fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/swap.h>
  32 #include <linux/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/mount.h>
  46 #include <linux/security.h>
  47 #include <linux/syscalls.h>
  48 #include <linux/tsacct_kern.h>
  49 #include <linux/cn_proc.h>
  50 #include <linux/audit.h>
  51 #include <linux/tracehook.h>
  52 #include <linux/kmod.h>
  53 #include <linux/fsnotify.h>
  54 #include <linux/fs_struct.h>
  55 #include <linux/pipe_fs_i.h>
  56 #include <linux/oom.h>
  57 #include <linux/compat.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include <asm/exec.h>
  63
  64 #include <trace/events/task.h>
  65 #include "internal.h"
  66
  67 #include <trace/events/sched.h>
  68
  69 int core_uses_pid;
  70 char core_pattern[CORENAME_MAX_SIZE] = "core";
  71 unsigned int core_pipe_limit;
  72 int suid_dumpable = 0;
  73
  74 struct core_name {
  75         char *corename;
  76         int used, size;
  77 };
  78 static atomic_t call_count = ATOMIC_INIT(1);
  79
  80 /* The maximal length of core_pattern is also specified in sysctl.c */
  81
  82 static LIST_HEAD(formats);
  83 static DEFINE_RWLOCK(binfmt_lock);
  84
  85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  86 {
  87         BUG_ON(!fmt);
  88         write_lock(&binfmt_lock);
  89         insert ? list_add(&fmt->lh, &formats) :
  90                  list_add_tail(&fmt->lh, &formats);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(__register_binfmt);
  95
  96 void unregister_binfmt(struct linux_binfmt * fmt)
  97 {
  98         write_lock(&binfmt_lock);
  99         list_del(&fmt->lh);
 100         write_unlock(&binfmt_lock);
 101 }
 102
 103 EXPORT_SYMBOL(unregister_binfmt);
 104
 105 static inline void put_binfmt(struct linux_binfmt * fmt)
 106 {
 107         module_put(fmt->module);
 108 }
 109
 110 /*
 111  * Note that a shared library must be both readable and executable due to
 112  * security reasons.
 113  *
 114  * Also note that we take the address to load from from the file itself.
 115  */
 116 SYSCALL_DEFINE1(uselib, const char __user *, library)
 117 {
 118         struct file *file;
 119         char *tmp = getname(library);
 120         int error = PTR_ERR(tmp);
 121         static const struct open_flags uselib_flags = {
 122                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 123                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 124                 .intent = LOOKUP_OPEN
 125         };
 126
 127         if (IS_ERR(tmp))
 128                 goto out;
 129
 130         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 131         putname(tmp);
 132         error = PTR_ERR(file);
 133         if (IS_ERR(file))
 134                 goto out;
 135
 136         error = -EINVAL;
 137         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 138                 goto exit;
 139
 140         error = -EACCES;
 141         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 142                 goto exit;
 143
 144         fsnotify_open(file);
 145
 146         error = -ENOEXEC;
 147         if(file->f_op) {
 148                 struct linux_binfmt * fmt;
 149
 150                 read_lock(&binfmt_lock);
 151                 list_for_each_entry(fmt, &formats, lh) {
 152                         if (!fmt->load_shlib)
 153                                 continue;
 154                         if (!try_module_get(fmt->module))
 155                                 continue;
 156                         read_unlock(&binfmt_lock);
 157                         error = fmt->load_shlib(file);
 158                         read_lock(&binfmt_lock);
 159                         put_binfmt(fmt);
 160                         if (error != -ENOEXEC)
 161                                 break;
 162                 }
 163                 read_unlock(&binfmt_lock);
 164         }
 165 exit:
 166         fput(file);
 167 out:
 168         return error;
 169 }
 170
 171 #ifdef CONFIG_MMU
 172 /*
 173  * The nascent bprm->mm is not visible until exec_mmap() but it can
 174  * use a lot of memory, account these pages in current->mm temporary
 175  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 176  * change the counter back via acct_arg_size(0).
 177  */
 178 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 179 {
 180         struct mm_struct *mm = current->mm;
 181         long diff = (long)(pages - bprm->vma_pages);
 182
 183         if (!mm || !diff)
 184                 return;
 185
 186         bprm->vma_pages = pages;
 187         add_mm_counter(mm, MM_ANONPAGES, diff);
 188 }
 189
 190 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 191                 int write)
 192 {
 193         struct page *page;
 194         int ret;
 195
 196 #ifdef CONFIG_STACK_GROWSUP
 197         if (write) {
 198                 ret = expand_downwards(bprm->vma, pos);
 199                 if (ret < 0)
 200                         return NULL;
 201         }
 202 #endif
 203         ret = get_user_pages(current, bprm->mm, pos,
 204                         1, write, 1, &page, NULL);
 205         if (ret <= 0)
 206                 return NULL;
 207
 208         if (write) {
 209                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 210                 struct rlimit *rlim;
 211
 212                 acct_arg_size(bprm, size / PAGE_SIZE);
 213
 214                 /*
 215                  * We've historically supported up to 32 pages (ARG_MAX)
 216                  * of argument strings even with small stacks
 217                  */
 218                 if (size <= ARG_MAX)
 219                         return page;
 220
 221                 /*
 222                  * Limit to 1/4-th the stack size for the argv+env strings.
 223                  * This ensures that:
 224                  *  - the remaining binfmt code will not run out of stack space,
 225                  *  - the program will have a reasonable amount of stack left
 226                  *    to work from.
 227                  */
 228                 rlim = current->signal->rlim;
 229                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 230                         put_page(page);
 231                         return NULL;
 232                 }
 233         }
 234
 235         return page;
 236 }
 237
 238 static void put_arg_page(struct page *page)
 239 {
 240         put_page(page);
 241 }
 242
 243 static void free_arg_page(struct linux_binprm *bprm, int i)
 244 {
 245 }
 246
 247 static void free_arg_pages(struct linux_binprm *bprm)
 248 {
 249 }
 250
 251 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 252                 struct page *page)
 253 {
 254         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 255 }
 256
 257 static int __bprm_mm_init(struct linux_binprm *bprm)
 258 {
 259         int err;
 260         struct vm_area_struct *vma = NULL;
 261         struct mm_struct *mm = bprm->mm;
 262
 263         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 264         if (!vma)
 265                 return -ENOMEM;
 266
 267         down_write(&mm->mmap_sem);
 268         vma->vm_mm = mm;
 269
 270         /*
 271          * Place the stack at the largest stack address the architecture
 272          * supports. Later, we'll move this to an appropriate place. We don't
 273          * use STACK_TOP because that can depend on attributes which aren't
 274          * configured yet.
 275          */
 276         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 277         vma->vm_end = STACK_TOP_MAX;
 278         vma->vm_start = vma->vm_end - PAGE_SIZE;
 279         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 280         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 281         INIT_LIST_HEAD(&vma->anon_vma_chain);
 282
 283         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 284         if (err)
 285                 goto err;
 286
 287         err = insert_vm_struct(mm, vma);
 288         if (err)
 289                 goto err;
 290
 291         mm->stack_vm = mm->total_vm = 1;
 292         up_write(&mm->mmap_sem);
 293         bprm->p = vma->vm_end - sizeof(void *);
 294         return 0;
 295 err:
 296         up_write(&mm->mmap_sem);
 297         bprm->vma = NULL;
 298         kmem_cache_free(vm_area_cachep, vma);
 299         return err;
 300 }
 301
 302 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 303 {
 304         return len <= MAX_ARG_STRLEN;
 305 }
 306
 307 #else
 308
 309 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 310 {
 311 }
 312
 313 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 314                 int write)
 315 {
 316         struct page *page;
 317
 318         page = bprm->page[pos / PAGE_SIZE];
 319         if (!page && write) {
 320                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 321                 if (!page)
 322                         return NULL;
 323                 bprm->page[pos / PAGE_SIZE] = page;
 324         }
 325
 326         return page;
 327 }
 328
 329 static void put_arg_page(struct page *page)
 330 {
 331 }
 332
 333 static void free_arg_page(struct linux_binprm *bprm, int i)
 334 {
 335         if (bprm->page[i]) {
 336                 __free_page(bprm->page[i]);
 337                 bprm->page[i] = NULL;
 338         }
 339 }
 340
 341 static void free_arg_pages(struct linux_binprm *bprm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < MAX_ARG_PAGES; i++)
 346                 free_arg_page(bprm, i);
 347 }
 348
 349 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 350                 struct page *page)
 351 {
 352 }
 353
 354 static int __bprm_mm_init(struct linux_binprm *bprm)
 355 {
 356         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 357         return 0;
 358 }
 359
 360 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 361 {
 362         return len <= bprm->p;
 363 }
 364
 365 #endif /* CONFIG_MMU */
 366
 367 /*
 368  * Create a new mm_struct and populate it with a temporary stack
 369  * vm_area_struct.  We don't have enough context at this point to set the stack
 370  * flags, permissions, and offset, so we use temporary values.  We'll update
 371  * them later in setup_arg_pages().
 372  */
 373 int bprm_mm_init(struct linux_binprm *bprm)
 374 {
 375         int err;
 376         struct mm_struct *mm = NULL;
 377
 378         bprm->mm = mm = mm_alloc();
 379         err = -ENOMEM;
 380         if (!mm)
 381                 goto err;
 382
 383         err = init_new_context(current, mm);
 384         if (err)
 385                 goto err;
 386
 387         err = __bprm_mm_init(bprm);
 388         if (err)
 389                 goto err;
 390
 391         return 0;
 392
 393 err:
 394         if (mm) {
 395                 bprm->mm = NULL;
 396                 mmdrop(mm);
 397         }
 398
 399         return err;
 400 }
 401
 402 struct user_arg_ptr {
 403 #ifdef CONFIG_COMPAT
 404         bool is_compat;
 405 #endif
 406         union {
 407                 const char __user *const __user *native;
 408 #ifdef CONFIG_COMPAT
 409                 compat_uptr_t __user *compat;
 410 #endif
 411         } ptr;
 412 };
 413
 414 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 415 {
 416         const char __user *native;
 417
 418 #ifdef CONFIG_COMPAT
 419         if (unlikely(argv.is_compat)) {
 420                 compat_uptr_t compat;
 421
 422                 if (get_user(compat, argv.ptr.compat + nr))
 423                         return ERR_PTR(-EFAULT);
 424
 425                 return compat_ptr(compat);
 426         }
 427 #endif
 428
 429         if (get_user(native, argv.ptr.native + nr))
 430                 return ERR_PTR(-EFAULT);
 431
 432         return native;
 433 }
 434
 435 /*
 436  * count() counts the number of strings in array ARGV.
 437  */
 438 static int count(struct user_arg_ptr argv, int max)
 439 {
 440         int i = 0;
 441
 442         if (argv.ptr.native != NULL) {
 443                 for (;;) {
 444                         const char __user *p = get_user_arg_ptr(argv, i);
 445
 446                         if (!p)
 447                                 break;
 448
 449                         if (IS_ERR(p))
 450                                 return -EFAULT;
 451
 452                         if (i++ >= max)
 453                                 return -E2BIG;
 454
 455                         if (fatal_signal_pending(current))
 456                                 return -ERESTARTNOHAND;
 457                         cond_resched();
 458                 }
 459         }
 460         return i;
 461 }
 462
 463 /*
 464  * 'copy_strings()' copies argument/environment strings from the old
 465  * processes's memory to the new process's stack.  The call to get_user_pages()
 466  * ensures the destination page is created and not swapped out.
 467  */
 468 static int copy_strings(int argc, struct user_arg_ptr argv,
 469                         struct linux_binprm *bprm)
 470 {
 471         struct page *kmapped_page = NULL;
 472         char *kaddr = NULL;
 473         unsigned long kpos = 0;
 474         int ret;
 475
 476         while (argc-- > 0) {
 477                 const char __user *str;
 478                 int len;
 479                 unsigned long pos;
 480
 481                 ret = -EFAULT;
 482                 str = get_user_arg_ptr(argv, argc);
 483                 if (IS_ERR(str))
 484                         goto out;
 485
 486                 len = strnlen_user(str, MAX_ARG_STRLEN);
 487                 if (!len)
 488                         goto out;
 489
 490                 ret = -E2BIG;
 491                 if (!valid_arg_len(bprm, len))
 492                         goto out;
 493
 494                 /* We're going to work our way backwords. */
 495                 pos = bprm->p;
 496                 str += len;
 497                 bprm->p -= len;
 498
 499                 while (len > 0) {
 500                         int offset, bytes_to_copy;
 501
 502                         if (fatal_signal_pending(current)) {
 503                                 ret = -ERESTARTNOHAND;
 504                                 goto out;
 505                         }
 506                         cond_resched();
 507
 508                         offset = pos % PAGE_SIZE;
 509                         if (offset == 0)
 510                                 offset = PAGE_SIZE;
 511
 512                         bytes_to_copy = offset;
 513                         if (bytes_to_copy > len)
 514                                 bytes_to_copy = len;
 515
 516                         offset -= bytes_to_copy;
 517                         pos -= bytes_to_copy;
 518                         str -= bytes_to_copy;
 519                         len -= bytes_to_copy;
 520
 521                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 522                                 struct page *page;
 523
 524                                 page = get_arg_page(bprm, pos, 1);
 525                                 if (!page) {
 526                                         ret = -E2BIG;
 527                                         goto out;
 528                                 }
 529
 530                                 if (kmapped_page) {
 531                                         flush_kernel_dcache_page(kmapped_page);
 532                                         kunmap(kmapped_page);
 533                                         put_arg_page(kmapped_page);
 534                                 }
 535                                 kmapped_page = page;
 536                                 kaddr = kmap(kmapped_page);
 537                                 kpos = pos & PAGE_MASK;
 538                                 flush_arg_page(bprm, kpos, kmapped_page);
 539                         }
 540                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 541                                 ret = -EFAULT;
 542                                 goto out;
 543                         }
 544                 }
 545         }
 546         ret = 0;
 547 out:
 548         if (kmapped_page) {
 549                 flush_kernel_dcache_page(kmapped_page);
 550                 kunmap(kmapped_page);
 551                 put_arg_page(kmapped_page);
 552         }
 553         return ret;
 554 }
 555
 556 /*
 557  * Like copy_strings, but get argv and its values from kernel memory.
 558  */
 559 int copy_strings_kernel(int argc, const char *const *__argv,
 560                         struct linux_binprm *bprm)
 561 {
 562         int r;
 563         mm_segment_t oldfs = get_fs();
 564         struct user_arg_ptr argv = {
 565                 .ptr.native = (const char __user *const  __user *)__argv,
 566         };
 567
 568         set_fs(KERNEL_DS);
 569         r = copy_strings(argc, argv, bprm);
 570         set_fs(oldfs);
 571
 572         return r;
 573 }
 574 EXPORT_SYMBOL(copy_strings_kernel);
 575
 576 #ifdef CONFIG_MMU
 577
 578 /*
 579  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 580  * the binfmt code determines where the new stack should reside, we shift it to
 581  * its final location.  The process proceeds as follows:
 582  *
 583  * 1) Use shift to calculate the new vma endpoints.
 584  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 585  *    arguments passed to subsequent functions are consistent.
 586  * 3) Move vma's page tables to the new range.
 587  * 4) Free up any cleared pgd range.
 588  * 5) Shrink the vma to cover only the new range.
 589  */
 590 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 591 {
 592         struct mm_struct *mm = vma->vm_mm;
 593         unsigned long old_start = vma->vm_start;
 594         unsigned long old_end = vma->vm_end;
 595         unsigned long length = old_end - old_start;
 596         unsigned long new_start = old_start - shift;
 597         unsigned long new_end = old_end - shift;
 598         struct mmu_gather tlb;
 599
 600         BUG_ON(new_start > new_end);
 601
 602         /*
 603          * ensure there are no vmas between where we want to go
 604          * and where we are
 605          */
 606         if (vma != find_vma(mm, new_start))
 607                 return -EFAULT;
 608
 609         /*
 610          * cover the whole range: [new_start, old_end)
 611          */
 612         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 613                 return -ENOMEM;
 614
 615         /*
 616          * move the page tables downwards, on failure we rely on
 617          * process cleanup to remove whatever mess we made.
 618          */
 619         if (length != move_page_tables(vma, old_start,
 620                                        vma, new_start, length))
 621                 return -ENOMEM;
 622
 623         lru_add_drain();
 624         tlb_gather_mmu(&tlb, mm, 0);
 625         if (new_end > old_start) {
 626                 /*
 627                  * when the old and new regions overlap clear from new_end.
 628                  */
 629                 free_pgd_range(&tlb, new_end, old_end, new_end,
 630                         vma->vm_next ? vma->vm_next->vm_start : 0);
 631         } else {
 632                 /*
 633                  * otherwise, clean from old_start; this is done to not touch
 634                  * the address space in [new_end, old_start) some architectures
 635                  * have constraints on va-space that make this illegal (IA64) -
 636                  * for the others its just a little faster.
 637                  */
 638                 free_pgd_range(&tlb, old_start, old_end, new_end,
 639                         vma->vm_next ? vma->vm_next->vm_start : 0);
 640         }
 641         tlb_finish_mmu(&tlb, new_end, old_end);
 642
 643         /*
 644          * Shrink the vma to just the new range.  Always succeeds.
 645          */
 646         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 653  * the stack is optionally relocated, and some extra space is added.
 654  */
 655 int setup_arg_pages(struct linux_binprm *bprm,
 656                     unsigned long stack_top,
 657                     int executable_stack)
 658 {
 659         unsigned long ret;
 660         unsigned long stack_shift;
 661         struct mm_struct *mm = current->mm;
 662         struct vm_area_struct *vma = bprm->vma;
 663         struct vm_area_struct *prev = NULL;
 664         unsigned long vm_flags;
 665         unsigned long stack_base;
 666         unsigned long stack_size;
 667         unsigned long stack_expand;
 668         unsigned long rlim_stack;
 669
 670 #ifdef CONFIG_STACK_GROWSUP
 671         /* Limit stack size to 1GB */
 672         stack_base = rlimit_max(RLIMIT_STACK);
 673         if (stack_base > (1 << 30))
 674                 stack_base = 1 << 30;
 675
 676         /* Make sure we didn't let the argument array grow too large. */
 677         if (vma->vm_end - vma->vm_start > stack_base)
 678                 return -ENOMEM;
 679
 680         stack_base = PAGE_ALIGN(stack_top - stack_base);
 681
 682         stack_shift = vma->vm_start - stack_base;
 683         mm->arg_start = bprm->p - stack_shift;
 684         bprm->p = vma->vm_end - stack_shift;
 685 #else
 686         stack_top = arch_align_stack(stack_top);
 687         stack_top = PAGE_ALIGN(stack_top);
 688
 689         if (unlikely(stack_top < mmap_min_addr) ||
 690             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 691                 return -ENOMEM;
 692
 693         stack_shift = vma->vm_end - stack_top;
 694
 695         bprm->p -= stack_shift;
 696         mm->arg_start = bprm->p;
 697 #endif
 698
 699         if (bprm->loader)
 700                 bprm->loader -= stack_shift;
 701         bprm->exec -= stack_shift;
 702
 703         down_write(&mm->mmap_sem);
 704         vm_flags = VM_STACK_FLAGS;
 705
 706         /*
 707          * Adjust stack execute permissions; explicitly enable for
 708          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 709          * (arch default) otherwise.
 710          */
 711         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 712                 vm_flags |= VM_EXEC;
 713         else if (executable_stack == EXSTACK_DISABLE_X)
 714                 vm_flags &= ~VM_EXEC;
 715         vm_flags |= mm->def_flags;
 716         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 717
 718         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 719                         vm_flags);
 720         if (ret)
 721                 goto out_unlock;
 722         BUG_ON(prev != vma);
 723
 724         /* Move stack pages down in memory. */
 725         if (stack_shift) {
 726                 ret = shift_arg_pages(vma, stack_shift);
 727                 if (ret)
 728                         goto out_unlock;
 729         }
 730
 731         /* mprotect_fixup is overkill to remove the temporary stack flags */
 732         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 733
 734         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 735         stack_size = vma->vm_end - vma->vm_start;
 736         /*
 737          * Align this down to a page boundary as expand_stack
 738          * will align it up.
 739          */
 740         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 741 #ifdef CONFIG_STACK_GROWSUP
 742         if (stack_size + stack_expand > rlim_stack)
 743                 stack_base = vma->vm_start + rlim_stack;
 744         else
 745                 stack_base = vma->vm_end + stack_expand;
 746 #else
 747         if (stack_size + stack_expand > rlim_stack)
 748                 stack_base = vma->vm_end - rlim_stack;
 749         else
 750                 stack_base = vma->vm_start - stack_expand;
 751 #endif
 752         current->mm->start_stack = bprm->p;
 753         ret = expand_stack(vma, stack_base);
 754         if (ret)
 755                 ret = -EFAULT;
 756
 757 out_unlock:
 758         up_write(&mm->mmap_sem);
 759         return ret;
 760 }
 761 EXPORT_SYMBOL(setup_arg_pages);
 762
 763 #endif /* CONFIG_MMU */
 764
 765 struct file *open_exec(const char *name)
 766 {
 767         struct file *file;
 768         int err;
 769         static const struct open_flags open_exec_flags = {
 770                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 771                 .acc_mode = MAY_EXEC | MAY_OPEN,
 772                 .intent = LOOKUP_OPEN
 773         };
 774
 775         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 776         if (IS_ERR(file))
 777                 goto out;
 778
 779         err = -EACCES;
 780         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 781                 goto exit;
 782
 783         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 784                 goto exit;
 785
 786         fsnotify_open(file);
 787
 788         err = deny_write_access(file);
 789         if (err)
 790                 goto exit;
 791
 792 out:
 793         return file;
 794
 795 exit:
 796         fput(file);
 797         return ERR_PTR(err);
 798 }
 799 EXPORT_SYMBOL(open_exec);
 800
 801 int kernel_read(struct file *file, loff_t offset,
 802                 char *addr, unsigned long count)
 803 {
 804         mm_segment_t old_fs;
 805         loff_t pos = offset;
 806         int result;
 807
 808         old_fs = get_fs();
 809         set_fs(get_ds());
 810         /* The cast to a user pointer is valid due to the set_fs() */
 811         result = vfs_read(file, (void __user *)addr, count, &pos);
 812         set_fs(old_fs);
 813         return result;
 814 }
 815
 816 EXPORT_SYMBOL(kernel_read);
 817
 818 static int exec_mmap(struct mm_struct *mm)
 819 {
 820         struct task_struct *tsk;
 821         struct mm_struct * old_mm, *active_mm;
 822
 823         /* Notify parent that we're no longer interested in the old VM */
 824         tsk = current;
 825         old_mm = current->mm;
 826         sync_mm_rss(old_mm);
 827         mm_release(tsk, old_mm);
 828
 829         if (old_mm) {
 830                 /*
 831                  * Make sure that if there is a core dump in progress
 832                  * for the old mm, we get out and die instead of going
 833                  * through with the exec.  We must hold mmap_sem around
 834                  * checking core_state and changing tsk->mm.
 835                  */
 836                 down_read(&old_mm->mmap_sem);
 837                 if (unlikely(old_mm->core_state)) {
 838                         up_read(&old_mm->mmap_sem);
 839                         return -EINTR;
 840                 }
 841         }
 842         task_lock(tsk);
 843         active_mm = tsk->active_mm;
 844         tsk->mm = mm;
 845         tsk->active_mm = mm;
 846         activate_mm(active_mm, mm);
 847         task_unlock(tsk);
 848         arch_pick_mmap_layout(mm);
 849         if (old_mm) {
 850                 up_read(&old_mm->mmap_sem);
 851                 BUG_ON(active_mm != old_mm);
 852                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 853                 mm_update_next_owner(old_mm);
 854                 mmput(old_mm);
 855                 return 0;
 856         }
 857         mmdrop(active_mm);
 858         return 0;
 859 }
 860
 861 /*
 862  * This function makes sure the current process has its own signal table,
 863  * so that flush_signal_handlers can later reset the handlers without
 864  * disturbing other processes.  (Other processes might share the signal
 865  * table via the CLONE_SIGHAND option to clone().)
 866  */
 867 static int de_thread(struct task_struct *tsk)
 868 {
 869         struct signal_struct *sig = tsk->signal;
 870         struct sighand_struct *oldsighand = tsk->sighand;
 871         spinlock_t *lock = &oldsighand->siglock;
 872
 873         if (thread_group_empty(tsk))
 874                 goto no_thread_group;
 875
 876         /*
 877          * Kill all other threads in the thread group.
 878          */
 879         spin_lock_irq(lock);
 880         if (signal_group_exit(sig)) {
 881                 /*
 882                  * Another group action in progress, just
 883                  * return so that the signal is processed.
 884                  */
 885                 spin_unlock_irq(lock);
 886                 return -EAGAIN;
 887         }
 888
 889         sig->group_exit_task = tsk;
 890         sig->notify_count = zap_other_threads(tsk);
 891         if (!thread_group_leader(tsk))
 892                 sig->notify_count--;
 893
 894         while (sig->notify_count) {
 895                 __set_current_state(TASK_UNINTERRUPTIBLE);
 896                 spin_unlock_irq(lock);
 897                 schedule();
 898                 spin_lock_irq(lock);
 899         }
 900         spin_unlock_irq(lock);
 901
 902         /*
 903          * At this point all other threads have exited, all we have to
 904          * do is to wait for the thread group leader to become inactive,
 905          * and to assume its PID:
 906          */
 907         if (!thread_group_leader(tsk)) {
 908                 struct task_struct *leader = tsk->group_leader;
 909
 910                 sig->notify_count = -1; /* for exit_notify() */
 911                 for (;;) {
 912                         write_lock_irq(&tasklist_lock);
 913                         if (likely(leader->exit_state))
 914                                 break;
 915                         __set_current_state(TASK_UNINTERRUPTIBLE);
 916                         write_unlock_irq(&tasklist_lock);
 917                         schedule();
 918                 }
 919
 920                 /*
 921                  * The only record we have of the real-time age of a
 922                  * process, regardless of execs it's done, is start_time.
 923                  * All the past CPU time is accumulated in signal_struct
 924                  * from sister threads now dead.  But in this non-leader
 925                  * exec, nothing survives from the original leader thread,
 926                  * whose birth marks the true age of this process now.
 927                  * When we take on its identity by switching to its PID, we
 928                  * also take its birthdate (always earlier than our own).
 929                  */
 930                 tsk->start_time = leader->start_time;
 931
 932                 BUG_ON(!same_thread_group(leader, tsk));
 933                 BUG_ON(has_group_leader_pid(tsk));
 934                 /*
 935                  * An exec() starts a new thread group with the
 936                  * TGID of the previous thread group. Rehash the
 937                  * two threads with a switched PID, and release
 938                  * the former thread group leader:
 939                  */
 940
 941                 /* Become a process group leader with the old leader's pid.
 942                  * The old leader becomes a thread of the this thread group.
 943                  * Note: The old leader also uses this pid until release_task
 944                  *       is called.  Odd but simple and correct.
 945                  */
 946                 detach_pid(tsk, PIDTYPE_PID);
 947                 tsk->pid = leader->pid;
 948                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 949                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 950                 transfer_pid(leader, tsk, PIDTYPE_SID);
 951
 952                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 953                 list_replace_init(&leader->sibling, &tsk->sibling);
 954
 955                 tsk->group_leader = tsk;
 956                 leader->group_leader = tsk;
 957
 958                 tsk->exit_signal = SIGCHLD;
 959                 leader->exit_signal = -1;
 960
 961                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 962                 leader->exit_state = EXIT_DEAD;
 963
 964                 /*
 965                  * We are going to release_task()->ptrace_unlink() silently,
 966                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 967                  * the tracer wont't block again waiting for this thread.
 968                  */
 969                 if (unlikely(leader->ptrace))
 970                         __wake_up_parent(leader, leader->parent);
 971                 write_unlock_irq(&tasklist_lock);
 972
 973                 release_task(leader);
 974         }
 975
 976         sig->group_exit_task = NULL;
 977         sig->notify_count = 0;
 978
 979 no_thread_group:
 980         /* we have changed execution domain */
 981         tsk->exit_signal = SIGCHLD;
 982
 983         exit_itimers(sig);
 984         flush_itimer_signals();
 985
 986         if (atomic_read(&oldsighand->count) != 1) {
 987                 struct sighand_struct *newsighand;
 988                 /*
 989                  * This ->sighand is shared with the CLONE_SIGHAND
 990                  * but not CLONE_THREAD task, switch to the new one.
 991                  */
 992                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 993                 if (!newsighand)
 994                         return -ENOMEM;
 995
 996                 atomic_set(&newsighand->count, 1);
 997                 memcpy(newsighand->action, oldsighand->action,
 998                        sizeof(newsighand->action));
 999
1000                 write_lock_irq(&tasklist_lock);
1001                 spin_lock(&oldsighand->siglock);
1002                 rcu_assign_pointer(tsk->sighand, newsighand);
1003                 spin_unlock(&oldsighand->siglock);
1004                 write_unlock_irq(&tasklist_lock);
1005
1006                 __cleanup_sighand(oldsighand);
1007         }
1008
1009         BUG_ON(!thread_group_leader(tsk));
1010         return 0;
1011 }
1012
1013 /*
1014  * These functions flushes out all traces of the currently running executable
1015  * so that a new one can be started
1016  */
1017 static void flush_old_files(struct files_struct * files)
1018 {
1019         long j = -1;
1020         struct fdtable *fdt;
1021
1022         spin_lock(&files->file_lock);
1023         for (;;) {
1024                 unsigned long set, i;
1025
1026                 j++;
1027                 i = j * __NFDBITS;
1028                 fdt = files_fdtable(files);
1029                 if (i >= fdt->max_fds)
1030                         break;
1031                 set = fdt->close_on_exec[j];
1032                 if (!set)
1033                         continue;
1034                 fdt->close_on_exec[j] = 0;
1035                 spin_unlock(&files->file_lock);
1036                 for ( ; set ; i++,set >>= 1) {
1037                         if (set & 1) {
1038                                 sys_close(i);
1039                         }
1040                 }
1041                 spin_lock(&files->file_lock);
1042
1043         }
1044         spin_unlock(&files->file_lock);
1045 }
1046
1047 char *get_task_comm(char *buf, struct task_struct *tsk)
1048 {
1049         /* buf must be at least sizeof(tsk->comm) in size */
1050         task_lock(tsk);
1051         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1052         task_unlock(tsk);
1053         return buf;
1054 }
1055 EXPORT_SYMBOL_GPL(get_task_comm);
1056
1057 void set_task_comm(struct task_struct *tsk, char *buf)
1058 {
1059         task_lock(tsk);
1060
1061         trace_task_rename(tsk, buf);
1062
1063         /*
1064          * Threads may access current->comm without holding
1065          * the task lock, so write the string carefully.
1066          * Readers without a lock may see incomplete new
1067          * names but are safe from non-terminating string reads.
1068          */
1069         memset(tsk->comm, 0, TASK_COMM_LEN);
1070         wmb();
1071         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1072         task_unlock(tsk);
1073         perf_event_comm(tsk);
1074 }
1075
1076 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1077 {
1078         int i, ch;
1079
1080         /* Copies the binary name from after last slash */
1081         for (i = 0; (ch = *(fn++)) != '\0';) {
1082                 if (ch == '/')
1083                         i = 0; /* overwrite what we wrote */
1084                 else
1085                         if (i < len - 1)
1086                                 tcomm[i++] = ch;
1087         }
1088         tcomm[i] = '\0';
1089 }
1090
1091 int flush_old_exec(struct linux_binprm * bprm)
1092 {
1093         int retval;
1094
1095         /*
1096          * Make sure we have a private signal table and that
1097          * we are unassociated from the previous thread group.
1098          */
1099         retval = de_thread(current);
1100         if (retval)
1101                 goto out;
1102
1103         set_mm_exe_file(bprm->mm, bprm->file);
1104
1105         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1106         /*
1107          * Release all of the old mmap stuff
1108          */
1109         acct_arg_size(bprm, 0);
1110         retval = exec_mmap(bprm->mm);
1111         if (retval)
1112                 goto out;
1113
1114         bprm->mm = NULL;                /* We're using it now */
1115
1116         set_fs(USER_DS);
1117         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1118         flush_thread();
1119         current->personality &= ~bprm->per_clear;
1120
1121         return 0;
1122
1123 out:
1124         return retval;
1125 }
1126 EXPORT_SYMBOL(flush_old_exec);
1127
1128 void would_dump(struct linux_binprm *bprm, struct file *file)
1129 {
1130         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1131                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1132 }
1133 EXPORT_SYMBOL(would_dump);
1134
1135 void setup_new_exec(struct linux_binprm * bprm)
1136 {
1137         arch_pick_mmap_layout(current->mm);
1138
1139         /* This is the point of no return */
1140         current->sas_ss_sp = current->sas_ss_size = 0;
1141
1142         if (current_euid() == current_uid() && current_egid() == current_gid())
1143                 set_dumpable(current->mm, 1);
1144         else
1145                 set_dumpable(current->mm, suid_dumpable);
1146
1147         set_task_comm(current, bprm->tcomm);
1148
1149         /* Set the new mm task size. We have to do that late because it may
1150          * depend on TIF_32BIT which is only updated in flush_thread() on
1151          * some architectures like powerpc
1152          */
1153         current->mm->task_size = TASK_SIZE;
1154
1155         /* install the new credentials */
1156         if (bprm->cred->uid != current_euid() ||
1157             bprm->cred->gid != current_egid()) {
1158                 current->pdeath_signal = 0;
1159         } else {
1160                 would_dump(bprm, bprm->file);
1161                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1162                         set_dumpable(current->mm, suid_dumpable);
1163         }
1164
1165         /*
1166          * Flush performance counters when crossing a
1167          * security domain:
1168          */
1169         if (!get_dumpable(current->mm))
1170                 perf_event_exit_task(current);
1171
1172         /* An exec changes our domain. We are no longer part of the thread
1173            group */
1174
1175         current->self_exec_id++;
1176
1177         flush_signal_handlers(current, 0);
1178         flush_old_files(current->files);
1179 }
1180 EXPORT_SYMBOL(setup_new_exec);
1181
1182 /*
1183  * Prepare credentials and lock ->cred_guard_mutex.
1184  * install_exec_creds() commits the new creds and drops the lock.
1185  * Or, if exec fails before, free_bprm() should release ->cred and
1186  * and unlock.
1187  */
1188 int prepare_bprm_creds(struct linux_binprm *bprm)
1189 {
1190         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1191                 return -ERESTARTNOINTR;
1192
1193         bprm->cred = prepare_exec_creds();
1194         if (likely(bprm->cred))
1195                 return 0;
1196
1197         mutex_unlock(&current->signal->cred_guard_mutex);
1198         return -ENOMEM;
1199 }
1200
1201 void free_bprm(struct linux_binprm *bprm)
1202 {
1203         free_arg_pages(bprm);
1204         if (bprm->cred) {
1205                 mutex_unlock(&current->signal->cred_guard_mutex);
1206                 abort_creds(bprm->cred);
1207         }
1208         kfree(bprm);
1209 }
1210
1211 /*
1212  * install the new credentials for this executable
1213  */
1214 void install_exec_creds(struct linux_binprm *bprm)
1215 {
1216         security_bprm_committing_creds(bprm);
1217
1218         commit_creds(bprm->cred);
1219         bprm->cred = NULL;
1220         /*
1221          * cred_guard_mutex must be held at least to this point to prevent
1222          * ptrace_attach() from altering our determination of the task's
1223          * credentials; any time after this it may be unlocked.
1224          */
1225         security_bprm_committed_creds(bprm);
1226         mutex_unlock(&current->signal->cred_guard_mutex);
1227 }
1228 EXPORT_SYMBOL(install_exec_creds);
1229
1230 /*
1231  * determine how safe it is to execute the proposed program
1232  * - the caller must hold ->cred_guard_mutex to protect against
1233  *   PTRACE_ATTACH
1234  */
1235 static int check_unsafe_exec(struct linux_binprm *bprm)
1236 {
1237         struct task_struct *p = current, *t;
1238         unsigned n_fs;
1239         int res = 0;
1240
1241         if (p->ptrace) {
1242                 if (p->ptrace & PT_PTRACE_CAP)
1243                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1244                 else
1245                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1246         }
1247
1248         /*
1249          * This isn't strictly necessary, but it makes it harder for LSMs to
1250          * mess up.
1251          */
1252         if (current->no_new_privs)
1253                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1254
1255         n_fs = 1;
1256         spin_lock(&p->fs->lock);
1257         rcu_read_lock();
1258         for (t = next_thread(p); t != p; t = next_thread(t)) {
1259                 if (t->fs == p->fs)
1260                         n_fs++;
1261         }
1262         rcu_read_unlock();
1263
1264         if (p->fs->users > n_fs) {
1265                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1266         } else {
1267                 res = -EAGAIN;
1268                 if (!p->fs->in_exec) {
1269                         p->fs->in_exec = 1;
1270                         res = 1;
1271                 }
1272         }
1273         spin_unlock(&p->fs->lock);
1274
1275         return res;
1276 }
1277
1278 /*
1279  * Fill the binprm structure from the inode.
1280  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1281  *
1282  * This may be called multiple times for binary chains (scripts for example).
1283  */
1284 int prepare_binprm(struct linux_binprm *bprm)
1285 {
1286         umode_t mode;
1287         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1288         int retval;
1289
1290         mode = inode->i_mode;
1291         if (bprm->file->f_op == NULL)
1292                 return -EACCES;
1293
1294         /* clear any previous set[ug]id data from a previous binary */
1295         bprm->cred->euid = current_euid();
1296         bprm->cred->egid = current_egid();
1297
1298         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1299             !current->no_new_privs) {
1300                 /* Set-uid? */
1301                 if (mode & S_ISUID) {
1302                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1303                         bprm->cred->euid = inode->i_uid;
1304                 }
1305
1306                 /* Set-gid? */
1307                 /*
1308                  * If setgid is set but no group execute bit then this
1309                  * is a candidate for mandatory locking, not a setgid
1310                  * executable.
1311                  */
1312                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1313                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1314                         bprm->cred->egid = inode->i_gid;
1315                 }
1316         }
1317
1318         /* fill in binprm security blob */
1319         retval = security_bprm_set_creds(bprm);
1320         if (retval)
1321                 return retval;
1322         bprm->cred_prepared = 1;
1323
1324         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1325         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1326 }
1327
1328 EXPORT_SYMBOL(prepare_binprm);
1329
1330 /*
1331  * Arguments are '\0' separated strings found at the location bprm->p
1332  * points to; chop off the first by relocating brpm->p to right after
1333  * the first '\0' encountered.
1334  */
1335 int remove_arg_zero(struct linux_binprm *bprm)
1336 {
1337         int ret = 0;
1338         unsigned long offset;
1339         char *kaddr;
1340         struct page *page;
1341
1342         if (!bprm->argc)
1343                 return 0;
1344
1345         do {
1346                 offset = bprm->p & ~PAGE_MASK;
1347                 page = get_arg_page(bprm, bprm->p, 0);
1348                 if (!page) {
1349                         ret = -EFAULT;
1350                         goto out;
1351                 }
1352                 kaddr = kmap_atomic(page);
1353
1354                 for (; offset < PAGE_SIZE && kaddr[offset];
1355                                 offset++, bprm->p++)
1356                         ;
1357
1358                 kunmap_atomic(kaddr);
1359                 put_arg_page(page);
1360
1361                 if (offset == PAGE_SIZE)
1362                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1363         } while (offset == PAGE_SIZE);
1364
1365         bprm->p++;
1366         bprm->argc--;
1367         ret = 0;
1368
1369 out:
1370         return ret;
1371 }
1372 EXPORT_SYMBOL(remove_arg_zero);
1373
1374 /*
1375  * cycle the list of binary formats handler, until one recognizes the image
1376  */
1377 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1378 {
1379         unsigned int depth = bprm->recursion_depth;
1380         int try,retval;
1381         struct linux_binfmt *fmt;
1382         pid_t old_pid, old_vpid;
1383
1384         retval = security_bprm_check(bprm);
1385         if (retval)
1386                 return retval;
1387
1388         retval = audit_bprm(bprm);
1389         if (retval)
1390                 return retval;
1391
1392         /* Need to fetch pid before load_binary changes it */
1393         old_pid = current->pid;
1394         rcu_read_lock();
1395         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1396         rcu_read_unlock();
1397
1398         retval = -ENOENT;
1399         for (try=0; try<2; try++) {
1400                 read_lock(&binfmt_lock);
1401                 list_for_each_entry(fmt, &formats, lh) {
1402                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1403                         if (!fn)
1404                                 continue;
1405                         if (!try_module_get(fmt->module))
1406                                 continue;
1407                         read_unlock(&binfmt_lock);
1408                         retval = fn(bprm, regs);
1409                         /*
1410                          * Restore the depth counter to its starting value
1411                          * in this call, so we don't have to rely on every
1412                          * load_binary function to restore it on return.
1413                          */
1414                         bprm->recursion_depth = depth;
1415                         if (retval >= 0) {
1416                                 if (depth == 0) {
1417                                         trace_sched_process_exec(current, old_pid, bprm);
1418                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1419                                 }
1420                                 put_binfmt(fmt);
1421                                 allow_write_access(bprm->file);
1422                                 if (bprm->file)
1423                                         fput(bprm->file);
1424                                 bprm->file = NULL;
1425                                 current->did_exec = 1;
1426                                 proc_exec_connector(current);
1427                                 return retval;
1428                         }
1429                         read_lock(&binfmt_lock);
1430                         put_binfmt(fmt);
1431                         if (retval != -ENOEXEC || bprm->mm == NULL)
1432                                 break;
1433                         if (!bprm->file) {
1434                                 read_unlock(&binfmt_lock);
1435                                 return retval;
1436                         }
1437                 }
1438                 read_unlock(&binfmt_lock);
1439 #ifdef CONFIG_MODULES
1440                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1441                         break;
1442                 } else {
1443 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1444                         if (printable(bprm->buf[0]) &&
1445                             printable(bprm->buf[1]) &&
1446                             printable(bprm->buf[2]) &&
1447                             printable(bprm->buf[3]))
1448                                 break; /* -ENOEXEC */
1449                         if (try)
1450                                 break; /* -ENOEXEC */
1451                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1452                 }
1453 #else
1454                 break;
1455 #endif
1456         }
1457         return retval;
1458 }
1459
1460 EXPORT_SYMBOL(search_binary_handler);
1461
1462 /*
1463  * sys_execve() executes a new program.
1464  */
1465 static int do_execve_common(const char *filename,
1466                                 struct user_arg_ptr argv,
1467                                 struct user_arg_ptr envp,
1468                                 struct pt_regs *regs)
1469 {
1470         struct linux_binprm *bprm;
1471         struct file *file;
1472         struct files_struct *displaced;
1473         bool clear_in_exec;
1474         int retval;
1475         const struct cred *cred = current_cred();
1476
1477         /*
1478          * We move the actual failure in case of RLIMIT_NPROC excess from
1479          * set*uid() to execve() because too many poorly written programs
1480          * don't check setuid() return code.  Here we additionally recheck
1481          * whether NPROC limit is still exceeded.
1482          */
1483         if ((current->flags & PF_NPROC_EXCEEDED) &&
1484             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1485                 retval = -EAGAIN;
1486                 goto out_ret;
1487         }
1488
1489         /* We're below the limit (still or again), so we don't want to make
1490          * further execve() calls fail. */
1491         current->flags &= ~PF_NPROC_EXCEEDED;
1492
1493         retval = unshare_files(&displaced);
1494         if (retval)
1495                 goto out_ret;
1496
1497         retval = -ENOMEM;
1498         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1499         if (!bprm)
1500                 goto out_files;
1501
1502         retval = prepare_bprm_creds(bprm);
1503         if (retval)
1504                 goto out_free;
1505
1506         retval = check_unsafe_exec(bprm);
1507         if (retval < 0)
1508                 goto out_free;
1509         clear_in_exec = retval;
1510         current->in_execve = 1;
1511
1512         file = open_exec(filename);
1513         retval = PTR_ERR(file);
1514         if (IS_ERR(file))
1515                 goto out_unmark;
1516
1517         sched_exec();
1518
1519         bprm->file = file;
1520         bprm->filename = filename;
1521         bprm->interp = filename;
1522
1523         retval = bprm_mm_init(bprm);
1524         if (retval)
1525                 goto out_file;
1526
1527         bprm->argc = count(argv, MAX_ARG_STRINGS);
1528         if ((retval = bprm->argc) < 0)
1529                 goto out;
1530
1531         bprm->envc = count(envp, MAX_ARG_STRINGS);
1532         if ((retval = bprm->envc) < 0)
1533                 goto out;
1534
1535         retval = prepare_binprm(bprm);
1536         if (retval < 0)
1537                 goto out;
1538
1539         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1540         if (retval < 0)
1541                 goto out;
1542
1543         bprm->exec = bprm->p;
1544         retval = copy_strings(bprm->envc, envp, bprm);
1545         if (retval < 0)
1546                 goto out;
1547
1548         retval = copy_strings(bprm->argc, argv, bprm);
1549         if (retval < 0)
1550                 goto out;
1551
1552         retval = search_binary_handler(bprm,regs);
1553         if (retval < 0)
1554                 goto out;
1555
1556         /* execve succeeded */
1557         current->fs->in_exec = 0;
1558         current->in_execve = 0;
1559         acct_update_integrals(current);
1560         free_bprm(bprm);
1561         if (displaced)
1562                 put_files_struct(displaced);
1563         return retval;
1564
1565 out:
1566         if (bprm->mm) {
1567                 acct_arg_size(bprm, 0);
1568                 mmput(bprm->mm);
1569         }
1570
1571 out_file:
1572         if (bprm->file) {
1573                 allow_write_access(bprm->file);
1574                 fput(bprm->file);
1575         }
1576
1577 out_unmark:
1578         if (clear_in_exec)
1579                 current->fs->in_exec = 0;
1580         current->in_execve = 0;
1581
1582 out_free:
1583         free_bprm(bprm);
1584
1585 out_files:
1586         if (displaced)
1587                 reset_files_struct(displaced);
1588 out_ret:
1589         return retval;
1590 }
1591
1592 int do_execve(const char *filename,
1593         const char __user *const __user *__argv,
1594         const char __user *const __user *__envp,
1595         struct pt_regs *regs)
1596 {
1597         struct user_arg_ptr argv = { .ptr.native = __argv };
1598         struct user_arg_ptr envp = { .ptr.native = __envp };
1599         return do_execve_common(filename, argv, envp, regs);
1600 }
1601
1602 #ifdef CONFIG_COMPAT
1603 int compat_do_execve(char *filename,
1604         compat_uptr_t __user *__argv,
1605         compat_uptr_t __user *__envp,
1606         struct pt_regs *regs)
1607 {
1608         struct user_arg_ptr argv = {
1609                 .is_compat = true,
1610                 .ptr.compat = __argv,
1611         };
1612         struct user_arg_ptr envp = {
1613                 .is_compat = true,
1614                 .ptr.compat = __envp,
1615         };
1616         return do_execve_common(filename, argv, envp, regs);
1617 }
1618 #endif
1619
1620 void set_binfmt(struct linux_binfmt *new)
1621 {
1622         struct mm_struct *mm = current->mm;
1623
1624         if (mm->binfmt)
1625                 module_put(mm->binfmt->module);
1626
1627         mm->binfmt = new;
1628         if (new)
1629                 __module_get(new->module);
1630 }
1631
1632 EXPORT_SYMBOL(set_binfmt);
1633
1634 static int expand_corename(struct core_name *cn)
1635 {
1636         char *old_corename = cn->corename;
1637
1638         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1639         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1640
1641         if (!cn->corename) {
1642                 kfree(old_corename);
1643                 return -ENOMEM;
1644         }
1645
1646         return 0;
1647 }
1648
1649 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1650 {
1651         char *cur;
1652         int need;
1653         int ret;
1654         va_list arg;
1655
1656         va_start(arg, fmt);
1657         need = vsnprintf(NULL, 0, fmt, arg);
1658         va_end(arg);
1659
1660         if (likely(need < cn->size - cn->used - 1))
1661                 goto out_printf;
1662
1663         ret = expand_corename(cn);
1664         if (ret)
1665                 goto expand_fail;
1666
1667 out_printf:
1668         cur = cn->corename + cn->used;
1669         va_start(arg, fmt);
1670         vsnprintf(cur, need + 1, fmt, arg);
1671         va_end(arg);
1672         cn->used += need;
1673         return 0;
1674
1675 expand_fail:
1676         return ret;
1677 }
1678
1679 static void cn_escape(char *str)
1680 {
1681         for (; *str; str++)
1682                 if (*str == '/')
1683                         *str = '!';
1684 }
1685
1686 static int cn_print_exe_file(struct core_name *cn)
1687 {
1688         struct file *exe_file;
1689         char *pathbuf, *path;
1690         int ret;
1691
1692         exe_file = get_mm_exe_file(current->mm);
1693         if (!exe_file) {
1694                 char *commstart = cn->corename + cn->used;
1695                 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1696                 cn_escape(commstart);
1697                 return ret;
1698         }
1699
1700         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1701         if (!pathbuf) {
1702                 ret = -ENOMEM;
1703                 goto put_exe_file;
1704         }
1705
1706         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1707         if (IS_ERR(path)) {
1708                 ret = PTR_ERR(path);
1709                 goto free_buf;
1710         }
1711
1712         cn_escape(path);
1713
1714         ret = cn_printf(cn, "%s", path);
1715
1716 free_buf:
1717         kfree(pathbuf);
1718 put_exe_file:
1719         fput(exe_file);
1720         return ret;
1721 }
1722
1723 /* format_corename will inspect the pattern parameter, and output a
1724  * name into corename, which must have space for at least
1725  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1726  */
1727 static int format_corename(struct core_name *cn, long signr)
1728 {
1729         const struct cred *cred = current_cred();
1730         const char *pat_ptr = core_pattern;
1731         int ispipe = (*pat_ptr == '|');
1732         int pid_in_pattern = 0;
1733         int err = 0;
1734
1735         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1736         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1737         cn->used = 0;
1738
1739         if (!cn->corename)
1740                 return -ENOMEM;
1741
1742         /* Repeat as long as we have more pattern to process and more output
1743            space */
1744         while (*pat_ptr) {
1745                 if (*pat_ptr != '%') {
1746                         if (*pat_ptr == 0)
1747                                 goto out;
1748                         err = cn_printf(cn, "%c", *pat_ptr++);
1749                 } else {
1750                         switch (*++pat_ptr) {
1751                         /* single % at the end, drop that */
1752                         case 0:
1753                                 goto out;
1754                         /* Double percent, output one percent */
1755                         case '%':
1756                                 err = cn_printf(cn, "%c", '%');
1757                                 break;
1758                         /* pid */
1759                         case 'p':
1760                                 pid_in_pattern = 1;
1761                                 err = cn_printf(cn, "%d",
1762                                               task_tgid_vnr(current));
1763                                 break;
1764                         /* uid */
1765                         case 'u':
1766                                 err = cn_printf(cn, "%d", cred->uid);
1767                                 break;
1768                         /* gid */
1769                         case 'g':
1770                                 err = cn_printf(cn, "%d", cred->gid);
1771                                 break;
1772                         /* signal that caused the coredump */
1773                         case 's':
1774                                 err = cn_printf(cn, "%ld", signr);
1775                                 break;
1776                         /* UNIX time of coredump */
1777                         case 't': {
1778                                 struct timeval tv;
1779                                 do_gettimeofday(&tv);
1780                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1781                                 break;
1782                         }
1783                         /* hostname */
1784                         case 'h': {
1785                                 char *namestart = cn->corename + cn->used;
1786                                 down_read(&uts_sem);
1787                                 err = cn_printf(cn, "%s",
1788                                               utsname()->nodename);
1789                                 up_read(&uts_sem);
1790                                 cn_escape(namestart);
1791                                 break;
1792                         }
1793                         /* executable */
1794                         case 'e': {
1795                                 char *commstart = cn->corename + cn->used;
1796                                 err = cn_printf(cn, "%s", current->comm);
1797                                 cn_escape(commstart);
1798                                 break;
1799                         }
1800                         case 'E':
1801                                 err = cn_print_exe_file(cn);
1802                                 break;
1803                         /* core limit size */
1804                         case 'c':
1805                                 err = cn_printf(cn, "%lu",
1806                                               rlimit(RLIMIT_CORE));
1807                                 break;
1808                         default:
1809                                 break;
1810                         }
1811                         ++pat_ptr;
1812                 }
1813
1814                 if (err)
1815                         return err;
1816         }
1817
1818         /* Backward compatibility with core_uses_pid:
1819          *
1820          * If core_pattern does not include a %p (as is the default)
1821          * and core_uses_pid is set, then .%pid will be appended to
1822          * the filename. Do not do this for piped commands. */
1823         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1824                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1825                 if (err)
1826                         return err;
1827         }
1828 out:
1829         return ispipe;
1830 }
1831
1832 static int zap_process(struct task_struct *start, int exit_code)
1833 {
1834         struct task_struct *t;
1835         int nr = 0;
1836
1837         start->signal->flags = SIGNAL_GROUP_EXIT;
1838         start->signal->group_exit_code = exit_code;
1839         start->signal->group_stop_count = 0;
1840
1841         t = start;
1842         do {
1843                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1844                 if (t != current && t->mm) {
1845                         sigaddset(&t->pending.signal, SIGKILL);
1846                         signal_wake_up(t, 1);
1847                         nr++;
1848                 }
1849         } while_each_thread(start, t);
1850
1851         return nr;
1852 }
1853
1854 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1855                                 struct core_state *core_state, int exit_code)
1856 {
1857         struct task_struct *g, *p;
1858         unsigned long flags;
1859         int nr = -EAGAIN;
1860
1861         spin_lock_irq(&tsk->sighand->siglock);
1862         if (!signal_group_exit(tsk->signal)) {
1863                 mm->core_state = core_state;
1864                 nr = zap_process(tsk, exit_code);
1865         }
1866         spin_unlock_irq(&tsk->sighand->siglock);
1867         if (unlikely(nr < 0))
1868                 return nr;
1869
1870         if (atomic_read(&mm->mm_users) == nr + 1)
1871                 goto done;
1872         /*
1873          * We should find and kill all tasks which use this mm, and we should
1874          * count them correctly into ->nr_threads. We don't take tasklist
1875          * lock, but this is safe wrt:
1876          *
1877          * fork:
1878          *      None of sub-threads can fork after zap_process(leader). All
1879          *      processes which were created before this point should be
1880          *      visible to zap_threads() because copy_process() adds the new
1881          *      process to the tail of init_task.tasks list, and lock/unlock
1882          *      of ->siglock provides a memory barrier.
1883          *
1884          * do_exit:
1885          *      The caller holds mm->mmap_sem. This means that the task which
1886          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1887          *      its ->mm.
1888          *
1889          * de_thread:
1890          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1891          *      we must see either old or new leader, this does not matter.
1892          *      However, it can change p->sighand, so lock_task_sighand(p)
1893          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1894          *      it can't fail.
1895          *
1896          *      Note also that "g" can be the old leader with ->mm == NULL
1897          *      and already unhashed and thus removed from ->thread_group.
1898          *      This is OK, __unhash_process()->list_del_rcu() does not
1899          *      clear the ->next pointer, we will find the new leader via
1900          *      next_thread().
1901          */
1902         rcu_read_lock();
1903         for_each_process(g) {
1904                 if (g == tsk->group_leader)
1905                         continue;
1906                 if (g->flags & PF_KTHREAD)
1907                         continue;
1908                 p = g;
1909                 do {
1910                         if (p->mm) {
1911                                 if (unlikely(p->mm == mm)) {
1912                                         lock_task_sighand(p, &flags);
1913                                         nr += zap_process(p, exit_code);
1914                                         unlock_task_sighand(p, &flags);
1915                                 }
1916                                 break;
1917                         }
1918                 } while_each_thread(g, p);
1919         }
1920         rcu_read_unlock();
1921 done:
1922         atomic_set(&core_state->nr_threads, nr);
1923         return nr;
1924 }
1925
1926 static int coredump_wait(int exit_code, struct core_state *core_state)
1927 {
1928         struct task_struct *tsk = current;
1929         struct mm_struct *mm = tsk->mm;
1930         int core_waiters = -EBUSY;
1931
1932         init_completion(&core_state->startup);
1933         core_state->dumper.task = tsk;
1934         core_state->dumper.next = NULL;
1935
1936         down_write(&mm->mmap_sem);
1937         if (!mm->core_state)
1938                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1939         up_write(&mm->mmap_sem);
1940
1941         if (core_waiters > 0)
1942                 wait_for_completion(&core_state->startup);
1943
1944         return core_waiters;
1945 }
1946
1947 static void coredump_finish(struct mm_struct *mm)
1948 {
1949         struct core_thread *curr, *next;
1950         struct task_struct *task;
1951
1952         next = mm->core_state->dumper.next;
1953         while ((curr = next) != NULL) {
1954                 next = curr->next;
1955                 task = curr->task;
1956                 /*
1957                  * see exit_mm(), curr->task must not see
1958                  * ->task == NULL before we read ->next.
1959                  */
1960                 smp_mb();
1961                 curr->task = NULL;
1962                 wake_up_process(task);
1963         }
1964
1965         mm->core_state = NULL;
1966 }
1967
1968 /*
1969  * set_dumpable converts traditional three-value dumpable to two flags and
1970  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1971  * these bits are not changed atomically.  So get_dumpable can observe the
1972  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1973  * return either old dumpable or new one by paying attention to the order of
1974  * modifying the bits.
1975  *
1976  * dumpable |   mm->flags (binary)
1977  * old  new | initial interim  final
1978  * ---------+-----------------------
1979  *  0    1  |   00      01      01
1980  *  0    2  |   00      10(*)   11
1981  *  1    0  |   01      00      00
1982  *  1    2  |   01      11      11
1983  *  2    0  |   11      10(*)   00
1984  *  2    1  |   11      11      01
1985  *
1986  * (*) get_dumpable regards interim value of 10 as 11.
1987  */
1988 void set_dumpable(struct mm_struct *mm, int value)
1989 {
1990         switch (value) {
1991         case 0:
1992                 clear_bit(MMF_DUMPABLE, &mm->flags);
1993                 smp_wmb();
1994                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1995                 break;
1996         case 1:
1997                 set_bit(MMF_DUMPABLE, &mm->flags);
1998                 smp_wmb();
1999                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2000                 break;
2001         case 2:
2002                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2003                 smp_wmb();
2004                 set_bit(MMF_DUMPABLE, &mm->flags);
2005                 break;
2006         }
2007 }
2008
2009 static int __get_dumpable(unsigned long mm_flags)
2010 {
2011         int ret;
2012
2013         ret = mm_flags & MMF_DUMPABLE_MASK;
2014         return (ret >= 2) ? 2 : ret;
2015 }
2016
2017 int get_dumpable(struct mm_struct *mm)
2018 {
2019         return __get_dumpable(mm->flags);
2020 }
2021
2022 static void wait_for_dump_helpers(struct file *file)
2023 {
2024         struct pipe_inode_info *pipe;
2025
2026         pipe = file->f_path.dentry->d_inode->i_pipe;
2027
2028         pipe_lock(pipe);
2029         pipe->readers++;
2030         pipe->writers--;
2031
2032         while ((pipe->readers > 1) && (!signal_pending(current))) {
2033                 wake_up_interruptible_sync(&pipe->wait);
2034                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2035                 pipe_wait(pipe);
2036         }
2037
2038         pipe->readers--;
2039         pipe->writers++;
2040         pipe_unlock(pipe);
2041
2042 }
2043
2044
2045 /*
2046  * umh_pipe_setup
2047  * helper function to customize the process used
2048  * to collect the core in userspace.  Specifically
2049  * it sets up a pipe and installs it as fd 0 (stdin)
2050  * for the process.  Returns 0 on success, or
2051  * PTR_ERR on failure.
2052  * Note that it also sets the core limit to 1.  This
2053  * is a special value that we use to trap recursive
2054  * core dumps
2055  */
2056 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2057 {
2058         struct file *rp, *wp;
2059         struct fdtable *fdt;
2060         struct coredump_params *cp = (struct coredump_params *)info->data;
2061         struct files_struct *cf = current->files;
2062
2063         wp = create_write_pipe(0);
2064         if (IS_ERR(wp))
2065                 return PTR_ERR(wp);
2066
2067         rp = create_read_pipe(wp, 0);
2068         if (IS_ERR(rp)) {
2069                 free_write_pipe(wp);
2070                 return PTR_ERR(rp);
2071         }
2072
2073         cp->file = wp;
2074
2075         sys_close(0);
2076         fd_install(0, rp);
2077         spin_lock(&cf->file_lock);
2078         fdt = files_fdtable(cf);
2079         __set_open_fd(0, fdt);
2080         __clear_close_on_exec(0, fdt);
2081         spin_unlock(&cf->file_lock);
2082
2083         /* and disallow core files too */
2084         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2085
2086         return 0;
2087 }
2088
2089 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2090 {
2091         struct core_state core_state;
2092         struct core_name cn;
2093         struct mm_struct *mm = current->mm;
2094         struct linux_binfmt * binfmt;
2095         const struct cred *old_cred;
2096         struct cred *cred;
2097         int retval = 0;
2098         int flag = 0;
2099         int ispipe;
2100         static atomic_t core_dump_count = ATOMIC_INIT(0);
2101         struct coredump_params cprm = {
2102                 .signr = signr,
2103                 .regs = regs,
2104                 .limit = rlimit(RLIMIT_CORE),
2105                 /*
2106                  * We must use the same mm->flags while dumping core to avoid
2107                  * inconsistency of bit flags, since this flag is not protected
2108                  * by any locks.
2109                  */
2110                 .mm_flags = mm->flags,
2111         };
2112
2113         audit_core_dumps(signr);
2114
2115         binfmt = mm->binfmt;
2116         if (!binfmt || !binfmt->core_dump)
2117                 goto fail;
2118         if (!__get_dumpable(cprm.mm_flags))
2119                 goto fail;
2120
2121         cred = prepare_creds();
2122         if (!cred)
2123                 goto fail;
2124         /*
2125          *      We cannot trust fsuid as being the "true" uid of the
2126          *      process nor do we know its entire history. We only know it
2127          *      was tainted so we dump it as root in mode 2.
2128          */
2129         if (__get_dumpable(cprm.mm_flags) == 2) {
2130                 /* Setuid core dump mode */
2131                 flag = O_EXCL;          /* Stop rewrite attacks */
2132                 cred->fsuid = 0;        /* Dump root private */
2133         }
2134
2135         retval = coredump_wait(exit_code, &core_state);
2136         if (retval < 0)
2137                 goto fail_creds;
2138
2139         old_cred = override_creds(cred);
2140
2141         /*
2142          * Clear any false indication of pending signals that might
2143          * be seen by the filesystem code called to write the core file.
2144          */
2145         clear_thread_flag(TIF_SIGPENDING);
2146
2147         ispipe = format_corename(&cn, signr);
2148
2149         if (ispipe) {
2150                 int dump_count;
2151                 char **helper_argv;
2152
2153                 if (ispipe < 0) {
2154                         printk(KERN_WARNING "format_corename failed\n");
2155                         printk(KERN_WARNING "Aborting core\n");
2156                         goto fail_corename;
2157                 }
2158
2159                 if (cprm.limit == 1) {
2160                         /*
2161                          * Normally core limits are irrelevant to pipes, since
2162                          * we're not writing to the file system, but we use
2163                          * cprm.limit of 1 here as a speacial value. Any
2164                          * non-1 limit gets set to RLIM_INFINITY below, but
2165                          * a limit of 0 skips the dump.  This is a consistent
2166                          * way to catch recursive crashes.  We can still crash
2167                          * if the core_pattern binary sets RLIM_CORE =  !1
2168                          * but it runs as root, and can do lots of stupid things
2169                          * Note that we use task_tgid_vnr here to grab the pid
2170                          * of the process group leader.  That way we get the
2171                          * right pid if a thread in a multi-threaded
2172                          * core_pattern process dies.
2173                          */
2174                         printk(KERN_WARNING
2175                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2176                                 task_tgid_vnr(current), current->comm);
2177                         printk(KERN_WARNING "Aborting core\n");
2178                         goto fail_unlock;
2179                 }
2180                 cprm.limit = RLIM_INFINITY;
2181
2182                 dump_count = atomic_inc_return(&core_dump_count);
2183                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2184                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2185                                task_tgid_vnr(current), current->comm);
2186                         printk(KERN_WARNING "Skipping core dump\n");
2187                         goto fail_dropcount;
2188                 }
2189
2190                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2191                 if (!helper_argv) {
2192                         printk(KERN_WARNING "%s failed to allocate memory\n",
2193                                __func__);
2194                         goto fail_dropcount;
2195                 }
2196
2197                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2198                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2199                                         NULL, &cprm);
2200                 argv_free(helper_argv);
2201                 if (retval) {
2202                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2203                                cn.corename);
2204                         goto close_fail;
2205                 }
2206         } else {
2207                 struct inode *inode;
2208
2209                 if (cprm.limit < binfmt->min_coredump)
2210                         goto fail_unlock;
2211
2212                 cprm.file = filp_open(cn.corename,
2213                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2214                                  0600);
2215                 if (IS_ERR(cprm.file))
2216                         goto fail_unlock;
2217
2218                 inode = cprm.file->f_path.dentry->d_inode;
2219                 if (inode->i_nlink > 1)
2220                         goto close_fail;
2221                 if (d_unhashed(cprm.file->f_path.dentry))
2222                         goto close_fail;
2223                 /*
2224                  * AK: actually i see no reason to not allow this for named
2225                  * pipes etc, but keep the previous behaviour for now.
2226                  */
2227                 if (!S_ISREG(inode->i_mode))
2228                         goto close_fail;
2229                 /*
2230                  * Dont allow local users get cute and trick others to coredump
2231                  * into their pre-created files.
2232                  */
2233                 if (inode->i_uid != current_fsuid())
2234                         goto close_fail;
2235                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2236                         goto close_fail;
2237                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2238                         goto close_fail;
2239         }
2240
2241         retval = binfmt->core_dump(&cprm);
2242         if (retval)
2243                 current->signal->group_exit_code |= 0x80;
2244
2245         if (ispipe && core_pipe_limit)
2246                 wait_for_dump_helpers(cprm.file);
2247 close_fail:
2248         if (cprm.file)
2249                 filp_close(cprm.file, NULL);
2250 fail_dropcount:
2251         if (ispipe)
2252                 atomic_dec(&core_dump_count);
2253 fail_unlock:
2254         kfree(cn.corename);
2255 fail_corename:
2256         coredump_finish(mm);
2257         revert_creds(old_cred);
2258 fail_creds:
2259         put_cred(cred);
2260 fail:
2261         return;
2262 }
2263
2264 /*
2265  * Core dumping helper functions.  These are the only things you should
2266  * do on a core-file: use only these functions to write out all the
2267  * necessary info.
2268  */
2269 int dump_write(struct file *file, const void *addr, int nr)
2270 {
2271         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2272 }
2273 EXPORT_SYMBOL(dump_write);
2274
2275 int dump_seek(struct file *file, loff_t off)
2276 {
2277         int ret = 1;
2278
2279         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2280                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2281                         return 0;
2282         } else {
2283                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2284
2285                 if (!buf)
2286                         return 0;
2287                 while (off > 0) {
2288                         unsigned long n = off;
2289
2290                         if (n > PAGE_SIZE)
2291                                 n = PAGE_SIZE;
2292                         if (!dump_write(file, buf, n)) {
2293                                 ret = 0;
2294                                 break;
2295                         }
2296                         off -= n;
2297                 }
2298                 free_page((unsigned long)buf);
2299         }
2300         return ret;
2301 }
2302 EXPORT_SYMBOL(dump_seek);