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/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/utsname.h>
41 #include <linux/pid_namespace.h>
42 #include <linux/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
58 #include <linux/kmod.h>
62 /* for /sbin/loader handling in search_binary_handler() */
63 #include <linux/a.out.h>
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
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)
79 write_lock(&binfmt_lock);
80 list_add(&fmt->lh, &formats);
81 write_unlock(&binfmt_lock);
85 EXPORT_SYMBOL(register_binfmt);
87 void unregister_binfmt(struct linux_binfmt * fmt)
89 write_lock(&binfmt_lock);
91 write_unlock(&binfmt_lock);
94 EXPORT_SYMBOL(unregister_binfmt);
96 static inline void put_binfmt(struct linux_binfmt * fmt)
98 module_put(fmt->module);
102 * Note that a shared library must be both readable and executable due to
105 * Also note that we take the address to load from from the file itself.
107 asmlinkage long sys_uselib(const char __user * library)
113 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
118 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
121 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
125 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
126 error = PTR_ERR(file);
132 struct linux_binfmt * fmt;
134 read_lock(&binfmt_lock);
135 list_for_each_entry(fmt, &formats, lh) {
136 if (!fmt->load_shlib)
138 if (!try_module_get(fmt->module))
140 read_unlock(&binfmt_lock);
141 error = fmt->load_shlib(file);
142 read_lock(&binfmt_lock);
144 if (error != -ENOEXEC)
147 read_unlock(&binfmt_lock);
153 release_open_intent(&nd);
160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
166 #ifdef CONFIG_STACK_GROWSUP
168 ret = expand_stack_downwards(bprm->vma, pos);
173 ret = get_user_pages(current, bprm->mm, pos,
174 1, write, 1, &page, NULL);
179 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
183 * We've historically supported up to 32 pages (ARG_MAX)
184 * of argument strings even with small stacks
190 * Limit to 1/4-th the stack size for the argv+env strings.
192 * - the remaining binfmt code will not run out of stack space,
193 * - the program will have a reasonable amount of stack left
196 rlim = current->signal->rlim;
197 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
206 static void put_arg_page(struct page *page)
211 static void free_arg_page(struct linux_binprm *bprm, int i)
215 static void free_arg_pages(struct linux_binprm *bprm)
219 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
222 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
225 static int __bprm_mm_init(struct linux_binprm *bprm)
228 struct vm_area_struct *vma = NULL;
229 struct mm_struct *mm = bprm->mm;
231 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
235 down_write(&mm->mmap_sem);
239 * Place the stack at the largest stack address the architecture
240 * supports. Later, we'll move this to an appropriate place. We don't
241 * use STACK_TOP because that can depend on attributes which aren't
244 vma->vm_end = STACK_TOP_MAX;
245 vma->vm_start = vma->vm_end - PAGE_SIZE;
247 vma->vm_flags = VM_STACK_FLAGS;
248 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
249 err = insert_vm_struct(mm, vma);
251 up_write(&mm->mmap_sem);
255 mm->stack_vm = mm->total_vm = 1;
256 up_write(&mm->mmap_sem);
258 bprm->p = vma->vm_end - sizeof(void *);
265 kmem_cache_free(vm_area_cachep, vma);
271 static bool valid_arg_len(struct linux_binprm *bprm, long len)
273 return len <= MAX_ARG_STRLEN;
278 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
283 page = bprm->page[pos / PAGE_SIZE];
284 if (!page && write) {
285 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
288 bprm->page[pos / PAGE_SIZE] = page;
294 static void put_arg_page(struct page *page)
298 static void free_arg_page(struct linux_binprm *bprm, int i)
301 __free_page(bprm->page[i]);
302 bprm->page[i] = NULL;
306 static void free_arg_pages(struct linux_binprm *bprm)
310 for (i = 0; i < MAX_ARG_PAGES; i++)
311 free_arg_page(bprm, i);
314 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
319 static int __bprm_mm_init(struct linux_binprm *bprm)
321 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
325 static bool valid_arg_len(struct linux_binprm *bprm, long len)
327 return len <= bprm->p;
330 #endif /* CONFIG_MMU */
333 * Create a new mm_struct and populate it with a temporary stack
334 * vm_area_struct. We don't have enough context at this point to set the stack
335 * flags, permissions, and offset, so we use temporary values. We'll update
336 * them later in setup_arg_pages().
338 int bprm_mm_init(struct linux_binprm *bprm)
341 struct mm_struct *mm = NULL;
343 bprm->mm = mm = mm_alloc();
348 err = init_new_context(current, mm);
352 err = __bprm_mm_init(bprm);
368 * count() counts the number of strings in array ARGV.
370 static int count(char __user * __user * argv, int max)
378 if (get_user(p, argv))
392 * 'copy_strings()' copies argument/environment strings from the old
393 * processes's memory to the new process's stack. The call to get_user_pages()
394 * ensures the destination page is created and not swapped out.
396 static int copy_strings(int argc, char __user * __user * argv,
397 struct linux_binprm *bprm)
399 struct page *kmapped_page = NULL;
401 unsigned long kpos = 0;
409 if (get_user(str, argv+argc) ||
410 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
415 if (!valid_arg_len(bprm, len)) {
420 /* We're going to work our way backwords. */
426 int offset, bytes_to_copy;
428 offset = pos % PAGE_SIZE;
432 bytes_to_copy = offset;
433 if (bytes_to_copy > len)
436 offset -= bytes_to_copy;
437 pos -= bytes_to_copy;
438 str -= bytes_to_copy;
439 len -= bytes_to_copy;
441 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
444 page = get_arg_page(bprm, pos, 1);
451 flush_kernel_dcache_page(kmapped_page);
452 kunmap(kmapped_page);
453 put_arg_page(kmapped_page);
456 kaddr = kmap(kmapped_page);
457 kpos = pos & PAGE_MASK;
458 flush_arg_page(bprm, kpos, kmapped_page);
460 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
469 flush_kernel_dcache_page(kmapped_page);
470 kunmap(kmapped_page);
471 put_arg_page(kmapped_page);
477 * Like copy_strings, but get argv and its values from kernel memory.
479 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
482 mm_segment_t oldfs = get_fs();
484 r = copy_strings(argc, (char __user * __user *)argv, bprm);
488 EXPORT_SYMBOL(copy_strings_kernel);
493 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
494 * the binfmt code determines where the new stack should reside, we shift it to
495 * its final location. The process proceeds as follows:
497 * 1) Use shift to calculate the new vma endpoints.
498 * 2) Extend vma to cover both the old and new ranges. This ensures the
499 * arguments passed to subsequent functions are consistent.
500 * 3) Move vma's page tables to the new range.
501 * 4) Free up any cleared pgd range.
502 * 5) Shrink the vma to cover only the new range.
504 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
506 struct mm_struct *mm = vma->vm_mm;
507 unsigned long old_start = vma->vm_start;
508 unsigned long old_end = vma->vm_end;
509 unsigned long length = old_end - old_start;
510 unsigned long new_start = old_start - shift;
511 unsigned long new_end = old_end - shift;
512 struct mmu_gather *tlb;
514 BUG_ON(new_start > new_end);
517 * ensure there are no vmas between where we want to go
520 if (vma != find_vma(mm, new_start))
524 * cover the whole range: [new_start, old_end)
526 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
529 * move the page tables downwards, on failure we rely on
530 * process cleanup to remove whatever mess we made.
532 if (length != move_page_tables(vma, old_start,
533 vma, new_start, length))
537 tlb = tlb_gather_mmu(mm, 0);
538 if (new_end > old_start) {
540 * when the old and new regions overlap clear from new_end.
542 free_pgd_range(tlb, new_end, old_end, new_end,
543 vma->vm_next ? vma->vm_next->vm_start : 0);
546 * otherwise, clean from old_start; this is done to not touch
547 * the address space in [new_end, old_start) some architectures
548 * have constraints on va-space that make this illegal (IA64) -
549 * for the others its just a little faster.
551 free_pgd_range(tlb, old_start, old_end, new_end,
552 vma->vm_next ? vma->vm_next->vm_start : 0);
554 tlb_finish_mmu(tlb, new_end, old_end);
557 * shrink the vma to just the new range.
559 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
564 #define EXTRA_STACK_VM_PAGES 20 /* random */
567 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
568 * the stack is optionally relocated, and some extra space is added.
570 int setup_arg_pages(struct linux_binprm *bprm,
571 unsigned long stack_top,
572 int executable_stack)
575 unsigned long stack_shift;
576 struct mm_struct *mm = current->mm;
577 struct vm_area_struct *vma = bprm->vma;
578 struct vm_area_struct *prev = NULL;
579 unsigned long vm_flags;
580 unsigned long stack_base;
582 #ifdef CONFIG_STACK_GROWSUP
583 /* Limit stack size to 1GB */
584 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
585 if (stack_base > (1 << 30))
586 stack_base = 1 << 30;
588 /* Make sure we didn't let the argument array grow too large. */
589 if (vma->vm_end - vma->vm_start > stack_base)
592 stack_base = PAGE_ALIGN(stack_top - stack_base);
594 stack_shift = vma->vm_start - stack_base;
595 mm->arg_start = bprm->p - stack_shift;
596 bprm->p = vma->vm_end - stack_shift;
598 stack_top = arch_align_stack(stack_top);
599 stack_top = PAGE_ALIGN(stack_top);
600 stack_shift = vma->vm_end - stack_top;
602 bprm->p -= stack_shift;
603 mm->arg_start = bprm->p;
607 bprm->loader -= stack_shift;
608 bprm->exec -= stack_shift;
610 down_write(&mm->mmap_sem);
611 vm_flags = VM_STACK_FLAGS;
614 * Adjust stack execute permissions; explicitly enable for
615 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
616 * (arch default) otherwise.
618 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
620 else if (executable_stack == EXSTACK_DISABLE_X)
621 vm_flags &= ~VM_EXEC;
622 vm_flags |= mm->def_flags;
624 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
630 /* Move stack pages down in memory. */
632 ret = shift_arg_pages(vma, stack_shift);
634 up_write(&mm->mmap_sem);
639 #ifdef CONFIG_STACK_GROWSUP
640 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
642 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
644 ret = expand_stack(vma, stack_base);
649 up_write(&mm->mmap_sem);
652 EXPORT_SYMBOL(setup_arg_pages);
654 #endif /* CONFIG_MMU */
656 struct file *open_exec(const char *name)
662 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
666 struct inode *inode = nd.path.dentry->d_inode;
667 file = ERR_PTR(-EACCES);
668 if (S_ISREG(inode->i_mode)) {
669 int err = vfs_permission(&nd, MAY_EXEC);
672 file = nameidata_to_filp(&nd,
673 O_RDONLY|O_LARGEFILE);
675 err = deny_write_access(file);
685 release_open_intent(&nd);
691 EXPORT_SYMBOL(open_exec);
693 int kernel_read(struct file *file, unsigned long offset,
694 char *addr, unsigned long count)
702 /* The cast to a user pointer is valid due to the set_fs() */
703 result = vfs_read(file, (void __user *)addr, count, &pos);
708 EXPORT_SYMBOL(kernel_read);
710 static int exec_mmap(struct mm_struct *mm)
712 struct task_struct *tsk;
713 struct mm_struct * old_mm, *active_mm;
715 /* Notify parent that we're no longer interested in the old VM */
717 old_mm = current->mm;
718 mm_release(tsk, old_mm);
722 * Make sure that if there is a core dump in progress
723 * for the old mm, we get out and die instead of going
724 * through with the exec. We must hold mmap_sem around
725 * checking core_state and changing tsk->mm.
727 down_read(&old_mm->mmap_sem);
728 if (unlikely(old_mm->core_state)) {
729 up_read(&old_mm->mmap_sem);
734 active_mm = tsk->active_mm;
737 activate_mm(active_mm, mm);
739 mm_update_next_owner(old_mm);
740 arch_pick_mmap_layout(mm);
742 up_read(&old_mm->mmap_sem);
743 BUG_ON(active_mm != old_mm);
752 * This function makes sure the current process has its own signal table,
753 * so that flush_signal_handlers can later reset the handlers without
754 * disturbing other processes. (Other processes might share the signal
755 * table via the CLONE_SIGHAND option to clone().)
757 static int de_thread(struct task_struct *tsk)
759 struct signal_struct *sig = tsk->signal;
760 struct sighand_struct *oldsighand = tsk->sighand;
761 spinlock_t *lock = &oldsighand->siglock;
762 struct task_struct *leader = NULL;
765 if (thread_group_empty(tsk))
766 goto no_thread_group;
769 * Kill all other threads in the thread group.
772 if (signal_group_exit(sig)) {
774 * Another group action in progress, just
775 * return so that the signal is processed.
777 spin_unlock_irq(lock);
780 sig->group_exit_task = tsk;
781 zap_other_threads(tsk);
783 /* Account for the thread group leader hanging around: */
784 count = thread_group_leader(tsk) ? 1 : 2;
785 sig->notify_count = count;
786 while (atomic_read(&sig->count) > count) {
787 __set_current_state(TASK_UNINTERRUPTIBLE);
788 spin_unlock_irq(lock);
792 spin_unlock_irq(lock);
795 * At this point all other threads have exited, all we have to
796 * do is to wait for the thread group leader to become inactive,
797 * and to assume its PID:
799 if (!thread_group_leader(tsk)) {
800 leader = tsk->group_leader;
802 sig->notify_count = -1; /* for exit_notify() */
804 write_lock_irq(&tasklist_lock);
805 if (likely(leader->exit_state))
807 __set_current_state(TASK_UNINTERRUPTIBLE);
808 write_unlock_irq(&tasklist_lock);
812 if (unlikely(task_child_reaper(tsk) == leader))
813 task_active_pid_ns(tsk)->child_reaper = tsk;
815 * The only record we have of the real-time age of a
816 * process, regardless of execs it's done, is start_time.
817 * All the past CPU time is accumulated in signal_struct
818 * from sister threads now dead. But in this non-leader
819 * exec, nothing survives from the original leader thread,
820 * whose birth marks the true age of this process now.
821 * When we take on its identity by switching to its PID, we
822 * also take its birthdate (always earlier than our own).
824 tsk->start_time = leader->start_time;
826 BUG_ON(!same_thread_group(leader, tsk));
827 BUG_ON(has_group_leader_pid(tsk));
829 * An exec() starts a new thread group with the
830 * TGID of the previous thread group. Rehash the
831 * two threads with a switched PID, and release
832 * the former thread group leader:
835 /* Become a process group leader with the old leader's pid.
836 * The old leader becomes a thread of the this thread group.
837 * Note: The old leader also uses this pid until release_task
838 * is called. Odd but simple and correct.
840 detach_pid(tsk, PIDTYPE_PID);
841 tsk->pid = leader->pid;
842 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
843 transfer_pid(leader, tsk, PIDTYPE_PGID);
844 transfer_pid(leader, tsk, PIDTYPE_SID);
845 list_replace_rcu(&leader->tasks, &tsk->tasks);
847 tsk->group_leader = tsk;
848 leader->group_leader = tsk;
850 tsk->exit_signal = SIGCHLD;
852 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
853 leader->exit_state = EXIT_DEAD;
855 write_unlock_irq(&tasklist_lock);
858 sig->group_exit_task = NULL;
859 sig->notify_count = 0;
863 flush_itimer_signals();
865 release_task(leader);
867 if (atomic_read(&oldsighand->count) != 1) {
868 struct sighand_struct *newsighand;
870 * This ->sighand is shared with the CLONE_SIGHAND
871 * but not CLONE_THREAD task, switch to the new one.
873 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
877 atomic_set(&newsighand->count, 1);
878 memcpy(newsighand->action, oldsighand->action,
879 sizeof(newsighand->action));
881 write_lock_irq(&tasklist_lock);
882 spin_lock(&oldsighand->siglock);
883 rcu_assign_pointer(tsk->sighand, newsighand);
884 spin_unlock(&oldsighand->siglock);
885 write_unlock_irq(&tasklist_lock);
887 __cleanup_sighand(oldsighand);
890 BUG_ON(!thread_group_leader(tsk));
895 * These functions flushes out all traces of the currently running executable
896 * so that a new one can be started
898 static void flush_old_files(struct files_struct * files)
903 spin_lock(&files->file_lock);
905 unsigned long set, i;
909 fdt = files_fdtable(files);
910 if (i >= fdt->max_fds)
912 set = fdt->close_on_exec->fds_bits[j];
915 fdt->close_on_exec->fds_bits[j] = 0;
916 spin_unlock(&files->file_lock);
917 for ( ; set ; i++,set >>= 1) {
922 spin_lock(&files->file_lock);
925 spin_unlock(&files->file_lock);
928 char *get_task_comm(char *buf, struct task_struct *tsk)
930 /* buf must be at least sizeof(tsk->comm) in size */
932 strncpy(buf, tsk->comm, sizeof(tsk->comm));
937 void set_task_comm(struct task_struct *tsk, char *buf)
940 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
944 int flush_old_exec(struct linux_binprm * bprm)
948 char tcomm[sizeof(current->comm)];
951 * Make sure we have a private signal table and that
952 * we are unassociated from the previous thread group.
954 retval = de_thread(current);
958 set_mm_exe_file(bprm->mm, bprm->file);
961 * Release all of the old mmap stuff
963 retval = exec_mmap(bprm->mm);
967 bprm->mm = NULL; /* We're using it now */
969 /* This is the point of no return */
970 current->sas_ss_sp = current->sas_ss_size = 0;
972 if (current->euid == current->uid && current->egid == current->gid)
973 set_dumpable(current->mm, 1);
975 set_dumpable(current->mm, suid_dumpable);
977 name = bprm->filename;
979 /* Copies the binary name from after last slash */
980 for (i=0; (ch = *(name++)) != '\0';) {
982 i = 0; /* overwrite what we wrote */
984 if (i < (sizeof(tcomm) - 1))
988 set_task_comm(current, tcomm);
990 current->flags &= ~PF_RANDOMIZE;
993 /* Set the new mm task size. We have to do that late because it may
994 * depend on TIF_32BIT which is only updated in flush_thread() on
995 * some architectures like powerpc
997 current->mm->task_size = TASK_SIZE;
999 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1001 set_dumpable(current->mm, suid_dumpable);
1002 current->pdeath_signal = 0;
1003 } else if (file_permission(bprm->file, MAY_READ) ||
1004 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1006 set_dumpable(current->mm, suid_dumpable);
1009 /* An exec changes our domain. We are no longer part of the thread
1012 current->self_exec_id++;
1014 flush_signal_handlers(current, 0);
1015 flush_old_files(current->files);
1023 EXPORT_SYMBOL(flush_old_exec);
1026 * Fill the binprm structure from the inode.
1027 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1029 int prepare_binprm(struct linux_binprm *bprm)
1032 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1035 mode = inode->i_mode;
1036 if (bprm->file->f_op == NULL)
1039 bprm->e_uid = current->euid;
1040 bprm->e_gid = current->egid;
1042 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1044 if (mode & S_ISUID) {
1045 current->personality &= ~PER_CLEAR_ON_SETID;
1046 bprm->e_uid = inode->i_uid;
1051 * If setgid is set but no group execute bit then this
1052 * is a candidate for mandatory locking, not a setgid
1055 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1056 current->personality &= ~PER_CLEAR_ON_SETID;
1057 bprm->e_gid = inode->i_gid;
1061 /* fill in binprm security blob */
1062 retval = security_bprm_set(bprm);
1066 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1067 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1070 EXPORT_SYMBOL(prepare_binprm);
1072 static int unsafe_exec(struct task_struct *p)
1074 int unsafe = tracehook_unsafe_exec(p);
1076 if (atomic_read(&p->fs->count) > 1 ||
1077 atomic_read(&p->files->count) > 1 ||
1078 atomic_read(&p->sighand->count) > 1)
1079 unsafe |= LSM_UNSAFE_SHARE;
1084 void compute_creds(struct linux_binprm *bprm)
1088 if (bprm->e_uid != current->uid) {
1090 current->pdeath_signal = 0;
1095 unsafe = unsafe_exec(current);
1096 security_bprm_apply_creds(bprm, unsafe);
1097 task_unlock(current);
1098 security_bprm_post_apply_creds(bprm);
1100 EXPORT_SYMBOL(compute_creds);
1103 * Arguments are '\0' separated strings found at the location bprm->p
1104 * points to; chop off the first by relocating brpm->p to right after
1105 * the first '\0' encountered.
1107 int remove_arg_zero(struct linux_binprm *bprm)
1110 unsigned long offset;
1118 offset = bprm->p & ~PAGE_MASK;
1119 page = get_arg_page(bprm, bprm->p, 0);
1124 kaddr = kmap_atomic(page, KM_USER0);
1126 for (; offset < PAGE_SIZE && kaddr[offset];
1127 offset++, bprm->p++)
1130 kunmap_atomic(kaddr, KM_USER0);
1133 if (offset == PAGE_SIZE)
1134 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1135 } while (offset == PAGE_SIZE);
1144 EXPORT_SYMBOL(remove_arg_zero);
1147 * cycle the list of binary formats handler, until one recognizes the image
1149 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1152 struct linux_binfmt *fmt;
1154 /* handle /sbin/loader.. */
1156 struct exec * eh = (struct exec *) bprm->buf;
1158 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1159 (eh->fh.f_flags & 0x3000) == 0x3000)
1162 unsigned long loader;
1164 allow_write_access(bprm->file);
1168 loader = bprm->vma->vm_end - sizeof(void *);
1170 file = open_exec("/sbin/loader");
1171 retval = PTR_ERR(file);
1175 /* Remember if the application is TASO. */
1176 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1179 bprm->loader = loader;
1180 retval = prepare_binprm(bprm);
1183 /* should call search_binary_handler recursively here,
1184 but it does not matter */
1188 retval = security_bprm_check(bprm);
1192 /* kernel module loader fixup */
1193 /* so we don't try to load run modprobe in kernel space. */
1196 retval = audit_bprm(bprm);
1201 for (try=0; try<2; try++) {
1202 read_lock(&binfmt_lock);
1203 list_for_each_entry(fmt, &formats, lh) {
1204 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1207 if (!try_module_get(fmt->module))
1209 read_unlock(&binfmt_lock);
1210 retval = fn(bprm, regs);
1212 tracehook_report_exec(fmt, bprm, regs);
1214 allow_write_access(bprm->file);
1218 current->did_exec = 1;
1219 proc_exec_connector(current);
1222 read_lock(&binfmt_lock);
1224 if (retval != -ENOEXEC || bprm->mm == NULL)
1227 read_unlock(&binfmt_lock);
1231 read_unlock(&binfmt_lock);
1232 if (retval != -ENOEXEC || bprm->mm == NULL) {
1236 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1237 if (printable(bprm->buf[0]) &&
1238 printable(bprm->buf[1]) &&
1239 printable(bprm->buf[2]) &&
1240 printable(bprm->buf[3]))
1241 break; /* -ENOEXEC */
1242 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1249 EXPORT_SYMBOL(search_binary_handler);
1251 void free_bprm(struct linux_binprm *bprm)
1253 free_arg_pages(bprm);
1258 * sys_execve() executes a new program.
1260 int do_execve(char * filename,
1261 char __user *__user *argv,
1262 char __user *__user *envp,
1263 struct pt_regs * regs)
1265 struct linux_binprm *bprm;
1267 struct files_struct *displaced;
1270 retval = unshare_files(&displaced);
1275 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1279 file = open_exec(filename);
1280 retval = PTR_ERR(file);
1287 bprm->filename = filename;
1288 bprm->interp = filename;
1290 retval = bprm_mm_init(bprm);
1294 bprm->argc = count(argv, MAX_ARG_STRINGS);
1295 if ((retval = bprm->argc) < 0)
1298 bprm->envc = count(envp, MAX_ARG_STRINGS);
1299 if ((retval = bprm->envc) < 0)
1302 retval = security_bprm_alloc(bprm);
1306 retval = prepare_binprm(bprm);
1310 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1314 bprm->exec = bprm->p;
1315 retval = copy_strings(bprm->envc, envp, bprm);
1319 retval = copy_strings(bprm->argc, argv, bprm);
1323 current->flags &= ~PF_KTHREAD;
1324 retval = search_binary_handler(bprm,regs);
1326 /* execve success */
1327 security_bprm_free(bprm);
1328 acct_update_integrals(current);
1331 put_files_struct(displaced);
1337 security_bprm_free(bprm);
1345 allow_write_access(bprm->file);
1353 reset_files_struct(displaced);
1358 int set_binfmt(struct linux_binfmt *new)
1360 struct linux_binfmt *old = current->binfmt;
1363 if (!try_module_get(new->module))
1366 current->binfmt = new;
1368 module_put(old->module);
1372 EXPORT_SYMBOL(set_binfmt);
1374 /* format_corename will inspect the pattern parameter, and output a
1375 * name into corename, which must have space for at least
1376 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1378 static int format_corename(char *corename, int nr_threads, long signr)
1380 const char *pat_ptr = core_pattern;
1381 int ispipe = (*pat_ptr == '|');
1382 char *out_ptr = corename;
1383 char *const out_end = corename + CORENAME_MAX_SIZE;
1385 int pid_in_pattern = 0;
1387 /* Repeat as long as we have more pattern to process and more output
1390 if (*pat_ptr != '%') {
1391 if (out_ptr == out_end)
1393 *out_ptr++ = *pat_ptr++;
1395 switch (*++pat_ptr) {
1398 /* Double percent, output one percent */
1400 if (out_ptr == out_end)
1407 rc = snprintf(out_ptr, out_end - out_ptr,
1408 "%d", task_tgid_vnr(current));
1409 if (rc > out_end - out_ptr)
1415 rc = snprintf(out_ptr, out_end - out_ptr,
1416 "%d", current->uid);
1417 if (rc > out_end - out_ptr)
1423 rc = snprintf(out_ptr, out_end - out_ptr,
1424 "%d", current->gid);
1425 if (rc > out_end - out_ptr)
1429 /* signal that caused the coredump */
1431 rc = snprintf(out_ptr, out_end - out_ptr,
1433 if (rc > out_end - out_ptr)
1437 /* UNIX time of coredump */
1440 do_gettimeofday(&tv);
1441 rc = snprintf(out_ptr, out_end - out_ptr,
1443 if (rc > out_end - out_ptr)
1450 down_read(&uts_sem);
1451 rc = snprintf(out_ptr, out_end - out_ptr,
1452 "%s", utsname()->nodename);
1454 if (rc > out_end - out_ptr)
1460 rc = snprintf(out_ptr, out_end - out_ptr,
1461 "%s", current->comm);
1462 if (rc > out_end - out_ptr)
1466 /* core limit size */
1468 rc = snprintf(out_ptr, out_end - out_ptr,
1469 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1470 if (rc > out_end - out_ptr)
1480 /* Backward compatibility with core_uses_pid:
1482 * If core_pattern does not include a %p (as is the default)
1483 * and core_uses_pid is set, then .%pid will be appended to
1484 * the filename. Do not do this for piped commands. */
1485 if (!ispipe && !pid_in_pattern
1486 && (core_uses_pid || nr_threads)) {
1487 rc = snprintf(out_ptr, out_end - out_ptr,
1488 ".%d", task_tgid_vnr(current));
1489 if (rc > out_end - out_ptr)
1498 static int zap_process(struct task_struct *start)
1500 struct task_struct *t;
1503 start->signal->flags = SIGNAL_GROUP_EXIT;
1504 start->signal->group_stop_count = 0;
1508 if (t != current && t->mm) {
1509 sigaddset(&t->pending.signal, SIGKILL);
1510 signal_wake_up(t, 1);
1513 } while_each_thread(start, t);
1518 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1519 struct core_state *core_state, int exit_code)
1521 struct task_struct *g, *p;
1522 unsigned long flags;
1525 spin_lock_irq(&tsk->sighand->siglock);
1526 if (!signal_group_exit(tsk->signal)) {
1527 mm->core_state = core_state;
1528 tsk->signal->group_exit_code = exit_code;
1529 nr = zap_process(tsk);
1531 spin_unlock_irq(&tsk->sighand->siglock);
1532 if (unlikely(nr < 0))
1535 if (atomic_read(&mm->mm_users) == nr + 1)
1538 * We should find and kill all tasks which use this mm, and we should
1539 * count them correctly into ->nr_threads. We don't take tasklist
1540 * lock, but this is safe wrt:
1543 * None of sub-threads can fork after zap_process(leader). All
1544 * processes which were created before this point should be
1545 * visible to zap_threads() because copy_process() adds the new
1546 * process to the tail of init_task.tasks list, and lock/unlock
1547 * of ->siglock provides a memory barrier.
1550 * The caller holds mm->mmap_sem. This means that the task which
1551 * uses this mm can't pass exit_mm(), so it can't exit or clear
1555 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1556 * we must see either old or new leader, this does not matter.
1557 * However, it can change p->sighand, so lock_task_sighand(p)
1558 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1561 * Note also that "g" can be the old leader with ->mm == NULL
1562 * and already unhashed and thus removed from ->thread_group.
1563 * This is OK, __unhash_process()->list_del_rcu() does not
1564 * clear the ->next pointer, we will find the new leader via
1568 for_each_process(g) {
1569 if (g == tsk->group_leader)
1571 if (g->flags & PF_KTHREAD)
1576 if (unlikely(p->mm == mm)) {
1577 lock_task_sighand(p, &flags);
1578 nr += zap_process(p);
1579 unlock_task_sighand(p, &flags);
1583 } while_each_thread(g, p);
1587 atomic_set(&core_state->nr_threads, nr);
1591 static int coredump_wait(int exit_code, struct core_state *core_state)
1593 struct task_struct *tsk = current;
1594 struct mm_struct *mm = tsk->mm;
1595 struct completion *vfork_done;
1598 init_completion(&core_state->startup);
1599 core_state->dumper.task = tsk;
1600 core_state->dumper.next = NULL;
1601 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1602 up_write(&mm->mmap_sem);
1604 if (unlikely(core_waiters < 0))
1608 * Make sure nobody is waiting for us to release the VM,
1609 * otherwise we can deadlock when we wait on each other
1611 vfork_done = tsk->vfork_done;
1613 tsk->vfork_done = NULL;
1614 complete(vfork_done);
1618 wait_for_completion(&core_state->startup);
1620 return core_waiters;
1623 static void coredump_finish(struct mm_struct *mm)
1625 struct core_thread *curr, *next;
1626 struct task_struct *task;
1628 next = mm->core_state->dumper.next;
1629 while ((curr = next) != NULL) {
1633 * see exit_mm(), curr->task must not see
1634 * ->task == NULL before we read ->next.
1638 wake_up_process(task);
1641 mm->core_state = NULL;
1645 * set_dumpable converts traditional three-value dumpable to two flags and
1646 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1647 * these bits are not changed atomically. So get_dumpable can observe the
1648 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1649 * return either old dumpable or new one by paying attention to the order of
1650 * modifying the bits.
1652 * dumpable | mm->flags (binary)
1653 * old new | initial interim final
1654 * ---------+-----------------------
1662 * (*) get_dumpable regards interim value of 10 as 11.
1664 void set_dumpable(struct mm_struct *mm, int value)
1668 clear_bit(MMF_DUMPABLE, &mm->flags);
1670 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1673 set_bit(MMF_DUMPABLE, &mm->flags);
1675 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1678 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1680 set_bit(MMF_DUMPABLE, &mm->flags);
1685 int get_dumpable(struct mm_struct *mm)
1689 ret = mm->flags & 0x3;
1690 return (ret >= 2) ? 2 : ret;
1693 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1695 struct core_state core_state;
1696 char corename[CORENAME_MAX_SIZE + 1];
1697 struct mm_struct *mm = current->mm;
1698 struct linux_binfmt * binfmt;
1699 struct inode * inode;
1702 int fsuid = current->fsuid;
1705 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1706 char **helper_argv = NULL;
1707 int helper_argc = 0;
1710 audit_core_dumps(signr);
1712 binfmt = current->binfmt;
1713 if (!binfmt || !binfmt->core_dump)
1715 down_write(&mm->mmap_sem);
1717 * If another thread got here first, or we are not dumpable, bail out.
1719 if (mm->core_state || !get_dumpable(mm)) {
1720 up_write(&mm->mmap_sem);
1725 * We cannot trust fsuid as being the "true" uid of the
1726 * process nor do we know its entire history. We only know it
1727 * was tainted so we dump it as root in mode 2.
1729 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1730 flag = O_EXCL; /* Stop rewrite attacks */
1731 current->fsuid = 0; /* Dump root private */
1734 retval = coredump_wait(exit_code, &core_state);
1739 * Clear any false indication of pending signals that might
1740 * be seen by the filesystem code called to write the core file.
1742 clear_thread_flag(TIF_SIGPENDING);
1745 * lock_kernel() because format_corename() is controlled by sysctl, which
1746 * uses lock_kernel()
1749 ispipe = format_corename(corename, retval, signr);
1752 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1753 * to a pipe. Since we're not writing directly to the filesystem
1754 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1755 * created unless the pipe reader choses to write out the core file
1756 * at which point file size limits and permissions will be imposed
1757 * as it does with any other process
1759 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1763 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1764 /* Terminate the string before the first option */
1765 delimit = strchr(corename, ' ');
1768 delimit = strrchr(helper_argv[0], '/');
1772 delimit = helper_argv[0];
1773 if (!strcmp(delimit, current->comm)) {
1774 printk(KERN_NOTICE "Recursive core dump detected, "
1779 core_limit = RLIM_INFINITY;
1781 /* SIGPIPE can happen, but it's just never processed */
1782 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1784 printk(KERN_INFO "Core dump to %s pipe failed\n",
1789 file = filp_open(corename,
1790 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1794 inode = file->f_path.dentry->d_inode;
1795 if (inode->i_nlink > 1)
1796 goto close_fail; /* multiple links - don't dump */
1797 if (!ispipe && d_unhashed(file->f_path.dentry))
1800 /* AK: actually i see no reason to not allow this for named pipes etc.,
1801 but keep the previous behaviour for now. */
1802 if (!ispipe && !S_ISREG(inode->i_mode))
1805 * Dont allow local users get cute and trick others to coredump
1806 * into their pre-created files:
1808 if (inode->i_uid != current->fsuid)
1812 if (!file->f_op->write)
1814 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1817 retval = binfmt->core_dump(signr, regs, file, core_limit);
1820 current->signal->group_exit_code |= 0x80;
1822 filp_close(file, NULL);
1825 argv_free(helper_argv);
1827 current->fsuid = fsuid;
1828 coredump_finish(mm);