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/sched/mm.h>
  36 #include <linux/sched/coredump.h>
  37 #include <linux/sched/signal.h>
  38 #include <linux/sched/numa_balancing.h>
  39 #include <linux/pagemap.h>
  40 #include <linux/perf_event.h>
  41 #include <linux/highmem.h>
  42 #include <linux/spinlock.h>
  43 #include <linux/key.h>
  44 #include <linux/personality.h>
  45 #include <linux/binfmts.h>
  46 #include <linux/utsname.h>
  47 #include <linux/pid_namespace.h>
  48 #include <linux/module.h>
  49 #include <linux/namei.h>
  50 #include <linux/mount.h>
  51 #include <linux/security.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/tsacct_kern.h>
  54 #include <linux/cn_proc.h>
  55 #include <linux/audit.h>
  56 #include <linux/tracehook.h>
  57 #include <linux/kmod.h>
  58 #include <linux/fsnotify.h>
  59 #include <linux/fs_struct.h>
  60 #include <linux/pipe_fs_i.h>
  61 #include <linux/oom.h>
  62 #include <linux/compat.h>
  63 #include <linux/vmalloc.h>
  64
  65 #include <linux/uaccess.h>
  66 #include <asm/mmu_context.h>
  67 #include <asm/tlb.h>
  68
  69 #include <trace/events/task.h>
  70 #include "internal.h"
  71
  72 #include <trace/events/sched.h>
  73
  74 int suid_dumpable = 0;
  75
  76 static LIST_HEAD(formats);
  77 static DEFINE_RWLOCK(binfmt_lock);
  78
  79 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  80 {
  81         BUG_ON(!fmt);
  82         if (WARN_ON(!fmt->load_binary))
  83                 return;
  84         write_lock(&binfmt_lock);
  85         insert ? list_add(&fmt->lh, &formats) :
  86                  list_add_tail(&fmt->lh, &formats);
  87         write_unlock(&binfmt_lock);
  88 }
  89
  90 EXPORT_SYMBOL(__register_binfmt);
  91
  92 void unregister_binfmt(struct linux_binfmt * fmt)
  93 {
  94         write_lock(&binfmt_lock);
  95         list_del(&fmt->lh);
  96         write_unlock(&binfmt_lock);
  97 }
  98
  99 EXPORT_SYMBOL(unregister_binfmt);
 100
 101 static inline void put_binfmt(struct linux_binfmt * fmt)
 102 {
 103         module_put(fmt->module);
 104 }
 105
 106 bool path_noexec(const struct path *path)
 107 {
 108         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 109                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 110 }
 111
 112 #ifdef CONFIG_USELIB
 113 /*
 114  * Note that a shared library must be both readable and executable due to
 115  * security reasons.
 116  *
 117  * Also note that we take the address to load from from the file itself.
 118  */
 119 SYSCALL_DEFINE1(uselib, const char __user *, library)
 120 {
 121         struct linux_binfmt *fmt;
 122         struct file *file;
 123         struct filename *tmp = getname(library);
 124         int error = PTR_ERR(tmp);
 125         static const struct open_flags uselib_flags = {
 126                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 127                 .acc_mode = MAY_READ | MAY_EXEC,
 128                 .intent = LOOKUP_OPEN,
 129                 .lookup_flags = LOOKUP_FOLLOW,
 130         };
 131
 132         if (IS_ERR(tmp))
 133                 goto out;
 134
 135         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 136         putname(tmp);
 137         error = PTR_ERR(file);
 138         if (IS_ERR(file))
 139                 goto out;
 140
 141         error = -EINVAL;
 142         if (!S_ISREG(file_inode(file)->i_mode))
 143                 goto exit;
 144
 145         error = -EACCES;
 146         if (path_noexec(&file->f_path))
 147                 goto exit;
 148
 149         fsnotify_open(file);
 150
 151         error = -ENOEXEC;
 152
 153         read_lock(&binfmt_lock);
 154         list_for_each_entry(fmt, &formats, lh) {
 155                 if (!fmt->load_shlib)
 156                         continue;
 157                 if (!try_module_get(fmt->module))
 158                         continue;
 159                 read_unlock(&binfmt_lock);
 160                 error = fmt->load_shlib(file);
 161                 read_lock(&binfmt_lock);
 162                 put_binfmt(fmt);
 163                 if (error != -ENOEXEC)
 164                         break;
 165         }
 166         read_unlock(&binfmt_lock);
 167 exit:
 168         fput(file);
 169 out:
 170         return error;
 171 }
 172 #endif /* #ifdef CONFIG_USELIB */
 173
 174 #ifdef CONFIG_MMU
 175 /*
 176  * The nascent bprm->mm is not visible until exec_mmap() but it can
 177  * use a lot of memory, account these pages in current->mm temporary
 178  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 179  * change the counter back via acct_arg_size(0).
 180  */
 181 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 182 {
 183         struct mm_struct *mm = current->mm;
 184         long diff = (long)(pages - bprm->vma_pages);
 185
 186         if (!mm || !diff)
 187                 return;
 188
 189         bprm->vma_pages = pages;
 190         add_mm_counter(mm, MM_ANONPAGES, diff);
 191 }
 192
 193 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 194                 int write)
 195 {
 196         struct page *page;
 197         int ret;
 198         unsigned int gup_flags = FOLL_FORCE;
 199
 200 #ifdef CONFIG_STACK_GROWSUP
 201         if (write) {
 202                 ret = expand_downwards(bprm->vma, pos);
 203                 if (ret < 0)
 204                         return NULL;
 205         }
 206 #endif
 207
 208         if (write)
 209                 gup_flags |= FOLL_WRITE;
 210
 211         /*
 212          * We are doing an exec().  'current' is the process
 213          * doing the exec and bprm->mm is the new process's mm.
 214          */
 215         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 216                         &page, NULL, NULL);
 217         if (ret <= 0)
 218                 return NULL;
 219
 220         if (write) {
 221                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 222                 struct rlimit *rlim;
 223
 224                 acct_arg_size(bprm, size / PAGE_SIZE);
 225
 226                 /*
 227                  * We've historically supported up to 32 pages (ARG_MAX)
 228                  * of argument strings even with small stacks
 229                  */
 230                 if (size <= ARG_MAX)
 231                         return page;
 232
 233                 /*
 234                  * Limit to 1/4-th the stack size for the argv+env strings.
 235                  * This ensures that:
 236                  *  - the remaining binfmt code will not run out of stack space,
 237                  *  - the program will have a reasonable amount of stack left
 238                  *    to work from.
 239                  */
 240                 rlim = current->signal->rlim;
 241                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 242                         put_page(page);
 243                         return NULL;
 244                 }
 245         }
 246
 247         return page;
 248 }
 249
 250 static void put_arg_page(struct page *page)
 251 {
 252         put_page(page);
 253 }
 254
 255 static void free_arg_pages(struct linux_binprm *bprm)
 256 {
 257 }
 258
 259 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 260                 struct page *page)
 261 {
 262         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 263 }
 264
 265 static int __bprm_mm_init(struct linux_binprm *bprm)
 266 {
 267         int err;
 268         struct vm_area_struct *vma = NULL;
 269         struct mm_struct *mm = bprm->mm;
 270
 271         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 272         if (!vma)
 273                 return -ENOMEM;
 274
 275         if (down_write_killable(&mm->mmap_sem)) {
 276                 err = -EINTR;
 277                 goto err_free;
 278         }
 279         vma->vm_mm = mm;
 280
 281         /*
 282          * Place the stack at the largest stack address the architecture
 283          * supports. Later, we'll move this to an appropriate place. We don't
 284          * use STACK_TOP because that can depend on attributes which aren't
 285          * configured yet.
 286          */
 287         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 288         vma->vm_end = STACK_TOP_MAX;
 289         vma->vm_start = vma->vm_end - PAGE_SIZE;
 290         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 291         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 292         INIT_LIST_HEAD(&vma->anon_vma_chain);
 293
 294         err = insert_vm_struct(mm, vma);
 295         if (err)
 296                 goto err;
 297
 298         mm->stack_vm = mm->total_vm = 1;
 299         arch_bprm_mm_init(mm, vma);
 300         up_write(&mm->mmap_sem);
 301         bprm->p = vma->vm_end - sizeof(void *);
 302         return 0;
 303 err:
 304         up_write(&mm->mmap_sem);
 305 err_free:
 306         bprm->vma = NULL;
 307         kmem_cache_free(vm_area_cachep, vma);
 308         return err;
 309 }
 310
 311 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 312 {
 313         return len <= MAX_ARG_STRLEN;
 314 }
 315
 316 #else
 317
 318 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 319 {
 320 }
 321
 322 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 323                 int write)
 324 {
 325         struct page *page;
 326
 327         page = bprm->page[pos / PAGE_SIZE];
 328         if (!page && write) {
 329                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 330                 if (!page)
 331                         return NULL;
 332                 bprm->page[pos / PAGE_SIZE] = page;
 333         }
 334
 335         return page;
 336 }
 337
 338 static void put_arg_page(struct page *page)
 339 {
 340 }
 341
 342 static void free_arg_page(struct linux_binprm *bprm, int i)
 343 {
 344         if (bprm->page[i]) {
 345                 __free_page(bprm->page[i]);
 346                 bprm->page[i] = NULL;
 347         }
 348 }
 349
 350 static void free_arg_pages(struct linux_binprm *bprm)
 351 {
 352         int i;
 353
 354         for (i = 0; i < MAX_ARG_PAGES; i++)
 355                 free_arg_page(bprm, i);
 356 }
 357
 358 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 359                 struct page *page)
 360 {
 361 }
 362
 363 static int __bprm_mm_init(struct linux_binprm *bprm)
 364 {
 365         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 366         return 0;
 367 }
 368
 369 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 370 {
 371         return len <= bprm->p;
 372 }
 373
 374 #endif /* CONFIG_MMU */
 375
 376 /*
 377  * Create a new mm_struct and populate it with a temporary stack
 378  * vm_area_struct.  We don't have enough context at this point to set the stack
 379  * flags, permissions, and offset, so we use temporary values.  We'll update
 380  * them later in setup_arg_pages().
 381  */
 382 static int bprm_mm_init(struct linux_binprm *bprm)
 383 {
 384         int err;
 385         struct mm_struct *mm = NULL;
 386
 387         bprm->mm = mm = mm_alloc();
 388         err = -ENOMEM;
 389         if (!mm)
 390                 goto err;
 391
 392         err = __bprm_mm_init(bprm);
 393         if (err)
 394                 goto err;
 395
 396         return 0;
 397
 398 err:
 399         if (mm) {
 400                 bprm->mm = NULL;
 401                 mmdrop(mm);
 402         }
 403
 404         return err;
 405 }
 406
 407 struct user_arg_ptr {
 408 #ifdef CONFIG_COMPAT
 409         bool is_compat;
 410 #endif
 411         union {
 412                 const char __user *const __user *native;
 413 #ifdef CONFIG_COMPAT
 414                 const compat_uptr_t __user *compat;
 415 #endif
 416         } ptr;
 417 };
 418
 419 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 420 {
 421         const char __user *native;
 422
 423 #ifdef CONFIG_COMPAT
 424         if (unlikely(argv.is_compat)) {
 425                 compat_uptr_t compat;
 426
 427                 if (get_user(compat, argv.ptr.compat + nr))
 428                         return ERR_PTR(-EFAULT);
 429
 430                 return compat_ptr(compat);
 431         }
 432 #endif
 433
 434         if (get_user(native, argv.ptr.native + nr))
 435                 return ERR_PTR(-EFAULT);
 436
 437         return native;
 438 }
 439
 440 /*
 441  * count() counts the number of strings in array ARGV.
 442  */
 443 static int count(struct user_arg_ptr argv, int max)
 444 {
 445         int i = 0;
 446
 447         if (argv.ptr.native != NULL) {
 448                 for (;;) {
 449                         const char __user *p = get_user_arg_ptr(argv, i);
 450
 451                         if (!p)
 452                                 break;
 453
 454                         if (IS_ERR(p))
 455                                 return -EFAULT;
 456
 457                         if (i >= max)
 458                                 return -E2BIG;
 459                         ++i;
 460
 461                         if (fatal_signal_pending(current))
 462                                 return -ERESTARTNOHAND;
 463                         cond_resched();
 464                 }
 465         }
 466         return i;
 467 }
 468
 469 /*
 470  * 'copy_strings()' copies argument/environment strings from the old
 471  * processes's memory to the new process's stack.  The call to get_user_pages()
 472  * ensures the destination page is created and not swapped out.
 473  */
 474 static int copy_strings(int argc, struct user_arg_ptr argv,
 475                         struct linux_binprm *bprm)
 476 {
 477         struct page *kmapped_page = NULL;
 478         char *kaddr = NULL;
 479         unsigned long kpos = 0;
 480         int ret;
 481
 482         while (argc-- > 0) {
 483                 const char __user *str;
 484                 int len;
 485                 unsigned long pos;
 486
 487                 ret = -EFAULT;
 488                 str = get_user_arg_ptr(argv, argc);
 489                 if (IS_ERR(str))
 490                         goto out;
 491
 492                 len = strnlen_user(str, MAX_ARG_STRLEN);
 493                 if (!len)
 494                         goto out;
 495
 496                 ret = -E2BIG;
 497                 if (!valid_arg_len(bprm, len))
 498                         goto out;
 499
 500                 /* We're going to work our way backwords. */
 501                 pos = bprm->p;
 502                 str += len;
 503                 bprm->p -= len;
 504
 505                 while (len > 0) {
 506                         int offset, bytes_to_copy;
 507
 508                         if (fatal_signal_pending(current)) {
 509                                 ret = -ERESTARTNOHAND;
 510                                 goto out;
 511                         }
 512                         cond_resched();
 513
 514                         offset = pos % PAGE_SIZE;
 515                         if (offset == 0)
 516                                 offset = PAGE_SIZE;
 517
 518                         bytes_to_copy = offset;
 519                         if (bytes_to_copy > len)
 520                                 bytes_to_copy = len;
 521
 522                         offset -= bytes_to_copy;
 523                         pos -= bytes_to_copy;
 524                         str -= bytes_to_copy;
 525                         len -= bytes_to_copy;
 526
 527                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 528                                 struct page *page;
 529
 530                                 page = get_arg_page(bprm, pos, 1);
 531                                 if (!page) {
 532                                         ret = -E2BIG;
 533                                         goto out;
 534                                 }
 535
 536                                 if (kmapped_page) {
 537                                         flush_kernel_dcache_page(kmapped_page);
 538                                         kunmap(kmapped_page);
 539                                         put_arg_page(kmapped_page);
 540                                 }
 541                                 kmapped_page = page;
 542                                 kaddr = kmap(kmapped_page);
 543                                 kpos = pos & PAGE_MASK;
 544                                 flush_arg_page(bprm, kpos, kmapped_page);
 545                         }
 546                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 547                                 ret = -EFAULT;
 548                                 goto out;
 549                         }
 550                 }
 551         }
 552         ret = 0;
 553 out:
 554         if (kmapped_page) {
 555                 flush_kernel_dcache_page(kmapped_page);
 556                 kunmap(kmapped_page);
 557                 put_arg_page(kmapped_page);
 558         }
 559         return ret;
 560 }
 561
 562 /*
 563  * Like copy_strings, but get argv and its values from kernel memory.
 564  */
 565 int copy_strings_kernel(int argc, const char *const *__argv,
 566                         struct linux_binprm *bprm)
 567 {
 568         int r;
 569         mm_segment_t oldfs = get_fs();
 570         struct user_arg_ptr argv = {
 571                 .ptr.native = (const char __user *const  __user *)__argv,
 572         };
 573
 574         set_fs(KERNEL_DS);
 575         r = copy_strings(argc, argv, bprm);
 576         set_fs(oldfs);
 577
 578         return r;
 579 }
 580 EXPORT_SYMBOL(copy_strings_kernel);
 581
 582 #ifdef CONFIG_MMU
 583
 584 /*
 585  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 586  * the binfmt code determines where the new stack should reside, we shift it to
 587  * its final location.  The process proceeds as follows:
 588  *
 589  * 1) Use shift to calculate the new vma endpoints.
 590  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 591  *    arguments passed to subsequent functions are consistent.
 592  * 3) Move vma's page tables to the new range.
 593  * 4) Free up any cleared pgd range.
 594  * 5) Shrink the vma to cover only the new range.
 595  */
 596 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 597 {
 598         struct mm_struct *mm = vma->vm_mm;
 599         unsigned long old_start = vma->vm_start;
 600         unsigned long old_end = vma->vm_end;
 601         unsigned long length = old_end - old_start;
 602         unsigned long new_start = old_start - shift;
 603         unsigned long new_end = old_end - shift;
 604         struct mmu_gather tlb;
 605
 606         BUG_ON(new_start > new_end);
 607
 608         /*
 609          * ensure there are no vmas between where we want to go
 610          * and where we are
 611          */
 612         if (vma != find_vma(mm, new_start))
 613                 return -EFAULT;
 614
 615         /*
 616          * cover the whole range: [new_start, old_end)
 617          */
 618         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 619                 return -ENOMEM;
 620
 621         /*
 622          * move the page tables downwards, on failure we rely on
 623          * process cleanup to remove whatever mess we made.
 624          */
 625         if (length != move_page_tables(vma, old_start,
 626                                        vma, new_start, length, false))
 627                 return -ENOMEM;
 628
 629         lru_add_drain();
 630         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 631         if (new_end > old_start) {
 632                 /*
 633                  * when the old and new regions overlap clear from new_end.
 634                  */
 635                 free_pgd_range(&tlb, new_end, old_end, new_end,
 636                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 637         } else {
 638                 /*
 639                  * otherwise, clean from old_start; this is done to not touch
 640                  * the address space in [new_end, old_start) some architectures
 641                  * have constraints on va-space that make this illegal (IA64) -
 642                  * for the others its just a little faster.
 643                  */
 644                 free_pgd_range(&tlb, old_start, old_end, new_end,
 645                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 646         }
 647         tlb_finish_mmu(&tlb, old_start, old_end);
 648
 649         /*
 650          * Shrink the vma to just the new range.  Always succeeds.
 651          */
 652         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 653
 654         return 0;
 655 }
 656
 657 /*
 658  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 659  * the stack is optionally relocated, and some extra space is added.
 660  */
 661 int setup_arg_pages(struct linux_binprm *bprm,
 662                     unsigned long stack_top,
 663                     int executable_stack)
 664 {
 665         unsigned long ret;
 666         unsigned long stack_shift;
 667         struct mm_struct *mm = current->mm;
 668         struct vm_area_struct *vma = bprm->vma;
 669         struct vm_area_struct *prev = NULL;
 670         unsigned long vm_flags;
 671         unsigned long stack_base;
 672         unsigned long stack_size;
 673         unsigned long stack_expand;
 674         unsigned long rlim_stack;
 675
 676 #ifdef CONFIG_STACK_GROWSUP
 677         /* Limit stack size */
 678         stack_base = rlimit_max(RLIMIT_STACK);
 679         if (stack_base > STACK_SIZE_MAX)
 680                 stack_base = STACK_SIZE_MAX;
 681
 682         /* Add space for stack randomization. */
 683         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 684
 685         /* Make sure we didn't let the argument array grow too large. */
 686         if (vma->vm_end - vma->vm_start > stack_base)
 687                 return -ENOMEM;
 688
 689         stack_base = PAGE_ALIGN(stack_top - stack_base);
 690
 691         stack_shift = vma->vm_start - stack_base;
 692         mm->arg_start = bprm->p - stack_shift;
 693         bprm->p = vma->vm_end - stack_shift;
 694 #else
 695         stack_top = arch_align_stack(stack_top);
 696         stack_top = PAGE_ALIGN(stack_top);
 697
 698         if (unlikely(stack_top < mmap_min_addr) ||
 699             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 700                 return -ENOMEM;
 701
 702         stack_shift = vma->vm_end - stack_top;
 703
 704         bprm->p -= stack_shift;
 705         mm->arg_start = bprm->p;
 706 #endif
 707
 708         if (bprm->loader)
 709                 bprm->loader -= stack_shift;
 710         bprm->exec -= stack_shift;
 711
 712         if (down_write_killable(&mm->mmap_sem))
 713                 return -EINTR;
 714
 715         vm_flags = VM_STACK_FLAGS;
 716
 717         /*
 718          * Adjust stack execute permissions; explicitly enable for
 719          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 720          * (arch default) otherwise.
 721          */
 722         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 723                 vm_flags |= VM_EXEC;
 724         else if (executable_stack == EXSTACK_DISABLE_X)
 725                 vm_flags &= ~VM_EXEC;
 726         vm_flags |= mm->def_flags;
 727         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 728
 729         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 730                         vm_flags);
 731         if (ret)
 732                 goto out_unlock;
 733         BUG_ON(prev != vma);
 734
 735         /* Move stack pages down in memory. */
 736         if (stack_shift) {
 737                 ret = shift_arg_pages(vma, stack_shift);
 738                 if (ret)
 739                         goto out_unlock;
 740         }
 741
 742         /* mprotect_fixup is overkill to remove the temporary stack flags */
 743         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 744
 745         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 746         stack_size = vma->vm_end - vma->vm_start;
 747         /*
 748          * Align this down to a page boundary as expand_stack
 749          * will align it up.
 750          */
 751         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 752 #ifdef CONFIG_STACK_GROWSUP
 753         if (stack_size + stack_expand > rlim_stack)
 754                 stack_base = vma->vm_start + rlim_stack;
 755         else
 756                 stack_base = vma->vm_end + stack_expand;
 757 #else
 758         if (stack_size + stack_expand > rlim_stack)
 759                 stack_base = vma->vm_end - rlim_stack;
 760         else
 761                 stack_base = vma->vm_start - stack_expand;
 762 #endif
 763         current->mm->start_stack = bprm->p;
 764         ret = expand_stack(vma, stack_base);
 765         if (ret)
 766                 ret = -EFAULT;
 767
 768 out_unlock:
 769         up_write(&mm->mmap_sem);
 770         return ret;
 771 }
 772 EXPORT_SYMBOL(setup_arg_pages);
 773
 774 #else
 775
 776 /*
 777  * Transfer the program arguments and environment from the holding pages
 778  * onto the stack. The provided stack pointer is adjusted accordingly.
 779  */
 780 int transfer_args_to_stack(struct linux_binprm *bprm,
 781                            unsigned long *sp_location)
 782 {
 783         unsigned long index, stop, sp;
 784         int ret = 0;
 785
 786         stop = bprm->p >> PAGE_SHIFT;
 787         sp = *sp_location;
 788
 789         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 790                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 791                 char *src = kmap(bprm->page[index]) + offset;
 792                 sp -= PAGE_SIZE - offset;
 793                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 794                         ret = -EFAULT;
 795                 kunmap(bprm->page[index]);
 796                 if (ret)
 797                         goto out;
 798         }
 799
 800         *sp_location = sp;
 801
 802 out:
 803         return ret;
 804 }
 805 EXPORT_SYMBOL(transfer_args_to_stack);
 806
 807 #endif /* CONFIG_MMU */
 808
 809 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 810 {
 811         struct file *file;
 812         int err;
 813         struct open_flags open_exec_flags = {
 814                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 815                 .acc_mode = MAY_EXEC,
 816                 .intent = LOOKUP_OPEN,
 817                 .lookup_flags = LOOKUP_FOLLOW,
 818         };
 819
 820         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 821                 return ERR_PTR(-EINVAL);
 822         if (flags & AT_SYMLINK_NOFOLLOW)
 823                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 824         if (flags & AT_EMPTY_PATH)
 825                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 826
 827         file = do_filp_open(fd, name, &open_exec_flags);
 828         if (IS_ERR(file))
 829                 goto out;
 830
 831         err = -EACCES;
 832         if (!S_ISREG(file_inode(file)->i_mode))
 833                 goto exit;
 834
 835         if (path_noexec(&file->f_path))
 836                 goto exit;
 837
 838         err = deny_write_access(file);
 839         if (err)
 840                 goto exit;
 841
 842         if (name->name[0] != '\0')
 843                 fsnotify_open(file);
 844
 845 out:
 846         return file;
 847
 848 exit:
 849         fput(file);
 850         return ERR_PTR(err);
 851 }
 852
 853 struct file *open_exec(const char *name)
 854 {
 855         struct filename *filename = getname_kernel(name);
 856         struct file *f = ERR_CAST(filename);
 857
 858         if (!IS_ERR(filename)) {
 859                 f = do_open_execat(AT_FDCWD, filename, 0);
 860                 putname(filename);
 861         }
 862         return f;
 863 }
 864 EXPORT_SYMBOL(open_exec);
 865
 866 int kernel_read(struct file *file, loff_t offset,
 867                 char *addr, unsigned long count)
 868 {
 869         mm_segment_t old_fs;
 870         loff_t pos = offset;
 871         int result;
 872
 873         old_fs = get_fs();
 874         set_fs(get_ds());
 875         /* The cast to a user pointer is valid due to the set_fs() */
 876         result = vfs_read(file, (void __user *)addr, count, &pos);
 877         set_fs(old_fs);
 878         return result;
 879 }
 880
 881 EXPORT_SYMBOL(kernel_read);
 882
 883 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 884                      loff_t max_size, enum kernel_read_file_id id)
 885 {
 886         loff_t i_size, pos;
 887         ssize_t bytes = 0;
 888         int ret;
 889
 890         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 891                 return -EINVAL;
 892
 893         ret = security_kernel_read_file(file, id);
 894         if (ret)
 895                 return ret;
 896
 897         ret = deny_write_access(file);
 898         if (ret)
 899                 return ret;
 900
 901         i_size = i_size_read(file_inode(file));
 902         if (max_size > 0 && i_size > max_size) {
 903                 ret = -EFBIG;
 904                 goto out;
 905         }
 906         if (i_size <= 0) {
 907                 ret = -EINVAL;
 908                 goto out;
 909         }
 910
 911         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 912                 *buf = vmalloc(i_size);
 913         if (!*buf) {
 914                 ret = -ENOMEM;
 915                 goto out;
 916         }
 917
 918         pos = 0;
 919         while (pos < i_size) {
 920                 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 921                                     i_size - pos);
 922                 if (bytes < 0) {
 923                         ret = bytes;
 924                         goto out;
 925                 }
 926
 927                 if (bytes == 0)
 928                         break;
 929                 pos += bytes;
 930         }
 931
 932         if (pos != i_size) {
 933                 ret = -EIO;
 934                 goto out_free;
 935         }
 936
 937         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 938         if (!ret)
 939                 *size = pos;
 940
 941 out_free:
 942         if (ret < 0) {
 943                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 944                         vfree(*buf);
 945                         *buf = NULL;
 946                 }
 947         }
 948
 949 out:
 950         allow_write_access(file);
 951         return ret;
 952 }
 953 EXPORT_SYMBOL_GPL(kernel_read_file);
 954
 955 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 956                                loff_t max_size, enum kernel_read_file_id id)
 957 {
 958         struct file *file;
 959         int ret;
 960
 961         if (!path || !*path)
 962                 return -EINVAL;
 963
 964         file = filp_open(path, O_RDONLY, 0);
 965         if (IS_ERR(file))
 966                 return PTR_ERR(file);
 967
 968         ret = kernel_read_file(file, buf, size, max_size, id);
 969         fput(file);
 970         return ret;
 971 }
 972 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 973
 974 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 975                              enum kernel_read_file_id id)
 976 {
 977         struct fd f = fdget(fd);
 978         int ret = -EBADF;
 979
 980         if (!f.file)
 981                 goto out;
 982
 983         ret = kernel_read_file(f.file, buf, size, max_size, id);
 984 out:
 985         fdput(f);
 986         return ret;
 987 }
 988 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 989
 990 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 991 {
 992         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 993         if (res > 0)
 994                 flush_icache_range(addr, addr + len);
 995         return res;
 996 }
 997 EXPORT_SYMBOL(read_code);
 998
 999 static int exec_mmap(struct mm_struct *mm)
1000 {
1001         struct task_struct *tsk;
1002         struct mm_struct *old_mm, *active_mm;
1003
1004         /* Notify parent that we're no longer interested in the old VM */
1005         tsk = current;
1006         old_mm = current->mm;
1007         mm_release(tsk, old_mm);
1008
1009         if (old_mm) {
1010                 sync_mm_rss(old_mm);
1011                 /*
1012                  * Make sure that if there is a core dump in progress
1013                  * for the old mm, we get out and die instead of going
1014                  * through with the exec.  We must hold mmap_sem around
1015                  * checking core_state and changing tsk->mm.
1016                  */
1017                 down_read(&old_mm->mmap_sem);
1018                 if (unlikely(old_mm->core_state)) {
1019                         up_read(&old_mm->mmap_sem);
1020                         return -EINTR;
1021                 }
1022         }
1023         task_lock(tsk);
1024         active_mm = tsk->active_mm;
1025         tsk->mm = mm;
1026         tsk->active_mm = mm;
1027         activate_mm(active_mm, mm);
1028         tsk->mm->vmacache_seqnum = 0;
1029         vmacache_flush(tsk);
1030         task_unlock(tsk);
1031         if (old_mm) {
1032                 up_read(&old_mm->mmap_sem);
1033                 BUG_ON(active_mm != old_mm);
1034                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1035                 mm_update_next_owner(old_mm);
1036                 mmput(old_mm);
1037                 return 0;
1038         }
1039         mmdrop(active_mm);
1040         return 0;
1041 }
1042
1043 /*
1044  * This function makes sure the current process has its own signal table,
1045  * so that flush_signal_handlers can later reset the handlers without
1046  * disturbing other processes.  (Other processes might share the signal
1047  * table via the CLONE_SIGHAND option to clone().)
1048  */
1049 static int de_thread(struct task_struct *tsk)
1050 {
1051         struct signal_struct *sig = tsk->signal;
1052         struct sighand_struct *oldsighand = tsk->sighand;
1053         spinlock_t *lock = &oldsighand->siglock;
1054
1055         if (thread_group_empty(tsk))
1056                 goto no_thread_group;
1057
1058         /*
1059          * Kill all other threads in the thread group.
1060          */
1061         spin_lock_irq(lock);
1062         if (signal_group_exit(sig)) {
1063                 /*
1064                  * Another group action in progress, just
1065                  * return so that the signal is processed.
1066                  */
1067                 spin_unlock_irq(lock);
1068                 return -EAGAIN;
1069         }
1070
1071         sig->group_exit_task = tsk;
1072         sig->notify_count = zap_other_threads(tsk);
1073         if (!thread_group_leader(tsk))
1074                 sig->notify_count--;
1075
1076         while (sig->notify_count) {
1077                 __set_current_state(TASK_KILLABLE);
1078                 spin_unlock_irq(lock);
1079                 schedule();
1080                 if (unlikely(__fatal_signal_pending(tsk)))
1081                         goto killed;
1082                 spin_lock_irq(lock);
1083         }
1084         spin_unlock_irq(lock);
1085
1086         /*
1087          * At this point all other threads have exited, all we have to
1088          * do is to wait for the thread group leader to become inactive,
1089          * and to assume its PID:
1090          */
1091         if (!thread_group_leader(tsk)) {
1092                 struct task_struct *leader = tsk->group_leader;
1093
1094                 for (;;) {
1095                         cgroup_threadgroup_change_begin(tsk);
1096                         write_lock_irq(&tasklist_lock);
1097                         /*
1098                          * Do this under tasklist_lock to ensure that
1099                          * exit_notify() can't miss ->group_exit_task
1100                          */
1101                         sig->notify_count = -1;
1102                         if (likely(leader->exit_state))
1103                                 break;
1104                         __set_current_state(TASK_KILLABLE);
1105                         write_unlock_irq(&tasklist_lock);
1106                         cgroup_threadgroup_change_end(tsk);
1107                         schedule();
1108                         if (unlikely(__fatal_signal_pending(tsk)))
1109                                 goto killed;
1110                 }
1111
1112                 /*
1113                  * The only record we have of the real-time age of a
1114                  * process, regardless of execs it's done, is start_time.
1115                  * All the past CPU time is accumulated in signal_struct
1116                  * from sister threads now dead.  But in this non-leader
1117                  * exec, nothing survives from the original leader thread,
1118                  * whose birth marks the true age of this process now.
1119                  * When we take on its identity by switching to its PID, we
1120                  * also take its birthdate (always earlier than our own).
1121                  */
1122                 tsk->start_time = leader->start_time;
1123                 tsk->real_start_time = leader->real_start_time;
1124
1125                 BUG_ON(!same_thread_group(leader, tsk));
1126                 BUG_ON(has_group_leader_pid(tsk));
1127                 /*
1128                  * An exec() starts a new thread group with the
1129                  * TGID of the previous thread group. Rehash the
1130                  * two threads with a switched PID, and release
1131                  * the former thread group leader:
1132                  */
1133
1134                 /* Become a process group leader with the old leader's pid.
1135                  * The old leader becomes a thread of the this thread group.
1136                  * Note: The old leader also uses this pid until release_task
1137                  *       is called.  Odd but simple and correct.
1138                  */
1139                 tsk->pid = leader->pid;
1140                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1141                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1142                 transfer_pid(leader, tsk, PIDTYPE_SID);
1143
1144                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1145                 list_replace_init(&leader->sibling, &tsk->sibling);
1146
1147                 tsk->group_leader = tsk;
1148                 leader->group_leader = tsk;
1149
1150                 tsk->exit_signal = SIGCHLD;
1151                 leader->exit_signal = -1;
1152
1153                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1154                 leader->exit_state = EXIT_DEAD;
1155
1156                 /*
1157                  * We are going to release_task()->ptrace_unlink() silently,
1158                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1159                  * the tracer wont't block again waiting for this thread.
1160                  */
1161                 if (unlikely(leader->ptrace))
1162                         __wake_up_parent(leader, leader->parent);
1163                 write_unlock_irq(&tasklist_lock);
1164                 cgroup_threadgroup_change_end(tsk);
1165
1166                 release_task(leader);
1167         }
1168
1169         sig->group_exit_task = NULL;
1170         sig->notify_count = 0;
1171
1172 no_thread_group:
1173         /* we have changed execution domain */
1174         tsk->exit_signal = SIGCHLD;
1175
1176 #ifdef CONFIG_POSIX_TIMERS
1177         exit_itimers(sig);
1178         flush_itimer_signals();
1179 #endif
1180
1181         if (atomic_read(&oldsighand->count) != 1) {
1182                 struct sighand_struct *newsighand;
1183                 /*
1184                  * This ->sighand is shared with the CLONE_SIGHAND
1185                  * but not CLONE_THREAD task, switch to the new one.
1186                  */
1187                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1188                 if (!newsighand)
1189                         return -ENOMEM;
1190
1191                 atomic_set(&newsighand->count, 1);
1192                 memcpy(newsighand->action, oldsighand->action,
1193                        sizeof(newsighand->action));
1194
1195                 write_lock_irq(&tasklist_lock);
1196                 spin_lock(&oldsighand->siglock);
1197                 rcu_assign_pointer(tsk->sighand, newsighand);
1198                 spin_unlock(&oldsighand->siglock);
1199                 write_unlock_irq(&tasklist_lock);
1200
1201                 __cleanup_sighand(oldsighand);
1202         }
1203
1204         BUG_ON(!thread_group_leader(tsk));
1205         return 0;
1206
1207 killed:
1208         /* protects against exit_notify() and __exit_signal() */
1209         read_lock(&tasklist_lock);
1210         sig->group_exit_task = NULL;
1211         sig->notify_count = 0;
1212         read_unlock(&tasklist_lock);
1213         return -EAGAIN;
1214 }
1215
1216 char *get_task_comm(char *buf, struct task_struct *tsk)
1217 {
1218         /* buf must be at least sizeof(tsk->comm) in size */
1219         task_lock(tsk);
1220         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1221         task_unlock(tsk);
1222         return buf;
1223 }
1224 EXPORT_SYMBOL_GPL(get_task_comm);
1225
1226 /*
1227  * These functions flushes out all traces of the currently running executable
1228  * so that a new one can be started
1229  */
1230
1231 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1232 {
1233         task_lock(tsk);
1234         trace_task_rename(tsk, buf);
1235         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1236         task_unlock(tsk);
1237         perf_event_comm(tsk, exec);
1238 }
1239
1240 int flush_old_exec(struct linux_binprm * bprm)
1241 {
1242         int retval;
1243
1244         /*
1245          * Make sure we have a private signal table and that
1246          * we are unassociated from the previous thread group.
1247          */
1248         retval = de_thread(current);
1249         if (retval)
1250                 goto out;
1251
1252         /*
1253          * Must be called _before_ exec_mmap() as bprm->mm is
1254          * not visibile until then. This also enables the update
1255          * to be lockless.
1256          */
1257         set_mm_exe_file(bprm->mm, bprm->file);
1258
1259         /*
1260          * Release all of the old mmap stuff
1261          */
1262         acct_arg_size(bprm, 0);
1263         retval = exec_mmap(bprm->mm);
1264         if (retval)
1265                 goto out;
1266
1267         bprm->mm = NULL;                /* We're using it now */
1268
1269         set_fs(USER_DS);
1270         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1271                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1272         flush_thread();
1273         current->personality &= ~bprm->per_clear;
1274
1275         /*
1276          * We have to apply CLOEXEC before we change whether the process is
1277          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1278          * trying to access the should-be-closed file descriptors of a process
1279          * undergoing exec(2).
1280          */
1281         do_close_on_exec(current->files);
1282         return 0;
1283
1284 out:
1285         return retval;
1286 }
1287 EXPORT_SYMBOL(flush_old_exec);
1288
1289 void would_dump(struct linux_binprm *bprm, struct file *file)
1290 {
1291         struct inode *inode = file_inode(file);
1292         if (inode_permission(inode, MAY_READ) < 0) {
1293                 struct user_namespace *old, *user_ns;
1294                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1295
1296                 /* Ensure mm->user_ns contains the executable */
1297                 user_ns = old = bprm->mm->user_ns;
1298                 while ((user_ns != &init_user_ns) &&
1299                        !privileged_wrt_inode_uidgid(user_ns, inode))
1300                         user_ns = user_ns->parent;
1301
1302                 if (old != user_ns) {
1303                         bprm->mm->user_ns = get_user_ns(user_ns);
1304                         put_user_ns(old);
1305                 }
1306         }
1307 }
1308 EXPORT_SYMBOL(would_dump);
1309
1310 void setup_new_exec(struct linux_binprm * bprm)
1311 {
1312         arch_pick_mmap_layout(current->mm);
1313
1314         /* This is the point of no return */
1315         current->sas_ss_sp = current->sas_ss_size = 0;
1316
1317         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1318                 set_dumpable(current->mm, SUID_DUMP_USER);
1319         else
1320                 set_dumpable(current->mm, suid_dumpable);
1321
1322         perf_event_exec();
1323         __set_task_comm(current, kbasename(bprm->filename), true);
1324
1325         /* Set the new mm task size. We have to do that late because it may
1326          * depend on TIF_32BIT which is only updated in flush_thread() on
1327          * some architectures like powerpc
1328          */
1329         current->mm->task_size = TASK_SIZE;
1330
1331         /* install the new credentials */
1332         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1333             !gid_eq(bprm->cred->gid, current_egid())) {
1334                 current->pdeath_signal = 0;
1335         } else {
1336                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1337                         set_dumpable(current->mm, suid_dumpable);
1338         }
1339
1340         /* An exec changes our domain. We are no longer part of the thread
1341            group */
1342         current->self_exec_id++;
1343         flush_signal_handlers(current, 0);
1344 }
1345 EXPORT_SYMBOL(setup_new_exec);
1346
1347 /*
1348  * Prepare credentials and lock ->cred_guard_mutex.
1349  * install_exec_creds() commits the new creds and drops the lock.
1350  * Or, if exec fails before, free_bprm() should release ->cred and
1351  * and unlock.
1352  */
1353 int prepare_bprm_creds(struct linux_binprm *bprm)
1354 {
1355         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1356                 return -ERESTARTNOINTR;
1357
1358         bprm->cred = prepare_exec_creds();
1359         if (likely(bprm->cred))
1360                 return 0;
1361
1362         mutex_unlock(&current->signal->cred_guard_mutex);
1363         return -ENOMEM;
1364 }
1365
1366 static void free_bprm(struct linux_binprm *bprm)
1367 {
1368         free_arg_pages(bprm);
1369         if (bprm->cred) {
1370                 mutex_unlock(&current->signal->cred_guard_mutex);
1371                 abort_creds(bprm->cred);
1372         }
1373         if (bprm->file) {
1374                 allow_write_access(bprm->file);
1375                 fput(bprm->file);
1376         }
1377         /* If a binfmt changed the interp, free it. */
1378         if (bprm->interp != bprm->filename)
1379                 kfree(bprm->interp);
1380         kfree(bprm);
1381 }
1382
1383 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1384 {
1385         /* If a binfmt changed the interp, free it first. */
1386         if (bprm->interp != bprm->filename)
1387                 kfree(bprm->interp);
1388         bprm->interp = kstrdup(interp, GFP_KERNEL);
1389         if (!bprm->interp)
1390                 return -ENOMEM;
1391         return 0;
1392 }
1393 EXPORT_SYMBOL(bprm_change_interp);
1394
1395 /*
1396  * install the new credentials for this executable
1397  */
1398 void install_exec_creds(struct linux_binprm *bprm)
1399 {
1400         security_bprm_committing_creds(bprm);
1401
1402         commit_creds(bprm->cred);
1403         bprm->cred = NULL;
1404
1405         /*
1406          * Disable monitoring for regular users
1407          * when executing setuid binaries. Must
1408          * wait until new credentials are committed
1409          * by commit_creds() above
1410          */
1411         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1412                 perf_event_exit_task(current);
1413         /*
1414          * cred_guard_mutex must be held at least to this point to prevent
1415          * ptrace_attach() from altering our determination of the task's
1416          * credentials; any time after this it may be unlocked.
1417          */
1418         security_bprm_committed_creds(bprm);
1419         mutex_unlock(&current->signal->cred_guard_mutex);
1420 }
1421 EXPORT_SYMBOL(install_exec_creds);
1422
1423 /*
1424  * determine how safe it is to execute the proposed program
1425  * - the caller must hold ->cred_guard_mutex to protect against
1426  *   PTRACE_ATTACH or seccomp thread-sync
1427  */
1428 static void check_unsafe_exec(struct linux_binprm *bprm)
1429 {
1430         struct task_struct *p = current, *t;
1431         unsigned n_fs;
1432
1433         if (p->ptrace)
1434                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1435
1436         /*
1437          * This isn't strictly necessary, but it makes it harder for LSMs to
1438          * mess up.
1439          */
1440         if (task_no_new_privs(current))
1441                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1442
1443         t = p;
1444         n_fs = 1;
1445         spin_lock(&p->fs->lock);
1446         rcu_read_lock();
1447         while_each_thread(p, t) {
1448                 if (t->fs == p->fs)
1449                         n_fs++;
1450         }
1451         rcu_read_unlock();
1452
1453         if (p->fs->users > n_fs)
1454                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1455         else
1456                 p->fs->in_exec = 1;
1457         spin_unlock(&p->fs->lock);
1458 }
1459
1460 static void bprm_fill_uid(struct linux_binprm *bprm)
1461 {
1462         struct inode *inode;
1463         unsigned int mode;
1464         kuid_t uid;
1465         kgid_t gid;
1466
1467         /*
1468          * Since this can be called multiple times (via prepare_binprm),
1469          * we must clear any previous work done when setting set[ug]id
1470          * bits from any earlier bprm->file uses (for example when run
1471          * first for a setuid script then again for its interpreter).
1472          */
1473         bprm->cred->euid = current_euid();
1474         bprm->cred->egid = current_egid();
1475
1476         if (!mnt_may_suid(bprm->file->f_path.mnt))
1477                 return;
1478
1479         if (task_no_new_privs(current))
1480                 return;
1481
1482         inode = bprm->file->f_path.dentry->d_inode;
1483         mode = READ_ONCE(inode->i_mode);
1484         if (!(mode & (S_ISUID|S_ISGID)))
1485                 return;
1486
1487         /* Be careful if suid/sgid is set */
1488         inode_lock(inode);
1489
1490         /* reload atomically mode/uid/gid now that lock held */
1491         mode = inode->i_mode;
1492         uid = inode->i_uid;
1493         gid = inode->i_gid;
1494         inode_unlock(inode);
1495
1496         /* We ignore suid/sgid if there are no mappings for them in the ns */
1497         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1498                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1499                 return;
1500
1501         if (mode & S_ISUID) {
1502                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1503                 bprm->cred->euid = uid;
1504         }
1505
1506         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1507                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1508                 bprm->cred->egid = gid;
1509         }
1510 }
1511
1512 /*
1513  * Fill the binprm structure from the inode.
1514  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1515  *
1516  * This may be called multiple times for binary chains (scripts for example).
1517  */
1518 int prepare_binprm(struct linux_binprm *bprm)
1519 {
1520         int retval;
1521
1522         bprm_fill_uid(bprm);
1523
1524         /* fill in binprm security blob */
1525         retval = security_bprm_set_creds(bprm);
1526         if (retval)
1527                 return retval;
1528         bprm->cred_prepared = 1;
1529
1530         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1531         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1532 }
1533
1534 EXPORT_SYMBOL(prepare_binprm);
1535
1536 /*
1537  * Arguments are '\0' separated strings found at the location bprm->p
1538  * points to; chop off the first by relocating brpm->p to right after
1539  * the first '\0' encountered.
1540  */
1541 int remove_arg_zero(struct linux_binprm *bprm)
1542 {
1543         int ret = 0;
1544         unsigned long offset;
1545         char *kaddr;
1546         struct page *page;
1547
1548         if (!bprm->argc)
1549                 return 0;
1550
1551         do {
1552                 offset = bprm->p & ~PAGE_MASK;
1553                 page = get_arg_page(bprm, bprm->p, 0);
1554                 if (!page) {
1555                         ret = -EFAULT;
1556                         goto out;
1557                 }
1558                 kaddr = kmap_atomic(page);
1559
1560                 for (; offset < PAGE_SIZE && kaddr[offset];
1561                                 offset++, bprm->p++)
1562                         ;
1563
1564                 kunmap_atomic(kaddr);
1565                 put_arg_page(page);
1566         } while (offset == PAGE_SIZE);
1567
1568         bprm->p++;
1569         bprm->argc--;
1570         ret = 0;
1571
1572 out:
1573         return ret;
1574 }
1575 EXPORT_SYMBOL(remove_arg_zero);
1576
1577 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1578 /*
1579  * cycle the list of binary formats handler, until one recognizes the image
1580  */
1581 int search_binary_handler(struct linux_binprm *bprm)
1582 {
1583         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1584         struct linux_binfmt *fmt;
1585         int retval;
1586
1587         /* This allows 4 levels of binfmt rewrites before failing hard. */
1588         if (bprm->recursion_depth > 5)
1589                 return -ELOOP;
1590
1591         retval = security_bprm_check(bprm);
1592         if (retval)
1593                 return retval;
1594
1595         retval = -ENOENT;
1596  retry:
1597         read_lock(&binfmt_lock);
1598         list_for_each_entry(fmt, &formats, lh) {
1599                 if (!try_module_get(fmt->module))
1600                         continue;
1601                 read_unlock(&binfmt_lock);
1602                 bprm->recursion_depth++;
1603                 retval = fmt->load_binary(bprm);
1604                 read_lock(&binfmt_lock);
1605                 put_binfmt(fmt);
1606                 bprm->recursion_depth--;
1607                 if (retval < 0 && !bprm->mm) {
1608                         /* we got to flush_old_exec() and failed after it */
1609                         read_unlock(&binfmt_lock);
1610                         force_sigsegv(SIGSEGV, current);
1611                         return retval;
1612                 }
1613                 if (retval != -ENOEXEC || !bprm->file) {
1614                         read_unlock(&binfmt_lock);
1615                         return retval;
1616                 }
1617         }
1618         read_unlock(&binfmt_lock);
1619
1620         if (need_retry) {
1621                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1622                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1623                         return retval;
1624                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1625                         return retval;
1626                 need_retry = false;
1627                 goto retry;
1628         }
1629
1630         return retval;
1631 }
1632 EXPORT_SYMBOL(search_binary_handler);
1633
1634 static int exec_binprm(struct linux_binprm *bprm)
1635 {
1636         pid_t old_pid, old_vpid;
1637         int ret;
1638
1639         /* Need to fetch pid before load_binary changes it */
1640         old_pid = current->pid;
1641         rcu_read_lock();
1642         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1643         rcu_read_unlock();
1644
1645         ret = search_binary_handler(bprm);
1646         if (ret >= 0) {
1647                 audit_bprm(bprm);
1648                 trace_sched_process_exec(current, old_pid, bprm);
1649                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1650                 proc_exec_connector(current);
1651         }
1652
1653         return ret;
1654 }
1655
1656 /*
1657  * sys_execve() executes a new program.
1658  */
1659 static int do_execveat_common(int fd, struct filename *filename,
1660                               struct user_arg_ptr argv,
1661                               struct user_arg_ptr envp,
1662                               int flags)
1663 {
1664         char *pathbuf = NULL;
1665         struct linux_binprm *bprm;
1666         struct file *file;
1667         struct files_struct *displaced;
1668         int retval;
1669
1670         if (IS_ERR(filename))
1671                 return PTR_ERR(filename);
1672
1673         /*
1674          * We move the actual failure in case of RLIMIT_NPROC excess from
1675          * set*uid() to execve() because too many poorly written programs
1676          * don't check setuid() return code.  Here we additionally recheck
1677          * whether NPROC limit is still exceeded.
1678          */
1679         if ((current->flags & PF_NPROC_EXCEEDED) &&
1680             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1681                 retval = -EAGAIN;
1682                 goto out_ret;
1683         }
1684
1685         /* We're below the limit (still or again), so we don't want to make
1686          * further execve() calls fail. */
1687         current->flags &= ~PF_NPROC_EXCEEDED;
1688
1689         retval = unshare_files(&displaced);
1690         if (retval)
1691                 goto out_ret;
1692
1693         retval = -ENOMEM;
1694         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1695         if (!bprm)
1696                 goto out_files;
1697
1698         retval = prepare_bprm_creds(bprm);
1699         if (retval)
1700                 goto out_free;
1701
1702         check_unsafe_exec(bprm);
1703         current->in_execve = 1;
1704
1705         file = do_open_execat(fd, filename, flags);
1706         retval = PTR_ERR(file);
1707         if (IS_ERR(file))
1708                 goto out_unmark;
1709
1710         sched_exec();
1711
1712         bprm->file = file;
1713         if (fd == AT_FDCWD || filename->name[0] == '/') {
1714                 bprm->filename = filename->name;
1715         } else {
1716                 if (filename->name[0] == '\0')
1717                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1718                 else
1719                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1720                                             fd, filename->name);
1721                 if (!pathbuf) {
1722                         retval = -ENOMEM;
1723                         goto out_unmark;
1724                 }
1725                 /*
1726                  * Record that a name derived from an O_CLOEXEC fd will be
1727                  * inaccessible after exec. Relies on having exclusive access to
1728                  * current->files (due to unshare_files above).
1729                  */
1730                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1731                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1732                 bprm->filename = pathbuf;
1733         }
1734         bprm->interp = bprm->filename;
1735
1736         retval = bprm_mm_init(bprm);
1737         if (retval)
1738                 goto out_unmark;
1739
1740         bprm->argc = count(argv, MAX_ARG_STRINGS);
1741         if ((retval = bprm->argc) < 0)
1742                 goto out;
1743
1744         bprm->envc = count(envp, MAX_ARG_STRINGS);
1745         if ((retval = bprm->envc) < 0)
1746                 goto out;
1747
1748         retval = prepare_binprm(bprm);
1749         if (retval < 0)
1750                 goto out;
1751
1752         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1753         if (retval < 0)
1754                 goto out;
1755
1756         bprm->exec = bprm->p;
1757         retval = copy_strings(bprm->envc, envp, bprm);
1758         if (retval < 0)
1759                 goto out;
1760
1761         retval = copy_strings(bprm->argc, argv, bprm);
1762         if (retval < 0)
1763                 goto out;
1764
1765         would_dump(bprm, bprm->file);
1766
1767         retval = exec_binprm(bprm);
1768         if (retval < 0)
1769                 goto out;
1770
1771         /* execve succeeded */
1772         current->fs->in_exec = 0;
1773         current->in_execve = 0;
1774         acct_update_integrals(current);
1775         task_numa_free(current);
1776         free_bprm(bprm);
1777         kfree(pathbuf);
1778         putname(filename);
1779         if (displaced)
1780                 put_files_struct(displaced);
1781         return retval;
1782
1783 out:
1784         if (bprm->mm) {
1785                 acct_arg_size(bprm, 0);
1786                 mmput(bprm->mm);
1787         }
1788
1789 out_unmark:
1790         current->fs->in_exec = 0;
1791         current->in_execve = 0;
1792
1793 out_free:
1794         free_bprm(bprm);
1795         kfree(pathbuf);
1796
1797 out_files:
1798         if (displaced)
1799                 reset_files_struct(displaced);
1800 out_ret:
1801         putname(filename);
1802         return retval;
1803 }
1804
1805 int do_execve(struct filename *filename,
1806         const char __user *const __user *__argv,
1807         const char __user *const __user *__envp)
1808 {
1809         struct user_arg_ptr argv = { .ptr.native = __argv };
1810         struct user_arg_ptr envp = { .ptr.native = __envp };
1811         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1812 }
1813
1814 int do_execveat(int fd, struct filename *filename,
1815                 const char __user *const __user *__argv,
1816                 const char __user *const __user *__envp,
1817                 int flags)
1818 {
1819         struct user_arg_ptr argv = { .ptr.native = __argv };
1820         struct user_arg_ptr envp = { .ptr.native = __envp };
1821
1822         return do_execveat_common(fd, filename, argv, envp, flags);
1823 }
1824
1825 #ifdef CONFIG_COMPAT
1826 static int compat_do_execve(struct filename *filename,
1827         const compat_uptr_t __user *__argv,
1828         const compat_uptr_t __user *__envp)
1829 {
1830         struct user_arg_ptr argv = {
1831                 .is_compat = true,
1832                 .ptr.compat = __argv,
1833         };
1834         struct user_arg_ptr envp = {
1835                 .is_compat = true,
1836                 .ptr.compat = __envp,
1837         };
1838         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1839 }
1840
1841 static int compat_do_execveat(int fd, struct filename *filename,
1842                               const compat_uptr_t __user *__argv,
1843                               const compat_uptr_t __user *__envp,
1844                               int flags)
1845 {
1846         struct user_arg_ptr argv = {
1847                 .is_compat = true,
1848                 .ptr.compat = __argv,
1849         };
1850         struct user_arg_ptr envp = {
1851                 .is_compat = true,
1852                 .ptr.compat = __envp,
1853         };
1854         return do_execveat_common(fd, filename, argv, envp, flags);
1855 }
1856 #endif
1857
1858 void set_binfmt(struct linux_binfmt *new)
1859 {
1860         struct mm_struct *mm = current->mm;
1861
1862         if (mm->binfmt)
1863                 module_put(mm->binfmt->module);
1864
1865         mm->binfmt = new;
1866         if (new)
1867                 __module_get(new->module);
1868 }
1869 EXPORT_SYMBOL(set_binfmt);
1870
1871 /*
1872  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1873  */
1874 void set_dumpable(struct mm_struct *mm, int value)
1875 {
1876         unsigned long old, new;
1877
1878         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1879                 return;
1880
1881         do {
1882                 old = ACCESS_ONCE(mm->flags);
1883                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1884         } while (cmpxchg(&mm->flags, old, new) != old);
1885 }
1886
1887 SYSCALL_DEFINE3(execve,
1888                 const char __user *, filename,
1889                 const char __user *const __user *, argv,
1890                 const char __user *const __user *, envp)
1891 {
1892         return do_execve(getname(filename), argv, envp);
1893 }
1894
1895 SYSCALL_DEFINE5(execveat,
1896                 int, fd, const char __user *, filename,
1897                 const char __user *const __user *, argv,
1898                 const char __user *const __user *, envp,
1899                 int, flags)
1900 {
1901         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1902
1903         return do_execveat(fd,
1904                            getname_flags(filename, lookup_flags, NULL),
1905                            argv, envp, flags);
1906 }
1907
1908 #ifdef CONFIG_COMPAT
1909 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1910         const compat_uptr_t __user *, argv,
1911         const compat_uptr_t __user *, envp)
1912 {
1913         return compat_do_execve(getname(filename), argv, envp);
1914 }
1915
1916 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1917                        const char __user *, filename,
1918                        const compat_uptr_t __user *, argv,
1919                        const compat_uptr_t __user *, envp,
1920                        int,  flags)
1921 {
1922         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1923
1924         return compat_do_execveat(fd,
1925                                   getname_flags(filename, lookup_flags, NULL),
1926                                   argv, envp, flags);
1927 }
1928 #endif