fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/fdtable.h>
  29 #include <linux/mm.h>
  30 #include <linux/vmacache.h>
  31 #include <linux/stat.h>
  32 #include <linux/fcntl.h>
  33 #include <linux/swap.h>
  34 #include <linux/string.h>
  35 #include <linux/init.h>
  36 #include <linux/sched/mm.h>
  37 #include <linux/sched/coredump.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/sched/numa_balancing.h>
  40 #include <linux/sched/task.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/perf_event.h>
  43 #include <linux/highmem.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/key.h>
  46 #include <linux/personality.h>
  47 #include <linux/binfmts.h>
  48 #include <linux/utsname.h>
  49 #include <linux/pid_namespace.h>
  50 #include <linux/module.h>
  51 #include <linux/namei.h>
  52 #include <linux/mount.h>
  53 #include <linux/security.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/audit.h>
  58 #include <linux/tracehook.h>
  59 #include <linux/kmod.h>
  60 #include <linux/fsnotify.h>
  61 #include <linux/fs_struct.h>
  62 #include <linux/pipe_fs_i.h>
  63 #include <linux/oom.h>
  64 #include <linux/compat.h>
  65 #include <linux/vmalloc.h>
  66
  67 #include <linux/uaccess.h>
  68 #include <asm/mmu_context.h>
  69 #include <asm/tlb.h>
  70
  71 #include <trace/events/task.h>
  72 #include "internal.h"
  73
  74 #include <trace/events/sched.h>
  75
  76 int suid_dumpable = 0;
  77
  78 static LIST_HEAD(formats);
  79 static DEFINE_RWLOCK(binfmt_lock);
  80
  81 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  82 {
  83         BUG_ON(!fmt);
  84         if (WARN_ON(!fmt->load_binary))
  85                 return;
  86         write_lock(&binfmt_lock);
  87         insert ? list_add(&fmt->lh, &formats) :
  88                  list_add_tail(&fmt->lh, &formats);
  89         write_unlock(&binfmt_lock);
  90 }
  91
  92 EXPORT_SYMBOL(__register_binfmt);
  93
  94 void unregister_binfmt(struct linux_binfmt * fmt)
  95 {
  96         write_lock(&binfmt_lock);
  97         list_del(&fmt->lh);
  98         write_unlock(&binfmt_lock);
  99 }
 100
 101 EXPORT_SYMBOL(unregister_binfmt);
 102
 103 static inline void put_binfmt(struct linux_binfmt * fmt)
 104 {
 105         module_put(fmt->module);
 106 }
 107
 108 bool path_noexec(const struct path *path)
 109 {
 110         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 111                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 112 }
 113
 114 #ifdef CONFIG_USELIB
 115 /*
 116  * Note that a shared library must be both readable and executable due to
 117  * security reasons.
 118  *
 119  * Also note that we take the address to load from from the file itself.
 120  */
 121 SYSCALL_DEFINE1(uselib, const char __user *, library)
 122 {
 123         struct linux_binfmt *fmt;
 124         struct file *file;
 125         struct filename *tmp = getname(library);
 126         int error = PTR_ERR(tmp);
 127         static const struct open_flags uselib_flags = {
 128                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 129                 .acc_mode = MAY_READ | MAY_EXEC,
 130                 .intent = LOOKUP_OPEN,
 131                 .lookup_flags = LOOKUP_FOLLOW,
 132         };
 133
 134         if (IS_ERR(tmp))
 135                 goto out;
 136
 137         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 138         putname(tmp);
 139         error = PTR_ERR(file);
 140         if (IS_ERR(file))
 141                 goto out;
 142
 143         error = -EINVAL;
 144         if (!S_ISREG(file_inode(file)->i_mode))
 145                 goto exit;
 146
 147         error = -EACCES;
 148         if (path_noexec(&file->f_path))
 149                 goto exit;
 150
 151         fsnotify_open(file);
 152
 153         error = -ENOEXEC;
 154
 155         read_lock(&binfmt_lock);
 156         list_for_each_entry(fmt, &formats, lh) {
 157                 if (!fmt->load_shlib)
 158                         continue;
 159                 if (!try_module_get(fmt->module))
 160                         continue;
 161                 read_unlock(&binfmt_lock);
 162                 error = fmt->load_shlib(file);
 163                 read_lock(&binfmt_lock);
 164                 put_binfmt(fmt);
 165                 if (error != -ENOEXEC)
 166                         break;
 167         }
 168         read_unlock(&binfmt_lock);
 169 exit:
 170         fput(file);
 171 out:
 172         return error;
 173 }
 174 #endif /* #ifdef CONFIG_USELIB */
 175
 176 #ifdef CONFIG_MMU
 177 /*
 178  * The nascent bprm->mm is not visible until exec_mmap() but it can
 179  * use a lot of memory, account these pages in current->mm temporary
 180  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 181  * change the counter back via acct_arg_size(0).
 182  */
 183 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 184 {
 185         struct mm_struct *mm = current->mm;
 186         long diff = (long)(pages - bprm->vma_pages);
 187
 188         if (!mm || !diff)
 189                 return;
 190
 191         bprm->vma_pages = pages;
 192         add_mm_counter(mm, MM_ANONPAGES, diff);
 193 }
 194
 195 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 196                 int write)
 197 {
 198         struct page *page;
 199         int ret;
 200         unsigned int gup_flags = FOLL_FORCE;
 201
 202 #ifdef CONFIG_STACK_GROWSUP
 203         if (write) {
 204                 ret = expand_downwards(bprm->vma, pos);
 205                 if (ret < 0)
 206                         return NULL;
 207         }
 208 #endif
 209
 210         if (write)
 211                 gup_flags |= FOLL_WRITE;
 212
 213         /*
 214          * We are doing an exec().  'current' is the process
 215          * doing the exec and bprm->mm is the new process's mm.
 216          */
 217         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 218                         &page, NULL, NULL);
 219         if (ret <= 0)
 220                 return NULL;
 221
 222         if (write)
 223                 acct_arg_size(bprm, vma_pages(bprm->vma));
 224
 225         return page;
 226 }
 227
 228 static void put_arg_page(struct page *page)
 229 {
 230         put_page(page);
 231 }
 232
 233 static void free_arg_pages(struct linux_binprm *bprm)
 234 {
 235 }
 236
 237 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 238                 struct page *page)
 239 {
 240         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 241 }
 242
 243 static int __bprm_mm_init(struct linux_binprm *bprm)
 244 {
 245         int err;
 246         struct vm_area_struct *vma = NULL;
 247         struct mm_struct *mm = bprm->mm;
 248
 249         bprm->vma = vma = vm_area_alloc(mm);
 250         if (!vma)
 251                 return -ENOMEM;
 252         vma_set_anonymous(vma);
 253
 254         if (down_write_killable(&mm->mmap_sem)) {
 255                 err = -EINTR;
 256                 goto err_free;
 257         }
 258
 259         /*
 260          * Place the stack at the largest stack address the architecture
 261          * supports. Later, we'll move this to an appropriate place. We don't
 262          * use STACK_TOP because that can depend on attributes which aren't
 263          * configured yet.
 264          */
 265         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 266         vma->vm_end = STACK_TOP_MAX;
 267         vma->vm_start = vma->vm_end - PAGE_SIZE;
 268         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 269         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 270
 271         err = insert_vm_struct(mm, vma);
 272         if (err)
 273                 goto err;
 274
 275         mm->stack_vm = mm->total_vm = 1;
 276         arch_bprm_mm_init(mm, vma);
 277         up_write(&mm->mmap_sem);
 278         bprm->p = vma->vm_end - sizeof(void *);
 279         return 0;
 280 err:
 281         up_write(&mm->mmap_sem);
 282 err_free:
 283         bprm->vma = NULL;
 284         vm_area_free(vma);
 285         return err;
 286 }
 287
 288 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 289 {
 290         return len <= MAX_ARG_STRLEN;
 291 }
 292
 293 #else
 294
 295 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 296 {
 297 }
 298
 299 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 300                 int write)
 301 {
 302         struct page *page;
 303
 304         page = bprm->page[pos / PAGE_SIZE];
 305         if (!page && write) {
 306                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 307                 if (!page)
 308                         return NULL;
 309                 bprm->page[pos / PAGE_SIZE] = page;
 310         }
 311
 312         return page;
 313 }
 314
 315 static void put_arg_page(struct page *page)
 316 {
 317 }
 318
 319 static void free_arg_page(struct linux_binprm *bprm, int i)
 320 {
 321         if (bprm->page[i]) {
 322                 __free_page(bprm->page[i]);
 323                 bprm->page[i] = NULL;
 324         }
 325 }
 326
 327 static void free_arg_pages(struct linux_binprm *bprm)
 328 {
 329         int i;
 330
 331         for (i = 0; i < MAX_ARG_PAGES; i++)
 332                 free_arg_page(bprm, i);
 333 }
 334
 335 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 336                 struct page *page)
 337 {
 338 }
 339
 340 static int __bprm_mm_init(struct linux_binprm *bprm)
 341 {
 342         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 343         return 0;
 344 }
 345
 346 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 347 {
 348         return len <= bprm->p;
 349 }
 350
 351 #endif /* CONFIG_MMU */
 352
 353 /*
 354  * Create a new mm_struct and populate it with a temporary stack
 355  * vm_area_struct.  We don't have enough context at this point to set the stack
 356  * flags, permissions, and offset, so we use temporary values.  We'll update
 357  * them later in setup_arg_pages().
 358  */
 359 static int bprm_mm_init(struct linux_binprm *bprm)
 360 {
 361         int err;
 362         struct mm_struct *mm = NULL;
 363
 364         bprm->mm = mm = mm_alloc();
 365         err = -ENOMEM;
 366         if (!mm)
 367                 goto err;
 368
 369         /* Save current stack limit for all calculations made during exec. */
 370         task_lock(current->group_leader);
 371         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 372         task_unlock(current->group_leader);
 373
 374         err = __bprm_mm_init(bprm);
 375         if (err)
 376                 goto err;
 377
 378         return 0;
 379
 380 err:
 381         if (mm) {
 382                 bprm->mm = NULL;
 383                 mmdrop(mm);
 384         }
 385
 386         return err;
 387 }
 388
 389 struct user_arg_ptr {
 390 #ifdef CONFIG_COMPAT
 391         bool is_compat;
 392 #endif
 393         union {
 394                 const char __user *const __user *native;
 395 #ifdef CONFIG_COMPAT
 396                 const compat_uptr_t __user *compat;
 397 #endif
 398         } ptr;
 399 };
 400
 401 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 402 {
 403         const char __user *native;
 404
 405 #ifdef CONFIG_COMPAT
 406         if (unlikely(argv.is_compat)) {
 407                 compat_uptr_t compat;
 408
 409                 if (get_user(compat, argv.ptr.compat + nr))
 410                         return ERR_PTR(-EFAULT);
 411
 412                 return compat_ptr(compat);
 413         }
 414 #endif
 415
 416         if (get_user(native, argv.ptr.native + nr))
 417                 return ERR_PTR(-EFAULT);
 418
 419         return native;
 420 }
 421
 422 /*
 423  * count() counts the number of strings in array ARGV.
 424  */
 425 static int count(struct user_arg_ptr argv, int max)
 426 {
 427         int i = 0;
 428
 429         if (argv.ptr.native != NULL) {
 430                 for (;;) {
 431                         const char __user *p = get_user_arg_ptr(argv, i);
 432
 433                         if (!p)
 434                                 break;
 435
 436                         if (IS_ERR(p))
 437                                 return -EFAULT;
 438
 439                         if (i >= max)
 440                                 return -E2BIG;
 441                         ++i;
 442
 443                         if (fatal_signal_pending(current))
 444                                 return -ERESTARTNOHAND;
 445                         cond_resched();
 446                 }
 447         }
 448         return i;
 449 }
 450
 451 static int prepare_arg_pages(struct linux_binprm *bprm,
 452                         struct user_arg_ptr argv, struct user_arg_ptr envp)
 453 {
 454         unsigned long limit, ptr_size;
 455
 456         bprm->argc = count(argv, MAX_ARG_STRINGS);
 457         if (bprm->argc == 0)
 458                 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
 459                              current->comm, bprm->filename);
 460         if (bprm->argc < 0)
 461                 return bprm->argc;
 462
 463         bprm->envc = count(envp, MAX_ARG_STRINGS);
 464         if (bprm->envc < 0)
 465                 return bprm->envc;
 466
 467         /*
 468          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 469          * (whichever is smaller) for the argv+env strings.
 470          * This ensures that:
 471          *  - the remaining binfmt code will not run out of stack space,
 472          *  - the program will have a reasonable amount of stack left
 473          *    to work from.
 474          */
 475         limit = _STK_LIM / 4 * 3;
 476         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 477         /*
 478          * We've historically supported up to 32 pages (ARG_MAX)
 479          * of argument strings even with small stacks
 480          */
 481         limit = max_t(unsigned long, limit, ARG_MAX);
 482         /*
 483          * We must account for the size of all the argv and envp pointers to
 484          * the argv and envp strings, since they will also take up space in
 485          * the stack. They aren't stored until much later when we can't
 486          * signal to the parent that the child has run out of stack space.
 487          * Instead, calculate it here so it's possible to fail gracefully.
 488          *
 489          * In the case of argc = 0, make sure there is space for adding a
 490          * empty string (which will bump argc to 1), to ensure confused
 491          * userspace programs don't start processing from argv[1], thinking
 492          * argc can never be 0, to keep them from walking envp by accident.
 493          * See do_execveat_common().
 494          */
 495         ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
 496         if (limit <= ptr_size)
 497                 return -E2BIG;
 498         limit -= ptr_size;
 499
 500         bprm->argmin = bprm->p - limit;
 501         return 0;
 502 }
 503
 504 /*
 505  * 'copy_strings()' copies argument/environment strings from the old
 506  * processes's memory to the new process's stack.  The call to get_user_pages()
 507  * ensures the destination page is created and not swapped out.
 508  */
 509 static int copy_strings(int argc, struct user_arg_ptr argv,
 510                         struct linux_binprm *bprm)
 511 {
 512         struct page *kmapped_page = NULL;
 513         char *kaddr = NULL;
 514         unsigned long kpos = 0;
 515         int ret;
 516
 517         while (argc-- > 0) {
 518                 const char __user *str;
 519                 int len;
 520                 unsigned long pos;
 521
 522                 ret = -EFAULT;
 523                 str = get_user_arg_ptr(argv, argc);
 524                 if (IS_ERR(str))
 525                         goto out;
 526
 527                 len = strnlen_user(str, MAX_ARG_STRLEN);
 528                 if (!len)
 529                         goto out;
 530
 531                 ret = -E2BIG;
 532                 if (!valid_arg_len(bprm, len))
 533                         goto out;
 534
 535                 /* We're going to work our way backwords. */
 536                 pos = bprm->p;
 537                 str += len;
 538                 bprm->p -= len;
 539 #ifdef CONFIG_MMU
 540                 if (bprm->p < bprm->argmin)
 541                         goto out;
 542 #endif
 543
 544                 while (len > 0) {
 545                         int offset, bytes_to_copy;
 546
 547                         if (fatal_signal_pending(current)) {
 548                                 ret = -ERESTARTNOHAND;
 549                                 goto out;
 550                         }
 551                         cond_resched();
 552
 553                         offset = pos % PAGE_SIZE;
 554                         if (offset == 0)
 555                                 offset = PAGE_SIZE;
 556
 557                         bytes_to_copy = offset;
 558                         if (bytes_to_copy > len)
 559                                 bytes_to_copy = len;
 560
 561                         offset -= bytes_to_copy;
 562                         pos -= bytes_to_copy;
 563                         str -= bytes_to_copy;
 564                         len -= bytes_to_copy;
 565
 566                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 567                                 struct page *page;
 568
 569                                 page = get_arg_page(bprm, pos, 1);
 570                                 if (!page) {
 571                                         ret = -E2BIG;
 572                                         goto out;
 573                                 }
 574
 575                                 if (kmapped_page) {
 576                                         flush_kernel_dcache_page(kmapped_page);
 577                                         kunmap(kmapped_page);
 578                                         put_arg_page(kmapped_page);
 579                                 }
 580                                 kmapped_page = page;
 581                                 kaddr = kmap(kmapped_page);
 582                                 kpos = pos & PAGE_MASK;
 583                                 flush_arg_page(bprm, kpos, kmapped_page);
 584                         }
 585                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 586                                 ret = -EFAULT;
 587                                 goto out;
 588                         }
 589                 }
 590         }
 591         ret = 0;
 592 out:
 593         if (kmapped_page) {
 594                 flush_kernel_dcache_page(kmapped_page);
 595                 kunmap(kmapped_page);
 596                 put_arg_page(kmapped_page);
 597         }
 598         return ret;
 599 }
 600
 601 /*
 602  * Like copy_strings, but get argv and its values from kernel memory.
 603  */
 604 int copy_strings_kernel(int argc, const char *const *__argv,
 605                         struct linux_binprm *bprm)
 606 {
 607         int r;
 608         mm_segment_t oldfs = get_fs();
 609         struct user_arg_ptr argv = {
 610                 .ptr.native = (const char __user *const  __user *)__argv,
 611         };
 612
 613         set_fs(KERNEL_DS);
 614         r = copy_strings(argc, argv, bprm);
 615         set_fs(oldfs);
 616
 617         return r;
 618 }
 619 EXPORT_SYMBOL(copy_strings_kernel);
 620
 621 #ifdef CONFIG_MMU
 622
 623 /*
 624  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 625  * the binfmt code determines where the new stack should reside, we shift it to
 626  * its final location.  The process proceeds as follows:
 627  *
 628  * 1) Use shift to calculate the new vma endpoints.
 629  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 630  *    arguments passed to subsequent functions are consistent.
 631  * 3) Move vma's page tables to the new range.
 632  * 4) Free up any cleared pgd range.
 633  * 5) Shrink the vma to cover only the new range.
 634  */
 635 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 636 {
 637         struct mm_struct *mm = vma->vm_mm;
 638         unsigned long old_start = vma->vm_start;
 639         unsigned long old_end = vma->vm_end;
 640         unsigned long length = old_end - old_start;
 641         unsigned long new_start = old_start - shift;
 642         unsigned long new_end = old_end - shift;
 643         struct mmu_gather tlb;
 644
 645         BUG_ON(new_start > new_end);
 646
 647         /*
 648          * ensure there are no vmas between where we want to go
 649          * and where we are
 650          */
 651         if (vma != find_vma(mm, new_start))
 652                 return -EFAULT;
 653
 654         /*
 655          * cover the whole range: [new_start, old_end)
 656          */
 657         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 658                 return -ENOMEM;
 659
 660         /*
 661          * move the page tables downwards, on failure we rely on
 662          * process cleanup to remove whatever mess we made.
 663          */
 664         if (length != move_page_tables(vma, old_start,
 665                                        vma, new_start, length, false))
 666                 return -ENOMEM;
 667
 668         lru_add_drain();
 669         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 670         if (new_end > old_start) {
 671                 /*
 672                  * when the old and new regions overlap clear from new_end.
 673                  */
 674                 free_pgd_range(&tlb, new_end, old_end, new_end,
 675                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 676         } else {
 677                 /*
 678                  * otherwise, clean from old_start; this is done to not touch
 679                  * the address space in [new_end, old_start) some architectures
 680                  * have constraints on va-space that make this illegal (IA64) -
 681                  * for the others its just a little faster.
 682                  */
 683                 free_pgd_range(&tlb, old_start, old_end, new_end,
 684                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 685         }
 686         tlb_finish_mmu(&tlb, old_start, old_end);
 687
 688         /*
 689          * Shrink the vma to just the new range.  Always succeeds.
 690          */
 691         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 692
 693         return 0;
 694 }
 695
 696 /*
 697  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 698  * the stack is optionally relocated, and some extra space is added.
 699  */
 700 int setup_arg_pages(struct linux_binprm *bprm,
 701                     unsigned long stack_top,
 702                     int executable_stack)
 703 {
 704         unsigned long ret;
 705         unsigned long stack_shift;
 706         struct mm_struct *mm = current->mm;
 707         struct vm_area_struct *vma = bprm->vma;
 708         struct vm_area_struct *prev = NULL;
 709         unsigned long vm_flags;
 710         unsigned long stack_base;
 711         unsigned long stack_size;
 712         unsigned long stack_expand;
 713         unsigned long rlim_stack;
 714
 715 #ifdef CONFIG_STACK_GROWSUP
 716         /* Limit stack size */
 717         stack_base = bprm->rlim_stack.rlim_max;
 718         if (stack_base > STACK_SIZE_MAX)
 719                 stack_base = STACK_SIZE_MAX;
 720
 721         /* Add space for stack randomization. */
 722         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 723
 724         /* Make sure we didn't let the argument array grow too large. */
 725         if (vma->vm_end - vma->vm_start > stack_base)
 726                 return -ENOMEM;
 727
 728         stack_base = PAGE_ALIGN(stack_top - stack_base);
 729
 730         stack_shift = vma->vm_start - stack_base;
 731         mm->arg_start = bprm->p - stack_shift;
 732         bprm->p = vma->vm_end - stack_shift;
 733 #else
 734         stack_top = arch_align_stack(stack_top);
 735         stack_top = PAGE_ALIGN(stack_top);
 736
 737         if (unlikely(stack_top < mmap_min_addr) ||
 738             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 739                 return -ENOMEM;
 740
 741         stack_shift = vma->vm_end - stack_top;
 742
 743         bprm->p -= stack_shift;
 744         mm->arg_start = bprm->p;
 745 #endif
 746
 747         if (bprm->loader)
 748                 bprm->loader -= stack_shift;
 749         bprm->exec -= stack_shift;
 750
 751         if (down_write_killable(&mm->mmap_sem))
 752                 return -EINTR;
 753
 754         vm_flags = VM_STACK_FLAGS;
 755
 756         /*
 757          * Adjust stack execute permissions; explicitly enable for
 758          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 759          * (arch default) otherwise.
 760          */
 761         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 762                 vm_flags |= VM_EXEC;
 763         else if (executable_stack == EXSTACK_DISABLE_X)
 764                 vm_flags &= ~VM_EXEC;
 765         vm_flags |= mm->def_flags;
 766         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 767
 768         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 769                         vm_flags);
 770         if (ret)
 771                 goto out_unlock;
 772         BUG_ON(prev != vma);
 773
 774         /* Move stack pages down in memory. */
 775         if (stack_shift) {
 776                 ret = shift_arg_pages(vma, stack_shift);
 777                 if (ret)
 778                         goto out_unlock;
 779         }
 780
 781         /* mprotect_fixup is overkill to remove the temporary stack flags */
 782         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 783
 784         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 785         stack_size = vma->vm_end - vma->vm_start;
 786         /*
 787          * Align this down to a page boundary as expand_stack
 788          * will align it up.
 789          */
 790         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 791 #ifdef CONFIG_STACK_GROWSUP
 792         if (stack_size + stack_expand > rlim_stack)
 793                 stack_base = vma->vm_start + rlim_stack;
 794         else
 795                 stack_base = vma->vm_end + stack_expand;
 796 #else
 797         if (stack_size + stack_expand > rlim_stack)
 798                 stack_base = vma->vm_end - rlim_stack;
 799         else
 800                 stack_base = vma->vm_start - stack_expand;
 801 #endif
 802         current->mm->start_stack = bprm->p;
 803         ret = expand_stack(vma, stack_base);
 804         if (ret)
 805                 ret = -EFAULT;
 806
 807 out_unlock:
 808         up_write(&mm->mmap_sem);
 809         return ret;
 810 }
 811 EXPORT_SYMBOL(setup_arg_pages);
 812
 813 #else
 814
 815 /*
 816  * Transfer the program arguments and environment from the holding pages
 817  * onto the stack. The provided stack pointer is adjusted accordingly.
 818  */
 819 int transfer_args_to_stack(struct linux_binprm *bprm,
 820                            unsigned long *sp_location)
 821 {
 822         unsigned long index, stop, sp;
 823         int ret = 0;
 824
 825         stop = bprm->p >> PAGE_SHIFT;
 826         sp = *sp_location;
 827
 828         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 829                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 830                 char *src = kmap(bprm->page[index]) + offset;
 831                 sp -= PAGE_SIZE - offset;
 832                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 833                         ret = -EFAULT;
 834                 kunmap(bprm->page[index]);
 835                 if (ret)
 836                         goto out;
 837         }
 838
 839         *sp_location = sp;
 840
 841 out:
 842         return ret;
 843 }
 844 EXPORT_SYMBOL(transfer_args_to_stack);
 845
 846 #endif /* CONFIG_MMU */
 847
 848 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 849 {
 850         struct file *file;
 851         int err;
 852         struct open_flags open_exec_flags = {
 853                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 854                 .acc_mode = MAY_EXEC,
 855                 .intent = LOOKUP_OPEN,
 856                 .lookup_flags = LOOKUP_FOLLOW,
 857         };
 858
 859         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 860                 return ERR_PTR(-EINVAL);
 861         if (flags & AT_SYMLINK_NOFOLLOW)
 862                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 863         if (flags & AT_EMPTY_PATH)
 864                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 865
 866         file = do_filp_open(fd, name, &open_exec_flags);
 867         if (IS_ERR(file))
 868                 goto out;
 869
 870         err = -EACCES;
 871         if (!S_ISREG(file_inode(file)->i_mode))
 872                 goto exit;
 873
 874         if (path_noexec(&file->f_path))
 875                 goto exit;
 876
 877         err = deny_write_access(file);
 878         if (err)
 879                 goto exit;
 880
 881         if (name->name[0] != '\0')
 882                 fsnotify_open(file);
 883
 884 out:
 885         return file;
 886
 887 exit:
 888         fput(file);
 889         return ERR_PTR(err);
 890 }
 891
 892 struct file *open_exec(const char *name)
 893 {
 894         struct filename *filename = getname_kernel(name);
 895         struct file *f = ERR_CAST(filename);
 896
 897         if (!IS_ERR(filename)) {
 898                 f = do_open_execat(AT_FDCWD, filename, 0);
 899                 putname(filename);
 900         }
 901         return f;
 902 }
 903 EXPORT_SYMBOL(open_exec);
 904
 905 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 906                      loff_t max_size, enum kernel_read_file_id id)
 907 {
 908         loff_t i_size, pos;
 909         ssize_t bytes = 0;
 910         int ret;
 911
 912         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 913                 return -EINVAL;
 914
 915         ret = deny_write_access(file);
 916         if (ret)
 917                 return ret;
 918
 919         ret = security_kernel_read_file(file, id);
 920         if (ret)
 921                 goto out;
 922
 923         i_size = i_size_read(file_inode(file));
 924         if (i_size <= 0) {
 925                 ret = -EINVAL;
 926                 goto out;
 927         }
 928         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 929                 ret = -EFBIG;
 930                 goto out;
 931         }
 932
 933         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 934                 *buf = vmalloc(i_size);
 935         if (!*buf) {
 936                 ret = -ENOMEM;
 937                 goto out;
 938         }
 939
 940         pos = 0;
 941         while (pos < i_size) {
 942                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 943                 if (bytes < 0) {
 944                         ret = bytes;
 945                         goto out_free;
 946                 }
 947
 948                 if (bytes == 0)
 949                         break;
 950         }
 951
 952         if (pos != i_size) {
 953                 ret = -EIO;
 954                 goto out_free;
 955         }
 956
 957         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 958         if (!ret)
 959                 *size = pos;
 960
 961 out_free:
 962         if (ret < 0) {
 963                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 964                         vfree(*buf);
 965                         *buf = NULL;
 966                 }
 967         }
 968
 969 out:
 970         allow_write_access(file);
 971         return ret;
 972 }
 973 EXPORT_SYMBOL_GPL(kernel_read_file);
 974
 975 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
 976                                loff_t max_size, enum kernel_read_file_id id)
 977 {
 978         struct file *file;
 979         int ret;
 980
 981         if (!path || !*path)
 982                 return -EINVAL;
 983
 984         file = filp_open(path, O_RDONLY, 0);
 985         if (IS_ERR(file))
 986                 return PTR_ERR(file);
 987
 988         ret = kernel_read_file(file, buf, size, max_size, id);
 989         fput(file);
 990         return ret;
 991 }
 992 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 993
 994 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 995                              enum kernel_read_file_id id)
 996 {
 997         struct fd f = fdget(fd);
 998         int ret = -EBADF;
 999
1000         if (!f.file || !(f.file->f_mode & FMODE_READ))
1001                 goto out;
1002
1003         ret = kernel_read_file(f.file, buf, size, max_size, id);
1004 out:
1005         fdput(f);
1006         return ret;
1007 }
1008 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1009
1010 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1011 {
1012         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1013         if (res > 0)
1014                 flush_icache_range(addr, addr + len);
1015         return res;
1016 }
1017 EXPORT_SYMBOL(read_code);
1018
1019 /*
1020  * Maps the mm_struct mm into the current task struct.
1021  * On success, this function returns with exec_update_lock
1022  * held for writing.
1023  */
1024 static int exec_mmap(struct mm_struct *mm)
1025 {
1026         struct task_struct *tsk;
1027         struct mm_struct *old_mm, *active_mm;
1028         int ret;
1029
1030         /* Notify parent that we're no longer interested in the old VM */
1031         tsk = current;
1032         old_mm = current->mm;
1033         exec_mm_release(tsk, old_mm);
1034
1035         ret = down_write_killable(&tsk->signal->exec_update_lock);
1036         if (ret)
1037                 return ret;
1038
1039         if (old_mm) {
1040                 sync_mm_rss(old_mm);
1041                 /*
1042                  * Make sure that if there is a core dump in progress
1043                  * for the old mm, we get out and die instead of going
1044                  * through with the exec.  We must hold mmap_sem around
1045                  * checking core_state and changing tsk->mm.
1046                  */
1047                 down_read(&old_mm->mmap_sem);
1048                 if (unlikely(old_mm->core_state)) {
1049                         up_read(&old_mm->mmap_sem);
1050                         up_write(&tsk->signal->exec_update_lock);
1051                         return -EINTR;
1052                 }
1053         }
1054
1055         task_lock(tsk);
1056         membarrier_exec_mmap(mm);
1057
1058         local_irq_disable();
1059         active_mm = tsk->active_mm;
1060         tsk->active_mm = mm;
1061         tsk->mm = mm;
1062         /*
1063          * This prevents preemption while active_mm is being loaded and
1064          * it and mm are being updated, which could cause problems for
1065          * lazy tlb mm refcounting when these are updated by context
1066          * switches. Not all architectures can handle irqs off over
1067          * activate_mm yet.
1068          */
1069         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1070                 local_irq_enable();
1071         activate_mm(active_mm, mm);
1072         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1073                 local_irq_enable();
1074         tsk->mm->vmacache_seqnum = 0;
1075         vmacache_flush(tsk);
1076         task_unlock(tsk);
1077         if (old_mm) {
1078                 up_read(&old_mm->mmap_sem);
1079                 BUG_ON(active_mm != old_mm);
1080                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1081                 mm_update_next_owner(old_mm);
1082                 mmput(old_mm);
1083                 return 0;
1084         }
1085         mmdrop(active_mm);
1086         return 0;
1087 }
1088
1089 /*
1090  * This function makes sure the current process has its own signal table,
1091  * so that flush_signal_handlers can later reset the handlers without
1092  * disturbing other processes.  (Other processes might share the signal
1093  * table via the CLONE_SIGHAND option to clone().)
1094  */
1095 static int de_thread(struct task_struct *tsk)
1096 {
1097         struct signal_struct *sig = tsk->signal;
1098         struct sighand_struct *oldsighand = tsk->sighand;
1099         spinlock_t *lock = &oldsighand->siglock;
1100
1101         if (thread_group_empty(tsk))
1102                 goto no_thread_group;
1103
1104         /*
1105          * Kill all other threads in the thread group.
1106          */
1107         spin_lock_irq(lock);
1108         if (signal_group_exit(sig)) {
1109                 /*
1110                  * Another group action in progress, just
1111                  * return so that the signal is processed.
1112                  */
1113                 spin_unlock_irq(lock);
1114                 return -EAGAIN;
1115         }
1116
1117         sig->group_exit_task = tsk;
1118         sig->notify_count = zap_other_threads(tsk);
1119         if (!thread_group_leader(tsk))
1120                 sig->notify_count--;
1121
1122         while (sig->notify_count) {
1123                 __set_current_state(TASK_KILLABLE);
1124                 spin_unlock_irq(lock);
1125                 schedule();
1126                 if (__fatal_signal_pending(tsk))
1127                         goto killed;
1128                 spin_lock_irq(lock);
1129         }
1130         spin_unlock_irq(lock);
1131
1132         /*
1133          * At this point all other threads have exited, all we have to
1134          * do is to wait for the thread group leader to become inactive,
1135          * and to assume its PID:
1136          */
1137         if (!thread_group_leader(tsk)) {
1138                 struct task_struct *leader = tsk->group_leader;
1139
1140                 for (;;) {
1141                         cgroup_threadgroup_change_begin(tsk);
1142                         write_lock_irq(&tasklist_lock);
1143                         /*
1144                          * Do this under tasklist_lock to ensure that
1145                          * exit_notify() can't miss ->group_exit_task
1146                          */
1147                         sig->notify_count = -1;
1148                         if (likely(leader->exit_state))
1149                                 break;
1150                         __set_current_state(TASK_KILLABLE);
1151                         write_unlock_irq(&tasklist_lock);
1152                         cgroup_threadgroup_change_end(tsk);
1153                         schedule();
1154                         if (__fatal_signal_pending(tsk))
1155                                 goto killed;
1156                 }
1157
1158                 /*
1159                  * The only record we have of the real-time age of a
1160                  * process, regardless of execs it's done, is start_time.
1161                  * All the past CPU time is accumulated in signal_struct
1162                  * from sister threads now dead.  But in this non-leader
1163                  * exec, nothing survives from the original leader thread,
1164                  * whose birth marks the true age of this process now.
1165                  * When we take on its identity by switching to its PID, we
1166                  * also take its birthdate (always earlier than our own).
1167                  */
1168                 tsk->start_time = leader->start_time;
1169                 tsk->real_start_time = leader->real_start_time;
1170
1171                 BUG_ON(!same_thread_group(leader, tsk));
1172                 BUG_ON(has_group_leader_pid(tsk));
1173                 /*
1174                  * An exec() starts a new thread group with the
1175                  * TGID of the previous thread group. Rehash the
1176                  * two threads with a switched PID, and release
1177                  * the former thread group leader:
1178                  */
1179
1180                 /* Become a process group leader with the old leader's pid.
1181                  * The old leader becomes a thread of the this thread group.
1182                  * Note: The old leader also uses this pid until release_task
1183                  *       is called.  Odd but simple and correct.
1184                  */
1185                 tsk->pid = leader->pid;
1186                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1187                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1188                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1189                 transfer_pid(leader, tsk, PIDTYPE_SID);
1190
1191                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1192                 list_replace_init(&leader->sibling, &tsk->sibling);
1193
1194                 tsk->group_leader = tsk;
1195                 leader->group_leader = tsk;
1196
1197                 tsk->exit_signal = SIGCHLD;
1198                 leader->exit_signal = -1;
1199
1200                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1201                 leader->exit_state = EXIT_DEAD;
1202
1203                 /*
1204                  * We are going to release_task()->ptrace_unlink() silently,
1205                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1206                  * the tracer wont't block again waiting for this thread.
1207                  */
1208                 if (unlikely(leader->ptrace))
1209                         __wake_up_parent(leader, leader->parent);
1210                 write_unlock_irq(&tasklist_lock);
1211                 cgroup_threadgroup_change_end(tsk);
1212
1213                 release_task(leader);
1214         }
1215
1216         sig->group_exit_task = NULL;
1217         sig->notify_count = 0;
1218
1219 no_thread_group:
1220         /* we have changed execution domain */
1221         tsk->exit_signal = SIGCHLD;
1222
1223 #ifdef CONFIG_POSIX_TIMERS
1224         exit_itimers(sig);
1225         flush_itimer_signals();
1226 #endif
1227
1228         if (refcount_read(&oldsighand->count) != 1) {
1229                 struct sighand_struct *newsighand;
1230                 /*
1231                  * This ->sighand is shared with the CLONE_SIGHAND
1232                  * but not CLONE_THREAD task, switch to the new one.
1233                  */
1234                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1235                 if (!newsighand)
1236                         return -ENOMEM;
1237
1238                 refcount_set(&newsighand->count, 1);
1239                 memcpy(newsighand->action, oldsighand->action,
1240                        sizeof(newsighand->action));
1241
1242                 write_lock_irq(&tasklist_lock);
1243                 spin_lock(&oldsighand->siglock);
1244                 rcu_assign_pointer(tsk->sighand, newsighand);
1245                 spin_unlock(&oldsighand->siglock);
1246                 write_unlock_irq(&tasklist_lock);
1247
1248                 __cleanup_sighand(oldsighand);
1249         }
1250
1251         BUG_ON(!thread_group_leader(tsk));
1252         return 0;
1253
1254 killed:
1255         /* protects against exit_notify() and __exit_signal() */
1256         read_lock(&tasklist_lock);
1257         sig->group_exit_task = NULL;
1258         sig->notify_count = 0;
1259         read_unlock(&tasklist_lock);
1260         return -EAGAIN;
1261 }
1262
1263 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1264 {
1265         task_lock(tsk);
1266         strncpy(buf, tsk->comm, buf_size);
1267         task_unlock(tsk);
1268         return buf;
1269 }
1270 EXPORT_SYMBOL_GPL(__get_task_comm);
1271
1272 /*
1273  * These functions flushes out all traces of the currently running executable
1274  * so that a new one can be started
1275  */
1276
1277 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1278 {
1279         task_lock(tsk);
1280         trace_task_rename(tsk, buf);
1281         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1282         task_unlock(tsk);
1283         perf_event_comm(tsk, exec);
1284 }
1285
1286 /*
1287  * Calling this is the point of no return. None of the failures will be
1288  * seen by userspace since either the process is already taking a fatal
1289  * signal (via de_thread() or coredump), or will have SEGV raised
1290  * (after exec_mmap()) by search_binary_handlers (see below).
1291  */
1292 int flush_old_exec(struct linux_binprm * bprm)
1293 {
1294         int retval;
1295
1296         /*
1297          * Make sure we have a private signal table and that
1298          * we are unassociated from the previous thread group.
1299          */
1300         retval = de_thread(current);
1301         if (retval)
1302                 goto out;
1303
1304         /*
1305          * Must be called _before_ exec_mmap() as bprm->mm is
1306          * not visibile until then. This also enables the update
1307          * to be lockless.
1308          */
1309         set_mm_exe_file(bprm->mm, bprm->file);
1310
1311         would_dump(bprm, bprm->file);
1312
1313         /*
1314          * Release all of the old mmap stuff
1315          */
1316         acct_arg_size(bprm, 0);
1317         retval = exec_mmap(bprm->mm);
1318         if (retval)
1319                 goto out;
1320
1321         /*
1322          * After setting bprm->called_exec_mmap (to mark that current is
1323          * using the prepared mm now), we have nothing left of the original
1324          * process. If anything from here on returns an error, the check
1325          * in search_binary_handler() will SEGV current.
1326          */
1327         bprm->called_exec_mmap = 1;
1328         bprm->mm = NULL;
1329
1330         set_fs(USER_DS);
1331         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1332                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1333         flush_thread();
1334         current->personality &= ~bprm->per_clear;
1335
1336         /*
1337          * We have to apply CLOEXEC before we change whether the process is
1338          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1339          * trying to access the should-be-closed file descriptors of a process
1340          * undergoing exec(2).
1341          */
1342         do_close_on_exec(current->files);
1343         return 0;
1344
1345 out:
1346         return retval;
1347 }
1348 EXPORT_SYMBOL(flush_old_exec);
1349
1350 void would_dump(struct linux_binprm *bprm, struct file *file)
1351 {
1352         struct inode *inode = file_inode(file);
1353         if (inode_permission(inode, MAY_READ) < 0) {
1354                 struct user_namespace *old, *user_ns;
1355                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1356
1357                 /* Ensure mm->user_ns contains the executable */
1358                 user_ns = old = bprm->mm->user_ns;
1359                 while ((user_ns != &init_user_ns) &&
1360                        !privileged_wrt_inode_uidgid(user_ns, inode))
1361                         user_ns = user_ns->parent;
1362
1363                 if (old != user_ns) {
1364                         bprm->mm->user_ns = get_user_ns(user_ns);
1365                         put_user_ns(old);
1366                 }
1367         }
1368 }
1369 EXPORT_SYMBOL(would_dump);
1370
1371 void setup_new_exec(struct linux_binprm * bprm)
1372 {
1373         /*
1374          * Once here, prepare_binrpm() will not be called any more, so
1375          * the final state of setuid/setgid/fscaps can be merged into the
1376          * secureexec flag.
1377          */
1378         bprm->secureexec |= bprm->cap_elevated;
1379
1380         if (bprm->secureexec) {
1381                 /* Make sure parent cannot signal privileged process. */
1382                 current->pdeath_signal = 0;
1383
1384                 /*
1385                  * For secureexec, reset the stack limit to sane default to
1386                  * avoid bad behavior from the prior rlimits. This has to
1387                  * happen before arch_pick_mmap_layout(), which examines
1388                  * RLIMIT_STACK, but after the point of no return to avoid
1389                  * needing to clean up the change on failure.
1390                  */
1391                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1392                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1393         }
1394
1395         arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1396
1397         current->sas_ss_sp = current->sas_ss_size = 0;
1398
1399         /*
1400          * Figure out dumpability. Note that this checking only of current
1401          * is wrong, but userspace depends on it. This should be testing
1402          * bprm->secureexec instead.
1403          */
1404         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1405             !(uid_eq(current_euid(), current_uid()) &&
1406               gid_eq(current_egid(), current_gid())))
1407                 set_dumpable(current->mm, suid_dumpable);
1408         else
1409                 set_dumpable(current->mm, SUID_DUMP_USER);
1410
1411         arch_setup_new_exec();
1412         perf_event_exec();
1413         __set_task_comm(current, kbasename(bprm->filename), true);
1414
1415         /* Set the new mm task size. We have to do that late because it may
1416          * depend on TIF_32BIT which is only updated in flush_thread() on
1417          * some architectures like powerpc
1418          */
1419         current->mm->task_size = TASK_SIZE;
1420
1421         /* An exec changes our domain. We are no longer part of the thread
1422            group */
1423         WRITE_ONCE(current->self_exec_id, current->self_exec_id + 1);
1424         flush_signal_handlers(current, 0);
1425 }
1426 EXPORT_SYMBOL(setup_new_exec);
1427
1428 /* Runs immediately before start_thread() takes over. */
1429 void finalize_exec(struct linux_binprm *bprm)
1430 {
1431         /* Store any stack rlimit changes before starting thread. */
1432         task_lock(current->group_leader);
1433         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1434         task_unlock(current->group_leader);
1435 }
1436 EXPORT_SYMBOL(finalize_exec);
1437
1438 /*
1439  * Prepare credentials and lock ->cred_guard_mutex.
1440  * install_exec_creds() commits the new creds and drops the lock.
1441  * Or, if exec fails before, free_bprm() should release ->cred and
1442  * and unlock.
1443  */
1444 static int prepare_bprm_creds(struct linux_binprm *bprm)
1445 {
1446         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1447                 return -ERESTARTNOINTR;
1448
1449         bprm->cred = prepare_exec_creds();
1450         if (likely(bprm->cred))
1451                 return 0;
1452
1453         mutex_unlock(&current->signal->cred_guard_mutex);
1454         return -ENOMEM;
1455 }
1456
1457 static void free_bprm(struct linux_binprm *bprm)
1458 {
1459         free_arg_pages(bprm);
1460         if (bprm->cred) {
1461                 if (bprm->called_exec_mmap)
1462                         up_write(&current->signal->exec_update_lock);
1463                 mutex_unlock(&current->signal->cred_guard_mutex);
1464                 abort_creds(bprm->cred);
1465         }
1466         if (bprm->file) {
1467                 allow_write_access(bprm->file);
1468                 fput(bprm->file);
1469         }
1470         /* If a binfmt changed the interp, free it. */
1471         if (bprm->interp != bprm->filename)
1472                 kfree(bprm->interp);
1473         kfree(bprm);
1474 }
1475
1476 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1477 {
1478         /* If a binfmt changed the interp, free it first. */
1479         if (bprm->interp != bprm->filename)
1480                 kfree(bprm->interp);
1481         bprm->interp = kstrdup(interp, GFP_KERNEL);
1482         if (!bprm->interp)
1483                 return -ENOMEM;
1484         return 0;
1485 }
1486 EXPORT_SYMBOL(bprm_change_interp);
1487
1488 /*
1489  * install the new credentials for this executable
1490  */
1491 void install_exec_creds(struct linux_binprm *bprm)
1492 {
1493         security_bprm_committing_creds(bprm);
1494
1495         commit_creds(bprm->cred);
1496         bprm->cred = NULL;
1497
1498         /*
1499          * Disable monitoring for regular users
1500          * when executing setuid binaries. Must
1501          * wait until new credentials are committed
1502          * by commit_creds() above
1503          */
1504         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1505                 perf_event_exit_task(current);
1506         /*
1507          * cred_guard_mutex must be held at least to this point to prevent
1508          * ptrace_attach() from altering our determination of the task's
1509          * credentials; any time after this it may be unlocked.
1510          */
1511         security_bprm_committed_creds(bprm);
1512         up_write(&current->signal->exec_update_lock);
1513         mutex_unlock(&current->signal->cred_guard_mutex);
1514 }
1515 EXPORT_SYMBOL(install_exec_creds);
1516
1517 /*
1518  * determine how safe it is to execute the proposed program
1519  * - the caller must hold ->cred_guard_mutex to protect against
1520  *   PTRACE_ATTACH or seccomp thread-sync
1521  */
1522 static void check_unsafe_exec(struct linux_binprm *bprm)
1523 {
1524         struct task_struct *p = current, *t;
1525         unsigned n_fs;
1526
1527         if (p->ptrace)
1528                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1529
1530         /*
1531          * This isn't strictly necessary, but it makes it harder for LSMs to
1532          * mess up.
1533          */
1534         if (task_no_new_privs(current))
1535                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1536
1537         t = p;
1538         n_fs = 1;
1539         spin_lock(&p->fs->lock);
1540         rcu_read_lock();
1541         while_each_thread(p, t) {
1542                 if (t->fs == p->fs)
1543                         n_fs++;
1544         }
1545         rcu_read_unlock();
1546
1547         if (p->fs->users > n_fs)
1548                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1549         else
1550                 p->fs->in_exec = 1;
1551         spin_unlock(&p->fs->lock);
1552 }
1553
1554 static void bprm_fill_uid(struct linux_binprm *bprm)
1555 {
1556         struct inode *inode;
1557         unsigned int mode;
1558         kuid_t uid;
1559         kgid_t gid;
1560
1561         /*
1562          * Since this can be called multiple times (via prepare_binprm),
1563          * we must clear any previous work done when setting set[ug]id
1564          * bits from any earlier bprm->file uses (for example when run
1565          * first for a setuid script then again for its interpreter).
1566          */
1567         bprm->cred->euid = current_euid();
1568         bprm->cred->egid = current_egid();
1569
1570         if (!mnt_may_suid(bprm->file->f_path.mnt))
1571                 return;
1572
1573         if (task_no_new_privs(current))
1574                 return;
1575
1576         inode = bprm->file->f_path.dentry->d_inode;
1577         mode = READ_ONCE(inode->i_mode);
1578         if (!(mode & (S_ISUID|S_ISGID)))
1579                 return;
1580
1581         /* Be careful if suid/sgid is set */
1582         inode_lock(inode);
1583
1584         /* reload atomically mode/uid/gid now that lock held */
1585         mode = inode->i_mode;
1586         uid = inode->i_uid;
1587         gid = inode->i_gid;
1588         inode_unlock(inode);
1589
1590         /* We ignore suid/sgid if there are no mappings for them in the ns */
1591         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1592                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1593                 return;
1594
1595         if (mode & S_ISUID) {
1596                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1597                 bprm->cred->euid = uid;
1598         }
1599
1600         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1601                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1602                 bprm->cred->egid = gid;
1603         }
1604 }
1605
1606 /*
1607  * Fill the binprm structure from the inode.
1608  * Check permissions, then read the first BINPRM_BUF_SIZE bytes
1609  *
1610  * This may be called multiple times for binary chains (scripts for example).
1611  */
1612 int prepare_binprm(struct linux_binprm *bprm)
1613 {
1614         int retval;
1615         loff_t pos = 0;
1616
1617         bprm_fill_uid(bprm);
1618
1619         /* fill in binprm security blob */
1620         retval = security_bprm_set_creds(bprm);
1621         if (retval)
1622                 return retval;
1623         bprm->called_set_creds = 1;
1624
1625         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1626         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1627 }
1628
1629 EXPORT_SYMBOL(prepare_binprm);
1630
1631 /*
1632  * Arguments are '\0' separated strings found at the location bprm->p
1633  * points to; chop off the first by relocating brpm->p to right after
1634  * the first '\0' encountered.
1635  */
1636 int remove_arg_zero(struct linux_binprm *bprm)
1637 {
1638         int ret = 0;
1639         unsigned long offset;
1640         char *kaddr;
1641         struct page *page;
1642
1643         if (!bprm->argc)
1644                 return 0;
1645
1646         do {
1647                 offset = bprm->p & ~PAGE_MASK;
1648                 page = get_arg_page(bprm, bprm->p, 0);
1649                 if (!page) {
1650                         ret = -EFAULT;
1651                         goto out;
1652                 }
1653                 kaddr = kmap_atomic(page);
1654
1655                 for (; offset < PAGE_SIZE && kaddr[offset];
1656                                 offset++, bprm->p++)
1657                         ;
1658
1659                 kunmap_atomic(kaddr);
1660                 put_arg_page(page);
1661         } while (offset == PAGE_SIZE);
1662
1663         bprm->p++;
1664         bprm->argc--;
1665         ret = 0;
1666
1667 out:
1668         return ret;
1669 }
1670 EXPORT_SYMBOL(remove_arg_zero);
1671
1672 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1673 /*
1674  * cycle the list of binary formats handler, until one recognizes the image
1675  */
1676 int search_binary_handler(struct linux_binprm *bprm)
1677 {
1678         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1679         struct linux_binfmt *fmt;
1680         int retval;
1681
1682         /* This allows 4 levels of binfmt rewrites before failing hard. */
1683         if (bprm->recursion_depth > 5)
1684                 return -ELOOP;
1685
1686         retval = security_bprm_check(bprm);
1687         if (retval)
1688                 return retval;
1689
1690         retval = -ENOENT;
1691  retry:
1692         read_lock(&binfmt_lock);
1693         list_for_each_entry(fmt, &formats, lh) {
1694                 if (!try_module_get(fmt->module))
1695                         continue;
1696                 read_unlock(&binfmt_lock);
1697
1698                 bprm->recursion_depth++;
1699                 retval = fmt->load_binary(bprm);
1700                 bprm->recursion_depth--;
1701
1702                 read_lock(&binfmt_lock);
1703                 put_binfmt(fmt);
1704                 if (retval < 0 && bprm->called_exec_mmap) {
1705                         /* we got to flush_old_exec() and failed after it */
1706                         read_unlock(&binfmt_lock);
1707                         force_sigsegv(SIGSEGV);
1708                         return retval;
1709                 }
1710                 if (retval != -ENOEXEC || !bprm->file) {
1711                         read_unlock(&binfmt_lock);
1712                         return retval;
1713                 }
1714         }
1715         read_unlock(&binfmt_lock);
1716
1717         if (need_retry) {
1718                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1719                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1720                         return retval;
1721                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1722                         return retval;
1723                 need_retry = false;
1724                 goto retry;
1725         }
1726
1727         return retval;
1728 }
1729 EXPORT_SYMBOL(search_binary_handler);
1730
1731 static int exec_binprm(struct linux_binprm *bprm)
1732 {
1733         pid_t old_pid, old_vpid;
1734         int ret;
1735
1736         /* Need to fetch pid before load_binary changes it */
1737         old_pid = current->pid;
1738         rcu_read_lock();
1739         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1740         rcu_read_unlock();
1741
1742         ret = search_binary_handler(bprm);
1743         if (ret >= 0) {
1744                 audit_bprm(bprm);
1745                 trace_sched_process_exec(current, old_pid, bprm);
1746                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1747                 proc_exec_connector(current);
1748         }
1749
1750         return ret;
1751 }
1752
1753 /*
1754  * sys_execve() executes a new program.
1755  */
1756 static int __do_execve_file(int fd, struct filename *filename,
1757                             struct user_arg_ptr argv,
1758                             struct user_arg_ptr envp,
1759                             int flags, struct file *file)
1760 {
1761         char *pathbuf = NULL;
1762         struct linux_binprm *bprm;
1763         struct files_struct *displaced;
1764         int retval;
1765
1766         if (IS_ERR(filename))
1767                 return PTR_ERR(filename);
1768
1769         /*
1770          * We move the actual failure in case of RLIMIT_NPROC excess from
1771          * set*uid() to execve() because too many poorly written programs
1772          * don't check setuid() return code.  Here we additionally recheck
1773          * whether NPROC limit is still exceeded.
1774          */
1775         if ((current->flags & PF_NPROC_EXCEEDED) &&
1776             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1777                 retval = -EAGAIN;
1778                 goto out_ret;
1779         }
1780
1781         /* We're below the limit (still or again), so we don't want to make
1782          * further execve() calls fail. */
1783         current->flags &= ~PF_NPROC_EXCEEDED;
1784
1785         retval = unshare_files(&displaced);
1786         if (retval)
1787                 goto out_ret;
1788
1789         retval = -ENOMEM;
1790         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1791         if (!bprm)
1792                 goto out_files;
1793
1794         retval = prepare_bprm_creds(bprm);
1795         if (retval)
1796                 goto out_free;
1797
1798         check_unsafe_exec(bprm);
1799         current->in_execve = 1;
1800
1801         if (!file)
1802                 file = do_open_execat(fd, filename, flags);
1803         retval = PTR_ERR(file);
1804         if (IS_ERR(file))
1805                 goto out_unmark;
1806
1807         sched_exec();
1808
1809         bprm->file = file;
1810         if (!filename) {
1811                 bprm->filename = "none";
1812         } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1813                 bprm->filename = filename->name;
1814         } else {
1815                 if (filename->name[0] == '\0')
1816                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1817                 else
1818                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1819                                             fd, filename->name);
1820                 if (!pathbuf) {
1821                         retval = -ENOMEM;
1822                         goto out_unmark;
1823                 }
1824                 /*
1825                  * Record that a name derived from an O_CLOEXEC fd will be
1826                  * inaccessible after exec. Relies on having exclusive access to
1827                  * current->files (due to unshare_files above).
1828                  */
1829                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1830                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1831                 bprm->filename = pathbuf;
1832         }
1833         bprm->interp = bprm->filename;
1834
1835         retval = bprm_mm_init(bprm);
1836         if (retval)
1837                 goto out_unmark;
1838
1839         retval = prepare_arg_pages(bprm, argv, envp);
1840         if (retval < 0)
1841                 goto out;
1842
1843         retval = prepare_binprm(bprm);
1844         if (retval < 0)
1845                 goto out;
1846
1847         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1848         if (retval < 0)
1849                 goto out;
1850
1851         bprm->exec = bprm->p;
1852         retval = copy_strings(bprm->envc, envp, bprm);
1853         if (retval < 0)
1854                 goto out;
1855
1856         retval = copy_strings(bprm->argc, argv, bprm);
1857         if (retval < 0)
1858                 goto out;
1859
1860         /*
1861          * When argv is empty, add an empty string ("") as argv[0] to
1862          * ensure confused userspace programs that start processing
1863          * from argv[1] won't end up walking envp. See also
1864          * bprm_stack_limits().
1865          */
1866         if (bprm->argc == 0) {
1867                 const char *argv[] = { "", NULL };
1868                 retval = copy_strings_kernel(1, argv, bprm);
1869                 if (retval < 0)
1870                         goto out;
1871                 bprm->argc = 1;
1872         }
1873
1874         retval = exec_binprm(bprm);
1875         if (retval < 0)
1876                 goto out;
1877
1878         /* execve succeeded */
1879         current->fs->in_exec = 0;
1880         current->in_execve = 0;
1881         rseq_execve(current);
1882         acct_update_integrals(current);
1883         task_numa_free(current, false);
1884         free_bprm(bprm);
1885         kfree(pathbuf);
1886         if (filename)
1887                 putname(filename);
1888         if (displaced)
1889                 put_files_struct(displaced);
1890         return retval;
1891
1892 out:
1893         if (bprm->mm) {
1894                 acct_arg_size(bprm, 0);
1895                 mmput(bprm->mm);
1896         }
1897
1898 out_unmark:
1899         current->fs->in_exec = 0;
1900         current->in_execve = 0;
1901
1902 out_free:
1903         free_bprm(bprm);
1904         kfree(pathbuf);
1905
1906 out_files:
1907         if (displaced)
1908                 reset_files_struct(displaced);
1909 out_ret:
1910         if (filename)
1911                 putname(filename);
1912         return retval;
1913 }
1914
1915 static int do_execveat_common(int fd, struct filename *filename,
1916                               struct user_arg_ptr argv,
1917                               struct user_arg_ptr envp,
1918                               int flags)
1919 {
1920         return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1921 }
1922
1923 int do_execve_file(struct file *file, void *__argv, void *__envp)
1924 {
1925         struct user_arg_ptr argv = { .ptr.native = __argv };
1926         struct user_arg_ptr envp = { .ptr.native = __envp };
1927
1928         return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1929 }
1930
1931 int do_execve(struct filename *filename,
1932         const char __user *const __user *__argv,
1933         const char __user *const __user *__envp)
1934 {
1935         struct user_arg_ptr argv = { .ptr.native = __argv };
1936         struct user_arg_ptr envp = { .ptr.native = __envp };
1937         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1938 }
1939
1940 int do_execveat(int fd, struct filename *filename,
1941                 const char __user *const __user *__argv,
1942                 const char __user *const __user *__envp,
1943                 int flags)
1944 {
1945         struct user_arg_ptr argv = { .ptr.native = __argv };
1946         struct user_arg_ptr envp = { .ptr.native = __envp };
1947
1948         return do_execveat_common(fd, filename, argv, envp, flags);
1949 }
1950
1951 #ifdef CONFIG_COMPAT
1952 static int compat_do_execve(struct filename *filename,
1953         const compat_uptr_t __user *__argv,
1954         const compat_uptr_t __user *__envp)
1955 {
1956         struct user_arg_ptr argv = {
1957                 .is_compat = true,
1958                 .ptr.compat = __argv,
1959         };
1960         struct user_arg_ptr envp = {
1961                 .is_compat = true,
1962                 .ptr.compat = __envp,
1963         };
1964         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1965 }
1966
1967 static int compat_do_execveat(int fd, struct filename *filename,
1968                               const compat_uptr_t __user *__argv,
1969                               const compat_uptr_t __user *__envp,
1970                               int flags)
1971 {
1972         struct user_arg_ptr argv = {
1973                 .is_compat = true,
1974                 .ptr.compat = __argv,
1975         };
1976         struct user_arg_ptr envp = {
1977                 .is_compat = true,
1978                 .ptr.compat = __envp,
1979         };
1980         return do_execveat_common(fd, filename, argv, envp, flags);
1981 }
1982 #endif
1983
1984 void set_binfmt(struct linux_binfmt *new)
1985 {
1986         struct mm_struct *mm = current->mm;
1987
1988         if (mm->binfmt)
1989                 module_put(mm->binfmt->module);
1990
1991         mm->binfmt = new;
1992         if (new)
1993                 __module_get(new->module);
1994 }
1995 EXPORT_SYMBOL(set_binfmt);
1996
1997 /*
1998  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1999  */
2000 void set_dumpable(struct mm_struct *mm, int value)
2001 {
2002         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2003                 return;
2004
2005         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2006 }
2007
2008 SYSCALL_DEFINE3(execve,
2009                 const char __user *, filename,
2010                 const char __user *const __user *, argv,
2011                 const char __user *const __user *, envp)
2012 {
2013         return do_execve(getname(filename), argv, envp);
2014 }
2015
2016 SYSCALL_DEFINE5(execveat,
2017                 int, fd, const char __user *, filename,
2018                 const char __user *const __user *, argv,
2019                 const char __user *const __user *, envp,
2020                 int, flags)
2021 {
2022         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2023
2024         return do_execveat(fd,
2025                            getname_flags(filename, lookup_flags, NULL),
2026                            argv, envp, flags);
2027 }
2028
2029 #ifdef CONFIG_COMPAT
2030 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2031         const compat_uptr_t __user *, argv,
2032         const compat_uptr_t __user *, envp)
2033 {
2034         return compat_do_execve(getname(filename), argv, envp);
2035 }
2036
2037 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2038                        const char __user *, filename,
2039                        const compat_uptr_t __user *, argv,
2040                        const compat_uptr_t __user *, envp,
2041                        int,  flags)
2042 {
2043         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2044
2045         return compat_do_execveat(fd,
2046                                   getname_flags(filename, lookup_flags, NULL),
2047                                   argv, envp, flags);
2048 }
2049 #endif