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