Back to home page

LXR

 
 

    


0001 /*
0002  *  linux/fs/exec.c
0003  *
0004  *  Copyright (C) 1991, 1992  Linus Torvalds
0005  */
0006 
0007 /*
0008  * #!-checking implemented by tytso.
0009  */
0010 /*
0011  * Demand-loading implemented 01.12.91 - no need to read anything but
0012  * the header into memory. The inode of the executable is put into
0013  * "current->executable", and page faults do the actual loading. Clean.
0014  *
0015  * Once more I can proudly say that linux stood up to being changed: it
0016  * was less than 2 hours work to get demand-loading completely implemented.
0017  *
0018  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
0019  * current->executable is only used by the procfs.  This allows a dispatch
0020  * table to check for several different types  of binary formats.  We keep
0021  * trying until we recognize the file or we run out of supported binary
0022  * formats.
0023  */
0024 
0025 #include <linux/slab.h>
0026 #include <linux/file.h>
0027 #include <linux/fdtable.h>
0028 #include <linux/mm.h>
0029 #include <linux/vmacache.h>
0030 #include <linux/stat.h>
0031 #include <linux/fcntl.h>
0032 #include <linux/swap.h>
0033 #include <linux/string.h>
0034 #include <linux/init.h>
0035 #include <linux/pagemap.h>
0036 #include <linux/perf_event.h>
0037 #include <linux/highmem.h>
0038 #include <linux/spinlock.h>
0039 #include <linux/key.h>
0040 #include <linux/personality.h>
0041 #include <linux/binfmts.h>
0042 #include <linux/utsname.h>
0043 #include <linux/pid_namespace.h>
0044 #include <linux/module.h>
0045 #include <linux/namei.h>
0046 #include <linux/mount.h>
0047 #include <linux/security.h>
0048 #include <linux/syscalls.h>
0049 #include <linux/tsacct_kern.h>
0050 #include <linux/cn_proc.h>
0051 #include <linux/audit.h>
0052 #include <linux/tracehook.h>
0053 #include <linux/kmod.h>
0054 #include <linux/fsnotify.h>
0055 #include <linux/fs_struct.h>
0056 #include <linux/pipe_fs_i.h>
0057 #include <linux/oom.h>
0058 #include <linux/compat.h>
0059 #include <linux/vmalloc.h>
0060 
0061 #include <linux/uaccess.h>
0062 #include <asm/mmu_context.h>
0063 #include <asm/tlb.h>
0064 
0065 #include <trace/events/task.h>
0066 #include "internal.h"
0067 
0068 #include <trace/events/sched.h>
0069 
0070 int suid_dumpable = 0;
0071 
0072 static LIST_HEAD(formats);
0073 static DEFINE_RWLOCK(binfmt_lock);
0074 
0075 void __register_binfmt(struct linux_binfmt * fmt, int insert)
0076 {
0077     BUG_ON(!fmt);
0078     if (WARN_ON(!fmt->load_binary))
0079         return;
0080     write_lock(&binfmt_lock);
0081     insert ? list_add(&fmt->lh, &formats) :
0082          list_add_tail(&fmt->lh, &formats);
0083     write_unlock(&binfmt_lock);
0084 }
0085 
0086 EXPORT_SYMBOL(__register_binfmt);
0087 
0088 void unregister_binfmt(struct linux_binfmt * fmt)
0089 {
0090     write_lock(&binfmt_lock);
0091     list_del(&fmt->lh);
0092     write_unlock(&binfmt_lock);
0093 }
0094 
0095 EXPORT_SYMBOL(unregister_binfmt);
0096 
0097 static inline void put_binfmt(struct linux_binfmt * fmt)
0098 {
0099     module_put(fmt->module);
0100 }
0101 
0102 bool path_noexec(const struct path *path)
0103 {
0104     return (path->mnt->mnt_flags & MNT_NOEXEC) ||
0105            (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
0106 }
0107 
0108 #ifdef CONFIG_USELIB
0109 /*
0110  * Note that a shared library must be both readable and executable due to
0111  * security reasons.
0112  *
0113  * Also note that we take the address to load from from the file itself.
0114  */
0115 SYSCALL_DEFINE1(uselib, const char __user *, library)
0116 {
0117     struct linux_binfmt *fmt;
0118     struct file *file;
0119     struct filename *tmp = getname(library);
0120     int error = PTR_ERR(tmp);
0121     static const struct open_flags uselib_flags = {
0122         .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
0123         .acc_mode = MAY_READ | MAY_EXEC,
0124         .intent = LOOKUP_OPEN,
0125         .lookup_flags = LOOKUP_FOLLOW,
0126     };
0127 
0128     if (IS_ERR(tmp))
0129         goto out;
0130 
0131     file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
0132     putname(tmp);
0133     error = PTR_ERR(file);
0134     if (IS_ERR(file))
0135         goto out;
0136 
0137     error = -EINVAL;
0138     if (!S_ISREG(file_inode(file)->i_mode))
0139         goto exit;
0140 
0141     error = -EACCES;
0142     if (path_noexec(&file->f_path))
0143         goto exit;
0144 
0145     fsnotify_open(file);
0146 
0147     error = -ENOEXEC;
0148 
0149     read_lock(&binfmt_lock);
0150     list_for_each_entry(fmt, &formats, lh) {
0151         if (!fmt->load_shlib)
0152             continue;
0153         if (!try_module_get(fmt->module))
0154             continue;
0155         read_unlock(&binfmt_lock);
0156         error = fmt->load_shlib(file);
0157         read_lock(&binfmt_lock);
0158         put_binfmt(fmt);
0159         if (error != -ENOEXEC)
0160             break;
0161     }
0162     read_unlock(&binfmt_lock);
0163 exit:
0164     fput(file);
0165 out:
0166     return error;
0167 }
0168 #endif /* #ifdef CONFIG_USELIB */
0169 
0170 #ifdef CONFIG_MMU
0171 /*
0172  * The nascent bprm->mm is not visible until exec_mmap() but it can
0173  * use a lot of memory, account these pages in current->mm temporary
0174  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
0175  * change the counter back via acct_arg_size(0).
0176  */
0177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
0178 {
0179     struct mm_struct *mm = current->mm;
0180     long diff = (long)(pages - bprm->vma_pages);
0181 
0182     if (!mm || !diff)
0183         return;
0184 
0185     bprm->vma_pages = pages;
0186     add_mm_counter(mm, MM_ANONPAGES, diff);
0187 }
0188 
0189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
0190         int write)
0191 {
0192     struct page *page;
0193     int ret;
0194     unsigned int gup_flags = FOLL_FORCE;
0195 
0196 #ifdef CONFIG_STACK_GROWSUP
0197     if (write) {
0198         ret = expand_downwards(bprm->vma, pos);
0199         if (ret < 0)
0200             return NULL;
0201     }
0202 #endif
0203 
0204     if (write)
0205         gup_flags |= FOLL_WRITE;
0206 
0207     /*
0208      * We are doing an exec().  'current' is the process
0209      * doing the exec and bprm->mm is the new process's mm.
0210      */
0211     ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
0212             &page, NULL, NULL);
0213     if (ret <= 0)
0214         return NULL;
0215 
0216     if (write) {
0217         unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
0218         struct rlimit *rlim;
0219 
0220         acct_arg_size(bprm, size / PAGE_SIZE);
0221 
0222         /*
0223          * We've historically supported up to 32 pages (ARG_MAX)
0224          * of argument strings even with small stacks
0225          */
0226         if (size <= ARG_MAX)
0227             return page;
0228 
0229         /*
0230          * Limit to 1/4-th the stack size for the argv+env strings.
0231          * This ensures that:
0232          *  - the remaining binfmt code will not run out of stack space,
0233          *  - the program will have a reasonable amount of stack left
0234          *    to work from.
0235          */
0236         rlim = current->signal->rlim;
0237         if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
0238             put_page(page);
0239             return NULL;
0240         }
0241     }
0242 
0243     return page;
0244 }
0245 
0246 static void put_arg_page(struct page *page)
0247 {
0248     put_page(page);
0249 }
0250 
0251 static void free_arg_pages(struct linux_binprm *bprm)
0252 {
0253 }
0254 
0255 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
0256         struct page *page)
0257 {
0258     flush_cache_page(bprm->vma, pos, page_to_pfn(page));
0259 }
0260 
0261 static int __bprm_mm_init(struct linux_binprm *bprm)
0262 {
0263     int err;
0264     struct vm_area_struct *vma = NULL;
0265     struct mm_struct *mm = bprm->mm;
0266 
0267     bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
0268     if (!vma)
0269         return -ENOMEM;
0270 
0271     if (down_write_killable(&mm->mmap_sem)) {
0272         err = -EINTR;
0273         goto err_free;
0274     }
0275     vma->vm_mm = mm;
0276 
0277     /*
0278      * Place the stack at the largest stack address the architecture
0279      * supports. Later, we'll move this to an appropriate place. We don't
0280      * use STACK_TOP because that can depend on attributes which aren't
0281      * configured yet.
0282      */
0283     BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
0284     vma->vm_end = STACK_TOP_MAX;
0285     vma->vm_start = vma->vm_end - PAGE_SIZE;
0286     vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
0287     vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
0288     INIT_LIST_HEAD(&vma->anon_vma_chain);
0289 
0290     err = insert_vm_struct(mm, vma);
0291     if (err)
0292         goto err;
0293 
0294     mm->stack_vm = mm->total_vm = 1;
0295     arch_bprm_mm_init(mm, vma);
0296     up_write(&mm->mmap_sem);
0297     bprm->p = vma->vm_end - sizeof(void *);
0298     return 0;
0299 err:
0300     up_write(&mm->mmap_sem);
0301 err_free:
0302     bprm->vma = NULL;
0303     kmem_cache_free(vm_area_cachep, vma);
0304     return err;
0305 }
0306 
0307 static bool valid_arg_len(struct linux_binprm *bprm, long len)
0308 {
0309     return len <= MAX_ARG_STRLEN;
0310 }
0311 
0312 #else
0313 
0314 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
0315 {
0316 }
0317 
0318 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
0319         int write)
0320 {
0321     struct page *page;
0322 
0323     page = bprm->page[pos / PAGE_SIZE];
0324     if (!page && write) {
0325         page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
0326         if (!page)
0327             return NULL;
0328         bprm->page[pos / PAGE_SIZE] = page;
0329     }
0330 
0331     return page;
0332 }
0333 
0334 static void put_arg_page(struct page *page)
0335 {
0336 }
0337 
0338 static void free_arg_page(struct linux_binprm *bprm, int i)
0339 {
0340     if (bprm->page[i]) {
0341         __free_page(bprm->page[i]);
0342         bprm->page[i] = NULL;
0343     }
0344 }
0345 
0346 static void free_arg_pages(struct linux_binprm *bprm)
0347 {
0348     int i;
0349 
0350     for (i = 0; i < MAX_ARG_PAGES; i++)
0351         free_arg_page(bprm, i);
0352 }
0353 
0354 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
0355         struct page *page)
0356 {
0357 }
0358 
0359 static int __bprm_mm_init(struct linux_binprm *bprm)
0360 {
0361     bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
0362     return 0;
0363 }
0364 
0365 static bool valid_arg_len(struct linux_binprm *bprm, long len)
0366 {
0367     return len <= bprm->p;
0368 }
0369 
0370 #endif /* CONFIG_MMU */
0371 
0372 /*
0373  * Create a new mm_struct and populate it with a temporary stack
0374  * vm_area_struct.  We don't have enough context at this point to set the stack
0375  * flags, permissions, and offset, so we use temporary values.  We'll update
0376  * them later in setup_arg_pages().
0377  */
0378 static int bprm_mm_init(struct linux_binprm *bprm)
0379 {
0380     int err;
0381     struct mm_struct *mm = NULL;
0382 
0383     bprm->mm = mm = mm_alloc();
0384     err = -ENOMEM;
0385     if (!mm)
0386         goto err;
0387 
0388     err = __bprm_mm_init(bprm);
0389     if (err)
0390         goto err;
0391 
0392     return 0;
0393 
0394 err:
0395     if (mm) {
0396         bprm->mm = NULL;
0397         mmdrop(mm);
0398     }
0399 
0400     return err;
0401 }
0402 
0403 struct user_arg_ptr {
0404 #ifdef CONFIG_COMPAT
0405     bool is_compat;
0406 #endif
0407     union {
0408         const char __user *const __user *native;
0409 #ifdef CONFIG_COMPAT
0410         const compat_uptr_t __user *compat;
0411 #endif
0412     } ptr;
0413 };
0414 
0415 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
0416 {
0417     const char __user *native;
0418 
0419 #ifdef CONFIG_COMPAT
0420     if (unlikely(argv.is_compat)) {
0421         compat_uptr_t compat;
0422 
0423         if (get_user(compat, argv.ptr.compat + nr))
0424             return ERR_PTR(-EFAULT);
0425 
0426         return compat_ptr(compat);
0427     }
0428 #endif
0429 
0430     if (get_user(native, argv.ptr.native + nr))
0431         return ERR_PTR(-EFAULT);
0432 
0433     return native;
0434 }
0435 
0436 /*
0437  * count() counts the number of strings in array ARGV.
0438  */
0439 static int count(struct user_arg_ptr argv, int max)
0440 {
0441     int i = 0;
0442 
0443     if (argv.ptr.native != NULL) {
0444         for (;;) {
0445             const char __user *p = get_user_arg_ptr(argv, i);
0446 
0447             if (!p)
0448                 break;
0449 
0450             if (IS_ERR(p))
0451                 return -EFAULT;
0452 
0453             if (i >= max)
0454                 return -E2BIG;
0455             ++i;
0456 
0457             if (fatal_signal_pending(current))
0458                 return -ERESTARTNOHAND;
0459             cond_resched();
0460         }
0461     }
0462     return i;
0463 }
0464 
0465 /*
0466  * 'copy_strings()' copies argument/environment strings from the old
0467  * processes's memory to the new process's stack.  The call to get_user_pages()
0468  * ensures the destination page is created and not swapped out.
0469  */
0470 static int copy_strings(int argc, struct user_arg_ptr argv,
0471             struct linux_binprm *bprm)
0472 {
0473     struct page *kmapped_page = NULL;
0474     char *kaddr = NULL;
0475     unsigned long kpos = 0;
0476     int ret;
0477 
0478     while (argc-- > 0) {
0479         const char __user *str;
0480         int len;
0481         unsigned long pos;
0482 
0483         ret = -EFAULT;
0484         str = get_user_arg_ptr(argv, argc);
0485         if (IS_ERR(str))
0486             goto out;
0487 
0488         len = strnlen_user(str, MAX_ARG_STRLEN);
0489         if (!len)
0490             goto out;
0491 
0492         ret = -E2BIG;
0493         if (!valid_arg_len(bprm, len))
0494             goto out;
0495 
0496         /* We're going to work our way backwords. */
0497         pos = bprm->p;
0498         str += len;
0499         bprm->p -= len;
0500 
0501         while (len > 0) {
0502             int offset, bytes_to_copy;
0503 
0504             if (fatal_signal_pending(current)) {
0505                 ret = -ERESTARTNOHAND;
0506                 goto out;
0507             }
0508             cond_resched();
0509 
0510             offset = pos % PAGE_SIZE;
0511             if (offset == 0)
0512                 offset = PAGE_SIZE;
0513 
0514             bytes_to_copy = offset;
0515             if (bytes_to_copy > len)
0516                 bytes_to_copy = len;
0517 
0518             offset -= bytes_to_copy;
0519             pos -= bytes_to_copy;
0520             str -= bytes_to_copy;
0521             len -= bytes_to_copy;
0522 
0523             if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
0524                 struct page *page;
0525 
0526                 page = get_arg_page(bprm, pos, 1);
0527                 if (!page) {
0528                     ret = -E2BIG;
0529                     goto out;
0530                 }
0531 
0532                 if (kmapped_page) {
0533                     flush_kernel_dcache_page(kmapped_page);
0534                     kunmap(kmapped_page);
0535                     put_arg_page(kmapped_page);
0536                 }
0537                 kmapped_page = page;
0538                 kaddr = kmap(kmapped_page);
0539                 kpos = pos & PAGE_MASK;
0540                 flush_arg_page(bprm, kpos, kmapped_page);
0541             }
0542             if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
0543                 ret = -EFAULT;
0544                 goto out;
0545             }
0546         }
0547     }
0548     ret = 0;
0549 out:
0550     if (kmapped_page) {
0551         flush_kernel_dcache_page(kmapped_page);
0552         kunmap(kmapped_page);
0553         put_arg_page(kmapped_page);
0554     }
0555     return ret;
0556 }
0557 
0558 /*
0559  * Like copy_strings, but get argv and its values from kernel memory.
0560  */
0561 int copy_strings_kernel(int argc, const char *const *__argv,
0562             struct linux_binprm *bprm)
0563 {
0564     int r;
0565     mm_segment_t oldfs = get_fs();
0566     struct user_arg_ptr argv = {
0567         .ptr.native = (const char __user *const  __user *)__argv,
0568     };
0569 
0570     set_fs(KERNEL_DS);
0571     r = copy_strings(argc, argv, bprm);
0572     set_fs(oldfs);
0573 
0574     return r;
0575 }
0576 EXPORT_SYMBOL(copy_strings_kernel);
0577 
0578 #ifdef CONFIG_MMU
0579 
0580 /*
0581  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
0582  * the binfmt code determines where the new stack should reside, we shift it to
0583  * its final location.  The process proceeds as follows:
0584  *
0585  * 1) Use shift to calculate the new vma endpoints.
0586  * 2) Extend vma to cover both the old and new ranges.  This ensures the
0587  *    arguments passed to subsequent functions are consistent.
0588  * 3) Move vma's page tables to the new range.
0589  * 4) Free up any cleared pgd range.
0590  * 5) Shrink the vma to cover only the new range.
0591  */
0592 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
0593 {
0594     struct mm_struct *mm = vma->vm_mm;
0595     unsigned long old_start = vma->vm_start;
0596     unsigned long old_end = vma->vm_end;
0597     unsigned long length = old_end - old_start;
0598     unsigned long new_start = old_start - shift;
0599     unsigned long new_end = old_end - shift;
0600     struct mmu_gather tlb;
0601 
0602     BUG_ON(new_start > new_end);
0603 
0604     /*
0605      * ensure there are no vmas between where we want to go
0606      * and where we are
0607      */
0608     if (vma != find_vma(mm, new_start))
0609         return -EFAULT;
0610 
0611     /*
0612      * cover the whole range: [new_start, old_end)
0613      */
0614     if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
0615         return -ENOMEM;
0616 
0617     /*
0618      * move the page tables downwards, on failure we rely on
0619      * process cleanup to remove whatever mess we made.
0620      */
0621     if (length != move_page_tables(vma, old_start,
0622                        vma, new_start, length, false))
0623         return -ENOMEM;
0624 
0625     lru_add_drain();
0626     tlb_gather_mmu(&tlb, mm, old_start, old_end);
0627     if (new_end > old_start) {
0628         /*
0629          * when the old and new regions overlap clear from new_end.
0630          */
0631         free_pgd_range(&tlb, new_end, old_end, new_end,
0632             vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
0633     } else {
0634         /*
0635          * otherwise, clean from old_start; this is done to not touch
0636          * the address space in [new_end, old_start) some architectures
0637          * have constraints on va-space that make this illegal (IA64) -
0638          * for the others its just a little faster.
0639          */
0640         free_pgd_range(&tlb, old_start, old_end, new_end,
0641             vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
0642     }
0643     tlb_finish_mmu(&tlb, old_start, old_end);
0644 
0645     /*
0646      * Shrink the vma to just the new range.  Always succeeds.
0647      */
0648     vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
0649 
0650     return 0;
0651 }
0652 
0653 /*
0654  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
0655  * the stack is optionally relocated, and some extra space is added.
0656  */
0657 int setup_arg_pages(struct linux_binprm *bprm,
0658             unsigned long stack_top,
0659             int executable_stack)
0660 {
0661     unsigned long ret;
0662     unsigned long stack_shift;
0663     struct mm_struct *mm = current->mm;
0664     struct vm_area_struct *vma = bprm->vma;
0665     struct vm_area_struct *prev = NULL;
0666     unsigned long vm_flags;
0667     unsigned long stack_base;
0668     unsigned long stack_size;
0669     unsigned long stack_expand;
0670     unsigned long rlim_stack;
0671 
0672 #ifdef CONFIG_STACK_GROWSUP
0673     /* Limit stack size */
0674     stack_base = rlimit_max(RLIMIT_STACK);
0675     if (stack_base > STACK_SIZE_MAX)
0676         stack_base = STACK_SIZE_MAX;
0677 
0678     /* Add space for stack randomization. */
0679     stack_base += (STACK_RND_MASK << PAGE_SHIFT);
0680 
0681     /* Make sure we didn't let the argument array grow too large. */
0682     if (vma->vm_end - vma->vm_start > stack_base)
0683         return -ENOMEM;
0684 
0685     stack_base = PAGE_ALIGN(stack_top - stack_base);
0686 
0687     stack_shift = vma->vm_start - stack_base;
0688     mm->arg_start = bprm->p - stack_shift;
0689     bprm->p = vma->vm_end - stack_shift;
0690 #else
0691     stack_top = arch_align_stack(stack_top);
0692     stack_top = PAGE_ALIGN(stack_top);
0693 
0694     if (unlikely(stack_top < mmap_min_addr) ||
0695         unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
0696         return -ENOMEM;
0697 
0698     stack_shift = vma->vm_end - stack_top;
0699 
0700     bprm->p -= stack_shift;
0701     mm->arg_start = bprm->p;
0702 #endif
0703 
0704     if (bprm->loader)
0705         bprm->loader -= stack_shift;
0706     bprm->exec -= stack_shift;
0707 
0708     if (down_write_killable(&mm->mmap_sem))
0709         return -EINTR;
0710 
0711     vm_flags = VM_STACK_FLAGS;
0712 
0713     /*
0714      * Adjust stack execute permissions; explicitly enable for
0715      * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
0716      * (arch default) otherwise.
0717      */
0718     if (unlikely(executable_stack == EXSTACK_ENABLE_X))
0719         vm_flags |= VM_EXEC;
0720     else if (executable_stack == EXSTACK_DISABLE_X)
0721         vm_flags &= ~VM_EXEC;
0722     vm_flags |= mm->def_flags;
0723     vm_flags |= VM_STACK_INCOMPLETE_SETUP;
0724 
0725     ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
0726             vm_flags);
0727     if (ret)
0728         goto out_unlock;
0729     BUG_ON(prev != vma);
0730 
0731     /* Move stack pages down in memory. */
0732     if (stack_shift) {
0733         ret = shift_arg_pages(vma, stack_shift);
0734         if (ret)
0735             goto out_unlock;
0736     }
0737 
0738     /* mprotect_fixup is overkill to remove the temporary stack flags */
0739     vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
0740 
0741     stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
0742     stack_size = vma->vm_end - vma->vm_start;
0743     /*
0744      * Align this down to a page boundary as expand_stack
0745      * will align it up.
0746      */
0747     rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
0748 #ifdef CONFIG_STACK_GROWSUP
0749     if (stack_size + stack_expand > rlim_stack)
0750         stack_base = vma->vm_start + rlim_stack;
0751     else
0752         stack_base = vma->vm_end + stack_expand;
0753 #else
0754     if (stack_size + stack_expand > rlim_stack)
0755         stack_base = vma->vm_end - rlim_stack;
0756     else
0757         stack_base = vma->vm_start - stack_expand;
0758 #endif
0759     current->mm->start_stack = bprm->p;
0760     ret = expand_stack(vma, stack_base);
0761     if (ret)
0762         ret = -EFAULT;
0763 
0764 out_unlock:
0765     up_write(&mm->mmap_sem);
0766     return ret;
0767 }
0768 EXPORT_SYMBOL(setup_arg_pages);
0769 
0770 #else
0771 
0772 /*
0773  * Transfer the program arguments and environment from the holding pages
0774  * onto the stack. The provided stack pointer is adjusted accordingly.
0775  */
0776 int transfer_args_to_stack(struct linux_binprm *bprm,
0777                unsigned long *sp_location)
0778 {
0779     unsigned long index, stop, sp;
0780     int ret = 0;
0781 
0782     stop = bprm->p >> PAGE_SHIFT;
0783     sp = *sp_location;
0784 
0785     for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
0786         unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
0787         char *src = kmap(bprm->page[index]) + offset;
0788         sp -= PAGE_SIZE - offset;
0789         if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
0790             ret = -EFAULT;
0791         kunmap(bprm->page[index]);
0792         if (ret)
0793             goto out;
0794     }
0795 
0796     *sp_location = sp;
0797 
0798 out:
0799     return ret;
0800 }
0801 EXPORT_SYMBOL(transfer_args_to_stack);
0802 
0803 #endif /* CONFIG_MMU */
0804 
0805 static struct file *do_open_execat(int fd, struct filename *name, int flags)
0806 {
0807     struct file *file;
0808     int err;
0809     struct open_flags open_exec_flags = {
0810         .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
0811         .acc_mode = MAY_EXEC,
0812         .intent = LOOKUP_OPEN,
0813         .lookup_flags = LOOKUP_FOLLOW,
0814     };
0815 
0816     if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
0817         return ERR_PTR(-EINVAL);
0818     if (flags & AT_SYMLINK_NOFOLLOW)
0819         open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
0820     if (flags & AT_EMPTY_PATH)
0821         open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
0822 
0823     file = do_filp_open(fd, name, &open_exec_flags);
0824     if (IS_ERR(file))
0825         goto out;
0826 
0827     err = -EACCES;
0828     if (!S_ISREG(file_inode(file)->i_mode))
0829         goto exit;
0830 
0831     if (path_noexec(&file->f_path))
0832         goto exit;
0833 
0834     err = deny_write_access(file);
0835     if (err)
0836         goto exit;
0837 
0838     if (name->name[0] != '\0')
0839         fsnotify_open(file);
0840 
0841 out:
0842     return file;
0843 
0844 exit:
0845     fput(file);
0846     return ERR_PTR(err);
0847 }
0848 
0849 struct file *open_exec(const char *name)
0850 {
0851     struct filename *filename = getname_kernel(name);
0852     struct file *f = ERR_CAST(filename);
0853 
0854     if (!IS_ERR(filename)) {
0855         f = do_open_execat(AT_FDCWD, filename, 0);
0856         putname(filename);
0857     }
0858     return f;
0859 }
0860 EXPORT_SYMBOL(open_exec);
0861 
0862 int kernel_read(struct file *file, loff_t offset,
0863         char *addr, unsigned long count)
0864 {
0865     mm_segment_t old_fs;
0866     loff_t pos = offset;
0867     int result;
0868 
0869     old_fs = get_fs();
0870     set_fs(get_ds());
0871     /* The cast to a user pointer is valid due to the set_fs() */
0872     result = vfs_read(file, (void __user *)addr, count, &pos);
0873     set_fs(old_fs);
0874     return result;
0875 }
0876 
0877 EXPORT_SYMBOL(kernel_read);
0878 
0879 int kernel_read_file(struct file *file, void **buf, loff_t *size,
0880              loff_t max_size, enum kernel_read_file_id id)
0881 {
0882     loff_t i_size, pos;
0883     ssize_t bytes = 0;
0884     int ret;
0885 
0886     if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
0887         return -EINVAL;
0888 
0889     ret = security_kernel_read_file(file, id);
0890     if (ret)
0891         return ret;
0892 
0893     ret = deny_write_access(file);
0894     if (ret)
0895         return ret;
0896 
0897     i_size = i_size_read(file_inode(file));
0898     if (max_size > 0 && i_size > max_size) {
0899         ret = -EFBIG;
0900         goto out;
0901     }
0902     if (i_size <= 0) {
0903         ret = -EINVAL;
0904         goto out;
0905     }
0906 
0907     if (id != READING_FIRMWARE_PREALLOC_BUFFER)
0908         *buf = vmalloc(i_size);
0909     if (!*buf) {
0910         ret = -ENOMEM;
0911         goto out;
0912     }
0913 
0914     pos = 0;
0915     while (pos < i_size) {
0916         bytes = kernel_read(file, pos, (char *)(*buf) + pos,
0917                     i_size - pos);
0918         if (bytes < 0) {
0919             ret = bytes;
0920             goto out;
0921         }
0922 
0923         if (bytes == 0)
0924             break;
0925         pos += bytes;
0926     }
0927 
0928     if (pos != i_size) {
0929         ret = -EIO;
0930         goto out_free;
0931     }
0932 
0933     ret = security_kernel_post_read_file(file, *buf, i_size, id);
0934     if (!ret)
0935         *size = pos;
0936 
0937 out_free:
0938     if (ret < 0) {
0939         if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
0940             vfree(*buf);
0941             *buf = NULL;
0942         }
0943     }
0944 
0945 out:
0946     allow_write_access(file);
0947     return ret;
0948 }
0949 EXPORT_SYMBOL_GPL(kernel_read_file);
0950 
0951 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
0952                    loff_t max_size, enum kernel_read_file_id id)
0953 {
0954     struct file *file;
0955     int ret;
0956 
0957     if (!path || !*path)
0958         return -EINVAL;
0959 
0960     file = filp_open(path, O_RDONLY, 0);
0961     if (IS_ERR(file))
0962         return PTR_ERR(file);
0963 
0964     ret = kernel_read_file(file, buf, size, max_size, id);
0965     fput(file);
0966     return ret;
0967 }
0968 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
0969 
0970 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
0971                  enum kernel_read_file_id id)
0972 {
0973     struct fd f = fdget(fd);
0974     int ret = -EBADF;
0975 
0976     if (!f.file)
0977         goto out;
0978 
0979     ret = kernel_read_file(f.file, buf, size, max_size, id);
0980 out:
0981     fdput(f);
0982     return ret;
0983 }
0984 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
0985 
0986 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
0987 {
0988     ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
0989     if (res > 0)
0990         flush_icache_range(addr, addr + len);
0991     return res;
0992 }
0993 EXPORT_SYMBOL(read_code);
0994 
0995 static int exec_mmap(struct mm_struct *mm)
0996 {
0997     struct task_struct *tsk;
0998     struct mm_struct *old_mm, *active_mm;
0999 
1000     /* Notify parent that we're no longer interested in the old VM */
1001     tsk = current;
1002     old_mm = current->mm;
1003     mm_release(tsk, old_mm);
1004 
1005     if (old_mm) {
1006         sync_mm_rss(old_mm);
1007         /*
1008          * Make sure that if there is a core dump in progress
1009          * for the old mm, we get out and die instead of going
1010          * through with the exec.  We must hold mmap_sem around
1011          * checking core_state and changing tsk->mm.
1012          */
1013         down_read(&old_mm->mmap_sem);
1014         if (unlikely(old_mm->core_state)) {
1015             up_read(&old_mm->mmap_sem);
1016             return -EINTR;
1017         }
1018     }
1019     task_lock(tsk);
1020     active_mm = tsk->active_mm;
1021     tsk->mm = mm;
1022     tsk->active_mm = mm;
1023     activate_mm(active_mm, mm);
1024     tsk->mm->vmacache_seqnum = 0;
1025     vmacache_flush(tsk);
1026     task_unlock(tsk);
1027     if (old_mm) {
1028         up_read(&old_mm->mmap_sem);
1029         BUG_ON(active_mm != old_mm);
1030         setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1031         mm_update_next_owner(old_mm);
1032         mmput(old_mm);
1033         return 0;
1034     }
1035     mmdrop(active_mm);
1036     return 0;
1037 }
1038 
1039 /*
1040  * This function makes sure the current process has its own signal table,
1041  * so that flush_signal_handlers can later reset the handlers without
1042  * disturbing other processes.  (Other processes might share the signal
1043  * table via the CLONE_SIGHAND option to clone().)
1044  */
1045 static int de_thread(struct task_struct *tsk)
1046 {
1047     struct signal_struct *sig = tsk->signal;
1048     struct sighand_struct *oldsighand = tsk->sighand;
1049     spinlock_t *lock = &oldsighand->siglock;
1050 
1051     if (thread_group_empty(tsk))
1052         goto no_thread_group;
1053 
1054     /*
1055      * Kill all other threads in the thread group.
1056      */
1057     spin_lock_irq(lock);
1058     if (signal_group_exit(sig)) {
1059         /*
1060          * Another group action in progress, just
1061          * return so that the signal is processed.
1062          */
1063         spin_unlock_irq(lock);
1064         return -EAGAIN;
1065     }
1066 
1067     sig->group_exit_task = tsk;
1068     sig->notify_count = zap_other_threads(tsk);
1069     if (!thread_group_leader(tsk))
1070         sig->notify_count--;
1071 
1072     while (sig->notify_count) {
1073         __set_current_state(TASK_KILLABLE);
1074         spin_unlock_irq(lock);
1075         schedule();
1076         if (unlikely(__fatal_signal_pending(tsk)))
1077             goto killed;
1078         spin_lock_irq(lock);
1079     }
1080     spin_unlock_irq(lock);
1081 
1082     /*
1083      * At this point all other threads have exited, all we have to
1084      * do is to wait for the thread group leader to become inactive,
1085      * and to assume its PID:
1086      */
1087     if (!thread_group_leader(tsk)) {
1088         struct task_struct *leader = tsk->group_leader;
1089 
1090         for (;;) {
1091             threadgroup_change_begin(tsk);
1092             write_lock_irq(&tasklist_lock);
1093             /*
1094              * Do this under tasklist_lock to ensure that
1095              * exit_notify() can't miss ->group_exit_task
1096              */
1097             sig->notify_count = -1;
1098             if (likely(leader->exit_state))
1099                 break;
1100             __set_current_state(TASK_KILLABLE);
1101             write_unlock_irq(&tasklist_lock);
1102             threadgroup_change_end(tsk);
1103             schedule();
1104             if (unlikely(__fatal_signal_pending(tsk)))
1105                 goto killed;
1106         }
1107 
1108         /*
1109          * The only record we have of the real-time age of a
1110          * process, regardless of execs it's done, is start_time.
1111          * All the past CPU time is accumulated in signal_struct
1112          * from sister threads now dead.  But in this non-leader
1113          * exec, nothing survives from the original leader thread,
1114          * whose birth marks the true age of this process now.
1115          * When we take on its identity by switching to its PID, we
1116          * also take its birthdate (always earlier than our own).
1117          */
1118         tsk->start_time = leader->start_time;
1119         tsk->real_start_time = leader->real_start_time;
1120 
1121         BUG_ON(!same_thread_group(leader, tsk));
1122         BUG_ON(has_group_leader_pid(tsk));
1123         /*
1124          * An exec() starts a new thread group with the
1125          * TGID of the previous thread group. Rehash the
1126          * two threads with a switched PID, and release
1127          * the former thread group leader:
1128          */
1129 
1130         /* Become a process group leader with the old leader's pid.
1131          * The old leader becomes a thread of the this thread group.
1132          * Note: The old leader also uses this pid until release_task
1133          *       is called.  Odd but simple and correct.
1134          */
1135         tsk->pid = leader->pid;
1136         change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1137         transfer_pid(leader, tsk, PIDTYPE_PGID);
1138         transfer_pid(leader, tsk, PIDTYPE_SID);
1139 
1140         list_replace_rcu(&leader->tasks, &tsk->tasks);
1141         list_replace_init(&leader->sibling, &tsk->sibling);
1142 
1143         tsk->group_leader = tsk;
1144         leader->group_leader = tsk;
1145 
1146         tsk->exit_signal = SIGCHLD;
1147         leader->exit_signal = -1;
1148 
1149         BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1150         leader->exit_state = EXIT_DEAD;
1151 
1152         /*
1153          * We are going to release_task()->ptrace_unlink() silently,
1154          * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1155          * the tracer wont't block again waiting for this thread.
1156          */
1157         if (unlikely(leader->ptrace))
1158             __wake_up_parent(leader, leader->parent);
1159         write_unlock_irq(&tasklist_lock);
1160         threadgroup_change_end(tsk);
1161 
1162         release_task(leader);
1163     }
1164 
1165     sig->group_exit_task = NULL;
1166     sig->notify_count = 0;
1167 
1168 no_thread_group:
1169     /* we have changed execution domain */
1170     tsk->exit_signal = SIGCHLD;
1171 
1172 #ifdef CONFIG_POSIX_TIMERS
1173     exit_itimers(sig);
1174     flush_itimer_signals();
1175 #endif
1176 
1177     if (atomic_read(&oldsighand->count) != 1) {
1178         struct sighand_struct *newsighand;
1179         /*
1180          * This ->sighand is shared with the CLONE_SIGHAND
1181          * but not CLONE_THREAD task, switch to the new one.
1182          */
1183         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1184         if (!newsighand)
1185             return -ENOMEM;
1186 
1187         atomic_set(&newsighand->count, 1);
1188         memcpy(newsighand->action, oldsighand->action,
1189                sizeof(newsighand->action));
1190 
1191         write_lock_irq(&tasklist_lock);
1192         spin_lock(&oldsighand->siglock);
1193         rcu_assign_pointer(tsk->sighand, newsighand);
1194         spin_unlock(&oldsighand->siglock);
1195         write_unlock_irq(&tasklist_lock);
1196 
1197         __cleanup_sighand(oldsighand);
1198     }
1199 
1200     BUG_ON(!thread_group_leader(tsk));
1201     return 0;
1202 
1203 killed:
1204     /* protects against exit_notify() and __exit_signal() */
1205     read_lock(&tasklist_lock);
1206     sig->group_exit_task = NULL;
1207     sig->notify_count = 0;
1208     read_unlock(&tasklist_lock);
1209     return -EAGAIN;
1210 }
1211 
1212 char *get_task_comm(char *buf, struct task_struct *tsk)
1213 {
1214     /* buf must be at least sizeof(tsk->comm) in size */
1215     task_lock(tsk);
1216     strncpy(buf, tsk->comm, sizeof(tsk->comm));
1217     task_unlock(tsk);
1218     return buf;
1219 }
1220 EXPORT_SYMBOL_GPL(get_task_comm);
1221 
1222 /*
1223  * These functions flushes out all traces of the currently running executable
1224  * so that a new one can be started
1225  */
1226 
1227 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1228 {
1229     task_lock(tsk);
1230     trace_task_rename(tsk, buf);
1231     strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1232     task_unlock(tsk);
1233     perf_event_comm(tsk, exec);
1234 }
1235 
1236 int flush_old_exec(struct linux_binprm * bprm)
1237 {
1238     int retval;
1239 
1240     /*
1241      * Make sure we have a private signal table and that
1242      * we are unassociated from the previous thread group.
1243      */
1244     retval = de_thread(current);
1245     if (retval)
1246         goto out;
1247 
1248     /*
1249      * Must be called _before_ exec_mmap() as bprm->mm is
1250      * not visibile until then. This also enables the update
1251      * to be lockless.
1252      */
1253     set_mm_exe_file(bprm->mm, bprm->file);
1254 
1255     /*
1256      * Release all of the old mmap stuff
1257      */
1258     acct_arg_size(bprm, 0);
1259     retval = exec_mmap(bprm->mm);
1260     if (retval)
1261         goto out;
1262 
1263     bprm->mm = NULL;        /* We're using it now */
1264 
1265     set_fs(USER_DS);
1266     current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1267                     PF_NOFREEZE | PF_NO_SETAFFINITY);
1268     flush_thread();
1269     current->personality &= ~bprm->per_clear;
1270 
1271     /*
1272      * We have to apply CLOEXEC before we change whether the process is
1273      * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1274      * trying to access the should-be-closed file descriptors of a process
1275      * undergoing exec(2).
1276      */
1277     do_close_on_exec(current->files);
1278     return 0;
1279 
1280 out:
1281     return retval;
1282 }
1283 EXPORT_SYMBOL(flush_old_exec);
1284 
1285 void would_dump(struct linux_binprm *bprm, struct file *file)
1286 {
1287     struct inode *inode = file_inode(file);
1288     if (inode_permission(inode, MAY_READ) < 0) {
1289         struct user_namespace *old, *user_ns;
1290         bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1291 
1292         /* Ensure mm->user_ns contains the executable */
1293         user_ns = old = bprm->mm->user_ns;
1294         while ((user_ns != &init_user_ns) &&
1295                !privileged_wrt_inode_uidgid(user_ns, inode))
1296             user_ns = user_ns->parent;
1297 
1298         if (old != user_ns) {
1299             bprm->mm->user_ns = get_user_ns(user_ns);
1300             put_user_ns(old);
1301         }
1302     }
1303 }
1304 EXPORT_SYMBOL(would_dump);
1305 
1306 void setup_new_exec(struct linux_binprm * bprm)
1307 {
1308     arch_pick_mmap_layout(current->mm);
1309 
1310     /* This is the point of no return */
1311     current->sas_ss_sp = current->sas_ss_size = 0;
1312 
1313     if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1314         set_dumpable(current->mm, SUID_DUMP_USER);
1315     else
1316         set_dumpable(current->mm, suid_dumpable);
1317 
1318     perf_event_exec();
1319     __set_task_comm(current, kbasename(bprm->filename), true);
1320 
1321     /* Set the new mm task size. We have to do that late because it may
1322      * depend on TIF_32BIT which is only updated in flush_thread() on
1323      * some architectures like powerpc
1324      */
1325     current->mm->task_size = TASK_SIZE;
1326 
1327     /* install the new credentials */
1328     if (!uid_eq(bprm->cred->uid, current_euid()) ||
1329         !gid_eq(bprm->cred->gid, current_egid())) {
1330         current->pdeath_signal = 0;
1331     } else {
1332         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1333             set_dumpable(current->mm, suid_dumpable);
1334     }
1335 
1336     /* An exec changes our domain. We are no longer part of the thread
1337        group */
1338     current->self_exec_id++;
1339     flush_signal_handlers(current, 0);
1340 }
1341 EXPORT_SYMBOL(setup_new_exec);
1342 
1343 /*
1344  * Prepare credentials and lock ->cred_guard_mutex.
1345  * install_exec_creds() commits the new creds and drops the lock.
1346  * Or, if exec fails before, free_bprm() should release ->cred and
1347  * and unlock.
1348  */
1349 int prepare_bprm_creds(struct linux_binprm *bprm)
1350 {
1351     if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1352         return -ERESTARTNOINTR;
1353 
1354     bprm->cred = prepare_exec_creds();
1355     if (likely(bprm->cred))
1356         return 0;
1357 
1358     mutex_unlock(&current->signal->cred_guard_mutex);
1359     return -ENOMEM;
1360 }
1361 
1362 static void free_bprm(struct linux_binprm *bprm)
1363 {
1364     free_arg_pages(bprm);
1365     if (bprm->cred) {
1366         mutex_unlock(&current->signal->cred_guard_mutex);
1367         abort_creds(bprm->cred);
1368     }
1369     if (bprm->file) {
1370         allow_write_access(bprm->file);
1371         fput(bprm->file);
1372     }
1373     /* If a binfmt changed the interp, free it. */
1374     if (bprm->interp != bprm->filename)
1375         kfree(bprm->interp);
1376     kfree(bprm);
1377 }
1378 
1379 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1380 {
1381     /* If a binfmt changed the interp, free it first. */
1382     if (bprm->interp != bprm->filename)
1383         kfree(bprm->interp);
1384     bprm->interp = kstrdup(interp, GFP_KERNEL);
1385     if (!bprm->interp)
1386         return -ENOMEM;
1387     return 0;
1388 }
1389 EXPORT_SYMBOL(bprm_change_interp);
1390 
1391 /*
1392  * install the new credentials for this executable
1393  */
1394 void install_exec_creds(struct linux_binprm *bprm)
1395 {
1396     security_bprm_committing_creds(bprm);
1397 
1398     commit_creds(bprm->cred);
1399     bprm->cred = NULL;
1400 
1401     /*
1402      * Disable monitoring for regular users
1403      * when executing setuid binaries. Must
1404      * wait until new credentials are committed
1405      * by commit_creds() above
1406      */
1407     if (get_dumpable(current->mm) != SUID_DUMP_USER)
1408         perf_event_exit_task(current);
1409     /*
1410      * cred_guard_mutex must be held at least to this point to prevent
1411      * ptrace_attach() from altering our determination of the task's
1412      * credentials; any time after this it may be unlocked.
1413      */
1414     security_bprm_committed_creds(bprm);
1415     mutex_unlock(&current->signal->cred_guard_mutex);
1416 }
1417 EXPORT_SYMBOL(install_exec_creds);
1418 
1419 /*
1420  * determine how safe it is to execute the proposed program
1421  * - the caller must hold ->cred_guard_mutex to protect against
1422  *   PTRACE_ATTACH or seccomp thread-sync
1423  */
1424 static void check_unsafe_exec(struct linux_binprm *bprm)
1425 {
1426     struct task_struct *p = current, *t;
1427     unsigned n_fs;
1428 
1429     if (p->ptrace) {
1430         if (ptracer_capable(p, current_user_ns()))
1431             bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1432         else
1433             bprm->unsafe |= LSM_UNSAFE_PTRACE;
1434     }
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 = file_inode(bprm->file);
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