4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/ima.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
55 #include <linux/fsnotify.h>
56 #include <linux/fs_struct.h>
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
67 /* The maximal length of core_pattern is also specified in sysctl.c */
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
72 int __register_binfmt(struct linux_binfmt * fmt, int insert)
76 write_lock(&binfmt_lock);
77 insert ? list_add(&fmt->lh, &formats) :
78 list_add_tail(&fmt->lh, &formats);
79 write_unlock(&binfmt_lock);
83 EXPORT_SYMBOL(__register_binfmt);
85 void unregister_binfmt(struct linux_binfmt * fmt)
87 write_lock(&binfmt_lock);
89 write_unlock(&binfmt_lock);
92 EXPORT_SYMBOL(unregister_binfmt);
94 static inline void put_binfmt(struct linux_binfmt * fmt)
96 module_put(fmt->module);
100 * Note that a shared library must be both readable and executable due to
103 * Also note that we take the address to load from from the file itself.
105 SYSCALL_DEFINE1(uselib, const char __user *, library)
109 char *tmp = getname(library);
110 int error = PTR_ERR(tmp);
113 error = path_lookup_open(AT_FDCWD, tmp,
115 FMODE_READ|FMODE_EXEC);
122 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
126 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
129 error = may_open(&nd.path, MAY_READ | MAY_EXEC | MAY_OPEN, 0);
133 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
134 error = PTR_ERR(file);
138 fsnotify_open(file->f_path.dentry);
142 struct linux_binfmt * fmt;
144 read_lock(&binfmt_lock);
145 list_for_each_entry(fmt, &formats, lh) {
146 if (!fmt->load_shlib)
148 if (!try_module_get(fmt->module))
150 read_unlock(&binfmt_lock);
151 error = fmt->load_shlib(file);
152 read_lock(&binfmt_lock);
154 if (error != -ENOEXEC)
157 read_unlock(&binfmt_lock);
163 release_open_intent(&nd);
170 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
176 #ifdef CONFIG_STACK_GROWSUP
178 ret = expand_stack_downwards(bprm->vma, pos);
183 ret = get_user_pages(current, bprm->mm, pos,
184 1, write, 1, &page, NULL);
189 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
193 * We've historically supported up to 32 pages (ARG_MAX)
194 * of argument strings even with small stacks
200 * Limit to 1/4-th the stack size for the argv+env strings.
202 * - the remaining binfmt code will not run out of stack space,
203 * - the program will have a reasonable amount of stack left
206 rlim = current->signal->rlim;
207 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
216 static void put_arg_page(struct page *page)
221 static void free_arg_page(struct linux_binprm *bprm, int i)
225 static void free_arg_pages(struct linux_binprm *bprm)
229 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
232 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
235 static int __bprm_mm_init(struct linux_binprm *bprm)
238 struct vm_area_struct *vma = NULL;
239 struct mm_struct *mm = bprm->mm;
241 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
245 down_write(&mm->mmap_sem);
249 * Place the stack at the largest stack address the architecture
250 * supports. Later, we'll move this to an appropriate place. We don't
251 * use STACK_TOP because that can depend on attributes which aren't
254 vma->vm_end = STACK_TOP_MAX;
255 vma->vm_start = vma->vm_end - PAGE_SIZE;
256 vma->vm_flags = VM_STACK_FLAGS;
257 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
258 err = insert_vm_struct(mm, vma);
262 mm->stack_vm = mm->total_vm = 1;
263 up_write(&mm->mmap_sem);
264 bprm->p = vma->vm_end - sizeof(void *);
267 up_write(&mm->mmap_sem);
269 kmem_cache_free(vm_area_cachep, vma);
273 static bool valid_arg_len(struct linux_binprm *bprm, long len)
275 return len <= MAX_ARG_STRLEN;
280 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
285 page = bprm->page[pos / PAGE_SIZE];
286 if (!page && write) {
287 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
290 bprm->page[pos / PAGE_SIZE] = page;
296 static void put_arg_page(struct page *page)
300 static void free_arg_page(struct linux_binprm *bprm, int i)
303 __free_page(bprm->page[i]);
304 bprm->page[i] = NULL;
308 static void free_arg_pages(struct linux_binprm *bprm)
312 for (i = 0; i < MAX_ARG_PAGES; i++)
313 free_arg_page(bprm, i);
316 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
321 static int __bprm_mm_init(struct linux_binprm *bprm)
323 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
327 static bool valid_arg_len(struct linux_binprm *bprm, long len)
329 return len <= bprm->p;
332 #endif /* CONFIG_MMU */
335 * Create a new mm_struct and populate it with a temporary stack
336 * vm_area_struct. We don't have enough context at this point to set the stack
337 * flags, permissions, and offset, so we use temporary values. We'll update
338 * them later in setup_arg_pages().
340 int bprm_mm_init(struct linux_binprm *bprm)
343 struct mm_struct *mm = NULL;
345 bprm->mm = mm = mm_alloc();
350 err = init_new_context(current, mm);
354 err = __bprm_mm_init(bprm);
370 * count() counts the number of strings in array ARGV.
372 static int count(char __user * __user * argv, int max)
380 if (get_user(p, argv))
394 * 'copy_strings()' copies argument/environment strings from the old
395 * processes's memory to the new process's stack. The call to get_user_pages()
396 * ensures the destination page is created and not swapped out.
398 static int copy_strings(int argc, char __user * __user * argv,
399 struct linux_binprm *bprm)
401 struct page *kmapped_page = NULL;
403 unsigned long kpos = 0;
411 if (get_user(str, argv+argc) ||
412 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
417 if (!valid_arg_len(bprm, len)) {
422 /* We're going to work our way backwords. */
428 int offset, bytes_to_copy;
430 offset = pos % PAGE_SIZE;
434 bytes_to_copy = offset;
435 if (bytes_to_copy > len)
438 offset -= bytes_to_copy;
439 pos -= bytes_to_copy;
440 str -= bytes_to_copy;
441 len -= bytes_to_copy;
443 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
446 page = get_arg_page(bprm, pos, 1);
453 flush_kernel_dcache_page(kmapped_page);
454 kunmap(kmapped_page);
455 put_arg_page(kmapped_page);
458 kaddr = kmap(kmapped_page);
459 kpos = pos & PAGE_MASK;
460 flush_arg_page(bprm, kpos, kmapped_page);
462 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
471 flush_kernel_dcache_page(kmapped_page);
472 kunmap(kmapped_page);
473 put_arg_page(kmapped_page);
479 * Like copy_strings, but get argv and its values from kernel memory.
481 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
484 mm_segment_t oldfs = get_fs();
486 r = copy_strings(argc, (char __user * __user *)argv, bprm);
490 EXPORT_SYMBOL(copy_strings_kernel);
495 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
496 * the binfmt code determines where the new stack should reside, we shift it to
497 * its final location. The process proceeds as follows:
499 * 1) Use shift to calculate the new vma endpoints.
500 * 2) Extend vma to cover both the old and new ranges. This ensures the
501 * arguments passed to subsequent functions are consistent.
502 * 3) Move vma's page tables to the new range.
503 * 4) Free up any cleared pgd range.
504 * 5) Shrink the vma to cover only the new range.
506 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
508 struct mm_struct *mm = vma->vm_mm;
509 unsigned long old_start = vma->vm_start;
510 unsigned long old_end = vma->vm_end;
511 unsigned long length = old_end - old_start;
512 unsigned long new_start = old_start - shift;
513 unsigned long new_end = old_end - shift;
514 struct mmu_gather *tlb;
516 BUG_ON(new_start > new_end);
519 * ensure there are no vmas between where we want to go
522 if (vma != find_vma(mm, new_start))
526 * cover the whole range: [new_start, old_end)
528 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
531 * move the page tables downwards, on failure we rely on
532 * process cleanup to remove whatever mess we made.
534 if (length != move_page_tables(vma, old_start,
535 vma, new_start, length))
539 tlb = tlb_gather_mmu(mm, 0);
540 if (new_end > old_start) {
542 * when the old and new regions overlap clear from new_end.
544 free_pgd_range(tlb, new_end, old_end, new_end,
545 vma->vm_next ? vma->vm_next->vm_start : 0);
548 * otherwise, clean from old_start; this is done to not touch
549 * the address space in [new_end, old_start) some architectures
550 * have constraints on va-space that make this illegal (IA64) -
551 * for the others its just a little faster.
553 free_pgd_range(tlb, old_start, old_end, new_end,
554 vma->vm_next ? vma->vm_next->vm_start : 0);
556 tlb_finish_mmu(tlb, new_end, old_end);
559 * shrink the vma to just the new range.
561 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
566 #define EXTRA_STACK_VM_PAGES 20 /* random */
569 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
570 * the stack is optionally relocated, and some extra space is added.
572 int setup_arg_pages(struct linux_binprm *bprm,
573 unsigned long stack_top,
574 int executable_stack)
577 unsigned long stack_shift;
578 struct mm_struct *mm = current->mm;
579 struct vm_area_struct *vma = bprm->vma;
580 struct vm_area_struct *prev = NULL;
581 unsigned long vm_flags;
582 unsigned long stack_base;
584 #ifdef CONFIG_STACK_GROWSUP
585 /* Limit stack size to 1GB */
586 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
587 if (stack_base > (1 << 30))
588 stack_base = 1 << 30;
590 /* Make sure we didn't let the argument array grow too large. */
591 if (vma->vm_end - vma->vm_start > stack_base)
594 stack_base = PAGE_ALIGN(stack_top - stack_base);
596 stack_shift = vma->vm_start - stack_base;
597 mm->arg_start = bprm->p - stack_shift;
598 bprm->p = vma->vm_end - stack_shift;
600 stack_top = arch_align_stack(stack_top);
601 stack_top = PAGE_ALIGN(stack_top);
602 stack_shift = vma->vm_end - stack_top;
604 bprm->p -= stack_shift;
605 mm->arg_start = bprm->p;
609 bprm->loader -= stack_shift;
610 bprm->exec -= stack_shift;
612 down_write(&mm->mmap_sem);
613 vm_flags = VM_STACK_FLAGS;
616 * Adjust stack execute permissions; explicitly enable for
617 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
618 * (arch default) otherwise.
620 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
622 else if (executable_stack == EXSTACK_DISABLE_X)
623 vm_flags &= ~VM_EXEC;
624 vm_flags |= mm->def_flags;
626 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
632 /* Move stack pages down in memory. */
634 ret = shift_arg_pages(vma, stack_shift);
636 up_write(&mm->mmap_sem);
641 #ifdef CONFIG_STACK_GROWSUP
642 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
644 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
646 ret = expand_stack(vma, stack_base);
651 up_write(&mm->mmap_sem);
654 EXPORT_SYMBOL(setup_arg_pages);
656 #endif /* CONFIG_MMU */
658 struct file *open_exec(const char *name)
664 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
665 FMODE_READ|FMODE_EXEC);
670 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
673 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
676 err = may_open(&nd.path, MAY_EXEC | MAY_OPEN, 0);
680 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
684 fsnotify_open(file->f_path.dentry);
686 err = deny_write_access(file);
695 release_open_intent(&nd);
700 EXPORT_SYMBOL(open_exec);
702 int kernel_read(struct file *file, unsigned long offset,
703 char *addr, unsigned long count)
711 /* The cast to a user pointer is valid due to the set_fs() */
712 result = vfs_read(file, (void __user *)addr, count, &pos);
717 EXPORT_SYMBOL(kernel_read);
719 static int exec_mmap(struct mm_struct *mm)
721 struct task_struct *tsk;
722 struct mm_struct * old_mm, *active_mm;
724 /* Notify parent that we're no longer interested in the old VM */
726 old_mm = current->mm;
727 mm_release(tsk, old_mm);
731 * Make sure that if there is a core dump in progress
732 * for the old mm, we get out and die instead of going
733 * through with the exec. We must hold mmap_sem around
734 * checking core_state and changing tsk->mm.
736 down_read(&old_mm->mmap_sem);
737 if (unlikely(old_mm->core_state)) {
738 up_read(&old_mm->mmap_sem);
743 active_mm = tsk->active_mm;
746 activate_mm(active_mm, mm);
748 arch_pick_mmap_layout(mm);
750 up_read(&old_mm->mmap_sem);
751 BUG_ON(active_mm != old_mm);
752 mm_update_next_owner(old_mm);
761 * This function makes sure the current process has its own signal table,
762 * so that flush_signal_handlers can later reset the handlers without
763 * disturbing other processes. (Other processes might share the signal
764 * table via the CLONE_SIGHAND option to clone().)
766 static int de_thread(struct task_struct *tsk)
768 struct signal_struct *sig = tsk->signal;
769 struct sighand_struct *oldsighand = tsk->sighand;
770 spinlock_t *lock = &oldsighand->siglock;
773 if (thread_group_empty(tsk))
774 goto no_thread_group;
777 * Kill all other threads in the thread group.
780 if (signal_group_exit(sig)) {
782 * Another group action in progress, just
783 * return so that the signal is processed.
785 spin_unlock_irq(lock);
788 sig->group_exit_task = tsk;
789 zap_other_threads(tsk);
791 /* Account for the thread group leader hanging around: */
792 count = thread_group_leader(tsk) ? 1 : 2;
793 sig->notify_count = count;
794 while (atomic_read(&sig->count) > count) {
795 __set_current_state(TASK_UNINTERRUPTIBLE);
796 spin_unlock_irq(lock);
800 spin_unlock_irq(lock);
803 * At this point all other threads have exited, all we have to
804 * do is to wait for the thread group leader to become inactive,
805 * and to assume its PID:
807 if (!thread_group_leader(tsk)) {
808 struct task_struct *leader = tsk->group_leader;
810 sig->notify_count = -1; /* for exit_notify() */
812 write_lock_irq(&tasklist_lock);
813 if (likely(leader->exit_state))
815 __set_current_state(TASK_UNINTERRUPTIBLE);
816 write_unlock_irq(&tasklist_lock);
821 * The only record we have of the real-time age of a
822 * process, regardless of execs it's done, is start_time.
823 * All the past CPU time is accumulated in signal_struct
824 * from sister threads now dead. But in this non-leader
825 * exec, nothing survives from the original leader thread,
826 * whose birth marks the true age of this process now.
827 * When we take on its identity by switching to its PID, we
828 * also take its birthdate (always earlier than our own).
830 tsk->start_time = leader->start_time;
832 BUG_ON(!same_thread_group(leader, tsk));
833 BUG_ON(has_group_leader_pid(tsk));
835 * An exec() starts a new thread group with the
836 * TGID of the previous thread group. Rehash the
837 * two threads with a switched PID, and release
838 * the former thread group leader:
841 /* Become a process group leader with the old leader's pid.
842 * The old leader becomes a thread of the this thread group.
843 * Note: The old leader also uses this pid until release_task
844 * is called. Odd but simple and correct.
846 detach_pid(tsk, PIDTYPE_PID);
847 tsk->pid = leader->pid;
848 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
849 transfer_pid(leader, tsk, PIDTYPE_PGID);
850 transfer_pid(leader, tsk, PIDTYPE_SID);
851 list_replace_rcu(&leader->tasks, &tsk->tasks);
853 tsk->group_leader = tsk;
854 leader->group_leader = tsk;
856 tsk->exit_signal = SIGCHLD;
858 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
859 leader->exit_state = EXIT_DEAD;
860 write_unlock_irq(&tasklist_lock);
862 release_task(leader);
865 sig->group_exit_task = NULL;
866 sig->notify_count = 0;
870 flush_itimer_signals();
872 if (atomic_read(&oldsighand->count) != 1) {
873 struct sighand_struct *newsighand;
875 * This ->sighand is shared with the CLONE_SIGHAND
876 * but not CLONE_THREAD task, switch to the new one.
878 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
882 atomic_set(&newsighand->count, 1);
883 memcpy(newsighand->action, oldsighand->action,
884 sizeof(newsighand->action));
886 write_lock_irq(&tasklist_lock);
887 spin_lock(&oldsighand->siglock);
888 rcu_assign_pointer(tsk->sighand, newsighand);
889 spin_unlock(&oldsighand->siglock);
890 write_unlock_irq(&tasklist_lock);
892 __cleanup_sighand(oldsighand);
895 BUG_ON(!thread_group_leader(tsk));
900 * These functions flushes out all traces of the currently running executable
901 * so that a new one can be started
903 static void flush_old_files(struct files_struct * files)
908 spin_lock(&files->file_lock);
910 unsigned long set, i;
914 fdt = files_fdtable(files);
915 if (i >= fdt->max_fds)
917 set = fdt->close_on_exec->fds_bits[j];
920 fdt->close_on_exec->fds_bits[j] = 0;
921 spin_unlock(&files->file_lock);
922 for ( ; set ; i++,set >>= 1) {
927 spin_lock(&files->file_lock);
930 spin_unlock(&files->file_lock);
933 char *get_task_comm(char *buf, struct task_struct *tsk)
935 /* buf must be at least sizeof(tsk->comm) in size */
937 strncpy(buf, tsk->comm, sizeof(tsk->comm));
942 void set_task_comm(struct task_struct *tsk, char *buf)
945 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
949 int flush_old_exec(struct linux_binprm * bprm)
953 char tcomm[sizeof(current->comm)];
956 * Make sure we have a private signal table and that
957 * we are unassociated from the previous thread group.
959 retval = de_thread(current);
963 set_mm_exe_file(bprm->mm, bprm->file);
966 * Release all of the old mmap stuff
968 retval = exec_mmap(bprm->mm);
972 bprm->mm = NULL; /* We're using it now */
974 /* This is the point of no return */
975 current->sas_ss_sp = current->sas_ss_size = 0;
977 if (current_euid() == current_uid() && current_egid() == current_gid())
978 set_dumpable(current->mm, 1);
980 set_dumpable(current->mm, suid_dumpable);
982 name = bprm->filename;
984 /* Copies the binary name from after last slash */
985 for (i=0; (ch = *(name++)) != '\0';) {
987 i = 0; /* overwrite what we wrote */
989 if (i < (sizeof(tcomm) - 1))
993 set_task_comm(current, tcomm);
995 current->flags &= ~PF_RANDOMIZE;
998 /* Set the new mm task size. We have to do that late because it may
999 * depend on TIF_32BIT which is only updated in flush_thread() on
1000 * some architectures like powerpc
1002 current->mm->task_size = TASK_SIZE;
1004 /* install the new credentials */
1005 if (bprm->cred->uid != current_euid() ||
1006 bprm->cred->gid != current_egid()) {
1007 current->pdeath_signal = 0;
1008 } else if (file_permission(bprm->file, MAY_READ) ||
1009 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1010 set_dumpable(current->mm, suid_dumpable);
1013 current->personality &= ~bprm->per_clear;
1015 /* An exec changes our domain. We are no longer part of the thread
1018 current->self_exec_id++;
1020 flush_signal_handlers(current, 0);
1021 flush_old_files(current->files);
1029 EXPORT_SYMBOL(flush_old_exec);
1032 * install the new credentials for this executable
1034 void install_exec_creds(struct linux_binprm *bprm)
1036 security_bprm_committing_creds(bprm);
1038 commit_creds(bprm->cred);
1041 /* cred_exec_mutex must be held at least to this point to prevent
1042 * ptrace_attach() from altering our determination of the task's
1043 * credentials; any time after this it may be unlocked */
1045 security_bprm_committed_creds(bprm);
1047 EXPORT_SYMBOL(install_exec_creds);
1050 * determine how safe it is to execute the proposed program
1051 * - the caller must hold current->cred_exec_mutex to protect against
1054 int check_unsafe_exec(struct linux_binprm *bprm)
1056 struct task_struct *p = current, *t;
1060 bprm->unsafe = tracehook_unsafe_exec(p);
1063 write_lock(&p->fs->lock);
1065 for (t = next_thread(p); t != p; t = next_thread(t)) {
1071 if (p->fs->users > n_fs) {
1072 bprm->unsafe |= LSM_UNSAFE_SHARE;
1075 if (!p->fs->in_exec) {
1080 write_unlock(&p->fs->lock);
1086 * Fill the binprm structure from the inode.
1087 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1089 * This may be called multiple times for binary chains (scripts for example).
1091 int prepare_binprm(struct linux_binprm *bprm)
1094 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1097 mode = inode->i_mode;
1098 if (bprm->file->f_op == NULL)
1101 /* clear any previous set[ug]id data from a previous binary */
1102 bprm->cred->euid = current_euid();
1103 bprm->cred->egid = current_egid();
1105 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1107 if (mode & S_ISUID) {
1108 bprm->per_clear |= PER_CLEAR_ON_SETID;
1109 bprm->cred->euid = inode->i_uid;
1114 * If setgid is set but no group execute bit then this
1115 * is a candidate for mandatory locking, not a setgid
1118 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1119 bprm->per_clear |= PER_CLEAR_ON_SETID;
1120 bprm->cred->egid = inode->i_gid;
1124 /* fill in binprm security blob */
1125 retval = security_bprm_set_creds(bprm);
1128 bprm->cred_prepared = 1;
1130 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1131 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1134 EXPORT_SYMBOL(prepare_binprm);
1137 * Arguments are '\0' separated strings found at the location bprm->p
1138 * points to; chop off the first by relocating brpm->p to right after
1139 * the first '\0' encountered.
1141 int remove_arg_zero(struct linux_binprm *bprm)
1144 unsigned long offset;
1152 offset = bprm->p & ~PAGE_MASK;
1153 page = get_arg_page(bprm, bprm->p, 0);
1158 kaddr = kmap_atomic(page, KM_USER0);
1160 for (; offset < PAGE_SIZE && kaddr[offset];
1161 offset++, bprm->p++)
1164 kunmap_atomic(kaddr, KM_USER0);
1167 if (offset == PAGE_SIZE)
1168 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1169 } while (offset == PAGE_SIZE);
1178 EXPORT_SYMBOL(remove_arg_zero);
1181 * cycle the list of binary formats handler, until one recognizes the image
1183 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1185 unsigned int depth = bprm->recursion_depth;
1187 struct linux_binfmt *fmt;
1189 retval = security_bprm_check(bprm);
1192 retval = ima_bprm_check(bprm);
1196 /* kernel module loader fixup */
1197 /* so we don't try to load run modprobe in kernel space. */
1200 retval = audit_bprm(bprm);
1205 for (try=0; try<2; try++) {
1206 read_lock(&binfmt_lock);
1207 list_for_each_entry(fmt, &formats, lh) {
1208 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1211 if (!try_module_get(fmt->module))
1213 read_unlock(&binfmt_lock);
1214 retval = fn(bprm, regs);
1216 * Restore the depth counter to its starting value
1217 * in this call, so we don't have to rely on every
1218 * load_binary function to restore it on return.
1220 bprm->recursion_depth = depth;
1223 tracehook_report_exec(fmt, bprm, regs);
1225 allow_write_access(bprm->file);
1229 current->did_exec = 1;
1230 proc_exec_connector(current);
1233 read_lock(&binfmt_lock);
1235 if (retval != -ENOEXEC || bprm->mm == NULL)
1238 read_unlock(&binfmt_lock);
1242 read_unlock(&binfmt_lock);
1243 if (retval != -ENOEXEC || bprm->mm == NULL) {
1245 #ifdef CONFIG_MODULES
1247 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1248 if (printable(bprm->buf[0]) &&
1249 printable(bprm->buf[1]) &&
1250 printable(bprm->buf[2]) &&
1251 printable(bprm->buf[3]))
1252 break; /* -ENOEXEC */
1253 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1260 EXPORT_SYMBOL(search_binary_handler);
1262 void free_bprm(struct linux_binprm *bprm)
1264 free_arg_pages(bprm);
1266 abort_creds(bprm->cred);
1271 * sys_execve() executes a new program.
1273 int do_execve(char * filename,
1274 char __user *__user *argv,
1275 char __user *__user *envp,
1276 struct pt_regs * regs)
1278 struct linux_binprm *bprm;
1280 struct files_struct *displaced;
1284 retval = unshare_files(&displaced);
1289 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1293 retval = mutex_lock_interruptible(¤t->cred_exec_mutex);
1296 current->in_execve = 1;
1299 bprm->cred = prepare_exec_creds();
1303 retval = check_unsafe_exec(bprm);
1306 clear_in_exec = retval;
1308 file = open_exec(filename);
1309 retval = PTR_ERR(file);
1316 bprm->filename = filename;
1317 bprm->interp = filename;
1319 retval = bprm_mm_init(bprm);
1323 bprm->argc = count(argv, MAX_ARG_STRINGS);
1324 if ((retval = bprm->argc) < 0)
1327 bprm->envc = count(envp, MAX_ARG_STRINGS);
1328 if ((retval = bprm->envc) < 0)
1331 retval = prepare_binprm(bprm);
1335 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1339 bprm->exec = bprm->p;
1340 retval = copy_strings(bprm->envc, envp, bprm);
1344 retval = copy_strings(bprm->argc, argv, bprm);
1348 current->flags &= ~PF_KTHREAD;
1349 retval = search_binary_handler(bprm,regs);
1353 /* execve succeeded */
1354 current->fs->in_exec = 0;
1355 current->in_execve = 0;
1356 mutex_unlock(¤t->cred_exec_mutex);
1357 acct_update_integrals(current);
1360 put_files_struct(displaced);
1369 allow_write_access(bprm->file);
1375 current->fs->in_exec = 0;
1378 current->in_execve = 0;
1379 mutex_unlock(¤t->cred_exec_mutex);
1386 reset_files_struct(displaced);
1391 int set_binfmt(struct linux_binfmt *new)
1393 struct linux_binfmt *old = current->binfmt;
1396 if (!try_module_get(new->module))
1399 current->binfmt = new;
1401 module_put(old->module);
1405 EXPORT_SYMBOL(set_binfmt);
1407 /* format_corename will inspect the pattern parameter, and output a
1408 * name into corename, which must have space for at least
1409 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1411 static int format_corename(char *corename, long signr)
1413 const struct cred *cred = current_cred();
1414 const char *pat_ptr = core_pattern;
1415 int ispipe = (*pat_ptr == '|');
1416 char *out_ptr = corename;
1417 char *const out_end = corename + CORENAME_MAX_SIZE;
1419 int pid_in_pattern = 0;
1421 /* Repeat as long as we have more pattern to process and more output
1424 if (*pat_ptr != '%') {
1425 if (out_ptr == out_end)
1427 *out_ptr++ = *pat_ptr++;
1429 switch (*++pat_ptr) {
1432 /* Double percent, output one percent */
1434 if (out_ptr == out_end)
1441 rc = snprintf(out_ptr, out_end - out_ptr,
1442 "%d", task_tgid_vnr(current));
1443 if (rc > out_end - out_ptr)
1449 rc = snprintf(out_ptr, out_end - out_ptr,
1451 if (rc > out_end - out_ptr)
1457 rc = snprintf(out_ptr, out_end - out_ptr,
1459 if (rc > out_end - out_ptr)
1463 /* signal that caused the coredump */
1465 rc = snprintf(out_ptr, out_end - out_ptr,
1467 if (rc > out_end - out_ptr)
1471 /* UNIX time of coredump */
1474 do_gettimeofday(&tv);
1475 rc = snprintf(out_ptr, out_end - out_ptr,
1477 if (rc > out_end - out_ptr)
1484 down_read(&uts_sem);
1485 rc = snprintf(out_ptr, out_end - out_ptr,
1486 "%s", utsname()->nodename);
1488 if (rc > out_end - out_ptr)
1494 rc = snprintf(out_ptr, out_end - out_ptr,
1495 "%s", current->comm);
1496 if (rc > out_end - out_ptr)
1500 /* core limit size */
1502 rc = snprintf(out_ptr, out_end - out_ptr,
1503 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1504 if (rc > out_end - out_ptr)
1514 /* Backward compatibility with core_uses_pid:
1516 * If core_pattern does not include a %p (as is the default)
1517 * and core_uses_pid is set, then .%pid will be appended to
1518 * the filename. Do not do this for piped commands. */
1519 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1520 rc = snprintf(out_ptr, out_end - out_ptr,
1521 ".%d", task_tgid_vnr(current));
1522 if (rc > out_end - out_ptr)
1531 static int zap_process(struct task_struct *start)
1533 struct task_struct *t;
1536 start->signal->flags = SIGNAL_GROUP_EXIT;
1537 start->signal->group_stop_count = 0;
1541 if (t != current && t->mm) {
1542 sigaddset(&t->pending.signal, SIGKILL);
1543 signal_wake_up(t, 1);
1546 } while_each_thread(start, t);
1551 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1552 struct core_state *core_state, int exit_code)
1554 struct task_struct *g, *p;
1555 unsigned long flags;
1558 spin_lock_irq(&tsk->sighand->siglock);
1559 if (!signal_group_exit(tsk->signal)) {
1560 mm->core_state = core_state;
1561 tsk->signal->group_exit_code = exit_code;
1562 nr = zap_process(tsk);
1564 spin_unlock_irq(&tsk->sighand->siglock);
1565 if (unlikely(nr < 0))
1568 if (atomic_read(&mm->mm_users) == nr + 1)
1571 * We should find and kill all tasks which use this mm, and we should
1572 * count them correctly into ->nr_threads. We don't take tasklist
1573 * lock, but this is safe wrt:
1576 * None of sub-threads can fork after zap_process(leader). All
1577 * processes which were created before this point should be
1578 * visible to zap_threads() because copy_process() adds the new
1579 * process to the tail of init_task.tasks list, and lock/unlock
1580 * of ->siglock provides a memory barrier.
1583 * The caller holds mm->mmap_sem. This means that the task which
1584 * uses this mm can't pass exit_mm(), so it can't exit or clear
1588 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1589 * we must see either old or new leader, this does not matter.
1590 * However, it can change p->sighand, so lock_task_sighand(p)
1591 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1594 * Note also that "g" can be the old leader with ->mm == NULL
1595 * and already unhashed and thus removed from ->thread_group.
1596 * This is OK, __unhash_process()->list_del_rcu() does not
1597 * clear the ->next pointer, we will find the new leader via
1601 for_each_process(g) {
1602 if (g == tsk->group_leader)
1604 if (g->flags & PF_KTHREAD)
1609 if (unlikely(p->mm == mm)) {
1610 lock_task_sighand(p, &flags);
1611 nr += zap_process(p);
1612 unlock_task_sighand(p, &flags);
1616 } while_each_thread(g, p);
1620 atomic_set(&core_state->nr_threads, nr);
1624 static int coredump_wait(int exit_code, struct core_state *core_state)
1626 struct task_struct *tsk = current;
1627 struct mm_struct *mm = tsk->mm;
1628 struct completion *vfork_done;
1631 init_completion(&core_state->startup);
1632 core_state->dumper.task = tsk;
1633 core_state->dumper.next = NULL;
1634 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1635 up_write(&mm->mmap_sem);
1637 if (unlikely(core_waiters < 0))
1641 * Make sure nobody is waiting for us to release the VM,
1642 * otherwise we can deadlock when we wait on each other
1644 vfork_done = tsk->vfork_done;
1646 tsk->vfork_done = NULL;
1647 complete(vfork_done);
1651 wait_for_completion(&core_state->startup);
1653 return core_waiters;
1656 static void coredump_finish(struct mm_struct *mm)
1658 struct core_thread *curr, *next;
1659 struct task_struct *task;
1661 next = mm->core_state->dumper.next;
1662 while ((curr = next) != NULL) {
1666 * see exit_mm(), curr->task must not see
1667 * ->task == NULL before we read ->next.
1671 wake_up_process(task);
1674 mm->core_state = NULL;
1678 * set_dumpable converts traditional three-value dumpable to two flags and
1679 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1680 * these bits are not changed atomically. So get_dumpable can observe the
1681 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1682 * return either old dumpable or new one by paying attention to the order of
1683 * modifying the bits.
1685 * dumpable | mm->flags (binary)
1686 * old new | initial interim final
1687 * ---------+-----------------------
1695 * (*) get_dumpable regards interim value of 10 as 11.
1697 void set_dumpable(struct mm_struct *mm, int value)
1701 clear_bit(MMF_DUMPABLE, &mm->flags);
1703 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1706 set_bit(MMF_DUMPABLE, &mm->flags);
1708 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1711 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1713 set_bit(MMF_DUMPABLE, &mm->flags);
1718 int get_dumpable(struct mm_struct *mm)
1722 ret = mm->flags & 0x3;
1723 return (ret >= 2) ? 2 : ret;
1726 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1728 struct core_state core_state;
1729 char corename[CORENAME_MAX_SIZE + 1];
1730 struct mm_struct *mm = current->mm;
1731 struct linux_binfmt * binfmt;
1732 struct inode * inode;
1734 const struct cred *old_cred;
1739 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1740 char **helper_argv = NULL;
1741 int helper_argc = 0;
1744 audit_core_dumps(signr);
1746 binfmt = current->binfmt;
1747 if (!binfmt || !binfmt->core_dump)
1750 cred = prepare_creds();
1756 down_write(&mm->mmap_sem);
1758 * If another thread got here first, or we are not dumpable, bail out.
1760 if (mm->core_state || !get_dumpable(mm)) {
1761 up_write(&mm->mmap_sem);
1767 * We cannot trust fsuid as being the "true" uid of the
1768 * process nor do we know its entire history. We only know it
1769 * was tainted so we dump it as root in mode 2.
1771 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1772 flag = O_EXCL; /* Stop rewrite attacks */
1773 cred->fsuid = 0; /* Dump root private */
1776 retval = coredump_wait(exit_code, &core_state);
1782 old_cred = override_creds(cred);
1785 * Clear any false indication of pending signals that might
1786 * be seen by the filesystem code called to write the core file.
1788 clear_thread_flag(TIF_SIGPENDING);
1791 * lock_kernel() because format_corename() is controlled by sysctl, which
1792 * uses lock_kernel()
1795 ispipe = format_corename(corename, signr);
1798 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1799 * to a pipe. Since we're not writing directly to the filesystem
1800 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1801 * created unless the pipe reader choses to write out the core file
1802 * at which point file size limits and permissions will be imposed
1803 * as it does with any other process
1805 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1809 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1811 printk(KERN_WARNING "%s failed to allocate memory\n",
1815 /* Terminate the string before the first option */
1816 delimit = strchr(corename, ' ');
1819 delimit = strrchr(helper_argv[0], '/');
1823 delimit = helper_argv[0];
1824 if (!strcmp(delimit, current->comm)) {
1825 printk(KERN_NOTICE "Recursive core dump detected, "
1830 core_limit = RLIM_INFINITY;
1832 /* SIGPIPE can happen, but it's just never processed */
1833 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1835 printk(KERN_INFO "Core dump to %s pipe failed\n",
1840 file = filp_open(corename,
1841 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1845 inode = file->f_path.dentry->d_inode;
1846 if (inode->i_nlink > 1)
1847 goto close_fail; /* multiple links - don't dump */
1848 if (!ispipe && d_unhashed(file->f_path.dentry))
1851 /* AK: actually i see no reason to not allow this for named pipes etc.,
1852 but keep the previous behaviour for now. */
1853 if (!ispipe && !S_ISREG(inode->i_mode))
1856 * Dont allow local users get cute and trick others to coredump
1857 * into their pre-created files:
1859 if (inode->i_uid != current_fsuid())
1863 if (!file->f_op->write)
1865 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1868 retval = binfmt->core_dump(signr, regs, file, core_limit);
1871 current->signal->group_exit_code |= 0x80;
1873 filp_close(file, NULL);
1876 argv_free(helper_argv);
1878 revert_creds(old_cred);
1880 coredump_finish(mm);