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/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.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/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 unsigned int core_pipe_limit;
66 int suid_dumpable = 0;
68 /* The maximal length of core_pattern is also specified in sysctl.c */
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
73 int __register_binfmt(struct linux_binfmt * fmt, int insert)
77 write_lock(&binfmt_lock);
78 insert ? list_add(&fmt->lh, &formats) :
79 list_add_tail(&fmt->lh, &formats);
80 write_unlock(&binfmt_lock);
84 EXPORT_SYMBOL(__register_binfmt);
86 void unregister_binfmt(struct linux_binfmt * fmt)
88 write_lock(&binfmt_lock);
90 write_unlock(&binfmt_lock);
93 EXPORT_SYMBOL(unregister_binfmt);
95 static inline void put_binfmt(struct linux_binfmt * fmt)
97 module_put(fmt->module);
101 * Note that a shared library must be both readable and executable due to
104 * Also note that we take the address to load from from the file itself.
106 SYSCALL_DEFINE1(uselib, const char __user *, library)
109 char *tmp = getname(library);
110 int error = PTR_ERR(tmp);
115 file = do_filp_open(AT_FDCWD, tmp,
116 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
117 MAY_READ | MAY_EXEC | MAY_OPEN);
119 error = PTR_ERR(file);
124 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
128 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
135 struct linux_binfmt * fmt;
137 read_lock(&binfmt_lock);
138 list_for_each_entry(fmt, &formats, lh) {
139 if (!fmt->load_shlib)
141 if (!try_module_get(fmt->module))
143 read_unlock(&binfmt_lock);
144 error = fmt->load_shlib(file);
145 read_lock(&binfmt_lock);
147 if (error != -ENOEXEC)
150 read_unlock(&binfmt_lock);
160 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
162 struct mm_struct *mm = current->mm;
163 long diff = (long)(pages - bprm->vma_pages);
168 bprm->vma_pages = pages;
170 #ifdef SPLIT_RSS_COUNTING
171 add_mm_counter(mm, MM_ANONPAGES, diff);
173 spin_lock(&mm->page_table_lock);
174 add_mm_counter(mm, MM_ANONPAGES, diff);
175 spin_unlock(&mm->page_table_lock);
179 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
185 #ifdef CONFIG_STACK_GROWSUP
187 ret = expand_stack_downwards(bprm->vma, pos);
192 ret = get_user_pages(current, bprm->mm, pos,
193 1, write, 1, &page, NULL);
198 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
201 acct_arg_size(bprm, size / PAGE_SIZE);
204 * We've historically supported up to 32 pages (ARG_MAX)
205 * of argument strings even with small stacks
211 * Limit to 1/4-th the stack size for the argv+env strings.
213 * - the remaining binfmt code will not run out of stack space,
214 * - the program will have a reasonable amount of stack left
217 rlim = current->signal->rlim;
218 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
227 static void put_arg_page(struct page *page)
232 static void free_arg_page(struct linux_binprm *bprm, int i)
236 static void free_arg_pages(struct linux_binprm *bprm)
240 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
243 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
246 static int __bprm_mm_init(struct linux_binprm *bprm)
249 struct vm_area_struct *vma = NULL;
250 struct mm_struct *mm = bprm->mm;
252 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
256 down_write(&mm->mmap_sem);
260 * Place the stack at the largest stack address the architecture
261 * supports. Later, we'll move this to an appropriate place. We don't
262 * use STACK_TOP because that can depend on attributes which aren't
265 BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
266 vma->vm_end = STACK_TOP_MAX;
267 vma->vm_start = vma->vm_end - PAGE_SIZE;
268 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
269 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
270 INIT_LIST_HEAD(&vma->anon_vma_chain);
272 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
276 err = insert_vm_struct(mm, vma);
280 mm->stack_vm = mm->total_vm = 1;
281 up_write(&mm->mmap_sem);
282 bprm->p = vma->vm_end - sizeof(void *);
285 up_write(&mm->mmap_sem);
287 kmem_cache_free(vm_area_cachep, vma);
291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
293 return len <= MAX_ARG_STRLEN;
298 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
302 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
307 page = bprm->page[pos / PAGE_SIZE];
308 if (!page && write) {
309 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
312 bprm->page[pos / PAGE_SIZE] = page;
318 static void put_arg_page(struct page *page)
322 static void free_arg_page(struct linux_binprm *bprm, int i)
325 __free_page(bprm->page[i]);
326 bprm->page[i] = NULL;
330 static void free_arg_pages(struct linux_binprm *bprm)
334 for (i = 0; i < MAX_ARG_PAGES; i++)
335 free_arg_page(bprm, i);
338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
343 static int __bprm_mm_init(struct linux_binprm *bprm)
345 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
351 return len <= bprm->p;
354 #endif /* CONFIG_MMU */
357 * Create a new mm_struct and populate it with a temporary stack
358 * vm_area_struct. We don't have enough context at this point to set the stack
359 * flags, permissions, and offset, so we use temporary values. We'll update
360 * them later in setup_arg_pages().
362 int bprm_mm_init(struct linux_binprm *bprm)
365 struct mm_struct *mm = NULL;
367 bprm->mm = mm = mm_alloc();
372 err = init_new_context(current, mm);
376 err = __bprm_mm_init(bprm);
392 * count() counts the number of strings in array ARGV.
394 static int count(const char __user * const __user * argv, int max)
400 const char __user * p;
402 if (get_user(p, argv))
410 if (fatal_signal_pending(current))
411 return -ERESTARTNOHAND;
419 * 'copy_strings()' copies argument/environment strings from the old
420 * processes's memory to the new process's stack. The call to get_user_pages()
421 * ensures the destination page is created and not swapped out.
423 static int copy_strings(int argc, const char __user *const __user *argv,
424 struct linux_binprm *bprm)
426 struct page *kmapped_page = NULL;
428 unsigned long kpos = 0;
432 const char __user *str;
436 if (get_user(str, argv+argc) ||
437 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
442 if (!valid_arg_len(bprm, len)) {
447 /* We're going to work our way backwords. */
453 int offset, bytes_to_copy;
455 if (fatal_signal_pending(current)) {
456 ret = -ERESTARTNOHAND;
461 offset = pos % PAGE_SIZE;
465 bytes_to_copy = offset;
466 if (bytes_to_copy > len)
469 offset -= bytes_to_copy;
470 pos -= bytes_to_copy;
471 str -= bytes_to_copy;
472 len -= bytes_to_copy;
474 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
477 page = get_arg_page(bprm, pos, 1);
484 flush_kernel_dcache_page(kmapped_page);
485 kunmap(kmapped_page);
486 put_arg_page(kmapped_page);
489 kaddr = kmap(kmapped_page);
490 kpos = pos & PAGE_MASK;
491 flush_arg_page(bprm, kpos, kmapped_page);
493 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
502 flush_kernel_dcache_page(kmapped_page);
503 kunmap(kmapped_page);
504 put_arg_page(kmapped_page);
510 * Like copy_strings, but get argv and its values from kernel memory.
512 int copy_strings_kernel(int argc, const char *const *argv,
513 struct linux_binprm *bprm)
516 mm_segment_t oldfs = get_fs();
518 r = copy_strings(argc, (const char __user *const __user *)argv, bprm);
522 EXPORT_SYMBOL(copy_strings_kernel);
527 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
528 * the binfmt code determines where the new stack should reside, we shift it to
529 * its final location. The process proceeds as follows:
531 * 1) Use shift to calculate the new vma endpoints.
532 * 2) Extend vma to cover both the old and new ranges. This ensures the
533 * arguments passed to subsequent functions are consistent.
534 * 3) Move vma's page tables to the new range.
535 * 4) Free up any cleared pgd range.
536 * 5) Shrink the vma to cover only the new range.
538 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
540 struct mm_struct *mm = vma->vm_mm;
541 unsigned long old_start = vma->vm_start;
542 unsigned long old_end = vma->vm_end;
543 unsigned long length = old_end - old_start;
544 unsigned long new_start = old_start - shift;
545 unsigned long new_end = old_end - shift;
546 struct mmu_gather *tlb;
548 BUG_ON(new_start > new_end);
551 * ensure there are no vmas between where we want to go
554 if (vma != find_vma(mm, new_start))
558 * cover the whole range: [new_start, old_end)
560 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
564 * move the page tables downwards, on failure we rely on
565 * process cleanup to remove whatever mess we made.
567 if (length != move_page_tables(vma, old_start,
568 vma, new_start, length))
572 tlb = tlb_gather_mmu(mm, 0);
573 if (new_end > old_start) {
575 * when the old and new regions overlap clear from new_end.
577 free_pgd_range(tlb, new_end, old_end, new_end,
578 vma->vm_next ? vma->vm_next->vm_start : 0);
581 * otherwise, clean from old_start; this is done to not touch
582 * the address space in [new_end, old_start) some architectures
583 * have constraints on va-space that make this illegal (IA64) -
584 * for the others its just a little faster.
586 free_pgd_range(tlb, old_start, old_end, new_end,
587 vma->vm_next ? vma->vm_next->vm_start : 0);
589 tlb_finish_mmu(tlb, new_end, old_end);
592 * Shrink the vma to just the new range. Always succeeds.
594 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
600 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
601 * the stack is optionally relocated, and some extra space is added.
603 int setup_arg_pages(struct linux_binprm *bprm,
604 unsigned long stack_top,
605 int executable_stack)
608 unsigned long stack_shift;
609 struct mm_struct *mm = current->mm;
610 struct vm_area_struct *vma = bprm->vma;
611 struct vm_area_struct *prev = NULL;
612 unsigned long vm_flags;
613 unsigned long stack_base;
614 unsigned long stack_size;
615 unsigned long stack_expand;
616 unsigned long rlim_stack;
618 #ifdef CONFIG_STACK_GROWSUP
619 /* Limit stack size to 1GB */
620 stack_base = rlimit_max(RLIMIT_STACK);
621 if (stack_base > (1 << 30))
622 stack_base = 1 << 30;
624 /* Make sure we didn't let the argument array grow too large. */
625 if (vma->vm_end - vma->vm_start > stack_base)
628 stack_base = PAGE_ALIGN(stack_top - stack_base);
630 stack_shift = vma->vm_start - stack_base;
631 mm->arg_start = bprm->p - stack_shift;
632 bprm->p = vma->vm_end - stack_shift;
634 stack_top = arch_align_stack(stack_top);
635 stack_top = PAGE_ALIGN(stack_top);
637 if (unlikely(stack_top < mmap_min_addr) ||
638 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
641 stack_shift = vma->vm_end - stack_top;
643 bprm->p -= stack_shift;
644 mm->arg_start = bprm->p;
648 bprm->loader -= stack_shift;
649 bprm->exec -= stack_shift;
651 down_write(&mm->mmap_sem);
652 vm_flags = VM_STACK_FLAGS;
655 * Adjust stack execute permissions; explicitly enable for
656 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
657 * (arch default) otherwise.
659 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
661 else if (executable_stack == EXSTACK_DISABLE_X)
662 vm_flags &= ~VM_EXEC;
663 vm_flags |= mm->def_flags;
664 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
666 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
672 /* Move stack pages down in memory. */
674 ret = shift_arg_pages(vma, stack_shift);
679 /* mprotect_fixup is overkill to remove the temporary stack flags */
680 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
682 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
683 stack_size = vma->vm_end - vma->vm_start;
685 * Align this down to a page boundary as expand_stack
688 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
689 #ifdef CONFIG_STACK_GROWSUP
690 if (stack_size + stack_expand > rlim_stack)
691 stack_base = vma->vm_start + rlim_stack;
693 stack_base = vma->vm_end + stack_expand;
695 if (stack_size + stack_expand > rlim_stack)
696 stack_base = vma->vm_end - rlim_stack;
698 stack_base = vma->vm_start - stack_expand;
700 current->mm->start_stack = bprm->p;
701 ret = expand_stack(vma, stack_base);
706 up_write(&mm->mmap_sem);
709 EXPORT_SYMBOL(setup_arg_pages);
711 #endif /* CONFIG_MMU */
713 struct file *open_exec(const char *name)
718 file = do_filp_open(AT_FDCWD, name,
719 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
720 MAY_EXEC | MAY_OPEN);
725 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
728 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
733 err = deny_write_access(file);
744 EXPORT_SYMBOL(open_exec);
746 int kernel_read(struct file *file, loff_t offset,
747 char *addr, unsigned long count)
755 /* The cast to a user pointer is valid due to the set_fs() */
756 result = vfs_read(file, (void __user *)addr, count, &pos);
761 EXPORT_SYMBOL(kernel_read);
763 static int exec_mmap(struct mm_struct *mm)
765 struct task_struct *tsk;
766 struct mm_struct * old_mm, *active_mm;
768 /* Notify parent that we're no longer interested in the old VM */
770 old_mm = current->mm;
771 sync_mm_rss(tsk, old_mm);
772 mm_release(tsk, old_mm);
776 * Make sure that if there is a core dump in progress
777 * for the old mm, we get out and die instead of going
778 * through with the exec. We must hold mmap_sem around
779 * checking core_state and changing tsk->mm.
781 down_read(&old_mm->mmap_sem);
782 if (unlikely(old_mm->core_state)) {
783 up_read(&old_mm->mmap_sem);
788 active_mm = tsk->active_mm;
791 activate_mm(active_mm, mm);
793 arch_pick_mmap_layout(mm);
795 up_read(&old_mm->mmap_sem);
796 BUG_ON(active_mm != old_mm);
797 mm_update_next_owner(old_mm);
806 * This function makes sure the current process has its own signal table,
807 * so that flush_signal_handlers can later reset the handlers without
808 * disturbing other processes. (Other processes might share the signal
809 * table via the CLONE_SIGHAND option to clone().)
811 static int de_thread(struct task_struct *tsk)
813 struct signal_struct *sig = tsk->signal;
814 struct sighand_struct *oldsighand = tsk->sighand;
815 spinlock_t *lock = &oldsighand->siglock;
817 if (thread_group_empty(tsk))
818 goto no_thread_group;
821 * Kill all other threads in the thread group.
824 if (signal_group_exit(sig)) {
826 * Another group action in progress, just
827 * return so that the signal is processed.
829 spin_unlock_irq(lock);
833 sig->group_exit_task = tsk;
834 sig->notify_count = zap_other_threads(tsk);
835 if (!thread_group_leader(tsk))
838 while (sig->notify_count) {
839 __set_current_state(TASK_UNINTERRUPTIBLE);
840 spin_unlock_irq(lock);
844 spin_unlock_irq(lock);
847 * At this point all other threads have exited, all we have to
848 * do is to wait for the thread group leader to become inactive,
849 * and to assume its PID:
851 if (!thread_group_leader(tsk)) {
852 struct task_struct *leader = tsk->group_leader;
854 sig->notify_count = -1; /* for exit_notify() */
856 write_lock_irq(&tasklist_lock);
857 if (likely(leader->exit_state))
859 __set_current_state(TASK_UNINTERRUPTIBLE);
860 write_unlock_irq(&tasklist_lock);
865 * The only record we have of the real-time age of a
866 * process, regardless of execs it's done, is start_time.
867 * All the past CPU time is accumulated in signal_struct
868 * from sister threads now dead. But in this non-leader
869 * exec, nothing survives from the original leader thread,
870 * whose birth marks the true age of this process now.
871 * When we take on its identity by switching to its PID, we
872 * also take its birthdate (always earlier than our own).
874 tsk->start_time = leader->start_time;
876 BUG_ON(!same_thread_group(leader, tsk));
877 BUG_ON(has_group_leader_pid(tsk));
879 * An exec() starts a new thread group with the
880 * TGID of the previous thread group. Rehash the
881 * two threads with a switched PID, and release
882 * the former thread group leader:
885 /* Become a process group leader with the old leader's pid.
886 * The old leader becomes a thread of the this thread group.
887 * Note: The old leader also uses this pid until release_task
888 * is called. Odd but simple and correct.
890 detach_pid(tsk, PIDTYPE_PID);
891 tsk->pid = leader->pid;
892 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
893 transfer_pid(leader, tsk, PIDTYPE_PGID);
894 transfer_pid(leader, tsk, PIDTYPE_SID);
896 list_replace_rcu(&leader->tasks, &tsk->tasks);
897 list_replace_init(&leader->sibling, &tsk->sibling);
899 tsk->group_leader = tsk;
900 leader->group_leader = tsk;
902 tsk->exit_signal = SIGCHLD;
904 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
905 leader->exit_state = EXIT_DEAD;
906 write_unlock_irq(&tasklist_lock);
908 release_task(leader);
911 sig->group_exit_task = NULL;
912 sig->notify_count = 0;
916 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
919 flush_itimer_signals();
921 if (atomic_read(&oldsighand->count) != 1) {
922 struct sighand_struct *newsighand;
924 * This ->sighand is shared with the CLONE_SIGHAND
925 * but not CLONE_THREAD task, switch to the new one.
927 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
931 atomic_set(&newsighand->count, 1);
932 memcpy(newsighand->action, oldsighand->action,
933 sizeof(newsighand->action));
935 write_lock_irq(&tasklist_lock);
936 spin_lock(&oldsighand->siglock);
937 rcu_assign_pointer(tsk->sighand, newsighand);
938 spin_unlock(&oldsighand->siglock);
939 write_unlock_irq(&tasklist_lock);
941 __cleanup_sighand(oldsighand);
944 BUG_ON(!thread_group_leader(tsk));
949 * These functions flushes out all traces of the currently running executable
950 * so that a new one can be started
952 static void flush_old_files(struct files_struct * files)
957 spin_lock(&files->file_lock);
959 unsigned long set, i;
963 fdt = files_fdtable(files);
964 if (i >= fdt->max_fds)
966 set = fdt->close_on_exec->fds_bits[j];
969 fdt->close_on_exec->fds_bits[j] = 0;
970 spin_unlock(&files->file_lock);
971 for ( ; set ; i++,set >>= 1) {
976 spin_lock(&files->file_lock);
979 spin_unlock(&files->file_lock);
982 char *get_task_comm(char *buf, struct task_struct *tsk)
984 /* buf must be at least sizeof(tsk->comm) in size */
986 strncpy(buf, tsk->comm, sizeof(tsk->comm));
991 void set_task_comm(struct task_struct *tsk, char *buf)
996 * Threads may access current->comm without holding
997 * the task lock, so write the string carefully.
998 * Readers without a lock may see incomplete new
999 * names but are safe from non-terminating string reads.
1001 memset(tsk->comm, 0, TASK_COMM_LEN);
1003 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1005 perf_event_comm(tsk);
1008 int flush_old_exec(struct linux_binprm * bprm)
1013 * Make sure we have a private signal table and that
1014 * we are unassociated from the previous thread group.
1016 retval = de_thread(current);
1020 set_mm_exe_file(bprm->mm, bprm->file);
1023 * Release all of the old mmap stuff
1025 acct_arg_size(bprm, 0);
1026 retval = exec_mmap(bprm->mm);
1030 bprm->mm = NULL; /* We're using it now */
1032 current->flags &= ~PF_RANDOMIZE;
1034 current->personality &= ~bprm->per_clear;
1041 EXPORT_SYMBOL(flush_old_exec);
1043 void setup_new_exec(struct linux_binprm * bprm)
1047 char tcomm[sizeof(current->comm)];
1049 arch_pick_mmap_layout(current->mm);
1051 /* This is the point of no return */
1052 current->sas_ss_sp = current->sas_ss_size = 0;
1054 if (current_euid() == current_uid() && current_egid() == current_gid())
1055 set_dumpable(current->mm, 1);
1057 set_dumpable(current->mm, suid_dumpable);
1059 name = bprm->filename;
1061 /* Copies the binary name from after last slash */
1062 for (i=0; (ch = *(name++)) != '\0';) {
1064 i = 0; /* overwrite what we wrote */
1066 if (i < (sizeof(tcomm) - 1))
1070 set_task_comm(current, tcomm);
1072 /* Set the new mm task size. We have to do that late because it may
1073 * depend on TIF_32BIT which is only updated in flush_thread() on
1074 * some architectures like powerpc
1076 current->mm->task_size = TASK_SIZE;
1078 /* install the new credentials */
1079 if (bprm->cred->uid != current_euid() ||
1080 bprm->cred->gid != current_egid()) {
1081 current->pdeath_signal = 0;
1082 } else if (file_permission(bprm->file, MAY_READ) ||
1083 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1084 set_dumpable(current->mm, suid_dumpable);
1088 * Flush performance counters when crossing a
1091 if (!get_dumpable(current->mm))
1092 perf_event_exit_task(current);
1094 /* An exec changes our domain. We are no longer part of the thread
1097 current->self_exec_id++;
1099 flush_signal_handlers(current, 0);
1100 flush_old_files(current->files);
1102 EXPORT_SYMBOL(setup_new_exec);
1105 * Prepare credentials and lock ->cred_guard_mutex.
1106 * install_exec_creds() commits the new creds and drops the lock.
1107 * Or, if exec fails before, free_bprm() should release ->cred and
1110 int prepare_bprm_creds(struct linux_binprm *bprm)
1112 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1113 return -ERESTARTNOINTR;
1115 bprm->cred = prepare_exec_creds();
1116 if (likely(bprm->cred))
1119 mutex_unlock(¤t->cred_guard_mutex);
1123 void free_bprm(struct linux_binprm *bprm)
1125 free_arg_pages(bprm);
1127 mutex_unlock(¤t->cred_guard_mutex);
1128 abort_creds(bprm->cred);
1134 * install the new credentials for this executable
1136 void install_exec_creds(struct linux_binprm *bprm)
1138 security_bprm_committing_creds(bprm);
1140 commit_creds(bprm->cred);
1143 * cred_guard_mutex must be held at least to this point to prevent
1144 * ptrace_attach() from altering our determination of the task's
1145 * credentials; any time after this it may be unlocked.
1147 security_bprm_committed_creds(bprm);
1148 mutex_unlock(¤t->cred_guard_mutex);
1150 EXPORT_SYMBOL(install_exec_creds);
1153 * determine how safe it is to execute the proposed program
1154 * - the caller must hold current->cred_guard_mutex to protect against
1157 int check_unsafe_exec(struct linux_binprm *bprm)
1159 struct task_struct *p = current, *t;
1163 bprm->unsafe = tracehook_unsafe_exec(p);
1166 spin_lock(&p->fs->lock);
1168 for (t = next_thread(p); t != p; t = next_thread(t)) {
1174 if (p->fs->users > n_fs) {
1175 bprm->unsafe |= LSM_UNSAFE_SHARE;
1178 if (!p->fs->in_exec) {
1183 spin_unlock(&p->fs->lock);
1189 * Fill the binprm structure from the inode.
1190 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1192 * This may be called multiple times for binary chains (scripts for example).
1194 int prepare_binprm(struct linux_binprm *bprm)
1197 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1200 mode = inode->i_mode;
1201 if (bprm->file->f_op == NULL)
1204 /* clear any previous set[ug]id data from a previous binary */
1205 bprm->cred->euid = current_euid();
1206 bprm->cred->egid = current_egid();
1208 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1210 if (mode & S_ISUID) {
1211 bprm->per_clear |= PER_CLEAR_ON_SETID;
1212 bprm->cred->euid = inode->i_uid;
1217 * If setgid is set but no group execute bit then this
1218 * is a candidate for mandatory locking, not a setgid
1221 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1222 bprm->per_clear |= PER_CLEAR_ON_SETID;
1223 bprm->cred->egid = inode->i_gid;
1227 /* fill in binprm security blob */
1228 retval = security_bprm_set_creds(bprm);
1231 bprm->cred_prepared = 1;
1233 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1234 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1237 EXPORT_SYMBOL(prepare_binprm);
1240 * Arguments are '\0' separated strings found at the location bprm->p
1241 * points to; chop off the first by relocating brpm->p to right after
1242 * the first '\0' encountered.
1244 int remove_arg_zero(struct linux_binprm *bprm)
1247 unsigned long offset;
1255 offset = bprm->p & ~PAGE_MASK;
1256 page = get_arg_page(bprm, bprm->p, 0);
1261 kaddr = kmap_atomic(page, KM_USER0);
1263 for (; offset < PAGE_SIZE && kaddr[offset];
1264 offset++, bprm->p++)
1267 kunmap_atomic(kaddr, KM_USER0);
1270 if (offset == PAGE_SIZE)
1271 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1272 } while (offset == PAGE_SIZE);
1281 EXPORT_SYMBOL(remove_arg_zero);
1284 * cycle the list of binary formats handler, until one recognizes the image
1286 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1288 unsigned int depth = bprm->recursion_depth;
1290 struct linux_binfmt *fmt;
1292 retval = security_bprm_check(bprm);
1296 /* kernel module loader fixup */
1297 /* so we don't try to load run modprobe in kernel space. */
1300 retval = audit_bprm(bprm);
1305 for (try=0; try<2; try++) {
1306 read_lock(&binfmt_lock);
1307 list_for_each_entry(fmt, &formats, lh) {
1308 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1311 if (!try_module_get(fmt->module))
1313 read_unlock(&binfmt_lock);
1314 retval = fn(bprm, regs);
1316 * Restore the depth counter to its starting value
1317 * in this call, so we don't have to rely on every
1318 * load_binary function to restore it on return.
1320 bprm->recursion_depth = depth;
1323 tracehook_report_exec(fmt, bprm, regs);
1325 allow_write_access(bprm->file);
1329 current->did_exec = 1;
1330 proc_exec_connector(current);
1333 read_lock(&binfmt_lock);
1335 if (retval != -ENOEXEC || bprm->mm == NULL)
1338 read_unlock(&binfmt_lock);
1342 read_unlock(&binfmt_lock);
1343 if (retval != -ENOEXEC || bprm->mm == NULL) {
1345 #ifdef CONFIG_MODULES
1347 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1348 if (printable(bprm->buf[0]) &&
1349 printable(bprm->buf[1]) &&
1350 printable(bprm->buf[2]) &&
1351 printable(bprm->buf[3]))
1352 break; /* -ENOEXEC */
1353 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1360 EXPORT_SYMBOL(search_binary_handler);
1363 * sys_execve() executes a new program.
1365 int do_execve(const char * filename,
1366 const char __user *const __user *argv,
1367 const char __user *const __user *envp,
1368 struct pt_regs * regs)
1370 struct linux_binprm *bprm;
1372 struct files_struct *displaced;
1376 retval = unshare_files(&displaced);
1381 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1385 retval = prepare_bprm_creds(bprm);
1389 retval = check_unsafe_exec(bprm);
1392 clear_in_exec = retval;
1393 current->in_execve = 1;
1395 file = open_exec(filename);
1396 retval = PTR_ERR(file);
1403 bprm->filename = filename;
1404 bprm->interp = filename;
1406 retval = bprm_mm_init(bprm);
1410 bprm->argc = count(argv, MAX_ARG_STRINGS);
1411 if ((retval = bprm->argc) < 0)
1414 bprm->envc = count(envp, MAX_ARG_STRINGS);
1415 if ((retval = bprm->envc) < 0)
1418 retval = prepare_binprm(bprm);
1422 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1426 bprm->exec = bprm->p;
1427 retval = copy_strings(bprm->envc, envp, bprm);
1431 retval = copy_strings(bprm->argc, argv, bprm);
1435 current->flags &= ~PF_KTHREAD;
1436 retval = search_binary_handler(bprm,regs);
1440 /* execve succeeded */
1441 current->fs->in_exec = 0;
1442 current->in_execve = 0;
1443 acct_update_integrals(current);
1446 put_files_struct(displaced);
1451 acct_arg_size(bprm, 0);
1457 allow_write_access(bprm->file);
1463 current->fs->in_exec = 0;
1464 current->in_execve = 0;
1471 reset_files_struct(displaced);
1476 void set_binfmt(struct linux_binfmt *new)
1478 struct mm_struct *mm = current->mm;
1481 module_put(mm->binfmt->module);
1485 __module_get(new->module);
1488 EXPORT_SYMBOL(set_binfmt);
1490 /* format_corename will inspect the pattern parameter, and output a
1491 * name into corename, which must have space for at least
1492 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1494 static int format_corename(char *corename, long signr)
1496 const struct cred *cred = current_cred();
1497 const char *pat_ptr = core_pattern;
1498 int ispipe = (*pat_ptr == '|');
1499 char *out_ptr = corename;
1500 char *const out_end = corename + CORENAME_MAX_SIZE;
1502 int pid_in_pattern = 0;
1504 /* Repeat as long as we have more pattern to process and more output
1507 if (*pat_ptr != '%') {
1508 if (out_ptr == out_end)
1510 *out_ptr++ = *pat_ptr++;
1512 switch (*++pat_ptr) {
1515 /* Double percent, output one percent */
1517 if (out_ptr == out_end)
1524 rc = snprintf(out_ptr, out_end - out_ptr,
1525 "%d", task_tgid_vnr(current));
1526 if (rc > out_end - out_ptr)
1532 rc = snprintf(out_ptr, out_end - out_ptr,
1534 if (rc > out_end - out_ptr)
1540 rc = snprintf(out_ptr, out_end - out_ptr,
1542 if (rc > out_end - out_ptr)
1546 /* signal that caused the coredump */
1548 rc = snprintf(out_ptr, out_end - out_ptr,
1550 if (rc > out_end - out_ptr)
1554 /* UNIX time of coredump */
1557 do_gettimeofday(&tv);
1558 rc = snprintf(out_ptr, out_end - out_ptr,
1560 if (rc > out_end - out_ptr)
1567 down_read(&uts_sem);
1568 rc = snprintf(out_ptr, out_end - out_ptr,
1569 "%s", utsname()->nodename);
1571 if (rc > out_end - out_ptr)
1577 rc = snprintf(out_ptr, out_end - out_ptr,
1578 "%s", current->comm);
1579 if (rc > out_end - out_ptr)
1583 /* core limit size */
1585 rc = snprintf(out_ptr, out_end - out_ptr,
1586 "%lu", rlimit(RLIMIT_CORE));
1587 if (rc > out_end - out_ptr)
1597 /* Backward compatibility with core_uses_pid:
1599 * If core_pattern does not include a %p (as is the default)
1600 * and core_uses_pid is set, then .%pid will be appended to
1601 * the filename. Do not do this for piped commands. */
1602 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1603 rc = snprintf(out_ptr, out_end - out_ptr,
1604 ".%d", task_tgid_vnr(current));
1605 if (rc > out_end - out_ptr)
1614 static int zap_process(struct task_struct *start, int exit_code)
1616 struct task_struct *t;
1619 start->signal->flags = SIGNAL_GROUP_EXIT;
1620 start->signal->group_exit_code = exit_code;
1621 start->signal->group_stop_count = 0;
1625 if (t != current && t->mm) {
1626 sigaddset(&t->pending.signal, SIGKILL);
1627 signal_wake_up(t, 1);
1630 } while_each_thread(start, t);
1635 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1636 struct core_state *core_state, int exit_code)
1638 struct task_struct *g, *p;
1639 unsigned long flags;
1642 spin_lock_irq(&tsk->sighand->siglock);
1643 if (!signal_group_exit(tsk->signal)) {
1644 mm->core_state = core_state;
1645 nr = zap_process(tsk, exit_code);
1647 spin_unlock_irq(&tsk->sighand->siglock);
1648 if (unlikely(nr < 0))
1651 if (atomic_read(&mm->mm_users) == nr + 1)
1654 * We should find and kill all tasks which use this mm, and we should
1655 * count them correctly into ->nr_threads. We don't take tasklist
1656 * lock, but this is safe wrt:
1659 * None of sub-threads can fork after zap_process(leader). All
1660 * processes which were created before this point should be
1661 * visible to zap_threads() because copy_process() adds the new
1662 * process to the tail of init_task.tasks list, and lock/unlock
1663 * of ->siglock provides a memory barrier.
1666 * The caller holds mm->mmap_sem. This means that the task which
1667 * uses this mm can't pass exit_mm(), so it can't exit or clear
1671 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1672 * we must see either old or new leader, this does not matter.
1673 * However, it can change p->sighand, so lock_task_sighand(p)
1674 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1677 * Note also that "g" can be the old leader with ->mm == NULL
1678 * and already unhashed and thus removed from ->thread_group.
1679 * This is OK, __unhash_process()->list_del_rcu() does not
1680 * clear the ->next pointer, we will find the new leader via
1684 for_each_process(g) {
1685 if (g == tsk->group_leader)
1687 if (g->flags & PF_KTHREAD)
1692 if (unlikely(p->mm == mm)) {
1693 lock_task_sighand(p, &flags);
1694 nr += zap_process(p, exit_code);
1695 unlock_task_sighand(p, &flags);
1699 } while_each_thread(g, p);
1703 atomic_set(&core_state->nr_threads, nr);
1707 static int coredump_wait(int exit_code, struct core_state *core_state)
1709 struct task_struct *tsk = current;
1710 struct mm_struct *mm = tsk->mm;
1711 struct completion *vfork_done;
1712 int core_waiters = -EBUSY;
1714 init_completion(&core_state->startup);
1715 core_state->dumper.task = tsk;
1716 core_state->dumper.next = NULL;
1718 down_write(&mm->mmap_sem);
1719 if (!mm->core_state)
1720 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1721 up_write(&mm->mmap_sem);
1723 if (unlikely(core_waiters < 0))
1727 * Make sure nobody is waiting for us to release the VM,
1728 * otherwise we can deadlock when we wait on each other
1730 vfork_done = tsk->vfork_done;
1732 tsk->vfork_done = NULL;
1733 complete(vfork_done);
1737 wait_for_completion(&core_state->startup);
1739 return core_waiters;
1742 static void coredump_finish(struct mm_struct *mm)
1744 struct core_thread *curr, *next;
1745 struct task_struct *task;
1747 next = mm->core_state->dumper.next;
1748 while ((curr = next) != NULL) {
1752 * see exit_mm(), curr->task must not see
1753 * ->task == NULL before we read ->next.
1757 wake_up_process(task);
1760 mm->core_state = NULL;
1764 * set_dumpable converts traditional three-value dumpable to two flags and
1765 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1766 * these bits are not changed atomically. So get_dumpable can observe the
1767 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1768 * return either old dumpable or new one by paying attention to the order of
1769 * modifying the bits.
1771 * dumpable | mm->flags (binary)
1772 * old new | initial interim final
1773 * ---------+-----------------------
1781 * (*) get_dumpable regards interim value of 10 as 11.
1783 void set_dumpable(struct mm_struct *mm, int value)
1787 clear_bit(MMF_DUMPABLE, &mm->flags);
1789 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1792 set_bit(MMF_DUMPABLE, &mm->flags);
1794 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1797 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1799 set_bit(MMF_DUMPABLE, &mm->flags);
1804 static int __get_dumpable(unsigned long mm_flags)
1808 ret = mm_flags & MMF_DUMPABLE_MASK;
1809 return (ret >= 2) ? 2 : ret;
1812 int get_dumpable(struct mm_struct *mm)
1814 return __get_dumpable(mm->flags);
1817 static void wait_for_dump_helpers(struct file *file)
1819 struct pipe_inode_info *pipe;
1821 pipe = file->f_path.dentry->d_inode->i_pipe;
1827 while ((pipe->readers > 1) && (!signal_pending(current))) {
1828 wake_up_interruptible_sync(&pipe->wait);
1829 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1842 * helper function to customize the process used
1843 * to collect the core in userspace. Specifically
1844 * it sets up a pipe and installs it as fd 0 (stdin)
1845 * for the process. Returns 0 on success, or
1846 * PTR_ERR on failure.
1847 * Note that it also sets the core limit to 1. This
1848 * is a special value that we use to trap recursive
1851 static int umh_pipe_setup(struct subprocess_info *info)
1853 struct file *rp, *wp;
1854 struct fdtable *fdt;
1855 struct coredump_params *cp = (struct coredump_params *)info->data;
1856 struct files_struct *cf = current->files;
1858 wp = create_write_pipe(0);
1862 rp = create_read_pipe(wp, 0);
1864 free_write_pipe(wp);
1872 spin_lock(&cf->file_lock);
1873 fdt = files_fdtable(cf);
1874 FD_SET(0, fdt->open_fds);
1875 FD_CLR(0, fdt->close_on_exec);
1876 spin_unlock(&cf->file_lock);
1878 /* and disallow core files too */
1879 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1884 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1886 struct core_state core_state;
1887 char corename[CORENAME_MAX_SIZE + 1];
1888 struct mm_struct *mm = current->mm;
1889 struct linux_binfmt * binfmt;
1890 const struct cred *old_cred;
1895 static atomic_t core_dump_count = ATOMIC_INIT(0);
1896 struct coredump_params cprm = {
1899 .limit = rlimit(RLIMIT_CORE),
1901 * We must use the same mm->flags while dumping core to avoid
1902 * inconsistency of bit flags, since this flag is not protected
1905 .mm_flags = mm->flags,
1908 audit_core_dumps(signr);
1910 binfmt = mm->binfmt;
1911 if (!binfmt || !binfmt->core_dump)
1913 if (!__get_dumpable(cprm.mm_flags))
1916 cred = prepare_creds();
1920 * We cannot trust fsuid as being the "true" uid of the
1921 * process nor do we know its entire history. We only know it
1922 * was tainted so we dump it as root in mode 2.
1924 if (__get_dumpable(cprm.mm_flags) == 2) {
1925 /* Setuid core dump mode */
1926 flag = O_EXCL; /* Stop rewrite attacks */
1927 cred->fsuid = 0; /* Dump root private */
1930 retval = coredump_wait(exit_code, &core_state);
1934 old_cred = override_creds(cred);
1937 * Clear any false indication of pending signals that might
1938 * be seen by the filesystem code called to write the core file.
1940 clear_thread_flag(TIF_SIGPENDING);
1942 ispipe = format_corename(corename, signr);
1948 if (cprm.limit == 1) {
1950 * Normally core limits are irrelevant to pipes, since
1951 * we're not writing to the file system, but we use
1952 * cprm.limit of 1 here as a speacial value. Any
1953 * non-1 limit gets set to RLIM_INFINITY below, but
1954 * a limit of 0 skips the dump. This is a consistent
1955 * way to catch recursive crashes. We can still crash
1956 * if the core_pattern binary sets RLIM_CORE = !1
1957 * but it runs as root, and can do lots of stupid things
1958 * Note that we use task_tgid_vnr here to grab the pid
1959 * of the process group leader. That way we get the
1960 * right pid if a thread in a multi-threaded
1961 * core_pattern process dies.
1964 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1965 task_tgid_vnr(current), current->comm);
1966 printk(KERN_WARNING "Aborting core\n");
1969 cprm.limit = RLIM_INFINITY;
1971 dump_count = atomic_inc_return(&core_dump_count);
1972 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1973 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1974 task_tgid_vnr(current), current->comm);
1975 printk(KERN_WARNING "Skipping core dump\n");
1976 goto fail_dropcount;
1979 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1981 printk(KERN_WARNING "%s failed to allocate memory\n",
1983 goto fail_dropcount;
1986 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1987 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1989 argv_free(helper_argv);
1991 printk(KERN_INFO "Core dump to %s pipe failed\n",
1996 struct inode *inode;
1998 if (cprm.limit < binfmt->min_coredump)
2001 cprm.file = filp_open(corename,
2002 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2004 if (IS_ERR(cprm.file))
2007 inode = cprm.file->f_path.dentry->d_inode;
2008 if (inode->i_nlink > 1)
2010 if (d_unhashed(cprm.file->f_path.dentry))
2013 * AK: actually i see no reason to not allow this for named
2014 * pipes etc, but keep the previous behaviour for now.
2016 if (!S_ISREG(inode->i_mode))
2019 * Dont allow local users get cute and trick others to coredump
2020 * into their pre-created files.
2022 if (inode->i_uid != current_fsuid())
2024 if (!cprm.file->f_op || !cprm.file->f_op->write)
2026 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2030 retval = binfmt->core_dump(&cprm);
2032 current->signal->group_exit_code |= 0x80;
2034 if (ispipe && core_pipe_limit)
2035 wait_for_dump_helpers(cprm.file);
2038 filp_close(cprm.file, NULL);
2041 atomic_dec(&core_dump_count);
2043 coredump_finish(mm);
2044 revert_creds(old_cred);
2052 * Core dumping helper functions. These are the only things you should
2053 * do on a core-file: use only these functions to write out all the
2056 int dump_write(struct file *file, const void *addr, int nr)
2058 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2060 EXPORT_SYMBOL(dump_write);
2062 int dump_seek(struct file *file, loff_t off)
2066 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2067 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2070 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2075 unsigned long n = off;
2079 if (!dump_write(file, buf, n)) {
2085 free_page((unsigned long)buf);
2089 EXPORT_SYMBOL(dump_seek);