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/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.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/swap.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/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
60 #include <linux/kmod.h>
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
67 /* The maximal length of core_pattern is also specified in sysctl.c */
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
72 int register_binfmt(struct linux_binfmt * fmt)
76 write_lock(&binfmt_lock);
77 list_add(&fmt->lh, &formats);
78 write_unlock(&binfmt_lock);
82 EXPORT_SYMBOL(register_binfmt);
84 void unregister_binfmt(struct linux_binfmt * fmt)
86 write_lock(&binfmt_lock);
88 write_unlock(&binfmt_lock);
91 EXPORT_SYMBOL(unregister_binfmt);
93 static inline void put_binfmt(struct linux_binfmt * fmt)
95 module_put(fmt->module);
99 * Note that a shared library must be both readable and executable due to
102 * Also note that we take the address to load from from the file itself.
104 asmlinkage long sys_uselib(const char __user * library)
110 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
115 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
118 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
122 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
123 error = PTR_ERR(file);
129 struct linux_binfmt * fmt;
131 read_lock(&binfmt_lock);
132 list_for_each_entry(fmt, &formats, lh) {
133 if (!fmt->load_shlib)
135 if (!try_module_get(fmt->module))
137 read_unlock(&binfmt_lock);
138 error = fmt->load_shlib(file);
139 read_lock(&binfmt_lock);
141 if (error != -ENOEXEC)
144 read_unlock(&binfmt_lock);
150 release_open_intent(&nd);
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
163 #ifdef CONFIG_STACK_GROWSUP
165 ret = expand_stack_downwards(bprm->vma, pos);
170 ret = get_user_pages(current, bprm->mm, pos,
171 1, write, 1, &page, NULL);
176 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
180 * We've historically supported up to 32 pages (ARG_MAX)
181 * of argument strings even with small stacks
187 * Limit to 1/4-th the stack size for the argv+env strings.
189 * - the remaining binfmt code will not run out of stack space,
190 * - the program will have a reasonable amount of stack left
193 rlim = current->signal->rlim;
194 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
203 static void put_arg_page(struct page *page)
208 static void free_arg_page(struct linux_binprm *bprm, int i)
212 static void free_arg_pages(struct linux_binprm *bprm)
216 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
219 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
222 static int __bprm_mm_init(struct linux_binprm *bprm)
225 struct vm_area_struct *vma = NULL;
226 struct mm_struct *mm = bprm->mm;
228 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
232 down_write(&mm->mmap_sem);
236 * Place the stack at the largest stack address the architecture
237 * supports. Later, we'll move this to an appropriate place. We don't
238 * use STACK_TOP because that can depend on attributes which aren't
241 vma->vm_end = STACK_TOP_MAX;
242 vma->vm_start = vma->vm_end - PAGE_SIZE;
244 vma->vm_flags = VM_STACK_FLAGS;
245 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
246 err = insert_vm_struct(mm, vma);
248 up_write(&mm->mmap_sem);
252 mm->stack_vm = mm->total_vm = 1;
253 up_write(&mm->mmap_sem);
255 bprm->p = vma->vm_end - sizeof(void *);
262 kmem_cache_free(vm_area_cachep, vma);
268 static bool valid_arg_len(struct linux_binprm *bprm, long len)
270 return len <= MAX_ARG_STRLEN;
275 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
280 page = bprm->page[pos / PAGE_SIZE];
281 if (!page && write) {
282 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
285 bprm->page[pos / PAGE_SIZE] = page;
291 static void put_arg_page(struct page *page)
295 static void free_arg_page(struct linux_binprm *bprm, int i)
298 __free_page(bprm->page[i]);
299 bprm->page[i] = NULL;
303 static void free_arg_pages(struct linux_binprm *bprm)
307 for (i = 0; i < MAX_ARG_PAGES; i++)
308 free_arg_page(bprm, i);
311 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
316 static int __bprm_mm_init(struct linux_binprm *bprm)
318 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
322 static bool valid_arg_len(struct linux_binprm *bprm, long len)
324 return len <= bprm->p;
327 #endif /* CONFIG_MMU */
330 * Create a new mm_struct and populate it with a temporary stack
331 * vm_area_struct. We don't have enough context at this point to set the stack
332 * flags, permissions, and offset, so we use temporary values. We'll update
333 * them later in setup_arg_pages().
335 int bprm_mm_init(struct linux_binprm *bprm)
338 struct mm_struct *mm = NULL;
340 bprm->mm = mm = mm_alloc();
345 err = init_new_context(current, mm);
349 err = __bprm_mm_init(bprm);
365 * count() counts the number of strings in array ARGV.
367 static int count(char __user * __user * argv, int max)
375 if (get_user(p, argv))
389 * 'copy_strings()' copies argument/environment strings from the old
390 * processes's memory to the new process's stack. The call to get_user_pages()
391 * ensures the destination page is created and not swapped out.
393 static int copy_strings(int argc, char __user * __user * argv,
394 struct linux_binprm *bprm)
396 struct page *kmapped_page = NULL;
398 unsigned long kpos = 0;
406 if (get_user(str, argv+argc) ||
407 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
412 if (!valid_arg_len(bprm, len)) {
417 /* We're going to work our way backwords. */
423 int offset, bytes_to_copy;
425 offset = pos % PAGE_SIZE;
429 bytes_to_copy = offset;
430 if (bytes_to_copy > len)
433 offset -= bytes_to_copy;
434 pos -= bytes_to_copy;
435 str -= bytes_to_copy;
436 len -= bytes_to_copy;
438 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
441 page = get_arg_page(bprm, pos, 1);
448 flush_kernel_dcache_page(kmapped_page);
449 kunmap(kmapped_page);
450 put_arg_page(kmapped_page);
453 kaddr = kmap(kmapped_page);
454 kpos = pos & PAGE_MASK;
455 flush_arg_page(bprm, kpos, kmapped_page);
457 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
466 flush_kernel_dcache_page(kmapped_page);
467 kunmap(kmapped_page);
468 put_arg_page(kmapped_page);
474 * Like copy_strings, but get argv and its values from kernel memory.
476 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
479 mm_segment_t oldfs = get_fs();
481 r = copy_strings(argc, (char __user * __user *)argv, bprm);
485 EXPORT_SYMBOL(copy_strings_kernel);
490 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
491 * the binfmt code determines where the new stack should reside, we shift it to
492 * its final location. The process proceeds as follows:
494 * 1) Use shift to calculate the new vma endpoints.
495 * 2) Extend vma to cover both the old and new ranges. This ensures the
496 * arguments passed to subsequent functions are consistent.
497 * 3) Move vma's page tables to the new range.
498 * 4) Free up any cleared pgd range.
499 * 5) Shrink the vma to cover only the new range.
501 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
503 struct mm_struct *mm = vma->vm_mm;
504 unsigned long old_start = vma->vm_start;
505 unsigned long old_end = vma->vm_end;
506 unsigned long length = old_end - old_start;
507 unsigned long new_start = old_start - shift;
508 unsigned long new_end = old_end - shift;
509 struct mmu_gather *tlb;
511 BUG_ON(new_start > new_end);
514 * ensure there are no vmas between where we want to go
517 if (vma != find_vma(mm, new_start))
521 * cover the whole range: [new_start, old_end)
523 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
526 * move the page tables downwards, on failure we rely on
527 * process cleanup to remove whatever mess we made.
529 if (length != move_page_tables(vma, old_start,
530 vma, new_start, length))
534 tlb = tlb_gather_mmu(mm, 0);
535 if (new_end > old_start) {
537 * when the old and new regions overlap clear from new_end.
539 free_pgd_range(&tlb, new_end, old_end, new_end,
540 vma->vm_next ? vma->vm_next->vm_start : 0);
543 * otherwise, clean from old_start; this is done to not touch
544 * the address space in [new_end, old_start) some architectures
545 * have constraints on va-space that make this illegal (IA64) -
546 * for the others its just a little faster.
548 free_pgd_range(&tlb, old_start, old_end, new_end,
549 vma->vm_next ? vma->vm_next->vm_start : 0);
551 tlb_finish_mmu(tlb, new_end, old_end);
554 * shrink the vma to just the new range.
556 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
561 #define EXTRA_STACK_VM_PAGES 20 /* random */
564 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
565 * the stack is optionally relocated, and some extra space is added.
567 int setup_arg_pages(struct linux_binprm *bprm,
568 unsigned long stack_top,
569 int executable_stack)
572 unsigned long stack_shift;
573 struct mm_struct *mm = current->mm;
574 struct vm_area_struct *vma = bprm->vma;
575 struct vm_area_struct *prev = NULL;
576 unsigned long vm_flags;
577 unsigned long stack_base;
579 #ifdef CONFIG_STACK_GROWSUP
580 /* Limit stack size to 1GB */
581 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
582 if (stack_base > (1 << 30))
583 stack_base = 1 << 30;
585 /* Make sure we didn't let the argument array grow too large. */
586 if (vma->vm_end - vma->vm_start > stack_base)
589 stack_base = PAGE_ALIGN(stack_top - stack_base);
591 stack_shift = vma->vm_start - stack_base;
592 mm->arg_start = bprm->p - stack_shift;
593 bprm->p = vma->vm_end - stack_shift;
595 stack_top = arch_align_stack(stack_top);
596 stack_top = PAGE_ALIGN(stack_top);
597 stack_shift = vma->vm_end - stack_top;
599 bprm->p -= stack_shift;
600 mm->arg_start = bprm->p;
604 bprm->loader -= stack_shift;
605 bprm->exec -= stack_shift;
607 down_write(&mm->mmap_sem);
608 vm_flags = vma->vm_flags;
611 * Adjust stack execute permissions; explicitly enable for
612 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
613 * (arch default) otherwise.
615 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
617 else if (executable_stack == EXSTACK_DISABLE_X)
618 vm_flags &= ~VM_EXEC;
619 vm_flags |= mm->def_flags;
621 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
627 /* Move stack pages down in memory. */
629 ret = shift_arg_pages(vma, stack_shift);
631 up_write(&mm->mmap_sem);
636 #ifdef CONFIG_STACK_GROWSUP
637 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
639 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
641 ret = expand_stack(vma, stack_base);
646 up_write(&mm->mmap_sem);
649 EXPORT_SYMBOL(setup_arg_pages);
651 #endif /* CONFIG_MMU */
653 struct file *open_exec(const char *name)
659 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
663 struct inode *inode = nd.path.dentry->d_inode;
664 file = ERR_PTR(-EACCES);
665 if (S_ISREG(inode->i_mode)) {
666 int err = vfs_permission(&nd, MAY_EXEC);
669 file = nameidata_to_filp(&nd,
670 O_RDONLY|O_LARGEFILE);
672 err = deny_write_access(file);
682 release_open_intent(&nd);
688 EXPORT_SYMBOL(open_exec);
690 int kernel_read(struct file *file, unsigned long offset,
691 char *addr, unsigned long count)
699 /* The cast to a user pointer is valid due to the set_fs() */
700 result = vfs_read(file, (void __user *)addr, count, &pos);
705 EXPORT_SYMBOL(kernel_read);
707 static int exec_mmap(struct mm_struct *mm)
709 struct task_struct *tsk;
710 struct mm_struct * old_mm, *active_mm;
712 /* Notify parent that we're no longer interested in the old VM */
714 old_mm = current->mm;
715 mm_release(tsk, old_mm);
719 * Make sure that if there is a core dump in progress
720 * for the old mm, we get out and die instead of going
721 * through with the exec. We must hold mmap_sem around
722 * checking core_waiters and changing tsk->mm. The
723 * core-inducing thread will increment core_waiters for
724 * each thread whose ->mm == old_mm.
726 down_read(&old_mm->mmap_sem);
727 if (unlikely(old_mm->core_waiters)) {
728 up_read(&old_mm->mmap_sem);
733 active_mm = tsk->active_mm;
736 activate_mm(active_mm, mm);
738 mm_update_next_owner(mm);
739 arch_pick_mmap_layout(mm);
741 up_read(&old_mm->mmap_sem);
742 BUG_ON(active_mm != old_mm);
751 * This function makes sure the current process has its own signal table,
752 * so that flush_signal_handlers can later reset the handlers without
753 * disturbing other processes. (Other processes might share the signal
754 * table via the CLONE_SIGHAND option to clone().)
756 static int de_thread(struct task_struct *tsk)
758 struct signal_struct *sig = tsk->signal;
759 struct sighand_struct *oldsighand = tsk->sighand;
760 spinlock_t *lock = &oldsighand->siglock;
761 struct task_struct *leader = NULL;
764 if (thread_group_empty(tsk))
765 goto no_thread_group;
768 * Kill all other threads in the thread group.
771 if (signal_group_exit(sig)) {
773 * Another group action in progress, just
774 * return so that the signal is processed.
776 spin_unlock_irq(lock);
779 sig->group_exit_task = tsk;
780 zap_other_threads(tsk);
782 /* Account for the thread group leader hanging around: */
783 count = thread_group_leader(tsk) ? 1 : 2;
784 sig->notify_count = count;
785 while (atomic_read(&sig->count) > count) {
786 __set_current_state(TASK_UNINTERRUPTIBLE);
787 spin_unlock_irq(lock);
791 spin_unlock_irq(lock);
794 * At this point all other threads have exited, all we have to
795 * do is to wait for the thread group leader to become inactive,
796 * and to assume its PID:
798 if (!thread_group_leader(tsk)) {
799 leader = tsk->group_leader;
801 sig->notify_count = -1; /* for exit_notify() */
803 write_lock_irq(&tasklist_lock);
804 if (likely(leader->exit_state))
806 __set_current_state(TASK_UNINTERRUPTIBLE);
807 write_unlock_irq(&tasklist_lock);
811 if (unlikely(task_child_reaper(tsk) == leader))
812 task_active_pid_ns(tsk)->child_reaper = tsk;
814 * The only record we have of the real-time age of a
815 * process, regardless of execs it's done, is start_time.
816 * All the past CPU time is accumulated in signal_struct
817 * from sister threads now dead. But in this non-leader
818 * exec, nothing survives from the original leader thread,
819 * whose birth marks the true age of this process now.
820 * When we take on its identity by switching to its PID, we
821 * also take its birthdate (always earlier than our own).
823 tsk->start_time = leader->start_time;
825 BUG_ON(!same_thread_group(leader, tsk));
826 BUG_ON(has_group_leader_pid(tsk));
828 * An exec() starts a new thread group with the
829 * TGID of the previous thread group. Rehash the
830 * two threads with a switched PID, and release
831 * the former thread group leader:
834 /* Become a process group leader with the old leader's pid.
835 * The old leader becomes a thread of the this thread group.
836 * Note: The old leader also uses this pid until release_task
837 * is called. Odd but simple and correct.
839 detach_pid(tsk, PIDTYPE_PID);
840 tsk->pid = leader->pid;
841 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
842 transfer_pid(leader, tsk, PIDTYPE_PGID);
843 transfer_pid(leader, tsk, PIDTYPE_SID);
844 list_replace_rcu(&leader->tasks, &tsk->tasks);
846 tsk->group_leader = tsk;
847 leader->group_leader = tsk;
849 tsk->exit_signal = SIGCHLD;
851 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
852 leader->exit_state = EXIT_DEAD;
854 write_unlock_irq(&tasklist_lock);
857 sig->group_exit_task = NULL;
858 sig->notify_count = 0;
863 release_task(leader);
865 if (atomic_read(&oldsighand->count) != 1) {
866 struct sighand_struct *newsighand;
868 * This ->sighand is shared with the CLONE_SIGHAND
869 * but not CLONE_THREAD task, switch to the new one.
871 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
875 atomic_set(&newsighand->count, 1);
876 memcpy(newsighand->action, oldsighand->action,
877 sizeof(newsighand->action));
879 write_lock_irq(&tasklist_lock);
880 spin_lock(&oldsighand->siglock);
881 rcu_assign_pointer(tsk->sighand, newsighand);
882 spin_unlock(&oldsighand->siglock);
883 write_unlock_irq(&tasklist_lock);
885 __cleanup_sighand(oldsighand);
888 BUG_ON(!thread_group_leader(tsk));
893 * These functions flushes out all traces of the currently running executable
894 * so that a new one can be started
896 static void flush_old_files(struct files_struct * files)
901 spin_lock(&files->file_lock);
903 unsigned long set, i;
907 fdt = files_fdtable(files);
908 if (i >= fdt->max_fds)
910 set = fdt->close_on_exec->fds_bits[j];
913 fdt->close_on_exec->fds_bits[j] = 0;
914 spin_unlock(&files->file_lock);
915 for ( ; set ; i++,set >>= 1) {
920 spin_lock(&files->file_lock);
923 spin_unlock(&files->file_lock);
926 char *get_task_comm(char *buf, struct task_struct *tsk)
928 /* buf must be at least sizeof(tsk->comm) in size */
930 strncpy(buf, tsk->comm, sizeof(tsk->comm));
935 void set_task_comm(struct task_struct *tsk, char *buf)
938 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
942 int flush_old_exec(struct linux_binprm * bprm)
946 char tcomm[sizeof(current->comm)];
949 * Make sure we have a private signal table and that
950 * we are unassociated from the previous thread group.
952 retval = de_thread(current);
956 set_mm_exe_file(bprm->mm, bprm->file);
959 * Release all of the old mmap stuff
961 retval = exec_mmap(bprm->mm);
965 bprm->mm = NULL; /* We're using it now */
967 /* This is the point of no return */
968 current->sas_ss_sp = current->sas_ss_size = 0;
970 if (current->euid == current->uid && current->egid == current->gid)
971 set_dumpable(current->mm, 1);
973 set_dumpable(current->mm, suid_dumpable);
975 name = bprm->filename;
977 /* Copies the binary name from after last slash */
978 for (i=0; (ch = *(name++)) != '\0';) {
980 i = 0; /* overwrite what we wrote */
982 if (i < (sizeof(tcomm) - 1))
986 set_task_comm(current, tcomm);
988 current->flags &= ~PF_RANDOMIZE;
991 /* Set the new mm task size. We have to do that late because it may
992 * depend on TIF_32BIT which is only updated in flush_thread() on
993 * some architectures like powerpc
995 current->mm->task_size = TASK_SIZE;
997 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
999 set_dumpable(current->mm, suid_dumpable);
1000 current->pdeath_signal = 0;
1001 } else if (file_permission(bprm->file, MAY_READ) ||
1002 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1004 set_dumpable(current->mm, suid_dumpable);
1007 /* An exec changes our domain. We are no longer part of the thread
1010 current->self_exec_id++;
1012 flush_signal_handlers(current, 0);
1013 flush_old_files(current->files);
1021 EXPORT_SYMBOL(flush_old_exec);
1024 * Fill the binprm structure from the inode.
1025 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1027 int prepare_binprm(struct linux_binprm *bprm)
1030 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1033 mode = inode->i_mode;
1034 if (bprm->file->f_op == NULL)
1037 bprm->e_uid = current->euid;
1038 bprm->e_gid = current->egid;
1040 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1042 if (mode & S_ISUID) {
1043 current->personality &= ~PER_CLEAR_ON_SETID;
1044 bprm->e_uid = inode->i_uid;
1049 * If setgid is set but no group execute bit then this
1050 * is a candidate for mandatory locking, not a setgid
1053 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1054 current->personality &= ~PER_CLEAR_ON_SETID;
1055 bprm->e_gid = inode->i_gid;
1059 /* fill in binprm security blob */
1060 retval = security_bprm_set(bprm);
1064 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1065 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1068 EXPORT_SYMBOL(prepare_binprm);
1070 static int unsafe_exec(struct task_struct *p)
1073 if (p->ptrace & PT_PTRACED) {
1074 if (p->ptrace & PT_PTRACE_CAP)
1075 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1077 unsafe |= LSM_UNSAFE_PTRACE;
1079 if (atomic_read(&p->fs->count) > 1 ||
1080 atomic_read(&p->files->count) > 1 ||
1081 atomic_read(&p->sighand->count) > 1)
1082 unsafe |= LSM_UNSAFE_SHARE;
1087 void compute_creds(struct linux_binprm *bprm)
1091 if (bprm->e_uid != current->uid) {
1093 current->pdeath_signal = 0;
1098 unsafe = unsafe_exec(current);
1099 security_bprm_apply_creds(bprm, unsafe);
1100 task_unlock(current);
1101 security_bprm_post_apply_creds(bprm);
1103 EXPORT_SYMBOL(compute_creds);
1106 * Arguments are '\0' separated strings found at the location bprm->p
1107 * points to; chop off the first by relocating brpm->p to right after
1108 * the first '\0' encountered.
1110 int remove_arg_zero(struct linux_binprm *bprm)
1113 unsigned long offset;
1121 offset = bprm->p & ~PAGE_MASK;
1122 page = get_arg_page(bprm, bprm->p, 0);
1127 kaddr = kmap_atomic(page, KM_USER0);
1129 for (; offset < PAGE_SIZE && kaddr[offset];
1130 offset++, bprm->p++)
1133 kunmap_atomic(kaddr, KM_USER0);
1136 if (offset == PAGE_SIZE)
1137 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1138 } while (offset == PAGE_SIZE);
1147 EXPORT_SYMBOL(remove_arg_zero);
1150 * cycle the list of binary formats handler, until one recognizes the image
1152 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1155 struct linux_binfmt *fmt;
1156 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1157 /* handle /sbin/loader.. */
1159 struct exec * eh = (struct exec *) bprm->buf;
1161 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1162 (eh->fh.f_flags & 0x3000) == 0x3000)
1165 unsigned long loader;
1167 allow_write_access(bprm->file);
1171 loader = bprm->vma->vm_end - sizeof(void *);
1173 file = open_exec("/sbin/loader");
1174 retval = PTR_ERR(file);
1178 /* Remember if the application is TASO. */
1179 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1182 bprm->loader = loader;
1183 retval = prepare_binprm(bprm);
1186 /* should call search_binary_handler recursively here,
1187 but it does not matter */
1191 retval = security_bprm_check(bprm);
1195 /* kernel module loader fixup */
1196 /* so we don't try to load run modprobe in kernel space. */
1199 retval = audit_bprm(bprm);
1204 for (try=0; try<2; try++) {
1205 read_lock(&binfmt_lock);
1206 list_for_each_entry(fmt, &formats, lh) {
1207 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1210 if (!try_module_get(fmt->module))
1212 read_unlock(&binfmt_lock);
1213 retval = fn(bprm, regs);
1216 allow_write_access(bprm->file);
1220 current->did_exec = 1;
1221 proc_exec_connector(current);
1224 read_lock(&binfmt_lock);
1226 if (retval != -ENOEXEC || bprm->mm == NULL)
1229 read_unlock(&binfmt_lock);
1233 read_unlock(&binfmt_lock);
1234 if (retval != -ENOEXEC || bprm->mm == NULL) {
1238 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1239 if (printable(bprm->buf[0]) &&
1240 printable(bprm->buf[1]) &&
1241 printable(bprm->buf[2]) &&
1242 printable(bprm->buf[3]))
1243 break; /* -ENOEXEC */
1244 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1251 EXPORT_SYMBOL(search_binary_handler);
1254 * sys_execve() executes a new program.
1256 int do_execve(char * filename,
1257 char __user *__user *argv,
1258 char __user *__user *envp,
1259 struct pt_regs * regs)
1261 struct linux_binprm *bprm;
1263 struct files_struct *displaced;
1266 retval = unshare_files(&displaced);
1271 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1275 file = open_exec(filename);
1276 retval = PTR_ERR(file);
1283 bprm->filename = filename;
1284 bprm->interp = filename;
1286 retval = bprm_mm_init(bprm);
1290 bprm->argc = count(argv, MAX_ARG_STRINGS);
1291 if ((retval = bprm->argc) < 0)
1294 bprm->envc = count(envp, MAX_ARG_STRINGS);
1295 if ((retval = bprm->envc) < 0)
1298 retval = security_bprm_alloc(bprm);
1302 retval = prepare_binprm(bprm);
1306 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1310 bprm->exec = bprm->p;
1311 retval = copy_strings(bprm->envc, envp, bprm);
1315 retval = copy_strings(bprm->argc, argv, bprm);
1319 retval = search_binary_handler(bprm,regs);
1321 /* execve success */
1322 free_arg_pages(bprm);
1323 security_bprm_free(bprm);
1324 acct_update_integrals(current);
1327 put_files_struct(displaced);
1332 free_arg_pages(bprm);
1334 security_bprm_free(bprm);
1342 allow_write_access(bprm->file);
1350 reset_files_struct(displaced);
1355 int set_binfmt(struct linux_binfmt *new)
1357 struct linux_binfmt *old = current->binfmt;
1360 if (!try_module_get(new->module))
1363 current->binfmt = new;
1365 module_put(old->module);
1369 EXPORT_SYMBOL(set_binfmt);
1371 /* format_corename will inspect the pattern parameter, and output a
1372 * name into corename, which must have space for at least
1373 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1375 static int format_corename(char *corename, const char *pattern, long signr)
1377 const char *pat_ptr = pattern;
1378 char *out_ptr = corename;
1379 char *const out_end = corename + CORENAME_MAX_SIZE;
1381 int pid_in_pattern = 0;
1384 if (*pattern == '|')
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 || atomic_read(¤t->mm->mm_users) != 1)) {
1487 rc = snprintf(out_ptr, out_end - out_ptr,
1488 ".%d", task_tgid_vnr(current));
1489 if (rc > out_end - out_ptr)
1498 static void zap_process(struct task_struct *start)
1500 struct task_struct *t;
1502 start->signal->flags = SIGNAL_GROUP_EXIT;
1503 start->signal->group_stop_count = 0;
1507 if (t != current && t->mm) {
1508 t->mm->core_waiters++;
1509 sigaddset(&t->pending.signal, SIGKILL);
1510 signal_wake_up(t, 1);
1512 } while ((t = next_thread(t)) != start);
1515 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1518 struct task_struct *g, *p;
1519 unsigned long flags;
1522 spin_lock_irq(&tsk->sighand->siglock);
1523 if (!signal_group_exit(tsk->signal)) {
1524 tsk->signal->group_exit_code = exit_code;
1528 spin_unlock_irq(&tsk->sighand->siglock);
1532 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1536 for_each_process(g) {
1537 if (g == tsk->group_leader)
1545 * p->sighand can't disappear, but
1546 * may be changed by de_thread()
1548 lock_task_sighand(p, &flags);
1550 unlock_task_sighand(p, &flags);
1554 } while ((p = next_thread(p)) != g);
1558 return mm->core_waiters;
1561 static int coredump_wait(int exit_code)
1563 struct task_struct *tsk = current;
1564 struct mm_struct *mm = tsk->mm;
1565 struct completion startup_done;
1566 struct completion *vfork_done;
1569 init_completion(&mm->core_done);
1570 init_completion(&startup_done);
1571 mm->core_startup_done = &startup_done;
1573 core_waiters = zap_threads(tsk, mm, exit_code);
1574 up_write(&mm->mmap_sem);
1576 if (unlikely(core_waiters < 0))
1580 * Make sure nobody is waiting for us to release the VM,
1581 * otherwise we can deadlock when we wait on each other
1583 vfork_done = tsk->vfork_done;
1585 tsk->vfork_done = NULL;
1586 complete(vfork_done);
1590 wait_for_completion(&startup_done);
1592 BUG_ON(mm->core_waiters);
1593 return core_waiters;
1597 * set_dumpable converts traditional three-value dumpable to two flags and
1598 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1599 * these bits are not changed atomically. So get_dumpable can observe the
1600 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1601 * return either old dumpable or new one by paying attention to the order of
1602 * modifying the bits.
1604 * dumpable | mm->flags (binary)
1605 * old new | initial interim final
1606 * ---------+-----------------------
1614 * (*) get_dumpable regards interim value of 10 as 11.
1616 void set_dumpable(struct mm_struct *mm, int value)
1620 clear_bit(MMF_DUMPABLE, &mm->flags);
1622 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1625 set_bit(MMF_DUMPABLE, &mm->flags);
1627 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1630 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1632 set_bit(MMF_DUMPABLE, &mm->flags);
1637 int get_dumpable(struct mm_struct *mm)
1641 ret = mm->flags & 0x3;
1642 return (ret >= 2) ? 2 : ret;
1645 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1647 char corename[CORENAME_MAX_SIZE + 1];
1648 struct mm_struct *mm = current->mm;
1649 struct linux_binfmt * binfmt;
1650 struct inode * inode;
1653 int fsuid = current->fsuid;
1656 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1657 char **helper_argv = NULL;
1658 int helper_argc = 0;
1661 audit_core_dumps(signr);
1663 binfmt = current->binfmt;
1664 if (!binfmt || !binfmt->core_dump)
1666 down_write(&mm->mmap_sem);
1668 * If another thread got here first, or we are not dumpable, bail out.
1670 if (mm->core_waiters || !get_dumpable(mm)) {
1671 up_write(&mm->mmap_sem);
1676 * We cannot trust fsuid as being the "true" uid of the
1677 * process nor do we know its entire history. We only know it
1678 * was tainted so we dump it as root in mode 2.
1680 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1681 flag = O_EXCL; /* Stop rewrite attacks */
1682 current->fsuid = 0; /* Dump root private */
1685 retval = coredump_wait(exit_code);
1690 * Clear any false indication of pending signals that might
1691 * be seen by the filesystem code called to write the core file.
1693 clear_thread_flag(TIF_SIGPENDING);
1696 * lock_kernel() because format_corename() is controlled by sysctl, which
1697 * uses lock_kernel()
1700 ispipe = format_corename(corename, core_pattern, signr);
1703 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1704 * to a pipe. Since we're not writing directly to the filesystem
1705 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1706 * created unless the pipe reader choses to write out the core file
1707 * at which point file size limits and permissions will be imposed
1708 * as it does with any other process
1710 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1714 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1715 /* Terminate the string before the first option */
1716 delimit = strchr(corename, ' ');
1719 delimit = strrchr(helper_argv[0], '/');
1723 delimit = helper_argv[0];
1724 if (!strcmp(delimit, current->comm)) {
1725 printk(KERN_NOTICE "Recursive core dump detected, "
1730 core_limit = RLIM_INFINITY;
1732 /* SIGPIPE can happen, but it's just never processed */
1733 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1735 printk(KERN_INFO "Core dump to %s pipe failed\n",
1740 file = filp_open(corename,
1741 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1745 inode = file->f_path.dentry->d_inode;
1746 if (inode->i_nlink > 1)
1747 goto close_fail; /* multiple links - don't dump */
1748 if (!ispipe && d_unhashed(file->f_path.dentry))
1751 /* AK: actually i see no reason to not allow this for named pipes etc.,
1752 but keep the previous behaviour for now. */
1753 if (!ispipe && !S_ISREG(inode->i_mode))
1756 * Dont allow local users get cute and trick others to coredump
1757 * into their pre-created files:
1759 if (inode->i_uid != current->fsuid)
1763 if (!file->f_op->write)
1765 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1768 retval = binfmt->core_dump(signr, regs, file, core_limit);
1771 current->signal->group_exit_code |= 0x80;
1773 filp_close(file, NULL);
1776 argv_free(helper_argv);
1778 current->fsuid = fsuid;
1779 complete_all(&mm->core_done);