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/smp_lock.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/proc_fs.h>
  47 #include <linux/mount.h>
  48 #include <linux/security.h>
  49 #include <linux/ima.h>
  50 #include <linux/syscalls.h>
  51 #include <linux/tsacct_kern.h>
  52 #include <linux/cn_proc.h>
  53 #include <linux/audit.h>
  54 #include <linux/tracehook.h>
  55 #include <linux/kmod.h>
  56 #include <linux/fsnotify.h>
  57 #include <linux/fs_struct.h>
  58 #include <linux/pipe_fs_i.h>
  59
  60 #include <asm/uaccess.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/tlb.h>
  63 #include "internal.h"
  64
  65 int core_uses_pid;
  66 char core_pattern[CORENAME_MAX_SIZE] = "core";
  67 unsigned int core_pipe_limit;
  68 int suid_dumpable = 0;
  69
  70 /* The maximal length of core_pattern is also specified in sysctl.c */
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  76 {
  77         if (!fmt)
  78                 return -EINVAL;
  79         write_lock(&binfmt_lock);
  80         insert ? list_add(&fmt->lh, &formats) :
  81                  list_add_tail(&fmt->lh, &formats);
  82         write_unlock(&binfmt_lock);
  83         return 0;
  84 }
  85
  86 EXPORT_SYMBOL(__register_binfmt);
  87
  88 void unregister_binfmt(struct linux_binfmt * fmt)
  89 {
  90         write_lock(&binfmt_lock);
  91         list_del(&fmt->lh);
  92         write_unlock(&binfmt_lock);
  93 }
  94
  95 EXPORT_SYMBOL(unregister_binfmt);
  96
  97 static inline void put_binfmt(struct linux_binfmt * fmt)
  98 {
  99         module_put(fmt->module);
 100 }
 101
 102 /*
 103  * Note that a shared library must be both readable and executable due to
 104  * security reasons.
 105  *
 106  * Also note that we take the address to load from from the file itself.
 107  */
 108 SYSCALL_DEFINE1(uselib, const char __user *, library)
 109 {
 110         struct file *file;
 111         char *tmp = getname(library);
 112         int error = PTR_ERR(tmp);
 113
 114         if (IS_ERR(tmp))
 115                 goto out;
 116
 117         file = do_filp_open(AT_FDCWD, tmp,
 118                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 119                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 120         putname(tmp);
 121         error = PTR_ERR(file);
 122         if (IS_ERR(file))
 123                 goto out;
 124
 125         error = -EINVAL;
 126         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 127                 goto exit;
 128
 129         error = -EACCES;
 130         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 131                 goto exit;
 132
 133         fsnotify_open(file->f_path.dentry);
 134
 135         error = -ENOEXEC;
 136         if(file->f_op) {
 137                 struct linux_binfmt * fmt;
 138
 139                 read_lock(&binfmt_lock);
 140                 list_for_each_entry(fmt, &formats, lh) {
 141                         if (!fmt->load_shlib)
 142                                 continue;
 143                         if (!try_module_get(fmt->module))
 144                                 continue;
 145                         read_unlock(&binfmt_lock);
 146                         error = fmt->load_shlib(file);
 147                         read_lock(&binfmt_lock);
 148                         put_binfmt(fmt);
 149                         if (error != -ENOEXEC)
 150                                 break;
 151                 }
 152                 read_unlock(&binfmt_lock);
 153         }
 154 exit:
 155         fput(file);
 156 out:
 157         return error;
 158 }
 159
 160 #ifdef CONFIG_MMU
 161
 162 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 163                 int write)
 164 {
 165         struct page *page;
 166         int ret;
 167
 168 #ifdef CONFIG_STACK_GROWSUP
 169         if (write) {
 170                 ret = expand_stack_downwards(bprm->vma, pos);
 171                 if (ret < 0)
 172                         return NULL;
 173         }
 174 #endif
 175         ret = get_user_pages(current, bprm->mm, pos,
 176                         1, write, 1, &page, NULL);
 177         if (ret <= 0)
 178                 return NULL;
 179
 180         if (write) {
 181                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 182                 struct rlimit *rlim;
 183
 184                 /*
 185                  * We've historically supported up to 32 pages (ARG_MAX)
 186                  * of argument strings even with small stacks
 187                  */
 188                 if (size <= ARG_MAX)
 189                         return page;
 190
 191                 /*
 192                  * Limit to 1/4-th the stack size for the argv+env strings.
 193                  * This ensures that:
 194                  *  - the remaining binfmt code will not run out of stack space,
 195                  *  - the program will have a reasonable amount of stack left
 196                  *    to work from.
 197                  */
 198                 rlim = current->signal->rlim;
 199                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 200                         put_page(page);
 201                         return NULL;
 202                 }
 203         }
 204
 205         return page;
 206 }
 207
 208 static void put_arg_page(struct page *page)
 209 {
 210         put_page(page);
 211 }
 212
 213 static void free_arg_page(struct linux_binprm *bprm, int i)
 214 {
 215 }
 216
 217 static void free_arg_pages(struct linux_binprm *bprm)
 218 {
 219 }
 220
 221 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 222                 struct page *page)
 223 {
 224         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 225 }
 226
 227 static int __bprm_mm_init(struct linux_binprm *bprm)
 228 {
 229         int err;
 230         struct vm_area_struct *vma = NULL;
 231         struct mm_struct *mm = bprm->mm;
 232
 233         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 234         if (!vma)
 235                 return -ENOMEM;
 236
 237         down_write(&mm->mmap_sem);
 238         vma->vm_mm = mm;
 239
 240         /*
 241          * Place the stack at the largest stack address the architecture
 242          * supports. Later, we'll move this to an appropriate place. We don't
 243          * use STACK_TOP because that can depend on attributes which aren't
 244          * configured yet.
 245          */
 246         vma->vm_end = STACK_TOP_MAX;
 247         vma->vm_start = vma->vm_end - PAGE_SIZE;
 248         vma->vm_flags = VM_STACK_FLAGS;
 249         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 250
 251         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 252         if (err)
 253                 goto err;
 254
 255         err = insert_vm_struct(mm, vma);
 256         if (err)
 257                 goto err;
 258
 259         mm->stack_vm = mm->total_vm = 1;
 260         up_write(&mm->mmap_sem);
 261         bprm->p = vma->vm_end - sizeof(void *);
 262         return 0;
 263 err:
 264         up_write(&mm->mmap_sem);
 265         bprm->vma = NULL;
 266         kmem_cache_free(vm_area_cachep, vma);
 267         return err;
 268 }
 269
 270 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 271 {
 272         return len <= MAX_ARG_STRLEN;
 273 }
 274
 275 #else
 276
 277 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 278                 int write)
 279 {
 280         struct page *page;
 281
 282         page = bprm->page[pos / PAGE_SIZE];
 283         if (!page && write) {
 284                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 285                 if (!page)
 286                         return NULL;
 287                 bprm->page[pos / PAGE_SIZE] = page;
 288         }
 289
 290         return page;
 291 }
 292
 293 static void put_arg_page(struct page *page)
 294 {
 295 }
 296
 297 static void free_arg_page(struct linux_binprm *bprm, int i)
 298 {
 299         if (bprm->page[i]) {
 300                 __free_page(bprm->page[i]);
 301                 bprm->page[i] = NULL;
 302         }
 303 }
 304
 305 static void free_arg_pages(struct linux_binprm *bprm)
 306 {
 307         int i;
 308
 309         for (i = 0; i < MAX_ARG_PAGES; i++)
 310                 free_arg_page(bprm, i);
 311 }
 312
 313 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 314                 struct page *page)
 315 {
 316 }
 317
 318 static int __bprm_mm_init(struct linux_binprm *bprm)
 319 {
 320         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 321         return 0;
 322 }
 323
 324 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 325 {
 326         return len <= bprm->p;
 327 }
 328
 329 #endif /* CONFIG_MMU */
 330
 331 /*
 332  * Create a new mm_struct and populate it with a temporary stack
 333  * vm_area_struct.  We don't have enough context at this point to set the stack
 334  * flags, permissions, and offset, so we use temporary values.  We'll update
 335  * them later in setup_arg_pages().
 336  */
 337 int bprm_mm_init(struct linux_binprm *bprm)
 338 {
 339         int err;
 340         struct mm_struct *mm = NULL;
 341
 342         bprm->mm = mm = mm_alloc();
 343         err = -ENOMEM;
 344         if (!mm)
 345                 goto err;
 346
 347         err = init_new_context(current, mm);
 348         if (err)
 349                 goto err;
 350
 351         err = __bprm_mm_init(bprm);
 352         if (err)
 353                 goto err;
 354
 355         return 0;
 356
 357 err:
 358         if (mm) {
 359                 bprm->mm = NULL;
 360                 mmdrop(mm);
 361         }
 362
 363         return err;
 364 }
 365
 366 /*
 367  * count() counts the number of strings in array ARGV.
 368  */
 369 static int count(char __user * __user * argv, int max)
 370 {
 371         int i = 0;
 372
 373         if (argv != NULL) {
 374                 for (;;) {
 375                         char __user * p;
 376
 377                         if (get_user(p, argv))
 378                                 return -EFAULT;
 379                         if (!p)
 380                                 break;
 381                         argv++;
 382                         if (i++ >= max)
 383                                 return -E2BIG;
 384
 385                         if (fatal_signal_pending(current))
 386                                 return -ERESTARTNOHAND;
 387                         cond_resched();
 388                 }
 389         }
 390         return i;
 391 }
 392
 393 /*
 394  * 'copy_strings()' copies argument/environment strings from the old
 395  * processes's memory to the new process's stack.  The call to get_user_pages()
 396  * ensures the destination page is created and not swapped out.
 397  */
 398 static int copy_strings(int argc, char __user * __user * argv,
 399                         struct linux_binprm *bprm)
 400 {
 401         struct page *kmapped_page = NULL;
 402         char *kaddr = NULL;
 403         unsigned long kpos = 0;
 404         int ret;
 405
 406         while (argc-- > 0) {
 407                 char __user *str;
 408                 int len;
 409                 unsigned long pos;
 410
 411                 if (get_user(str, argv+argc) ||
 412                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 413                         ret = -EFAULT;
 414                         goto out;
 415                 }
 416
 417                 if (!valid_arg_len(bprm, len)) {
 418                         ret = -E2BIG;
 419                         goto out;
 420                 }
 421
 422                 /* We're going to work our way backwords. */
 423                 pos = bprm->p;
 424                 str += len;
 425                 bprm->p -= len;
 426
 427                 while (len > 0) {
 428                         int offset, bytes_to_copy;
 429
 430                         if (fatal_signal_pending(current)) {
 431                                 ret = -ERESTARTNOHAND;
 432                                 goto out;
 433                         }
 434                         cond_resched();
 435
 436                         offset = pos % PAGE_SIZE;
 437                         if (offset == 0)
 438                                 offset = PAGE_SIZE;
 439
 440                         bytes_to_copy = offset;
 441                         if (bytes_to_copy > len)
 442                                 bytes_to_copy = len;
 443
 444                         offset -= bytes_to_copy;
 445                         pos -= bytes_to_copy;
 446                         str -= bytes_to_copy;
 447                         len -= bytes_to_copy;
 448
 449                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 450                                 struct page *page;
 451
 452                                 page = get_arg_page(bprm, pos, 1);
 453                                 if (!page) {
 454                                         ret = -E2BIG;
 455                                         goto out;
 456                                 }
 457
 458                                 if (kmapped_page) {
 459                                         flush_kernel_dcache_page(kmapped_page);
 460                                         kunmap(kmapped_page);
 461                                         put_arg_page(kmapped_page);
 462                                 }
 463                                 kmapped_page = page;
 464                                 kaddr = kmap(kmapped_page);
 465                                 kpos = pos & PAGE_MASK;
 466                                 flush_arg_page(bprm, kpos, kmapped_page);
 467                         }
 468                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 469                                 ret = -EFAULT;
 470                                 goto out;
 471                         }
 472                 }
 473         }
 474         ret = 0;
 475 out:
 476         if (kmapped_page) {
 477                 flush_kernel_dcache_page(kmapped_page);
 478                 kunmap(kmapped_page);
 479                 put_arg_page(kmapped_page);
 480         }
 481         return ret;
 482 }
 483
 484 /*
 485  * Like copy_strings, but get argv and its values from kernel memory.
 486  */
 487 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 488 {
 489         int r;
 490         mm_segment_t oldfs = get_fs();
 491         set_fs(KERNEL_DS);
 492         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 493         set_fs(oldfs);
 494         return r;
 495 }
 496 EXPORT_SYMBOL(copy_strings_kernel);
 497
 498 #ifdef CONFIG_MMU
 499
 500 /*
 501  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 502  * the binfmt code determines where the new stack should reside, we shift it to
 503  * its final location.  The process proceeds as follows:
 504  *
 505  * 1) Use shift to calculate the new vma endpoints.
 506  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 507  *    arguments passed to subsequent functions are consistent.
 508  * 3) Move vma's page tables to the new range.
 509  * 4) Free up any cleared pgd range.
 510  * 5) Shrink the vma to cover only the new range.
 511  */
 512 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 513 {
 514         struct mm_struct *mm = vma->vm_mm;
 515         unsigned long old_start = vma->vm_start;
 516         unsigned long old_end = vma->vm_end;
 517         unsigned long length = old_end - old_start;
 518         unsigned long new_start = old_start - shift;
 519         unsigned long new_end = old_end - shift;
 520         struct mmu_gather *tlb;
 521
 522         BUG_ON(new_start > new_end);
 523
 524         /*
 525          * ensure there are no vmas between where we want to go
 526          * and where we are
 527          */
 528         if (vma != find_vma(mm, new_start))
 529                 return -EFAULT;
 530
 531         /*
 532          * cover the whole range: [new_start, old_end)
 533          */
 534         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 535
 536         /*
 537          * move the page tables downwards, on failure we rely on
 538          * process cleanup to remove whatever mess we made.
 539          */
 540         if (length != move_page_tables(vma, old_start,
 541                                        vma, new_start, length))
 542                 return -ENOMEM;
 543
 544         lru_add_drain();
 545         tlb = tlb_gather_mmu(mm, 0);
 546         if (new_end > old_start) {
 547                 /*
 548                  * when the old and new regions overlap clear from new_end.
 549                  */
 550                 free_pgd_range(tlb, new_end, old_end, new_end,
 551                         vma->vm_next ? vma->vm_next->vm_start : 0);
 552         } else {
 553                 /*
 554                  * otherwise, clean from old_start; this is done to not touch
 555                  * the address space in [new_end, old_start) some architectures
 556                  * have constraints on va-space that make this illegal (IA64) -
 557                  * for the others its just a little faster.
 558                  */
 559                 free_pgd_range(tlb, old_start, old_end, new_end,
 560                         vma->vm_next ? vma->vm_next->vm_start : 0);
 561         }
 562         tlb_finish_mmu(tlb, new_end, old_end);
 563
 564         /*
 565          * shrink the vma to just the new range.
 566          */
 567         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 568
 569         return 0;
 570 }
 571
 572 #define EXTRA_STACK_VM_PAGES    20      /* random */
 573
 574 /*
 575  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 576  * the stack is optionally relocated, and some extra space is added.
 577  */
 578 int setup_arg_pages(struct linux_binprm *bprm,
 579                     unsigned long stack_top,
 580                     int executable_stack)
 581 {
 582         unsigned long ret;
 583         unsigned long stack_shift;
 584         struct mm_struct *mm = current->mm;
 585         struct vm_area_struct *vma = bprm->vma;
 586         struct vm_area_struct *prev = NULL;
 587         unsigned long vm_flags;
 588         unsigned long stack_base;
 589         unsigned long stack_size;
 590         unsigned long stack_expand;
 591         unsigned long rlim_stack;
 592
 593 #ifdef CONFIG_STACK_GROWSUP
 594         /* Limit stack size to 1GB */
 595         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 596         if (stack_base > (1 << 30))
 597                 stack_base = 1 << 30;
 598
 599         /* Make sure we didn't let the argument array grow too large. */
 600         if (vma->vm_end - vma->vm_start > stack_base)
 601                 return -ENOMEM;
 602
 603         stack_base = PAGE_ALIGN(stack_top - stack_base);
 604
 605         stack_shift = vma->vm_start - stack_base;
 606         mm->arg_start = bprm->p - stack_shift;
 607         bprm->p = vma->vm_end - stack_shift;
 608 #else
 609         stack_top = arch_align_stack(stack_top);
 610         stack_top = PAGE_ALIGN(stack_top);
 611
 612         if (unlikely(stack_top < mmap_min_addr) ||
 613             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 614                 return -ENOMEM;
 615
 616         stack_shift = vma->vm_end - stack_top;
 617
 618         bprm->p -= stack_shift;
 619         mm->arg_start = bprm->p;
 620 #endif
 621
 622         if (bprm->loader)
 623                 bprm->loader -= stack_shift;
 624         bprm->exec -= stack_shift;
 625
 626         down_write(&mm->mmap_sem);
 627         vm_flags = VM_STACK_FLAGS;
 628
 629         /*
 630          * Adjust stack execute permissions; explicitly enable for
 631          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 632          * (arch default) otherwise.
 633          */
 634         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 635                 vm_flags |= VM_EXEC;
 636         else if (executable_stack == EXSTACK_DISABLE_X)
 637                 vm_flags &= ~VM_EXEC;
 638         vm_flags |= mm->def_flags;
 639
 640         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 641                         vm_flags);
 642         if (ret)
 643                 goto out_unlock;
 644         BUG_ON(prev != vma);
 645
 646         /* Move stack pages down in memory. */
 647         if (stack_shift) {
 648                 ret = shift_arg_pages(vma, stack_shift);
 649                 if (ret)
 650                         goto out_unlock;
 651         }
 652
 653         stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 654         stack_size = vma->vm_end - vma->vm_start;
 655         /*
 656          * Align this down to a page boundary as expand_stack
 657          * will align it up.
 658          */
 659         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 660 #ifdef CONFIG_STACK_GROWSUP
 661         if (stack_size + stack_expand > rlim_stack)
 662                 stack_base = vma->vm_start + rlim_stack;
 663         else
 664                 stack_base = vma->vm_end + stack_expand;
 665 #else
 666         if (stack_size + stack_expand > rlim_stack)
 667                 stack_base = vma->vm_end - rlim_stack;
 668         else
 669                 stack_base = vma->vm_start - stack_expand;
 670 #endif
 671         ret = expand_stack(vma, stack_base);
 672         if (ret)
 673                 ret = -EFAULT;
 674
 675 out_unlock:
 676         up_write(&mm->mmap_sem);
 677         return ret;
 678 }
 679 EXPORT_SYMBOL(setup_arg_pages);
 680
 681 #endif /* CONFIG_MMU */
 682
 683 struct file *open_exec(const char *name)
 684 {
 685         struct file *file;
 686         int err;
 687
 688         file = do_filp_open(AT_FDCWD, name,
 689                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 690                                 MAY_EXEC | MAY_OPEN);
 691         if (IS_ERR(file))
 692                 goto out;
 693
 694         err = -EACCES;
 695         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 696                 goto exit;
 697
 698         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 699                 goto exit;
 700
 701         fsnotify_open(file->f_path.dentry);
 702
 703         err = deny_write_access(file);
 704         if (err)
 705                 goto exit;
 706
 707 out:
 708         return file;
 709
 710 exit:
 711         fput(file);
 712         return ERR_PTR(err);
 713 }
 714 EXPORT_SYMBOL(open_exec);
 715
 716 int kernel_read(struct file *file, loff_t offset,
 717                 char *addr, unsigned long count)
 718 {
 719         mm_segment_t old_fs;
 720         loff_t pos = offset;
 721         int result;
 722
 723         old_fs = get_fs();
 724         set_fs(get_ds());
 725         /* The cast to a user pointer is valid due to the set_fs() */
 726         result = vfs_read(file, (void __user *)addr, count, &pos);
 727         set_fs(old_fs);
 728         return result;
 729 }
 730
 731 EXPORT_SYMBOL(kernel_read);
 732
 733 static int exec_mmap(struct mm_struct *mm)
 734 {
 735         struct task_struct *tsk;
 736         struct mm_struct * old_mm, *active_mm;
 737
 738         /* Notify parent that we're no longer interested in the old VM */
 739         tsk = current;
 740         old_mm = current->mm;
 741         mm_release(tsk, old_mm);
 742
 743         if (old_mm) {
 744                 /*
 745                  * Make sure that if there is a core dump in progress
 746                  * for the old mm, we get out and die instead of going
 747                  * through with the exec.  We must hold mmap_sem around
 748                  * checking core_state and changing tsk->mm.
 749                  */
 750                 down_read(&old_mm->mmap_sem);
 751                 if (unlikely(old_mm->core_state)) {
 752                         up_read(&old_mm->mmap_sem);
 753                         return -EINTR;
 754                 }
 755         }
 756         task_lock(tsk);
 757         active_mm = tsk->active_mm;
 758         tsk->mm = mm;
 759         tsk->active_mm = mm;
 760         activate_mm(active_mm, mm);
 761         task_unlock(tsk);
 762         arch_pick_mmap_layout(mm);
 763         if (old_mm) {
 764                 up_read(&old_mm->mmap_sem);
 765                 BUG_ON(active_mm != old_mm);
 766                 mm_update_next_owner(old_mm);
 767                 mmput(old_mm);
 768                 return 0;
 769         }
 770         mmdrop(active_mm);
 771         return 0;
 772 }
 773
 774 /*
 775  * This function makes sure the current process has its own signal table,
 776  * so that flush_signal_handlers can later reset the handlers without
 777  * disturbing other processes.  (Other processes might share the signal
 778  * table via the CLONE_SIGHAND option to clone().)
 779  */
 780 static int de_thread(struct task_struct *tsk)
 781 {
 782         struct signal_struct *sig = tsk->signal;
 783         struct sighand_struct *oldsighand = tsk->sighand;
 784         spinlock_t *lock = &oldsighand->siglock;
 785         int count;
 786
 787         if (thread_group_empty(tsk))
 788                 goto no_thread_group;
 789
 790         /*
 791          * Kill all other threads in the thread group.
 792          */
 793         spin_lock_irq(lock);
 794         if (signal_group_exit(sig)) {
 795                 /*
 796                  * Another group action in progress, just
 797                  * return so that the signal is processed.
 798                  */
 799                 spin_unlock_irq(lock);
 800                 return -EAGAIN;
 801         }
 802         sig->group_exit_task = tsk;
 803         zap_other_threads(tsk);
 804
 805         /* Account for the thread group leader hanging around: */
 806         count = thread_group_leader(tsk) ? 1 : 2;
 807         sig->notify_count = count;
 808         while (atomic_read(&sig->count) > count) {
 809                 __set_current_state(TASK_UNINTERRUPTIBLE);
 810                 spin_unlock_irq(lock);
 811                 schedule();
 812                 spin_lock_irq(lock);
 813         }
 814         spin_unlock_irq(lock);
 815
 816         /*
 817          * At this point all other threads have exited, all we have to
 818          * do is to wait for the thread group leader to become inactive,
 819          * and to assume its PID:
 820          */
 821         if (!thread_group_leader(tsk)) {
 822                 struct task_struct *leader = tsk->group_leader;
 823
 824                 sig->notify_count = -1; /* for exit_notify() */
 825                 for (;;) {
 826                         write_lock_irq(&tasklist_lock);
 827                         if (likely(leader->exit_state))
 828                                 break;
 829                         __set_current_state(TASK_UNINTERRUPTIBLE);
 830                         write_unlock_irq(&tasklist_lock);
 831                         schedule();
 832                 }
 833
 834                 /*
 835                  * The only record we have of the real-time age of a
 836                  * process, regardless of execs it's done, is start_time.
 837                  * All the past CPU time is accumulated in signal_struct
 838                  * from sister threads now dead.  But in this non-leader
 839                  * exec, nothing survives from the original leader thread,
 840                  * whose birth marks the true age of this process now.
 841                  * When we take on its identity by switching to its PID, we
 842                  * also take its birthdate (always earlier than our own).
 843                  */
 844                 tsk->start_time = leader->start_time;
 845
 846                 BUG_ON(!same_thread_group(leader, tsk));
 847                 BUG_ON(has_group_leader_pid(tsk));
 848                 /*
 849                  * An exec() starts a new thread group with the
 850                  * TGID of the previous thread group. Rehash the
 851                  * two threads with a switched PID, and release
 852                  * the former thread group leader:
 853                  */
 854
 855                 /* Become a process group leader with the old leader's pid.
 856                  * The old leader becomes a thread of the this thread group.
 857                  * Note: The old leader also uses this pid until release_task
 858                  *       is called.  Odd but simple and correct.
 859                  */
 860                 detach_pid(tsk, PIDTYPE_PID);
 861                 tsk->pid = leader->pid;
 862                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 863                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 864                 transfer_pid(leader, tsk, PIDTYPE_SID);
 865                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 866
 867                 tsk->group_leader = tsk;
 868                 leader->group_leader = tsk;
 869
 870                 tsk->exit_signal = SIGCHLD;
 871
 872                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 873                 leader->exit_state = EXIT_DEAD;
 874                 write_unlock_irq(&tasklist_lock);
 875
 876                 release_task(leader);
 877         }
 878
 879         sig->group_exit_task = NULL;
 880         sig->notify_count = 0;
 881
 882 no_thread_group:
 883         if (current->mm)
 884                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 885
 886         exit_itimers(sig);
 887         flush_itimer_signals();
 888
 889         if (atomic_read(&oldsighand->count) != 1) {
 890                 struct sighand_struct *newsighand;
 891                 /*
 892                  * This ->sighand is shared with the CLONE_SIGHAND
 893                  * but not CLONE_THREAD task, switch to the new one.
 894                  */
 895                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 896                 if (!newsighand)
 897                         return -ENOMEM;
 898
 899                 atomic_set(&newsighand->count, 1);
 900                 memcpy(newsighand->action, oldsighand->action,
 901                        sizeof(newsighand->action));
 902
 903                 write_lock_irq(&tasklist_lock);
 904                 spin_lock(&oldsighand->siglock);
 905                 rcu_assign_pointer(tsk->sighand, newsighand);
 906                 spin_unlock(&oldsighand->siglock);
 907                 write_unlock_irq(&tasklist_lock);
 908
 909                 __cleanup_sighand(oldsighand);
 910         }
 911
 912         BUG_ON(!thread_group_leader(tsk));
 913         return 0;
 914 }
 915
 916 /*
 917  * These functions flushes out all traces of the currently running executable
 918  * so that a new one can be started
 919  */
 920 static void flush_old_files(struct files_struct * files)
 921 {
 922         long j = -1;
 923         struct fdtable *fdt;
 924
 925         spin_lock(&files->file_lock);
 926         for (;;) {
 927                 unsigned long set, i;
 928
 929                 j++;
 930                 i = j * __NFDBITS;
 931                 fdt = files_fdtable(files);
 932                 if (i >= fdt->max_fds)
 933                         break;
 934                 set = fdt->close_on_exec->fds_bits[j];
 935                 if (!set)
 936                         continue;
 937                 fdt->close_on_exec->fds_bits[j] = 0;
 938                 spin_unlock(&files->file_lock);
 939                 for ( ; set ; i++,set >>= 1) {
 940                         if (set & 1) {
 941                                 sys_close(i);
 942                         }
 943                 }
 944                 spin_lock(&files->file_lock);
 945
 946         }
 947         spin_unlock(&files->file_lock);
 948 }
 949
 950 char *get_task_comm(char *buf, struct task_struct *tsk)
 951 {
 952         /* buf must be at least sizeof(tsk->comm) in size */
 953         task_lock(tsk);
 954         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 955         task_unlock(tsk);
 956         return buf;
 957 }
 958
 959 void set_task_comm(struct task_struct *tsk, char *buf)
 960 {
 961         task_lock(tsk);
 962         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 963         task_unlock(tsk);
 964         perf_event_comm(tsk);
 965 }
 966
 967 int flush_old_exec(struct linux_binprm * bprm)
 968 {
 969         int retval;
 970
 971         /*
 972          * Make sure we have a private signal table and that
 973          * we are unassociated from the previous thread group.
 974          */
 975         retval = de_thread(current);
 976         if (retval)
 977                 goto out;
 978
 979         set_mm_exe_file(bprm->mm, bprm->file);
 980
 981         /*
 982          * Release all of the old mmap stuff
 983          */
 984         retval = exec_mmap(bprm->mm);
 985         if (retval)
 986                 goto out;
 987
 988         bprm->mm = NULL;                /* We're using it now */
 989
 990         current->flags &= ~PF_RANDOMIZE;
 991         flush_thread();
 992         current->personality &= ~bprm->per_clear;
 993
 994         return 0;
 995
 996 out:
 997         return retval;
 998 }
 999 EXPORT_SYMBOL(flush_old_exec);
1000
1001 void setup_new_exec(struct linux_binprm * bprm)
1002 {
1003         int i, ch;
1004         char * name;
1005         char tcomm[sizeof(current->comm)];
1006
1007         arch_pick_mmap_layout(current->mm);
1008
1009         /* This is the point of no return */
1010         current->sas_ss_sp = current->sas_ss_size = 0;
1011
1012         if (current_euid() == current_uid() && current_egid() == current_gid())
1013                 set_dumpable(current->mm, 1);
1014         else
1015                 set_dumpable(current->mm, suid_dumpable);
1016
1017         name = bprm->filename;
1018
1019         /* Copies the binary name from after last slash */
1020         for (i=0; (ch = *(name++)) != '\0';) {
1021                 if (ch == '/')
1022                         i = 0; /* overwrite what we wrote */
1023                 else
1024                         if (i < (sizeof(tcomm) - 1))
1025                                 tcomm[i++] = ch;
1026         }
1027         tcomm[i] = '\0';
1028         set_task_comm(current, tcomm);
1029
1030         /* Set the new mm task size. We have to do that late because it may
1031          * depend on TIF_32BIT which is only updated in flush_thread() on
1032          * some architectures like powerpc
1033          */
1034         current->mm->task_size = TASK_SIZE;
1035
1036         /* install the new credentials */
1037         if (bprm->cred->uid != current_euid() ||
1038             bprm->cred->gid != current_egid()) {
1039                 current->pdeath_signal = 0;
1040         } else if (file_permission(bprm->file, MAY_READ) ||
1041                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1042                 set_dumpable(current->mm, suid_dumpable);
1043         }
1044
1045         /*
1046          * Flush performance counters when crossing a
1047          * security domain:
1048          */
1049         if (!get_dumpable(current->mm))
1050                 perf_event_exit_task(current);
1051
1052         /* An exec changes our domain. We are no longer part of the thread
1053            group */
1054
1055         current->self_exec_id++;
1056
1057         flush_signal_handlers(current, 0);
1058         flush_old_files(current->files);
1059 }
1060 EXPORT_SYMBOL(setup_new_exec);
1061
1062 /*
1063  * Prepare credentials and lock ->cred_guard_mutex.
1064  * install_exec_creds() commits the new creds and drops the lock.
1065  * Or, if exec fails before, free_bprm() should release ->cred and
1066  * and unlock.
1067  */
1068 int prepare_bprm_creds(struct linux_binprm *bprm)
1069 {
1070         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1071                 return -ERESTARTNOINTR;
1072
1073         bprm->cred = prepare_exec_creds();
1074         if (likely(bprm->cred))
1075                 return 0;
1076
1077         mutex_unlock(&current->cred_guard_mutex);
1078         return -ENOMEM;
1079 }
1080
1081 void free_bprm(struct linux_binprm *bprm)
1082 {
1083         free_arg_pages(bprm);
1084         if (bprm->cred) {
1085                 mutex_unlock(&current->cred_guard_mutex);
1086                 abort_creds(bprm->cred);
1087         }
1088         kfree(bprm);
1089 }
1090
1091 /*
1092  * install the new credentials for this executable
1093  */
1094 void install_exec_creds(struct linux_binprm *bprm)
1095 {
1096         security_bprm_committing_creds(bprm);
1097
1098         commit_creds(bprm->cred);
1099         bprm->cred = NULL;
1100         /*
1101          * cred_guard_mutex must be held at least to this point to prevent
1102          * ptrace_attach() from altering our determination of the task's
1103          * credentials; any time after this it may be unlocked.
1104          */
1105         security_bprm_committed_creds(bprm);
1106         mutex_unlock(&current->cred_guard_mutex);
1107 }
1108 EXPORT_SYMBOL(install_exec_creds);
1109
1110 /*
1111  * determine how safe it is to execute the proposed program
1112  * - the caller must hold current->cred_guard_mutex to protect against
1113  *   PTRACE_ATTACH
1114  */
1115 int check_unsafe_exec(struct linux_binprm *bprm)
1116 {
1117         struct task_struct *p = current, *t;
1118         unsigned n_fs;
1119         int res = 0;
1120
1121         bprm->unsafe = tracehook_unsafe_exec(p);
1122
1123         n_fs = 1;
1124         write_lock(&p->fs->lock);
1125         rcu_read_lock();
1126         for (t = next_thread(p); t != p; t = next_thread(t)) {
1127                 if (t->fs == p->fs)
1128                         n_fs++;
1129         }
1130         rcu_read_unlock();
1131
1132         if (p->fs->users > n_fs) {
1133                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1134         } else {
1135                 res = -EAGAIN;
1136                 if (!p->fs->in_exec) {
1137                         p->fs->in_exec = 1;
1138                         res = 1;
1139                 }
1140         }
1141         write_unlock(&p->fs->lock);
1142
1143         return res;
1144 }
1145
1146 /*
1147  * Fill the binprm structure from the inode.
1148  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1149  *
1150  * This may be called multiple times for binary chains (scripts for example).
1151  */
1152 int prepare_binprm(struct linux_binprm *bprm)
1153 {
1154         umode_t mode;
1155         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1156         int retval;
1157
1158         mode = inode->i_mode;
1159         if (bprm->file->f_op == NULL)
1160                 return -EACCES;
1161
1162         /* clear any previous set[ug]id data from a previous binary */
1163         bprm->cred->euid = current_euid();
1164         bprm->cred->egid = current_egid();
1165
1166         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1167                 /* Set-uid? */
1168                 if (mode & S_ISUID) {
1169                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1170                         bprm->cred->euid = inode->i_uid;
1171                 }
1172
1173                 /* Set-gid? */
1174                 /*
1175                  * If setgid is set but no group execute bit then this
1176                  * is a candidate for mandatory locking, not a setgid
1177                  * executable.
1178                  */
1179                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1180                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1181                         bprm->cred->egid = inode->i_gid;
1182                 }
1183         }
1184
1185         /* fill in binprm security blob */
1186         retval = security_bprm_set_creds(bprm);
1187         if (retval)
1188                 return retval;
1189         bprm->cred_prepared = 1;
1190
1191         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1192         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1193 }
1194
1195 EXPORT_SYMBOL(prepare_binprm);
1196
1197 /*
1198  * Arguments are '\0' separated strings found at the location bprm->p
1199  * points to; chop off the first by relocating brpm->p to right after
1200  * the first '\0' encountered.
1201  */
1202 int remove_arg_zero(struct linux_binprm *bprm)
1203 {
1204         int ret = 0;
1205         unsigned long offset;
1206         char *kaddr;
1207         struct page *page;
1208
1209         if (!bprm->argc)
1210                 return 0;
1211
1212         do {
1213                 offset = bprm->p & ~PAGE_MASK;
1214                 page = get_arg_page(bprm, bprm->p, 0);
1215                 if (!page) {
1216                         ret = -EFAULT;
1217                         goto out;
1218                 }
1219                 kaddr = kmap_atomic(page, KM_USER0);
1220
1221                 for (; offset < PAGE_SIZE && kaddr[offset];
1222                                 offset++, bprm->p++)
1223                         ;
1224
1225                 kunmap_atomic(kaddr, KM_USER0);
1226                 put_arg_page(page);
1227
1228                 if (offset == PAGE_SIZE)
1229                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1230         } while (offset == PAGE_SIZE);
1231
1232         bprm->p++;
1233         bprm->argc--;
1234         ret = 0;
1235
1236 out:
1237         return ret;
1238 }
1239 EXPORT_SYMBOL(remove_arg_zero);
1240
1241 /*
1242  * cycle the list of binary formats handler, until one recognizes the image
1243  */
1244 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1245 {
1246         unsigned int depth = bprm->recursion_depth;
1247         int try,retval;
1248         struct linux_binfmt *fmt;
1249
1250         retval = security_bprm_check(bprm);
1251         if (retval)
1252                 return retval;
1253         retval = ima_bprm_check(bprm);
1254         if (retval)
1255                 return retval;
1256
1257         /* kernel module loader fixup */
1258         /* so we don't try to load run modprobe in kernel space. */
1259         set_fs(USER_DS);
1260
1261         retval = audit_bprm(bprm);
1262         if (retval)
1263                 return retval;
1264
1265         retval = -ENOENT;
1266         for (try=0; try<2; try++) {
1267                 read_lock(&binfmt_lock);
1268                 list_for_each_entry(fmt, &formats, lh) {
1269                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1270                         if (!fn)
1271                                 continue;
1272                         if (!try_module_get(fmt->module))
1273                                 continue;
1274                         read_unlock(&binfmt_lock);
1275                         retval = fn(bprm, regs);
1276                         /*
1277                          * Restore the depth counter to its starting value
1278                          * in this call, so we don't have to rely on every
1279                          * load_binary function to restore it on return.
1280                          */
1281                         bprm->recursion_depth = depth;
1282                         if (retval >= 0) {
1283                                 if (depth == 0)
1284                                         tracehook_report_exec(fmt, bprm, regs);
1285                                 put_binfmt(fmt);
1286                                 allow_write_access(bprm->file);
1287                                 if (bprm->file)
1288                                         fput(bprm->file);
1289                                 bprm->file = NULL;
1290                                 current->did_exec = 1;
1291                                 proc_exec_connector(current);
1292                                 return retval;
1293                         }
1294                         read_lock(&binfmt_lock);
1295                         put_binfmt(fmt);
1296                         if (retval != -ENOEXEC || bprm->mm == NULL)
1297                                 break;
1298                         if (!bprm->file) {
1299                                 read_unlock(&binfmt_lock);
1300                                 return retval;
1301                         }
1302                 }
1303                 read_unlock(&binfmt_lock);
1304                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1305                         break;
1306 #ifdef CONFIG_MODULES
1307                 } else {
1308 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1309                         if (printable(bprm->buf[0]) &&
1310                             printable(bprm->buf[1]) &&
1311                             printable(bprm->buf[2]) &&
1312                             printable(bprm->buf[3]))
1313                                 break; /* -ENOEXEC */
1314                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1315 #endif
1316                 }
1317         }
1318         return retval;
1319 }
1320
1321 EXPORT_SYMBOL(search_binary_handler);
1322
1323 /*
1324  * sys_execve() executes a new program.
1325  */
1326 int do_execve(char * filename,
1327         char __user *__user *argv,
1328         char __user *__user *envp,
1329         struct pt_regs * regs)
1330 {
1331         struct linux_binprm *bprm;
1332         struct file *file;
1333         struct files_struct *displaced;
1334         bool clear_in_exec;
1335         int retval;
1336
1337         retval = unshare_files(&displaced);
1338         if (retval)
1339                 goto out_ret;
1340
1341         retval = -ENOMEM;
1342         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1343         if (!bprm)
1344                 goto out_files;
1345
1346         retval = prepare_bprm_creds(bprm);
1347         if (retval)
1348                 goto out_free;
1349
1350         retval = check_unsafe_exec(bprm);
1351         if (retval < 0)
1352                 goto out_free;
1353         clear_in_exec = retval;
1354         current->in_execve = 1;
1355
1356         file = open_exec(filename);
1357         retval = PTR_ERR(file);
1358         if (IS_ERR(file))
1359                 goto out_unmark;
1360
1361         sched_exec();
1362
1363         bprm->file = file;
1364         bprm->filename = filename;
1365         bprm->interp = filename;
1366
1367         retval = bprm_mm_init(bprm);
1368         if (retval)
1369                 goto out_file;
1370
1371         bprm->argc = count(argv, MAX_ARG_STRINGS);
1372         if ((retval = bprm->argc) < 0)
1373                 goto out;
1374
1375         bprm->envc = count(envp, MAX_ARG_STRINGS);
1376         if ((retval = bprm->envc) < 0)
1377                 goto out;
1378
1379         retval = prepare_binprm(bprm);
1380         if (retval < 0)
1381                 goto out;
1382
1383         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1384         if (retval < 0)
1385                 goto out;
1386
1387         bprm->exec = bprm->p;
1388         retval = copy_strings(bprm->envc, envp, bprm);
1389         if (retval < 0)
1390                 goto out;
1391
1392         retval = copy_strings(bprm->argc, argv, bprm);
1393         if (retval < 0)
1394                 goto out;
1395
1396         current->flags &= ~PF_KTHREAD;
1397         retval = search_binary_handler(bprm,regs);
1398         if (retval < 0)
1399                 goto out;
1400
1401         /* execve succeeded */
1402         current->fs->in_exec = 0;
1403         current->in_execve = 0;
1404         acct_update_integrals(current);
1405         free_bprm(bprm);
1406         if (displaced)
1407                 put_files_struct(displaced);
1408         return retval;
1409
1410 out:
1411         if (bprm->mm)
1412                 mmput (bprm->mm);
1413
1414 out_file:
1415         if (bprm->file) {
1416                 allow_write_access(bprm->file);
1417                 fput(bprm->file);
1418         }
1419
1420 out_unmark:
1421         if (clear_in_exec)
1422                 current->fs->in_exec = 0;
1423         current->in_execve = 0;
1424
1425 out_free:
1426         free_bprm(bprm);
1427
1428 out_files:
1429         if (displaced)
1430                 reset_files_struct(displaced);
1431 out_ret:
1432         return retval;
1433 }
1434
1435 void set_binfmt(struct linux_binfmt *new)
1436 {
1437         struct mm_struct *mm = current->mm;
1438
1439         if (mm->binfmt)
1440                 module_put(mm->binfmt->module);
1441
1442         mm->binfmt = new;
1443         if (new)
1444                 __module_get(new->module);
1445 }
1446
1447 EXPORT_SYMBOL(set_binfmt);
1448
1449 /* format_corename will inspect the pattern parameter, and output a
1450  * name into corename, which must have space for at least
1451  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1452  */
1453 static int format_corename(char *corename, long signr)
1454 {
1455         const struct cred *cred = current_cred();
1456         const char *pat_ptr = core_pattern;
1457         int ispipe = (*pat_ptr == '|');
1458         char *out_ptr = corename;
1459         char *const out_end = corename + CORENAME_MAX_SIZE;
1460         int rc;
1461         int pid_in_pattern = 0;
1462
1463         /* Repeat as long as we have more pattern to process and more output
1464            space */
1465         while (*pat_ptr) {
1466                 if (*pat_ptr != '%') {
1467                         if (out_ptr == out_end)
1468                                 goto out;
1469                         *out_ptr++ = *pat_ptr++;
1470                 } else {
1471                         switch (*++pat_ptr) {
1472                         case 0:
1473                                 goto out;
1474                         /* Double percent, output one percent */
1475                         case '%':
1476                                 if (out_ptr == out_end)
1477                                         goto out;
1478                                 *out_ptr++ = '%';
1479                                 break;
1480                         /* pid */
1481                         case 'p':
1482                                 pid_in_pattern = 1;
1483                                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                                               "%d", task_tgid_vnr(current));
1485                                 if (rc > out_end - out_ptr)
1486                                         goto out;
1487                                 out_ptr += rc;
1488                                 break;
1489                         /* uid */
1490                         case 'u':
1491                                 rc = snprintf(out_ptr, out_end - out_ptr,
1492                                               "%d", cred->uid);
1493                                 if (rc > out_end - out_ptr)
1494                                         goto out;
1495                                 out_ptr += rc;
1496                                 break;
1497                         /* gid */
1498                         case 'g':
1499                                 rc = snprintf(out_ptr, out_end - out_ptr,
1500                                               "%d", cred->gid);
1501                                 if (rc > out_end - out_ptr)
1502                                         goto out;
1503                                 out_ptr += rc;
1504                                 break;
1505                         /* signal that caused the coredump */
1506                         case 's':
1507                                 rc = snprintf(out_ptr, out_end - out_ptr,
1508                                               "%ld", signr);
1509                                 if (rc > out_end - out_ptr)
1510                                         goto out;
1511                                 out_ptr += rc;
1512                                 break;
1513                         /* UNIX time of coredump */
1514                         case 't': {
1515                                 struct timeval tv;
1516                                 do_gettimeofday(&tv);
1517                                 rc = snprintf(out_ptr, out_end - out_ptr,
1518                                               "%lu", tv.tv_sec);
1519                                 if (rc > out_end - out_ptr)
1520                                         goto out;
1521                                 out_ptr += rc;
1522                                 break;
1523                         }
1524                         /* hostname */
1525                         case 'h':
1526                                 down_read(&uts_sem);
1527                                 rc = snprintf(out_ptr, out_end - out_ptr,
1528                                               "%s", utsname()->nodename);
1529                                 up_read(&uts_sem);
1530                                 if (rc > out_end - out_ptr)
1531                                         goto out;
1532                                 out_ptr += rc;
1533                                 break;
1534                         /* executable */
1535                         case 'e':
1536                                 rc = snprintf(out_ptr, out_end - out_ptr,
1537                                               "%s", current->comm);
1538                                 if (rc > out_end - out_ptr)
1539                                         goto out;
1540                                 out_ptr += rc;
1541                                 break;
1542                         /* core limit size */
1543                         case 'c':
1544                                 rc = snprintf(out_ptr, out_end - out_ptr,
1545                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1546                                 if (rc > out_end - out_ptr)
1547                                         goto out;
1548                                 out_ptr += rc;
1549                                 break;
1550                         default:
1551                                 break;
1552                         }
1553                         ++pat_ptr;
1554                 }
1555         }
1556         /* Backward compatibility with core_uses_pid:
1557          *
1558          * If core_pattern does not include a %p (as is the default)
1559          * and core_uses_pid is set, then .%pid will be appended to
1560          * the filename. Do not do this for piped commands. */
1561         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1562                 rc = snprintf(out_ptr, out_end - out_ptr,
1563                               ".%d", task_tgid_vnr(current));
1564                 if (rc > out_end - out_ptr)
1565                         goto out;
1566                 out_ptr += rc;
1567         }
1568 out:
1569         *out_ptr = 0;
1570         return ispipe;
1571 }
1572
1573 static int zap_process(struct task_struct *start)
1574 {
1575         struct task_struct *t;
1576         int nr = 0;
1577
1578         start->signal->flags = SIGNAL_GROUP_EXIT;
1579         start->signal->group_stop_count = 0;
1580
1581         t = start;
1582         do {
1583                 if (t != current && t->mm) {
1584                         sigaddset(&t->pending.signal, SIGKILL);
1585                         signal_wake_up(t, 1);
1586                         nr++;
1587                 }
1588         } while_each_thread(start, t);
1589
1590         return nr;
1591 }
1592
1593 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1594                                 struct core_state *core_state, int exit_code)
1595 {
1596         struct task_struct *g, *p;
1597         unsigned long flags;
1598         int nr = -EAGAIN;
1599
1600         spin_lock_irq(&tsk->sighand->siglock);
1601         if (!signal_group_exit(tsk->signal)) {
1602                 mm->core_state = core_state;
1603                 tsk->signal->group_exit_code = exit_code;
1604                 nr = zap_process(tsk);
1605         }
1606         spin_unlock_irq(&tsk->sighand->siglock);
1607         if (unlikely(nr < 0))
1608                 return nr;
1609
1610         if (atomic_read(&mm->mm_users) == nr + 1)
1611                 goto done;
1612         /*
1613          * We should find and kill all tasks which use this mm, and we should
1614          * count them correctly into ->nr_threads. We don't take tasklist
1615          * lock, but this is safe wrt:
1616          *
1617          * fork:
1618          *      None of sub-threads can fork after zap_process(leader). All
1619          *      processes which were created before this point should be
1620          *      visible to zap_threads() because copy_process() adds the new
1621          *      process to the tail of init_task.tasks list, and lock/unlock
1622          *      of ->siglock provides a memory barrier.
1623          *
1624          * do_exit:
1625          *      The caller holds mm->mmap_sem. This means that the task which
1626          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1627          *      its ->mm.
1628          *
1629          * de_thread:
1630          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1631          *      we must see either old or new leader, this does not matter.
1632          *      However, it can change p->sighand, so lock_task_sighand(p)
1633          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1634          *      it can't fail.
1635          *
1636          *      Note also that "g" can be the old leader with ->mm == NULL
1637          *      and already unhashed and thus removed from ->thread_group.
1638          *      This is OK, __unhash_process()->list_del_rcu() does not
1639          *      clear the ->next pointer, we will find the new leader via
1640          *      next_thread().
1641          */
1642         rcu_read_lock();
1643         for_each_process(g) {
1644                 if (g == tsk->group_leader)
1645                         continue;
1646                 if (g->flags & PF_KTHREAD)
1647                         continue;
1648                 p = g;
1649                 do {
1650                         if (p->mm) {
1651                                 if (unlikely(p->mm == mm)) {
1652                                         lock_task_sighand(p, &flags);
1653                                         nr += zap_process(p);
1654                                         unlock_task_sighand(p, &flags);
1655                                 }
1656                                 break;
1657                         }
1658                 } while_each_thread(g, p);
1659         }
1660         rcu_read_unlock();
1661 done:
1662         atomic_set(&core_state->nr_threads, nr);
1663         return nr;
1664 }
1665
1666 static int coredump_wait(int exit_code, struct core_state *core_state)
1667 {
1668         struct task_struct *tsk = current;
1669         struct mm_struct *mm = tsk->mm;
1670         struct completion *vfork_done;
1671         int core_waiters;
1672
1673         init_completion(&core_state->startup);
1674         core_state->dumper.task = tsk;
1675         core_state->dumper.next = NULL;
1676         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1677         up_write(&mm->mmap_sem);
1678
1679         if (unlikely(core_waiters < 0))
1680                 goto fail;
1681
1682         /*
1683          * Make sure nobody is waiting for us to release the VM,
1684          * otherwise we can deadlock when we wait on each other
1685          */
1686         vfork_done = tsk->vfork_done;
1687         if (vfork_done) {
1688                 tsk->vfork_done = NULL;
1689                 complete(vfork_done);
1690         }
1691
1692         if (core_waiters)
1693                 wait_for_completion(&core_state->startup);
1694 fail:
1695         return core_waiters;
1696 }
1697
1698 static void coredump_finish(struct mm_struct *mm)
1699 {
1700         struct core_thread *curr, *next;
1701         struct task_struct *task;
1702
1703         next = mm->core_state->dumper.next;
1704         while ((curr = next) != NULL) {
1705                 next = curr->next;
1706                 task = curr->task;
1707                 /*
1708                  * see exit_mm(), curr->task must not see
1709                  * ->task == NULL before we read ->next.
1710                  */
1711                 smp_mb();
1712                 curr->task = NULL;
1713                 wake_up_process(task);
1714         }
1715
1716         mm->core_state = NULL;
1717 }
1718
1719 /*
1720  * set_dumpable converts traditional three-value dumpable to two flags and
1721  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1722  * these bits are not changed atomically.  So get_dumpable can observe the
1723  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1724  * return either old dumpable or new one by paying attention to the order of
1725  * modifying the bits.
1726  *
1727  * dumpable |   mm->flags (binary)
1728  * old  new | initial interim  final
1729  * ---------+-----------------------
1730  *  0    1  |   00      01      01
1731  *  0    2  |   00      10(*)   11
1732  *  1    0  |   01      00      00
1733  *  1    2  |   01      11      11
1734  *  2    0  |   11      10(*)   00
1735  *  2    1  |   11      11      01
1736  *
1737  * (*) get_dumpable regards interim value of 10 as 11.
1738  */
1739 void set_dumpable(struct mm_struct *mm, int value)
1740 {
1741         switch (value) {
1742         case 0:
1743                 clear_bit(MMF_DUMPABLE, &mm->flags);
1744                 smp_wmb();
1745                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1746                 break;
1747         case 1:
1748                 set_bit(MMF_DUMPABLE, &mm->flags);
1749                 smp_wmb();
1750                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1751                 break;
1752         case 2:
1753                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1754                 smp_wmb();
1755                 set_bit(MMF_DUMPABLE, &mm->flags);
1756                 break;
1757         }
1758 }
1759
1760 int get_dumpable(struct mm_struct *mm)
1761 {
1762         int ret;
1763
1764         ret = mm->flags & 0x3;
1765         return (ret >= 2) ? 2 : ret;
1766 }
1767
1768 static void wait_for_dump_helpers(struct file *file)
1769 {
1770         struct pipe_inode_info *pipe;
1771
1772         pipe = file->f_path.dentry->d_inode->i_pipe;
1773
1774         pipe_lock(pipe);
1775         pipe->readers++;
1776         pipe->writers--;
1777
1778         while ((pipe->readers > 1) && (!signal_pending(current))) {
1779                 wake_up_interruptible_sync(&pipe->wait);
1780                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1781                 pipe_wait(pipe);
1782         }
1783
1784         pipe->readers--;
1785         pipe->writers++;
1786         pipe_unlock(pipe);
1787
1788 }
1789
1790
1791 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1792 {
1793         struct core_state core_state;
1794         char corename[CORENAME_MAX_SIZE + 1];
1795         struct mm_struct *mm = current->mm;
1796         struct linux_binfmt * binfmt;
1797         struct inode * inode;
1798         struct file * file;
1799         const struct cred *old_cred;
1800         struct cred *cred;
1801         int retval = 0;
1802         int flag = 0;
1803         int ispipe = 0;
1804         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1805         char **helper_argv = NULL;
1806         int helper_argc = 0;
1807         int dump_count = 0;
1808         static atomic_t core_dump_count = ATOMIC_INIT(0);
1809
1810         audit_core_dumps(signr);
1811
1812         binfmt = mm->binfmt;
1813         if (!binfmt || !binfmt->core_dump)
1814                 goto fail;
1815
1816         cred = prepare_creds();
1817         if (!cred) {
1818                 retval = -ENOMEM;
1819                 goto fail;
1820         }
1821
1822         down_write(&mm->mmap_sem);
1823         /*
1824          * If another thread got here first, or we are not dumpable, bail out.
1825          */
1826         if (mm->core_state || !get_dumpable(mm)) {
1827                 up_write(&mm->mmap_sem);
1828                 put_cred(cred);
1829                 goto fail;
1830         }
1831
1832         /*
1833          *      We cannot trust fsuid as being the "true" uid of the
1834          *      process nor do we know its entire history. We only know it
1835          *      was tainted so we dump it as root in mode 2.
1836          */
1837         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1838                 flag = O_EXCL;          /* Stop rewrite attacks */
1839                 cred->fsuid = 0;        /* Dump root private */
1840         }
1841
1842         retval = coredump_wait(exit_code, &core_state);
1843         if (retval < 0) {
1844                 put_cred(cred);
1845                 goto fail;
1846         }
1847
1848         old_cred = override_creds(cred);
1849
1850         /*
1851          * Clear any false indication of pending signals that might
1852          * be seen by the filesystem code called to write the core file.
1853          */
1854         clear_thread_flag(TIF_SIGPENDING);
1855
1856         /*
1857          * lock_kernel() because format_corename() is controlled by sysctl, which
1858          * uses lock_kernel()
1859          */
1860         lock_kernel();
1861         ispipe = format_corename(corename, signr);
1862         unlock_kernel();
1863
1864         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1865                 goto fail_unlock;
1866
1867         if (ispipe) {
1868                 if (core_limit == 0) {
1869                         /*
1870                          * Normally core limits are irrelevant to pipes, since
1871                          * we're not writing to the file system, but we use
1872                          * core_limit of 0 here as a speacial value. Any
1873                          * non-zero limit gets set to RLIM_INFINITY below, but
1874                          * a limit of 0 skips the dump.  This is a consistent
1875                          * way to catch recursive crashes.  We can still crash
1876                          * if the core_pattern binary sets RLIM_CORE =  !0
1877                          * but it runs as root, and can do lots of stupid things
1878                          * Note that we use task_tgid_vnr here to grab the pid
1879                          * of the process group leader.  That way we get the
1880                          * right pid if a thread in a multi-threaded
1881                          * core_pattern process dies.
1882                          */
1883                         printk(KERN_WARNING
1884                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1885                                 task_tgid_vnr(current), current->comm);
1886                         printk(KERN_WARNING "Aborting core\n");
1887                         goto fail_unlock;
1888                 }
1889
1890                 dump_count = atomic_inc_return(&core_dump_count);
1891                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1892                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1893                                task_tgid_vnr(current), current->comm);
1894                         printk(KERN_WARNING "Skipping core dump\n");
1895                         goto fail_dropcount;
1896                 }
1897
1898                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1899                 if (!helper_argv) {
1900                         printk(KERN_WARNING "%s failed to allocate memory\n",
1901                                __func__);
1902                         goto fail_dropcount;
1903                 }
1904
1905                 core_limit = RLIM_INFINITY;
1906
1907                 /* SIGPIPE can happen, but it's just never processed */
1908                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1909                                 &file)) {
1910                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1911                                corename);
1912                         goto fail_dropcount;
1913                 }
1914         } else
1915                 file = filp_open(corename,
1916                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1917                                  0600);
1918         if (IS_ERR(file))
1919                 goto fail_dropcount;
1920         inode = file->f_path.dentry->d_inode;
1921         if (inode->i_nlink > 1)
1922                 goto close_fail;        /* multiple links - don't dump */
1923         if (!ispipe && d_unhashed(file->f_path.dentry))
1924                 goto close_fail;
1925
1926         /* AK: actually i see no reason to not allow this for named pipes etc.,
1927            but keep the previous behaviour for now. */
1928         if (!ispipe && !S_ISREG(inode->i_mode))
1929                 goto close_fail;
1930         /*
1931          * Dont allow local users get cute and trick others to coredump
1932          * into their pre-created files:
1933          * Note, this is not relevant for pipes
1934          */
1935         if (!ispipe && (inode->i_uid != current_fsuid()))
1936                 goto close_fail;
1937         if (!file->f_op)
1938                 goto close_fail;
1939         if (!file->f_op->write)
1940                 goto close_fail;
1941         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1942                 goto close_fail;
1943
1944         retval = binfmt->core_dump(signr, regs, file, core_limit);
1945
1946         if (retval)
1947                 current->signal->group_exit_code |= 0x80;
1948 close_fail:
1949         if (ispipe && core_pipe_limit)
1950                 wait_for_dump_helpers(file);
1951         filp_close(file, NULL);
1952 fail_dropcount:
1953         if (dump_count)
1954                 atomic_dec(&core_dump_count);
1955 fail_unlock:
1956         if (helper_argv)
1957                 argv_free(helper_argv);
1958
1959         revert_creds(old_cred);
1960         put_cred(cred);
1961         coredump_finish(mm);
1962 fail:
1963         return;
1964 }