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/vmalloc.h>
  31 #include <linux/security.h>
  32 #include <linux/random.h>
  33 #include <linux/elf.h>
  34 #include <linux/utsname.h>
  35 #include <linux/coredump.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/param.h>
  38 #include <asm/page.h>
  39
  40 #ifndef user_long_t
  41 #define user_long_t long
  42 #endif
  43 #ifndef user_siginfo_t
  44 #define user_siginfo_t siginfo_t
  45 #endif
  46
  47 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
  48 static int load_elf_library(struct file *);
  49 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
  50                                 int, int, unsigned long);
  51
  52 /*
  53  * If we don't support core dumping, then supply a NULL so we
  54  * don't even try.
  55  */
  56 #ifdef CONFIG_ELF_CORE
  57 static int elf_core_dump(struct coredump_params *cprm);
  58 #else
  59 #define elf_core_dump   NULL
  60 #endif
  61
  62 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  63 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  64 #else
  65 #define ELF_MIN_ALIGN   PAGE_SIZE
  66 #endif
  67
  68 #ifndef ELF_CORE_EFLAGS
  69 #define ELF_CORE_EFLAGS 0
  70 #endif
  71
  72 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  73 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  74 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  75
  76 static struct linux_binfmt elf_format = {
  77         .module         = THIS_MODULE,
  78         .load_binary    = load_elf_binary,
  79         .load_shlib     = load_elf_library,
  80         .core_dump      = elf_core_dump,
  81         .min_coredump   = ELF_EXEC_PAGESIZE,
  82 };
  83
  84 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
  85
  86 static int set_brk(unsigned long start, unsigned long end)
  87 {
  88         start = ELF_PAGEALIGN(start);
  89         end = ELF_PAGEALIGN(end);
  90         if (end > start) {
  91                 unsigned long addr;
  92                 addr = vm_brk(start, end - start);
  93                 if (BAD_ADDR(addr))
  94                         return addr;
  95         }
  96         current->mm->start_brk = current->mm->brk = end;
  97         return 0;
  98 }
  99
 100 /* We need to explicitly zero any fractional pages
 101    after the data section (i.e. bss).  This would
 102    contain the junk from the file that should not
 103    be in memory
 104  */
 105 static int padzero(unsigned long elf_bss)
 106 {
 107         unsigned long nbyte;
 108
 109         nbyte = ELF_PAGEOFFSET(elf_bss);
 110         if (nbyte) {
 111                 nbyte = ELF_MIN_ALIGN - nbyte;
 112                 if (clear_user((void __user *) elf_bss, nbyte))
 113                         return -EFAULT;
 114         }
 115         return 0;
 116 }
 117
 118 /* Let's use some macros to make this stack manipulation a little clearer */
 119 #ifdef CONFIG_STACK_GROWSUP
 120 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
 121 #define STACK_ROUND(sp, items) \
 122         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
 123 #define STACK_ALLOC(sp, len) ({ \
 124         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
 125         old_sp; })
 126 #else
 127 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
 128 #define STACK_ROUND(sp, items) \
 129         (((unsigned long) (sp - items)) &~ 15UL)
 130 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
 131 #endif
 132
 133 #ifndef ELF_BASE_PLATFORM
 134 /*
 135  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
 136  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
 137  * will be copied to the user stack in the same manner as AT_PLATFORM.
 138  */
 139 #define ELF_BASE_PLATFORM NULL
 140 #endif
 141
 142 static int
 143 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 144                 unsigned long load_addr, unsigned long interp_load_addr)
 145 {
 146         unsigned long p = bprm->p;
 147         int argc = bprm->argc;
 148         int envc = bprm->envc;
 149         elf_addr_t __user *argv;
 150         elf_addr_t __user *envp;
 151         elf_addr_t __user *sp;
 152         elf_addr_t __user *u_platform;
 153         elf_addr_t __user *u_base_platform;
 154         elf_addr_t __user *u_rand_bytes;
 155         const char *k_platform = ELF_PLATFORM;
 156         const char *k_base_platform = ELF_BASE_PLATFORM;
 157         unsigned char k_rand_bytes[16];
 158         int items;
 159         elf_addr_t *elf_info;
 160         int ei_index = 0;
 161         const struct cred *cred = current_cred();
 162         struct vm_area_struct *vma;
 163
 164         /*
 165          * In some cases (e.g. Hyper-Threading), we want to avoid L1
 166          * evictions by the processes running on the same package. One
 167          * thing we can do is to shuffle the initial stack for them.
 168          */
 169
 170         p = arch_align_stack(p);
 171
 172         /*
 173          * If this architecture has a platform capability string, copy it
 174          * to userspace.  In some cases (Sparc), this info is impossible
 175          * for userspace to get any other way, in others (i386) it is
 176          * merely difficult.
 177          */
 178         u_platform = NULL;
 179         if (k_platform) {
 180                 size_t len = strlen(k_platform) + 1;
 181
 182                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 183                 if (__copy_to_user(u_platform, k_platform, len))
 184                         return -EFAULT;
 185         }
 186
 187         /*
 188          * If this architecture has a "base" platform capability
 189          * string, copy it to userspace.
 190          */
 191         u_base_platform = NULL;
 192         if (k_base_platform) {
 193                 size_t len = strlen(k_base_platform) + 1;
 194
 195                 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 196                 if (__copy_to_user(u_base_platform, k_base_platform, len))
 197                         return -EFAULT;
 198         }
 199
 200         /*
 201          * Generate 16 random bytes for userspace PRNG seeding.
 202          */
 203         get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
 204         u_rand_bytes = (elf_addr_t __user *)
 205                        STACK_ALLOC(p, sizeof(k_rand_bytes));
 206         if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
 207                 return -EFAULT;
 208
 209         /* Create the ELF interpreter info */
 210         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 211         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
 212 #define NEW_AUX_ENT(id, val) \
 213         do { \
 214                 elf_info[ei_index++] = id; \
 215                 elf_info[ei_index++] = val; \
 216         } while (0)
 217
 218 #ifdef ARCH_DLINFO
 219         /*
 220          * ARCH_DLINFO must come first so PPC can do its special alignment of
 221          * AUXV.
 222          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
 223          * ARCH_DLINFO changes
 224          */
 225         ARCH_DLINFO;
 226 #endif
 227         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 228         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 229         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 230         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 231         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 232         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 233         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 234         NEW_AUX_ENT(AT_FLAGS, 0);
 235         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 236         NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
 237         NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
 238         NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
 239         NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
 240         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 241         NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
 242         NEW_AUX_ENT(AT_EXECFN, bprm->exec);
 243         if (k_platform) {
 244                 NEW_AUX_ENT(AT_PLATFORM,
 245                             (elf_addr_t)(unsigned long)u_platform);
 246         }
 247         if (k_base_platform) {
 248                 NEW_AUX_ENT(AT_BASE_PLATFORM,
 249                             (elf_addr_t)(unsigned long)u_base_platform);
 250         }
 251         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 252                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 253         }
 254 #undef NEW_AUX_ENT
 255         /* AT_NULL is zero; clear the rest too */
 256         memset(&elf_info[ei_index], 0,
 257                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 258
 259         /* And advance past the AT_NULL entry.  */
 260         ei_index += 2;
 261
 262         sp = STACK_ADD(p, ei_index);
 263
 264         items = (argc + 1) + (envc + 1) + 1;
 265         bprm->p = STACK_ROUND(sp, items);
 266
 267         /* Point sp at the lowest address on the stack */
 268 #ifdef CONFIG_STACK_GROWSUP
 269         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 270         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 271 #else
 272         sp = (elf_addr_t __user *)bprm->p;
 273 #endif
 274
 275
 276         /*
 277          * Grow the stack manually; some architectures have a limit on how
 278          * far ahead a user-space access may be in order to grow the stack.
 279          */
 280         vma = find_extend_vma(current->mm, bprm->p);
 281         if (!vma)
 282                 return -EFAULT;
 283
 284         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 285         if (__put_user(argc, sp++))
 286                 return -EFAULT;
 287         argv = sp;
 288         envp = argv + argc + 1;
 289
 290         /* Populate argv and envp */
 291         p = current->mm->arg_end = current->mm->arg_start;
 292         while (argc-- > 0) {
 293                 size_t len;
 294                 if (__put_user((elf_addr_t)p, argv++))
 295                         return -EFAULT;
 296                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 297                 if (!len || len > MAX_ARG_STRLEN)
 298                         return -EINVAL;
 299                 p += len;
 300         }
 301         if (__put_user(0, argv))
 302                 return -EFAULT;
 303         current->mm->arg_end = current->mm->env_start = p;
 304         while (envc-- > 0) {
 305                 size_t len;
 306                 if (__put_user((elf_addr_t)p, envp++))
 307                         return -EFAULT;
 308                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 309                 if (!len || len > MAX_ARG_STRLEN)
 310                         return -EINVAL;
 311                 p += len;
 312         }
 313         if (__put_user(0, envp))
 314                 return -EFAULT;
 315         current->mm->env_end = p;
 316
 317         /* Put the elf_info on the stack in the right place.  */
 318         sp = (elf_addr_t __user *)envp + 1;
 319         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 320                 return -EFAULT;
 321         return 0;
 322 }
 323
 324 static unsigned long elf_map(struct file *filep, unsigned long addr,
 325                 struct elf_phdr *eppnt, int prot, int type,
 326                 unsigned long total_size)
 327 {
 328         unsigned long map_addr;
 329         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
 330         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
 331         addr = ELF_PAGESTART(addr);
 332         size = ELF_PAGEALIGN(size);
 333
 334         /* mmap() will return -EINVAL if given a zero size, but a
 335          * segment with zero filesize is perfectly valid */
 336         if (!size)
 337                 return addr;
 338
 339         /*
 340         * total_size is the size of the ELF (interpreter) image.
 341         * The _first_ mmap needs to know the full size, otherwise
 342         * randomization might put this image into an overlapping
 343         * position with the ELF binary image. (since size < total_size)
 344         * So we first map the 'big' image - and unmap the remainder at
 345         * the end. (which unmap is needed for ELF images with holes.)
 346         */
 347         if (total_size) {
 348                 total_size = ELF_PAGEALIGN(total_size);
 349                 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
 350                 if (!BAD_ADDR(map_addr))
 351                         vm_munmap(map_addr+size, total_size-size);
 352         } else
 353                 map_addr = vm_mmap(filep, addr, size, prot, type, off);
 354
 355         return(map_addr);
 356 }
 357
 358 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
 359 {
 360         int i, first_idx = -1, last_idx = -1;
 361
 362         for (i = 0; i < nr; i++) {
 363                 if (cmds[i].p_type == PT_LOAD) {
 364                         last_idx = i;
 365                         if (first_idx == -1)
 366                                 first_idx = i;
 367                 }
 368         }
 369         if (first_idx == -1)
 370                 return 0;
 371
 372         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
 373                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
 374 }
 375
 376
 377 /* This is much more generalized than the library routine read function,
 378    so we keep this separate.  Technically the library read function
 379    is only provided so that we can read a.out libraries that have
 380    an ELF header */
 381
 382 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 383                 struct file *interpreter, unsigned long *interp_map_addr,
 384                 unsigned long no_base)
 385 {
 386         struct elf_phdr *elf_phdata;
 387         struct elf_phdr *eppnt;
 388         unsigned long load_addr = 0;
 389         int load_addr_set = 0;
 390         unsigned long last_bss = 0, elf_bss = 0;
 391         unsigned long error = ~0UL;
 392         unsigned long total_size;
 393         int retval, i, size;
 394
 395         /* First of all, some simple consistency checks */
 396         if (interp_elf_ex->e_type != ET_EXEC &&
 397             interp_elf_ex->e_type != ET_DYN)
 398                 goto out;
 399         if (!elf_check_arch(interp_elf_ex))
 400                 goto out;
 401         if (!interpreter->f_op || !interpreter->f_op->mmap)
 402                 goto out;
 403
 404         /*
 405          * If the size of this structure has changed, then punt, since
 406          * we will be doing the wrong thing.
 407          */
 408         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 409                 goto out;
 410         if (interp_elf_ex->e_phnum < 1 ||
 411                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 412                 goto out;
 413
 414         /* Now read in all of the header information */
 415         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 416         if (size > ELF_MIN_ALIGN)
 417                 goto out;
 418         elf_phdata = kmalloc(size, GFP_KERNEL);
 419         if (!elf_phdata)
 420                 goto out;
 421
 422         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 423                              (char *)elf_phdata, size);
 424         error = -EIO;
 425         if (retval != size) {
 426                 if (retval < 0)
 427                         error = retval;
 428                 goto out_close;
 429         }
 430
 431         total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
 432         if (!total_size) {
 433                 error = -EINVAL;
 434                 goto out_close;
 435         }
 436
 437         eppnt = elf_phdata;
 438         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 439                 if (eppnt->p_type == PT_LOAD) {
 440                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 441                         int elf_prot = 0;
 442                         unsigned long vaddr = 0;
 443                         unsigned long k, map_addr;
 444
 445                         if (eppnt->p_flags & PF_R)
 446                                 elf_prot = PROT_READ;
 447                         if (eppnt->p_flags & PF_W)
 448                                 elf_prot |= PROT_WRITE;
 449                         if (eppnt->p_flags & PF_X)
 450                                 elf_prot |= PROT_EXEC;
 451                         vaddr = eppnt->p_vaddr;
 452                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 453                                 elf_type |= MAP_FIXED;
 454                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
 455                                 load_addr = -vaddr;
 456
 457                         map_addr = elf_map(interpreter, load_addr + vaddr,
 458                                         eppnt, elf_prot, elf_type, total_size);
 459                         total_size = 0;
 460                         if (!*interp_map_addr)
 461                                 *interp_map_addr = map_addr;
 462                         error = map_addr;
 463                         if (BAD_ADDR(map_addr))
 464                                 goto out_close;
 465
 466                         if (!load_addr_set &&
 467                             interp_elf_ex->e_type == ET_DYN) {
 468                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 469                                 load_addr_set = 1;
 470                         }
 471
 472                         /*
 473                          * Check to see if the section's size will overflow the
 474                          * allowed task size. Note that p_filesz must always be
 475                          * <= p_memsize so it's only necessary to check p_memsz.
 476                          */
 477                         k = load_addr + eppnt->p_vaddr;
 478                         if (BAD_ADDR(k) ||
 479                             eppnt->p_filesz > eppnt->p_memsz ||
 480                             eppnt->p_memsz > TASK_SIZE ||
 481                             TASK_SIZE - eppnt->p_memsz < k) {
 482                                 error = -ENOMEM;
 483                                 goto out_close;
 484                         }
 485
 486                         /*
 487                          * Find the end of the file mapping for this phdr, and
 488                          * keep track of the largest address we see for this.
 489                          */
 490                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 491                         if (k > elf_bss)
 492                                 elf_bss = k;
 493
 494                         /*
 495                          * Do the same thing for the memory mapping - between
 496                          * elf_bss and last_bss is the bss section.
 497                          */
 498                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 499                         if (k > last_bss)
 500                                 last_bss = k;
 501                 }
 502         }
 503
 504         if (last_bss > elf_bss) {
 505                 /*
 506                  * Now fill out the bss section.  First pad the last page up
 507                  * to the page boundary, and then perform a mmap to make sure
 508                  * that there are zero-mapped pages up to and including the
 509                  * last bss page.
 510                  */
 511                 if (padzero(elf_bss)) {
 512                         error = -EFAULT;
 513                         goto out_close;
 514                 }
 515
 516                 /* What we have mapped so far */
 517                 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 518
 519                 /* Map the last of the bss segment */
 520                 error = vm_brk(elf_bss, last_bss - elf_bss);
 521                 if (BAD_ADDR(error))
 522                         goto out_close;
 523         }
 524
 525         error = load_addr;
 526
 527 out_close:
 528         kfree(elf_phdata);
 529 out:
 530         return error;
 531 }
 532
 533 /*
 534  * These are the functions used to load ELF style executables and shared
 535  * libraries.  There is no binary dependent code anywhere else.
 536  */
 537
 538 #define INTERPRETER_NONE 0
 539 #define INTERPRETER_ELF 2
 540
 541 #ifndef STACK_RND_MASK
 542 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 543 #endif
 544
 545 static unsigned long randomize_stack_top(unsigned long stack_top)
 546 {
 547         unsigned int random_variable = 0;
 548
 549         if ((current->flags & PF_RANDOMIZE) &&
 550                 !(current->personality & ADDR_NO_RANDOMIZE)) {
 551                 random_variable = get_random_int() & STACK_RND_MASK;
 552                 random_variable <<= PAGE_SHIFT;
 553         }
 554 #ifdef CONFIG_STACK_GROWSUP
 555         return PAGE_ALIGN(stack_top) + random_variable;
 556 #else
 557         return PAGE_ALIGN(stack_top) - random_variable;
 558 #endif
 559 }
 560
 561 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 562 {
 563         struct file *interpreter = NULL; /* to shut gcc up */
 564         unsigned long load_addr = 0, load_bias = 0;
 565         int load_addr_set = 0;
 566         char * elf_interpreter = NULL;
 567         unsigned long error;
 568         struct elf_phdr *elf_ppnt, *elf_phdata;
 569         unsigned long elf_bss, elf_brk;
 570         int retval, i;
 571         unsigned int size;
 572         unsigned long elf_entry;
 573         unsigned long interp_load_addr = 0;
 574         unsigned long start_code, end_code, start_data, end_data;
 575         unsigned long reloc_func_desc __maybe_unused = 0;
 576         int executable_stack = EXSTACK_DEFAULT;
 577         unsigned long def_flags = 0;
 578         struct {
 579                 struct elfhdr elf_ex;
 580                 struct elfhdr interp_elf_ex;
 581         } *loc;
 582
 583         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 584         if (!loc) {
 585                 retval = -ENOMEM;
 586                 goto out_ret;
 587         }
 588
 589         /* Get the exec-header */
 590         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 591
 592         retval = -ENOEXEC;
 593         /* First of all, some simple consistency checks */
 594         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 595                 goto out;
 596
 597         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 598                 goto out;
 599         if (!elf_check_arch(&loc->elf_ex))
 600                 goto out;
 601         if (!bprm->file->f_op || !bprm->file->f_op->mmap)
 602                 goto out;
 603
 604         /* Now read in all of the header information */
 605         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 606                 goto out;
 607         if (loc->elf_ex.e_phnum < 1 ||
 608                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 609                 goto out;
 610         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 611         retval = -ENOMEM;
 612         elf_phdata = kmalloc(size, GFP_KERNEL);
 613         if (!elf_phdata)
 614                 goto out;
 615
 616         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 617                              (char *)elf_phdata, size);
 618         if (retval != size) {
 619                 if (retval >= 0)
 620                         retval = -EIO;
 621                 goto out_free_ph;
 622         }
 623
 624         elf_ppnt = elf_phdata;
 625         elf_bss = 0;
 626         elf_brk = 0;
 627
 628         start_code = ~0UL;
 629         end_code = 0;
 630         start_data = 0;
 631         end_data = 0;
 632
 633         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 634                 if (elf_ppnt->p_type == PT_INTERP) {
 635                         /* This is the program interpreter used for
 636                          * shared libraries - for now assume that this
 637                          * is an a.out format binary
 638                          */
 639                         retval = -ENOEXEC;
 640                         if (elf_ppnt->p_filesz > PATH_MAX ||
 641                             elf_ppnt->p_filesz < 2)
 642                                 goto out_free_ph;
 643
 644                         retval = -ENOMEM;
 645                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 646                                                   GFP_KERNEL);
 647                         if (!elf_interpreter)
 648                                 goto out_free_ph;
 649
 650                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 651                                              elf_interpreter,
 652                                              elf_ppnt->p_filesz);
 653                         if (retval != elf_ppnt->p_filesz) {
 654                                 if (retval >= 0)
 655                                         retval = -EIO;
 656                                 goto out_free_interp;
 657                         }
 658                         /* make sure path is NULL terminated */
 659                         retval = -ENOEXEC;
 660                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 661                                 goto out_free_interp;
 662
 663                         interpreter = open_exec(elf_interpreter);
 664                         retval = PTR_ERR(interpreter);
 665                         if (IS_ERR(interpreter))
 666                                 goto out_free_interp;
 667
 668                         /*
 669                          * If the binary is not readable then enforce
 670                          * mm->dumpable = 0 regardless of the interpreter's
 671                          * permissions.
 672                          */
 673                         would_dump(bprm, interpreter);
 674
 675                         retval = kernel_read(interpreter, 0, bprm->buf,
 676                                              BINPRM_BUF_SIZE);
 677                         if (retval != BINPRM_BUF_SIZE) {
 678                                 if (retval >= 0)
 679                                         retval = -EIO;
 680                                 goto out_free_dentry;
 681                         }
 682
 683                         /* Get the exec headers */
 684                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
 685                         break;
 686                 }
 687                 elf_ppnt++;
 688         }
 689
 690         elf_ppnt = elf_phdata;
 691         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 692                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 693                         if (elf_ppnt->p_flags & PF_X)
 694                                 executable_stack = EXSTACK_ENABLE_X;
 695                         else
 696                                 executable_stack = EXSTACK_DISABLE_X;
 697                         break;
 698                 }
 699
 700         /* Some simple consistency checks for the interpreter */
 701         if (elf_interpreter) {
 702                 retval = -ELIBBAD;
 703                 /* Not an ELF interpreter */
 704                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 705                         goto out_free_dentry;
 706                 /* Verify the interpreter has a valid arch */
 707                 if (!elf_check_arch(&loc->interp_elf_ex))
 708                         goto out_free_dentry;
 709         }
 710
 711         /* Flush all traces of the currently running executable */
 712         retval = flush_old_exec(bprm);
 713         if (retval)
 714                 goto out_free_dentry;
 715
 716         /* OK, This is the point of no return */
 717         current->mm->def_flags = def_flags;
 718
 719         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 720            may depend on the personality.  */
 721         SET_PERSONALITY(loc->elf_ex);
 722         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 723                 current->personality |= READ_IMPLIES_EXEC;
 724
 725         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 726                 current->flags |= PF_RANDOMIZE;
 727
 728         setup_new_exec(bprm);
 729
 730         /* Do this so that we can load the interpreter, if need be.  We will
 731            change some of these later */
 732         current->mm->free_area_cache = current->mm->mmap_base;
 733         current->mm->cached_hole_size = 0;
 734         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 735                                  executable_stack);
 736         if (retval < 0) {
 737                 send_sig(SIGKILL, current, 0);
 738                 goto out_free_dentry;
 739         }
 740
 741         current->mm->start_stack = bprm->p;
 742
 743         /* Now we do a little grungy work by mmapping the ELF image into
 744            the correct location in memory. */
 745         for(i = 0, elf_ppnt = elf_phdata;
 746             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 747                 int elf_prot = 0, elf_flags;
 748                 unsigned long k, vaddr;
 749
 750                 if (elf_ppnt->p_type != PT_LOAD)
 751                         continue;
 752
 753                 if (unlikely (elf_brk > elf_bss)) {
 754                         unsigned long nbyte;
 755
 756                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 757                            before this one. Map anonymous pages, if needed,
 758                            and clear the area.  */
 759                         retval = set_brk(elf_bss + load_bias,
 760                                          elf_brk + load_bias);
 761                         if (retval) {
 762                                 send_sig(SIGKILL, current, 0);
 763                                 goto out_free_dentry;
 764                         }
 765                         nbyte = ELF_PAGEOFFSET(elf_bss);
 766                         if (nbyte) {
 767                                 nbyte = ELF_MIN_ALIGN - nbyte;
 768                                 if (nbyte > elf_brk - elf_bss)
 769                                         nbyte = elf_brk - elf_bss;
 770                                 if (clear_user((void __user *)elf_bss +
 771                                                         load_bias, nbyte)) {
 772                                         /*
 773                                          * This bss-zeroing can fail if the ELF
 774                                          * file specifies odd protections. So
 775                                          * we don't check the return value
 776                                          */
 777                                 }
 778                         }
 779                 }
 780
 781                 if (elf_ppnt->p_flags & PF_R)
 782                         elf_prot |= PROT_READ;
 783                 if (elf_ppnt->p_flags & PF_W)
 784                         elf_prot |= PROT_WRITE;
 785                 if (elf_ppnt->p_flags & PF_X)
 786                         elf_prot |= PROT_EXEC;
 787
 788                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 789
 790                 vaddr = elf_ppnt->p_vaddr;
 791                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 792                         elf_flags |= MAP_FIXED;
 793                 } else if (loc->elf_ex.e_type == ET_DYN) {
 794                         /* Try and get dynamic programs out of the way of the
 795                          * default mmap base, as well as whatever program they
 796                          * might try to exec.  This is because the brk will
 797                          * follow the loader, and is not movable.  */
 798 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
 799                         /* Memory randomization might have been switched off
 800                          * in runtime via sysctl.
 801                          * If that is the case, retain the original non-zero
 802                          * load_bias value in order to establish proper
 803                          * non-randomized mappings.
 804                          */
 805                         if (current->flags & PF_RANDOMIZE)
 806                                 load_bias = 0;
 807                         else
 808                                 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 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 interp_map_addr = 0;
 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         retval = create_elf_tables(bprm, &loc->elf_ex,
 939                           load_addr, interp_load_addr);
 940         if (retval < 0) {
 941                 send_sig(SIGKILL, current, 0);
 942                 goto out;
 943         }
 944         /* N.B. passed_fileno might not be initialized? */
 945         current->mm->end_code = end_code;
 946         current->mm->start_code = start_code;
 947         current->mm->start_data = start_data;
 948         current->mm->end_data = end_data;
 949         current->mm->start_stack = bprm->p;
 950
 951 #ifdef arch_randomize_brk
 952         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
 953                 current->mm->brk = current->mm->start_brk =
 954                         arch_randomize_brk(current->mm);
 955 #ifdef CONFIG_COMPAT_BRK
 956                 current->brk_randomized = 1;
 957 #endif
 958         }
 959 #endif
 960
 961         if (current->personality & MMAP_PAGE_ZERO) {
 962                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 963                    and some applications "depend" upon this behavior.
 964                    Since we do not have the power to recompile these, we
 965                    emulate the SVr4 behavior. Sigh. */
 966                 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
 967                                 MAP_FIXED | MAP_PRIVATE, 0);
 968         }
 969
 970 #ifdef ELF_PLAT_INIT
 971         /*
 972          * The ABI may specify that certain registers be set up in special
 973          * ways (on i386 %edx is the address of a DT_FINI function, for
 974          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
 975          * that the e_entry field is the address of the function descriptor
 976          * for the startup routine, rather than the address of the startup
 977          * routine itself.  This macro performs whatever initialization to
 978          * the regs structure is required as well as any relocations to the
 979          * function descriptor entries when executing dynamically links apps.
 980          */
 981         ELF_PLAT_INIT(regs, reloc_func_desc);
 982 #endif
 983
 984         start_thread(regs, elf_entry, bprm->p);
 985         retval = 0;
 986 out:
 987         kfree(loc);
 988 out_ret:
 989         return retval;
 990
 991         /* error cleanup */
 992 out_free_dentry:
 993         allow_write_access(interpreter);
 994         if (interpreter)
 995                 fput(interpreter);
 996 out_free_interp:
 997         kfree(elf_interpreter);
 998 out_free_ph:
 999         kfree(elf_phdata);
1000         goto out;
1001 }
1002
1003 /* This is really simpleminded and specialized - we are loading an
1004    a.out library that is given an ELF header. */
1005 static int load_elf_library(struct file *file)
1006 {
1007         struct elf_phdr *elf_phdata;
1008         struct elf_phdr *eppnt;
1009         unsigned long elf_bss, bss, len;
1010         int retval, error, i, j;
1011         struct elfhdr elf_ex;
1012
1013         error = -ENOEXEC;
1014         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1015         if (retval != sizeof(elf_ex))
1016                 goto out;
1017
1018         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1019                 goto out;
1020
1021         /* First of all, some simple consistency checks */
1022         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1023             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1024                 goto out;
1025
1026         /* Now read in all of the header information */
1027
1028         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1029         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1030
1031         error = -ENOMEM;
1032         elf_phdata = kmalloc(j, GFP_KERNEL);
1033         if (!elf_phdata)
1034                 goto out;
1035
1036         eppnt = elf_phdata;
1037         error = -ENOEXEC;
1038         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1039         if (retval != j)
1040                 goto out_free_ph;
1041
1042         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1043                 if ((eppnt + i)->p_type == PT_LOAD)
1044                         j++;
1045         if (j != 1)
1046                 goto out_free_ph;
1047
1048         while (eppnt->p_type != PT_LOAD)
1049                 eppnt++;
1050
1051         /* Now use mmap to map the library into memory. */
1052         error = vm_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         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1061                 goto out_free_ph;
1062
1063         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1064         if (padzero(elf_bss)) {
1065                 error = -EFAULT;
1066                 goto out_free_ph;
1067         }
1068
1069         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1070                             ELF_MIN_ALIGN - 1);
1071         bss = eppnt->p_memsz + eppnt->p_vaddr;
1072         if (bss > len)
1073                 vm_brk(len, bss - len);
1074         error = 0;
1075
1076 out_free_ph:
1077         kfree(elf_phdata);
1078 out:
1079         return error;
1080 }
1081
1082 #ifdef CONFIG_ELF_CORE
1083 /*
1084  * ELF core dumper
1085  *
1086  * Modelled on fs/exec.c:aout_core_dump()
1087  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1088  */
1089
1090 /*
1091  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1092  * that are useful for post-mortem analysis are included in every core dump.
1093  * In that way we ensure that the core dump is fully interpretable later
1094  * without matching up the same kernel and hardware config to see what PC values
1095  * meant. These special mappings include - vDSO, vsyscall, and other
1096  * architecture specific mappings
1097  */
1098 static bool always_dump_vma(struct vm_area_struct *vma)
1099 {
1100         /* Any vsyscall mappings? */
1101         if (vma == get_gate_vma(vma->vm_mm))
1102                 return true;
1103         /*
1104          * arch_vma_name() returns non-NULL for special architecture mappings,
1105          * such as vDSO sections.
1106          */
1107         if (arch_vma_name(vma))
1108                 return true;
1109
1110         return false;
1111 }
1112
1113 /*
1114  * Decide what to dump of a segment, part, all or none.
1115  */
1116 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1117                                    unsigned long mm_flags)
1118 {
1119 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1120
1121         /* always dump the vdso and vsyscall sections */
1122         if (always_dump_vma(vma))
1123                 goto whole;
1124
1125         if (vma->vm_flags & VM_DONTDUMP)
1126                 return 0;
1127
1128         /* Hugetlb memory check */
1129         if (vma->vm_flags & VM_HUGETLB) {
1130                 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1131                         goto whole;
1132                 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1133                         goto whole;
1134         }
1135
1136         /* Do not dump I/O mapped devices or special mappings */
1137         if (vma->vm_flags & VM_IO)
1138                 return 0;
1139
1140         /* By default, dump shared memory if mapped from an anonymous file. */
1141         if (vma->vm_flags & VM_SHARED) {
1142                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1143                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1144                         goto whole;
1145                 return 0;
1146         }
1147
1148         /* Dump segments that have been written to.  */
1149         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1150                 goto whole;
1151         if (vma->vm_file == NULL)
1152                 return 0;
1153
1154         if (FILTER(MAPPED_PRIVATE))
1155                 goto whole;
1156
1157         /*
1158          * If this looks like the beginning of a DSO or executable mapping,
1159          * check for an ELF header.  If we find one, dump the first page to
1160          * aid in determining what was mapped here.
1161          */
1162         if (FILTER(ELF_HEADERS) &&
1163             vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1164                 u32 __user *header = (u32 __user *) vma->vm_start;
1165                 u32 word;
1166                 mm_segment_t fs = get_fs();
1167                 /*
1168                  * Doing it this way gets the constant folded by GCC.
1169                  */
1170                 union {
1171                         u32 cmp;
1172                         char elfmag[SELFMAG];
1173                 } magic;
1174                 BUILD_BUG_ON(SELFMAG != sizeof word);
1175                 magic.elfmag[EI_MAG0] = ELFMAG0;
1176                 magic.elfmag[EI_MAG1] = ELFMAG1;
1177                 magic.elfmag[EI_MAG2] = ELFMAG2;
1178                 magic.elfmag[EI_MAG3] = ELFMAG3;
1179                 /*
1180                  * Switch to the user "segment" for get_user(),
1181                  * then put back what elf_core_dump() had in place.
1182                  */
1183                 set_fs(USER_DS);
1184                 if (unlikely(get_user(word, header)))
1185                         word = 0;
1186                 set_fs(fs);
1187                 if (word == magic.cmp)
1188                         return PAGE_SIZE;
1189         }
1190
1191 #undef  FILTER
1192
1193         return 0;
1194
1195 whole:
1196         return vma->vm_end - vma->vm_start;
1197 }
1198
1199 /* An ELF note in memory */
1200 struct memelfnote
1201 {
1202         const char *name;
1203         int type;
1204         unsigned int datasz;
1205         void *data;
1206 };
1207
1208 static int notesize(struct memelfnote *en)
1209 {
1210         int sz;
1211
1212         sz = sizeof(struct elf_note);
1213         sz += roundup(strlen(en->name) + 1, 4);
1214         sz += roundup(en->datasz, 4);
1215
1216         return sz;
1217 }
1218
1219 #define DUMP_WRITE(addr, nr, foffset)   \
1220         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1221
1222 static int alignfile(struct file *file, loff_t *foffset)
1223 {
1224         static const char buf[4] = { 0, };
1225         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1226         return 1;
1227 }
1228
1229 static int writenote(struct memelfnote *men, struct file *file,
1230                         loff_t *foffset)
1231 {
1232         struct elf_note en;
1233         en.n_namesz = strlen(men->name) + 1;
1234         en.n_descsz = men->datasz;
1235         en.n_type = men->type;
1236
1237         DUMP_WRITE(&en, sizeof(en), foffset);
1238         DUMP_WRITE(men->name, en.n_namesz, foffset);
1239         if (!alignfile(file, foffset))
1240                 return 0;
1241         DUMP_WRITE(men->data, men->datasz, foffset);
1242         if (!alignfile(file, foffset))
1243                 return 0;
1244
1245         return 1;
1246 }
1247 #undef DUMP_WRITE
1248
1249 static void fill_elf_header(struct elfhdr *elf, int segs,
1250                             u16 machine, u32 flags, u8 osabi)
1251 {
1252         memset(elf, 0, sizeof(*elf));
1253
1254         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1255         elf->e_ident[EI_CLASS] = ELF_CLASS;
1256         elf->e_ident[EI_DATA] = ELF_DATA;
1257         elf->e_ident[EI_VERSION] = EV_CURRENT;
1258         elf->e_ident[EI_OSABI] = ELF_OSABI;
1259
1260         elf->e_type = ET_CORE;
1261         elf->e_machine = machine;
1262         elf->e_version = EV_CURRENT;
1263         elf->e_phoff = sizeof(struct elfhdr);
1264         elf->e_flags = flags;
1265         elf->e_ehsize = sizeof(struct elfhdr);
1266         elf->e_phentsize = sizeof(struct elf_phdr);
1267         elf->e_phnum = segs;
1268
1269         return;
1270 }
1271
1272 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1273 {
1274         phdr->p_type = PT_NOTE;
1275         phdr->p_offset = offset;
1276         phdr->p_vaddr = 0;
1277         phdr->p_paddr = 0;
1278         phdr->p_filesz = sz;
1279         phdr->p_memsz = 0;
1280         phdr->p_flags = 0;
1281         phdr->p_align = 0;
1282         return;
1283 }
1284
1285 static void fill_note(struct memelfnote *note, const char *name, int type,
1286                 unsigned int sz, void *data)
1287 {
1288         note->name = name;
1289         note->type = type;
1290         note->datasz = sz;
1291         note->data = data;
1292         return;
1293 }
1294
1295 /*
1296  * fill up all the fields in prstatus from the given task struct, except
1297  * registers which need to be filled up separately.
1298  */
1299 static void fill_prstatus(struct elf_prstatus *prstatus,
1300                 struct task_struct *p, long signr)
1301 {
1302         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1303         prstatus->pr_sigpend = p->pending.signal.sig[0];
1304         prstatus->pr_sighold = p->blocked.sig[0];
1305         rcu_read_lock();
1306         prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1307         rcu_read_unlock();
1308         prstatus->pr_pid = task_pid_vnr(p);
1309         prstatus->pr_pgrp = task_pgrp_vnr(p);
1310         prstatus->pr_sid = task_session_vnr(p);
1311         if (thread_group_leader(p)) {
1312                 struct task_cputime cputime;
1313
1314                 /*
1315                  * This is the record for the group leader.  It shows the
1316                  * group-wide total, not its individual thread total.
1317                  */
1318                 thread_group_cputime(p, &cputime);
1319                 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1320                 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1321         } else {
1322                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1323                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1324         }
1325         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1326         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1327 }
1328
1329 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1330                        struct mm_struct *mm)
1331 {
1332         const struct cred *cred;
1333         unsigned int i, len;
1334
1335         /* first copy the parameters from user space */
1336         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1337
1338         len = mm->arg_end - mm->arg_start;
1339         if (len >= ELF_PRARGSZ)
1340                 len = ELF_PRARGSZ-1;
1341         if (copy_from_user(&psinfo->pr_psargs,
1342                            (const char __user *)mm->arg_start, len))
1343                 return -EFAULT;
1344         for(i = 0; i < len; i++)
1345                 if (psinfo->pr_psargs[i] == 0)
1346                         psinfo->pr_psargs[i] = ' ';
1347         psinfo->pr_psargs[len] = 0;
1348
1349         rcu_read_lock();
1350         psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1351         rcu_read_unlock();
1352         psinfo->pr_pid = task_pid_vnr(p);
1353         psinfo->pr_pgrp = task_pgrp_vnr(p);
1354         psinfo->pr_sid = task_session_vnr(p);
1355
1356         i = p->state ? ffz(~p->state) + 1 : 0;
1357         psinfo->pr_state = i;
1358         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1359         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1360         psinfo->pr_nice = task_nice(p);
1361         psinfo->pr_flag = p->flags;
1362         rcu_read_lock();
1363         cred = __task_cred(p);
1364         SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1365         SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1366         rcu_read_unlock();
1367         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1368
1369         return 0;
1370 }
1371
1372 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1373 {
1374         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1375         int i = 0;
1376         do
1377                 i += 2;
1378         while (auxv[i - 2] != AT_NULL);
1379         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1380 }
1381
1382 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1383                 siginfo_t *siginfo)
1384 {
1385         mm_segment_t old_fs = get_fs();
1386         set_fs(KERNEL_DS);
1387         copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1388         set_fs(old_fs);
1389         fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1390 }
1391
1392 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1393 /*
1394  * Format of NT_FILE note:
1395  *
1396  * long count     -- how many files are mapped
1397  * long page_size -- units for file_ofs
1398  * array of [COUNT] elements of
1399  *   long start
1400  *   long end
1401  *   long file_ofs
1402  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1403  */
1404 static void fill_files_note(struct memelfnote *note)
1405 {
1406         struct vm_area_struct *vma;
1407         unsigned count, size, names_ofs, remaining, n;
1408         user_long_t *data;
1409         user_long_t *start_end_ofs;
1410         char *name_base, *name_curpos;
1411
1412         /* *Estimated* file count and total data size needed */
1413         count = current->mm->map_count;
1414         size = count * 64;
1415
1416         names_ofs = (2 + 3 * count) * sizeof(data[0]);
1417  alloc:
1418         if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1419                 goto err;
1420         size = round_up(size, PAGE_SIZE);
1421         data = vmalloc(size);
1422         if (!data)
1423                 goto err;
1424
1425         start_end_ofs = data + 2;
1426         name_base = name_curpos = ((char *)data) + names_ofs;
1427         remaining = size - names_ofs;
1428         count = 0;
1429         for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1430                 struct file *file;
1431                 const char *filename;
1432
1433                 file = vma->vm_file;
1434                 if (!file)
1435                         continue;
1436                 filename = d_path(&file->f_path, name_curpos, remaining);
1437                 if (IS_ERR(filename)) {
1438                         if (PTR_ERR(filename) == -ENAMETOOLONG) {
1439                                 vfree(data);
1440                                 size = size * 5 / 4;
1441                                 goto alloc;
1442                         }
1443                         continue;
1444                 }
1445
1446                 /* d_path() fills at the end, move name down */
1447                 /* n = strlen(filename) + 1: */
1448                 n = (name_curpos + remaining) - filename;
1449                 remaining = filename - name_curpos;
1450                 memmove(name_curpos, filename, n);
1451                 name_curpos += n;
1452
1453                 *start_end_ofs++ = vma->vm_start;
1454                 *start_end_ofs++ = vma->vm_end;
1455                 *start_end_ofs++ = vma->vm_pgoff;
1456                 count++;
1457         }
1458
1459         /* Now we know exact count of files, can store it */
1460         data[0] = count;
1461         data[1] = PAGE_SIZE;
1462         /*
1463          * Count usually is less than current->mm->map_count,
1464          * we need to move filenames down.
1465          */
1466         n = current->mm->map_count - count;
1467         if (n != 0) {
1468                 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1469                 memmove(name_base - shift_bytes, name_base,
1470                         name_curpos - name_base);
1471                 name_curpos -= shift_bytes;
1472         }
1473
1474         size = name_curpos - (char *)data;
1475         fill_note(note, "CORE", NT_FILE, size, data);
1476  err: ;
1477 }
1478
1479 #ifdef CORE_DUMP_USE_REGSET
1480 #include <linux/regset.h>
1481
1482 struct elf_thread_core_info {
1483         struct elf_thread_core_info *next;
1484         struct task_struct *task;
1485         struct elf_prstatus prstatus;
1486         struct memelfnote notes[0];
1487 };
1488
1489 struct elf_note_info {
1490         struct elf_thread_core_info *thread;
1491         struct memelfnote psinfo;
1492         struct memelfnote signote;
1493         struct memelfnote auxv;
1494         struct memelfnote files;
1495         user_siginfo_t csigdata;
1496         size_t size;
1497         int thread_notes;
1498 };
1499
1500 /*
1501  * When a regset has a writeback hook, we call it on each thread before
1502  * dumping user memory.  On register window machines, this makes sure the
1503  * user memory backing the register data is up to date before we read it.
1504  */
1505 static void do_thread_regset_writeback(struct task_struct *task,
1506                                        const struct user_regset *regset)
1507 {
1508         if (regset->writeback)
1509                 regset->writeback(task, regset, 1);
1510 }
1511
1512 #ifndef PR_REG_SIZE
1513 #define PR_REG_SIZE(S) sizeof(S)
1514 #endif
1515
1516 #ifndef PRSTATUS_SIZE
1517 #define PRSTATUS_SIZE(S) sizeof(S)
1518 #endif
1519
1520 #ifndef PR_REG_PTR
1521 #define PR_REG_PTR(S) (&((S)->pr_reg))
1522 #endif
1523
1524 #ifndef SET_PR_FPVALID
1525 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1526 #endif
1527
1528 static int fill_thread_core_info(struct elf_thread_core_info *t,
1529                                  const struct user_regset_view *view,
1530                                  long signr, size_t *total)
1531 {
1532         unsigned int i;
1533
1534         /*
1535          * NT_PRSTATUS is the one special case, because the regset data
1536          * goes into the pr_reg field inside the note contents, rather
1537          * than being the whole note contents.  We fill the reset in here.
1538          * We assume that regset 0 is NT_PRSTATUS.
1539          */
1540         fill_prstatus(&t->prstatus, t->task, signr);
1541         (void) view->regsets[0].get(t->task, &view->regsets[0],
1542                                     0, PR_REG_SIZE(t->prstatus.pr_reg),
1543                                     PR_REG_PTR(&t->prstatus), NULL);
1544
1545         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1546                   PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1547         *total += notesize(&t->notes[0]);
1548
1549         do_thread_regset_writeback(t->task, &view->regsets[0]);
1550
1551         /*
1552          * Each other regset might generate a note too.  For each regset
1553          * that has no core_note_type or is inactive, we leave t->notes[i]
1554          * all zero and we'll know to skip writing it later.
1555          */
1556         for (i = 1; i < view->n; ++i) {
1557                 const struct user_regset *regset = &view->regsets[i];
1558                 do_thread_regset_writeback(t->task, regset);
1559                 if (regset->core_note_type && regset->get &&
1560                     (!regset->active || regset->active(t->task, regset))) {
1561                         int ret;
1562                         size_t size = regset->n * regset->size;
1563                         void *data = kmalloc(size, GFP_KERNEL);
1564                         if (unlikely(!data))
1565                                 return 0;
1566                         ret = regset->get(t->task, regset,
1567                                           0, size, data, NULL);
1568                         if (unlikely(ret))
1569                                 kfree(data);
1570                         else {
1571                                 if (regset->core_note_type != NT_PRFPREG)
1572                                         fill_note(&t->notes[i], "LINUX",
1573                                                   regset->core_note_type,
1574                                                   size, data);
1575                                 else {
1576                                         SET_PR_FPVALID(&t->prstatus, 1);
1577                                         fill_note(&t->notes[i], "CORE",
1578                                                   NT_PRFPREG, size, data);
1579                                 }
1580                                 *total += notesize(&t->notes[i]);
1581                         }
1582                 }
1583         }
1584
1585         return 1;
1586 }
1587
1588 static int fill_note_info(struct elfhdr *elf, int phdrs,
1589                           struct elf_note_info *info,
1590                           siginfo_t *siginfo, struct pt_regs *regs)
1591 {
1592         struct task_struct *dump_task = current;
1593         const struct user_regset_view *view = task_user_regset_view(dump_task);
1594         struct elf_thread_core_info *t;
1595         struct elf_prpsinfo *psinfo;
1596         struct core_thread *ct;
1597         unsigned int i;
1598
1599         info->size = 0;
1600         info->thread = NULL;
1601
1602         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1603         if (psinfo == NULL)
1604                 return 0;
1605
1606         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1607
1608         /*
1609          * Figure out how many notes we're going to need for each thread.
1610          */
1611         info->thread_notes = 0;
1612         for (i = 0; i < view->n; ++i)
1613                 if (view->regsets[i].core_note_type != 0)
1614                         ++info->thread_notes;
1615
1616         /*
1617          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1618          * since it is our one special case.
1619          */
1620         if (unlikely(info->thread_notes == 0) ||
1621             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1622                 WARN_ON(1);
1623                 return 0;
1624         }
1625
1626         /*
1627          * Initialize the ELF file header.
1628          */
1629         fill_elf_header(elf, phdrs,
1630                         view->e_machine, view->e_flags, view->ei_osabi);
1631
1632         /*
1633          * Allocate a structure for each thread.
1634          */
1635         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1636                 t = kzalloc(offsetof(struct elf_thread_core_info,
1637                                      notes[info->thread_notes]),
1638                             GFP_KERNEL);
1639                 if (unlikely(!t))
1640                         return 0;
1641
1642                 t->task = ct->task;
1643                 if (ct->task == dump_task || !info->thread) {
1644                         t->next = info->thread;
1645                         info->thread = t;
1646                 } else {
1647                         /*
1648                          * Make sure to keep the original task at
1649                          * the head of the list.
1650                          */
1651                         t->next = info->thread->next;
1652                         info->thread->next = t;
1653                 }
1654         }
1655
1656         /*
1657          * Now fill in each thread's information.
1658          */
1659         for (t = info->thread; t != NULL; t = t->next)
1660                 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1661                         return 0;
1662
1663         /*
1664          * Fill in the two process-wide notes.
1665          */
1666         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1667         info->size += notesize(&info->psinfo);
1668
1669         fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1670         info->size += notesize(&info->signote);
1671
1672         fill_auxv_note(&info->auxv, current->mm);
1673         info->size += notesize(&info->auxv);
1674
1675         fill_files_note(&info->files);
1676         info->size += notesize(&info->files);
1677
1678         return 1;
1679 }
1680
1681 static size_t get_note_info_size(struct elf_note_info *info)
1682 {
1683         return info->size;
1684 }
1685
1686 /*
1687  * Write all the notes for each thread.  When writing the first thread, the
1688  * process-wide notes are interleaved after the first thread-specific note.
1689  */
1690 static int write_note_info(struct elf_note_info *info,
1691                            struct file *file, loff_t *foffset)
1692 {
1693         bool first = 1;
1694         struct elf_thread_core_info *t = info->thread;
1695
1696         do {
1697                 int i;
1698
1699                 if (!writenote(&t->notes[0], file, foffset))
1700                         return 0;
1701
1702                 if (first && !writenote(&info->psinfo, file, foffset))
1703                         return 0;
1704                 if (first && !writenote(&info->signote, file, foffset))
1705                         return 0;
1706                 if (first && !writenote(&info->auxv, file, foffset))
1707                         return 0;
1708                 if (first && !writenote(&info->files, file, foffset))
1709                         return 0;
1710
1711                 for (i = 1; i < info->thread_notes; ++i)
1712                         if (t->notes[i].data &&
1713                             !writenote(&t->notes[i], file, foffset))
1714                                 return 0;
1715
1716                 first = 0;
1717                 t = t->next;
1718         } while (t);
1719
1720         return 1;
1721 }
1722
1723 static void free_note_info(struct elf_note_info *info)
1724 {
1725         struct elf_thread_core_info *threads = info->thread;
1726         while (threads) {
1727                 unsigned int i;
1728                 struct elf_thread_core_info *t = threads;
1729                 threads = t->next;
1730                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1731                 for (i = 1; i < info->thread_notes; ++i)
1732                         kfree(t->notes[i].data);
1733                 kfree(t);
1734         }
1735         kfree(info->psinfo.data);
1736         vfree(info->files.data);
1737 }
1738
1739 #else
1740
1741 /* Here is the structure in which status of each thread is captured. */
1742 struct elf_thread_status
1743 {
1744         struct list_head list;
1745         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1746         elf_fpregset_t fpu;             /* NT_PRFPREG */
1747         struct task_struct *thread;
1748 #ifdef ELF_CORE_COPY_XFPREGS
1749         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1750 #endif
1751         struct memelfnote notes[3];
1752         int num_notes;
1753 };
1754
1755 /*
1756  * In order to add the specific thread information for the elf file format,
1757  * we need to keep a linked list of every threads pr_status and then create
1758  * a single section for them in the final core file.
1759  */
1760 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1761 {
1762         int sz = 0;
1763         struct task_struct *p = t->thread;
1764         t->num_notes = 0;
1765
1766         fill_prstatus(&t->prstatus, p, signr);
1767         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1768
1769         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1770                   &(t->prstatus));
1771         t->num_notes++;
1772         sz += notesize(&t->notes[0]);
1773
1774         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1775                                                                 &t->fpu))) {
1776                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1777                           &(t->fpu));
1778                 t->num_notes++;
1779                 sz += notesize(&t->notes[1]);
1780         }
1781
1782 #ifdef ELF_CORE_COPY_XFPREGS
1783         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1784                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1785                           sizeof(t->xfpu), &t->xfpu);
1786                 t->num_notes++;
1787                 sz += notesize(&t->notes[2]);
1788         }
1789 #endif
1790         return sz;
1791 }
1792
1793 struct elf_note_info {
1794         struct memelfnote *notes;
1795         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1796         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1797         struct list_head thread_list;
1798         elf_fpregset_t *fpu;
1799 #ifdef ELF_CORE_COPY_XFPREGS
1800         elf_fpxregset_t *xfpu;
1801 #endif
1802         user_siginfo_t csigdata;
1803         int thread_status_size;
1804         int numnote;
1805 };
1806
1807 static int elf_note_info_init(struct elf_note_info *info)
1808 {
1809         memset(info, 0, sizeof(*info));
1810         INIT_LIST_HEAD(&info->thread_list);
1811
1812         /* Allocate space for ELF notes */
1813         info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1814         if (!info->notes)
1815                 return 0;
1816         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1817         if (!info->psinfo)
1818                 return 0;
1819         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1820         if (!info->prstatus)
1821                 return 0;
1822         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1823         if (!info->fpu)
1824                 return 0;
1825 #ifdef ELF_CORE_COPY_XFPREGS
1826         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1827         if (!info->xfpu)
1828                 return 0;
1829 #endif
1830         return 1;
1831 }
1832
1833 static int fill_note_info(struct elfhdr *elf, int phdrs,
1834                           struct elf_note_info *info,
1835                           siginfo_t *siginfo, struct pt_regs *regs)
1836 {
1837         struct list_head *t;
1838
1839         if (!elf_note_info_init(info))
1840                 return 0;
1841
1842         if (siginfo->si_signo) {
1843                 struct core_thread *ct;
1844                 struct elf_thread_status *ets;
1845
1846                 for (ct = current->mm->core_state->dumper.next;
1847                                                 ct; ct = ct->next) {
1848                         ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1849                         if (!ets)
1850                                 return 0;
1851
1852                         ets->thread = ct->task;
1853                         list_add(&ets->list, &info->thread_list);
1854                 }
1855
1856                 list_for_each(t, &info->thread_list) {
1857                         int sz;
1858
1859                         ets = list_entry(t, struct elf_thread_status, list);
1860                         sz = elf_dump_thread_status(siginfo->si_signo, ets);
1861                         info->thread_status_size += sz;
1862                 }
1863         }
1864         /* now collect the dump for the current */
1865         memset(info->prstatus, 0, sizeof(*info->prstatus));
1866         fill_prstatus(info->prstatus, current, siginfo->si_signo);
1867         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1868
1869         /* Set up header */
1870         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1871
1872         /*
1873          * Set up the notes in similar form to SVR4 core dumps made
1874          * with info from their /proc.
1875          */
1876
1877         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1878                   sizeof(*info->prstatus), info->prstatus);
1879         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1880         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1881                   sizeof(*info->psinfo), info->psinfo);
1882
1883         fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1884         fill_auxv_note(info->notes + 3, current->mm);
1885         fill_files_note(info->notes + 4);
1886
1887         info->numnote = 5;
1888
1889         /* Try to dump the FPU. */
1890         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1891                                                                info->fpu);
1892         if (info->prstatus->pr_fpvalid)
1893                 fill_note(info->notes + info->numnote++,
1894                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1895 #ifdef ELF_CORE_COPY_XFPREGS
1896         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1897                 fill_note(info->notes + info->numnote++,
1898                           "LINUX", ELF_CORE_XFPREG_TYPE,
1899                           sizeof(*info->xfpu), info->xfpu);
1900 #endif
1901
1902         return 1;
1903 }
1904
1905 static size_t get_note_info_size(struct elf_note_info *info)
1906 {
1907         int sz = 0;
1908         int i;
1909
1910         for (i = 0; i < info->numnote; i++)
1911                 sz += notesize(info->notes + i);
1912
1913         sz += info->thread_status_size;
1914
1915         return sz;
1916 }
1917
1918 static int write_note_info(struct elf_note_info *info,
1919                            struct file *file, loff_t *foffset)
1920 {
1921         int i;
1922         struct list_head *t;
1923
1924         for (i = 0; i < info->numnote; i++)
1925                 if (!writenote(info->notes + i, file, foffset))
1926                         return 0;
1927
1928         /* write out the thread status notes section */
1929         list_for_each(t, &info->thread_list) {
1930                 struct elf_thread_status *tmp =
1931                                 list_entry(t, struct elf_thread_status, list);
1932
1933                 for (i = 0; i < tmp->num_notes; i++)
1934                         if (!writenote(&tmp->notes[i], file, foffset))
1935                                 return 0;
1936         }
1937
1938         return 1;
1939 }
1940
1941 static void free_note_info(struct elf_note_info *info)
1942 {
1943         while (!list_empty(&info->thread_list)) {
1944                 struct list_head *tmp = info->thread_list.next;
1945                 list_del(tmp);
1946                 kfree(list_entry(tmp, struct elf_thread_status, list));
1947         }
1948
1949         /* Free data allocated by fill_files_note(): */
1950         vfree(info->notes[4].data);
1951
1952         kfree(info->prstatus);
1953         kfree(info->psinfo);
1954         kfree(info->notes);
1955         kfree(info->fpu);
1956 #ifdef ELF_CORE_COPY_XFPREGS
1957         kfree(info->xfpu);
1958 #endif
1959 }
1960
1961 #endif
1962
1963 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1964                                         struct vm_area_struct *gate_vma)
1965 {
1966         struct vm_area_struct *ret = tsk->mm->mmap;
1967
1968         if (ret)
1969                 return ret;
1970         return gate_vma;
1971 }
1972 /*
1973  * Helper function for iterating across a vma list.  It ensures that the caller
1974  * will visit `gate_vma' prior to terminating the search.
1975  */
1976 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1977                                         struct vm_area_struct *gate_vma)
1978 {
1979         struct vm_area_struct *ret;
1980
1981         ret = this_vma->vm_next;
1982         if (ret)
1983                 return ret;
1984         if (this_vma == gate_vma)
1985                 return NULL;
1986         return gate_vma;
1987 }
1988
1989 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1990                              elf_addr_t e_shoff, int segs)
1991 {
1992         elf->e_shoff = e_shoff;
1993         elf->e_shentsize = sizeof(*shdr4extnum);
1994         elf->e_shnum = 1;
1995         elf->e_shstrndx = SHN_UNDEF;
1996
1997         memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1998
1999         shdr4extnum->sh_type = SHT_NULL;
2000         shdr4extnum->sh_size = elf->e_shnum;
2001         shdr4extnum->sh_link = elf->e_shstrndx;
2002         shdr4extnum->sh_info = segs;
2003 }
2004
2005 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2006                                      unsigned long mm_flags)
2007 {
2008         struct vm_area_struct *vma;
2009         size_t size = 0;
2010
2011         for (vma = first_vma(current, gate_vma); vma != NULL;
2012              vma = next_vma(vma, gate_vma))
2013                 size += vma_dump_size(vma, mm_flags);
2014         return size;
2015 }
2016
2017 /*
2018  * Actual dumper
2019  *
2020  * This is a two-pass process; first we find the offsets of the bits,
2021  * and then they are actually written out.  If we run out of core limit
2022  * we just truncate.
2023  */
2024 static int elf_core_dump(struct coredump_params *cprm)
2025 {
2026         int has_dumped = 0;
2027         mm_segment_t fs;
2028         int segs;
2029         size_t size = 0;
2030         struct vm_area_struct *vma, *gate_vma;
2031         struct elfhdr *elf = NULL;
2032         loff_t offset = 0, dataoff, foffset;
2033         struct elf_note_info info;
2034         struct elf_phdr *phdr4note = NULL;
2035         struct elf_shdr *shdr4extnum = NULL;
2036         Elf_Half e_phnum;
2037         elf_addr_t e_shoff;
2038
2039         /*
2040          * We no longer stop all VM operations.
2041          *
2042          * This is because those proceses that could possibly change map_count
2043          * or the mmap / vma pages are now blocked in do_exit on current
2044          * finishing this core dump.
2045          *
2046          * Only ptrace can touch these memory addresses, but it doesn't change
2047          * the map_count or the pages allocated. So no possibility of crashing
2048          * exists while dumping the mm->vm_next areas to the core file.
2049          */
2050
2051         /* alloc memory for large data structures: too large to be on stack */
2052         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2053         if (!elf)
2054                 goto out;
2055         /*
2056          * The number of segs are recored into ELF header as 16bit value.
2057          * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2058          */
2059         segs = current->mm->map_count;
2060         segs += elf_core_extra_phdrs();
2061
2062         gate_vma = get_gate_vma(current->mm);
2063         if (gate_vma != NULL)
2064                 segs++;
2065
2066         /* for notes section */
2067         segs++;
2068
2069         /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2070          * this, kernel supports extended numbering. Have a look at
2071          * include/linux/elf.h for further information. */
2072         e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2073
2074         /*
2075          * Collect all the non-memory information about the process for the
2076          * notes.  This also sets up the file header.
2077          */
2078         if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2079                 goto cleanup;
2080
2081         has_dumped = 1;
2082         current->flags |= PF_DUMPCORE;
2083
2084         fs = get_fs();
2085         set_fs(KERNEL_DS);
2086
2087         offset += sizeof(*elf);                         /* Elf header */
2088         offset += segs * sizeof(struct elf_phdr);       /* Program headers */
2089         foffset = offset;
2090
2091         /* Write notes phdr entry */
2092         {
2093                 size_t sz = get_note_info_size(&info);
2094
2095                 sz += elf_coredump_extra_notes_size();
2096
2097                 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2098                 if (!phdr4note)
2099                         goto end_coredump;
2100
2101                 fill_elf_note_phdr(phdr4note, sz, offset);
2102                 offset += sz;
2103         }
2104
2105         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2106
2107         offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2108         offset += elf_core_extra_data_size();
2109         e_shoff = offset;
2110
2111         if (e_phnum == PN_XNUM) {
2112                 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2113                 if (!shdr4extnum)
2114                         goto end_coredump;
2115                 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2116         }
2117
2118         offset = dataoff;
2119
2120         size += sizeof(*elf);
2121         if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2122                 goto end_coredump;
2123
2124         size += sizeof(*phdr4note);
2125         if (size > cprm->limit
2126             || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2127                 goto end_coredump;
2128
2129         /* Write program headers for segments dump */
2130         for (vma = first_vma(current, gate_vma); vma != NULL;
2131                         vma = next_vma(vma, gate_vma)) {
2132                 struct elf_phdr phdr;
2133
2134                 phdr.p_type = PT_LOAD;
2135                 phdr.p_offset = offset;
2136                 phdr.p_vaddr = vma->vm_start;
2137                 phdr.p_paddr = 0;
2138                 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2139                 phdr.p_memsz = vma->vm_end - vma->vm_start;
2140                 offset += phdr.p_filesz;
2141                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2142                 if (vma->vm_flags & VM_WRITE)
2143                         phdr.p_flags |= PF_W;
2144                 if (vma->vm_flags & VM_EXEC)
2145                         phdr.p_flags |= PF_X;
2146                 phdr.p_align = ELF_EXEC_PAGESIZE;
2147
2148                 size += sizeof(phdr);
2149                 if (size > cprm->limit
2150                     || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2151                         goto end_coredump;
2152         }
2153
2154         if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2155                 goto end_coredump;
2156
2157         /* write out the notes section */
2158         if (!write_note_info(&info, cprm->file, &foffset))
2159                 goto end_coredump;
2160
2161         if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2162                 goto end_coredump;
2163
2164         /* Align to page */
2165         if (!dump_seek(cprm->file, dataoff - foffset))
2166                 goto end_coredump;
2167
2168         for (vma = first_vma(current, gate_vma); vma != NULL;
2169                         vma = next_vma(vma, gate_vma)) {
2170                 unsigned long addr;
2171                 unsigned long end;
2172
2173                 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2174
2175                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2176                         struct page *page;
2177                         int stop;
2178
2179                         page = get_dump_page(addr);
2180                         if (page) {
2181                                 void *kaddr = kmap(page);
2182                                 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2183                                         !dump_write(cprm->file, kaddr,
2184                                                     PAGE_SIZE);
2185                                 kunmap(page);
2186                                 page_cache_release(page);
2187                         } else
2188                                 stop = !dump_seek(cprm->file, PAGE_SIZE);
2189                         if (stop)
2190                                 goto end_coredump;
2191                 }
2192         }
2193
2194         if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2195                 goto end_coredump;
2196
2197         if (e_phnum == PN_XNUM) {
2198                 size += sizeof(*shdr4extnum);
2199                 if (size > cprm->limit
2200                     || !dump_write(cprm->file, shdr4extnum,
2201                                    sizeof(*shdr4extnum)))
2202                         goto end_coredump;
2203         }
2204
2205 end_coredump:
2206         set_fs(fs);
2207
2208 cleanup:
2209         free_note_info(&info);
2210         kfree(shdr4extnum);
2211         kfree(phdr4note);
2212         kfree(elf);
2213 out:
2214         return has_dumped;
2215 }
2216
2217 #endif          /* CONFIG_ELF_CORE */
2218
2219 static int __init init_elf_binfmt(void)
2220 {
2221         register_binfmt(&elf_format);
2222         return 0;
2223 }
2224
2225 static void __exit exit_elf_binfmt(void)
2226 {
2227         /* Remove the COFF and ELF loaders. */
2228         unregister_binfmt(&elf_format);
2229 }
2230
2231 core_initcall(init_elf_binfmt);
2232 module_exit(exit_elf_binfmt);
2233 MODULE_LICENSE("GPL");