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/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/ima.h>
50 #include <linux/syscalls.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54 #include <linux/tracehook.h>
55 #include <linux/kmod.h>
56 #include <linux/fsnotify.h>
57 #include <linux/fs_struct.h>
58 #include <linux/pipe_fs_i.h>
60 #include <asm/uaccess.h>
61 #include <asm/mmu_context.h>
66 char core_pattern[CORENAME_MAX_SIZE] = "core";
67 unsigned int core_pipe_limit;
68 int suid_dumpable = 0;
70 /* The maximal length of core_pattern is also specified in sysctl.c */
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
75 int __register_binfmt(struct linux_binfmt * fmt, int insert)
79 write_lock(&binfmt_lock);
80 insert ? list_add(&fmt->lh, &formats) :
81 list_add_tail(&fmt->lh, &formats);
82 write_unlock(&binfmt_lock);
86 EXPORT_SYMBOL(__register_binfmt);
88 void unregister_binfmt(struct linux_binfmt * fmt)
90 write_lock(&binfmt_lock);
92 write_unlock(&binfmt_lock);
95 EXPORT_SYMBOL(unregister_binfmt);
97 static inline void put_binfmt(struct linux_binfmt * fmt)
99 module_put(fmt->module);
103 * Note that a shared library must be both readable and executable due to
106 * Also note that we take the address to load from from the file itself.
108 SYSCALL_DEFINE1(uselib, const char __user *, library)
111 char *tmp = getname(library);
112 int error = PTR_ERR(tmp);
117 file = do_filp_open(AT_FDCWD, tmp,
118 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
119 MAY_READ | MAY_EXEC | MAY_OPEN);
121 error = PTR_ERR(file);
126 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
130 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
133 fsnotify_open(file->f_path.dentry);
137 struct linux_binfmt * fmt;
139 read_lock(&binfmt_lock);
140 list_for_each_entry(fmt, &formats, lh) {
141 if (!fmt->load_shlib)
143 if (!try_module_get(fmt->module))
145 read_unlock(&binfmt_lock);
146 error = fmt->load_shlib(file);
147 read_lock(&binfmt_lock);
149 if (error != -ENOEXEC)
152 read_unlock(&binfmt_lock);
162 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
164 struct mm_struct *mm = current->mm;
165 long diff = (long)(pages - bprm->vma_pages);
170 bprm->vma_pages = pages;
172 down_write(&mm->mmap_sem);
173 mm->total_vm += diff;
174 up_write(&mm->mmap_sem);
177 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
183 #ifdef CONFIG_STACK_GROWSUP
185 ret = expand_stack_downwards(bprm->vma, pos);
190 ret = get_user_pages(current, bprm->mm, pos,
191 1, write, 1, &page, NULL);
196 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
199 acct_arg_size(bprm, size / PAGE_SIZE);
202 * We've historically supported up to 32 pages (ARG_MAX)
203 * of argument strings even with small stacks
209 * Limit to 1/4-th the stack size for the argv+env strings.
211 * - the remaining binfmt code will not run out of stack space,
212 * - the program will have a reasonable amount of stack left
215 rlim = current->signal->rlim;
216 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
225 static void put_arg_page(struct page *page)
230 static void free_arg_page(struct linux_binprm *bprm, int i)
234 static void free_arg_pages(struct linux_binprm *bprm)
238 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
241 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
244 static int __bprm_mm_init(struct linux_binprm *bprm)
247 struct vm_area_struct *vma = NULL;
248 struct mm_struct *mm = bprm->mm;
250 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
254 down_write(&mm->mmap_sem);
258 * Place the stack at the largest stack address the architecture
259 * supports. Later, we'll move this to an appropriate place. We don't
260 * use STACK_TOP because that can depend on attributes which aren't
263 vma->vm_end = STACK_TOP_MAX;
264 vma->vm_start = vma->vm_end - PAGE_SIZE;
265 vma->vm_flags = VM_STACK_FLAGS;
266 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
268 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
272 err = insert_vm_struct(mm, vma);
276 mm->stack_vm = mm->total_vm = 1;
277 up_write(&mm->mmap_sem);
278 bprm->p = vma->vm_end - sizeof(void *);
281 up_write(&mm->mmap_sem);
283 kmem_cache_free(vm_area_cachep, vma);
287 static bool valid_arg_len(struct linux_binprm *bprm, long len)
289 return len <= MAX_ARG_STRLEN;
294 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
298 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
303 page = bprm->page[pos / PAGE_SIZE];
304 if (!page && write) {
305 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
308 bprm->page[pos / PAGE_SIZE] = page;
314 static void put_arg_page(struct page *page)
318 static void free_arg_page(struct linux_binprm *bprm, int i)
321 __free_page(bprm->page[i]);
322 bprm->page[i] = NULL;
326 static void free_arg_pages(struct linux_binprm *bprm)
330 for (i = 0; i < MAX_ARG_PAGES; i++)
331 free_arg_page(bprm, i);
334 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
339 static int __bprm_mm_init(struct linux_binprm *bprm)
341 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
345 static bool valid_arg_len(struct linux_binprm *bprm, long len)
347 return len <= bprm->p;
350 #endif /* CONFIG_MMU */
353 * Create a new mm_struct and populate it with a temporary stack
354 * vm_area_struct. We don't have enough context at this point to set the stack
355 * flags, permissions, and offset, so we use temporary values. We'll update
356 * them later in setup_arg_pages().
358 int bprm_mm_init(struct linux_binprm *bprm)
361 struct mm_struct *mm = NULL;
363 bprm->mm = mm = mm_alloc();
368 err = init_new_context(current, mm);
372 err = __bprm_mm_init(bprm);
388 * count() counts the number of strings in array ARGV.
390 static int count(char __user * __user * argv, int max)
398 if (get_user(p, argv))
406 if (fatal_signal_pending(current))
407 return -ERESTARTNOHAND;
415 * 'copy_strings()' copies argument/environment strings from the old
416 * processes's memory to the new process's stack. The call to get_user_pages()
417 * ensures the destination page is created and not swapped out.
419 static int copy_strings(int argc, char __user * __user * argv,
420 struct linux_binprm *bprm)
422 struct page *kmapped_page = NULL;
424 unsigned long kpos = 0;
432 if (get_user(str, argv+argc) ||
433 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
438 if (!valid_arg_len(bprm, len)) {
443 /* We're going to work our way backwords. */
449 int offset, bytes_to_copy;
451 if (fatal_signal_pending(current)) {
452 ret = -ERESTARTNOHAND;
457 offset = pos % PAGE_SIZE;
461 bytes_to_copy = offset;
462 if (bytes_to_copy > len)
465 offset -= bytes_to_copy;
466 pos -= bytes_to_copy;
467 str -= bytes_to_copy;
468 len -= bytes_to_copy;
470 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
473 page = get_arg_page(bprm, pos, 1);
480 flush_kernel_dcache_page(kmapped_page);
481 kunmap(kmapped_page);
482 put_arg_page(kmapped_page);
485 kaddr = kmap(kmapped_page);
486 kpos = pos & PAGE_MASK;
487 flush_arg_page(bprm, kpos, kmapped_page);
489 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
498 flush_kernel_dcache_page(kmapped_page);
499 kunmap(kmapped_page);
500 put_arg_page(kmapped_page);
506 * Like copy_strings, but get argv and its values from kernel memory.
508 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
511 mm_segment_t oldfs = get_fs();
513 r = copy_strings(argc, (char __user * __user *)argv, bprm);
517 EXPORT_SYMBOL(copy_strings_kernel);
522 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
523 * the binfmt code determines where the new stack should reside, we shift it to
524 * its final location. The process proceeds as follows:
526 * 1) Use shift to calculate the new vma endpoints.
527 * 2) Extend vma to cover both the old and new ranges. This ensures the
528 * arguments passed to subsequent functions are consistent.
529 * 3) Move vma's page tables to the new range.
530 * 4) Free up any cleared pgd range.
531 * 5) Shrink the vma to cover only the new range.
533 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
535 struct mm_struct *mm = vma->vm_mm;
536 unsigned long old_start = vma->vm_start;
537 unsigned long old_end = vma->vm_end;
538 unsigned long length = old_end - old_start;
539 unsigned long new_start = old_start - shift;
540 unsigned long new_end = old_end - shift;
541 struct mmu_gather *tlb;
543 BUG_ON(new_start > new_end);
546 * ensure there are no vmas between where we want to go
549 if (vma != find_vma(mm, new_start))
553 * cover the whole range: [new_start, old_end)
555 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
558 * move the page tables downwards, on failure we rely on
559 * process cleanup to remove whatever mess we made.
561 if (length != move_page_tables(vma, old_start,
562 vma, new_start, length))
566 tlb = tlb_gather_mmu(mm, 0);
567 if (new_end > old_start) {
569 * when the old and new regions overlap clear from new_end.
571 free_pgd_range(tlb, new_end, old_end, new_end,
572 vma->vm_next ? vma->vm_next->vm_start : 0);
575 * otherwise, clean from old_start; this is done to not touch
576 * the address space in [new_end, old_start) some architectures
577 * have constraints on va-space that make this illegal (IA64) -
578 * for the others its just a little faster.
580 free_pgd_range(tlb, old_start, old_end, new_end,
581 vma->vm_next ? vma->vm_next->vm_start : 0);
583 tlb_finish_mmu(tlb, new_end, old_end);
586 * shrink the vma to just the new range.
588 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
593 #define EXTRA_STACK_VM_PAGES 20 /* random */
596 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
597 * the stack is optionally relocated, and some extra space is added.
599 int setup_arg_pages(struct linux_binprm *bprm,
600 unsigned long stack_top,
601 int executable_stack)
604 unsigned long stack_shift;
605 struct mm_struct *mm = current->mm;
606 struct vm_area_struct *vma = bprm->vma;
607 struct vm_area_struct *prev = NULL;
608 unsigned long vm_flags;
609 unsigned long stack_base;
610 unsigned long stack_size;
611 unsigned long stack_expand;
612 unsigned long rlim_stack;
614 #ifdef CONFIG_STACK_GROWSUP
615 /* Limit stack size to 1GB */
616 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
617 if (stack_base > (1 << 30))
618 stack_base = 1 << 30;
620 /* Make sure we didn't let the argument array grow too large. */
621 if (vma->vm_end - vma->vm_start > stack_base)
624 stack_base = PAGE_ALIGN(stack_top - stack_base);
626 stack_shift = vma->vm_start - stack_base;
627 mm->arg_start = bprm->p - stack_shift;
628 bprm->p = vma->vm_end - stack_shift;
630 stack_top = arch_align_stack(stack_top);
631 stack_top = PAGE_ALIGN(stack_top);
633 if (unlikely(stack_top < mmap_min_addr) ||
634 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
637 stack_shift = vma->vm_end - stack_top;
639 bprm->p -= stack_shift;
640 mm->arg_start = bprm->p;
644 bprm->loader -= stack_shift;
645 bprm->exec -= stack_shift;
647 down_write(&mm->mmap_sem);
648 vm_flags = VM_STACK_FLAGS;
651 * Adjust stack execute permissions; explicitly enable for
652 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
653 * (arch default) otherwise.
655 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
657 else if (executable_stack == EXSTACK_DISABLE_X)
658 vm_flags &= ~VM_EXEC;
659 vm_flags |= mm->def_flags;
661 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
667 /* Move stack pages down in memory. */
669 ret = shift_arg_pages(vma, stack_shift);
674 stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
675 stack_size = vma->vm_end - vma->vm_start;
677 * Align this down to a page boundary as expand_stack
680 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
681 #ifdef CONFIG_STACK_GROWSUP
682 if (stack_size + stack_expand > rlim_stack)
683 stack_base = vma->vm_start + rlim_stack;
685 stack_base = vma->vm_end + stack_expand;
687 if (stack_size + stack_expand > rlim_stack)
688 stack_base = vma->vm_end - rlim_stack;
690 stack_base = vma->vm_start - stack_expand;
692 ret = expand_stack(vma, stack_base);
697 up_write(&mm->mmap_sem);
700 EXPORT_SYMBOL(setup_arg_pages);
702 #endif /* CONFIG_MMU */
704 struct file *open_exec(const char *name)
709 file = do_filp_open(AT_FDCWD, name,
710 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
711 MAY_EXEC | MAY_OPEN);
716 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
719 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
722 fsnotify_open(file->f_path.dentry);
724 err = deny_write_access(file);
735 EXPORT_SYMBOL(open_exec);
737 int kernel_read(struct file *file, loff_t offset,
738 char *addr, unsigned long count)
746 /* The cast to a user pointer is valid due to the set_fs() */
747 result = vfs_read(file, (void __user *)addr, count, &pos);
752 EXPORT_SYMBOL(kernel_read);
754 static int exec_mmap(struct mm_struct *mm)
756 struct task_struct *tsk;
757 struct mm_struct * old_mm, *active_mm;
759 /* Notify parent that we're no longer interested in the old VM */
761 old_mm = current->mm;
762 mm_release(tsk, old_mm);
766 * Make sure that if there is a core dump in progress
767 * for the old mm, we get out and die instead of going
768 * through with the exec. We must hold mmap_sem around
769 * checking core_state and changing tsk->mm.
771 down_read(&old_mm->mmap_sem);
772 if (unlikely(old_mm->core_state)) {
773 up_read(&old_mm->mmap_sem);
778 active_mm = tsk->active_mm;
781 activate_mm(active_mm, mm);
783 arch_pick_mmap_layout(mm);
785 up_read(&old_mm->mmap_sem);
786 BUG_ON(active_mm != old_mm);
787 mm_update_next_owner(old_mm);
796 * This function makes sure the current process has its own signal table,
797 * so that flush_signal_handlers can later reset the handlers without
798 * disturbing other processes. (Other processes might share the signal
799 * table via the CLONE_SIGHAND option to clone().)
801 static int de_thread(struct task_struct *tsk)
803 struct signal_struct *sig = tsk->signal;
804 struct sighand_struct *oldsighand = tsk->sighand;
805 spinlock_t *lock = &oldsighand->siglock;
808 if (thread_group_empty(tsk))
809 goto no_thread_group;
812 * Kill all other threads in the thread group.
815 if (signal_group_exit(sig)) {
817 * Another group action in progress, just
818 * return so that the signal is processed.
820 spin_unlock_irq(lock);
823 sig->group_exit_task = tsk;
824 zap_other_threads(tsk);
826 /* Account for the thread group leader hanging around: */
827 count = thread_group_leader(tsk) ? 1 : 2;
828 sig->notify_count = count;
829 while (atomic_read(&sig->count) > count) {
830 __set_current_state(TASK_UNINTERRUPTIBLE);
831 spin_unlock_irq(lock);
835 spin_unlock_irq(lock);
838 * At this point all other threads have exited, all we have to
839 * do is to wait for the thread group leader to become inactive,
840 * and to assume its PID:
842 if (!thread_group_leader(tsk)) {
843 struct task_struct *leader = tsk->group_leader;
845 sig->notify_count = -1; /* for exit_notify() */
847 write_lock_irq(&tasklist_lock);
848 if (likely(leader->exit_state))
850 __set_current_state(TASK_UNINTERRUPTIBLE);
851 write_unlock_irq(&tasklist_lock);
856 * The only record we have of the real-time age of a
857 * process, regardless of execs it's done, is start_time.
858 * All the past CPU time is accumulated in signal_struct
859 * from sister threads now dead. But in this non-leader
860 * exec, nothing survives from the original leader thread,
861 * whose birth marks the true age of this process now.
862 * When we take on its identity by switching to its PID, we
863 * also take its birthdate (always earlier than our own).
865 tsk->start_time = leader->start_time;
867 BUG_ON(!same_thread_group(leader, tsk));
868 BUG_ON(has_group_leader_pid(tsk));
870 * An exec() starts a new thread group with the
871 * TGID of the previous thread group. Rehash the
872 * two threads with a switched PID, and release
873 * the former thread group leader:
876 /* Become a process group leader with the old leader's pid.
877 * The old leader becomes a thread of the this thread group.
878 * Note: The old leader also uses this pid until release_task
879 * is called. Odd but simple and correct.
881 detach_pid(tsk, PIDTYPE_PID);
882 tsk->pid = leader->pid;
883 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
884 transfer_pid(leader, tsk, PIDTYPE_PGID);
885 transfer_pid(leader, tsk, PIDTYPE_SID);
886 list_replace_rcu(&leader->tasks, &tsk->tasks);
888 tsk->group_leader = tsk;
889 leader->group_leader = tsk;
891 tsk->exit_signal = SIGCHLD;
893 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
894 leader->exit_state = EXIT_DEAD;
895 write_unlock_irq(&tasklist_lock);
897 release_task(leader);
900 sig->group_exit_task = NULL;
901 sig->notify_count = 0;
905 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
908 flush_itimer_signals();
910 if (atomic_read(&oldsighand->count) != 1) {
911 struct sighand_struct *newsighand;
913 * This ->sighand is shared with the CLONE_SIGHAND
914 * but not CLONE_THREAD task, switch to the new one.
916 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
920 atomic_set(&newsighand->count, 1);
921 memcpy(newsighand->action, oldsighand->action,
922 sizeof(newsighand->action));
924 write_lock_irq(&tasklist_lock);
925 spin_lock(&oldsighand->siglock);
926 rcu_assign_pointer(tsk->sighand, newsighand);
927 spin_unlock(&oldsighand->siglock);
928 write_unlock_irq(&tasklist_lock);
930 __cleanup_sighand(oldsighand);
933 BUG_ON(!thread_group_leader(tsk));
938 * These functions flushes out all traces of the currently running executable
939 * so that a new one can be started
941 static void flush_old_files(struct files_struct * files)
946 spin_lock(&files->file_lock);
948 unsigned long set, i;
952 fdt = files_fdtable(files);
953 if (i >= fdt->max_fds)
955 set = fdt->close_on_exec->fds_bits[j];
958 fdt->close_on_exec->fds_bits[j] = 0;
959 spin_unlock(&files->file_lock);
960 for ( ; set ; i++,set >>= 1) {
965 spin_lock(&files->file_lock);
968 spin_unlock(&files->file_lock);
971 char *get_task_comm(char *buf, struct task_struct *tsk)
973 /* buf must be at least sizeof(tsk->comm) in size */
975 strncpy(buf, tsk->comm, sizeof(tsk->comm));
980 void set_task_comm(struct task_struct *tsk, char *buf)
983 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
985 perf_event_comm(tsk);
988 int flush_old_exec(struct linux_binprm * bprm)
993 * Make sure we have a private signal table and that
994 * we are unassociated from the previous thread group.
996 retval = de_thread(current);
1000 set_mm_exe_file(bprm->mm, bprm->file);
1003 * Release all of the old mmap stuff
1005 acct_arg_size(bprm, 0);
1006 retval = exec_mmap(bprm->mm);
1010 bprm->mm = NULL; /* We're using it now */
1013 current->flags &= ~PF_RANDOMIZE;
1015 current->personality &= ~bprm->per_clear;
1022 EXPORT_SYMBOL(flush_old_exec);
1024 void setup_new_exec(struct linux_binprm * bprm)
1028 char tcomm[sizeof(current->comm)];
1030 arch_pick_mmap_layout(current->mm);
1032 /* This is the point of no return */
1033 current->sas_ss_sp = current->sas_ss_size = 0;
1035 if (current_euid() == current_uid() && current_egid() == current_gid())
1036 set_dumpable(current->mm, 1);
1038 set_dumpable(current->mm, suid_dumpable);
1040 name = bprm->filename;
1042 /* Copies the binary name from after last slash */
1043 for (i=0; (ch = *(name++)) != '\0';) {
1045 i = 0; /* overwrite what we wrote */
1047 if (i < (sizeof(tcomm) - 1))
1051 set_task_comm(current, tcomm);
1053 /* Set the new mm task size. We have to do that late because it may
1054 * depend on TIF_32BIT which is only updated in flush_thread() on
1055 * some architectures like powerpc
1057 current->mm->task_size = TASK_SIZE;
1059 /* install the new credentials */
1060 if (bprm->cred->uid != current_euid() ||
1061 bprm->cred->gid != current_egid()) {
1062 current->pdeath_signal = 0;
1063 } else if (file_permission(bprm->file, MAY_READ) ||
1064 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1065 set_dumpable(current->mm, suid_dumpable);
1069 * Flush performance counters when crossing a
1072 if (!get_dumpable(current->mm))
1073 perf_event_exit_task(current);
1075 /* An exec changes our domain. We are no longer part of the thread
1078 current->self_exec_id++;
1080 flush_signal_handlers(current, 0);
1081 flush_old_files(current->files);
1083 EXPORT_SYMBOL(setup_new_exec);
1086 * Prepare credentials and lock ->cred_guard_mutex.
1087 * install_exec_creds() commits the new creds and drops the lock.
1088 * Or, if exec fails before, free_bprm() should release ->cred and
1091 int prepare_bprm_creds(struct linux_binprm *bprm)
1093 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1094 return -ERESTARTNOINTR;
1096 bprm->cred = prepare_exec_creds();
1097 if (likely(bprm->cred))
1100 mutex_unlock(¤t->cred_guard_mutex);
1104 void free_bprm(struct linux_binprm *bprm)
1106 free_arg_pages(bprm);
1108 mutex_unlock(¤t->cred_guard_mutex);
1109 abort_creds(bprm->cred);
1115 * install the new credentials for this executable
1117 void install_exec_creds(struct linux_binprm *bprm)
1119 security_bprm_committing_creds(bprm);
1121 commit_creds(bprm->cred);
1124 * cred_guard_mutex must be held at least to this point to prevent
1125 * ptrace_attach() from altering our determination of the task's
1126 * credentials; any time after this it may be unlocked.
1128 security_bprm_committed_creds(bprm);
1129 mutex_unlock(¤t->cred_guard_mutex);
1131 EXPORT_SYMBOL(install_exec_creds);
1134 * determine how safe it is to execute the proposed program
1135 * - the caller must hold current->cred_guard_mutex to protect against
1138 int check_unsafe_exec(struct linux_binprm *bprm)
1140 struct task_struct *p = current, *t;
1144 bprm->unsafe = tracehook_unsafe_exec(p);
1147 write_lock(&p->fs->lock);
1149 for (t = next_thread(p); t != p; t = next_thread(t)) {
1155 if (p->fs->users > n_fs) {
1156 bprm->unsafe |= LSM_UNSAFE_SHARE;
1159 if (!p->fs->in_exec) {
1164 write_unlock(&p->fs->lock);
1170 * Fill the binprm structure from the inode.
1171 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1173 * This may be called multiple times for binary chains (scripts for example).
1175 int prepare_binprm(struct linux_binprm *bprm)
1178 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1181 mode = inode->i_mode;
1182 if (bprm->file->f_op == NULL)
1185 /* clear any previous set[ug]id data from a previous binary */
1186 bprm->cred->euid = current_euid();
1187 bprm->cred->egid = current_egid();
1189 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1191 if (mode & S_ISUID) {
1192 bprm->per_clear |= PER_CLEAR_ON_SETID;
1193 bprm->cred->euid = inode->i_uid;
1198 * If setgid is set but no group execute bit then this
1199 * is a candidate for mandatory locking, not a setgid
1202 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1203 bprm->per_clear |= PER_CLEAR_ON_SETID;
1204 bprm->cred->egid = inode->i_gid;
1208 /* fill in binprm security blob */
1209 retval = security_bprm_set_creds(bprm);
1212 bprm->cred_prepared = 1;
1214 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1215 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1218 EXPORT_SYMBOL(prepare_binprm);
1221 * Arguments are '\0' separated strings found at the location bprm->p
1222 * points to; chop off the first by relocating brpm->p to right after
1223 * the first '\0' encountered.
1225 int remove_arg_zero(struct linux_binprm *bprm)
1228 unsigned long offset;
1236 offset = bprm->p & ~PAGE_MASK;
1237 page = get_arg_page(bprm, bprm->p, 0);
1242 kaddr = kmap_atomic(page, KM_USER0);
1244 for (; offset < PAGE_SIZE && kaddr[offset];
1245 offset++, bprm->p++)
1248 kunmap_atomic(kaddr, KM_USER0);
1251 if (offset == PAGE_SIZE)
1252 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1253 } while (offset == PAGE_SIZE);
1262 EXPORT_SYMBOL(remove_arg_zero);
1265 * cycle the list of binary formats handler, until one recognizes the image
1267 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1269 unsigned int depth = bprm->recursion_depth;
1271 struct linux_binfmt *fmt;
1273 retval = security_bprm_check(bprm);
1276 retval = ima_bprm_check(bprm);
1280 retval = audit_bprm(bprm);
1285 for (try=0; try<2; try++) {
1286 read_lock(&binfmt_lock);
1287 list_for_each_entry(fmt, &formats, lh) {
1288 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1291 if (!try_module_get(fmt->module))
1293 read_unlock(&binfmt_lock);
1294 retval = fn(bprm, regs);
1296 * Restore the depth counter to its starting value
1297 * in this call, so we don't have to rely on every
1298 * load_binary function to restore it on return.
1300 bprm->recursion_depth = depth;
1303 tracehook_report_exec(fmt, bprm, regs);
1305 allow_write_access(bprm->file);
1309 current->did_exec = 1;
1310 proc_exec_connector(current);
1313 read_lock(&binfmt_lock);
1315 if (retval != -ENOEXEC || bprm->mm == NULL)
1318 read_unlock(&binfmt_lock);
1322 read_unlock(&binfmt_lock);
1323 if (retval != -ENOEXEC || bprm->mm == NULL) {
1325 #ifdef CONFIG_MODULES
1327 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1328 if (printable(bprm->buf[0]) &&
1329 printable(bprm->buf[1]) &&
1330 printable(bprm->buf[2]) &&
1331 printable(bprm->buf[3]))
1332 break; /* -ENOEXEC */
1333 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1340 EXPORT_SYMBOL(search_binary_handler);
1343 * sys_execve() executes a new program.
1345 int do_execve(char * filename,
1346 char __user *__user *argv,
1347 char __user *__user *envp,
1348 struct pt_regs * regs)
1350 struct linux_binprm *bprm;
1352 struct files_struct *displaced;
1356 retval = unshare_files(&displaced);
1361 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1365 retval = prepare_bprm_creds(bprm);
1369 retval = check_unsafe_exec(bprm);
1372 clear_in_exec = retval;
1373 current->in_execve = 1;
1375 file = open_exec(filename);
1376 retval = PTR_ERR(file);
1383 bprm->filename = filename;
1384 bprm->interp = filename;
1386 retval = bprm_mm_init(bprm);
1390 bprm->argc = count(argv, MAX_ARG_STRINGS);
1391 if ((retval = bprm->argc) < 0)
1394 bprm->envc = count(envp, MAX_ARG_STRINGS);
1395 if ((retval = bprm->envc) < 0)
1398 retval = prepare_binprm(bprm);
1402 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1406 bprm->exec = bprm->p;
1407 retval = copy_strings(bprm->envc, envp, bprm);
1411 retval = copy_strings(bprm->argc, argv, bprm);
1415 current->flags &= ~PF_KTHREAD;
1416 retval = search_binary_handler(bprm,regs);
1420 /* execve succeeded */
1421 current->fs->in_exec = 0;
1422 current->in_execve = 0;
1423 acct_update_integrals(current);
1426 put_files_struct(displaced);
1431 acct_arg_size(bprm, 0);
1437 allow_write_access(bprm->file);
1443 current->fs->in_exec = 0;
1444 current->in_execve = 0;
1451 reset_files_struct(displaced);
1456 void set_binfmt(struct linux_binfmt *new)
1458 struct mm_struct *mm = current->mm;
1461 module_put(mm->binfmt->module);
1465 __module_get(new->module);
1468 EXPORT_SYMBOL(set_binfmt);
1470 /* format_corename will inspect the pattern parameter, and output a
1471 * name into corename, which must have space for at least
1472 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1474 static int format_corename(char *corename, long signr)
1476 const struct cred *cred = current_cred();
1477 const char *pat_ptr = core_pattern;
1478 int ispipe = (*pat_ptr == '|');
1479 char *out_ptr = corename;
1480 char *const out_end = corename + CORENAME_MAX_SIZE;
1482 int pid_in_pattern = 0;
1484 /* Repeat as long as we have more pattern to process and more output
1487 if (*pat_ptr != '%') {
1488 if (out_ptr == out_end)
1490 *out_ptr++ = *pat_ptr++;
1492 switch (*++pat_ptr) {
1495 /* Double percent, output one percent */
1497 if (out_ptr == out_end)
1504 rc = snprintf(out_ptr, out_end - out_ptr,
1505 "%d", task_tgid_vnr(current));
1506 if (rc > out_end - out_ptr)
1512 rc = snprintf(out_ptr, out_end - out_ptr,
1514 if (rc > out_end - out_ptr)
1520 rc = snprintf(out_ptr, out_end - out_ptr,
1522 if (rc > out_end - out_ptr)
1526 /* signal that caused the coredump */
1528 rc = snprintf(out_ptr, out_end - out_ptr,
1530 if (rc > out_end - out_ptr)
1534 /* UNIX time of coredump */
1537 do_gettimeofday(&tv);
1538 rc = snprintf(out_ptr, out_end - out_ptr,
1540 if (rc > out_end - out_ptr)
1547 down_read(&uts_sem);
1548 rc = snprintf(out_ptr, out_end - out_ptr,
1549 "%s", utsname()->nodename);
1551 if (rc > out_end - out_ptr)
1557 rc = snprintf(out_ptr, out_end - out_ptr,
1558 "%s", current->comm);
1559 if (rc > out_end - out_ptr)
1563 /* core limit size */
1565 rc = snprintf(out_ptr, out_end - out_ptr,
1566 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1567 if (rc > out_end - out_ptr)
1577 /* Backward compatibility with core_uses_pid:
1579 * If core_pattern does not include a %p (as is the default)
1580 * and core_uses_pid is set, then .%pid will be appended to
1581 * the filename. Do not do this for piped commands. */
1582 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1583 rc = snprintf(out_ptr, out_end - out_ptr,
1584 ".%d", task_tgid_vnr(current));
1585 if (rc > out_end - out_ptr)
1594 static int zap_process(struct task_struct *start)
1596 struct task_struct *t;
1599 start->signal->flags = SIGNAL_GROUP_EXIT;
1600 start->signal->group_stop_count = 0;
1604 if (t != current && t->mm) {
1605 sigaddset(&t->pending.signal, SIGKILL);
1606 signal_wake_up(t, 1);
1609 } while_each_thread(start, t);
1614 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1615 struct core_state *core_state, int exit_code)
1617 struct task_struct *g, *p;
1618 unsigned long flags;
1621 spin_lock_irq(&tsk->sighand->siglock);
1622 if (!signal_group_exit(tsk->signal)) {
1623 mm->core_state = core_state;
1624 tsk->signal->group_exit_code = exit_code;
1625 nr = zap_process(tsk);
1627 spin_unlock_irq(&tsk->sighand->siglock);
1628 if (unlikely(nr < 0))
1631 if (atomic_read(&mm->mm_users) == nr + 1)
1634 * We should find and kill all tasks which use this mm, and we should
1635 * count them correctly into ->nr_threads. We don't take tasklist
1636 * lock, but this is safe wrt:
1639 * None of sub-threads can fork after zap_process(leader). All
1640 * processes which were created before this point should be
1641 * visible to zap_threads() because copy_process() adds the new
1642 * process to the tail of init_task.tasks list, and lock/unlock
1643 * of ->siglock provides a memory barrier.
1646 * The caller holds mm->mmap_sem. This means that the task which
1647 * uses this mm can't pass exit_mm(), so it can't exit or clear
1651 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1652 * we must see either old or new leader, this does not matter.
1653 * However, it can change p->sighand, so lock_task_sighand(p)
1654 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1657 * Note also that "g" can be the old leader with ->mm == NULL
1658 * and already unhashed and thus removed from ->thread_group.
1659 * This is OK, __unhash_process()->list_del_rcu() does not
1660 * clear the ->next pointer, we will find the new leader via
1664 for_each_process(g) {
1665 if (g == tsk->group_leader)
1667 if (g->flags & PF_KTHREAD)
1672 if (unlikely(p->mm == mm)) {
1673 lock_task_sighand(p, &flags);
1674 nr += zap_process(p);
1675 unlock_task_sighand(p, &flags);
1679 } while_each_thread(g, p);
1683 atomic_set(&core_state->nr_threads, nr);
1687 static int coredump_wait(int exit_code, struct core_state *core_state)
1689 struct task_struct *tsk = current;
1690 struct mm_struct *mm = tsk->mm;
1691 struct completion *vfork_done;
1694 init_completion(&core_state->startup);
1695 core_state->dumper.task = tsk;
1696 core_state->dumper.next = NULL;
1697 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1698 up_write(&mm->mmap_sem);
1700 if (unlikely(core_waiters < 0))
1704 * Make sure nobody is waiting for us to release the VM,
1705 * otherwise we can deadlock when we wait on each other
1707 vfork_done = tsk->vfork_done;
1709 tsk->vfork_done = NULL;
1710 complete(vfork_done);
1714 wait_for_completion(&core_state->startup);
1716 return core_waiters;
1719 static void coredump_finish(struct mm_struct *mm)
1721 struct core_thread *curr, *next;
1722 struct task_struct *task;
1724 next = mm->core_state->dumper.next;
1725 while ((curr = next) != NULL) {
1729 * see exit_mm(), curr->task must not see
1730 * ->task == NULL before we read ->next.
1734 wake_up_process(task);
1737 mm->core_state = NULL;
1741 * set_dumpable converts traditional three-value dumpable to two flags and
1742 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1743 * these bits are not changed atomically. So get_dumpable can observe the
1744 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1745 * return either old dumpable or new one by paying attention to the order of
1746 * modifying the bits.
1748 * dumpable | mm->flags (binary)
1749 * old new | initial interim final
1750 * ---------+-----------------------
1758 * (*) get_dumpable regards interim value of 10 as 11.
1760 void set_dumpable(struct mm_struct *mm, int value)
1764 clear_bit(MMF_DUMPABLE, &mm->flags);
1766 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1769 set_bit(MMF_DUMPABLE, &mm->flags);
1771 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1774 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1776 set_bit(MMF_DUMPABLE, &mm->flags);
1781 int get_dumpable(struct mm_struct *mm)
1785 ret = mm->flags & 0x3;
1786 return (ret >= 2) ? 2 : ret;
1789 static void wait_for_dump_helpers(struct file *file)
1791 struct pipe_inode_info *pipe;
1793 pipe = file->f_path.dentry->d_inode->i_pipe;
1799 while ((pipe->readers > 1) && (!signal_pending(current))) {
1800 wake_up_interruptible_sync(&pipe->wait);
1801 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1812 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1814 struct core_state core_state;
1815 char corename[CORENAME_MAX_SIZE + 1];
1816 struct mm_struct *mm = current->mm;
1817 struct linux_binfmt * binfmt;
1818 struct inode * inode;
1820 const struct cred *old_cred;
1825 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1826 char **helper_argv = NULL;
1827 int helper_argc = 0;
1829 static atomic_t core_dump_count = ATOMIC_INIT(0);
1831 audit_core_dumps(signr);
1833 binfmt = mm->binfmt;
1834 if (!binfmt || !binfmt->core_dump)
1837 cred = prepare_creds();
1843 down_write(&mm->mmap_sem);
1845 * If another thread got here first, or we are not dumpable, bail out.
1847 if (mm->core_state || !get_dumpable(mm)) {
1848 up_write(&mm->mmap_sem);
1854 * We cannot trust fsuid as being the "true" uid of the
1855 * process nor do we know its entire history. We only know it
1856 * was tainted so we dump it as root in mode 2.
1858 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1859 flag = O_EXCL; /* Stop rewrite attacks */
1860 cred->fsuid = 0; /* Dump root private */
1863 retval = coredump_wait(exit_code, &core_state);
1869 old_cred = override_creds(cred);
1872 * Clear any false indication of pending signals that might
1873 * be seen by the filesystem code called to write the core file.
1875 clear_thread_flag(TIF_SIGPENDING);
1878 * lock_kernel() because format_corename() is controlled by sysctl, which
1879 * uses lock_kernel()
1882 ispipe = format_corename(corename, signr);
1885 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1889 if (core_limit == 0) {
1891 * Normally core limits are irrelevant to pipes, since
1892 * we're not writing to the file system, but we use
1893 * core_limit of 0 here as a speacial value. Any
1894 * non-zero limit gets set to RLIM_INFINITY below, but
1895 * a limit of 0 skips the dump. This is a consistent
1896 * way to catch recursive crashes. We can still crash
1897 * if the core_pattern binary sets RLIM_CORE = !0
1898 * but it runs as root, and can do lots of stupid things
1899 * Note that we use task_tgid_vnr here to grab the pid
1900 * of the process group leader. That way we get the
1901 * right pid if a thread in a multi-threaded
1902 * core_pattern process dies.
1905 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1906 task_tgid_vnr(current), current->comm);
1907 printk(KERN_WARNING "Aborting core\n");
1911 dump_count = atomic_inc_return(&core_dump_count);
1912 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1913 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1914 task_tgid_vnr(current), current->comm);
1915 printk(KERN_WARNING "Skipping core dump\n");
1916 goto fail_dropcount;
1919 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1921 printk(KERN_WARNING "%s failed to allocate memory\n",
1923 goto fail_dropcount;
1926 core_limit = RLIM_INFINITY;
1928 /* SIGPIPE can happen, but it's just never processed */
1929 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1931 printk(KERN_INFO "Core dump to %s pipe failed\n",
1933 goto fail_dropcount;
1936 file = filp_open(corename,
1937 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1940 goto fail_dropcount;
1941 inode = file->f_path.dentry->d_inode;
1942 if (inode->i_nlink > 1)
1943 goto close_fail; /* multiple links - don't dump */
1944 if (!ispipe && d_unhashed(file->f_path.dentry))
1947 /* AK: actually i see no reason to not allow this for named pipes etc.,
1948 but keep the previous behaviour for now. */
1949 if (!ispipe && !S_ISREG(inode->i_mode))
1952 * Dont allow local users get cute and trick others to coredump
1953 * into their pre-created files:
1954 * Note, this is not relevant for pipes
1956 if (!ispipe && (inode->i_uid != current_fsuid()))
1960 if (!file->f_op->write)
1962 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1965 retval = binfmt->core_dump(signr, regs, file, core_limit);
1968 current->signal->group_exit_code |= 0x80;
1970 if (ispipe && core_pipe_limit)
1971 wait_for_dump_helpers(file);
1972 filp_close(file, NULL);
1975 atomic_dec(&core_dump_count);
1978 argv_free(helper_argv);
1980 revert_creds(old_cred);
1982 coredump_finish(mm);