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/vmacache.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.h>
  58 #include <linux/compat.h>
  59 #include <linux/vmalloc.h>
  60
  61 #include <asm/uaccess.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/tlb.h>
  64
  65 #include <trace/events/task.h>
  66 #include "internal.h"
  67
  68 #include <trace/events/sched.h>
  69
  70 int suid_dumpable = 0;
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  76 {
  77         BUG_ON(!fmt);
  78         if (WARN_ON(!fmt->load_binary))
  79                 return;
  80         write_lock(&binfmt_lock);
  81         insert ? list_add(&fmt->lh, &formats) :
  82                  list_add_tail(&fmt->lh, &formats);
  83         write_unlock(&binfmt_lock);
  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 bool path_noexec(const struct path *path)
 103 {
 104         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 105                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 106 }
 107
 108 #ifdef CONFIG_USELIB
 109 /*
 110  * Note that a shared library must be both readable and executable due to
 111  * security reasons.
 112  *
 113  * Also note that we take the address to load from from the file itself.
 114  */
 115 SYSCALL_DEFINE1(uselib, const char __user *, library)
 116 {
 117         struct linux_binfmt *fmt;
 118         struct file *file;
 119         struct filename *tmp = getname(library);
 120         int error = PTR_ERR(tmp);
 121         static const struct open_flags uselib_flags = {
 122                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 123                 .acc_mode = MAY_READ | MAY_EXEC,
 124                 .intent = LOOKUP_OPEN,
 125                 .lookup_flags = LOOKUP_FOLLOW,
 126         };
 127
 128         if (IS_ERR(tmp))
 129                 goto out;
 130
 131         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 132         putname(tmp);
 133         error = PTR_ERR(file);
 134         if (IS_ERR(file))
 135                 goto out;
 136
 137         error = -EINVAL;
 138         if (!S_ISREG(file_inode(file)->i_mode))
 139                 goto exit;
 140
 141         error = -EACCES;
 142         if (path_noexec(&file->f_path))
 143                 goto exit;
 144
 145         fsnotify_open(file);
 146
 147         error = -ENOEXEC;
 148
 149         read_lock(&binfmt_lock);
 150         list_for_each_entry(fmt, &formats, lh) {
 151                 if (!fmt->load_shlib)
 152                         continue;
 153                 if (!try_module_get(fmt->module))
 154                         continue;
 155                 read_unlock(&binfmt_lock);
 156                 error = fmt->load_shlib(file);
 157                 read_lock(&binfmt_lock);
 158                 put_binfmt(fmt);
 159                 if (error != -ENOEXEC)
 160                         break;
 161         }
 162         read_unlock(&binfmt_lock);
 163 exit:
 164         fput(file);
 165 out:
 166         return error;
 167 }
 168 #endif /* #ifdef CONFIG_USELIB */
 169
 170 #ifdef CONFIG_MMU
 171 /*
 172  * The nascent bprm->mm is not visible until exec_mmap() but it can
 173  * use a lot of memory, account these pages in current->mm temporary
 174  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 175  * change the counter back via acct_arg_size(0).
 176  */
 177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 178 {
 179         struct mm_struct *mm = current->mm;
 180         long diff = (long)(pages - bprm->vma_pages);
 181
 182         if (!mm || !diff)
 183                 return;
 184
 185         bprm->vma_pages = pages;
 186         add_mm_counter(mm, MM_ANONPAGES, diff);
 187 }
 188
 189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 190                 int write)
 191 {
 192         struct page *page;
 193         int ret;
 194
 195 #ifdef CONFIG_STACK_GROWSUP
 196         if (write) {
 197                 ret = expand_downwards(bprm->vma, pos);
 198                 if (ret < 0)
 199                         return NULL;
 200         }
 201 #endif
 202         /*
 203          * We are doing an exec().  'current' is the process
 204          * doing the exec and bprm->mm is the new process's mm.
 205          */
 206         ret = get_user_pages_remote(current, bprm->mm, pos, 1, write,
 207                         1, &page, NULL);
 208         if (ret <= 0)
 209                 return NULL;
 210
 211         if (write) {
 212                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 213                 struct rlimit *rlim;
 214
 215                 acct_arg_size(bprm, size / PAGE_SIZE);
 216
 217                 /*
 218                  * We've historically supported up to 32 pages (ARG_MAX)
 219                  * of argument strings even with small stacks
 220                  */
 221                 if (size <= ARG_MAX)
 222                         return page;
 223
 224                 /*
 225                  * Limit to 1/4-th the stack size for the argv+env strings.
 226                  * This ensures that:
 227                  *  - the remaining binfmt code will not run out of stack space,
 228                  *  - the program will have a reasonable amount of stack left
 229                  *    to work from.
 230                  */
 231                 rlim = current->signal->rlim;
 232                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 233                         put_page(page);
 234                         return NULL;
 235                 }
 236         }
 237
 238         return page;
 239 }
 240
 241 static void put_arg_page(struct page *page)
 242 {
 243         put_page(page);
 244 }
 245
 246 static void free_arg_pages(struct linux_binprm *bprm)
 247 {
 248 }
 249
 250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 251                 struct page *page)
 252 {
 253         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 254 }
 255
 256 static int __bprm_mm_init(struct linux_binprm *bprm)
 257 {
 258         int err;
 259         struct vm_area_struct *vma = NULL;
 260         struct mm_struct *mm = bprm->mm;
 261
 262         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 263         if (!vma)
 264                 return -ENOMEM;
 265
 266         if (down_write_killable(&mm->mmap_sem)) {
 267                 err = -EINTR;
 268                 goto err_free;
 269         }
 270         vma->vm_mm = mm;
 271
 272         /*
 273          * Place the stack at the largest stack address the architecture
 274          * supports. Later, we'll move this to an appropriate place. We don't
 275          * use STACK_TOP because that can depend on attributes which aren't
 276          * configured yet.
 277          */
 278         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 279         vma->vm_end = STACK_TOP_MAX;
 280         vma->vm_start = vma->vm_end - PAGE_SIZE;
 281         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 282         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 283         INIT_LIST_HEAD(&vma->anon_vma_chain);
 284
 285         err = insert_vm_struct(mm, vma);
 286         if (err)
 287                 goto err;
 288
 289         mm->stack_vm = mm->total_vm = 1;
 290         arch_bprm_mm_init(mm, vma);
 291         up_write(&mm->mmap_sem);
 292         bprm->p = vma->vm_end - sizeof(void *);
 293         return 0;
 294 err:
 295         up_write(&mm->mmap_sem);
 296 err_free:
 297         bprm->vma = NULL;
 298         kmem_cache_free(vm_area_cachep, vma);
 299         return err;
 300 }
 301
 302 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 303 {
 304         return len <= MAX_ARG_STRLEN;
 305 }
 306
 307 #else
 308
 309 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 310 {
 311 }
 312
 313 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 314                 int write)
 315 {
 316         struct page *page;
 317
 318         page = bprm->page[pos / PAGE_SIZE];
 319         if (!page && write) {
 320                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 321                 if (!page)
 322                         return NULL;
 323                 bprm->page[pos / PAGE_SIZE] = page;
 324         }
 325
 326         return page;
 327 }
 328
 329 static void put_arg_page(struct page *page)
 330 {
 331 }
 332
 333 static void free_arg_page(struct linux_binprm *bprm, int i)
 334 {
 335         if (bprm->page[i]) {
 336                 __free_page(bprm->page[i]);
 337                 bprm->page[i] = NULL;
 338         }
 339 }
 340
 341 static void free_arg_pages(struct linux_binprm *bprm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < MAX_ARG_PAGES; i++)
 346                 free_arg_page(bprm, i);
 347 }
 348
 349 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 350                 struct page *page)
 351 {
 352 }
 353
 354 static int __bprm_mm_init(struct linux_binprm *bprm)
 355 {
 356         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 357         return 0;
 358 }
 359
 360 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 361 {
 362         return len <= bprm->p;
 363 }
 364
 365 #endif /* CONFIG_MMU */
 366
 367 /*
 368  * Create a new mm_struct and populate it with a temporary stack
 369  * vm_area_struct.  We don't have enough context at this point to set the stack
 370  * flags, permissions, and offset, so we use temporary values.  We'll update
 371  * them later in setup_arg_pages().
 372  */
 373 static int bprm_mm_init(struct linux_binprm *bprm)
 374 {
 375         int err;
 376         struct mm_struct *mm = NULL;
 377
 378         bprm->mm = mm = mm_alloc();
 379         err = -ENOMEM;
 380         if (!mm)
 381                 goto err;
 382
 383         err = __bprm_mm_init(bprm);
 384         if (err)
 385                 goto err;
 386
 387         return 0;
 388
 389 err:
 390         if (mm) {
 391                 bprm->mm = NULL;
 392                 mmdrop(mm);
 393         }
 394
 395         return err;
 396 }
 397
 398 struct user_arg_ptr {
 399 #ifdef CONFIG_COMPAT
 400         bool is_compat;
 401 #endif
 402         union {
 403                 const char __user *const __user *native;
 404 #ifdef CONFIG_COMPAT
 405                 const compat_uptr_t __user *compat;
 406 #endif
 407         } ptr;
 408 };
 409
 410 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 411 {
 412         const char __user *native;
 413
 414 #ifdef CONFIG_COMPAT
 415         if (unlikely(argv.is_compat)) {
 416                 compat_uptr_t compat;
 417
 418                 if (get_user(compat, argv.ptr.compat + nr))
 419                         return ERR_PTR(-EFAULT);
 420
 421                 return compat_ptr(compat);
 422         }
 423 #endif
 424
 425         if (get_user(native, argv.ptr.native + nr))
 426                 return ERR_PTR(-EFAULT);
 427
 428         return native;
 429 }
 430
 431 /*
 432  * count() counts the number of strings in array ARGV.
 433  */
 434 static int count(struct user_arg_ptr argv, int max)
 435 {
 436         int i = 0;
 437
 438         if (argv.ptr.native != NULL) {
 439                 for (;;) {
 440                         const char __user *p = get_user_arg_ptr(argv, i);
 441
 442                         if (!p)
 443                                 break;
 444
 445                         if (IS_ERR(p))
 446                                 return -EFAULT;
 447
 448                         if (i >= max)
 449                                 return -E2BIG;
 450                         ++i;
 451
 452                         if (fatal_signal_pending(current))
 453                                 return -ERESTARTNOHAND;
 454                         cond_resched();
 455                 }
 456         }
 457         return i;
 458 }
 459
 460 /*
 461  * 'copy_strings()' copies argument/environment strings from the old
 462  * processes's memory to the new process's stack.  The call to get_user_pages()
 463  * ensures the destination page is created and not swapped out.
 464  */
 465 static int copy_strings(int argc, struct user_arg_ptr argv,
 466                         struct linux_binprm *bprm)
 467 {
 468         struct page *kmapped_page = NULL;
 469         char *kaddr = NULL;
 470         unsigned long kpos = 0;
 471         int ret;
 472
 473         while (argc-- > 0) {
 474                 const char __user *str;
 475                 int len;
 476                 unsigned long pos;
 477
 478                 ret = -EFAULT;
 479                 str = get_user_arg_ptr(argv, argc);
 480                 if (IS_ERR(str))
 481                         goto out;
 482
 483                 len = strnlen_user(str, MAX_ARG_STRLEN);
 484                 if (!len)
 485                         goto out;
 486
 487                 ret = -E2BIG;
 488                 if (!valid_arg_len(bprm, len))
 489                         goto out;
 490
 491                 /* We're going to work our way backwords. */
 492                 pos = bprm->p;
 493                 str += len;
 494                 bprm->p -= len;
 495
 496                 while (len > 0) {
 497                         int offset, bytes_to_copy;
 498
 499                         if (fatal_signal_pending(current)) {
 500                                 ret = -ERESTARTNOHAND;
 501                                 goto out;
 502                         }
 503                         cond_resched();
 504
 505                         offset = pos % PAGE_SIZE;
 506                         if (offset == 0)
 507                                 offset = PAGE_SIZE;
 508
 509                         bytes_to_copy = offset;
 510                         if (bytes_to_copy > len)
 511                                 bytes_to_copy = len;
 512
 513                         offset -= bytes_to_copy;
 514                         pos -= bytes_to_copy;
 515                         str -= bytes_to_copy;
 516                         len -= bytes_to_copy;
 517
 518                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 519                                 struct page *page;
 520
 521                                 page = get_arg_page(bprm, pos, 1);
 522                                 if (!page) {
 523                                         ret = -E2BIG;
 524                                         goto out;
 525                                 }
 526
 527                                 if (kmapped_page) {
 528                                         flush_kernel_dcache_page(kmapped_page);
 529                                         kunmap(kmapped_page);
 530                                         put_arg_page(kmapped_page);
 531                                 }
 532                                 kmapped_page = page;
 533                                 kaddr = kmap(kmapped_page);
 534                                 kpos = pos & PAGE_MASK;
 535                                 flush_arg_page(bprm, kpos, kmapped_page);
 536                         }
 537                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 538                                 ret = -EFAULT;
 539                                 goto out;
 540                         }
 541                 }
 542         }
 543         ret = 0;
 544 out:
 545         if (kmapped_page) {
 546                 flush_kernel_dcache_page(kmapped_page);
 547                 kunmap(kmapped_page);
 548                 put_arg_page(kmapped_page);
 549         }
 550         return ret;
 551 }
 552
 553 /*
 554  * Like copy_strings, but get argv and its values from kernel memory.
 555  */
 556 int copy_strings_kernel(int argc, const char *const *__argv,
 557                         struct linux_binprm *bprm)
 558 {
 559         int r;
 560         mm_segment_t oldfs = get_fs();
 561         struct user_arg_ptr argv = {
 562                 .ptr.native = (const char __user *const  __user *)__argv,
 563         };
 564
 565         set_fs(KERNEL_DS);
 566         r = copy_strings(argc, argv, bprm);
 567         set_fs(oldfs);
 568
 569         return r;
 570 }
 571 EXPORT_SYMBOL(copy_strings_kernel);
 572
 573 #ifdef CONFIG_MMU
 574
 575 /*
 576  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 577  * the binfmt code determines where the new stack should reside, we shift it to
 578  * its final location.  The process proceeds as follows:
 579  *
 580  * 1) Use shift to calculate the new vma endpoints.
 581  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 582  *    arguments passed to subsequent functions are consistent.
 583  * 3) Move vma's page tables to the new range.
 584  * 4) Free up any cleared pgd range.
 585  * 5) Shrink the vma to cover only the new range.
 586  */
 587 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 588 {
 589         struct mm_struct *mm = vma->vm_mm;
 590         unsigned long old_start = vma->vm_start;
 591         unsigned long old_end = vma->vm_end;
 592         unsigned long length = old_end - old_start;
 593         unsigned long new_start = old_start - shift;
 594         unsigned long new_end = old_end - shift;
 595         struct mmu_gather tlb;
 596
 597         BUG_ON(new_start > new_end);
 598
 599         /*
 600          * ensure there are no vmas between where we want to go
 601          * and where we are
 602          */
 603         if (vma != find_vma(mm, new_start))
 604                 return -EFAULT;
 605
 606         /*
 607          * cover the whole range: [new_start, old_end)
 608          */
 609         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 610                 return -ENOMEM;
 611
 612         /*
 613          * move the page tables downwards, on failure we rely on
 614          * process cleanup to remove whatever mess we made.
 615          */
 616         if (length != move_page_tables(vma, old_start,
 617                                        vma, new_start, length, false))
 618                 return -ENOMEM;
 619
 620         lru_add_drain();
 621         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 622         if (new_end > old_start) {
 623                 /*
 624                  * when the old and new regions overlap clear from new_end.
 625                  */
 626                 free_pgd_range(&tlb, new_end, old_end, new_end,
 627                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 628         } else {
 629                 /*
 630                  * otherwise, clean from old_start; this is done to not touch
 631                  * the address space in [new_end, old_start) some architectures
 632                  * have constraints on va-space that make this illegal (IA64) -
 633                  * for the others its just a little faster.
 634                  */
 635                 free_pgd_range(&tlb, old_start, old_end, new_end,
 636                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 637         }
 638         tlb_finish_mmu(&tlb, old_start, old_end);
 639
 640         /*
 641          * Shrink the vma to just the new range.  Always succeeds.
 642          */
 643         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 644
 645         return 0;
 646 }
 647
 648 /*
 649  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 650  * the stack is optionally relocated, and some extra space is added.
 651  */
 652 int setup_arg_pages(struct linux_binprm *bprm,
 653                     unsigned long stack_top,
 654                     int executable_stack)
 655 {
 656         unsigned long ret;
 657         unsigned long stack_shift;
 658         struct mm_struct *mm = current->mm;
 659         struct vm_area_struct *vma = bprm->vma;
 660         struct vm_area_struct *prev = NULL;
 661         unsigned long vm_flags;
 662         unsigned long stack_base;
 663         unsigned long stack_size;
 664         unsigned long stack_expand;
 665         unsigned long rlim_stack;
 666
 667 #ifdef CONFIG_STACK_GROWSUP
 668         /* Limit stack size */
 669         stack_base = rlimit_max(RLIMIT_STACK);
 670         if (stack_base > STACK_SIZE_MAX)
 671                 stack_base = STACK_SIZE_MAX;
 672
 673         /* Add space for stack randomization. */
 674         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 675
 676         /* Make sure we didn't let the argument array grow too large. */
 677         if (vma->vm_end - vma->vm_start > stack_base)
 678                 return -ENOMEM;
 679
 680         stack_base = PAGE_ALIGN(stack_top - stack_base);
 681
 682         stack_shift = vma->vm_start - stack_base;
 683         mm->arg_start = bprm->p - stack_shift;
 684         bprm->p = vma->vm_end - stack_shift;
 685 #else
 686         stack_top = arch_align_stack(stack_top);
 687         stack_top = PAGE_ALIGN(stack_top);
 688
 689         if (unlikely(stack_top < mmap_min_addr) ||
 690             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 691                 return -ENOMEM;
 692
 693         stack_shift = vma->vm_end - stack_top;
 694
 695         bprm->p -= stack_shift;
 696         mm->arg_start = bprm->p;
 697 #endif
 698
 699         if (bprm->loader)
 700                 bprm->loader -= stack_shift;
 701         bprm->exec -= stack_shift;
 702
 703         if (down_write_killable(&mm->mmap_sem))
 704                 return -EINTR;
 705
 706         vm_flags = VM_STACK_FLAGS;
 707
 708         /*
 709          * Adjust stack execute permissions; explicitly enable for
 710          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 711          * (arch default) otherwise.
 712          */
 713         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 714                 vm_flags |= VM_EXEC;
 715         else if (executable_stack == EXSTACK_DISABLE_X)
 716                 vm_flags &= ~VM_EXEC;
 717         vm_flags |= mm->def_flags;
 718         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 719
 720         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 721                         vm_flags);
 722         if (ret)
 723                 goto out_unlock;
 724         BUG_ON(prev != vma);
 725
 726         /* Move stack pages down in memory. */
 727         if (stack_shift) {
 728                 ret = shift_arg_pages(vma, stack_shift);
 729                 if (ret)
 730                         goto out_unlock;
 731         }
 732
 733         /* mprotect_fixup is overkill to remove the temporary stack flags */
 734         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 735
 736         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 737         stack_size = vma->vm_end - vma->vm_start;
 738         /*
 739          * Align this down to a page boundary as expand_stack
 740          * will align it up.
 741          */
 742         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 743 #ifdef CONFIG_STACK_GROWSUP
 744         if (stack_size + stack_expand > rlim_stack)
 745                 stack_base = vma->vm_start + rlim_stack;
 746         else
 747                 stack_base = vma->vm_end + stack_expand;
 748 #else
 749         if (stack_size + stack_expand > rlim_stack)
 750                 stack_base = vma->vm_end - rlim_stack;
 751         else
 752                 stack_base = vma->vm_start - stack_expand;
 753 #endif
 754         current->mm->start_stack = bprm->p;
 755         ret = expand_stack(vma, stack_base);
 756         if (ret)
 757                 ret = -EFAULT;
 758
 759 out_unlock:
 760         up_write(&mm->mmap_sem);
 761         return ret;
 762 }
 763 EXPORT_SYMBOL(setup_arg_pages);
 764
 765 #else
 766
 767 /*
 768  * Transfer the program arguments and environment from the holding pages
 769  * onto the stack. The provided stack pointer is adjusted accordingly.
 770  */
 771 int transfer_args_to_stack(struct linux_binprm *bprm,
 772                            unsigned long *sp_location)
 773 {
 774         unsigned long index, stop, sp;
 775         int ret = 0;
 776
 777         stop = bprm->p >> PAGE_SHIFT;
 778         sp = *sp_location;
 779
 780         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 781                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 782                 char *src = kmap(bprm->page[index]) + offset;
 783                 sp -= PAGE_SIZE - offset;
 784                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 785                         ret = -EFAULT;
 786                 kunmap(bprm->page[index]);
 787                 if (ret)
 788                         goto out;
 789         }
 790
 791         *sp_location = sp;
 792
 793 out:
 794         return ret;
 795 }
 796 EXPORT_SYMBOL(transfer_args_to_stack);
 797
 798 #endif /* CONFIG_MMU */
 799
 800 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 801 {
 802         struct file *file;
 803         int err;
 804         struct open_flags open_exec_flags = {
 805                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 806                 .acc_mode = MAY_EXEC,
 807                 .intent = LOOKUP_OPEN,
 808                 .lookup_flags = LOOKUP_FOLLOW,
 809         };
 810
 811         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 812                 return ERR_PTR(-EINVAL);
 813         if (flags & AT_SYMLINK_NOFOLLOW)
 814                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 815         if (flags & AT_EMPTY_PATH)
 816                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 817
 818         file = do_filp_open(fd, name, &open_exec_flags);
 819         if (IS_ERR(file))
 820                 goto out;
 821
 822         err = -EACCES;
 823         if (!S_ISREG(file_inode(file)->i_mode))
 824                 goto exit;
 825
 826         if (path_noexec(&file->f_path))
 827                 goto exit;
 828
 829         err = deny_write_access(file);
 830         if (err)
 831                 goto exit;
 832
 833         if (name->name[0] != '\0')
 834                 fsnotify_open(file);
 835
 836 out:
 837         return file;
 838
 839 exit:
 840         fput(file);
 841         return ERR_PTR(err);
 842 }
 843
 844 struct file *open_exec(const char *name)
 845 {
 846         struct filename *filename = getname_kernel(name);
 847         struct file *f = ERR_CAST(filename);
 848
 849         if (!IS_ERR(filename)) {
 850                 f = do_open_execat(AT_FDCWD, filename, 0);
 851                 putname(filename);
 852         }
 853         return f;
 854 }
 855 EXPORT_SYMBOL(open_exec);
 856
 857 int kernel_read(struct file *file, loff_t offset,
 858                 char *addr, unsigned long count)
 859 {
 860         mm_segment_t old_fs;
 861         loff_t pos = offset;
 862         int result;
 863
 864         old_fs = get_fs();
 865         set_fs(get_ds());
 866         /* The cast to a user pointer is valid due to the set_fs() */
 867         result = vfs_read(file, (void __user *)addr, count, &pos);
 868         set_fs(old_fs);
 869         return result;
 870 }
 871
 872 EXPORT_SYMBOL(kernel_read);
 873
 874 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 875                      loff_t max_size, enum kernel_read_file_id id)
 876 {
 877         loff_t i_size, pos;
 878         ssize_t bytes = 0;
 879         int ret;
 880
 881         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 882                 return -EINVAL;
 883
 884         ret = security_kernel_read_file(file, id);
 885         if (ret)
 886                 return ret;
 887
 888         ret = deny_write_access(file);
 889         if (ret)
 890                 return ret;
 891
 892         i_size = i_size_read(file_inode(file));
 893         if (max_size > 0 && i_size > max_size) {
 894                 ret = -EFBIG;
 895                 goto out;
 896         }
 897         if (i_size <= 0) {
 898                 ret = -EINVAL;
 899                 goto out;
 900         }
 901
 902         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 903                 *buf = vmalloc(i_size);
 904         if (!*buf) {
 905                 ret = -ENOMEM;
 906                 goto out;
 907         }
 908
 909         pos = 0;
 910         while (pos < i_size) {
 911                 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 912                                     i_size - pos);
 913                 if (bytes < 0) {
 914                         ret = bytes;
 915                         goto out;
 916                 }
 917
 918                 if (bytes == 0)
 919                         break;
 920                 pos += bytes;
 921         }
 922
 923         if (pos != i_size) {
 924                 ret = -EIO;
 925                 goto out_free;
 926         }
 927
 928         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 929         if (!ret)
 930                 *size = pos;
 931
 932 out_free:
 933         if (ret < 0) {
 934                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 935                         vfree(*buf);
 936                         *buf = NULL;
 937                 }
 938         }
 939
 940 out:
 941         allow_write_access(file);
 942         return ret;
 943 }
 944 EXPORT_SYMBOL_GPL(kernel_read_file);
 945
 946 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 947                                loff_t max_size, enum kernel_read_file_id id)
 948 {
 949         struct file *file;
 950         int ret;
 951
 952         if (!path || !*path)
 953                 return -EINVAL;
 954
 955         file = filp_open(path, O_RDONLY, 0);
 956         if (IS_ERR(file))
 957                 return PTR_ERR(file);
 958
 959         ret = kernel_read_file(file, buf, size, max_size, id);
 960         fput(file);
 961         return ret;
 962 }
 963 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 964
 965 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 966                              enum kernel_read_file_id id)
 967 {
 968         struct fd f = fdget(fd);
 969         int ret = -EBADF;
 970
 971         if (!f.file)
 972                 goto out;
 973
 974         ret = kernel_read_file(f.file, buf, size, max_size, id);
 975 out:
 976         fdput(f);
 977         return ret;
 978 }
 979 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 980
 981 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 982 {
 983         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 984         if (res > 0)
 985                 flush_icache_range(addr, addr + len);
 986         return res;
 987 }
 988 EXPORT_SYMBOL(read_code);
 989
 990 static int exec_mmap(struct mm_struct *mm)
 991 {
 992         struct task_struct *tsk;
 993         struct mm_struct *old_mm, *active_mm;
 994
 995         /* Notify parent that we're no longer interested in the old VM */
 996         tsk = current;
 997         old_mm = current->mm;
 998         mm_release(tsk, old_mm);
 999
1000         if (old_mm) {
1001                 sync_mm_rss(old_mm);
1002                 /*
1003                  * Make sure that if there is a core dump in progress
1004                  * for the old mm, we get out and die instead of going
1005                  * through with the exec.  We must hold mmap_sem around
1006                  * checking core_state and changing tsk->mm.
1007                  */
1008                 down_read(&old_mm->mmap_sem);
1009                 if (unlikely(old_mm->core_state)) {
1010                         up_read(&old_mm->mmap_sem);
1011                         return -EINTR;
1012                 }
1013         }
1014         task_lock(tsk);
1015         active_mm = tsk->active_mm;
1016         tsk->mm = mm;
1017         tsk->active_mm = mm;
1018         activate_mm(active_mm, mm);
1019         tsk->mm->vmacache_seqnum = 0;
1020         vmacache_flush(tsk);
1021         task_unlock(tsk);
1022         if (old_mm) {
1023                 up_read(&old_mm->mmap_sem);
1024                 BUG_ON(active_mm != old_mm);
1025                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1026                 mm_update_next_owner(old_mm);
1027                 mmput(old_mm);
1028                 return 0;
1029         }
1030         mmdrop(active_mm);
1031         return 0;
1032 }
1033
1034 /*
1035  * This function makes sure the current process has its own signal table,
1036  * so that flush_signal_handlers can later reset the handlers without
1037  * disturbing other processes.  (Other processes might share the signal
1038  * table via the CLONE_SIGHAND option to clone().)
1039  */
1040 static int de_thread(struct task_struct *tsk)
1041 {
1042         struct signal_struct *sig = tsk->signal;
1043         struct sighand_struct *oldsighand = tsk->sighand;
1044         spinlock_t *lock = &oldsighand->siglock;
1045
1046         if (thread_group_empty(tsk))
1047                 goto no_thread_group;
1048
1049         /*
1050          * Kill all other threads in the thread group.
1051          */
1052         spin_lock_irq(lock);
1053         if (signal_group_exit(sig)) {
1054                 /*
1055                  * Another group action in progress, just
1056                  * return so that the signal is processed.
1057                  */
1058                 spin_unlock_irq(lock);
1059                 return -EAGAIN;
1060         }
1061
1062         sig->group_exit_task = tsk;
1063         sig->notify_count = zap_other_threads(tsk);
1064         if (!thread_group_leader(tsk))
1065                 sig->notify_count--;
1066
1067         while (sig->notify_count) {
1068                 __set_current_state(TASK_KILLABLE);
1069                 spin_unlock_irq(lock);
1070                 schedule();
1071                 if (unlikely(__fatal_signal_pending(tsk)))
1072                         goto killed;
1073                 spin_lock_irq(lock);
1074         }
1075         spin_unlock_irq(lock);
1076
1077         /*
1078          * At this point all other threads have exited, all we have to
1079          * do is to wait for the thread group leader to become inactive,
1080          * and to assume its PID:
1081          */
1082         if (!thread_group_leader(tsk)) {
1083                 struct task_struct *leader = tsk->group_leader;
1084
1085                 for (;;) {
1086                         threadgroup_change_begin(tsk);
1087                         write_lock_irq(&tasklist_lock);
1088                         /*
1089                          * Do this under tasklist_lock to ensure that
1090                          * exit_notify() can't miss ->group_exit_task
1091                          */
1092                         sig->notify_count = -1;
1093                         if (likely(leader->exit_state))
1094                                 break;
1095                         __set_current_state(TASK_KILLABLE);
1096                         write_unlock_irq(&tasklist_lock);
1097                         threadgroup_change_end(tsk);
1098                         schedule();
1099                         if (unlikely(__fatal_signal_pending(tsk)))
1100                                 goto killed;
1101                 }
1102
1103                 /*
1104                  * The only record we have of the real-time age of a
1105                  * process, regardless of execs it's done, is start_time.
1106                  * All the past CPU time is accumulated in signal_struct
1107                  * from sister threads now dead.  But in this non-leader
1108                  * exec, nothing survives from the original leader thread,
1109                  * whose birth marks the true age of this process now.
1110                  * When we take on its identity by switching to its PID, we
1111                  * also take its birthdate (always earlier than our own).
1112                  */
1113                 tsk->start_time = leader->start_time;
1114                 tsk->real_start_time = leader->real_start_time;
1115
1116                 BUG_ON(!same_thread_group(leader, tsk));
1117                 BUG_ON(has_group_leader_pid(tsk));
1118                 /*
1119                  * An exec() starts a new thread group with the
1120                  * TGID of the previous thread group. Rehash the
1121                  * two threads with a switched PID, and release
1122                  * the former thread group leader:
1123                  */
1124
1125                 /* Become a process group leader with the old leader's pid.
1126                  * The old leader becomes a thread of the this thread group.
1127                  * Note: The old leader also uses this pid until release_task
1128                  *       is called.  Odd but simple and correct.
1129                  */
1130                 tsk->pid = leader->pid;
1131                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1132                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1133                 transfer_pid(leader, tsk, PIDTYPE_SID);
1134
1135                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1136                 list_replace_init(&leader->sibling, &tsk->sibling);
1137
1138                 tsk->group_leader = tsk;
1139                 leader->group_leader = tsk;
1140
1141                 tsk->exit_signal = SIGCHLD;
1142                 leader->exit_signal = -1;
1143
1144                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1145                 leader->exit_state = EXIT_DEAD;
1146
1147                 /*
1148                  * We are going to release_task()->ptrace_unlink() silently,
1149                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1150                  * the tracer wont't block again waiting for this thread.
1151                  */
1152                 if (unlikely(leader->ptrace))
1153                         __wake_up_parent(leader, leader->parent);
1154                 write_unlock_irq(&tasklist_lock);
1155                 threadgroup_change_end(tsk);
1156
1157                 release_task(leader);
1158         }
1159
1160         sig->group_exit_task = NULL;
1161         sig->notify_count = 0;
1162
1163 no_thread_group:
1164         /* we have changed execution domain */
1165         tsk->exit_signal = SIGCHLD;
1166
1167         exit_itimers(sig);
1168         flush_itimer_signals();
1169
1170         if (atomic_read(&oldsighand->count) != 1) {
1171                 struct sighand_struct *newsighand;
1172                 /*
1173                  * This ->sighand is shared with the CLONE_SIGHAND
1174                  * but not CLONE_THREAD task, switch to the new one.
1175                  */
1176                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1177                 if (!newsighand)
1178                         return -ENOMEM;
1179
1180                 atomic_set(&newsighand->count, 1);
1181                 memcpy(newsighand->action, oldsighand->action,
1182                        sizeof(newsighand->action));
1183
1184                 write_lock_irq(&tasklist_lock);
1185                 spin_lock(&oldsighand->siglock);
1186                 rcu_assign_pointer(tsk->sighand, newsighand);
1187                 spin_unlock(&oldsighand->siglock);
1188                 write_unlock_irq(&tasklist_lock);
1189
1190                 __cleanup_sighand(oldsighand);
1191         }
1192
1193         BUG_ON(!thread_group_leader(tsk));
1194         return 0;
1195
1196 killed:
1197         /* protects against exit_notify() and __exit_signal() */
1198         read_lock(&tasklist_lock);
1199         sig->group_exit_task = NULL;
1200         sig->notify_count = 0;
1201         read_unlock(&tasklist_lock);
1202         return -EAGAIN;
1203 }
1204
1205 char *get_task_comm(char *buf, struct task_struct *tsk)
1206 {
1207         /* buf must be at least sizeof(tsk->comm) in size */
1208         task_lock(tsk);
1209         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1210         task_unlock(tsk);
1211         return buf;
1212 }
1213 EXPORT_SYMBOL_GPL(get_task_comm);
1214
1215 /*
1216  * These functions flushes out all traces of the currently running executable
1217  * so that a new one can be started
1218  */
1219
1220 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1221 {
1222         task_lock(tsk);
1223         trace_task_rename(tsk, buf);
1224         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1225         task_unlock(tsk);
1226         perf_event_comm(tsk, exec);
1227 }
1228
1229 int flush_old_exec(struct linux_binprm * bprm)
1230 {
1231         int retval;
1232
1233         /*
1234          * Make sure we have a private signal table and that
1235          * we are unassociated from the previous thread group.
1236          */
1237         retval = de_thread(current);
1238         if (retval)
1239                 goto out;
1240
1241         /*
1242          * Must be called _before_ exec_mmap() as bprm->mm is
1243          * not visibile until then. This also enables the update
1244          * to be lockless.
1245          */
1246         set_mm_exe_file(bprm->mm, bprm->file);
1247
1248         /*
1249          * Release all of the old mmap stuff
1250          */
1251         acct_arg_size(bprm, 0);
1252         retval = exec_mmap(bprm->mm);
1253         if (retval)
1254                 goto out;
1255
1256         bprm->mm = NULL;                /* We're using it now */
1257
1258         set_fs(USER_DS);
1259         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1260                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1261         flush_thread();
1262         current->personality &= ~bprm->per_clear;
1263
1264         return 0;
1265
1266 out:
1267         return retval;
1268 }
1269 EXPORT_SYMBOL(flush_old_exec);
1270
1271 void would_dump(struct linux_binprm *bprm, struct file *file)
1272 {
1273         if (inode_permission(file_inode(file), MAY_READ) < 0)
1274                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1275 }
1276 EXPORT_SYMBOL(would_dump);
1277
1278 void setup_new_exec(struct linux_binprm * bprm)
1279 {
1280         arch_pick_mmap_layout(current->mm);
1281
1282         /* This is the point of no return */
1283         current->sas_ss_sp = current->sas_ss_size = 0;
1284
1285         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1286                 set_dumpable(current->mm, SUID_DUMP_USER);
1287         else
1288                 set_dumpable(current->mm, suid_dumpable);
1289
1290         perf_event_exec();
1291         __set_task_comm(current, kbasename(bprm->filename), true);
1292
1293         /* Set the new mm task size. We have to do that late because it may
1294          * depend on TIF_32BIT which is only updated in flush_thread() on
1295          * some architectures like powerpc
1296          */
1297         current->mm->task_size = TASK_SIZE;
1298
1299         /* install the new credentials */
1300         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1301             !gid_eq(bprm->cred->gid, current_egid())) {
1302                 current->pdeath_signal = 0;
1303         } else {
1304                 would_dump(bprm, bprm->file);
1305                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1306                         set_dumpable(current->mm, suid_dumpable);
1307         }
1308
1309         /* An exec changes our domain. We are no longer part of the thread
1310            group */
1311         current->self_exec_id++;
1312         flush_signal_handlers(current, 0);
1313         do_close_on_exec(current->files);
1314 }
1315 EXPORT_SYMBOL(setup_new_exec);
1316
1317 /*
1318  * Prepare credentials and lock ->cred_guard_mutex.
1319  * install_exec_creds() commits the new creds and drops the lock.
1320  * Or, if exec fails before, free_bprm() should release ->cred and
1321  * and unlock.
1322  */
1323 int prepare_bprm_creds(struct linux_binprm *bprm)
1324 {
1325         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1326                 return -ERESTARTNOINTR;
1327
1328         bprm->cred = prepare_exec_creds();
1329         if (likely(bprm->cred))
1330                 return 0;
1331
1332         mutex_unlock(&current->signal->cred_guard_mutex);
1333         return -ENOMEM;
1334 }
1335
1336 static void free_bprm(struct linux_binprm *bprm)
1337 {
1338         free_arg_pages(bprm);
1339         if (bprm->cred) {
1340                 mutex_unlock(&current->signal->cred_guard_mutex);
1341                 abort_creds(bprm->cred);
1342         }
1343         if (bprm->file) {
1344                 allow_write_access(bprm->file);
1345                 fput(bprm->file);
1346         }
1347         /* If a binfmt changed the interp, free it. */
1348         if (bprm->interp != bprm->filename)
1349                 kfree(bprm->interp);
1350         kfree(bprm);
1351 }
1352
1353 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1354 {
1355         /* If a binfmt changed the interp, free it first. */
1356         if (bprm->interp != bprm->filename)
1357                 kfree(bprm->interp);
1358         bprm->interp = kstrdup(interp, GFP_KERNEL);
1359         if (!bprm->interp)
1360                 return -ENOMEM;
1361         return 0;
1362 }
1363 EXPORT_SYMBOL(bprm_change_interp);
1364
1365 /*
1366  * install the new credentials for this executable
1367  */
1368 void install_exec_creds(struct linux_binprm *bprm)
1369 {
1370         security_bprm_committing_creds(bprm);
1371
1372         commit_creds(bprm->cred);
1373         bprm->cred = NULL;
1374
1375         /*
1376          * Disable monitoring for regular users
1377          * when executing setuid binaries. Must
1378          * wait until new credentials are committed
1379          * by commit_creds() above
1380          */
1381         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1382                 perf_event_exit_task(current);
1383         /*
1384          * cred_guard_mutex must be held at least to this point to prevent
1385          * ptrace_attach() from altering our determination of the task's
1386          * credentials; any time after this it may be unlocked.
1387          */
1388         security_bprm_committed_creds(bprm);
1389         mutex_unlock(&current->signal->cred_guard_mutex);
1390 }
1391 EXPORT_SYMBOL(install_exec_creds);
1392
1393 /*
1394  * determine how safe it is to execute the proposed program
1395  * - the caller must hold ->cred_guard_mutex to protect against
1396  *   PTRACE_ATTACH or seccomp thread-sync
1397  */
1398 static void check_unsafe_exec(struct linux_binprm *bprm)
1399 {
1400         struct task_struct *p = current, *t;
1401         unsigned n_fs;
1402
1403         if (p->ptrace) {
1404                 if (p->ptrace & PT_PTRACE_CAP)
1405                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1406                 else
1407                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1408         }
1409
1410         /*
1411          * This isn't strictly necessary, but it makes it harder for LSMs to
1412          * mess up.
1413          */
1414         if (task_no_new_privs(current))
1415                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1416
1417         t = p;
1418         n_fs = 1;
1419         spin_lock(&p->fs->lock);
1420         rcu_read_lock();
1421         while_each_thread(p, t) {
1422                 if (t->fs == p->fs)
1423                         n_fs++;
1424         }
1425         rcu_read_unlock();
1426
1427         if (p->fs->users > n_fs)
1428                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1429         else
1430                 p->fs->in_exec = 1;
1431         spin_unlock(&p->fs->lock);
1432 }
1433
1434 static void bprm_fill_uid(struct linux_binprm *bprm)
1435 {
1436         struct inode *inode;
1437         unsigned int mode;
1438         kuid_t uid;
1439         kgid_t gid;
1440
1441         /*
1442          * Since this can be called multiple times (via prepare_binprm),
1443          * we must clear any previous work done when setting set[ug]id
1444          * bits from any earlier bprm->file uses (for example when run
1445          * first for a setuid script then again for its interpreter).
1446          */
1447         bprm->cred->euid = current_euid();
1448         bprm->cred->egid = current_egid();
1449
1450         if (!mnt_may_suid(bprm->file->f_path.mnt))
1451                 return;
1452
1453         if (task_no_new_privs(current))
1454                 return;
1455
1456         inode = file_inode(bprm->file);
1457         mode = READ_ONCE(inode->i_mode);
1458         if (!(mode & (S_ISUID|S_ISGID)))
1459                 return;
1460
1461         /* Be careful if suid/sgid is set */
1462         inode_lock(inode);
1463
1464         /* reload atomically mode/uid/gid now that lock held */
1465         mode = inode->i_mode;
1466         uid = inode->i_uid;
1467         gid = inode->i_gid;
1468         inode_unlock(inode);
1469
1470         /* We ignore suid/sgid if there are no mappings for them in the ns */
1471         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1472                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1473                 return;
1474
1475         if (mode & S_ISUID) {
1476                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1477                 bprm->cred->euid = uid;
1478         }
1479
1480         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1481                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1482                 bprm->cred->egid = gid;
1483         }
1484 }
1485
1486 /*
1487  * Fill the binprm structure from the inode.
1488  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1489  *
1490  * This may be called multiple times for binary chains (scripts for example).
1491  */
1492 int prepare_binprm(struct linux_binprm *bprm)
1493 {
1494         int retval;
1495
1496         bprm_fill_uid(bprm);
1497
1498         /* fill in binprm security blob */
1499         retval = security_bprm_set_creds(bprm);
1500         if (retval)
1501                 return retval;
1502         bprm->cred_prepared = 1;
1503
1504         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1505         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1506 }
1507
1508 EXPORT_SYMBOL(prepare_binprm);
1509
1510 /*
1511  * Arguments are '\0' separated strings found at the location bprm->p
1512  * points to; chop off the first by relocating brpm->p to right after
1513  * the first '\0' encountered.
1514  */
1515 int remove_arg_zero(struct linux_binprm *bprm)
1516 {
1517         int ret = 0;
1518         unsigned long offset;
1519         char *kaddr;
1520         struct page *page;
1521
1522         if (!bprm->argc)
1523                 return 0;
1524
1525         do {
1526                 offset = bprm->p & ~PAGE_MASK;
1527                 page = get_arg_page(bprm, bprm->p, 0);
1528                 if (!page) {
1529                         ret = -EFAULT;
1530                         goto out;
1531                 }
1532                 kaddr = kmap_atomic(page);
1533
1534                 for (; offset < PAGE_SIZE && kaddr[offset];
1535                                 offset++, bprm->p++)
1536                         ;
1537
1538                 kunmap_atomic(kaddr);
1539                 put_arg_page(page);
1540         } while (offset == PAGE_SIZE);
1541
1542         bprm->p++;
1543         bprm->argc--;
1544         ret = 0;
1545
1546 out:
1547         return ret;
1548 }
1549 EXPORT_SYMBOL(remove_arg_zero);
1550
1551 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1552 /*
1553  * cycle the list of binary formats handler, until one recognizes the image
1554  */
1555 int search_binary_handler(struct linux_binprm *bprm)
1556 {
1557         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1558         struct linux_binfmt *fmt;
1559         int retval;
1560
1561         /* This allows 4 levels of binfmt rewrites before failing hard. */
1562         if (bprm->recursion_depth > 5)
1563                 return -ELOOP;
1564
1565         retval = security_bprm_check(bprm);
1566         if (retval)
1567                 return retval;
1568
1569         retval = -ENOENT;
1570  retry:
1571         read_lock(&binfmt_lock);
1572         list_for_each_entry(fmt, &formats, lh) {
1573                 if (!try_module_get(fmt->module))
1574                         continue;
1575                 read_unlock(&binfmt_lock);
1576                 bprm->recursion_depth++;
1577                 retval = fmt->load_binary(bprm);
1578                 read_lock(&binfmt_lock);
1579                 put_binfmt(fmt);
1580                 bprm->recursion_depth--;
1581                 if (retval < 0 && !bprm->mm) {
1582                         /* we got to flush_old_exec() and failed after it */
1583                         read_unlock(&binfmt_lock);
1584                         force_sigsegv(SIGSEGV, current);
1585                         return retval;
1586                 }
1587                 if (retval != -ENOEXEC || !bprm->file) {
1588                         read_unlock(&binfmt_lock);
1589                         return retval;
1590                 }
1591         }
1592         read_unlock(&binfmt_lock);
1593
1594         if (need_retry) {
1595                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1596                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1597                         return retval;
1598                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1599                         return retval;
1600                 need_retry = false;
1601                 goto retry;
1602         }
1603
1604         return retval;
1605 }
1606 EXPORT_SYMBOL(search_binary_handler);
1607
1608 static int exec_binprm(struct linux_binprm *bprm)
1609 {
1610         pid_t old_pid, old_vpid;
1611         int ret;
1612
1613         /* Need to fetch pid before load_binary changes it */
1614         old_pid = current->pid;
1615         rcu_read_lock();
1616         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1617         rcu_read_unlock();
1618
1619         ret = search_binary_handler(bprm);
1620         if (ret >= 0) {
1621                 audit_bprm(bprm);
1622                 trace_sched_process_exec(current, old_pid, bprm);
1623                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1624                 proc_exec_connector(current);
1625         }
1626
1627         return ret;
1628 }
1629
1630 /*
1631  * sys_execve() executes a new program.
1632  */
1633 static int do_execveat_common(int fd, struct filename *filename,
1634                               struct user_arg_ptr argv,
1635                               struct user_arg_ptr envp,
1636                               int flags)
1637 {
1638         char *pathbuf = NULL;
1639         struct linux_binprm *bprm;
1640         struct file *file;
1641         struct files_struct *displaced;
1642         int retval;
1643
1644         if (IS_ERR(filename))
1645                 return PTR_ERR(filename);
1646
1647         /*
1648          * We move the actual failure in case of RLIMIT_NPROC excess from
1649          * set*uid() to execve() because too many poorly written programs
1650          * don't check setuid() return code.  Here we additionally recheck
1651          * whether NPROC limit is still exceeded.
1652          */
1653         if ((current->flags & PF_NPROC_EXCEEDED) &&
1654             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1655                 retval = -EAGAIN;
1656                 goto out_ret;
1657         }
1658
1659         /* We're below the limit (still or again), so we don't want to make
1660          * further execve() calls fail. */
1661         current->flags &= ~PF_NPROC_EXCEEDED;
1662
1663         retval = unshare_files(&displaced);
1664         if (retval)
1665                 goto out_ret;
1666
1667         retval = -ENOMEM;
1668         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1669         if (!bprm)
1670                 goto out_files;
1671
1672         retval = prepare_bprm_creds(bprm);
1673         if (retval)
1674                 goto out_free;
1675
1676         check_unsafe_exec(bprm);
1677         current->in_execve = 1;
1678
1679         file = do_open_execat(fd, filename, flags);
1680         retval = PTR_ERR(file);
1681         if (IS_ERR(file))
1682                 goto out_unmark;
1683
1684         sched_exec();
1685
1686         bprm->file = file;
1687         if (fd == AT_FDCWD || filename->name[0] == '/') {
1688                 bprm->filename = filename->name;
1689         } else {
1690                 if (filename->name[0] == '\0')
1691                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1692                 else
1693                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1694                                             fd, filename->name);
1695                 if (!pathbuf) {
1696                         retval = -ENOMEM;
1697                         goto out_unmark;
1698                 }
1699                 /*
1700                  * Record that a name derived from an O_CLOEXEC fd will be
1701                  * inaccessible after exec. Relies on having exclusive access to
1702                  * current->files (due to unshare_files above).
1703                  */
1704                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1705                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1706                 bprm->filename = pathbuf;
1707         }
1708         bprm->interp = bprm->filename;
1709
1710         retval = bprm_mm_init(bprm);
1711         if (retval)
1712                 goto out_unmark;
1713
1714         bprm->argc = count(argv, MAX_ARG_STRINGS);
1715         if ((retval = bprm->argc) < 0)
1716                 goto out;
1717
1718         bprm->envc = count(envp, MAX_ARG_STRINGS);
1719         if ((retval = bprm->envc) < 0)
1720                 goto out;
1721
1722         retval = prepare_binprm(bprm);
1723         if (retval < 0)
1724                 goto out;
1725
1726         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1727         if (retval < 0)
1728                 goto out;
1729
1730         bprm->exec = bprm->p;
1731         retval = copy_strings(bprm->envc, envp, bprm);
1732         if (retval < 0)
1733                 goto out;
1734
1735         retval = copy_strings(bprm->argc, argv, bprm);
1736         if (retval < 0)
1737                 goto out;
1738
1739         retval = exec_binprm(bprm);
1740         if (retval < 0)
1741                 goto out;
1742
1743         /* execve succeeded */
1744         current->fs->in_exec = 0;
1745         current->in_execve = 0;
1746         acct_update_integrals(current);
1747         task_numa_free(current);
1748         free_bprm(bprm);
1749         kfree(pathbuf);
1750         putname(filename);
1751         if (displaced)
1752                 put_files_struct(displaced);
1753         return retval;
1754
1755 out:
1756         if (bprm->mm) {
1757                 acct_arg_size(bprm, 0);
1758                 mmput(bprm->mm);
1759         }
1760
1761 out_unmark:
1762         current->fs->in_exec = 0;
1763         current->in_execve = 0;
1764
1765 out_free:
1766         free_bprm(bprm);
1767         kfree(pathbuf);
1768
1769 out_files:
1770         if (displaced)
1771                 reset_files_struct(displaced);
1772 out_ret:
1773         putname(filename);
1774         return retval;
1775 }
1776
1777 int do_execve(struct filename *filename,
1778         const char __user *const __user *__argv,
1779         const char __user *const __user *__envp)
1780 {
1781         struct user_arg_ptr argv = { .ptr.native = __argv };
1782         struct user_arg_ptr envp = { .ptr.native = __envp };
1783         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1784 }
1785
1786 int do_execveat(int fd, struct filename *filename,
1787                 const char __user *const __user *__argv,
1788                 const char __user *const __user *__envp,
1789                 int flags)
1790 {
1791         struct user_arg_ptr argv = { .ptr.native = __argv };
1792         struct user_arg_ptr envp = { .ptr.native = __envp };
1793
1794         return do_execveat_common(fd, filename, argv, envp, flags);
1795 }
1796
1797 #ifdef CONFIG_COMPAT
1798 static int compat_do_execve(struct filename *filename,
1799         const compat_uptr_t __user *__argv,
1800         const compat_uptr_t __user *__envp)
1801 {
1802         struct user_arg_ptr argv = {
1803                 .is_compat = true,
1804                 .ptr.compat = __argv,
1805         };
1806         struct user_arg_ptr envp = {
1807                 .is_compat = true,
1808                 .ptr.compat = __envp,
1809         };
1810         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1811 }
1812
1813 static int compat_do_execveat(int fd, struct filename *filename,
1814                               const compat_uptr_t __user *__argv,
1815                               const compat_uptr_t __user *__envp,
1816                               int flags)
1817 {
1818         struct user_arg_ptr argv = {
1819                 .is_compat = true,
1820                 .ptr.compat = __argv,
1821         };
1822         struct user_arg_ptr envp = {
1823                 .is_compat = true,
1824                 .ptr.compat = __envp,
1825         };
1826         return do_execveat_common(fd, filename, argv, envp, flags);
1827 }
1828 #endif
1829
1830 void set_binfmt(struct linux_binfmt *new)
1831 {
1832         struct mm_struct *mm = current->mm;
1833
1834         if (mm->binfmt)
1835                 module_put(mm->binfmt->module);
1836
1837         mm->binfmt = new;
1838         if (new)
1839                 __module_get(new->module);
1840 }
1841 EXPORT_SYMBOL(set_binfmt);
1842
1843 /*
1844  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1845  */
1846 void set_dumpable(struct mm_struct *mm, int value)
1847 {
1848         unsigned long old, new;
1849
1850         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1851                 return;
1852
1853         do {
1854                 old = ACCESS_ONCE(mm->flags);
1855                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1856         } while (cmpxchg(&mm->flags, old, new) != old);
1857 }
1858
1859 SYSCALL_DEFINE3(execve,
1860                 const char __user *, filename,
1861                 const char __user *const __user *, argv,
1862                 const char __user *const __user *, envp)
1863 {
1864         return do_execve(getname(filename), argv, envp);
1865 }
1866
1867 SYSCALL_DEFINE5(execveat,
1868                 int, fd, const char __user *, filename,
1869                 const char __user *const __user *, argv,
1870                 const char __user *const __user *, envp,
1871                 int, flags)
1872 {
1873         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1874
1875         return do_execveat(fd,
1876                            getname_flags(filename, lookup_flags, NULL),
1877                            argv, envp, flags);
1878 }
1879
1880 #ifdef CONFIG_COMPAT
1881 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1882         const compat_uptr_t __user *, argv,
1883         const compat_uptr_t __user *, envp)
1884 {
1885         return compat_do_execve(getname(filename), argv, envp);
1886 }
1887
1888 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1889                        const char __user *, filename,
1890                        const compat_uptr_t __user *, argv,
1891                        const compat_uptr_t __user *, envp,
1892                        int,  flags)
1893 {
1894         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1895
1896         return compat_do_execveat(fd,
1897                                   getname_flags(filename, lookup_flags, NULL),
1898                                   argv, envp, flags);
1899 }
1900 #endif