with the ELF binary image. (since size < total_size)
 339         * So we first map the 'big' image - and unmap the remainder at
 340         * the end. (which unmap is needed for ELF images with holes.)
 341         */
 342         if (total_size) {
 343                 total_size = ELF_PAGEALIGN(total_size);
 344                 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
 345                 if (!BAD_ADDR(map_addr))
 346                         do_munmap(current->mm, map_addr+size, total_size-size);
 347         } else
 348                 map_addr = do_mmap(filep, addr, size, prot, type, off);
 349
 350         up_write(&current->mm->mmap_sem);
 351         return(map_addr);
 352 }
 353
 354 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
 355 {
 356         int i, first_idx = -1, last_idx = -1;
 357
 358         for (i = 0; i < nr; i++) {
 359                 if (cmds[i].p_type == PT_LOAD) {
 360                         last_idx = i;
 361                         if (first_idx == -1)
 362                                 first_idx = i;
 363                 }
 364         }
 365         if (first_idx == -1)
 366                 return 0;
 367
 368         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
 369                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
 370 }
 371
 372
 373 /* This is much more generalized than the library routine read function,
 374    so we keep this separate.  Technically the library read function
 375    is only provided so that we can read a.out libraries that have
 376    an ELF header */
 377
 378 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 379                 struct file *interpreter, unsigned long *interp_map_addr,
 380                 unsigned long no_base)
 381 {
 382         struct elf_phdr *elf_phdata;
 383         struct elf_phdr *eppnt;
 384         unsigned long load_addr = 0;
 385         int load_addr_set = 0;
 386         unsigned long last_bss = 0, elf_bss = 0;
 387         unsigned long error = ~0UL;
 388         unsigned long total_size;
 389         int retval, i, size;
 390
 391         /* First of all, some simple consistency checks */
 392         if (interp_elf_ex->e_type != ET_EXEC &&
 393             interp_elf_ex->e_type != ET_DYN)
 394                 goto out;
 395         if (!elf_check_arch(interp_elf_ex))
 396                 goto out;
 397         if (!interpreter->f_op || !interpreter->f_op->mmap)
 398                 goto out;
 399
 400         /*
 401          * If the size of this structure has changed, then punt, since
 402          * we will be doing the wrong thing.
 403          */
 404         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 405                 goto out;
 406         if (interp_elf_ex->e_phnum < 1 ||
 407                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 408                 goto out;
 409
 410         /* Now read in all of the header information */
 411         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 412         if (size > ELF_MIN_ALIGN)
 413                 goto out;
 414         elf_phdata = kmalloc(size, GFP_KERNEL);
 415         if (!elf_phdata)
 416                 goto out;
 417
 418         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 419                              (char *)elf_phdata, size);
 420         error = -EIO;
 421         if (retval != size) {
 422                 if (retval < 0)
 423                         error = retval;
 424                 goto out_close;
 425         }
 426
 427         total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
 428         if (!total_size) {
 429                 error = -EINVAL;
 430                 goto out_close;
 431         }
 432
 433         eppnt = elf_phdata;
 434         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 435                 if (eppnt->p_type == PT_LOAD) {
 436                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 437                         int elf_prot = 0;
 438                         unsigned long vaddr = 0;
 439                         unsigned long k, map_addr;
 440
 441                         if (eppnt->p_flags & PF_R)
 442                                 elf_prot = PROT_READ;
 443                         if (eppnt->p_flags & PF_W)
 444                                 elf_prot |= PROT_WRITE;
 445                         if (eppnt->p_flags & PF_X)
 446                                 elf_prot |= PROT_EXEC;
 447                         vaddr = eppnt->p_vaddr;
 448                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 449                                 elf_type |= MAP_FIXED;
 450                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
 451                                 load_addr = -vaddr;
 452
 453                         map_addr = elf_map(interpreter, load_addr + vaddr,
 454                                         eppnt, elf_prot, elf_type, total_size);
 455                         total_size = 0;
 456                         if (!*interp_map_addr)
 457                                 *interp_map_addr = map_addr;
 458                         error = map_addr;
 459                         if (BAD_ADDR(map_addr))
 460                                 goto out_close;
 461
 462                         if (!load_addr_set &&
 463                             interp_elf_ex->e_type == ET_DYN) {
 464                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 465                                 load_addr_set = 1;
 466                         }
 467
 468                         /*
 469                          * Check to see if the section's size will overflow the
 470                          * allowed task size. Note that p_filesz must always be
 471                          * <= p_memsize so it's only necessary to check p_memsz.
 472                          */
 473                         k = load_addr + eppnt->p_vaddr;
 474                         if (BAD_ADDR(k) ||
 475                             eppnt->p_filesz > eppnt->p_memsz ||
 476                             eppnt->p_memsz > TASK_SIZE ||
 477                             TASK_SIZE - eppnt->p_memsz < k) {
 478                                 error = -ENOMEM;
 479                                 goto out_close;
 480                         }
 481
 482                         /*
 483                          * Find the end of the file mapping for this phdr, and
 484                          * keep track of the largest address we see for this.
 485                          */
 486                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 487                         if (k > elf_bss)
 488                                 elf_bss = k;
 489
 490                         /*
 491                          * Do the same thing for the memory mapping - between
 492                          * elf_bss and last_bss is the bss section.
 493                          */
 494                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 495                         if (k > last_bss)
 496                                 last_bss = k;
 497                 }
 498         }
 499
 500         if (last_bss > elf_bss) {
 501                 /*
 502                  * Now fill out the bss section.  First pad the last page up
 503                  * to the page boundary, and then perform a mmap to make sure
 504                  * that there are zero-mapped pages up to and including the
 505                  * last bss page.
 506                  */
 507                 if (padzero(elf_bss)) {
 508                         error = -EFAULT;
 509                         goto out_close;
 510                 }
 511
 512                 /* What we have mapped so far */
 513                 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 514
 515                 /* Map the last of the bss segment */
 516                 down_write(&current->mm->mmap_sem);
 517                 error = do_brk(elf_bss, last_bss - elf_bss);
 518                 up_write(&current->mm->mmap_sem);
 519                 if (BAD_ADDR(error))
 520                         goto out_close;
 521         }
 522
 523         error = load_addr;
 524
 525 out_close:
 526         kfree(elf_phdata);
 527 out:
 528         return error;
 529 }
 530
 531 /*
 532  * These are the functions used to load ELF style executables and shared
 533  * libraries.  There is no binary dependent code anywhere else.
 534  */
 535
 536 #define INTERPRETER_NONE 0
 537 #define INTERPRETER_ELF 2
 538
 539 #ifndef STACK_RND_MASK
 540 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 541 #endif
 542
 543 static unsigned long randomize_stack_top(unsigned long stack_top)
 544 {
 545         unsigned int random_variable = 0;
 546
 547         if ((current->flags & PF_RANDOMIZE) &&
 548                 !(current->personality & ADDR_NO_RANDOMIZE)) {
 549                 random_variable = get_random_int() & STACK_RND_MASK;
 550                 random_variable <<= PAGE_SHIFT;
 551         }
 552 #ifdef CONFIG_STACK_GROWSUP
 553         return PAGE_ALIGN(stack_top) + random_variable;
 554 #else
 555         return PAGE_ALIGN(stack_top) - random_variable;
 556 #endif
 557 }
 558
 559 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 560 {
 561         struct file *interpreter = NULL; /* to shut gcc up */
 562         unsigned long load_addr = 0, load_bias = 0;
 563         int load_addr_set = 0;
 564         char * elf_interpreter = NULL;
 565         unsigned long error;
 566         struct elf_phdr *elf_ppnt, *elf_phdata;
 567         unsigned long elf_bss, elf_brk;
 568         int retval, i;
 569         unsigned int size;
 570         unsigned long elf_entry;
 571         unsigned long interp_load_addr = 0;
 572         unsigned long start_code, end_code, start_data, end_data;
 573         unsigned long reloc_func_desc = 0;
 574         int executable_stack = EXSTACK_DEFAULT;
 575         unsigned long def_flags = 0;
 576         struct {
 577                 struct elfhdr elf_ex;
 578                 struct elfhdr interp_elf_ex;
 579         } *loc;
 580
 581         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 582         if (!loc) {
 583                 retval = -ENOMEM;
 584                 goto out_ret;
 585         }
 586
 587         /* Get the exec-header */
 588         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 589
 590         retval = -ENOEXEC;
 591         /* First of all, some simple consistency checks */
 592         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 593                 goto out;
 594
 595         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 596                 goto out;
 597         if (!elf_check_arch(&loc->elf_ex))
 598                 goto out;
 599         if (!bprm->file->f_op || !bprm->file->f_op->mmap)
 600                 goto out;
 601
 602         /* Now read in all of the header information */
 603         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 604                 goto out;
 605         if (loc->elf_ex.e_phnum < 1 ||
 606                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 607                 goto out;
 608         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 609         retval = -ENOMEM;
 610         elf_phdata = kmalloc(size, GFP_KERNEL);
 611         if (!elf_phdata)
 612                 goto out;
 613
 614         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 615                              (char *)elf_phdata, size);
 616         if (retval != size) {
 617                 if (retval >= 0)
 618                         retval = -EIO;
 619                 goto out_free_ph;
 620         }
 621
 622         elf_ppnt = elf_phdata;
 623         elf_bss = 0;
 624         elf_brk = 0;
 625
 626         start_code = ~0UL;
 627         end_code = 0;
 628         start_data = 0;
 629         end_data = 0;
 630
 631         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 632                 if (elf_ppnt->p_type == PT_INTERP) {
 633                         /* This is the program interpreter used for
 634                          * shared libraries - for now assume that this
 635                          * is an a.out format binary
 636                          */
 637                         retval = -ENOEXEC;
 638                         if (elf_ppnt->p_filesz > PATH_MAX ||
 639                             elf_ppnt->p_filesz < 2)
 640                                 goto out_free_ph;
 641
 642                         retval = -ENOMEM;
 643                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 644                                                   GFP_KERNEL);
 645                         if (!elf_interpreter)
 646                                 goto out_free_ph;
 647
 648                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 649                                              elf_interpreter,
 650                                              elf_ppnt->p_filesz);
 651                         if (retval != elf_ppnt->p_filesz) {
 652                                 if (retval >= 0)
 653                                         retval = -EIO;
 654                                 goto out_free_interp;
 655                         }
 656                         /* make sure path is NULL terminated */
 657                         retval = -ENOEXEC;
 658                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 659                                 goto out_free_interp;
 660
 661                         interpreter = open_exec(elf_interpreter);
 662                         retval = PTR_ERR(interpreter);
 663                         if (IS_ERR(interpreter))
 664                                 goto out_free_interp;
 665
 666                         /*
 667                          * If the binary is not readable then enforce
 668                          * mm->dumpable = 0 regardless of the interpreter's
 669                          * permissions.
 670                          */
 671                         if (file_permission(interpreter, MAY_READ) < 0)
 672                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
 673
 674                         retval = kernel_read(interpreter, 0, bprm->buf,
 675                                              BINPRM_BUF_SIZE);
 676                         if (retval != BINPRM_BUF_SIZE) {
 677                                 if (retval >= 0)
 678                                         retval = -EIO;
 679                                 goto out_free_dentry;
 680                         }
 681
 682                         /* Get the exec headers */
 683                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
 684                         break;
 685                 }
 686                 elf_ppnt++;
 687         }
 688
 689         elf_ppnt = elf_phdata;
 690         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 691                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 692                         if (elf_ppnt->p_flags & PF_X)
 693                                 executable_stack = EXSTACK_ENABLE_X;
 694                         else
 695                                 executable_stack = EXSTACK_DISABLE_X;
 696                         break;
 697                 }
 698
 699         /* Some simple consistency checks for the interpreter */
 700         if (elf_interpreter) {
 701                 retval = -ELIBBAD;
 702                 /* Not an ELF interpreter */
 703                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 704                         goto out_free_dentry;
 705                 /* Verify the interpreter has a valid arch */
 706                 if (!elf_check_arch(&loc->interp_elf_ex))
 707                         goto out_free_dentry;
 708         }
 709
 710         /* Flush all traces of the currently running executable */
 711         retval = flush_old_exec(bprm);
 712         if (retval)
 713                 goto out_free_dentry;
 714
 715 #ifdef CONFIG_X86_32
 716         /*
 717          * Turn off the CS limit completely if exec-shield disabled or
 718          * NX active:
 719          */
 720         if (!exec_shield || executable_stack != EXSTACK_DISABLE_X || (__supported_pte_mask & _PAGE_NX))
 721                 arch_add_exec_range(current->mm, -1);
 722 #endif
 723
 724         /* OK, This is the point of no return */
 725         current->flags &= ~PF_FORKNOEXEC;
 726         current->mm->def_flags = def_flags;
 727
 728         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 729            may depend on the personality.  */
 730         SET_PERSONALITY(loc->elf_ex);
 731         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 732                 current->personality |= READ_IMPLIES_EXEC;
 733
 734         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 735                 current->flags |= PF_RANDOMIZE;
 736
 737         setup_new_exec(bprm);
 738
 739         /* Do this so that we can load the interpreter, if need be.  We will
 740            change some of these later */
 741         current->mm->free_area_cache = current->mm->mmap_base;
 742         current->mm->cached_hole_size = 0;
 743         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 744                                  executable_stack);
 745         if (retval < 0) {
 746                 send_sig(SIGKILL, current, 0);
 747                 goto out_free_dentry;
 748         }
 749
 750         current->mm->start_stack = bprm->p;
 751
 752         /* Now we do a little grungy work by mmapping the ELF image into
 753            the correct location in memory. */
 754         for(i = 0, elf_ppnt = elf_phdata;
 755             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 756                 int elf_prot = 0, elf_flags;
 757                 unsigned long k, vaddr;
 758
 759                 if (elf_ppnt->p_type != PT_LOAD)
 760                         continue;
 761
 762                 if (unlikely (elf_brk > elf_bss)) {
 763                         unsigned long nbyte;
 764
 765                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 766                            before this one. Map anonymous pages, if needed,
 767                            and clear the area.  */
 768                         retval = set_brk(elf_bss + load_bias,
 769                                          elf_brk + load_bias);
 770                         if (retval) {
 771                                 send_sig(SIGKILL, current, 0);
 772                                 goto out_free_dentry;
 773                         }
 774                         nbyte = ELF_PAGEOFFSET(elf_bss);
 775                         if (nbyte) {
 776                                 nbyte = ELF_MIN_ALIGN - nbyte;
 777                                 if (nbyte > elf_brk - elf_bss)
 778                                         nbyte = elf_brk - elf_bss;
 779                                 if (clear_user((void __user *)elf_bss +
 780                                                         load_bias, nbyte)) {
 781                                         /*
 782                                          * This bss-zeroing can fail if the ELF
 783                                          * file specifies odd protections. So
 784                                          * we don't check the return value
 785                                          */
 786                                 }
 787                         }
 788                 }
 789
 790                 if (elf_ppnt->p_flags & PF_R)
 791                         elf_prot |= PROT_READ;
 792                 if (elf_ppnt->p_flags & PF_W)
 793                         elf_prot |= PROT_WRITE;
 794                 if (elf_ppnt->p_flags & PF_X)
 795                         elf_prot |= PROT_EXEC;
 796
 797                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 798
 799                 vaddr = elf_ppnt->p_vaddr;
 800                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 801                         elf_flags |= MAP_FIXED;
 802                 } else if (loc->elf_ex.e_type == ET_DYN) {
 803                         /* Try and get dynamic programs out of the way of the
 804                          * default mmap base, as well as whatever program they
 805                          * might try to exec.  This is because the brk will
 806                          * follow the loader, and is not movable.  */
 807 #if defined(CONFIG_X86) || defined(CONFIG_ARM)
 808                         load_bias = 0;
 809 #else
 810                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 811 #endif
 812                 }
 813
 814                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 815                                 elf_prot, elf_flags, 0);
 816                 if (BAD_ADDR(error)) {
 817                         send_sig(SIGKILL, current, 0);
 818                         retval = IS_ERR((void *)error) ?
 819                                 PTR_ERR((void*)error) : -EINVAL;
 820                         goto out_free_dentry;
 821                 }
 822
 823                 if (!load_addr_set) {
 824                         load_addr_set = 1;
 825                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 826                         if (loc->elf_ex.e_type == ET_DYN) {
 827                                 load_bias += error -
 828                                              ELF_PAGESTART(load_bias + vaddr);
 829                                 load_addr += load_bias;
 830                                 reloc_func_desc = load_bias;
 831                         }
 832                 }
 833                 k = elf_ppnt->p_vaddr;
 834                 if (k < start_code)
 835                         start_code = k;
 836                 if (start_data < k)
 837                         start_data = k;
 838
 839                 /*
 840                  * Check to see if the section's size will overflow the
 841                  * allowed task size. Note that p_filesz must always be
 842                  * <= p_memsz so it is only necessary to check p_memsz.
 843                  */
 844                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 845                     elf_ppnt->p_memsz > TASK_SIZE ||
 846                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 847                         /* set_brk can never work. Avoid overflows. */
 848                         send_sig(SIGKILL, current, 0);
 849                         retval = -EINVAL;
 850                         goto out_free_dentry;
 851                 }
 852
 853                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 854
 855                 if (k > elf_bss)
 856                         elf_bss = k;
 857                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 858                         end_code = k;
 859                 if (end_data < k)
 860                         end_data = k;
 861                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 862                 if (k > elf_brk)
 863                         elf_brk = k;
 864         }
 865
 866         loc->elf_ex.e_entry += load_bias;
 867         elf_bss += load_bias;
 868         elf_brk += load_bias;
 869         start_code += load_bias;
 870         end_code += load_bias;
 871         start_data += load_bias;
 872         end_data += load_bias;
 873
 874         /* Calling set_brk effectively mmaps the pages that we need
 875          * for the bss and break sections.  We must do this before
 876          * mapping in the interpreter, to make sure it doesn't wind
 877          * up getting placed where the bss needs to go.
 878          */
 879         retval = set_brk(elf_bss, elf_brk);
 880         if (retval) {
 881                 send_sig(SIGKILL, current, 0);
 882                 goto out_free_dentry;
 883         }
 884         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 885                 send_sig(SIGSEGV, current, 0);
 886                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 887                 goto out_free_dentry;
 888         }
 889
 890         if (elf_interpreter) {
 891                 unsigned long uninitialized_var(interp_map_addr);
 892
 893                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
 894                                             interpreter,
 895                                             &interp_map_addr,
 896                                             load_bias);
 897                 if (!IS_ERR((void *)elf_entry)) {
 898                         /*
 899                          * load_elf_interp() returns relocation
 900                          * adjustment
 901                          */
 902                         interp_load_addr = elf_entry;
 903                         elf_entry += loc->interp_elf_ex.e_entry;
 904                 }
 905                 if (BAD_ADDR(elf_entry)) {
 906                         force_sig(SIGSEGV, current);
 907                         retval = IS_ERR((void *)elf_entry) ?
 908                                         (int)elf_entry : -EINVAL;
 909                         goto out_free_dentry;
 910                 }
 911                 reloc_func_desc = interp_load_addr;
 912
 913                 allow_write_access(interpreter);
 914                 fput(interpreter);
 915                 kfree(elf_interpreter);
 916         } else {
 917                 elf_entry = loc->elf_ex.e_entry;
 918                 if (BAD_ADDR(elf_entry)) {
 919                         force_sig(SIGSEGV, current);
 920                         retval = -EINVAL;
 921                         goto out_free_dentry;
 922                 }
 923         }
 924
 925         kfree(elf_phdata);
 926
 927         set_binfmt(&elf_format);
 928
 929 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 930         retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
 931         if (retval < 0) {
 932                 send_sig(SIGKILL, current, 0);
 933                 goto out;
 934         }
 935 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 936
 937         install_exec_creds(bprm);
 938         current->flags &= ~PF_FORKNOEXEC;
 939         retval = create_elf_tables(bprm, &loc->elf_ex,
 940                           load_addr, interp_load_addr);
 941         if (retval < 0) {
 942                 send_sig(SIGKILL, current, 0);
 943                 goto out;
 944         }
 945         /* N.B. passed_fileno might not be initialized? */
 946         current->mm->end_code = end_code;
 947         current->mm->start_code = start_code;
 948         current->mm->start_data = start_data;
 949         current->mm->end_data = end_data;
 950         current->mm->start_stack = bprm->p;
 951
 952 #ifdef arch_randomize_brk
 953         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
 954                 current->mm->brk = current->mm->start_brk =
 955                         arch_randomize_brk(current->mm);
 956 #endif
 957
 958         if (current->personality & MMAP_PAGE_ZERO) {
 959                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 960                    and some applications "depend" upon this behavior.
 961                    Since we do not have the power to recompile these, we
 962                    emulate the SVr4 behavior. Sigh. */
 963                 down_write(&current->mm->mmap_sem);
 964                 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
 965                                 MAP_FIXED | MAP_PRIVATE, 0);
 966                 up_write(&current->mm->mmap_sem);
 967         }
 968
 969 #ifdef ELF_PLAT_INIT
 970         /*
 971          * The ABI may specify that certain registers be set up in special
 972          * ways (on i386 %edx is the address of a DT_FINI function, for
 973          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
 974          * that the e_entry field is the address of the function descriptor
 975          * for the startup routine, rather than the address of the startup
 976          * routine itself.  This macro performs whatever initialization to
 977          * the regs structure is required as well as any relocations to the
 978          * function descriptor entries when executing dynamically links apps.
 979          */
 980         ELF_PLAT_INIT(regs, reloc_func_desc);
 981 #endif
 982
 983         start_thread(regs, elf_entry, bprm->p);
 984         retval = 0;
 985 out:
 986         kfree(loc);
 987 out_ret:
 988         return retval;
 989
 990         /* error cleanup */
 991 out_free_dentry:
 992         allow_write_access(interpreter);
 993         if (interpreter)
 994                 fput(interpreter);
 995 out_free_interp:
 996         kfree(elf_interpreter);
 997 out_free_ph:
 998         kfree(elf_phdata);
 999         goto out;
1000 }
1001
1002 /* This is really simpleminded and specialized - we are loading an
1003    a.out library that is given an ELF header. */
1004 static int load_elf_library(struct file *file)
1005 {
1006         struct elf_phdr *elf_phdata;
1007         struct elf_phdr *eppnt;
1008         unsigned long elf_bss, bss, len;
1009         int retval, error, i, j;
1010         struct elfhdr elf_ex;
1011
1012         error = -ENOEXEC;
1013         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1014         if (retval != sizeof(elf_ex))
1015                 goto out;
1016
1017         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1018                 goto out;
1019
1020         /* First of all, some simple consistency checks */
1021         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1022             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1023                 goto out;
1024
1025         /* Now read in all of the header information */
1026
1027         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1028         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1029
1030         error = -ENOMEM;
1031         elf_phdata = kmalloc(j, GFP_KERNEL);
1032         if (!elf_phdata)
1033                 goto out;
1034
1035         eppnt = elf_phdata;
1036         error = -ENOEXEC;
1037         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1038         if (retval != j)
1039                 goto out_free_ph;
1040
1041         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1042                 if ((eppnt + i)->p_type == PT_LOAD)
1043                         j++;
1044         if (j != 1)
1045                 goto out_free_ph;
1046
1047         while (eppnt->p_type != PT_LOAD)
1048                 eppnt++;
1049
1050         /* Now use mmap to map the library into memory. */
1051         down_write(&current->mm->mmap_sem);
1052         error = do_mmap(file,
1053                         ELF_PAGESTART(eppnt->p_vaddr),
1054                         (eppnt->p_filesz +
1055                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1056                         PROT_READ | PROT_WRITE | PROT_EXEC,
1057                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1058                         (eppnt->p_offset -
1059                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1060         up_write(&current->mm->mmap_sem);
1061         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1062                 goto out_free_ph;
1063
1064         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1065         if (padzero(elf_bss)) {
1066                 error = -EFAULT;
1067                 goto out_free_ph;
1068         }
1069
1070         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1071                             ELF_MIN_ALIGN - 1);
1072         bss = eppnt->p_memsz + eppnt->p_vaddr;
1073         if (bss > len) {
1074                 down_write(&current->mm->mmap_sem);
1075                 do_brk(len, bss - len);
1076                 up_write(&current->mm->mmap_sem);
1077         }
1078         error = 0;
1079
1080 out_free_ph:
1081         kfree(elf_phdata);
1082 out:
1083         return error;
1084 }
1085
1086 #ifdef CONFIG_ELF_CORE
1087 /*
1088  * ELF core dumper
1089  *
1090  * Modelled on fs/exec.c:aout_core_dump()
1091  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1092  */
1093
1094 /*
1095  * Decide what to dump of a segment, part, all or none.
1096  */
1097 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1098                                    unsigned long mm_flags)
1099 {
1100 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1101
1102         /* The vma can be set up to tell us the answer directly.  */
1103         if (vma->vm_flags & VM_ALWAYSDUMP)
1104                 goto whole;
1105
1106         /* Hugetlb memory check */
1107         if (vma->vm_flags & VM_HUGETLB) {
1108                 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1109                         goto whole;
1110                 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1111                         goto whole;
1112         }
1113
1114         /* Do not dump I/O mapped devices or special mappings */
1115         if (vma->vm_flags & (VM_IO | VM_RESERVED))
1116                 return 0;
1117
1118         /* By default, dump shared memory if mapped from an anonymous file. */
1119         if (vma->vm_flags & VM_SHARED) {
1120                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1121                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1122                         goto whole;
1123                 return 0;
1124         }
1125
1126         /* Dump segments that have been written to.  */
1127         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1128                 goto whole;
1129         if (vma->vm_file == NULL)
1130                 return 0;
1131
1132         if (FILTER(MAPPED_PRIVATE))
1133                 goto whole;
1134
1135         /*
1136          * If this looks like the beginning of a DSO or executable mapping,
1137          * check for an ELF header.  If we find one, dump the first page to
1138          * aid in determining what was mapped here.
1139          */
1140         if (FILTER(ELF_HEADERS) &&
1141             vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1142                 u32 __user *header = (u32 __user *) vma->vm_start;
1143                 u32 word;
1144                 mm_segment_t fs = get_fs();
1145                 /*
1146                  * Doing it this way gets the constant folded by GCC.
1147                  */
1148                 union {
1149                         u32 cmp;
1150                         char elfmag[SELFMAG];
1151                 } magic;
1152                 BUILD_BUG_ON(SELFMAG != sizeof word);
1153                 magic.elfmag[EI_MAG0] = ELFMAG0;
1154                 magic.elfmag[EI_MAG1] = ELFMAG1;
1155                 magic.elfmag[EI_MAG2] = ELFMAG2;
1156                 magic.elfmag[EI_MAG3] = ELFMAG3;
1157                 /*
1158                  * Switch to the user "segment" for get_user(),
1159                  * then put back what elf_core_dump() had in place.
1160                  */
1161                 set_fs(USER_DS);
1162                 if (unlikely(get_user(word, header)))
1163                         word = 0;
1164                 set_fs(fs);
1165                 if (word == magic.cmp)
1166                         return PAGE_SIZE;
1167         }
1168
1169 #undef  FILTER
1170
1171         return 0;
1172
1173 whole:
1174         return vma->vm_end - vma->vm_start;
1175 }
1176
1177 /* An ELF note in memory */
1178 struct memelfnote
1179 {
1180         const char *name;
1181         int type;
1182         unsigned int datasz;
1183         void *data;
1184 };
1185
1186 static int notesize(struct memelfnote *en)
1187 {
1188         int sz;
1189
1190         sz = sizeof(struct elf_note);
1191         sz += roundup(strlen(en->name) + 1, 4);
1192         sz += roundup(en->datasz, 4);
1193
1194         return sz;
1195 }
1196
1197 #define DUMP_WRITE(addr, nr, foffset)   \
1198         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1199
1200 static int alignfile(struct file *file, loff_t *foffset)
1201 {
1202         static const char buf[4] = { 0, };
1203         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1204         return 1;
1205 }
1206
1207 static int writenote(struct memelfnote *men, struct file *file,
1208                         loff_t *foffset)
1209 {
1210         struct elf_note en;
1211         en.n_namesz = strlen(men->name) + 1;
1212         en.n_descsz = men->datasz;
1213         en.n_type = men->type;
1214
1215         DUMP_WRITE(&en, sizeof(en), foffset);
1216         DUMP_WRITE(men->name, en.n_namesz, foffset);
1217         if (!alignfile(file, foffset))
1218                 return 0;
1219         DUMP_WRITE(men->data, men->datasz, foffset);
1220         if (!alignfile(file, foffset))
1221                 return 0;
1222
1223         return 1;
1224 }
1225 #undef DUMP_WRITE
1226
1227 static void fill_elf_header(struct elfhdr *elf, int segs,
1228                             u16 machine, u32 flags, u8 osabi)
1229 {
1230         memset(elf, 0, sizeof(*elf));
1231
1232         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1233         elf->e_ident[EI_CLASS] = ELF_CLASS;
1234         elf->e_ident[EI_DATA] = ELF_DATA;
1235         elf->e_ident[EI_VERSION] = EV_CURRENT;
1236         elf->e_ident[EI_OSABI] = ELF_OSABI;
1237
1238         elf->e_type = ET_CORE;
1239         elf->e_machine = machine;
1240         elf->e_version = EV_CURRENT;
1241         elf->e_phoff = sizeof(struct elfhdr);
1242         elf->e_flags = flags;
1243         elf->e_ehsize = sizeof(struct elfhdr);
1244         elf->e_phentsize = sizeof(struct elf_phdr);
1245         elf->e_phnum = segs;
1246
1247         return;
1248 }
1249
1250 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1251 {
1252         phdr->p_type = PT_NOTE;
1253         phdr->p_offset = offset;
1254         phdr->p_vaddr = 0;
1255         phdr->p_paddr = 0;
1256         phdr->p_filesz = sz;
1257         phdr->p_memsz = 0;
1258         phdr->p_flags = 0;
1259         phdr->p_align = 0;
1260         return;
1261 }
1262
1263 static void fill_note(struct memelfnote *note, const char *name, int type,
1264                 unsigned int sz, void *data)
1265 {
1266         note->name = name;
1267         note->type = type;
1268         note->datasz = sz;
1269         note->data = data;
1270         return;
1271 }
1272
1273 /*
1274  * fill up all the fields in prstatus from the given task struct, except
1275  * registers which need to be filled up separately.
1276  */
1277 static void fill_prstatus(struct elf_prstatus *prstatus,
1278                 struct task_struct *p, long signr)
1279 {
1280         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1281         prstatus->pr_sigpend = p->pending.signal.sig[0];
1282         prstatus->pr_sighold = p->blocked.sig[0];
1283         rcu_read_lock();
1284         prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1285         rcu_read_unlock();
1286         prstatus->pr_pid = task_pid_vnr(p);
1287         prstatus->pr_pgrp = task_pgrp_vnr(p);
1288         prstatus->pr_sid = task_session_vnr(p);
1289         if (thread_group_leader(p)) {
1290                 struct task_cputime cputime;
1291
1292                 /*
1293                  * This is the record for the group leader.  It shows the
1294                  * group-wide total, not its individual thread total.
1295                  */
1296                 thread_group_cputime(p, &cputime);
1297                 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1298                 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1299         } else {
1300                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1301                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1302         }
1303         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1304         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1305 }
1306
1307 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1308                        struct mm_struct *mm)
1309 {
1310         const struct cred *cred;
1311         unsigned int i, len;
1312
1313         /* first copy the parameters from user space */
1314         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1315
1316         len = mm->arg_end - mm->arg_start;
1317         if (len >= ELF_PRARGSZ)
1318                 len = ELF_PRARGSZ-1;
1319         if (copy_from_user(&psinfo->pr_psargs,
1320                            (const char __user *)mm->arg_start, len))
1321                 return -EFAULT;
1322         for(i = 0; i < len; i++)
1323                 if (psinfo->pr_psargs[i] == 0)
1324                         psinfo->pr_psargs[i] = ' ';
1325         psinfo->pr_psargs[len] = 0;
1326
1327         rcu_read_lock();
1328         psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1329         rcu_read_unlock();
1330         psinfo->pr_pid = task_pid_vnr(p);
1331         psinfo->pr_pgrp = task_pgrp_vnr(p);
1332         psinfo->pr_sid = task_session_vnr(p);
1333
1334         i = p->state ? ffz(~p->state) + 1 : 0;
1335         psinfo->pr_state = i;
1336         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1337         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1338         psinfo->pr_nice = task_nice(p);
1339         psinfo->pr_flag = p->flags;
1340         rcu_read_lock();
1341         cred = __task_cred(p);
1342         SET_UID(psinfo->pr_uid, cred->uid);
1343         SET_GID(psinfo->pr_gid, cred->gid);
1344         rcu_read_unlock();
1345         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1346
1347         return 0;
1348 }
1349
1350 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1351 {
1352         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1353         int i = 0;
1354         do
1355                 i += 2;
1356         while (auxv[i - 2] != AT_NULL);
1357         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1358 }
1359
1360 #ifdef CORE_DUMP_USE_REGSET
1361 #include <linux/regset.h>
1362
1363 struct elf_thread_core_info {
1364         struct elf_thread_core_info *next;
1365         struct task_struct *task;
1366         struct elf_prstatus prstatus;
1367         struct memelfnote notes[0];
1368 };
1369
1370 struct elf_note_info {
1371         struct elf_thread_core_info *thread;
1372         struct memelfnote psinfo;
1373         struct memelfnote auxv;
1374         size_t size;
1375         int thread_notes;
1376 };
1377
1378 /*
1379  * When a regset has a writeback hook, we call it on each thread before
1380  * dumping user memory.  On register window machines, this makes sure the
1381  * user memory backing the register data is up to date before we read it.
1382  */
1383 static void do_thread_regset_writeback(struct task_struct *task,
1384                                        const struct user_regset *regset)
1385 {
1386         if (regset->writeback)
1387                 regset->writeback(task, regset, 1);
1388 }
1389
1390 static int fill_thread_core_info(struct elf_thread_core_info *t,
1391                                  const struct user_regset_view *view,
1392                                  long signr, size_t *total)
1393 {
1394         unsigned int i;
1395
1396         /*
1397          * NT_PRSTATUS is the one special case, because the regset data
1398          * goes into the pr_reg field inside the note contents, rather
1399          * than being the whole note contents.  We fill the reset in here.
1400          * We assume that regset 0 is NT_PRSTATUS.
1401          */
1402         fill_prstatus(&t->prstatus, t->task, signr);
1403         (void) view->regsets[0].get(t->task, &view->regsets[0],
1404                                     0, sizeof(t->prstatus.pr_reg),
1405                                     &t->prstatus.pr_reg, NULL);
1406
1407         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1408                   sizeof(t->prstatus), &t->prstatus);
1409         *total += notesize(&t->notes[0]);
1410
1411         do_thread_regset_writeback(t->task, &view->regsets[0]);
1412
1413         /*
1414          * Each other regset might generate a note too.  For each regset
1415          * that has no core_note_type or is inactive, we leave t->notes[i]
1416          * all zero and we'll know to skip writing it later.
1417          */
1418         for (i = 1; i < view->n; ++i) {
1419                 const struct user_regset *regset = &view->regsets[i];
1420                 do_thread_regset_writeback(t->task, regset);
1421                 if (regset->core_note_type &&
1422                     (!regset->active || regset->active(t->task, regset))) {
1423                         int ret;
1424                         size_t size = regset->n * regset->size;
1425                         void *data = kmalloc(size, GFP_KERNEL);
1426                         if (unlikely(!data))
1427                                 return 0;
1428                         ret = regset->get(t->task, regset,
1429                                           0, size, data, NULL);
1430                         if (unlikely(ret))
1431                                 kfree(data);
1432                         else {
1433                                 if (regset->core_note_type != NT_PRFPREG)
1434                                         fill_note(&t->notes[i], "LINUX",
1435                                                   regset->core_note_type,
1436                                                   size, data);
1437                                 else {
1438                                         t->prstatus.pr_fpvalid = 1;
1439                                         fill_note(&t->notes[i], "CORE",
1440                                                   NT_PRFPREG, size, data);
1441                                 }
1442                                 *total += notesize(&t->notes[i]);
1443                         }
1444                 }
1445         }
1446
1447         return 1;
1448 }
1449
1450 static int fill_note_info(struct elfhdr *elf, int phdrs,
1451                           struct elf_note_info *info,
1452                           long signr, struct pt_regs *regs)
1453 {
1454         struct task_struct *dump_task = current;
1455         const struct user_regset_view *view = task_user_regset_view(dump_task);
1456         struct elf_thread_core_info *t;
1457         struct elf_prpsinfo *psinfo;
1458         struct core_thread *ct;
1459         unsigned int i;
1460
1461         info->size = 0;
1462         info->thread = NULL;
1463
1464         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1465         if (psinfo == NULL)
1466                 return 0;
1467
1468         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1469
1470         /*
1471          * Figure out how many notes we're going to need for each thread.
1472          */
1473         info->thread_notes = 0;
1474         for (i = 0; i < view->n; ++i)
1475                 if (view->regsets[i].core_note_type != 0)
1476                         ++info->thread_notes;
1477
1478         /*
1479          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1480          * since it is our one special case.
1481          */
1482         if (unlikely(info->thread_notes == 0) ||
1483             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1484                 WARN_ON(1);
1485                 return 0;
1486         }
1487
1488         /*
1489          * Initialize the ELF file header.
1490          */
1491         fill_elf_header(elf, phdrs,
1492                         view->e_machine, view->e_flags, view->ei_osabi);
1493
1494         /*
1495          * Allocate a structure for each thread.
1496          */
1497         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1498                 t = kzalloc(offsetof(struct elf_thread_core_info,
1499                                      notes[info->thread_notes]),
1500                             GFP_KERNEL);
1501                 if (unlikely(!t))
1502                         return 0;
1503
1504                 t->task = ct->task;
1505                 if (ct->task == dump_task || !info->thread) {
1506                         t->next = info->thread;
1507                         info->thread = t;
1508                 } else {
1509                         /*
1510                          * Make sure to keep the original task at
1511                          * the head of the list.
1512                          */
1513                         t->next = info->thread->next;
1514                         info->thread->next = t;
1515                 }
1516         }
1517
1518         /*
1519          * Now fill in each thread's information.
1520          */
1521         for (t = info->thread; t != NULL; t = t->next)
1522                 if (!fill_thread_core_info(t, view, signr, &info->size))
1523                         return 0;
1524
1525         /*
1526          * Fill in the two process-wide notes.
1527          */
1528         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1529         info->size += notesize(&info->psinfo);
1530
1531         fill_auxv_note(&info->auxv, current->mm);
1532         info->size += notesize(&info->auxv);
1533
1534         return 1;
1535 }
1536
1537 static size_t get_note_info_size(struct elf_note_info *info)
1538 {
1539         return info->size;
1540 }
1541
1542 /*
1543  * Write all the notes for each thread.  When writing the first thread, the
1544  * process-wide notes are interleaved after the first thread-specific note.
1545  */
1546 static int write_note_info(struct elf_note_info *info,
1547                            struct file *file, loff_t *foffset)
1548 {
1549         bool first = 1;
1550         struct elf_thread_core_info *t = info->thread;
1551
1552         do {
1553                 int i;
1554
1555                 if (!writenote(&t->notes[0], file, foffset))
1556                         return 0;
1557
1558                 if (first && !writenote(&info->psinfo, file, foffset))
1559                         return 0;
1560                 if (first && !writenote(&info->auxv, file, foffset))
1561                         return 0;
1562
1563                 for (i = 1; i < info->thread_notes; ++i)
1564                         if (t->notes[i].data &&
1565                             !writenote(&t->notes[i], file, foffset))
1566                                 return 0;
1567
1568                 first = 0;
1569                 t = t->next;
1570         } while (t);
1571
1572         return 1;
1573 }
1574
1575 static void free_note_info(struct elf_note_info *info)
1576 {
1577         struct elf_thread_core_info *threads = info->thread;
1578         while (threads) {
1579                 unsigned int i;
1580                 struct elf_thread_core_info *t = threads;
1581                 threads = t->next;
1582                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1583                 for (i = 1; i < info->thread_notes; ++i)
1584                         kfree(t->notes[i].data);
1585                 kfree(t);
1586         }
1587         kfree(info->psinfo.data);
1588 }
1589
1590 #else
1591
1592 /* Here is the structure in which status of each thread is captured. */
1593 struct elf_thread_status
1594 {
1595         struct list_head list;
1596         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1597         elf_fpregset_t fpu;             /* NT_PRFPREG */
1598         struct task_struct *thread;
1599 #ifdef ELF_CORE_COPY_XFPREGS
1600         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1601 #endif
1602         struct memelfnote notes[3];
1603         int num_notes;
1604 };
1605
1606 /*
1607  * In order to add the specific thread information for the elf file format,
1608  * we need to keep a linked list of every threads pr_status and then create
1609  * a single section for them in the final core file.
1610  */
1611 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1612 {
1613         int sz = 0;
1614         struct task_struct *p = t->thread;
1615         t->num_notes = 0;
1616
1617         fill_prstatus(&t->prstatus, p, signr);
1618         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1619
1620         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1621                   &(t->prstatus));
1622         t->num_notes++;
1623         sz += notesize(&t->notes[0]);
1624
1625         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1626                                                                 &t->fpu))) {
1627                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1628                           &(t->fpu));
1629                 t->num_notes++;
1630                 sz += notesize(&t->notes[1]);
1631         }
1632
1633 #ifdef ELF_CORE_COPY_XFPREGS
1634         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1635                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1636                           sizeof(t->xfpu), &t->xfpu);
1637                 t->num_notes++;
1638                 sz += notesize(&t->notes[2]);
1639         }
1640 #endif
1641         return sz;
1642 }
1643
1644 struct elf_note_info {
1645         struct memelfnote *notes;
1646         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1647         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1648         struct list_head thread_list;
1649         elf_fpregset_t *fpu;
1650 #ifdef ELF_CORE_COPY_XFPREGS
1651         elf_fpxregset_t *xfpu;
1652 #endif
1653         int thread_status_size;
1654         int numnote;
1655 };
1656
1657 static int elf_note_info_init(struct elf_note_info *info)
1658 {
1659         memset(info, 0, sizeof(*info));
1660         INIT_LIST_HEAD(&info->thread_list);
1661
1662         /* Allocate space for six ELF notes */
1663         info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
1664         if (!info->notes)
1665                 return 0;
1666         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1667         if (!info->psinfo)
1668                 goto notes_free;
1669         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1670         if (!info->prstatus)
1671                 goto psinfo_free;
1672         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1673         if (!info->fpu)
1674                 goto prstatus_free;
1675 #ifdef ELF_CORE_COPY_XFPREGS
1676         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1677         if (!info->xfpu)
1678                 goto fpu_free;
1679 #endif
1680         return 1;
1681 #ifdef ELF_CORE_COPY_XFPREGS
1682  fpu_free:
1683         kfree(info->fpu);
1684 #endif
1685  prstatus_free:
1686         kfree(info->prstatus);
1687  psinfo_free:
1688         kfree(info->psinfo);
1689  notes_free:
1690         kfree(info->notes);
1691         return 0;
1692 }
1693
1694 static int fill_note_info(struct elfhdr *elf, int phdrs,
1695                           struct elf_note_info *info,
1696                           long signr, struct pt_regs *regs)
1697 {
1698         struct list_head *t;
1699
1700         if (!elf_note_info_init(info))
1701                 return 0;
1702
1703         if (signr) {
1704                 struct core_thread *ct;
1705                 struct elf_thread_status *ets;
1706
1707                 for (ct = current->mm->core_state->dumper.next;
1708                                                 ct; ct = ct->next) {
1709                         ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1710                         if (!ets)
1711                                 return 0;
1712
1713                         ets->thread = ct->task;
1714                         list_add(&ets->list, &info->thread_list);
1715                 }
1716
1717                 list_for_each(t, &info->thread_list) {
1718                         int sz;
1719
1720                         ets = list_entry(t, struct elf_thread_status, list);
1721                         sz = elf_dump_thread_status(signr, ets);
1722                         info->thread_status_size += sz;
1723                 }
1724         }
1725         /* now collect the dump for the current */
1726         memset(info->prstatus, 0, sizeof(*info->prstatus));
1727         fill_prstatus(info->prstatus, current, signr);
1728         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1729
1730         /* Set up header */
1731         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1732
1733         /*
1734          * Set up the notes in similar form to SVR4 core dumps made
1735          * with info from their /proc.
1736          */
1737
1738         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1739                   sizeof(*info->prstatus), info->prstatus);
1740         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1741         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1742                   sizeof(*info->psinfo), info->psinfo);
1743
1744         info->numnote = 2;
1745
1746         fill_auxv_note(&info->notes[info->numnote++], current->mm);
1747
1748         /* Try to dump the FPU. */
1749         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1750                                                                info->fpu);
1751         if (info->prstatus->pr_fpvalid)
1752                 fill_note(info->notes + info->numnote++,
1753                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1754 #ifdef ELF_CORE_COPY_XFPREGS
1755         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1756                 fill_note(info->notes + info->numnote++,
1757                           "LINUX", ELF_CORE_XFPREG_TYPE,
1758                           sizeof(*info->xfpu), info->xfpu);
1759 #endif
1760
1761         return 1;
1762 }
1763
1764 static size_t get_note_info_size(struct elf_note_info *info)
1765 {
1766         int sz = 0;
1767         int i;
1768
1769         for (i = 0; i < info->numnote; i++)
1770                 sz += notesize(info->notes + i);
1771
1772         sz += info->thread_status_size;
1773
1774         return sz;
1775 }
1776
1777 static int write_note_info(struct elf_note_info *info,
1778                            struct file *file, loff_t *foffset)
1779 {
1780         int i;
1781         struct list_head *t;
1782
1783         for (i = 0; i < info->numnote; i++)
1784                 if (!writenote(info->notes + i, file, foffset))
1785                         return 0;
1786
1787         /* write out the thread status notes section */
1788         list_for_each(t, &info->thread_list) {
1789                 struct elf_thread_status *tmp =
1790                                 list_entry(t, struct elf_thread_status, list);
1791
1792                 for (i = 0; i < tmp->num_notes; i++)
1793                         if (!writenote(&tmp->notes[i], file, foffset))
1794                                 return 0;
1795         }
1796
1797         return 1;
1798 }
1799
1800 static void free_note_info(struct elf_note_info *info)
1801 {
1802         while (!list_empty(&info->thread_list)) {
1803                 struct list_head *tmp = info->thread_list.next;
1804                 list_del(tmp);
1805                 kfree(list_entry(tmp, struct elf_thread_status, list));
1806         }
1807
1808         kfree(info->prstatus);
1809         kfree(info->psinfo);
1810         kfree(info->notes);
1811         kfree(info->fpu);
1812 #ifdef ELF_CORE_COPY_XFPREGS
1813         kfree(info->xfpu);
1814 #endif
1815 }
1816
1817 #endif
1818
1819 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1820                                         struct vm_area_struct *gate_vma)
1821 {
1822         struct vm_area_struct *ret = tsk->mm->mmap;
1823
1824         if (ret)
1825                 return ret;
1826         return gate_vma;
1827 }
1828 /*
1829  * Helper function for iterating across a vma list.  It ensures that the caller
1830  * will visit `gate_vma' prior to terminating the search.
1831  */
1832 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1833                                         struct vm_area_struct *gate_vma)
1834 {
1835         struct vm_area_struct *ret;
1836
1837         ret = this_vma->vm_next;
1838         if (ret)
1839                 return ret;
1840         if (this_vma == gate_vma)
1841                 return NULL;
1842         return gate_vma;
1843 }
1844
1845 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1846                              elf_addr_t e_shoff, int segs)
1847 {
1848         elf->e_shoff = e_shoff;
1849         elf->e_shentsize = sizeof(*shdr4extnum);
1850         elf->e_shnum = 1;
1851         elf->e_shstrndx = SHN_UNDEF;
1852
1853         memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1854
1855         shdr4extnum->sh_type = SHT_NULL;
1856         shdr4extnum->sh_size = elf->e_shnum;
1857         shdr4extnum->sh_link = elf->e_shstrndx;
1858         shdr4extnum->sh_info = segs;
1859 }
1860
1861 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1862                                      unsigned long mm_flags)
1863 {
1864         struct vm_area_struct *vma;
1865         size_t size = 0;
1866
1867         for (vma = first_vma(current, gate_vma); vma != NULL;
1868              vma = next_vma(vma, gate_vma))
1869                 size += vma_dump_size(vma, mm_flags);
1870         return size;
1871 }
1872
1873 /*
1874  * Actual dumper
1875  *
1876  * This is a two-pass process; first we find the offsets of the bits,
1877  * and then they are actually written out.  If we run out of core limit
1878  * we just truncate.
1879  */
1880 static int elf_core_dump(struct coredump_params *cprm)
1881 {
1882         int has_dumped = 0;
1883         mm_segment_t fs;
1884         int segs;
1885         size_t size = 0;
1886         struct vm_area_struct *vma, *gate_vma;
1887         struct elfhdr *elf = NULL;
1888         loff_t offset = 0, dataoff, foffset;
1889         struct elf_note_info info;
1890         struct elf_phdr *phdr4note = NULL;
1891         struct elf_shdr *shdr4extnum = NULL;
1892         Elf_Half e_phnum;
1893         elf_addr_t e_shoff;
1894
1895         /*
1896          * We no longer stop all VM operations.
1897          *
1898          * This is because those proceses that could possibly change map_count
1899          * or the mmap / vma pages are now blocked in do_exit on current
1900          * finishing this core dump.
1901          *
1902          * Only ptrace can touch these memory addresses, but it doesn't change
1903          * the map_count or the pages allocated. So no possibility of crashing
1904          * exists while dumping the mm->vm_next areas to the core file.
1905          */
1906
1907         /* alloc memory for large data structures: too large to be on stack */
1908         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1909         if (!elf)
1910                 goto out;
1911         /*
1912          * The number of segs are recored into ELF header as 16bit value.
1913          * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1914          */
1915         segs = current->mm->map_count;
1916         segs += elf_core_extra_phdrs();
1917
1918         gate_vma = get_gate_vma(current);
1919         if (gate_vma != NULL)
1920                 segs++;
1921
1922         /* for notes section */
1923         segs++;
1924
1925         /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1926          * this, kernel supports extended numbering. Have a look at
1927          * include/linux/elf.h for further information. */
1928         e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1929
1930         /*
1931          * Collect all the non-memory information about the process for the
1932          * notes.  This also sets up the file header.
1933          */
1934         if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
1935                 goto cleanup;
1936
1937         has_dumped = 1;
1938         current->flags |= PF_DUMPCORE;
1939
1940         fs = get_fs();
1941         set_fs(KERNEL_DS);
1942
1943         offset += sizeof(*elf);                         /* Elf header */
1944         offset += segs * sizeof(struct elf_phdr);       /* Program headers */
1945         foffset = offset;
1946
1947         /* Write notes phdr entry */
1948         {
1949                 size_t sz = get_note_info_size(&info);
1950
1951                 sz += elf_coredump_extra_notes_size();
1952
1953                 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1954                 if (!phdr4note)
1955                         goto end_coredump;
1956
1957                 fill_elf_note_phdr(phdr4note, sz, offset);
1958                 offset += sz;
1959         }
1960
1961         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1962
1963         offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
1964         offset += elf_core_extra_data_size();
1965         e_shoff = offset;
1966
1967         if (e_phnum == PN_XNUM) {
1968                 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1969                 if (!shdr4extnum)
1970                         goto end_coredump;
1971                 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1972         }
1973
1974         offset = dataoff;
1975
1976         size += sizeof(*elf);
1977         if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1978                 goto end_coredump;
1979
1980         size += sizeof(*phdr4note);
1981         if (size > cprm->limit
1982             || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1983                 goto end_coredump;
1984
1985         /* Write program headers for segments dump */
1986         for (vma = first_vma(current, gate_vma); vma != NULL;
1987                         vma = next_vma(vma, gate_vma)) {
1988                 struct elf_phdr phdr;
1989
1990                 phdr.p_type = PT_LOAD;
1991                 phdr.p_offset = offset;
1992                 phdr.p_vaddr = vma->vm_start;
1993                 phdr.p_paddr = 0;
1994                 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
1995                 phdr.p_memsz = vma->vm_end - vma->vm_start;
1996                 offset += phdr.p_filesz;
1997                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1998                 if (vma->vm_flags & VM_WRITE)
1999                         phdr.p_flags |= PF_W;
2000                 if (vma->vm_flags & VM_EXEC)
2001                         phdr.p_flags |= PF_X;
2002                 phdr.p_align = ELF_EXEC_PAGESIZE;
2003
2004                 size += sizeof(phdr);
2005                 if (size > cprm->limit
2006                     || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2007                         goto end_coredump;
2008         }
2009
2010         if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2011                 goto end_coredump;
2012
2013         /* write out the notes section */
2014         if (!write_note_info(&info, cprm->file, &foffset))
2015                 goto end_coredump;
2016
2017         if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2018                 goto end_coredump;
2019
2020         /* Align to page */
2021         if (!dump_seek(cprm->file, dataoff - foffset))
2022                 goto end_coredump;
2023
2024         for (vma = first_vma(current, gate_vma); vma != NULL;
2025                         vma = next_vma(vma, gate_vma)) {
2026                 unsigned long addr;
2027                 unsigned long end;
2028
2029                 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2030
2031                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2032                         struct page *page;
2033                         int stop;
2034
2035                         page = get_dump_page(addr);
2036                         if (page) {
2037                                 void *kaddr = kmap(page);
2038                                 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2039                                         !dump_write(cprm->file, kaddr,
2040                                                     PAGE_SIZE);
2041                                 kunmap(page);
2042                                 page_cache_release(page);
2043                         } else
2044                                 stop = !dump_seek(cprm->file, PAGE_SIZE);
2045                         if (stop)
2046                                 goto end_coredump;
2047                 }
2048         }
2049
2050         if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2051                 goto end_coredump;
2052
2053         if (e_phnum == PN_XNUM) {
2054                 size += sizeof(*shdr4extnum);
2055                 if (size > cprm->limit
2056                     || !dump_write(cprm->file, shdr4extnum,
2057                                    sizeof(*shdr4extnum)))
2058                         goto end_coredump;
2059         }
2060
2061 end_coredump:
2062         set_fs(fs);
2063
2064 cleanup:
2065         free_note_info(&info);
2066         kfree(shdr4extnum);
2067         kfree(phdr4note);
2068         kfree(elf);
2069 out:
2070         return has_dumped;
2071 }
2072
2073 #endif          /* CONFIG_ELF_CORE */
2074
2075 static int __init init_elf_binfmt(void)
2076 {
2077         return register_binfmt(&elf_format);
2078 }
2079
2080 static void __exit exit_elf_binfmt(void)
2081 {
2082         /* Remove the COFF and ELF loaders. */
2083         unregister_binfmt(&elf_format);
2084 }
2085
2086 core_initcall(init_elf_binfmt);
2087 module_exit(exit_elf_binfmt);
2088 MODULE_LICENSE("GPL");