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/freezer.h>
  29 #include <linux/mm.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.h>
  58 #include <linux/compat.h>
  59 #include <linux/sched.h>
  60 #include <linux/fs.h>
  61 #include <linux/path.h>
  62
  63 #include <asm/uaccess.h>
  64 #include <asm/mmu_context.h>
  65 #include <asm/tlb.h>
  66 #include "internal.h"
  67
  68 int core_uses_pid;
  69 char core_pattern[CORENAME_MAX_SIZE] = "core";
  70 unsigned int core_pipe_limit;
  71 int suid_dumpable = 0;
  72
  73 struct core_name {
  74         char *corename;
  75         int used, size;
  76 };
  77 static atomic_t call_count = ATOMIC_INIT(1);
  78
  79 /* The maximal length of core_pattern is also specified in sysctl.c */
  80
  81 static LIST_HEAD(formats);
  82 static DEFINE_RWLOCK(binfmt_lock);
  83
  84 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  85 {
  86         if (!fmt)
  87                 return -EINVAL;
  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         return 0;
  93 }
  94
  95 EXPORT_SYMBOL(__register_binfmt);
  96
  97 void unregister_binfmt(struct linux_binfmt * fmt)
  98 {
  99         write_lock(&binfmt_lock);
 100         list_del(&fmt->lh);
 101         write_unlock(&binfmt_lock);
 102 }
 103
 104 EXPORT_SYMBOL(unregister_binfmt);
 105
 106 static inline void put_binfmt(struct linux_binfmt * fmt)
 107 {
 108         module_put(fmt->module);
 109 }
 110
 111 /*
 112  * Note that a shared library must be both readable and executable due to
 113  * security reasons.
 114  *
 115  * Also note that we take the address to load from from the file itself.
 116  */
 117 SYSCALL_DEFINE1(uselib, const char __user *, library)
 118 {
 119         struct file *file;
 120         char *tmp = getname(library);
 121         int error = PTR_ERR(tmp);
 122         static const struct open_flags uselib_flags = {
 123                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 124                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 125                 .intent = LOOKUP_OPEN
 126         };
 127
 128         if (IS_ERR(tmp))
 129                 goto out;
 130
 131         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 132         putname(tmp);
 133         error = PTR_ERR(file);
 134         if (IS_ERR(file))
 135                 goto out;
 136
 137         error = -EINVAL;
 138         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 139                 goto exit;
 140
 141         error = -EACCES;
 142         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 143                 goto exit;
 144
 145         fsnotify_open(file);
 146
 147         error = -ENOEXEC;
 148         if(file->f_op) {
 149                 struct linux_binfmt * fmt;
 150
 151                 read_lock(&binfmt_lock);
 152                 list_for_each_entry(fmt, &formats, lh) {
 153                         if (!fmt->load_shlib)
 154                                 continue;
 155                         if (!try_module_get(fmt->module))
 156                                 continue;
 157                         read_unlock(&binfmt_lock);
 158                         error = fmt->load_shlib(file);
 159                         read_lock(&binfmt_lock);
 160                         put_binfmt(fmt);
 161                         if (error != -ENOEXEC)
 162                                 break;
 163                 }
 164                 read_unlock(&binfmt_lock);
 165         }
 166 exit:
 167         fput(file);
 168 out:
 169         return error;
 170 }
 171
 172 #ifdef CONFIG_MMU
 173 /*
 174  * The nascent bprm->mm is not visible until exec_mmap() but it can
 175  * use a lot of memory, account these pages in current->mm temporary
 176  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 177  * change the counter back via acct_arg_size(0).
 178  */
 179 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 180 {
 181         struct mm_struct *mm = current->mm;
 182         long diff = (long)(pages - bprm->vma_pages);
 183
 184         if (!mm || !diff)
 185                 return;
 186
 187         bprm->vma_pages = pages;
 188         add_mm_counter(mm, MM_ANONPAGES, diff);
 189 }
 190
 191 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 192                 int write)
 193 {
 194         struct page *page;
 195         int ret;
 196
 197 #ifdef CONFIG_STACK_GROWSUP
 198         if (write) {
 199                 ret = expand_downwards(bprm->vma, pos);
 200                 if (ret < 0)
 201                         return NULL;
 202         }
 203 #endif
 204         ret = get_user_pages(current, bprm->mm, pos,
 205                         1, write, 1, &page, NULL);
 206         if (ret <= 0)
 207                 return NULL;
 208
 209         if (write) {
 210                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 211                 unsigned long ptr_size;
 212                 struct rlimit *rlim;
 213
 214                 /*
 215                  * Since the stack will hold pointers to the strings, we
 216                  * must account for them as well.
 217                  *
 218                  * The size calculation is the entire vma while each arg page is
 219                  * built, so each time we get here it's calculating how far it
 220                  * is currently (rather than each call being just the newly
 221                  * added size from the arg page).  As a result, we need to
 222                  * always add the entire size of the pointers, so that on the
 223                  * last call to get_arg_page() we'll actually have the entire
 224                  * correct size.
 225                  */
 226                 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 227                 if (ptr_size > ULONG_MAX - size)
 228                         goto fail;
 229                 size += ptr_size;
 230
 231                 acct_arg_size(bprm, size / PAGE_SIZE);
 232
 233                 /*
 234                  * We've historically supported up to 32 pages (ARG_MAX)
 235                  * of argument strings even with small stacks
 236                  */
 237                 if (size <= ARG_MAX)
 238                         return page;
 239
 240                 /*
 241                  * Limit to 1/4-th the stack size for the argv+env strings.
 242                  * This ensures that:
 243                  *  - the remaining binfmt code will not run out of stack space,
 244                  *  - the program will have a reasonable amount of stack left
 245                  *    to work from.
 246                  */
 247                 rlim = current->signal->rlim;
 248                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4)
 249                         goto fail;
 250         }
 251
 252         return page;
 253
 254 fail:
 255         put_page(page);
 256         return NULL;
 257 }
 258
 259 static void put_arg_page(struct page *page)
 260 {
 261         put_page(page);
 262 }
 263
 264 static void free_arg_page(struct linux_binprm *bprm, int i)
 265 {
 266 }
 267
 268 static void free_arg_pages(struct linux_binprm *bprm)
 269 {
 270 }
 271
 272 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 273                 struct page *page)
 274 {
 275         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 276 }
 277
 278 static int __bprm_mm_init(struct linux_binprm *bprm)
 279 {
 280         int err;
 281         struct vm_area_struct *vma = NULL;
 282         struct mm_struct *mm = bprm->mm;
 283
 284         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 285         if (!vma)
 286                 return -ENOMEM;
 287
 288         down_write(&mm->mmap_sem);
 289         vma->vm_mm = mm;
 290
 291         /*
 292          * Place the stack at the largest stack address the architecture
 293          * supports. Later, we'll move this to an appropriate place. We don't
 294          * use STACK_TOP because that can depend on attributes which aren't
 295          * configured yet.
 296          */
 297         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 298         vma->vm_end = STACK_TOP_MAX;
 299         vma->vm_start = vma->vm_end - PAGE_SIZE;
 300         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 301         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 302         INIT_LIST_HEAD(&vma->anon_vma_chain);
 303
 304         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 305         if (err)
 306                 goto err;
 307
 308         err = insert_vm_struct(mm, vma);
 309         if (err)
 310                 goto err;
 311
 312         mm->stack_vm = mm->total_vm = 1;
 313         up_write(&mm->mmap_sem);
 314         bprm->p = vma->vm_end - sizeof(void *);
 315         return 0;
 316 err:
 317         up_write(&mm->mmap_sem);
 318         bprm->vma = NULL;
 319         kmem_cache_free(vm_area_cachep, vma);
 320         return err;
 321 }
 322
 323 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 324 {
 325         return len <= MAX_ARG_STRLEN;
 326 }
 327
 328 #else
 329
 330 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 331 {
 332 }
 333
 334 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 335                 int write)
 336 {
 337         struct page *page;
 338
 339         page = bprm->page[pos / PAGE_SIZE];
 340         if (!page && write) {
 341                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 342                 if (!page)
 343                         return NULL;
 344                 bprm->page[pos / PAGE_SIZE] = page;
 345         }
 346
 347         return page;
 348 }
 349
 350 static void put_arg_page(struct page *page)
 351 {
 352 }
 353
 354 static void free_arg_page(struct linux_binprm *bprm, int i)
 355 {
 356         if (bprm->page[i]) {
 357                 __free_page(bprm->page[i]);
 358                 bprm->page[i] = NULL;
 359         }
 360 }
 361
 362 static void free_arg_pages(struct linux_binprm *bprm)
 363 {
 364         int i;
 365
 366         for (i = 0; i < MAX_ARG_PAGES; i++)
 367                 free_arg_page(bprm, i);
 368 }
 369
 370 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 371                 struct page *page)
 372 {
 373 }
 374
 375 static int __bprm_mm_init(struct linux_binprm *bprm)
 376 {
 377         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 378         return 0;
 379 }
 380
 381 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 382 {
 383         return len <= bprm->p;
 384 }
 385
 386 #endif /* CONFIG_MMU */
 387
 388 /*
 389  * Create a new mm_struct and populate it with a temporary stack
 390  * vm_area_struct.  We don't have enough context at this point to set the stack
 391  * flags, permissions, and offset, so we use temporary values.  We'll update
 392  * them later in setup_arg_pages().
 393  */
 394 int bprm_mm_init(struct linux_binprm *bprm)
 395 {
 396         int err;
 397         struct mm_struct *mm = NULL;
 398
 399         bprm->mm = mm = mm_alloc();
 400         err = -ENOMEM;
 401         if (!mm)
 402                 goto err;
 403
 404         err = init_new_context(current, mm);
 405         if (err)
 406                 goto err;
 407
 408         err = __bprm_mm_init(bprm);
 409         if (err)
 410                 goto err;
 411
 412         return 0;
 413
 414 err:
 415         if (mm) {
 416                 bprm->mm = NULL;
 417                 mmdrop(mm);
 418         }
 419
 420         return err;
 421 }
 422
 423 struct user_arg_ptr {
 424 #ifdef CONFIG_COMPAT
 425         bool is_compat;
 426 #endif
 427         union {
 428                 const char __user *const __user *native;
 429 #ifdef CONFIG_COMPAT
 430                 compat_uptr_t __user *compat;
 431 #endif
 432         } ptr;
 433 };
 434
 435 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 436 {
 437         const char __user *native;
 438
 439 #ifdef CONFIG_COMPAT
 440         if (unlikely(argv.is_compat)) {
 441                 compat_uptr_t compat;
 442
 443                 if (get_user(compat, argv.ptr.compat + nr))
 444                         return ERR_PTR(-EFAULT);
 445
 446                 return compat_ptr(compat);
 447         }
 448 #endif
 449
 450         if (get_user(native, argv.ptr.native + nr))
 451                 return ERR_PTR(-EFAULT);
 452
 453         return native;
 454 }
 455
 456 /*
 457  * count() counts the number of strings in array ARGV.
 458  */
 459 static int count(struct user_arg_ptr argv, int max)
 460 {
 461         int i = 0;
 462
 463         if (argv.ptr.native != NULL) {
 464                 for (;;) {
 465                         const char __user *p = get_user_arg_ptr(argv, i);
 466
 467                         if (!p)
 468                                 break;
 469
 470                         if (IS_ERR(p))
 471                                 return -EFAULT;
 472
 473                         if (i++ >= max)
 474                                 return -E2BIG;
 475
 476                         if (fatal_signal_pending(current))
 477                                 return -ERESTARTNOHAND;
 478                         cond_resched();
 479                 }
 480         }
 481         return i;
 482 }
 483
 484 /*
 485  * 'copy_strings()' copies argument/environment strings from the old
 486  * processes's memory to the new process's stack.  The call to get_user_pages()
 487  * ensures the destination page is created and not swapped out.
 488  */
 489 static int copy_strings(int argc, struct user_arg_ptr argv,
 490                         struct linux_binprm *bprm)
 491 {
 492         struct page *kmapped_page = NULL;
 493         char *kaddr = NULL;
 494         unsigned long kpos = 0;
 495         int ret;
 496
 497         while (argc-- > 0) {
 498                 const char __user *str;
 499                 int len;
 500                 unsigned long pos;
 501
 502                 ret = -EFAULT;
 503                 str = get_user_arg_ptr(argv, argc);
 504                 if (IS_ERR(str))
 505                         goto out;
 506
 507                 len = strnlen_user(str, MAX_ARG_STRLEN);
 508                 if (!len)
 509                         goto out;
 510
 511                 ret = -E2BIG;
 512                 if (!valid_arg_len(bprm, len))
 513                         goto out;
 514
 515                 /* We're going to work our way backwords. */
 516                 pos = bprm->p;
 517                 str += len;
 518                 bprm->p -= len;
 519
 520                 while (len > 0) {
 521                         int offset, bytes_to_copy;
 522
 523                         if (fatal_signal_pending(current)) {
 524                                 ret = -ERESTARTNOHAND;
 525                                 goto out;
 526                         }
 527                         cond_resched();
 528
 529                         offset = pos % PAGE_SIZE;
 530                         if (offset == 0)
 531                                 offset = PAGE_SIZE;
 532
 533                         bytes_to_copy = offset;
 534                         if (bytes_to_copy > len)
 535                                 bytes_to_copy = len;
 536
 537                         offset -= bytes_to_copy;
 538                         pos -= bytes_to_copy;
 539                         str -= bytes_to_copy;
 540                         len -= bytes_to_copy;
 541
 542                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 543                                 struct page *page;
 544
 545                                 page = get_arg_page(bprm, pos, 1);
 546                                 if (!page) {
 547                                         ret = -E2BIG;
 548                                         goto out;
 549                                 }
 550
 551                                 if (kmapped_page) {
 552                                         flush_kernel_dcache_page(kmapped_page);
 553                                         kunmap(kmapped_page);
 554                                         put_arg_page(kmapped_page);
 555                                 }
 556                                 kmapped_page = page;
 557                                 kaddr = kmap(kmapped_page);
 558                                 kpos = pos & PAGE_MASK;
 559                                 flush_arg_page(bprm, kpos, kmapped_page);
 560                         }
 561                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 562                                 ret = -EFAULT;
 563                                 goto out;
 564                         }
 565                 }
 566         }
 567         ret = 0;
 568 out:
 569         if (kmapped_page) {
 570                 flush_kernel_dcache_page(kmapped_page);
 571                 kunmap(kmapped_page);
 572                 put_arg_page(kmapped_page);
 573         }
 574         return ret;
 575 }
 576
 577 /*
 578  * Like copy_strings, but get argv and its values from kernel memory.
 579  */
 580 int copy_strings_kernel(int argc, const char *const *__argv,
 581                         struct linux_binprm *bprm)
 582 {
 583         int r;
 584         mm_segment_t oldfs = get_fs();
 585         struct user_arg_ptr argv = {
 586                 .ptr.native = (const char __user *const  __user *)__argv,
 587         };
 588
 589         set_fs(KERNEL_DS);
 590         r = copy_strings(argc, argv, bprm);
 591         set_fs(oldfs);
 592
 593         return r;
 594 }
 595 EXPORT_SYMBOL(copy_strings_kernel);
 596
 597 #ifdef CONFIG_MMU
 598
 599 /*
 600  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 601  * the binfmt code determines where the new stack should reside, we shift it to
 602  * its final location.  The process proceeds as follows:
 603  *
 604  * 1) Use shift to calculate the new vma endpoints.
 605  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 606  *    arguments passed to subsequent functions are consistent.
 607  * 3) Move vma's page tables to the new range.
 608  * 4) Free up any cleared pgd range.
 609  * 5) Shrink the vma to cover only the new range.
 610  */
 611 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 612 {
 613         struct mm_struct *mm = vma->vm_mm;
 614         unsigned long old_start = vma->vm_start;
 615         unsigned long old_end = vma->vm_end;
 616         unsigned long length = old_end - old_start;
 617         unsigned long new_start = old_start - shift;
 618         unsigned long new_end = old_end - shift;
 619         struct mmu_gather tlb;
 620
 621         BUG_ON(new_start > new_end);
 622
 623         /*
 624          * ensure there are no vmas between where we want to go
 625          * and where we are
 626          */
 627         if (vma != find_vma(mm, new_start))
 628                 return -EFAULT;
 629
 630         /*
 631          * cover the whole range: [new_start, old_end)
 632          */
 633         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 634                 return -ENOMEM;
 635
 636         /*
 637          * move the page tables downwards, on failure we rely on
 638          * process cleanup to remove whatever mess we made.
 639          */
 640         if (length != move_page_tables(vma, old_start,
 641                                        vma, new_start, length))
 642                 return -ENOMEM;
 643
 644         lru_add_drain();
 645         tlb_gather_mmu(&tlb, mm, 0);
 646         if (new_end > old_start) {
 647                 /*
 648                  * when the old and new regions overlap clear from new_end.
 649                  */
 650                 free_pgd_range(&tlb, new_end, old_end, new_end,
 651                         vma->vm_next ? vma->vm_next->vm_start : 0);
 652         } else {
 653                 /*
 654                  * otherwise, clean from old_start; this is done to not touch
 655                  * the address space in [new_end, old_start) some architectures
 656                  * have constraints on va-space that make this illegal (IA64) -
 657                  * for the others its just a little faster.
 658                  */
 659                 free_pgd_range(&tlb, old_start, old_end, new_end,
 660                         vma->vm_next ? vma->vm_next->vm_start : 0);
 661         }
 662         tlb_finish_mmu(&tlb, new_end, old_end);
 663
 664         /*
 665          * Shrink the vma to just the new range.  Always succeeds.
 666          */
 667         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 674  * the stack is optionally relocated, and some extra space is added.
 675  */
 676 int setup_arg_pages(struct linux_binprm *bprm,
 677                     unsigned long stack_top,
 678                     int executable_stack)
 679 {
 680         unsigned long ret;
 681         unsigned long stack_shift;
 682         struct mm_struct *mm = current->mm;
 683         struct vm_area_struct *vma = bprm->vma;
 684         struct vm_area_struct *prev = NULL;
 685         unsigned long vm_flags;
 686         unsigned long stack_base;
 687         unsigned long stack_size;
 688         unsigned long stack_expand;
 689         unsigned long rlim_stack;
 690
 691 #ifdef CONFIG_STACK_GROWSUP
 692         /* Limit stack size to 1GB */
 693         stack_base = rlimit_max(RLIMIT_STACK);
 694         if (stack_base > (1 << 30))
 695                 stack_base = 1 << 30;
 696
 697         /* Make sure we didn't let the argument array grow too large. */
 698         if (vma->vm_end - vma->vm_start > stack_base)
 699                 return -ENOMEM;
 700
 701         stack_base = PAGE_ALIGN(stack_top - stack_base);
 702
 703         stack_shift = vma->vm_start - stack_base;
 704         mm->arg_start = bprm->p - stack_shift;
 705         bprm->p = vma->vm_end - stack_shift;
 706 #else
 707         stack_top = arch_align_stack(stack_top);
 708         stack_top = PAGE_ALIGN(stack_top);
 709
 710         if (unlikely(stack_top < mmap_min_addr) ||
 711             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 712                 return -ENOMEM;
 713
 714         stack_shift = vma->vm_end - stack_top;
 715
 716         bprm->p -= stack_shift;
 717         mm->arg_start = bprm->p;
 718 #endif
 719
 720         if (bprm->loader)
 721                 bprm->loader -= stack_shift;
 722         bprm->exec -= stack_shift;
 723
 724         down_write(&mm->mmap_sem);
 725         vm_flags = VM_STACK_FLAGS;
 726
 727         /*
 728          * Adjust stack execute permissions; explicitly enable for
 729          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 730          * (arch default) otherwise.
 731          */
 732         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 733                 vm_flags |= VM_EXEC;
 734         else if (executable_stack == EXSTACK_DISABLE_X)
 735                 vm_flags &= ~VM_EXEC;
 736         vm_flags |= mm->def_flags;
 737         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 738
 739         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 740                         vm_flags);
 741         if (ret)
 742                 goto out_unlock;
 743         BUG_ON(prev != vma);
 744
 745         /* Move stack pages down in memory. */
 746         if (stack_shift) {
 747                 ret = shift_arg_pages(vma, stack_shift);
 748                 if (ret)
 749                         goto out_unlock;
 750         }
 751
 752         /* mprotect_fixup is overkill to remove the temporary stack flags */
 753         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 754
 755         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 756         stack_size = vma->vm_end - vma->vm_start;
 757         /*
 758          * Align this down to a page boundary as expand_stack
 759          * will align it up.
 760          */
 761         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 762 #ifdef CONFIG_STACK_GROWSUP
 763         if (stack_size + stack_expand > rlim_stack)
 764                 stack_base = vma->vm_start + rlim_stack;
 765         else
 766                 stack_base = vma->vm_end + stack_expand;
 767 #else
 768         if (stack_size + stack_expand > rlim_stack)
 769                 stack_base = vma->vm_end - rlim_stack;
 770         else
 771                 stack_base = vma->vm_start - stack_expand;
 772 #endif
 773         current->mm->start_stack = bprm->p;
 774         ret = expand_stack(vma, stack_base);
 775         if (ret)
 776                 ret = -EFAULT;
 777
 778 out_unlock:
 779         up_write(&mm->mmap_sem);
 780         return ret;
 781 }
 782 EXPORT_SYMBOL(setup_arg_pages);
 783
 784 #endif /* CONFIG_MMU */
 785
 786 struct file *open_exec(const char *name)
 787 {
 788         struct file *file;
 789         int err;
 790         static const struct open_flags open_exec_flags = {
 791                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 792                 .acc_mode = MAY_EXEC | MAY_OPEN,
 793                 .intent = LOOKUP_OPEN
 794         };
 795
 796         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 797         if (IS_ERR(file))
 798                 goto out;
 799
 800         err = -EACCES;
 801         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 802                 goto exit;
 803
 804         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 805                 goto exit;
 806
 807         fsnotify_open(file);
 808
 809         err = deny_write_access(file);
 810         if (err)
 811                 goto exit;
 812
 813 out:
 814         return file;
 815
 816 exit:
 817         fput(file);
 818         return ERR_PTR(err);
 819 }
 820 EXPORT_SYMBOL(open_exec);
 821
 822 int kernel_read(struct file *file, loff_t offset,
 823                 char *addr, unsigned long count)
 824 {
 825         mm_segment_t old_fs;
 826         loff_t pos = offset;
 827         int result;
 828
 829         old_fs = get_fs();
 830         set_fs(get_ds());
 831         /* The cast to a user pointer is valid due to the set_fs() */
 832         result = vfs_read(file, (void __user *)addr, count, &pos);
 833         set_fs(old_fs);
 834         return result;
 835 }
 836
 837 EXPORT_SYMBOL(kernel_read);
 838
 839 static int exec_mmap(struct mm_struct *mm)
 840 {
 841         struct task_struct *tsk;
 842         struct mm_struct * old_mm, *active_mm;
 843
 844         /* Notify parent that we're no longer interested in the old VM */
 845         tsk = current;
 846         old_mm = current->mm;
 847         sync_mm_rss(tsk, old_mm);
 848         mm_release(tsk, old_mm);
 849
 850         if (old_mm) {
 851                 /*
 852                  * Make sure that if there is a core dump in progress
 853                  * for the old mm, we get out and die instead of going
 854                  * through with the exec.  We must hold mmap_sem around
 855                  * checking core_state and changing tsk->mm.
 856                  */
 857                 down_read(&old_mm->mmap_sem);
 858                 if (unlikely(old_mm->core_state)) {
 859                         up_read(&old_mm->mmap_sem);
 860                         return -EINTR;
 861                 }
 862         }
 863         task_lock(tsk);
 864         active_mm = tsk->active_mm;
 865         tsk->mm = mm;
 866         tsk->active_mm = mm;
 867         activate_mm(active_mm, mm);
 868         task_unlock(tsk);
 869         arch_pick_mmap_layout(mm);
 870         if (old_mm) {
 871                 up_read(&old_mm->mmap_sem);
 872                 BUG_ON(active_mm != old_mm);
 873                 mm_update_next_owner(old_mm);
 874                 mmput(old_mm);
 875                 return 0;
 876         }
 877         mmdrop(active_mm);
 878         return 0;
 879 }
 880
 881 /*
 882  * This function makes sure the current process has its own signal table,
 883  * so that flush_signal_handlers can later reset the handlers without
 884  * disturbing other processes.  (Other processes might share the signal
 885  * table via the CLONE_SIGHAND option to clone().)
 886  */
 887 static int de_thread(struct task_struct *tsk)
 888 {
 889         struct signal_struct *sig = tsk->signal;
 890         struct sighand_struct *oldsighand = tsk->sighand;
 891         spinlock_t *lock = &oldsighand->siglock;
 892
 893         if (thread_group_empty(tsk))
 894                 goto no_thread_group;
 895
 896         /*
 897          * Kill all other threads in the thread group.
 898          */
 899         spin_lock_irq(lock);
 900         if (signal_group_exit(sig)) {
 901                 /*
 902                  * Another group action in progress, just
 903                  * return so that the signal is processed.
 904                  */
 905                 spin_unlock_irq(lock);
 906                 return -EAGAIN;
 907         }
 908
 909         sig->group_exit_task = tsk;
 910         sig->notify_count = zap_other_threads(tsk);
 911         if (!thread_group_leader(tsk))
 912                 sig->notify_count--;
 913
 914         while (sig->notify_count) {
 915                 __set_current_state(TASK_UNINTERRUPTIBLE);
 916                 spin_unlock_irq(lock);
 917                 schedule();
 918                 spin_lock_irq(lock);
 919         }
 920         spin_unlock_irq(lock);
 921
 922         /*
 923          * At this point all other threads have exited, all we have to
 924          * do is to wait for the thread group leader to become inactive,
 925          * and to assume its PID:
 926          */
 927         if (!thread_group_leader(tsk)) {
 928                 struct task_struct *leader = tsk->group_leader;
 929
 930                 sig->notify_count = -1; /* for exit_notify() */
 931                 for (;;) {
 932                         write_lock_irq(&tasklist_lock);
 933                         if (likely(leader->exit_state))
 934                                 break;
 935                         __set_current_state(TASK_UNINTERRUPTIBLE);
 936                         write_unlock_irq(&tasklist_lock);
 937                         schedule();
 938                 }
 939
 940                 /*
 941                  * The only record we have of the real-time age of a
 942                  * process, regardless of execs it's done, is start_time.
 943                  * All the past CPU time is accumulated in signal_struct
 944                  * from sister threads now dead.  But in this non-leader
 945                  * exec, nothing survives from the original leader thread,
 946                  * whose birth marks the true age of this process now.
 947                  * When we take on its identity by switching to its PID, we
 948                  * also take its birthdate (always earlier than our own).
 949                  */
 950                 tsk->start_time = leader->start_time;
 951
 952                 BUG_ON(!same_thread_group(leader, tsk));
 953                 BUG_ON(has_group_leader_pid(tsk));
 954                 /*
 955                  * An exec() starts a new thread group with the
 956                  * TGID of the previous thread group. Rehash the
 957                  * two threads with a switched PID, and release
 958                  * the former thread group leader:
 959                  */
 960
 961                 /* Become a process group leader with the old leader's pid.
 962                  * The old leader becomes a thread of the this thread group.
 963                  * Note: The old leader also uses this pid until release_task
 964                  *       is called.  Odd but simple and correct.
 965                  */
 966                 detach_pid(tsk, PIDTYPE_PID);
 967                 tsk->pid = leader->pid;
 968                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 969                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 970                 transfer_pid(leader, tsk, PIDTYPE_SID);
 971
 972                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 973                 list_replace_init(&leader->sibling, &tsk->sibling);
 974
 975                 tsk->group_leader = tsk;
 976                 leader->group_leader = tsk;
 977
 978                 tsk->exit_signal = SIGCHLD;
 979                 leader->exit_signal = -1;
 980
 981                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 982                 leader->exit_state = EXIT_DEAD;
 983
 984                 /*
 985                  * We are going to release_task()->ptrace_unlink() silently,
 986                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 987                  * the tracer wont't block again waiting for this thread.
 988                  */
 989                 if (unlikely(leader->ptrace))
 990                         __wake_up_parent(leader, leader->parent);
 991                 write_unlock_irq(&tasklist_lock);
 992
 993                 release_task(leader);
 994         }
 995
 996         sig->group_exit_task = NULL;
 997         sig->notify_count = 0;
 998
 999 no_thread_group:
1000         /* we have changed execution domain */
1001         tsk->exit_signal = SIGCHLD;
1002
1003         if (current->mm)
1004                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
1005
1006         exit_itimers(sig);
1007         flush_itimer_signals();
1008
1009         if (atomic_read(&oldsighand->count) != 1) {
1010                 struct sighand_struct *newsighand;
1011                 /*
1012                  * This ->sighand is shared with the CLONE_SIGHAND
1013                  * but not CLONE_THREAD task, switch to the new one.
1014                  */
1015                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1016                 if (!newsighand)
1017                         return -ENOMEM;
1018
1019                 atomic_set(&newsighand->count, 1);
1020                 memcpy(newsighand->action, oldsighand->action,
1021                        sizeof(newsighand->action));
1022
1023                 write_lock_irq(&tasklist_lock);
1024                 spin_lock(&oldsighand->siglock);
1025                 rcu_assign_pointer(tsk->sighand, newsighand);
1026                 spin_unlock(&oldsighand->siglock);
1027                 write_unlock_irq(&tasklist_lock);
1028
1029                 __cleanup_sighand(oldsighand);
1030         }
1031
1032         BUG_ON(!thread_group_leader(tsk));
1033         return 0;
1034 }
1035
1036 /*
1037  * These functions flushes out all traces of the currently running executable
1038  * so that a new one can be started
1039  */
1040 static void flush_old_files(struct files_struct * files)
1041 {
1042         long j = -1;
1043         struct fdtable *fdt;
1044
1045         spin_lock(&files->file_lock);
1046         for (;;) {
1047                 unsigned long set, i;
1048
1049                 j++;
1050                 i = j * __NFDBITS;
1051                 fdt = files_fdtable(files);
1052                 if (i >= fdt->max_fds)
1053                         break;
1054                 set = fdt->close_on_exec->fds_bits[j];
1055                 if (!set)
1056                         continue;
1057                 fdt->close_on_exec->fds_bits[j] = 0;
1058                 spin_unlock(&files->file_lock);
1059                 for ( ; set ; i++,set >>= 1) {
1060                         if (set & 1) {
1061                                 sys_close(i);
1062                         }
1063                 }
1064                 spin_lock(&files->file_lock);
1065
1066         }
1067         spin_unlock(&files->file_lock);
1068 }
1069
1070 char *get_task_comm(char *buf, struct task_struct *tsk)
1071 {
1072         /* buf must be at least sizeof(tsk->comm) in size */
1073         task_lock(tsk);
1074         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1075         task_unlock(tsk);
1076         return buf;
1077 }
1078 EXPORT_SYMBOL_GPL(get_task_comm);
1079
1080 void set_task_comm(struct task_struct *tsk, char *buf)
1081 {
1082         task_lock(tsk);
1083
1084         /*
1085          * Threads may access current->comm without holding
1086          * the task lock, so write the string carefully.
1087          * Readers without a lock may see incomplete new
1088          * names but are safe from non-terminating string reads.
1089          */
1090         memset(tsk->comm, 0, TASK_COMM_LEN);
1091         wmb();
1092         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1093         task_unlock(tsk);
1094         perf_event_comm(tsk);
1095 }
1096
1097 int flush_old_exec(struct linux_binprm * bprm)
1098 {
1099         int retval;
1100
1101         /*
1102          * Make sure we have a private signal table and that
1103          * we are unassociated from the previous thread group.
1104          */
1105         retval = de_thread(current);
1106         if (retval)
1107                 goto out;
1108
1109         set_mm_exe_file(bprm->mm, bprm->file);
1110
1111         /*
1112          * Release all of the old mmap stuff
1113          */
1114         acct_arg_size(bprm, 0);
1115         retval = exec_mmap(bprm->mm);
1116         if (retval)
1117                 goto out;
1118
1119         bprm->mm = NULL;                /* We're using it now */
1120
1121         set_fs(USER_DS);
1122         current->flags &=
1123                 ~(PF_RANDOMIZE | PF_KTHREAD | PF_NOFREEZE | PF_FREEZER_NOSIG);
1124         flush_thread();
1125         current->personality &= ~bprm->per_clear;
1126
1127         return 0;
1128
1129 out:
1130         return retval;
1131 }
1132 EXPORT_SYMBOL(flush_old_exec);
1133
1134 void would_dump(struct linux_binprm *bprm, struct file *file)
1135 {
1136         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1137                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1138 }
1139 EXPORT_SYMBOL(would_dump);
1140
1141 void setup_new_exec(struct linux_binprm * bprm)
1142 {
1143         int i, ch;
1144         const char *name;
1145         char tcomm[sizeof(current->comm)];
1146
1147         arch_pick_mmap_layout(current->mm);
1148
1149         /* This is the point of no return */
1150         current->sas_ss_sp = current->sas_ss_size = 0;
1151
1152         if (current_euid() == current_uid() && current_egid() == current_gid())
1153                 set_dumpable(current->mm, 1);
1154         else
1155                 set_dumpable(current->mm, suid_dumpable);
1156
1157         name = bprm->filename;
1158
1159         /* Copies the binary name from after last slash */
1160         for (i=0; (ch = *(name++)) != '\0';) {
1161                 if (ch == '/')
1162                         i = 0; /* overwrite what we wrote */
1163                 else
1164                         if (i < (sizeof(tcomm) - 1))
1165                                 tcomm[i++] = ch;
1166         }
1167         tcomm[i] = '\0';
1168         set_task_comm(current, tcomm);
1169
1170         /* Set the new mm task size. We have to do that late because it may
1171          * depend on TIF_32BIT which is only updated in flush_thread() on
1172          * some architectures like powerpc
1173          */
1174         current->mm->task_size = TASK_SIZE;
1175
1176         /* install the new credentials */
1177         if (bprm->cred->uid != current_euid() ||
1178             bprm->cred->gid != current_egid()) {
1179                 current->pdeath_signal = 0;
1180         } else {
1181                 would_dump(bprm, bprm->file);
1182                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1183                         set_dumpable(current->mm, suid_dumpable);
1184         }
1185
1186         /* An exec changes our domain. We are no longer part of the thread
1187            group */
1188
1189         current->self_exec_id++;
1190
1191         flush_signal_handlers(current, 0);
1192         flush_old_files(current->files);
1193 }
1194 EXPORT_SYMBOL(setup_new_exec);
1195
1196 /*
1197  * Prepare credentials and lock ->cred_guard_mutex.
1198  * install_exec_creds() commits the new creds and drops the lock.
1199  * Or, if exec fails before, free_bprm() should release ->cred and
1200  * and unlock.
1201  */
1202 int prepare_bprm_creds(struct linux_binprm *bprm)
1203 {
1204         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1205                 return -ERESTARTNOINTR;
1206
1207         bprm->cred = prepare_exec_creds();
1208         if (likely(bprm->cred))
1209                 return 0;
1210
1211         mutex_unlock(&current->signal->cred_guard_mutex);
1212         return -ENOMEM;
1213 }
1214
1215 void free_bprm(struct linux_binprm *bprm)
1216 {
1217         free_arg_pages(bprm);
1218         if (bprm->cred) {
1219                 mutex_unlock(&current->signal->cred_guard_mutex);
1220                 abort_creds(bprm->cred);
1221         }
1222         /* If a binfmt changed the interp, free it. */
1223         if (bprm->interp != bprm->filename)
1224                 kfree(bprm->interp);
1225         kfree(bprm);
1226 }
1227
1228 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1229 {
1230         /* If a binfmt changed the interp, free it first. */
1231         if (bprm->interp != bprm->filename)
1232                 kfree(bprm->interp);
1233         bprm->interp = kstrdup(interp, GFP_KERNEL);
1234         if (!bprm->interp)
1235                 return -ENOMEM;
1236         return 0;
1237 }
1238 EXPORT_SYMBOL(bprm_change_interp);
1239
1240 /*
1241  * install the new credentials for this executable
1242  */
1243 void install_exec_creds(struct linux_binprm *bprm)
1244 {
1245         security_bprm_committing_creds(bprm);
1246
1247         commit_creds(bprm->cred);
1248         bprm->cred = NULL;
1249
1250         /*
1251          * Disable monitoring for regular users
1252          * when executing setuid binaries. Must
1253          * wait until new credentials are committed
1254          * by commit_creds() above
1255          */
1256         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1257                 perf_event_exit_task(current);
1258         /*
1259          * cred_guard_mutex must be held at least to this point to prevent
1260          * ptrace_attach() from altering our determination of the task's
1261          * credentials; any time after this it may be unlocked.
1262          */
1263         security_bprm_committed_creds(bprm);
1264         mutex_unlock(&current->signal->cred_guard_mutex);
1265 }
1266 EXPORT_SYMBOL(install_exec_creds);
1267
1268 /*
1269  * determine how safe it is to execute the proposed program
1270  * - the caller must hold ->cred_guard_mutex to protect against
1271  *   PTRACE_ATTACH
1272  */
1273 int check_unsafe_exec(struct linux_binprm *bprm)
1274 {
1275         struct task_struct *p = current, *t;
1276         unsigned n_fs;
1277         int res = 0;
1278
1279         if (p->ptrace) {
1280                 if (p->ptrace & PT_PTRACE_CAP)
1281                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1282                 else
1283                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1284         }
1285
1286         n_fs = 1;
1287         spin_lock(&p->fs->lock);
1288         rcu_read_lock();
1289         for (t = next_thread(p); t != p; t = next_thread(t)) {
1290                 if (t->fs == p->fs)
1291                         n_fs++;
1292         }
1293         rcu_read_unlock();
1294
1295         if (p->fs->users > n_fs) {
1296                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1297         } else {
1298                 res = -EAGAIN;
1299                 if (!p->fs->in_exec) {
1300                         p->fs->in_exec = 1;
1301                         res = 1;
1302                 }
1303         }
1304         spin_unlock(&p->fs->lock);
1305
1306         return res;
1307 }
1308
1309 static void bprm_fill_uid(struct linux_binprm *bprm)
1310 {
1311         struct inode *inode;
1312         unsigned int mode;
1313         uid_t uid;
1314         gid_t gid;
1315
1316         /* clear any previous set[ug]id data from a previous binary */
1317         bprm->cred->euid = current_euid();
1318         bprm->cred->egid = current_egid();
1319
1320         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1321                 return;
1322
1323         inode = bprm->file->f_path.dentry->d_inode;
1324         mode = ACCESS_ONCE(inode->i_mode);
1325         if (!(mode & (S_ISUID|S_ISGID)))
1326                 return;
1327
1328         /* Be careful if suid/sgid is set */
1329         mutex_lock(&inode->i_mutex);
1330
1331         /* reload atomically mode/uid/gid now that lock held */
1332         mode = inode->i_mode;
1333         uid = inode->i_uid;
1334         gid = inode->i_gid;
1335         mutex_unlock(&inode->i_mutex);
1336
1337         if (mode & S_ISUID) {
1338                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1339                 bprm->cred->euid = uid;
1340         }
1341
1342         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1343                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1344                 bprm->cred->egid = gid;
1345         }
1346 }
1347
1348 /*
1349  * Fill the binprm structure from the inode.
1350  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1351  *
1352  * This may be called multiple times for binary chains (scripts for example).
1353  */
1354 int prepare_binprm(struct linux_binprm *bprm)
1355 {
1356         int retval;
1357
1358         if (bprm->file->f_op == NULL)
1359                 return -EACCES;
1360
1361         bprm_fill_uid(bprm);
1362
1363         /* fill in binprm security blob */
1364         retval = security_bprm_set_creds(bprm);
1365         if (retval)
1366                 return retval;
1367         bprm->cred_prepared = 1;
1368
1369         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1370         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1371 }
1372
1373 EXPORT_SYMBOL(prepare_binprm);
1374
1375 /*
1376  * Arguments are '\0' separated strings found at the location bprm->p
1377  * points to; chop off the first by relocating brpm->p to right after
1378  * the first '\0' encountered.
1379  */
1380 int remove_arg_zero(struct linux_binprm *bprm)
1381 {
1382         int ret = 0;
1383         unsigned long offset;
1384         char *kaddr;
1385         struct page *page;
1386
1387         if (!bprm->argc)
1388                 return 0;
1389
1390         do {
1391                 offset = bprm->p & ~PAGE_MASK;
1392                 page = get_arg_page(bprm, bprm->p, 0);
1393                 if (!page) {
1394                         ret = -EFAULT;
1395                         goto out;
1396                 }
1397                 kaddr = kmap_atomic(page, KM_USER0);
1398
1399                 for (; offset < PAGE_SIZE && kaddr[offset];
1400                                 offset++, bprm->p++)
1401                         ;
1402
1403                 kunmap_atomic(kaddr, KM_USER0);
1404                 put_arg_page(page);
1405
1406                 if (offset == PAGE_SIZE)
1407                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1408         } while (offset == PAGE_SIZE);
1409
1410         bprm->p++;
1411         bprm->argc--;
1412         ret = 0;
1413
1414 out:
1415         return ret;
1416 }
1417 EXPORT_SYMBOL(remove_arg_zero);
1418
1419 /*
1420  * cycle the list of binary formats handler, until one recognizes the image
1421  */
1422 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1423 {
1424         unsigned int depth = bprm->recursion_depth;
1425         int try,retval;
1426         struct linux_binfmt *fmt;
1427         pid_t old_pid;
1428
1429         /* This allows 4 levels of binfmt rewrites before failing hard. */
1430         if (depth > 5)
1431                 return -ELOOP;
1432
1433         retval = security_bprm_check(bprm);
1434         if (retval)
1435                 return retval;
1436
1437         retval = audit_bprm(bprm);
1438         if (retval)
1439                 return retval;
1440
1441         /* Need to fetch pid before load_binary changes it */
1442         rcu_read_lock();
1443         old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1444         rcu_read_unlock();
1445
1446         retval = -ENOENT;
1447         for (try=0; try<2; try++) {
1448                 read_lock(&binfmt_lock);
1449                 list_for_each_entry(fmt, &formats, lh) {
1450                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1451                         if (!fn)
1452                                 continue;
1453                         if (!try_module_get(fmt->module))
1454                                 continue;
1455                         read_unlock(&binfmt_lock);
1456                         bprm->recursion_depth = depth + 1;
1457                         retval = fn(bprm, regs);
1458                         bprm->recursion_depth = depth;
1459                         if (retval >= 0) {
1460                                 if (depth == 0)
1461                                         ptrace_event(PTRACE_EVENT_EXEC,
1462                                                         old_pid);
1463                                 put_binfmt(fmt);
1464                                 allow_write_access(bprm->file);
1465                                 if (bprm->file)
1466                                         fput(bprm->file);
1467                                 bprm->file = NULL;
1468                                 current->did_exec = 1;
1469                                 proc_exec_connector(current);
1470                                 return retval;
1471                         }
1472                         read_lock(&binfmt_lock);
1473                         put_binfmt(fmt);
1474                         if (retval != -ENOEXEC || bprm->mm == NULL)
1475                                 break;
1476                         if (!bprm->file) {
1477                                 read_unlock(&binfmt_lock);
1478                                 return retval;
1479                         }
1480                 }
1481                 read_unlock(&binfmt_lock);
1482 #ifdef CONFIG_MODULES
1483                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1484                         break;
1485                 } else {
1486 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1487                         if (printable(bprm->buf[0]) &&
1488                             printable(bprm->buf[1]) &&
1489                             printable(bprm->buf[2]) &&
1490                             printable(bprm->buf[3]))
1491                                 break; /* -ENOEXEC */
1492                         if (try)
1493                                 break; /* -ENOEXEC */
1494                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1495                 }
1496 #else
1497                 break;
1498 #endif
1499         }
1500         return retval;
1501 }
1502
1503 EXPORT_SYMBOL(search_binary_handler);
1504
1505 /*
1506  * sys_execve() executes a new program.
1507  */
1508 static int do_execve_common(const char *filename,
1509                                 struct user_arg_ptr argv,
1510                                 struct user_arg_ptr envp,
1511                                 struct pt_regs *regs)
1512 {
1513         struct linux_binprm *bprm;
1514         struct file *file;
1515         struct files_struct *displaced;
1516         bool clear_in_exec;
1517         int retval;
1518         const struct cred *cred = current_cred();
1519
1520         /*
1521          * We move the actual failure in case of RLIMIT_NPROC excess from
1522          * set*uid() to execve() because too many poorly written programs
1523          * don't check setuid() return code.  Here we additionally recheck
1524          * whether NPROC limit is still exceeded.
1525          */
1526         if ((current->flags & PF_NPROC_EXCEEDED) &&
1527             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1528                 retval = -EAGAIN;
1529                 goto out_ret;
1530         }
1531
1532         /* We're below the limit (still or again), so we don't want to make
1533          * further execve() calls fail. */
1534         current->flags &= ~PF_NPROC_EXCEEDED;
1535
1536         retval = unshare_files(&displaced);
1537         if (retval)
1538                 goto out_ret;
1539
1540         retval = -ENOMEM;
1541         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1542         if (!bprm)
1543                 goto out_files;
1544
1545         retval = prepare_bprm_creds(bprm);
1546         if (retval)
1547                 goto out_free;
1548
1549         retval = check_unsafe_exec(bprm);
1550         if (retval < 0)
1551                 goto out_free;
1552         clear_in_exec = retval;
1553         current->in_execve = 1;
1554
1555         file = open_exec(filename);
1556         retval = PTR_ERR(file);
1557         if (IS_ERR(file))
1558                 goto out_unmark;
1559
1560         sched_exec();
1561
1562         bprm->file = file;
1563         bprm->filename = filename;
1564         bprm->interp = filename;
1565
1566         retval = bprm_mm_init(bprm);
1567         if (retval)
1568                 goto out_file;
1569
1570         bprm->argc = count(argv, MAX_ARG_STRINGS);
1571         if ((retval = bprm->argc) < 0)
1572                 goto out;
1573
1574         bprm->envc = count(envp, MAX_ARG_STRINGS);
1575         if ((retval = bprm->envc) < 0)
1576                 goto out;
1577
1578         retval = prepare_binprm(bprm);
1579         if (retval < 0)
1580                 goto out;
1581
1582         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1583         if (retval < 0)
1584                 goto out;
1585
1586         bprm->exec = bprm->p;
1587         retval = copy_strings(bprm->envc, envp, bprm);
1588         if (retval < 0)
1589                 goto out;
1590
1591         retval = copy_strings(bprm->argc, argv, bprm);
1592         if (retval < 0)
1593                 goto out;
1594
1595         retval = search_binary_handler(bprm,regs);
1596         if (retval < 0)
1597                 goto out;
1598
1599         /* execve succeeded */
1600         current->fs->in_exec = 0;
1601         current->in_execve = 0;
1602         acct_update_integrals(current);
1603         free_bprm(bprm);
1604         if (displaced)
1605                 put_files_struct(displaced);
1606         return retval;
1607
1608 out:
1609         if (bprm->mm) {
1610                 acct_arg_size(bprm, 0);
1611                 mmput(bprm->mm);
1612         }
1613
1614 out_file:
1615         if (bprm->file) {
1616                 allow_write_access(bprm->file);
1617                 fput(bprm->file);
1618         }
1619
1620 out_unmark:
1621         if (clear_in_exec)
1622                 current->fs->in_exec = 0;
1623         current->in_execve = 0;
1624
1625 out_free:
1626         free_bprm(bprm);
1627
1628 out_files:
1629         if (displaced)
1630                 reset_files_struct(displaced);
1631 out_ret:
1632         return retval;
1633 }
1634
1635 int do_execve(const char *filename,
1636         const char __user *const __user *__argv,
1637         const char __user *const __user *__envp,
1638         struct pt_regs *regs)
1639 {
1640         struct user_arg_ptr argv = { .ptr.native = __argv };
1641         struct user_arg_ptr envp = { .ptr.native = __envp };
1642         return do_execve_common(filename, argv, envp, regs);
1643 }
1644
1645 #ifdef CONFIG_COMPAT
1646 int compat_do_execve(char *filename,
1647         compat_uptr_t __user *__argv,
1648         compat_uptr_t __user *__envp,
1649         struct pt_regs *regs)
1650 {
1651         struct user_arg_ptr argv = {
1652                 .is_compat = true,
1653                 .ptr.compat = __argv,
1654         };
1655         struct user_arg_ptr envp = {
1656                 .is_compat = true,
1657                 .ptr.compat = __envp,
1658         };
1659         return do_execve_common(filename, argv, envp, regs);
1660 }
1661 #endif
1662
1663 void set_binfmt(struct linux_binfmt *new)
1664 {
1665         struct mm_struct *mm = current->mm;
1666
1667         if (mm->binfmt)
1668                 module_put(mm->binfmt->module);
1669
1670         mm->binfmt = new;
1671         if (new)
1672                 __module_get(new->module);
1673 }
1674
1675 EXPORT_SYMBOL(set_binfmt);
1676
1677 static int expand_corename(struct core_name *cn)
1678 {
1679         char *old_corename = cn->corename;
1680
1681         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1682         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1683
1684         if (!cn->corename) {
1685                 kfree(old_corename);
1686                 return -ENOMEM;
1687         }
1688
1689         return 0;
1690 }
1691
1692 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1693 {
1694         char *cur;
1695         int need;
1696         int ret;
1697         va_list arg;
1698
1699         va_start(arg, fmt);
1700         need = vsnprintf(NULL, 0, fmt, arg);
1701         va_end(arg);
1702
1703         if (likely(need < cn->size - cn->used - 1))
1704                 goto out_printf;
1705
1706         ret = expand_corename(cn);
1707         if (ret)
1708                 goto expand_fail;
1709
1710 out_printf:
1711         cur = cn->corename + cn->used;
1712         va_start(arg, fmt);
1713         vsnprintf(cur, need + 1, fmt, arg);
1714         va_end(arg);
1715         cn->used += need;
1716         return 0;
1717
1718 expand_fail:
1719         return ret;
1720 }
1721
1722 static void cn_escape(char *str)
1723 {
1724         for (; *str; str++)
1725                 if (*str == '/')
1726                         *str = '!';
1727 }
1728
1729 static int cn_print_exe_file(struct core_name *cn)
1730 {
1731         struct file *exe_file;
1732         char *pathbuf, *path;
1733         int ret;
1734
1735         exe_file = get_mm_exe_file(current->mm);
1736         if (!exe_file) {
1737                 char *commstart = cn->corename + cn->used;
1738                 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1739                 cn_escape(commstart);
1740                 return ret;
1741         }
1742
1743         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1744         if (!pathbuf) {
1745                 ret = -ENOMEM;
1746                 goto put_exe_file;
1747         }
1748
1749         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1750         if (IS_ERR(path)) {
1751                 ret = PTR_ERR(path);
1752                 goto free_buf;
1753         }
1754
1755         cn_escape(path);
1756
1757         ret = cn_printf(cn, "%s", path);
1758
1759 free_buf:
1760         kfree(pathbuf);
1761 put_exe_file:
1762         fput(exe_file);
1763         return ret;
1764 }
1765
1766 /* format_corename will inspect the pattern parameter, and output a
1767  * name into corename, which must have space for at least
1768  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1769  */
1770 static int format_corename(struct core_name *cn, long signr)
1771 {
1772         const struct cred *cred = current_cred();
1773         const char *pat_ptr = core_pattern;
1774         int ispipe = (*pat_ptr == '|');
1775         int pid_in_pattern = 0;
1776         int err = 0;
1777
1778         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1779         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1780         cn->used = 0;
1781
1782         if (!cn->corename)
1783                 return -ENOMEM;
1784
1785         /* Repeat as long as we have more pattern to process and more output
1786            space */
1787         while (*pat_ptr) {
1788                 if (*pat_ptr != '%') {
1789                         if (*pat_ptr == 0)
1790                                 goto out;
1791                         err = cn_printf(cn, "%c", *pat_ptr++);
1792                 } else {
1793                         switch (*++pat_ptr) {
1794                         /* single % at the end, drop that */
1795                         case 0:
1796                                 goto out;
1797                         /* Double percent, output one percent */
1798                         case '%':
1799                                 err = cn_printf(cn, "%c", '%');
1800                                 break;
1801                         /* pid */
1802                         case 'p':
1803                                 pid_in_pattern = 1;
1804                                 err = cn_printf(cn, "%d",
1805                                               task_tgid_vnr(current));
1806                                 break;
1807                         /* uid */
1808                         case 'u':
1809                                 err = cn_printf(cn, "%d", cred->uid);
1810                                 break;
1811                         /* gid */
1812                         case 'g':
1813                                 err = cn_printf(cn, "%d", cred->gid);
1814                                 break;
1815                         /* signal that caused the coredump */
1816                         case 's':
1817                                 err = cn_printf(cn, "%ld", signr);
1818                                 break;
1819                         /* UNIX time of coredump */
1820                         case 't': {
1821                                 struct timeval tv;
1822                                 do_gettimeofday(&tv);
1823                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1824                                 break;
1825                         }
1826                         /* hostname */
1827                         case 'h': {
1828                                 char *namestart = cn->corename + cn->used;
1829                                 down_read(&uts_sem);
1830                                 err = cn_printf(cn, "%s",
1831                                               utsname()->nodename);
1832                                 up_read(&uts_sem);
1833                                 cn_escape(namestart);
1834                                 break;
1835                         }
1836                         /* executable */
1837                         case 'e': {
1838                                 char *commstart = cn->corename + cn->used;
1839                                 err = cn_printf(cn, "%s", current->comm);
1840                                 cn_escape(commstart);
1841                                 break;
1842                         }
1843                         case 'E':
1844                                 err = cn_print_exe_file(cn);
1845                                 break;
1846                         /* core limit size */
1847                         case 'c':
1848                                 err = cn_printf(cn, "%lu",
1849                                               rlimit(RLIMIT_CORE));
1850                                 break;
1851                         default:
1852                                 break;
1853                         }
1854                         ++pat_ptr;
1855                 }
1856
1857                 if (err)
1858                         return err;
1859         }
1860
1861         /* Backward compatibility with core_uses_pid:
1862          *
1863          * If core_pattern does not include a %p (as is the default)
1864          * and core_uses_pid is set, then .%pid will be appended to
1865          * the filename. Do not do this for piped commands. */
1866         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1867                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1868                 if (err)
1869                         return err;
1870         }
1871 out:
1872         return ispipe;
1873 }
1874
1875 static int zap_process(struct task_struct *start, int exit_code)
1876 {
1877         struct task_struct *t;
1878         int nr = 0;
1879
1880         start->signal->flags = SIGNAL_GROUP_EXIT;
1881         start->signal->group_exit_code = exit_code;
1882         start->signal->group_stop_count = 0;
1883
1884         t = start;
1885         do {
1886                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1887                 if (t != current && t->mm) {
1888                         sigaddset(&t->pending.signal, SIGKILL);
1889                         signal_wake_up(t, 1);
1890                         nr++;
1891                 }
1892         } while_each_thread(start, t);
1893
1894         return nr;
1895 }
1896
1897 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1898                                 struct core_state *core_state, int exit_code)
1899 {
1900         struct task_struct *g, *p;
1901         unsigned long flags;
1902         int nr = -EAGAIN;
1903
1904         spin_lock_irq(&tsk->sighand->siglock);
1905         if (!signal_group_exit(tsk->signal)) {
1906                 mm->core_state = core_state;
1907                 nr = zap_process(tsk, exit_code);
1908         }
1909         spin_unlock_irq(&tsk->sighand->siglock);
1910         if (unlikely(nr < 0))
1911                 return nr;
1912
1913         if (atomic_read(&mm->mm_users) == nr + 1)
1914                 goto done;
1915         /*
1916          * We should find and kill all tasks which use this mm, and we should
1917          * count them correctly into ->nr_threads. We don't take tasklist
1918          * lock, but this is safe wrt:
1919          *
1920          * fork:
1921          *      None of sub-threads can fork after zap_process(leader). All
1922          *      processes which were created before this point should be
1923          *      visible to zap_threads() because copy_process() adds the new
1924          *      process to the tail of init_task.tasks list, and lock/unlock
1925          *      of ->siglock provides a memory barrier.
1926          *
1927          * do_exit:
1928          *      The caller holds mm->mmap_sem. This means that the task which
1929          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1930          *      its ->mm.
1931          *
1932          * de_thread:
1933          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1934          *      we must see either old or new leader, this does not matter.
1935          *      However, it can change p->sighand, so lock_task_sighand(p)
1936          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1937          *      it can't fail.
1938          *
1939          *      Note also that "g" can be the old leader with ->mm == NULL
1940          *      and already unhashed and thus removed from ->thread_group.
1941          *      This is OK, __unhash_process()->list_del_rcu() does not
1942          *      clear the ->next pointer, we will find the new leader via
1943          *      next_thread().
1944          */
1945         rcu_read_lock();
1946         for_each_process(g) {
1947                 if (g == tsk->group_leader)
1948                         continue;
1949                 if (g->flags & PF_KTHREAD)
1950                         continue;
1951                 p = g;
1952                 do {
1953                         if (p->mm) {
1954                                 if (unlikely(p->mm == mm)) {
1955                                         lock_task_sighand(p, &flags);
1956                                         nr += zap_process(p, exit_code);
1957                                         unlock_task_sighand(p, &flags);
1958                                 }
1959                                 break;
1960                         }
1961                 } while_each_thread(g, p);
1962         }
1963         rcu_read_unlock();
1964 done:
1965         atomic_set(&core_state->nr_threads, nr);
1966         return nr;
1967 }
1968
1969 static int coredump_wait(int exit_code, struct core_state *core_state)
1970 {
1971         struct task_struct *tsk = current;
1972         struct mm_struct *mm = tsk->mm;
1973         struct completion *vfork_done;
1974         int core_waiters = -EBUSY;
1975
1976         init_completion(&core_state->startup);
1977         core_state->dumper.task = tsk;
1978         core_state->dumper.next = NULL;
1979
1980         down_write(&mm->mmap_sem);
1981         if (!mm->core_state)
1982                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1983         up_write(&mm->mmap_sem);
1984
1985         if (unlikely(core_waiters < 0))
1986                 goto fail;
1987
1988         /*
1989          * Make sure nobody is waiting for us to release the VM,
1990          * otherwise we can deadlock when we wait on each other
1991          */
1992         vfork_done = tsk->vfork_done;
1993         if (vfork_done) {
1994                 tsk->vfork_done = NULL;
1995                 complete(vfork_done);
1996         }
1997
1998         if (core_waiters > 0) {
1999                 freezer_do_not_count();
2000                 wait_for_completion(&core_state->startup);
2001                 freezer_count();
2002         }
2003 fail:
2004         return core_waiters;
2005 }
2006
2007 static void coredump_finish(struct mm_struct *mm)
2008 {
2009         struct core_thread *curr, *next;
2010         struct task_struct *task;
2011
2012         next = mm->core_state->dumper.next;
2013         while ((curr = next) != NULL) {
2014                 next = curr->next;
2015                 task = curr->task;
2016                 /*
2017                  * see exit_mm(), curr->task must not see
2018                  * ->task == NULL before we read ->next.
2019                  */
2020                 smp_mb();
2021                 curr->task = NULL;
2022                 wake_up_process(task);
2023         }
2024
2025         mm->core_state = NULL;
2026 }
2027
2028 /*
2029  * set_dumpable converts traditional three-value dumpable to two flags and
2030  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
2031  * these bits are not changed atomically.  So get_dumpable can observe the
2032  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
2033  * return either old dumpable or new one by paying attention to the order of
2034  * modifying the bits.
2035  *
2036  * dumpable |   mm->flags (binary)
2037  * old  new | initial interim  final
2038  * ---------+-----------------------
2039  *  0    1  |   00      01      01
2040  *  0    2  |   00      10(*)   11
2041  *  1    0  |   01      00      00
2042  *  1    2  |   01      11      11
2043  *  2    0  |   11      10(*)   00
2044  *  2    1  |   11      11      01
2045  *
2046  * (*) get_dumpable regards interim value of 10 as 11.
2047  */
2048 void set_dumpable(struct mm_struct *mm, int value)
2049 {
2050         switch (value) {
2051         case 0:
2052                 clear_bit(MMF_DUMPABLE, &mm->flags);
2053                 smp_wmb();
2054                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2055                 break;
2056         case 1:
2057                 set_bit(MMF_DUMPABLE, &mm->flags);
2058                 smp_wmb();
2059                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2060                 break;
2061         case 2:
2062                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2063                 smp_wmb();
2064                 set_bit(MMF_DUMPABLE, &mm->flags);
2065                 break;
2066         }
2067 }
2068
2069 static int __get_dumpable(unsigned long mm_flags)
2070 {
2071         int ret;
2072
2073         ret = mm_flags & MMF_DUMPABLE_MASK;
2074         return (ret >= 2) ? 2 : ret;
2075 }
2076
2077 /*
2078  * This returns the actual value of the suid_dumpable flag. For things
2079  * that are using this for checking for privilege transitions, it must
2080  * test against SUID_DUMP_USER rather than treating it as a boolean
2081  * value.
2082  */
2083 int get_dumpable(struct mm_struct *mm)
2084 {
2085         return __get_dumpable(mm->flags);
2086 }
2087
2088 static void wait_for_dump_helpers(struct file *file)
2089 {
2090         struct pipe_inode_info *pipe;
2091
2092         pipe = file->f_path.dentry->d_inode->i_pipe;
2093
2094         pipe_lock(pipe);
2095         pipe->readers++;
2096         pipe->writers--;
2097
2098         while ((pipe->readers > 1) && (!signal_pending(current))) {
2099                 wake_up_interruptible_sync(&pipe->wait);
2100                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2101                 pipe_wait(pipe);
2102         }
2103
2104         pipe->readers--;
2105         pipe->writers++;
2106         pipe_unlock(pipe);
2107
2108 }
2109
2110
2111 /*
2112  * umh_pipe_setup
2113  * helper function to customize the process used
2114  * to collect the core in userspace.  Specifically
2115  * it sets up a pipe and installs it as fd 0 (stdin)
2116  * for the process.  Returns 0 on success, or
2117  * PTR_ERR on failure.
2118  * Note that it also sets the core limit to 1.  This
2119  * is a special value that we use to trap recursive
2120  * core dumps
2121  */
2122 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2123 {
2124         struct file *rp, *wp;
2125         struct fdtable *fdt;
2126         struct coredump_params *cp = (struct coredump_params *)info->data;
2127         struct files_struct *cf = current->files;
2128
2129         wp = create_write_pipe(0);
2130         if (IS_ERR(wp))
2131                 return PTR_ERR(wp);
2132
2133         rp = create_read_pipe(wp, 0);
2134         if (IS_ERR(rp)) {
2135                 free_write_pipe(wp);
2136                 return PTR_ERR(rp);
2137         }
2138
2139         cp->file = wp;
2140
2141         sys_close(0);
2142         fd_install(0, rp);
2143         spin_lock(&cf->file_lock);
2144         fdt = files_fdtable(cf);
2145         __set_open_fd(0, fdt);
2146         __clear_close_on_exec(0, fdt);
2147         spin_unlock(&cf->file_lock);
2148
2149         /* and disallow core files too */
2150         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2151
2152         return 0;
2153 }
2154
2155 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2156 {
2157         struct core_state core_state;
2158         struct core_name cn;
2159         struct mm_struct *mm = current->mm;
2160         struct linux_binfmt * binfmt;
2161         const struct cred *old_cred;
2162         struct cred *cred;
2163         int retval = 0;
2164         int ispipe;
2165         /* require nonrelative corefile path and be extra careful */
2166         bool need_suid_safe = false;
2167         static atomic_t core_dump_count = ATOMIC_INIT(0);
2168         struct coredump_params cprm = {
2169                 .signr = signr,
2170                 .regs = regs,
2171                 .limit = rlimit(RLIMIT_CORE),
2172                 /*
2173                  * We must use the same mm->flags while dumping core to avoid
2174                  * inconsistency of bit flags, since this flag is not protected
2175                  * by any locks.
2176                  */
2177                 .mm_flags = mm->flags,
2178         };
2179
2180         audit_core_dumps(signr);
2181
2182         binfmt = mm->binfmt;
2183         if (!binfmt || !binfmt->core_dump)
2184                 goto fail;
2185         if (!__get_dumpable(cprm.mm_flags))
2186                 goto fail;
2187
2188         cred = prepare_creds();
2189         if (!cred)
2190                 goto fail;
2191         /*
2192          * We cannot trust fsuid as being the "true" uid of the process
2193          * nor do we know its entire history. We only know it was tainted
2194          * so we dump it as root in mode 2, and only into a controlled
2195          * environment (pipe handler or fully qualified path).
2196          */
2197         if (__get_dumpable(cprm.mm_flags) == 2) {
2198                 /* Setuid core dump mode */
2199                 cred->fsuid = 0;        /* Dump root private */
2200                 need_suid_safe = true;
2201         }
2202
2203         retval = coredump_wait(exit_code, &core_state);
2204         if (retval < 0)
2205                 goto fail_creds;
2206
2207         old_cred = override_creds(cred);
2208
2209         /*
2210          * Clear any false indication of pending signals that might
2211          * be seen by the filesystem code called to write the core file.
2212          */
2213         clear_thread_flag(TIF_SIGPENDING);
2214
2215         ispipe = format_corename(&cn, signr);
2216
2217         if (ispipe) {
2218                 int dump_count;
2219                 char **helper_argv;
2220
2221                 if (ispipe < 0) {
2222                         printk(KERN_WARNING "format_corename failed\n");
2223                         printk(KERN_WARNING "Aborting core\n");
2224                         goto fail_corename;
2225                 }
2226
2227                 if (cprm.limit == 1) {
2228                         /*
2229                          * Normally core limits are irrelevant to pipes, since
2230                          * we're not writing to the file system, but we use
2231                          * cprm.limit of 1 here as a speacial value. Any
2232                          * non-1 limit gets set to RLIM_INFINITY below, but
2233                          * a limit of 0 skips the dump.  This is a consistent
2234                          * way to catch recursive crashes.  We can still crash
2235                          * if the core_pattern binary sets RLIM_CORE =  !1
2236                          * but it runs as root, and can do lots of stupid things
2237                          * Note that we use task_tgid_vnr here to grab the pid
2238                          * of the process group leader.  That way we get the
2239                          * right pid if a thread in a multi-threaded
2240                          * core_pattern process dies.
2241                          */
2242                         printk(KERN_WARNING
2243                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2244                                 task_tgid_vnr(current), current->comm);
2245                         printk(KERN_WARNING "Aborting core\n");
2246                         goto fail_unlock;
2247                 }
2248                 cprm.limit = RLIM_INFINITY;
2249
2250                 dump_count = atomic_inc_return(&core_dump_count);
2251                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2252                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2253                                task_tgid_vnr(current), current->comm);
2254                         printk(KERN_WARNING "Skipping core dump\n");
2255                         goto fail_dropcount;
2256                 }
2257
2258                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2259                 if (!helper_argv) {
2260                         printk(KERN_WARNING "%s failed to allocate memory\n",
2261                                __func__);
2262                         goto fail_dropcount;
2263                 }
2264
2265                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2266                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2267                                         NULL, &cprm);
2268                 argv_free(helper_argv);
2269                 if (retval) {
2270                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2271                                cn.corename);
2272                         goto close_fail;
2273                 }
2274         } else {
2275                 struct inode *inode;
2276                 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
2277                                  O_LARGEFILE | O_EXCL;
2278
2279                 if (cprm.limit < binfmt->min_coredump)
2280                         goto fail_unlock;
2281
2282                 if (need_suid_safe && cn.corename[0] != '/') {
2283                         printk(KERN_WARNING "Pid %d(%s) can only dump core "\
2284                                 "to fully qualified path!\n",
2285                                 task_tgid_vnr(current), current->comm);
2286                         printk(KERN_WARNING "Skipping core dump\n");
2287                         goto fail_unlock;
2288                 }
2289
2290                 /*
2291                  * Unlink the file if it exists unless this is a SUID
2292                  * binary - in that case, we're running around with root
2293                  * privs and don't want to unlink another user's coredump.
2294                  */
2295                 if (!need_suid_safe) {
2296                         mm_segment_t old_fs;
2297
2298                         old_fs = get_fs();
2299                         set_fs(KERNEL_DS);
2300                         /*
2301                          * If it doesn't exist, that's fine. If there's some
2302                          * other problem, we'll catch it at the filp_open().
2303                          */
2304                         (void) sys_unlink((const char __user *)cn.corename);
2305                         set_fs(old_fs);
2306                 }
2307
2308                 /*
2309                  * There is a race between unlinking and creating the
2310                  * file, but if that causes an EEXIST here, that's
2311                  * fine - another process raced with us while creating
2312                  * the corefile, and the other process won. To userspace,
2313                  * what matters is that at least one of the two processes
2314                  * writes its coredump successfully, not which one.
2315                  */
2316                 if (need_suid_safe) {
2317                         /*
2318                          * Using user namespaces, normal user tasks can change
2319                          * their current->fs->root to point to arbitrary
2320                          * directories. Since the intention of the "only dump
2321                          * with a fully qualified path" rule is to control where
2322                          * coredumps may be placed using root privileges,
2323                          * current->fs->root must not be used. Instead, use the
2324                          * root directory of init_task.
2325                          */
2326                         struct path root;
2327
2328                         task_lock(&init_task);
2329                         get_fs_root(init_task.fs, &root);
2330                         task_unlock(&init_task);
2331                         cprm.file = file_open_root(root.dentry, root.mnt,
2332                                 cn.corename, open_flags, 0600);
2333                         path_put(&root);
2334                 } else {
2335                         cprm.file = filp_open(cn.corename, open_flags, 0600);
2336                 }
2337                 if (IS_ERR(cprm.file))
2338                         goto fail_unlock;
2339
2340                 inode = cprm.file->f_path.dentry->d_inode;
2341                 if (inode->i_nlink > 1)
2342                         goto close_fail;
2343                 if (d_unhashed(cprm.file->f_path.dentry))
2344                         goto close_fail;
2345                 /*
2346                  * AK: actually i see no reason to not allow this for named
2347                  * pipes etc, but keep the previous behaviour for now.
2348                  */
2349                 if (!S_ISREG(inode->i_mode))
2350                         goto close_fail;
2351                 /*
2352                  * Dont allow local users get cute and trick others to coredump
2353                  * into their pre-created files.
2354                  */
2355                 if (inode->i_uid != current_fsuid())
2356                         goto close_fail;
2357                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2358                         goto close_fail;
2359                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2360                         goto close_fail;
2361         }
2362
2363         retval = binfmt->core_dump(&cprm);
2364         if (retval)
2365                 current->signal->group_exit_code |= 0x80;
2366
2367         if (ispipe && core_pipe_limit)
2368                 wait_for_dump_helpers(cprm.file);
2369 close_fail:
2370         if (cprm.file)
2371                 filp_close(cprm.file, NULL);
2372 fail_dropcount:
2373         if (ispipe)
2374                 atomic_dec(&core_dump_count);
2375 fail_unlock:
2376         kfree(cn.corename);
2377 fail_corename:
2378         coredump_finish(mm);
2379         revert_creds(old_cred);
2380 fail_creds:
2381         put_cred(cred);
2382 fail:
2383         return;
2384 }
2385
2386 /*
2387  * Core dumping helper functions.  These are the only things you should
2388  * do on a core-file: use only these functions to write out all the
2389  * necessary info.
2390  */
2391 int dump_write(struct file *file, const void *addr, int nr)
2392 {
2393         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2394 }
2395 EXPORT_SYMBOL(dump_write);
2396
2397 int dump_seek(struct file *file, loff_t off)
2398 {
2399         int ret = 1;
2400
2401         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2402                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2403                         return 0;
2404         } else {
2405                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2406
2407                 if (!buf)
2408                         return 0;
2409                 while (off > 0) {
2410                         unsigned long n = off;
2411
2412                         if (n > PAGE_SIZE)
2413                                 n = PAGE_SIZE;
2414                         if (!dump_write(file, buf, n)) {
2415                                 ret = 0;
2416                                 break;
2417                         }
2418                         off -= n;
2419                 }
2420                 free_page((unsigned long)buf);
2421         }
2422         return ret;
2423 }
2424 EXPORT_SYMBOL(dump_seek);