4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
54 #include <asm/uaccess.h>
55 #include <asm/mmu_context.h>
59 #include <linux/kmod.h>
63 /* for /sbin/loader handling in search_binary_handler() */
64 #include <linux/a.out.h>
68 char core_pattern[CORENAME_MAX_SIZE] = "core";
69 int suid_dumpable = 0;
71 /* The maximal length of core_pattern is also specified in sysctl.c */
73 static LIST_HEAD(formats);
74 static DEFINE_RWLOCK(binfmt_lock);
76 int register_binfmt(struct linux_binfmt * fmt)
80 write_lock(&binfmt_lock);
81 list_add(&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)
112 char *tmp = getname(library);
113 int error = PTR_ERR(tmp);
116 error = path_lookup_open(AT_FDCWD, tmp,
118 FMODE_READ|FMODE_EXEC);
125 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
129 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
132 error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
136 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
137 error = PTR_ERR(file);
143 struct linux_binfmt * fmt;
145 read_lock(&binfmt_lock);
146 list_for_each_entry(fmt, &formats, lh) {
147 if (!fmt->load_shlib)
149 if (!try_module_get(fmt->module))
151 read_unlock(&binfmt_lock);
152 error = fmt->load_shlib(file);
153 read_lock(&binfmt_lock);
155 if (error != -ENOEXEC)
158 read_unlock(&binfmt_lock);
164 release_open_intent(&nd);
171 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
177 #ifdef CONFIG_STACK_GROWSUP
179 ret = expand_stack_downwards(bprm->vma, pos);
184 ret = get_user_pages(current, bprm->mm, pos,
185 1, write, 1, &page, NULL);
190 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
194 * We've historically supported up to 32 pages (ARG_MAX)
195 * of argument strings even with small stacks
201 * Limit to 1/4-th the stack size for the argv+env strings.
203 * - the remaining binfmt code will not run out of stack space,
204 * - the program will have a reasonable amount of stack left
207 rlim = current->signal->rlim;
208 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
217 static void put_arg_page(struct page *page)
222 static void free_arg_page(struct linux_binprm *bprm, int i)
226 static void free_arg_pages(struct linux_binprm *bprm)
230 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
233 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
236 static int __bprm_mm_init(struct linux_binprm *bprm)
239 struct vm_area_struct *vma = NULL;
240 struct mm_struct *mm = bprm->mm;
242 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
246 down_write(&mm->mmap_sem);
250 * Place the stack at the largest stack address the architecture
251 * supports. Later, we'll move this to an appropriate place. We don't
252 * use STACK_TOP because that can depend on attributes which aren't
255 vma->vm_end = STACK_TOP_MAX;
256 vma->vm_start = vma->vm_end - PAGE_SIZE;
258 vma->vm_flags = VM_STACK_FLAGS;
259 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
260 err = insert_vm_struct(mm, vma);
262 up_write(&mm->mmap_sem);
266 mm->stack_vm = mm->total_vm = 1;
267 up_write(&mm->mmap_sem);
269 bprm->p = vma->vm_end - sizeof(void *);
276 kmem_cache_free(vm_area_cachep, vma);
282 static bool valid_arg_len(struct linux_binprm *bprm, long len)
284 return len <= MAX_ARG_STRLEN;
289 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
294 page = bprm->page[pos / PAGE_SIZE];
295 if (!page && write) {
296 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
299 bprm->page[pos / PAGE_SIZE] = page;
305 static void put_arg_page(struct page *page)
309 static void free_arg_page(struct linux_binprm *bprm, int i)
312 __free_page(bprm->page[i]);
313 bprm->page[i] = NULL;
317 static void free_arg_pages(struct linux_binprm *bprm)
321 for (i = 0; i < MAX_ARG_PAGES; i++)
322 free_arg_page(bprm, i);
325 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
330 static int __bprm_mm_init(struct linux_binprm *bprm)
332 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
336 static bool valid_arg_len(struct linux_binprm *bprm, long len)
338 return len <= bprm->p;
341 #endif /* CONFIG_MMU */
344 * Create a new mm_struct and populate it with a temporary stack
345 * vm_area_struct. We don't have enough context at this point to set the stack
346 * flags, permissions, and offset, so we use temporary values. We'll update
347 * them later in setup_arg_pages().
349 int bprm_mm_init(struct linux_binprm *bprm)
352 struct mm_struct *mm = NULL;
354 bprm->mm = mm = mm_alloc();
359 err = init_new_context(current, mm);
363 err = __bprm_mm_init(bprm);
379 * count() counts the number of strings in array ARGV.
381 static int count(char __user * __user * argv, int max)
389 if (get_user(p, argv))
403 * 'copy_strings()' copies argument/environment strings from the old
404 * processes's memory to the new process's stack. The call to get_user_pages()
405 * ensures the destination page is created and not swapped out.
407 static int copy_strings(int argc, char __user * __user * argv,
408 struct linux_binprm *bprm)
410 struct page *kmapped_page = NULL;
412 unsigned long kpos = 0;
420 if (get_user(str, argv+argc) ||
421 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
426 if (!valid_arg_len(bprm, len)) {
431 /* We're going to work our way backwords. */
437 int offset, bytes_to_copy;
439 offset = pos % PAGE_SIZE;
443 bytes_to_copy = offset;
444 if (bytes_to_copy > len)
447 offset -= bytes_to_copy;
448 pos -= bytes_to_copy;
449 str -= bytes_to_copy;
450 len -= bytes_to_copy;
452 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
455 page = get_arg_page(bprm, pos, 1);
462 flush_kernel_dcache_page(kmapped_page);
463 kunmap(kmapped_page);
464 put_arg_page(kmapped_page);
467 kaddr = kmap(kmapped_page);
468 kpos = pos & PAGE_MASK;
469 flush_arg_page(bprm, kpos, kmapped_page);
471 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
480 flush_kernel_dcache_page(kmapped_page);
481 kunmap(kmapped_page);
482 put_arg_page(kmapped_page);
488 * Like copy_strings, but get argv and its values from kernel memory.
490 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
493 mm_segment_t oldfs = get_fs();
495 r = copy_strings(argc, (char __user * __user *)argv, bprm);
499 EXPORT_SYMBOL(copy_strings_kernel);
504 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
505 * the binfmt code determines where the new stack should reside, we shift it to
506 * its final location. The process proceeds as follows:
508 * 1) Use shift to calculate the new vma endpoints.
509 * 2) Extend vma to cover both the old and new ranges. This ensures the
510 * arguments passed to subsequent functions are consistent.
511 * 3) Move vma's page tables to the new range.
512 * 4) Free up any cleared pgd range.
513 * 5) Shrink the vma to cover only the new range.
515 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
517 struct mm_struct *mm = vma->vm_mm;
518 unsigned long old_start = vma->vm_start;
519 unsigned long old_end = vma->vm_end;
520 unsigned long length = old_end - old_start;
521 unsigned long new_start = old_start - shift;
522 unsigned long new_end = old_end - shift;
523 struct mmu_gather *tlb;
525 BUG_ON(new_start > new_end);
528 * ensure there are no vmas between where we want to go
531 if (vma != find_vma(mm, new_start))
535 * cover the whole range: [new_start, old_end)
537 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
540 * move the page tables downwards, on failure we rely on
541 * process cleanup to remove whatever mess we made.
543 if (length != move_page_tables(vma, old_start,
544 vma, new_start, length))
548 tlb = tlb_gather_mmu(mm, 0);
549 if (new_end > old_start) {
551 * when the old and new regions overlap clear from new_end.
553 free_pgd_range(tlb, new_end, old_end, new_end,
554 vma->vm_next ? vma->vm_next->vm_start : 0);
557 * otherwise, clean from old_start; this is done to not touch
558 * the address space in [new_end, old_start) some architectures
559 * have constraints on va-space that make this illegal (IA64) -
560 * for the others its just a little faster.
562 free_pgd_range(tlb, old_start, old_end, new_end,
563 vma->vm_next ? vma->vm_next->vm_start : 0);
565 tlb_finish_mmu(tlb, new_end, old_end);
568 * shrink the vma to just the new range.
570 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
575 #define EXTRA_STACK_VM_PAGES 20 /* random */
578 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
579 * the stack is optionally relocated, and some extra space is added.
581 int setup_arg_pages(struct linux_binprm *bprm,
582 unsigned long stack_top,
583 int executable_stack)
586 unsigned long stack_shift;
587 struct mm_struct *mm = current->mm;
588 struct vm_area_struct *vma = bprm->vma;
589 struct vm_area_struct *prev = NULL;
590 unsigned long vm_flags;
591 unsigned long stack_base;
593 #ifdef CONFIG_STACK_GROWSUP
594 /* Limit stack size to 1GB */
595 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
596 if (stack_base > (1 << 30))
597 stack_base = 1 << 30;
599 /* Make sure we didn't let the argument array grow too large. */
600 if (vma->vm_end - vma->vm_start > stack_base)
603 stack_base = PAGE_ALIGN(stack_top - stack_base);
605 stack_shift = vma->vm_start - stack_base;
606 mm->arg_start = bprm->p - stack_shift;
607 bprm->p = vma->vm_end - stack_shift;
609 stack_top = arch_align_stack(stack_top);
610 stack_top = PAGE_ALIGN(stack_top);
611 stack_shift = vma->vm_end - stack_top;
613 bprm->p -= stack_shift;
614 mm->arg_start = bprm->p;
618 bprm->loader -= stack_shift;
619 bprm->exec -= stack_shift;
621 down_write(&mm->mmap_sem);
622 vm_flags = VM_STACK_FLAGS;
625 * Adjust stack execute permissions; explicitly enable for
626 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
627 * (arch default) otherwise.
629 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
631 else if (executable_stack == EXSTACK_DISABLE_X)
632 vm_flags &= ~VM_EXEC;
633 vm_flags |= mm->def_flags;
635 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
641 /* Move stack pages down in memory. */
643 ret = shift_arg_pages(vma, stack_shift);
645 up_write(&mm->mmap_sem);
650 #ifdef CONFIG_STACK_GROWSUP
651 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
653 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
655 ret = expand_stack(vma, stack_base);
660 up_write(&mm->mmap_sem);
663 EXPORT_SYMBOL(setup_arg_pages);
665 #endif /* CONFIG_MMU */
667 struct file *open_exec(const char *name)
673 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
674 FMODE_READ|FMODE_EXEC);
679 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
682 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
685 err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
689 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
693 err = deny_write_access(file);
702 release_open_intent(&nd);
707 EXPORT_SYMBOL(open_exec);
709 int kernel_read(struct file *file, unsigned long offset,
710 char *addr, unsigned long count)
718 /* The cast to a user pointer is valid due to the set_fs() */
719 result = vfs_read(file, (void __user *)addr, count, &pos);
724 EXPORT_SYMBOL(kernel_read);
726 static int exec_mmap(struct mm_struct *mm)
728 struct task_struct *tsk;
729 struct mm_struct * old_mm, *active_mm;
731 /* Notify parent that we're no longer interested in the old VM */
733 old_mm = current->mm;
734 mm_release(tsk, old_mm);
738 * Make sure that if there is a core dump in progress
739 * for the old mm, we get out and die instead of going
740 * through with the exec. We must hold mmap_sem around
741 * checking core_state and changing tsk->mm.
743 down_read(&old_mm->mmap_sem);
744 if (unlikely(old_mm->core_state)) {
745 up_read(&old_mm->mmap_sem);
750 active_mm = tsk->active_mm;
753 activate_mm(active_mm, mm);
755 arch_pick_mmap_layout(mm);
757 up_read(&old_mm->mmap_sem);
758 BUG_ON(active_mm != old_mm);
759 mm_update_next_owner(old_mm);
768 * This function makes sure the current process has its own signal table,
769 * so that flush_signal_handlers can later reset the handlers without
770 * disturbing other processes. (Other processes might share the signal
771 * table via the CLONE_SIGHAND option to clone().)
773 static int de_thread(struct task_struct *tsk)
775 struct signal_struct *sig = tsk->signal;
776 struct sighand_struct *oldsighand = tsk->sighand;
777 spinlock_t *lock = &oldsighand->siglock;
778 struct task_struct *leader = NULL;
781 if (thread_group_empty(tsk))
782 goto no_thread_group;
785 * Kill all other threads in the thread group.
788 if (signal_group_exit(sig)) {
790 * Another group action in progress, just
791 * return so that the signal is processed.
793 spin_unlock_irq(lock);
796 sig->group_exit_task = tsk;
797 zap_other_threads(tsk);
799 /* Account for the thread group leader hanging around: */
800 count = thread_group_leader(tsk) ? 1 : 2;
801 sig->notify_count = count;
802 while (atomic_read(&sig->count) > count) {
803 __set_current_state(TASK_UNINTERRUPTIBLE);
804 spin_unlock_irq(lock);
808 spin_unlock_irq(lock);
811 * At this point all other threads have exited, all we have to
812 * do is to wait for the thread group leader to become inactive,
813 * and to assume its PID:
815 if (!thread_group_leader(tsk)) {
816 leader = tsk->group_leader;
818 sig->notify_count = -1; /* for exit_notify() */
820 write_lock_irq(&tasklist_lock);
821 if (likely(leader->exit_state))
823 __set_current_state(TASK_UNINTERRUPTIBLE);
824 write_unlock_irq(&tasklist_lock);
828 if (unlikely(task_child_reaper(tsk) == leader))
829 task_active_pid_ns(tsk)->child_reaper = tsk;
831 * The only record we have of the real-time age of a
832 * process, regardless of execs it's done, is start_time.
833 * All the past CPU time is accumulated in signal_struct
834 * from sister threads now dead. But in this non-leader
835 * exec, nothing survives from the original leader thread,
836 * whose birth marks the true age of this process now.
837 * When we take on its identity by switching to its PID, we
838 * also take its birthdate (always earlier than our own).
840 tsk->start_time = leader->start_time;
842 BUG_ON(!same_thread_group(leader, tsk));
843 BUG_ON(has_group_leader_pid(tsk));
845 * An exec() starts a new thread group with the
846 * TGID of the previous thread group. Rehash the
847 * two threads with a switched PID, and release
848 * the former thread group leader:
851 /* Become a process group leader with the old leader's pid.
852 * The old leader becomes a thread of the this thread group.
853 * Note: The old leader also uses this pid until release_task
854 * is called. Odd but simple and correct.
856 detach_pid(tsk, PIDTYPE_PID);
857 tsk->pid = leader->pid;
858 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
859 transfer_pid(leader, tsk, PIDTYPE_PGID);
860 transfer_pid(leader, tsk, PIDTYPE_SID);
861 list_replace_rcu(&leader->tasks, &tsk->tasks);
863 tsk->group_leader = tsk;
864 leader->group_leader = tsk;
866 tsk->exit_signal = SIGCHLD;
868 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
869 leader->exit_state = EXIT_DEAD;
871 write_unlock_irq(&tasklist_lock);
874 sig->group_exit_task = NULL;
875 sig->notify_count = 0;
879 flush_itimer_signals();
881 release_task(leader);
883 if (atomic_read(&oldsighand->count) != 1) {
884 struct sighand_struct *newsighand;
886 * This ->sighand is shared with the CLONE_SIGHAND
887 * but not CLONE_THREAD task, switch to the new one.
889 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
893 atomic_set(&newsighand->count, 1);
894 memcpy(newsighand->action, oldsighand->action,
895 sizeof(newsighand->action));
897 write_lock_irq(&tasklist_lock);
898 spin_lock(&oldsighand->siglock);
899 rcu_assign_pointer(tsk->sighand, newsighand);
900 spin_unlock(&oldsighand->siglock);
901 write_unlock_irq(&tasklist_lock);
903 __cleanup_sighand(oldsighand);
906 BUG_ON(!thread_group_leader(tsk));
911 * These functions flushes out all traces of the currently running executable
912 * so that a new one can be started
914 static void flush_old_files(struct files_struct * files)
919 spin_lock(&files->file_lock);
921 unsigned long set, i;
925 fdt = files_fdtable(files);
926 if (i >= fdt->max_fds)
928 set = fdt->close_on_exec->fds_bits[j];
931 fdt->close_on_exec->fds_bits[j] = 0;
932 spin_unlock(&files->file_lock);
933 for ( ; set ; i++,set >>= 1) {
938 spin_lock(&files->file_lock);
941 spin_unlock(&files->file_lock);
944 char *get_task_comm(char *buf, struct task_struct *tsk)
946 /* buf must be at least sizeof(tsk->comm) in size */
948 strncpy(buf, tsk->comm, sizeof(tsk->comm));
953 void set_task_comm(struct task_struct *tsk, char *buf)
956 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
960 int flush_old_exec(struct linux_binprm * bprm)
964 char tcomm[sizeof(current->comm)];
967 * Make sure we have a private signal table and that
968 * we are unassociated from the previous thread group.
970 retval = de_thread(current);
974 set_mm_exe_file(bprm->mm, bprm->file);
977 * Release all of the old mmap stuff
979 retval = exec_mmap(bprm->mm);
983 bprm->mm = NULL; /* We're using it now */
985 /* This is the point of no return */
986 current->sas_ss_sp = current->sas_ss_size = 0;
988 if (current->euid == current->uid && current->egid == current->gid)
989 set_dumpable(current->mm, 1);
991 set_dumpable(current->mm, suid_dumpable);
993 name = bprm->filename;
995 /* Copies the binary name from after last slash */
996 for (i=0; (ch = *(name++)) != '\0';) {
998 i = 0; /* overwrite what we wrote */
1000 if (i < (sizeof(tcomm) - 1))
1004 set_task_comm(current, tcomm);
1006 current->flags &= ~PF_RANDOMIZE;
1009 /* Set the new mm task size. We have to do that late because it may
1010 * depend on TIF_32BIT which is only updated in flush_thread() on
1011 * some architectures like powerpc
1013 current->mm->task_size = TASK_SIZE;
1015 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1017 set_dumpable(current->mm, suid_dumpable);
1018 current->pdeath_signal = 0;
1019 } else if (file_permission(bprm->file, MAY_READ) ||
1020 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1022 set_dumpable(current->mm, suid_dumpable);
1025 /* An exec changes our domain. We are no longer part of the thread
1028 current->self_exec_id++;
1030 flush_signal_handlers(current, 0);
1031 flush_old_files(current->files);
1039 EXPORT_SYMBOL(flush_old_exec);
1042 * Fill the binprm structure from the inode.
1043 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1045 int prepare_binprm(struct linux_binprm *bprm)
1048 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1051 mode = inode->i_mode;
1052 if (bprm->file->f_op == NULL)
1055 bprm->e_uid = current->euid;
1056 bprm->e_gid = current->egid;
1058 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1060 if (mode & S_ISUID) {
1061 current->personality &= ~PER_CLEAR_ON_SETID;
1062 bprm->e_uid = inode->i_uid;
1067 * If setgid is set but no group execute bit then this
1068 * is a candidate for mandatory locking, not a setgid
1071 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1072 current->personality &= ~PER_CLEAR_ON_SETID;
1073 bprm->e_gid = inode->i_gid;
1077 /* fill in binprm security blob */
1078 retval = security_bprm_set(bprm);
1082 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1083 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1086 EXPORT_SYMBOL(prepare_binprm);
1088 static int unsafe_exec(struct task_struct *p)
1090 int unsafe = tracehook_unsafe_exec(p);
1092 if (atomic_read(&p->fs->count) > 1)
1093 unsafe |= LSM_UNSAFE_SHARE;
1098 void compute_creds(struct linux_binprm *bprm)
1102 if (bprm->e_uid != current->uid) {
1104 current->pdeath_signal = 0;
1109 unsafe = unsafe_exec(current);
1110 security_bprm_apply_creds(bprm, unsafe);
1111 task_unlock(current);
1112 security_bprm_post_apply_creds(bprm);
1114 EXPORT_SYMBOL(compute_creds);
1117 * Arguments are '\0' separated strings found at the location bprm->p
1118 * points to; chop off the first by relocating brpm->p to right after
1119 * the first '\0' encountered.
1121 int remove_arg_zero(struct linux_binprm *bprm)
1124 unsigned long offset;
1132 offset = bprm->p & ~PAGE_MASK;
1133 page = get_arg_page(bprm, bprm->p, 0);
1138 kaddr = kmap_atomic(page, KM_USER0);
1140 for (; offset < PAGE_SIZE && kaddr[offset];
1141 offset++, bprm->p++)
1144 kunmap_atomic(kaddr, KM_USER0);
1147 if (offset == PAGE_SIZE)
1148 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1149 } while (offset == PAGE_SIZE);
1158 EXPORT_SYMBOL(remove_arg_zero);
1161 * cycle the list of binary formats handler, until one recognizes the image
1163 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1165 unsigned int depth = bprm->recursion_depth;
1167 struct linux_binfmt *fmt;
1169 /* handle /sbin/loader.. */
1171 struct exec * eh = (struct exec *) bprm->buf;
1173 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1174 (eh->fh.f_flags & 0x3000) == 0x3000)
1177 unsigned long loader;
1179 allow_write_access(bprm->file);
1183 loader = bprm->vma->vm_end - sizeof(void *);
1185 file = open_exec("/sbin/loader");
1186 retval = PTR_ERR(file);
1190 /* Remember if the application is TASO. */
1191 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1194 bprm->loader = loader;
1195 retval = prepare_binprm(bprm);
1198 /* should call search_binary_handler recursively here,
1199 but it does not matter */
1203 retval = security_bprm_check(bprm);
1207 /* kernel module loader fixup */
1208 /* so we don't try to load run modprobe in kernel space. */
1211 retval = audit_bprm(bprm);
1216 for (try=0; try<2; try++) {
1217 read_lock(&binfmt_lock);
1218 list_for_each_entry(fmt, &formats, lh) {
1219 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1222 if (!try_module_get(fmt->module))
1224 read_unlock(&binfmt_lock);
1225 retval = fn(bprm, regs);
1227 * Restore the depth counter to its starting value
1228 * in this call, so we don't have to rely on every
1229 * load_binary function to restore it on return.
1231 bprm->recursion_depth = depth;
1234 tracehook_report_exec(fmt, bprm, regs);
1236 allow_write_access(bprm->file);
1240 current->did_exec = 1;
1241 proc_exec_connector(current);
1244 read_lock(&binfmt_lock);
1246 if (retval != -ENOEXEC || bprm->mm == NULL)
1249 read_unlock(&binfmt_lock);
1253 read_unlock(&binfmt_lock);
1254 if (retval != -ENOEXEC || bprm->mm == NULL) {
1258 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1259 if (printable(bprm->buf[0]) &&
1260 printable(bprm->buf[1]) &&
1261 printable(bprm->buf[2]) &&
1262 printable(bprm->buf[3]))
1263 break; /* -ENOEXEC */
1264 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1271 EXPORT_SYMBOL(search_binary_handler);
1273 void free_bprm(struct linux_binprm *bprm)
1275 free_arg_pages(bprm);
1280 * sys_execve() executes a new program.
1282 int do_execve(char * filename,
1283 char __user *__user *argv,
1284 char __user *__user *envp,
1285 struct pt_regs * regs)
1287 struct linux_binprm *bprm;
1289 struct files_struct *displaced;
1292 retval = unshare_files(&displaced);
1297 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1301 file = open_exec(filename);
1302 retval = PTR_ERR(file);
1309 bprm->filename = filename;
1310 bprm->interp = filename;
1312 retval = bprm_mm_init(bprm);
1316 bprm->argc = count(argv, MAX_ARG_STRINGS);
1317 if ((retval = bprm->argc) < 0)
1320 bprm->envc = count(envp, MAX_ARG_STRINGS);
1321 if ((retval = bprm->envc) < 0)
1324 retval = security_bprm_alloc(bprm);
1328 retval = prepare_binprm(bprm);
1332 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1336 bprm->exec = bprm->p;
1337 retval = copy_strings(bprm->envc, envp, bprm);
1341 retval = copy_strings(bprm->argc, argv, bprm);
1345 current->flags &= ~PF_KTHREAD;
1346 retval = search_binary_handler(bprm,regs);
1348 /* execve success */
1349 security_bprm_free(bprm);
1350 acct_update_integrals(current);
1353 put_files_struct(displaced);
1359 security_bprm_free(bprm);
1367 allow_write_access(bprm->file);
1375 reset_files_struct(displaced);
1380 int set_binfmt(struct linux_binfmt *new)
1382 struct linux_binfmt *old = current->binfmt;
1385 if (!try_module_get(new->module))
1388 current->binfmt = new;
1390 module_put(old->module);
1394 EXPORT_SYMBOL(set_binfmt);
1396 /* format_corename will inspect the pattern parameter, and output a
1397 * name into corename, which must have space for at least
1398 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1400 static int format_corename(char *corename, int nr_threads, long signr)
1402 const char *pat_ptr = core_pattern;
1403 int ispipe = (*pat_ptr == '|');
1404 char *out_ptr = corename;
1405 char *const out_end = corename + CORENAME_MAX_SIZE;
1407 int pid_in_pattern = 0;
1409 /* Repeat as long as we have more pattern to process and more output
1412 if (*pat_ptr != '%') {
1413 if (out_ptr == out_end)
1415 *out_ptr++ = *pat_ptr++;
1417 switch (*++pat_ptr) {
1420 /* Double percent, output one percent */
1422 if (out_ptr == out_end)
1429 rc = snprintf(out_ptr, out_end - out_ptr,
1430 "%d", task_tgid_vnr(current));
1431 if (rc > out_end - out_ptr)
1437 rc = snprintf(out_ptr, out_end - out_ptr,
1438 "%d", current->uid);
1439 if (rc > out_end - out_ptr)
1445 rc = snprintf(out_ptr, out_end - out_ptr,
1446 "%d", current->gid);
1447 if (rc > out_end - out_ptr)
1451 /* signal that caused the coredump */
1453 rc = snprintf(out_ptr, out_end - out_ptr,
1455 if (rc > out_end - out_ptr)
1459 /* UNIX time of coredump */
1462 do_gettimeofday(&tv);
1463 rc = snprintf(out_ptr, out_end - out_ptr,
1465 if (rc > out_end - out_ptr)
1472 down_read(&uts_sem);
1473 rc = snprintf(out_ptr, out_end - out_ptr,
1474 "%s", utsname()->nodename);
1476 if (rc > out_end - out_ptr)
1482 rc = snprintf(out_ptr, out_end - out_ptr,
1483 "%s", current->comm);
1484 if (rc > out_end - out_ptr)
1488 /* core limit size */
1490 rc = snprintf(out_ptr, out_end - out_ptr,
1491 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1492 if (rc > out_end - out_ptr)
1502 /* Backward compatibility with core_uses_pid:
1504 * If core_pattern does not include a %p (as is the default)
1505 * and core_uses_pid is set, then .%pid will be appended to
1506 * the filename. Do not do this for piped commands. */
1507 if (!ispipe && !pid_in_pattern
1508 && (core_uses_pid || nr_threads)) {
1509 rc = snprintf(out_ptr, out_end - out_ptr,
1510 ".%d", task_tgid_vnr(current));
1511 if (rc > out_end - out_ptr)
1520 static int zap_process(struct task_struct *start)
1522 struct task_struct *t;
1525 start->signal->flags = SIGNAL_GROUP_EXIT;
1526 start->signal->group_stop_count = 0;
1530 if (t != current && t->mm) {
1531 sigaddset(&t->pending.signal, SIGKILL);
1532 signal_wake_up(t, 1);
1535 } while_each_thread(start, t);
1540 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1541 struct core_state *core_state, int exit_code)
1543 struct task_struct *g, *p;
1544 unsigned long flags;
1547 spin_lock_irq(&tsk->sighand->siglock);
1548 if (!signal_group_exit(tsk->signal)) {
1549 mm->core_state = core_state;
1550 tsk->signal->group_exit_code = exit_code;
1551 nr = zap_process(tsk);
1553 spin_unlock_irq(&tsk->sighand->siglock);
1554 if (unlikely(nr < 0))
1557 if (atomic_read(&mm->mm_users) == nr + 1)
1560 * We should find and kill all tasks which use this mm, and we should
1561 * count them correctly into ->nr_threads. We don't take tasklist
1562 * lock, but this is safe wrt:
1565 * None of sub-threads can fork after zap_process(leader). All
1566 * processes which were created before this point should be
1567 * visible to zap_threads() because copy_process() adds the new
1568 * process to the tail of init_task.tasks list, and lock/unlock
1569 * of ->siglock provides a memory barrier.
1572 * The caller holds mm->mmap_sem. This means that the task which
1573 * uses this mm can't pass exit_mm(), so it can't exit or clear
1577 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1578 * we must see either old or new leader, this does not matter.
1579 * However, it can change p->sighand, so lock_task_sighand(p)
1580 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1583 * Note also that "g" can be the old leader with ->mm == NULL
1584 * and already unhashed and thus removed from ->thread_group.
1585 * This is OK, __unhash_process()->list_del_rcu() does not
1586 * clear the ->next pointer, we will find the new leader via
1590 for_each_process(g) {
1591 if (g == tsk->group_leader)
1593 if (g->flags & PF_KTHREAD)
1598 if (unlikely(p->mm == mm)) {
1599 lock_task_sighand(p, &flags);
1600 nr += zap_process(p);
1601 unlock_task_sighand(p, &flags);
1605 } while_each_thread(g, p);
1609 atomic_set(&core_state->nr_threads, nr);
1613 static int coredump_wait(int exit_code, struct core_state *core_state)
1615 struct task_struct *tsk = current;
1616 struct mm_struct *mm = tsk->mm;
1617 struct completion *vfork_done;
1620 init_completion(&core_state->startup);
1621 core_state->dumper.task = tsk;
1622 core_state->dumper.next = NULL;
1623 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1624 up_write(&mm->mmap_sem);
1626 if (unlikely(core_waiters < 0))
1630 * Make sure nobody is waiting for us to release the VM,
1631 * otherwise we can deadlock when we wait on each other
1633 vfork_done = tsk->vfork_done;
1635 tsk->vfork_done = NULL;
1636 complete(vfork_done);
1640 wait_for_completion(&core_state->startup);
1642 return core_waiters;
1645 static void coredump_finish(struct mm_struct *mm)
1647 struct core_thread *curr, *next;
1648 struct task_struct *task;
1650 next = mm->core_state->dumper.next;
1651 while ((curr = next) != NULL) {
1655 * see exit_mm(), curr->task must not see
1656 * ->task == NULL before we read ->next.
1660 wake_up_process(task);
1663 mm->core_state = NULL;
1667 * set_dumpable converts traditional three-value dumpable to two flags and
1668 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1669 * these bits are not changed atomically. So get_dumpable can observe the
1670 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1671 * return either old dumpable or new one by paying attention to the order of
1672 * modifying the bits.
1674 * dumpable | mm->flags (binary)
1675 * old new | initial interim final
1676 * ---------+-----------------------
1684 * (*) get_dumpable regards interim value of 10 as 11.
1686 void set_dumpable(struct mm_struct *mm, int value)
1690 clear_bit(MMF_DUMPABLE, &mm->flags);
1692 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1695 set_bit(MMF_DUMPABLE, &mm->flags);
1697 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1700 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1702 set_bit(MMF_DUMPABLE, &mm->flags);
1707 int get_dumpable(struct mm_struct *mm)
1711 ret = mm->flags & 0x3;
1712 return (ret >= 2) ? 2 : ret;
1715 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1717 struct core_state core_state;
1718 char corename[CORENAME_MAX_SIZE + 1];
1719 struct mm_struct *mm = current->mm;
1720 struct linux_binfmt * binfmt;
1721 struct inode * inode;
1724 int fsuid = current->fsuid;
1727 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1728 char **helper_argv = NULL;
1729 int helper_argc = 0;
1732 audit_core_dumps(signr);
1734 binfmt = current->binfmt;
1735 if (!binfmt || !binfmt->core_dump)
1737 down_write(&mm->mmap_sem);
1739 * If another thread got here first, or we are not dumpable, bail out.
1741 if (mm->core_state || !get_dumpable(mm)) {
1742 up_write(&mm->mmap_sem);
1747 * We cannot trust fsuid as being the "true" uid of the
1748 * process nor do we know its entire history. We only know it
1749 * was tainted so we dump it as root in mode 2.
1751 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1752 flag = O_EXCL; /* Stop rewrite attacks */
1753 current->fsuid = 0; /* Dump root private */
1756 retval = coredump_wait(exit_code, &core_state);
1761 * Clear any false indication of pending signals that might
1762 * be seen by the filesystem code called to write the core file.
1764 clear_thread_flag(TIF_SIGPENDING);
1767 * lock_kernel() because format_corename() is controlled by sysctl, which
1768 * uses lock_kernel()
1771 ispipe = format_corename(corename, retval, signr);
1774 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1775 * to a pipe. Since we're not writing directly to the filesystem
1776 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1777 * created unless the pipe reader choses to write out the core file
1778 * at which point file size limits and permissions will be imposed
1779 * as it does with any other process
1781 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1785 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1786 /* Terminate the string before the first option */
1787 delimit = strchr(corename, ' ');
1790 delimit = strrchr(helper_argv[0], '/');
1794 delimit = helper_argv[0];
1795 if (!strcmp(delimit, current->comm)) {
1796 printk(KERN_NOTICE "Recursive core dump detected, "
1801 core_limit = RLIM_INFINITY;
1803 /* SIGPIPE can happen, but it's just never processed */
1804 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1806 printk(KERN_INFO "Core dump to %s pipe failed\n",
1811 file = filp_open(corename,
1812 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1816 inode = file->f_path.dentry->d_inode;
1817 if (inode->i_nlink > 1)
1818 goto close_fail; /* multiple links - don't dump */
1819 if (!ispipe && d_unhashed(file->f_path.dentry))
1822 /* AK: actually i see no reason to not allow this for named pipes etc.,
1823 but keep the previous behaviour for now. */
1824 if (!ispipe && !S_ISREG(inode->i_mode))
1827 * Dont allow local users get cute and trick others to coredump
1828 * into their pre-created files:
1830 if (inode->i_uid != current->fsuid)
1834 if (!file->f_op->write)
1836 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1839 retval = binfmt->core_dump(signr, regs, file, core_limit);
1842 current->signal->group_exit_code |= 0x80;
1844 filp_close(file, NULL);
1847 argv_free(helper_argv);
1849 current->fsuid = fsuid;
1850 coredump_finish(mm);