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/syscalls.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/tracehook.h>
  54 #include <linux/kmod.h>
  55 #include <linux/fsnotify.h>
  56 #include <linux/fs_struct.h>
  57 #include <linux/pipe_fs_i.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include "internal.h"
  63
  64 int core_uses_pid;
  65 char core_pattern[CORENAME_MAX_SIZE] = "core";
  66 unsigned int core_pipe_limit;
  67 int suid_dumpable = 0;
  68
  69 /* The maximal length of core_pattern is also specified in sysctl.c */
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         if (!fmt)
  77                 return -EINVAL;
  78         write_lock(&binfmt_lock);
  79         insert ? list_add(&fmt->lh, &formats) :
  80                  list_add_tail(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 SYSCALL_DEFINE1(uselib, const char __user *, library)
 108 {
 109         struct file *file;
 110         char *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112
 113         if (IS_ERR(tmp))
 114                 goto out;
 115
 116         file = do_filp_open(AT_FDCWD, tmp,
 117                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 118                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 119         putname(tmp);
 120         error = PTR_ERR(file);
 121         if (IS_ERR(file))
 122                 goto out;
 123
 124         error = -EINVAL;
 125         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 126                 goto exit;
 127
 128         error = -EACCES;
 129         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 130                 goto exit;
 131
 132         fsnotify_open(file->f_path.dentry);
 133
 134         error = -ENOEXEC;
 135         if(file->f_op) {
 136                 struct linux_binfmt * fmt;
 137
 138                 read_lock(&binfmt_lock);
 139                 list_for_each_entry(fmt, &formats, lh) {
 140                         if (!fmt->load_shlib)
 141                                 continue;
 142                         if (!try_module_get(fmt->module))
 143                                 continue;
 144                         read_unlock(&binfmt_lock);
 145                         error = fmt->load_shlib(file);
 146                         read_lock(&binfmt_lock);
 147                         put_binfmt(fmt);
 148                         if (error != -ENOEXEC)
 149                                 break;
 150                 }
 151                 read_unlock(&binfmt_lock);
 152         }
 153 exit:
 154         fput(file);
 155 out:
 156         return error;
 157 }
 158
 159 #ifdef CONFIG_MMU
 160
 161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 162                 int write)
 163 {
 164         struct page *page;
 165         int ret;
 166
 167 #ifdef CONFIG_STACK_GROWSUP
 168         if (write) {
 169                 ret = expand_stack_downwards(bprm->vma, pos);
 170                 if (ret < 0)
 171                         return NULL;
 172         }
 173 #endif
 174         ret = get_user_pages(current, bprm->mm, pos,
 175                         1, write, 1, &page, NULL);
 176         if (ret <= 0)
 177                 return NULL;
 178
 179         if (write) {
 180                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 181                 struct rlimit *rlim;
 182
 183                 /*
 184                  * We've historically supported up to 32 pages (ARG_MAX)
 185                  * of argument strings even with small stacks
 186                  */
 187                 if (size <= ARG_MAX)
 188                         return page;
 189
 190                 /*
 191                  * Limit to 1/4-th the stack size for the argv+env strings.
 192                  * This ensures that:
 193                  *  - the remaining binfmt code will not run out of stack space,
 194                  *  - the program will have a reasonable amount of stack left
 195                  *    to work from.
 196                  */
 197                 rlim = current->signal->rlim;
 198                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 199                         put_page(page);
 200                         return NULL;
 201                 }
 202         }
 203
 204         return page;
 205 }
 206
 207 static void put_arg_page(struct page *page)
 208 {
 209         put_page(page);
 210 }
 211
 212 static void free_arg_page(struct linux_binprm *bprm, int i)
 213 {
 214 }
 215
 216 static void free_arg_pages(struct linux_binprm *bprm)
 217 {
 218 }
 219
 220 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 221                 struct page *page)
 222 {
 223         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 224 }
 225
 226 static int __bprm_mm_init(struct linux_binprm *bprm)
 227 {
 228         int err;
 229         struct vm_area_struct *vma = NULL;
 230         struct mm_struct *mm = bprm->mm;
 231
 232         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 233         if (!vma)
 234                 return -ENOMEM;
 235
 236         down_write(&mm->mmap_sem);
 237         vma->vm_mm = mm;
 238
 239         /*
 240          * Place the stack at the largest stack address the architecture
 241          * supports. Later, we'll move this to an appropriate place. We don't
 242          * use STACK_TOP because that can depend on attributes which aren't
 243          * configured yet.
 244          */
 245         vma->vm_end = STACK_TOP_MAX;
 246         vma->vm_start = vma->vm_end - PAGE_SIZE;
 247         vma->vm_flags = VM_STACK_FLAGS;
 248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 249
 250         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 251         if (err)
 252                 goto err;
 253
 254         err = insert_vm_struct(mm, vma);
 255         if (err)
 256                 goto err;
 257
 258         mm->stack_vm = mm->total_vm = 1;
 259         up_write(&mm->mmap_sem);
 260         bprm->p = vma->vm_end - sizeof(void *);
 261         return 0;
 262 err:
 263         up_write(&mm->mmap_sem);
 264         bprm->vma = NULL;
 265         kmem_cache_free(vm_area_cachep, vma);
 266         return err;
 267 }
 268
 269 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 270 {
 271         return len <= MAX_ARG_STRLEN;
 272 }
 273
 274 #else
 275
 276 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 277                 int write)
 278 {
 279         struct page *page;
 280
 281         page = bprm->page[pos / PAGE_SIZE];
 282         if (!page && write) {
 283                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 284                 if (!page)
 285                         return NULL;
 286                 bprm->page[pos / PAGE_SIZE] = page;
 287         }
 288
 289         return page;
 290 }
 291
 292 static void put_arg_page(struct page *page)
 293 {
 294 }
 295
 296 static void free_arg_page(struct linux_binprm *bprm, int i)
 297 {
 298         if (bprm->page[i]) {
 299                 __free_page(bprm->page[i]);
 300                 bprm->page[i] = NULL;
 301         }
 302 }
 303
 304 static void free_arg_pages(struct linux_binprm *bprm)
 305 {
 306         int i;
 307
 308         for (i = 0; i < MAX_ARG_PAGES; i++)
 309                 free_arg_page(bprm, i);
 310 }
 311
 312 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 313                 struct page *page)
 314 {
 315 }
 316
 317 static int __bprm_mm_init(struct linux_binprm *bprm)
 318 {
 319         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 320         return 0;
 321 }
 322
 323 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 324 {
 325         return len <= bprm->p;
 326 }
 327
 328 #endif /* CONFIG_MMU */
 329
 330 /*
 331  * Create a new mm_struct and populate it with a temporary stack
 332  * vm_area_struct.  We don't have enough context at this point to set the stack
 333  * flags, permissions, and offset, so we use temporary values.  We'll update
 334  * them later in setup_arg_pages().
 335  */
 336 int bprm_mm_init(struct linux_binprm *bprm)
 337 {
 338         int err;
 339         struct mm_struct *mm = NULL;
 340
 341         bprm->mm = mm = mm_alloc();
 342         err = -ENOMEM;
 343         if (!mm)
 344                 goto err;
 345
 346         err = init_new_context(current, mm);
 347         if (err)
 348                 goto err;
 349
 350         err = __bprm_mm_init(bprm);
 351         if (err)
 352                 goto err;
 353
 354         return 0;
 355
 356 err:
 357         if (mm) {
 358                 bprm->mm = NULL;
 359                 mmdrop(mm);
 360         }
 361
 362         return err;
 363 }
 364
 365 /*
 366  * count() counts the number of strings in array ARGV.
 367  */
 368 static int count(char __user * __user * argv, int max)
 369 {
 370         int i = 0;
 371
 372         if (argv != NULL) {
 373                 for (;;) {
 374                         char __user * p;
 375
 376                         if (get_user(p, argv))
 377                                 return -EFAULT;
 378                         if (!p)
 379                                 break;
 380                         argv++;
 381                         if (i++ >= max)
 382                                 return -E2BIG;
 383
 384                         if (fatal_signal_pending(current))
 385                                 return -ERESTARTNOHAND;
 386                         cond_resched();
 387                 }
 388         }
 389         return i;
 390 }
 391
 392 /*
 393  * 'copy_strings()' copies argument/environment strings from the old
 394  * processes's memory to the new process's stack.  The call to get_user_pages()
 395  * ensures the destination page is created and not swapped out.
 396  */
 397 static int copy_strings(int argc, char __user * __user * argv,
 398                         struct linux_binprm *bprm)
 399 {
 400         struct page *kmapped_page = NULL;
 401         char *kaddr = NULL;
 402         unsigned long kpos = 0;
 403         int ret;
 404
 405         while (argc-- > 0) {
 406                 char __user *str;
 407                 int len;
 408                 unsigned long pos;
 409
 410                 if (get_user(str, argv+argc) ||
 411                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 412                         ret = -EFAULT;
 413                         goto out;
 414                 }
 415
 416                 if (!valid_arg_len(bprm, len)) {
 417                         ret = -E2BIG;
 418                         goto out;
 419                 }
 420
 421                 /* We're going to work our way backwords. */
 422                 pos = bprm->p;
 423                 str += len;
 424                 bprm->p -= len;
 425
 426                 while (len > 0) {
 427                         int offset, bytes_to_copy;
 428
 429                         if (fatal_signal_pending(current)) {
 430                                 ret = -ERESTARTNOHAND;
 431                                 goto out;
 432                         }
 433                         cond_resched();
 434
 435                         offset = pos % PAGE_SIZE;
 436                         if (offset == 0)
 437                                 offset = PAGE_SIZE;
 438
 439                         bytes_to_copy = offset;
 440                         if (bytes_to_copy > len)
 441                                 bytes_to_copy = len;
 442
 443                         offset -= bytes_to_copy;
 444                         pos -= bytes_to_copy;
 445                         str -= bytes_to_copy;
 446                         len -= bytes_to_copy;
 447
 448                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 449                                 struct page *page;
 450
 451                                 page = get_arg_page(bprm, pos, 1);
 452                                 if (!page) {
 453                                         ret = -E2BIG;
 454                                         goto out;
 455                                 }
 456
 457                                 if (kmapped_page) {
 458                                         flush_kernel_dcache_page(kmapped_page);
 459                                         kunmap(kmapped_page);
 460                                         put_arg_page(kmapped_page);
 461                                 }
 462                                 kmapped_page = page;
 463                                 kaddr = kmap(kmapped_page);
 464                                 kpos = pos & PAGE_MASK;
 465                                 flush_arg_page(bprm, kpos, kmapped_page);
 466                         }
 467                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 468                                 ret = -EFAULT;
 469                                 goto out;
 470                         }
 471                 }
 472         }
 473         ret = 0;
 474 out:
 475         if (kmapped_page) {
 476                 flush_kernel_dcache_page(kmapped_page);
 477                 kunmap(kmapped_page);
 478                 put_arg_page(kmapped_page);
 479         }
 480         return ret;
 481 }
 482
 483 /*
 484  * Like copy_strings, but get argv and its values from kernel memory.
 485  */
 486 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 487 {
 488         int r;
 489         mm_segment_t oldfs = get_fs();
 490         set_fs(KERNEL_DS);
 491         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 492         set_fs(oldfs);
 493         return r;
 494 }
 495 EXPORT_SYMBOL(copy_strings_kernel);
 496
 497 #ifdef CONFIG_MMU
 498
 499 /*
 500  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 501  * the binfmt code determines where the new stack should reside, we shift it to
 502  * its final location.  The process proceeds as follows:
 503  *
 504  * 1) Use shift to calculate the new vma endpoints.
 505  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 506  *    arguments passed to subsequent functions are consistent.
 507  * 3) Move vma's page tables to the new range.
 508  * 4) Free up any cleared pgd range.
 509  * 5) Shrink the vma to cover only the new range.
 510  */
 511 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 512 {
 513         struct mm_struct *mm = vma->vm_mm;
 514         unsigned long old_start = vma->vm_start;
 515         unsigned long old_end = vma->vm_end;
 516         unsigned long length = old_end - old_start;
 517         unsigned long new_start = old_start - shift;
 518         unsigned long new_end = old_end - shift;
 519         struct mmu_gather *tlb;
 520
 521         BUG_ON(new_start > new_end);
 522
 523         /*
 524          * ensure there are no vmas between where we want to go
 525          * and where we are
 526          */
 527         if (vma != find_vma(mm, new_start))
 528                 return -EFAULT;
 529
 530         /*
 531          * cover the whole range: [new_start, old_end)
 532          */
 533         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 534
 535         /*
 536          * move the page tables downwards, on failure we rely on
 537          * process cleanup to remove whatever mess we made.
 538          */
 539         if (length != move_page_tables(vma, old_start,
 540                                        vma, new_start, length))
 541                 return -ENOMEM;
 542
 543         lru_add_drain();
 544         tlb = tlb_gather_mmu(mm, 0);
 545         if (new_end > old_start) {
 546                 /*
 547                  * when the old and new regions overlap clear from new_end.
 548                  */
 549                 free_pgd_range(tlb, new_end, old_end, new_end,
 550                         vma->vm_next ? vma->vm_next->vm_start : 0);
 551         } else {
 552                 /*
 553                  * otherwise, clean from old_start; this is done to not touch
 554                  * the address space in [new_end, old_start) some architectures
 555                  * have constraints on va-space that make this illegal (IA64) -
 556                  * for the others its just a little faster.
 557                  */
 558                 free_pgd_range(tlb, old_start, old_end, new_end,
 559                         vma->vm_next ? vma->vm_next->vm_start : 0);
 560         }
 561         tlb_finish_mmu(tlb, new_end, old_end);
 562
 563         /*
 564          * shrink the vma to just the new range.
 565          */
 566         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 567
 568         return 0;
 569 }
 570
 571 #define EXTRA_STACK_VM_PAGES    20      /* random */
 572
 573 /*
 574  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 575  * the stack is optionally relocated, and some extra space is added.
 576  */
 577 int setup_arg_pages(struct linux_binprm *bprm,
 578                     unsigned long stack_top,
 579                     int executable_stack)
 580 {
 581         unsigned long ret;
 582         unsigned long stack_shift;
 583         struct mm_struct *mm = current->mm;
 584         struct vm_area_struct *vma = bprm->vma;
 585         struct vm_area_struct *prev = NULL;
 586         unsigned long vm_flags;
 587         unsigned long stack_base;
 588         unsigned long stack_size;
 589         unsigned long stack_expand;
 590         unsigned long rlim_stack;
 591
 592 #ifdef CONFIG_STACK_GROWSUP
 593         /* Limit stack size to 1GB */
 594         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 595         if (stack_base > (1 << 30))
 596                 stack_base = 1 << 30;
 597
 598         /* Make sure we didn't let the argument array grow too large. */
 599         if (vma->vm_end - vma->vm_start > stack_base)
 600                 return -ENOMEM;
 601
 602         stack_base = PAGE_ALIGN(stack_top - stack_base);
 603
 604         stack_shift = vma->vm_start - stack_base;
 605         mm->arg_start = bprm->p - stack_shift;
 606         bprm->p = vma->vm_end - stack_shift;
 607 #else
 608         stack_top = arch_align_stack(stack_top);
 609         stack_top = PAGE_ALIGN(stack_top);
 610
 611         if (unlikely(stack_top < mmap_min_addr) ||
 612             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 613                 return -ENOMEM;
 614
 615         stack_shift = vma->vm_end - stack_top;
 616
 617         bprm->p -= stack_shift;
 618         mm->arg_start = bprm->p;
 619 #endif
 620
 621         if (bprm->loader)
 622                 bprm->loader -= stack_shift;
 623         bprm->exec -= stack_shift;
 624
 625         down_write(&mm->mmap_sem);
 626         vm_flags = VM_STACK_FLAGS;
 627
 628         /*
 629          * Adjust stack execute permissions; explicitly enable for
 630          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 631          * (arch default) otherwise.
 632          */
 633         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 634                 vm_flags |= VM_EXEC;
 635         else if (executable_stack == EXSTACK_DISABLE_X)
 636                 vm_flags &= ~VM_EXEC;
 637         vm_flags |= mm->def_flags;
 638
 639         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 640                         vm_flags);
 641         if (ret)
 642                 goto out_unlock;
 643         BUG_ON(prev != vma);
 644
 645         /* Move stack pages down in memory. */
 646         if (stack_shift) {
 647                 ret = shift_arg_pages(vma, stack_shift);
 648                 if (ret)
 649                         goto out_unlock;
 650         }
 651
 652         stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 653         stack_size = vma->vm_end - vma->vm_start;
 654         /*
 655          * Align this down to a page boundary as expand_stack
 656          * will align it up.
 657          */
 658         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 659 #ifdef CONFIG_STACK_GROWSUP
 660         if (stack_size + stack_expand > rlim_stack)
 661                 stack_base = vma->vm_start + rlim_stack;
 662         else
 663                 stack_base = vma->vm_end + stack_expand;
 664 #else
 665         if (stack_size + stack_expand > rlim_stack)
 666                 stack_base = vma->vm_end - rlim_stack;
 667         else
 668                 stack_base = vma->vm_start - stack_expand;
 669 #endif
 670         ret = expand_stack(vma, stack_base);
 671         if (ret)
 672                 ret = -EFAULT;
 673
 674 out_unlock:
 675         up_write(&mm->mmap_sem);
 676         return ret;
 677 }
 678 EXPORT_SYMBOL(setup_arg_pages);
 679
 680 #endif /* CONFIG_MMU */
 681
 682 struct file *open_exec(const char *name)
 683 {
 684         struct file *file;
 685         int err;
 686
 687         file = do_filp_open(AT_FDCWD, name,
 688                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 689                                 MAY_EXEC | MAY_OPEN);
 690         if (IS_ERR(file))
 691                 goto out;
 692
 693         err = -EACCES;
 694         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 695                 goto exit;
 696
 697         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 698                 goto exit;
 699
 700         fsnotify_open(file->f_path.dentry);
 701
 702         err = deny_write_access(file);
 703         if (err)
 704                 goto exit;
 705
 706 out:
 707         return file;
 708
 709 exit:
 710         fput(file);
 711         return ERR_PTR(err);
 712 }
 713 EXPORT_SYMBOL(open_exec);
 714
 715 int kernel_read(struct file *file, loff_t offset,
 716                 char *addr, unsigned long count)
 717 {
 718         mm_segment_t old_fs;
 719         loff_t pos = offset;
 720         int result;
 721
 722         old_fs = get_fs();
 723         set_fs(get_ds());
 724         /* The cast to a user pointer is valid due to the set_fs() */
 725         result = vfs_read(file, (void __user *)addr, count, &pos);
 726         set_fs(old_fs);
 727         return result;
 728 }
 729
 730 EXPORT_SYMBOL(kernel_read);
 731
 732 static int exec_mmap(struct mm_struct *mm)
 733 {
 734         struct task_struct *tsk;
 735         struct mm_struct * old_mm, *active_mm;
 736
 737         /* Notify parent that we're no longer interested in the old VM */
 738         tsk = current;
 739         old_mm = current->mm;
 740         mm_release(tsk, old_mm);
 741
 742         if (old_mm) {
 743                 /*
 744                  * Make sure that if there is a core dump in progress
 745                  * for the old mm, we get out and die instead of going
 746                  * through with the exec.  We must hold mmap_sem around
 747                  * checking core_state and changing tsk->mm.
 748                  */
 749                 down_read(&old_mm->mmap_sem);
 750                 if (unlikely(old_mm->core_state)) {
 751                         up_read(&old_mm->mmap_sem);
 752                         return -EINTR;
 753                 }
 754         }
 755         task_lock(tsk);
 756         active_mm = tsk->active_mm;
 757         tsk->mm = mm;
 758         tsk->active_mm = mm;
 759         activate_mm(active_mm, mm);
 760         task_unlock(tsk);
 761         arch_pick_mmap_layout(mm);
 762         if (old_mm) {
 763                 up_read(&old_mm->mmap_sem);
 764                 BUG_ON(active_mm != old_mm);
 765                 mm_update_next_owner(old_mm);
 766                 mmput(old_mm);
 767                 return 0;
 768         }
 769         mmdrop(active_mm);
 770         return 0;
 771 }
 772
 773 /*
 774  * This function makes sure the current process has its own signal table,
 775  * so that flush_signal_handlers can later reset the handlers without
 776  * disturbing other processes.  (Other processes might share the signal
 777  * table via the CLONE_SIGHAND option to clone().)
 778  */
 779 static int de_thread(struct task_struct *tsk)
 780 {
 781         struct signal_struct *sig = tsk->signal;
 782         struct sighand_struct *oldsighand = tsk->sighand;
 783         spinlock_t *lock = &oldsighand->siglock;
 784         int count;
 785
 786         if (thread_group_empty(tsk))
 787                 goto no_thread_group;
 788
 789         /*
 790          * Kill all other threads in the thread group.
 791          */
 792         spin_lock_irq(lock);
 793         if (signal_group_exit(sig)) {
 794                 /*
 795                  * Another group action in progress, just
 796                  * return so that the signal is processed.
 797                  */
 798                 spin_unlock_irq(lock);
 799                 return -EAGAIN;
 800         }
 801         sig->group_exit_task = tsk;
 802         zap_other_threads(tsk);
 803
 804         /* Account for the thread group leader hanging around: */
 805         count = thread_group_leader(tsk) ? 1 : 2;
 806         sig->notify_count = count;
 807         while (atomic_read(&sig->count) > count) {
 808                 __set_current_state(TASK_UNINTERRUPTIBLE);
 809                 spin_unlock_irq(lock);
 810                 schedule();
 811                 spin_lock_irq(lock);
 812         }
 813         spin_unlock_irq(lock);
 814
 815         /*
 816          * At this point all other threads have exited, all we have to
 817          * do is to wait for the thread group leader to become inactive,
 818          * and to assume its PID:
 819          */
 820         if (!thread_group_leader(tsk)) {
 821                 struct task_struct *leader = tsk->group_leader;
 822
 823                 sig->notify_count = -1; /* for exit_notify() */
 824                 for (;;) {
 825                         write_lock_irq(&tasklist_lock);
 826                         if (likely(leader->exit_state))
 827                                 break;
 828                         __set_current_state(TASK_UNINTERRUPTIBLE);
 829                         write_unlock_irq(&tasklist_lock);
 830                         schedule();
 831                 }
 832
 833                 /*
 834                  * The only record we have of the real-time age of a
 835                  * process, regardless of execs it's done, is start_time.
 836                  * All the past CPU time is accumulated in signal_struct
 837                  * from sister threads now dead.  But in this non-leader
 838                  * exec, nothing survives from the original leader thread,
 839                  * whose birth marks the true age of this process now.
 840                  * When we take on its identity by switching to its PID, we
 841                  * also take its birthdate (always earlier than our own).
 842                  */
 843                 tsk->start_time = leader->start_time;
 844
 845                 BUG_ON(!same_thread_group(leader, tsk));
 846                 BUG_ON(has_group_leader_pid(tsk));
 847                 /*
 848                  * An exec() starts a new thread group with the
 849                  * TGID of the previous thread group. Rehash the
 850                  * two threads with a switched PID, and release
 851                  * the former thread group leader:
 852                  */
 853
 854                 /* Become a process group leader with the old leader's pid.
 855                  * The old leader becomes a thread of the this thread group.
 856                  * Note: The old leader also uses this pid until release_task
 857                  *       is called.  Odd but simple and correct.
 858                  */
 859                 detach_pid(tsk, PIDTYPE_PID);
 860                 tsk->pid = leader->pid;
 861                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 862                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 863                 transfer_pid(leader, tsk, PIDTYPE_SID);
 864
 865                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 866                 list_replace_init(&leader->sibling, &tsk->sibling);
 867
 868                 tsk->group_leader = tsk;
 869                 leader->group_leader = tsk;
 870
 871                 tsk->exit_signal = SIGCHLD;
 872
 873                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 874                 leader->exit_state = EXIT_DEAD;
 875                 write_unlock_irq(&tasklist_lock);
 876
 877                 release_task(leader);
 878         }
 879
 880         sig->group_exit_task = NULL;
 881         sig->notify_count = 0;
 882
 883 no_thread_group:
 884         if (current->mm)
 885                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 886
 887         exit_itimers(sig);
 888         flush_itimer_signals();
 889
 890         if (atomic_read(&oldsighand->count) != 1) {
 891                 struct sighand_struct *newsighand;
 892                 /*
 893                  * This ->sighand is shared with the CLONE_SIGHAND
 894                  * but not CLONE_THREAD task, switch to the new one.
 895                  */
 896                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 897                 if (!newsighand)
 898                         return -ENOMEM;
 899
 900                 atomic_set(&newsighand->count, 1);
 901                 memcpy(newsighand->action, oldsighand->action,
 902                        sizeof(newsighand->action));
 903
 904                 write_lock_irq(&tasklist_lock);
 905                 spin_lock(&oldsighand->siglock);
 906                 rcu_assign_pointer(tsk->sighand, newsighand);
 907                 spin_unlock(&oldsighand->siglock);
 908                 write_unlock_irq(&tasklist_lock);
 909
 910                 __cleanup_sighand(oldsighand);
 911         }
 912
 913         BUG_ON(!thread_group_leader(tsk));
 914         return 0;
 915 }
 916
 917 /*
 918  * These functions flushes out all traces of the currently running executable
 919  * so that a new one can be started
 920  */
 921 static void flush_old_files(struct files_struct * files)
 922 {
 923         long j = -1;
 924         struct fdtable *fdt;
 925
 926         spin_lock(&files->file_lock);
 927         for (;;) {
 928                 unsigned long set, i;
 929
 930                 j++;
 931                 i = j * __NFDBITS;
 932                 fdt = files_fdtable(files);
 933                 if (i >= fdt->max_fds)
 934                         break;
 935                 set = fdt->close_on_exec->fds_bits[j];
 936                 if (!set)
 937                         continue;
 938                 fdt->close_on_exec->fds_bits[j] = 0;
 939                 spin_unlock(&files->file_lock);
 940                 for ( ; set ; i++,set >>= 1) {
 941                         if (set & 1) {
 942                                 sys_close(i);
 943                         }
 944                 }
 945                 spin_lock(&files->file_lock);
 946
 947         }
 948         spin_unlock(&files->file_lock);
 949 }
 950
 951 char *get_task_comm(char *buf, struct task_struct *tsk)
 952 {
 953         /* buf must be at least sizeof(tsk->comm) in size */
 954         task_lock(tsk);
 955         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 956         task_unlock(tsk);
 957         return buf;
 958 }
 959
 960 void set_task_comm(struct task_struct *tsk, char *buf)
 961 {
 962         task_lock(tsk);
 963
 964         /*
 965          * Threads may access current->comm without holding
 966          * the task lock, so write the string carefully.
 967          * Readers without a lock may see incomplete new
 968          * names but are safe from non-terminating string reads.
 969          */
 970         memset(tsk->comm, 0, TASK_COMM_LEN);
 971         wmb();
 972         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 973         task_unlock(tsk);
 974         perf_event_comm(tsk);
 975 }
 976
 977 int flush_old_exec(struct linux_binprm * bprm)
 978 {
 979         int retval;
 980
 981         /*
 982          * Make sure we have a private signal table and that
 983          * we are unassociated from the previous thread group.
 984          */
 985         retval = de_thread(current);
 986         if (retval)
 987                 goto out;
 988
 989         set_mm_exe_file(bprm->mm, bprm->file);
 990
 991         /*
 992          * Release all of the old mmap stuff
 993          */
 994         retval = exec_mmap(bprm->mm);
 995         if (retval)
 996                 goto out;
 997
 998         bprm->mm = NULL;                /* We're using it now */
 999
1000         current->flags &= ~PF_RANDOMIZE;
1001         flush_thread();
1002         current->personality &= ~bprm->per_clear;
1003
1004         return 0;
1005
1006 out:
1007         return retval;
1008 }
1009 EXPORT_SYMBOL(flush_old_exec);
1010
1011 void setup_new_exec(struct linux_binprm * bprm)
1012 {
1013         int i, ch;
1014         char * name;
1015         char tcomm[sizeof(current->comm)];
1016
1017         arch_pick_mmap_layout(current->mm);
1018
1019         /* This is the point of no return */
1020         current->sas_ss_sp = current->sas_ss_size = 0;
1021
1022         if (current_euid() == current_uid() && current_egid() == current_gid())
1023                 set_dumpable(current->mm, 1);
1024         else
1025                 set_dumpable(current->mm, suid_dumpable);
1026
1027         name = bprm->filename;
1028
1029         /* Copies the binary name from after last slash */
1030         for (i=0; (ch = *(name++)) != '\0';) {
1031                 if (ch == '/')
1032                         i = 0; /* overwrite what we wrote */
1033                 else
1034                         if (i < (sizeof(tcomm) - 1))
1035                                 tcomm[i++] = ch;
1036         }
1037         tcomm[i] = '\0';
1038         set_task_comm(current, tcomm);
1039
1040         /* Set the new mm task size. We have to do that late because it may
1041          * depend on TIF_32BIT which is only updated in flush_thread() on
1042          * some architectures like powerpc
1043          */
1044         current->mm->task_size = TASK_SIZE;
1045
1046         /* install the new credentials */
1047         if (bprm->cred->uid != current_euid() ||
1048             bprm->cred->gid != current_egid()) {
1049                 current->pdeath_signal = 0;
1050         } else if (file_permission(bprm->file, MAY_READ) ||
1051                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1052                 set_dumpable(current->mm, suid_dumpable);
1053         }
1054
1055         /*
1056          * Flush performance counters when crossing a
1057          * security domain:
1058          */
1059         if (!get_dumpable(current->mm))
1060                 perf_event_exit_task(current);
1061
1062         /* An exec changes our domain. We are no longer part of the thread
1063            group */
1064
1065         current->self_exec_id++;
1066
1067         flush_signal_handlers(current, 0);
1068         flush_old_files(current->files);
1069 }
1070 EXPORT_SYMBOL(setup_new_exec);
1071
1072 /*
1073  * Prepare credentials and lock ->cred_guard_mutex.
1074  * install_exec_creds() commits the new creds and drops the lock.
1075  * Or, if exec fails before, free_bprm() should release ->cred and
1076  * and unlock.
1077  */
1078 int prepare_bprm_creds(struct linux_binprm *bprm)
1079 {
1080         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1081                 return -ERESTARTNOINTR;
1082
1083         bprm->cred = prepare_exec_creds();
1084         if (likely(bprm->cred))
1085                 return 0;
1086
1087         mutex_unlock(&current->cred_guard_mutex);
1088         return -ENOMEM;
1089 }
1090
1091 void free_bprm(struct linux_binprm *bprm)
1092 {
1093         free_arg_pages(bprm);
1094         if (bprm->cred) {
1095                 mutex_unlock(&current->cred_guard_mutex);
1096                 abort_creds(bprm->cred);
1097         }
1098         kfree(bprm);
1099 }
1100
1101 /*
1102  * install the new credentials for this executable
1103  */
1104 void install_exec_creds(struct linux_binprm *bprm)
1105 {
1106         security_bprm_committing_creds(bprm);
1107
1108         commit_creds(bprm->cred);
1109         bprm->cred = NULL;
1110         /*
1111          * cred_guard_mutex must be held at least to this point to prevent
1112          * ptrace_attach() from altering our determination of the task's
1113          * credentials; any time after this it may be unlocked.
1114          */
1115         security_bprm_committed_creds(bprm);
1116         mutex_unlock(&current->cred_guard_mutex);
1117 }
1118 EXPORT_SYMBOL(install_exec_creds);
1119
1120 /*
1121  * determine how safe it is to execute the proposed program
1122  * - the caller must hold current->cred_guard_mutex to protect against
1123  *   PTRACE_ATTACH
1124  */
1125 int check_unsafe_exec(struct linux_binprm *bprm)
1126 {
1127         struct task_struct *p = current, *t;
1128         unsigned n_fs;
1129         int res = 0;
1130
1131         bprm->unsafe = tracehook_unsafe_exec(p);
1132
1133         n_fs = 1;
1134         write_lock(&p->fs->lock);
1135         rcu_read_lock();
1136         for (t = next_thread(p); t != p; t = next_thread(t)) {
1137                 if (t->fs == p->fs)
1138                         n_fs++;
1139         }
1140         rcu_read_unlock();
1141
1142         if (p->fs->users > n_fs) {
1143                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1144         } else {
1145                 res = -EAGAIN;
1146                 if (!p->fs->in_exec) {
1147                         p->fs->in_exec = 1;
1148                         res = 1;
1149                 }
1150         }
1151         write_unlock(&p->fs->lock);
1152
1153         return res;
1154 }
1155
1156 /*
1157  * Fill the binprm structure from the inode.
1158  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1159  *
1160  * This may be called multiple times for binary chains (scripts for example).
1161  */
1162 int prepare_binprm(struct linux_binprm *bprm)
1163 {
1164         umode_t mode;
1165         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1166         int retval;
1167
1168         mode = inode->i_mode;
1169         if (bprm->file->f_op == NULL)
1170                 return -EACCES;
1171
1172         /* clear any previous set[ug]id data from a previous binary */
1173         bprm->cred->euid = current_euid();
1174         bprm->cred->egid = current_egid();
1175
1176         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1177                 /* Set-uid? */
1178                 if (mode & S_ISUID) {
1179                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1180                         bprm->cred->euid = inode->i_uid;
1181                 }
1182
1183                 /* Set-gid? */
1184                 /*
1185                  * If setgid is set but no group execute bit then this
1186                  * is a candidate for mandatory locking, not a setgid
1187                  * executable.
1188                  */
1189                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1190                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1191                         bprm->cred->egid = inode->i_gid;
1192                 }
1193         }
1194
1195         /* fill in binprm security blob */
1196         retval = security_bprm_set_creds(bprm);
1197         if (retval)
1198                 return retval;
1199         bprm->cred_prepared = 1;
1200
1201         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1202         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1203 }
1204
1205 EXPORT_SYMBOL(prepare_binprm);
1206
1207 /*
1208  * Arguments are '\0' separated strings found at the location bprm->p
1209  * points to; chop off the first by relocating brpm->p to right after
1210  * the first '\0' encountered.
1211  */
1212 int remove_arg_zero(struct linux_binprm *bprm)
1213 {
1214         int ret = 0;
1215         unsigned long offset;
1216         char *kaddr;
1217         struct page *page;
1218
1219         if (!bprm->argc)
1220                 return 0;
1221
1222         do {
1223                 offset = bprm->p & ~PAGE_MASK;
1224                 page = get_arg_page(bprm, bprm->p, 0);
1225                 if (!page) {
1226                         ret = -EFAULT;
1227                         goto out;
1228                 }
1229                 kaddr = kmap_atomic(page, KM_USER0);
1230
1231                 for (; offset < PAGE_SIZE && kaddr[offset];
1232                                 offset++, bprm->p++)
1233                         ;
1234
1235                 kunmap_atomic(kaddr, KM_USER0);
1236                 put_arg_page(page);
1237
1238                 if (offset == PAGE_SIZE)
1239                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1240         } while (offset == PAGE_SIZE);
1241
1242         bprm->p++;
1243         bprm->argc--;
1244         ret = 0;
1245
1246 out:
1247         return ret;
1248 }
1249 EXPORT_SYMBOL(remove_arg_zero);
1250
1251 /*
1252  * cycle the list of binary formats handler, until one recognizes the image
1253  */
1254 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1255 {
1256         unsigned int depth = bprm->recursion_depth;
1257         int try,retval;
1258         struct linux_binfmt *fmt;
1259
1260         retval = security_bprm_check(bprm);
1261         if (retval)
1262                 return retval;
1263
1264         /* kernel module loader fixup */
1265         /* so we don't try to load run modprobe in kernel space. */
1266         set_fs(USER_DS);
1267
1268         retval = audit_bprm(bprm);
1269         if (retval)
1270                 return retval;
1271
1272         retval = -ENOENT;
1273         for (try=0; try<2; try++) {
1274                 read_lock(&binfmt_lock);
1275                 list_for_each_entry(fmt, &formats, lh) {
1276                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1277                         if (!fn)
1278                                 continue;
1279                         if (!try_module_get(fmt->module))
1280                                 continue;
1281                         read_unlock(&binfmt_lock);
1282                         retval = fn(bprm, regs);
1283                         /*
1284                          * Restore the depth counter to its starting value
1285                          * in this call, so we don't have to rely on every
1286                          * load_binary function to restore it on return.
1287                          */
1288                         bprm->recursion_depth = depth;
1289                         if (retval >= 0) {
1290                                 if (depth == 0)
1291                                         tracehook_report_exec(fmt, bprm, regs);
1292                                 put_binfmt(fmt);
1293                                 allow_write_access(bprm->file);
1294                                 if (bprm->file)
1295                                         fput(bprm->file);
1296                                 bprm->file = NULL;
1297                                 current->did_exec = 1;
1298                                 proc_exec_connector(current);
1299                                 return retval;
1300                         }
1301                         read_lock(&binfmt_lock);
1302                         put_binfmt(fmt);
1303                         if (retval != -ENOEXEC || bprm->mm == NULL)
1304                                 break;
1305                         if (!bprm->file) {
1306                                 read_unlock(&binfmt_lock);
1307                                 return retval;
1308                         }
1309                 }
1310                 read_unlock(&binfmt_lock);
1311                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1312                         break;
1313 #ifdef CONFIG_MODULES
1314                 } else {
1315 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1316                         if (printable(bprm->buf[0]) &&
1317                             printable(bprm->buf[1]) &&
1318                             printable(bprm->buf[2]) &&
1319                             printable(bprm->buf[3]))
1320                                 break; /* -ENOEXEC */
1321                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1322 #endif
1323                 }
1324         }
1325         return retval;
1326 }
1327
1328 EXPORT_SYMBOL(search_binary_handler);
1329
1330 /*
1331  * sys_execve() executes a new program.
1332  */
1333 int do_execve(char * filename,
1334         char __user *__user *argv,
1335         char __user *__user *envp,
1336         struct pt_regs * regs)
1337 {
1338         struct linux_binprm *bprm;
1339         struct file *file;
1340         struct files_struct *displaced;
1341         bool clear_in_exec;
1342         int retval;
1343
1344         retval = unshare_files(&displaced);
1345         if (retval)
1346                 goto out_ret;
1347
1348         retval = -ENOMEM;
1349         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1350         if (!bprm)
1351                 goto out_files;
1352
1353         retval = prepare_bprm_creds(bprm);
1354         if (retval)
1355                 goto out_free;
1356
1357         retval = check_unsafe_exec(bprm);
1358         if (retval < 0)
1359                 goto out_free;
1360         clear_in_exec = retval;
1361         current->in_execve = 1;
1362
1363         file = open_exec(filename);
1364         retval = PTR_ERR(file);
1365         if (IS_ERR(file))
1366                 goto out_unmark;
1367
1368         sched_exec();
1369
1370         bprm->file = file;
1371         bprm->filename = filename;
1372         bprm->interp = filename;
1373
1374         retval = bprm_mm_init(bprm);
1375         if (retval)
1376                 goto out_file;
1377
1378         bprm->argc = count(argv, MAX_ARG_STRINGS);
1379         if ((retval = bprm->argc) < 0)
1380                 goto out;
1381
1382         bprm->envc = count(envp, MAX_ARG_STRINGS);
1383         if ((retval = bprm->envc) < 0)
1384                 goto out;
1385
1386         retval = prepare_binprm(bprm);
1387         if (retval < 0)
1388                 goto out;
1389
1390         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1391         if (retval < 0)
1392                 goto out;
1393
1394         bprm->exec = bprm->p;
1395         retval = copy_strings(bprm->envc, envp, bprm);
1396         if (retval < 0)
1397                 goto out;
1398
1399         retval = copy_strings(bprm->argc, argv, bprm);
1400         if (retval < 0)
1401                 goto out;
1402
1403         current->flags &= ~PF_KTHREAD;
1404         retval = search_binary_handler(bprm,regs);
1405         if (retval < 0)
1406                 goto out;
1407
1408         /* execve succeeded */
1409         current->fs->in_exec = 0;
1410         current->in_execve = 0;
1411         acct_update_integrals(current);
1412         free_bprm(bprm);
1413         if (displaced)
1414                 put_files_struct(displaced);
1415         return retval;
1416
1417 out:
1418         if (bprm->mm)
1419                 mmput (bprm->mm);
1420
1421 out_file:
1422         if (bprm->file) {
1423                 allow_write_access(bprm->file);
1424                 fput(bprm->file);
1425         }
1426
1427 out_unmark:
1428         if (clear_in_exec)
1429                 current->fs->in_exec = 0;
1430         current->in_execve = 0;
1431
1432 out_free:
1433         free_bprm(bprm);
1434
1435 out_files:
1436         if (displaced)
1437                 reset_files_struct(displaced);
1438 out_ret:
1439         return retval;
1440 }
1441
1442 void set_binfmt(struct linux_binfmt *new)
1443 {
1444         struct mm_struct *mm = current->mm;
1445
1446         if (mm->binfmt)
1447                 module_put(mm->binfmt->module);
1448
1449         mm->binfmt = new;
1450         if (new)
1451                 __module_get(new->module);
1452 }
1453
1454 EXPORT_SYMBOL(set_binfmt);
1455
1456 /* format_corename will inspect the pattern parameter, and output a
1457  * name into corename, which must have space for at least
1458  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1459  */
1460 static int format_corename(char *corename, long signr)
1461 {
1462         const struct cred *cred = current_cred();
1463         const char *pat_ptr = core_pattern;
1464         int ispipe = (*pat_ptr == '|');
1465         char *out_ptr = corename;
1466         char *const out_end = corename + CORENAME_MAX_SIZE;
1467         int rc;
1468         int pid_in_pattern = 0;
1469
1470         /* Repeat as long as we have more pattern to process and more output
1471            space */
1472         while (*pat_ptr) {
1473                 if (*pat_ptr != '%') {
1474                         if (out_ptr == out_end)
1475                                 goto out;
1476                         *out_ptr++ = *pat_ptr++;
1477                 } else {
1478                         switch (*++pat_ptr) {
1479                         case 0:
1480                                 goto out;
1481                         /* Double percent, output one percent */
1482                         case '%':
1483                                 if (out_ptr == out_end)
1484                                         goto out;
1485                                 *out_ptr++ = '%';
1486                                 break;
1487                         /* pid */
1488                         case 'p':
1489                                 pid_in_pattern = 1;
1490                                 rc = snprintf(out_ptr, out_end - out_ptr,
1491                                               "%d", task_tgid_vnr(current));
1492                                 if (rc > out_end - out_ptr)
1493                                         goto out;
1494                                 out_ptr += rc;
1495                                 break;
1496                         /* uid */
1497                         case 'u':
1498                                 rc = snprintf(out_ptr, out_end - out_ptr,
1499                                               "%d", cred->uid);
1500                                 if (rc > out_end - out_ptr)
1501                                         goto out;
1502                                 out_ptr += rc;
1503                                 break;
1504                         /* gid */
1505                         case 'g':
1506                                 rc = snprintf(out_ptr, out_end - out_ptr,
1507                                               "%d", cred->gid);
1508                                 if (rc > out_end - out_ptr)
1509                                         goto out;
1510                                 out_ptr += rc;
1511                                 break;
1512                         /* signal that caused the coredump */
1513                         case 's':
1514                                 rc = snprintf(out_ptr, out_end - out_ptr,
1515                                               "%ld", signr);
1516                                 if (rc > out_end - out_ptr)
1517                                         goto out;
1518                                 out_ptr += rc;
1519                                 break;
1520                         /* UNIX time of coredump */
1521                         case 't': {
1522                                 struct timeval tv;
1523                                 do_gettimeofday(&tv);
1524                                 rc = snprintf(out_ptr, out_end - out_ptr,
1525                                               "%lu", tv.tv_sec);
1526                                 if (rc > out_end - out_ptr)
1527                                         goto out;
1528                                 out_ptr += rc;
1529                                 break;
1530                         }
1531                         /* hostname */
1532                         case 'h':
1533                                 down_read(&uts_sem);
1534                                 rc = snprintf(out_ptr, out_end - out_ptr,
1535                                               "%s", utsname()->nodename);
1536                                 up_read(&uts_sem);
1537                                 if (rc > out_end - out_ptr)
1538                                         goto out;
1539                                 out_ptr += rc;
1540                                 break;
1541                         /* executable */
1542                         case 'e':
1543                                 rc = snprintf(out_ptr, out_end - out_ptr,
1544                                               "%s", current->comm);
1545                                 if (rc > out_end - out_ptr)
1546                                         goto out;
1547                                 out_ptr += rc;
1548                                 break;
1549                         /* core limit size */
1550                         case 'c':
1551                                 rc = snprintf(out_ptr, out_end - out_ptr,
1552                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1553                                 if (rc > out_end - out_ptr)
1554                                         goto out;
1555                                 out_ptr += rc;
1556                                 break;
1557                         default:
1558                                 break;
1559                         }
1560                         ++pat_ptr;
1561                 }
1562         }
1563         /* Backward compatibility with core_uses_pid:
1564          *
1565          * If core_pattern does not include a %p (as is the default)
1566          * and core_uses_pid is set, then .%pid will be appended to
1567          * the filename. Do not do this for piped commands. */
1568         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1569                 rc = snprintf(out_ptr, out_end - out_ptr,
1570                               ".%d", task_tgid_vnr(current));
1571                 if (rc > out_end - out_ptr)
1572                         goto out;
1573                 out_ptr += rc;
1574         }
1575 out:
1576         *out_ptr = 0;
1577         return ispipe;
1578 }
1579
1580 static int zap_process(struct task_struct *start)
1581 {
1582         struct task_struct *t;
1583         int nr = 0;
1584
1585         start->signal->flags = SIGNAL_GROUP_EXIT;
1586         start->signal->group_stop_count = 0;
1587
1588         t = start;
1589         do {
1590                 if (t != current && t->mm) {
1591                         sigaddset(&t->pending.signal, SIGKILL);
1592                         signal_wake_up(t, 1);
1593                         nr++;
1594                 }
1595         } while_each_thread(start, t);
1596
1597         return nr;
1598 }
1599
1600 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1601                                 struct core_state *core_state, int exit_code)
1602 {
1603         struct task_struct *g, *p;
1604         unsigned long flags;
1605         int nr = -EAGAIN;
1606
1607         spin_lock_irq(&tsk->sighand->siglock);
1608         if (!signal_group_exit(tsk->signal)) {
1609                 mm->core_state = core_state;
1610                 tsk->signal->group_exit_code = exit_code;
1611                 nr = zap_process(tsk);
1612         }
1613         spin_unlock_irq(&tsk->sighand->siglock);
1614         if (unlikely(nr < 0))
1615                 return nr;
1616
1617         if (atomic_read(&mm->mm_users) == nr + 1)
1618                 goto done;
1619         /*
1620          * We should find and kill all tasks which use this mm, and we should
1621          * count them correctly into ->nr_threads. We don't take tasklist
1622          * lock, but this is safe wrt:
1623          *
1624          * fork:
1625          *      None of sub-threads can fork after zap_process(leader). All
1626          *      processes which were created before this point should be
1627          *      visible to zap_threads() because copy_process() adds the new
1628          *      process to the tail of init_task.tasks list, and lock/unlock
1629          *      of ->siglock provides a memory barrier.
1630          *
1631          * do_exit:
1632          *      The caller holds mm->mmap_sem. This means that the task which
1633          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1634          *      its ->mm.
1635          *
1636          * de_thread:
1637          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1638          *      we must see either old or new leader, this does not matter.
1639          *      However, it can change p->sighand, so lock_task_sighand(p)
1640          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1641          *      it can't fail.
1642          *
1643          *      Note also that "g" can be the old leader with ->mm == NULL
1644          *      and already unhashed and thus removed from ->thread_group.
1645          *      This is OK, __unhash_process()->list_del_rcu() does not
1646          *      clear the ->next pointer, we will find the new leader via
1647          *      next_thread().
1648          */
1649         rcu_read_lock();
1650         for_each_process(g) {
1651                 if (g == tsk->group_leader)
1652                         continue;
1653                 if (g->flags & PF_KTHREAD)
1654                         continue;
1655                 p = g;
1656                 do {
1657                         if (p->mm) {
1658                                 if (unlikely(p->mm == mm)) {
1659                                         lock_task_sighand(p, &flags);
1660                                         nr += zap_process(p);
1661                                         unlock_task_sighand(p, &flags);
1662                                 }
1663                                 break;
1664                         }
1665                 } while_each_thread(g, p);
1666         }
1667         rcu_read_unlock();
1668 done:
1669         atomic_set(&core_state->nr_threads, nr);
1670         return nr;
1671 }
1672
1673 static int coredump_wait(int exit_code, struct core_state *core_state)
1674 {
1675         struct task_struct *tsk = current;
1676         struct mm_struct *mm = tsk->mm;
1677         struct completion *vfork_done;
1678         int core_waiters;
1679
1680         init_completion(&core_state->startup);
1681         core_state->dumper.task = tsk;
1682         core_state->dumper.next = NULL;
1683         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1684         up_write(&mm->mmap_sem);
1685
1686         if (unlikely(core_waiters < 0))
1687                 goto fail;
1688
1689         /*
1690          * Make sure nobody is waiting for us to release the VM,
1691          * otherwise we can deadlock when we wait on each other
1692          */
1693         vfork_done = tsk->vfork_done;
1694         if (vfork_done) {
1695                 tsk->vfork_done = NULL;
1696                 complete(vfork_done);
1697         }
1698
1699         if (core_waiters)
1700                 wait_for_completion(&core_state->startup);
1701 fail:
1702         return core_waiters;
1703 }
1704
1705 static void coredump_finish(struct mm_struct *mm)
1706 {
1707         struct core_thread *curr, *next;
1708         struct task_struct *task;
1709
1710         next = mm->core_state->dumper.next;
1711         while ((curr = next) != NULL) {
1712                 next = curr->next;
1713                 task = curr->task;
1714                 /*
1715                  * see exit_mm(), curr->task must not see
1716                  * ->task == NULL before we read ->next.
1717                  */
1718                 smp_mb();
1719                 curr->task = NULL;
1720                 wake_up_process(task);
1721         }
1722
1723         mm->core_state = NULL;
1724 }
1725
1726 /*
1727  * set_dumpable converts traditional three-value dumpable to two flags and
1728  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1729  * these bits are not changed atomically.  So get_dumpable can observe the
1730  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1731  * return either old dumpable or new one by paying attention to the order of
1732  * modifying the bits.
1733  *
1734  * dumpable |   mm->flags (binary)
1735  * old  new | initial interim  final
1736  * ---------+-----------------------
1737  *  0    1  |   00      01      01
1738  *  0    2  |   00      10(*)   11
1739  *  1    0  |   01      00      00
1740  *  1    2  |   01      11      11
1741  *  2    0  |   11      10(*)   00
1742  *  2    1  |   11      11      01
1743  *
1744  * (*) get_dumpable regards interim value of 10 as 11.
1745  */
1746 void set_dumpable(struct mm_struct *mm, int value)
1747 {
1748         switch (value) {
1749         case 0:
1750                 clear_bit(MMF_DUMPABLE, &mm->flags);
1751                 smp_wmb();
1752                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1753                 break;
1754         case 1:
1755                 set_bit(MMF_DUMPABLE, &mm->flags);
1756                 smp_wmb();
1757                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1758                 break;
1759         case 2:
1760                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1761                 smp_wmb();
1762                 set_bit(MMF_DUMPABLE, &mm->flags);
1763                 break;
1764         }
1765 }
1766
1767 int get_dumpable(struct mm_struct *mm)
1768 {
1769         int ret;
1770
1771         ret = mm->flags & 0x3;
1772         return (ret >= 2) ? 2 : ret;
1773 }
1774
1775 static void wait_for_dump_helpers(struct file *file)
1776 {
1777         struct pipe_inode_info *pipe;
1778
1779         pipe = file->f_path.dentry->d_inode->i_pipe;
1780
1781         pipe_lock(pipe);
1782         pipe->readers++;
1783         pipe->writers--;
1784
1785         while ((pipe->readers > 1) && (!signal_pending(current))) {
1786                 wake_up_interruptible_sync(&pipe->wait);
1787                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1788                 pipe_wait(pipe);
1789         }
1790
1791         pipe->readers--;
1792         pipe->writers++;
1793         pipe_unlock(pipe);
1794
1795 }
1796
1797
1798 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1799 {
1800         struct core_state core_state;
1801         char corename[CORENAME_MAX_SIZE + 1];
1802         struct mm_struct *mm = current->mm;
1803         struct linux_binfmt * binfmt;
1804         struct inode * inode;
1805         const struct cred *old_cred;
1806         struct cred *cred;
1807         int retval = 0;
1808         int flag = 0;
1809         int ispipe = 0;
1810         char **helper_argv = NULL;
1811         int helper_argc = 0;
1812         int dump_count = 0;
1813         static atomic_t core_dump_count = ATOMIC_INIT(0);
1814         struct coredump_params cprm = {
1815                 .signr = signr,
1816                 .regs = regs,
1817                 .limit = current->signal->rlim[RLIMIT_CORE].rlim_cur,
1818         };
1819
1820         audit_core_dumps(signr);
1821
1822         binfmt = mm->binfmt;
1823         if (!binfmt || !binfmt->core_dump)
1824                 goto fail;
1825
1826         cred = prepare_creds();
1827         if (!cred) {
1828                 retval = -ENOMEM;
1829                 goto fail;
1830         }
1831
1832         down_write(&mm->mmap_sem);
1833         /*
1834          * If another thread got here first, or we are not dumpable, bail out.
1835          */
1836         if (mm->core_state || !get_dumpable(mm)) {
1837                 up_write(&mm->mmap_sem);
1838                 put_cred(cred);
1839                 goto fail;
1840         }
1841
1842         /*
1843          *      We cannot trust fsuid as being the "true" uid of the
1844          *      process nor do we know its entire history. We only know it
1845          *      was tainted so we dump it as root in mode 2.
1846          */
1847         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1848                 flag = O_EXCL;          /* Stop rewrite attacks */
1849                 cred->fsuid = 0;        /* Dump root private */
1850         }
1851
1852         retval = coredump_wait(exit_code, &core_state);
1853         if (retval < 0) {
1854                 put_cred(cred);
1855                 goto fail;
1856         }
1857
1858         old_cred = override_creds(cred);
1859
1860         /*
1861          * Clear any false indication of pending signals that might
1862          * be seen by the filesystem code called to write the core file.
1863          */
1864         clear_thread_flag(TIF_SIGPENDING);
1865
1866         /*
1867          * lock_kernel() because format_corename() is controlled by sysctl, which
1868          * uses lock_kernel()
1869          */
1870         lock_kernel();
1871         ispipe = format_corename(corename, signr);
1872         unlock_kernel();
1873
1874         if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
1875                 goto fail_unlock;
1876
1877         if (ispipe) {
1878                 if (cprm.limit == 0) {
1879                         /*
1880                          * Normally core limits are irrelevant to pipes, since
1881                          * we're not writing to the file system, but we use
1882                          * cprm.limit of 0 here as a speacial value. Any
1883                          * non-zero limit gets set to RLIM_INFINITY below, but
1884                          * a limit of 0 skips the dump.  This is a consistent
1885                          * way to catch recursive crashes.  We can still crash
1886                          * if the core_pattern binary sets RLIM_CORE =  !0
1887                          * but it runs as root, and can do lots of stupid things
1888                          * Note that we use task_tgid_vnr here to grab the pid
1889                          * of the process group leader.  That way we get the
1890                          * right pid if a thread in a multi-threaded
1891                          * core_pattern process dies.
1892                          */
1893                         printk(KERN_WARNING
1894                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1895                                 task_tgid_vnr(current), current->comm);
1896                         printk(KERN_WARNING "Aborting core\n");
1897                         goto fail_unlock;
1898                 }
1899
1900                 dump_count = atomic_inc_return(&core_dump_count);
1901                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1902                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1903                                task_tgid_vnr(current), current->comm);
1904                         printk(KERN_WARNING "Skipping core dump\n");
1905                         goto fail_dropcount;
1906                 }
1907
1908                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1909                 if (!helper_argv) {
1910                         printk(KERN_WARNING "%s failed to allocate memory\n",
1911                                __func__);
1912                         goto fail_dropcount;
1913                 }
1914
1915                 cprm.limit = RLIM_INFINITY;
1916
1917                 /* SIGPIPE can happen, but it's just never processed */
1918                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1919                                 &cprm.file)) {
1920                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1921                                corename);
1922                         goto fail_dropcount;
1923                 }
1924         } else
1925                 cprm.file = filp_open(corename,
1926                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1927                                  0600);
1928         if (IS_ERR(cprm.file))
1929                 goto fail_dropcount;
1930         inode = cprm.file->f_path.dentry->d_inode;
1931         if (inode->i_nlink > 1)
1932                 goto close_fail;        /* multiple links - don't dump */
1933         if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
1934                 goto close_fail;
1935
1936         /* AK: actually i see no reason to not allow this for named pipes etc.,
1937            but keep the previous behaviour for now. */
1938         if (!ispipe && !S_ISREG(inode->i_mode))
1939                 goto close_fail;
1940         /*
1941          * Dont allow local users get cute and trick others to coredump
1942          * into their pre-created files:
1943          * Note, this is not relevant for pipes
1944          */
1945         if (!ispipe && (inode->i_uid != current_fsuid()))
1946                 goto close_fail;
1947         if (!cprm.file->f_op)
1948                 goto close_fail;
1949         if (!cprm.file->f_op->write)
1950                 goto close_fail;
1951         if (!ispipe &&
1952             do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
1953                 goto close_fail;
1954
1955         retval = binfmt->core_dump(&cprm);
1956
1957         if (retval)
1958                 current->signal->group_exit_code |= 0x80;
1959 close_fail:
1960         if (ispipe && core_pipe_limit)
1961                 wait_for_dump_helpers(cprm.file);
1962         filp_close(cprm.file, NULL);
1963 fail_dropcount:
1964         if (dump_count)
1965                 atomic_dec(&core_dump_count);
1966 fail_unlock:
1967         if (helper_argv)
1968                 argv_free(helper_argv);
1969
1970         revert_creds(old_cred);
1971         put_cred(cred);
1972         coredump_finish(mm);
1973 fail:
1974         return;
1975 }