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.
769 * We must hold tasklist_lock to call zap_other_threads.
771 read_lock(&tasklist_lock);
773 if (signal_group_exit(sig)) {
775 * Another group action in progress, just
776 * return so that the signal is processed.
778 spin_unlock_irq(lock);
779 read_unlock(&tasklist_lock);
784 * child_reaper ignores SIGKILL, change it now.
785 * Reparenting needs write_lock on tasklist_lock,
786 * so it is safe to do it under read_lock.
788 if (unlikely(tsk->group_leader == task_child_reaper(tsk)))
789 task_active_pid_ns(tsk)->child_reaper = tsk;
791 sig->group_exit_task = tsk;
792 zap_other_threads(tsk);
793 read_unlock(&tasklist_lock);
795 /* Account for the thread group leader hanging around: */
796 count = thread_group_leader(tsk) ? 1 : 2;
797 sig->notify_count = count;
798 while (atomic_read(&sig->count) > count) {
799 __set_current_state(TASK_UNINTERRUPTIBLE);
800 spin_unlock_irq(lock);
804 spin_unlock_irq(lock);
807 * At this point all other threads have exited, all we have to
808 * do is to wait for the thread group leader to become inactive,
809 * and to assume its PID:
811 if (!thread_group_leader(tsk)) {
812 leader = tsk->group_leader;
814 sig->notify_count = -1;
816 write_lock_irq(&tasklist_lock);
817 if (likely(leader->exit_state))
819 __set_current_state(TASK_UNINTERRUPTIBLE);
820 write_unlock_irq(&tasklist_lock);
825 * The only record we have of the real-time age of a
826 * process, regardless of execs it's done, is start_time.
827 * All the past CPU time is accumulated in signal_struct
828 * from sister threads now dead. But in this non-leader
829 * exec, nothing survives from the original leader thread,
830 * whose birth marks the true age of this process now.
831 * When we take on its identity by switching to its PID, we
832 * also take its birthdate (always earlier than our own).
834 tsk->start_time = leader->start_time;
836 BUG_ON(!same_thread_group(leader, tsk));
837 BUG_ON(has_group_leader_pid(tsk));
839 * An exec() starts a new thread group with the
840 * TGID of the previous thread group. Rehash the
841 * two threads with a switched PID, and release
842 * the former thread group leader:
845 /* Become a process group leader with the old leader's pid.
846 * The old leader becomes a thread of the this thread group.
847 * Note: The old leader also uses this pid until release_task
848 * is called. Odd but simple and correct.
850 detach_pid(tsk, PIDTYPE_PID);
851 tsk->pid = leader->pid;
852 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
853 transfer_pid(leader, tsk, PIDTYPE_PGID);
854 transfer_pid(leader, tsk, PIDTYPE_SID);
855 list_replace_rcu(&leader->tasks, &tsk->tasks);
857 tsk->group_leader = tsk;
858 leader->group_leader = tsk;
860 tsk->exit_signal = SIGCHLD;
862 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
863 leader->exit_state = EXIT_DEAD;
865 write_unlock_irq(&tasklist_lock);
868 sig->group_exit_task = NULL;
869 sig->notify_count = 0;
874 release_task(leader);
876 if (atomic_read(&oldsighand->count) != 1) {
877 struct sighand_struct *newsighand;
879 * This ->sighand is shared with the CLONE_SIGHAND
880 * but not CLONE_THREAD task, switch to the new one.
882 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
886 atomic_set(&newsighand->count, 1);
887 memcpy(newsighand->action, oldsighand->action,
888 sizeof(newsighand->action));
890 write_lock_irq(&tasklist_lock);
891 spin_lock(&oldsighand->siglock);
892 rcu_assign_pointer(tsk->sighand, newsighand);
893 spin_unlock(&oldsighand->siglock);
894 write_unlock_irq(&tasklist_lock);
896 __cleanup_sighand(oldsighand);
899 BUG_ON(!thread_group_leader(tsk));
904 * These functions flushes out all traces of the currently running executable
905 * so that a new one can be started
907 static void flush_old_files(struct files_struct * files)
912 spin_lock(&files->file_lock);
914 unsigned long set, i;
918 fdt = files_fdtable(files);
919 if (i >= fdt->max_fds)
921 set = fdt->close_on_exec->fds_bits[j];
924 fdt->close_on_exec->fds_bits[j] = 0;
925 spin_unlock(&files->file_lock);
926 for ( ; set ; i++,set >>= 1) {
931 spin_lock(&files->file_lock);
934 spin_unlock(&files->file_lock);
937 char *get_task_comm(char *buf, struct task_struct *tsk)
939 /* buf must be at least sizeof(tsk->comm) in size */
941 strncpy(buf, tsk->comm, sizeof(tsk->comm));
946 void set_task_comm(struct task_struct *tsk, char *buf)
949 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
953 int flush_old_exec(struct linux_binprm * bprm)
957 char tcomm[sizeof(current->comm)];
960 * Make sure we have a private signal table and that
961 * we are unassociated from the previous thread group.
963 retval = de_thread(current);
968 * Release all of the old mmap stuff
970 retval = exec_mmap(bprm->mm);
974 bprm->mm = NULL; /* We're using it now */
976 /* This is the point of no return */
977 current->sas_ss_sp = current->sas_ss_size = 0;
979 if (current->euid == current->uid && current->egid == current->gid)
980 set_dumpable(current->mm, 1);
982 set_dumpable(current->mm, suid_dumpable);
984 name = bprm->filename;
986 /* Copies the binary name from after last slash */
987 for (i=0; (ch = *(name++)) != '\0';) {
989 i = 0; /* overwrite what we wrote */
991 if (i < (sizeof(tcomm) - 1))
995 set_task_comm(current, tcomm);
997 current->flags &= ~PF_RANDOMIZE;
1000 /* Set the new mm task size. We have to do that late because it may
1001 * depend on TIF_32BIT which is only updated in flush_thread() on
1002 * some architectures like powerpc
1004 current->mm->task_size = TASK_SIZE;
1006 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1008 set_dumpable(current->mm, suid_dumpable);
1009 current->pdeath_signal = 0;
1010 } else if (file_permission(bprm->file, MAY_READ) ||
1011 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1013 set_dumpable(current->mm, suid_dumpable);
1016 /* An exec changes our domain. We are no longer part of the thread
1019 current->self_exec_id++;
1021 flush_signal_handlers(current, 0);
1022 flush_old_files(current->files);
1030 EXPORT_SYMBOL(flush_old_exec);
1033 * Fill the binprm structure from the inode.
1034 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1036 int prepare_binprm(struct linux_binprm *bprm)
1039 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1042 mode = inode->i_mode;
1043 if (bprm->file->f_op == NULL)
1046 bprm->e_uid = current->euid;
1047 bprm->e_gid = current->egid;
1049 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1051 if (mode & S_ISUID) {
1052 current->personality &= ~PER_CLEAR_ON_SETID;
1053 bprm->e_uid = inode->i_uid;
1058 * If setgid is set but no group execute bit then this
1059 * is a candidate for mandatory locking, not a setgid
1062 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1063 current->personality &= ~PER_CLEAR_ON_SETID;
1064 bprm->e_gid = inode->i_gid;
1068 /* fill in binprm security blob */
1069 retval = security_bprm_set(bprm);
1073 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1074 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1077 EXPORT_SYMBOL(prepare_binprm);
1079 static int unsafe_exec(struct task_struct *p)
1082 if (p->ptrace & PT_PTRACED) {
1083 if (p->ptrace & PT_PTRACE_CAP)
1084 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1086 unsafe |= LSM_UNSAFE_PTRACE;
1088 if (atomic_read(&p->fs->count) > 1 ||
1089 atomic_read(&p->files->count) > 1 ||
1090 atomic_read(&p->sighand->count) > 1)
1091 unsafe |= LSM_UNSAFE_SHARE;
1096 void compute_creds(struct linux_binprm *bprm)
1100 if (bprm->e_uid != current->uid) {
1102 current->pdeath_signal = 0;
1107 unsafe = unsafe_exec(current);
1108 security_bprm_apply_creds(bprm, unsafe);
1109 task_unlock(current);
1110 security_bprm_post_apply_creds(bprm);
1112 EXPORT_SYMBOL(compute_creds);
1115 * Arguments are '\0' separated strings found at the location bprm->p
1116 * points to; chop off the first by relocating brpm->p to right after
1117 * the first '\0' encountered.
1119 int remove_arg_zero(struct linux_binprm *bprm)
1122 unsigned long offset;
1130 offset = bprm->p & ~PAGE_MASK;
1131 page = get_arg_page(bprm, bprm->p, 0);
1136 kaddr = kmap_atomic(page, KM_USER0);
1138 for (; offset < PAGE_SIZE && kaddr[offset];
1139 offset++, bprm->p++)
1142 kunmap_atomic(kaddr, KM_USER0);
1145 if (offset == PAGE_SIZE)
1146 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1147 } while (offset == PAGE_SIZE);
1156 EXPORT_SYMBOL(remove_arg_zero);
1159 * cycle the list of binary formats handler, until one recognizes the image
1161 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1164 struct linux_binfmt *fmt;
1165 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1166 /* handle /sbin/loader.. */
1168 struct exec * eh = (struct exec *) bprm->buf;
1170 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1171 (eh->fh.f_flags & 0x3000) == 0x3000)
1174 unsigned long loader;
1176 allow_write_access(bprm->file);
1180 loader = bprm->vma->vm_end - sizeof(void *);
1182 file = open_exec("/sbin/loader");
1183 retval = PTR_ERR(file);
1187 /* Remember if the application is TASO. */
1188 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1191 bprm->loader = loader;
1192 retval = prepare_binprm(bprm);
1195 /* should call search_binary_handler recursively here,
1196 but it does not matter */
1200 retval = security_bprm_check(bprm);
1204 /* kernel module loader fixup */
1205 /* so we don't try to load run modprobe in kernel space. */
1208 retval = audit_bprm(bprm);
1213 for (try=0; try<2; try++) {
1214 read_lock(&binfmt_lock);
1215 list_for_each_entry(fmt, &formats, lh) {
1216 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1219 if (!try_module_get(fmt->module))
1221 read_unlock(&binfmt_lock);
1222 retval = fn(bprm, regs);
1225 allow_write_access(bprm->file);
1229 current->did_exec = 1;
1230 proc_exec_connector(current);
1233 read_lock(&binfmt_lock);
1235 if (retval != -ENOEXEC || bprm->mm == NULL)
1238 read_unlock(&binfmt_lock);
1242 read_unlock(&binfmt_lock);
1243 if (retval != -ENOEXEC || bprm->mm == NULL) {
1247 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1248 if (printable(bprm->buf[0]) &&
1249 printable(bprm->buf[1]) &&
1250 printable(bprm->buf[2]) &&
1251 printable(bprm->buf[3]))
1252 break; /* -ENOEXEC */
1253 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1260 EXPORT_SYMBOL(search_binary_handler);
1263 * sys_execve() executes a new program.
1265 int do_execve(char * filename,
1266 char __user *__user *argv,
1267 char __user *__user *envp,
1268 struct pt_regs * regs)
1270 struct linux_binprm *bprm;
1272 struct files_struct *displaced;
1275 retval = unshare_files(&displaced);
1280 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1284 file = open_exec(filename);
1285 retval = PTR_ERR(file);
1292 bprm->filename = filename;
1293 bprm->interp = filename;
1295 retval = bprm_mm_init(bprm);
1299 bprm->argc = count(argv, MAX_ARG_STRINGS);
1300 if ((retval = bprm->argc) < 0)
1303 bprm->envc = count(envp, MAX_ARG_STRINGS);
1304 if ((retval = bprm->envc) < 0)
1307 retval = security_bprm_alloc(bprm);
1311 retval = prepare_binprm(bprm);
1315 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1319 bprm->exec = bprm->p;
1320 retval = copy_strings(bprm->envc, envp, bprm);
1324 retval = copy_strings(bprm->argc, argv, bprm);
1328 retval = search_binary_handler(bprm,regs);
1330 /* execve success */
1331 free_arg_pages(bprm);
1332 security_bprm_free(bprm);
1333 acct_update_integrals(current);
1336 put_files_struct(displaced);
1341 free_arg_pages(bprm);
1343 security_bprm_free(bprm);
1351 allow_write_access(bprm->file);
1359 reset_files_struct(displaced);
1364 int set_binfmt(struct linux_binfmt *new)
1366 struct linux_binfmt *old = current->binfmt;
1369 if (!try_module_get(new->module))
1372 current->binfmt = new;
1374 module_put(old->module);
1378 EXPORT_SYMBOL(set_binfmt);
1380 /* format_corename will inspect the pattern parameter, and output a
1381 * name into corename, which must have space for at least
1382 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1384 static int format_corename(char *corename, const char *pattern, long signr)
1386 const char *pat_ptr = pattern;
1387 char *out_ptr = corename;
1388 char *const out_end = corename + CORENAME_MAX_SIZE;
1390 int pid_in_pattern = 0;
1393 if (*pattern == '|')
1396 /* Repeat as long as we have more pattern to process and more output
1399 if (*pat_ptr != '%') {
1400 if (out_ptr == out_end)
1402 *out_ptr++ = *pat_ptr++;
1404 switch (*++pat_ptr) {
1407 /* Double percent, output one percent */
1409 if (out_ptr == out_end)
1416 rc = snprintf(out_ptr, out_end - out_ptr,
1417 "%d", task_tgid_vnr(current));
1418 if (rc > out_end - out_ptr)
1424 rc = snprintf(out_ptr, out_end - out_ptr,
1425 "%d", current->uid);
1426 if (rc > out_end - out_ptr)
1432 rc = snprintf(out_ptr, out_end - out_ptr,
1433 "%d", current->gid);
1434 if (rc > out_end - out_ptr)
1438 /* signal that caused the coredump */
1440 rc = snprintf(out_ptr, out_end - out_ptr,
1442 if (rc > out_end - out_ptr)
1446 /* UNIX time of coredump */
1449 do_gettimeofday(&tv);
1450 rc = snprintf(out_ptr, out_end - out_ptr,
1452 if (rc > out_end - out_ptr)
1459 down_read(&uts_sem);
1460 rc = snprintf(out_ptr, out_end - out_ptr,
1461 "%s", utsname()->nodename);
1463 if (rc > out_end - out_ptr)
1469 rc = snprintf(out_ptr, out_end - out_ptr,
1470 "%s", current->comm);
1471 if (rc > out_end - out_ptr)
1475 /* core limit size */
1477 rc = snprintf(out_ptr, out_end - out_ptr,
1478 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1479 if (rc > out_end - out_ptr)
1489 /* Backward compatibility with core_uses_pid:
1491 * If core_pattern does not include a %p (as is the default)
1492 * and core_uses_pid is set, then .%pid will be appended to
1493 * the filename. Do not do this for piped commands. */
1494 if (!ispipe && !pid_in_pattern
1495 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) {
1496 rc = snprintf(out_ptr, out_end - out_ptr,
1497 ".%d", task_tgid_vnr(current));
1498 if (rc > out_end - out_ptr)
1507 static void zap_process(struct task_struct *start)
1509 struct task_struct *t;
1511 start->signal->flags = SIGNAL_GROUP_EXIT;
1512 start->signal->group_stop_count = 0;
1516 if (t != current && t->mm) {
1517 t->mm->core_waiters++;
1518 sigaddset(&t->pending.signal, SIGKILL);
1519 signal_wake_up(t, 1);
1521 } while ((t = next_thread(t)) != start);
1524 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1527 struct task_struct *g, *p;
1528 unsigned long flags;
1531 spin_lock_irq(&tsk->sighand->siglock);
1532 if (!signal_group_exit(tsk->signal)) {
1533 tsk->signal->group_exit_code = exit_code;
1537 spin_unlock_irq(&tsk->sighand->siglock);
1541 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1545 for_each_process(g) {
1546 if (g == tsk->group_leader)
1554 * p->sighand can't disappear, but
1555 * may be changed by de_thread()
1557 lock_task_sighand(p, &flags);
1559 unlock_task_sighand(p, &flags);
1563 } while ((p = next_thread(p)) != g);
1567 return mm->core_waiters;
1570 static int coredump_wait(int exit_code)
1572 struct task_struct *tsk = current;
1573 struct mm_struct *mm = tsk->mm;
1574 struct completion startup_done;
1575 struct completion *vfork_done;
1578 init_completion(&mm->core_done);
1579 init_completion(&startup_done);
1580 mm->core_startup_done = &startup_done;
1582 core_waiters = zap_threads(tsk, mm, exit_code);
1583 up_write(&mm->mmap_sem);
1585 if (unlikely(core_waiters < 0))
1589 * Make sure nobody is waiting for us to release the VM,
1590 * otherwise we can deadlock when we wait on each other
1592 vfork_done = tsk->vfork_done;
1594 tsk->vfork_done = NULL;
1595 complete(vfork_done);
1599 wait_for_completion(&startup_done);
1601 BUG_ON(mm->core_waiters);
1602 return core_waiters;
1606 * set_dumpable converts traditional three-value dumpable to two flags and
1607 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1608 * these bits are not changed atomically. So get_dumpable can observe the
1609 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1610 * return either old dumpable or new one by paying attention to the order of
1611 * modifying the bits.
1613 * dumpable | mm->flags (binary)
1614 * old new | initial interim final
1615 * ---------+-----------------------
1623 * (*) get_dumpable regards interim value of 10 as 11.
1625 void set_dumpable(struct mm_struct *mm, int value)
1629 clear_bit(MMF_DUMPABLE, &mm->flags);
1631 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1634 set_bit(MMF_DUMPABLE, &mm->flags);
1636 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1639 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1641 set_bit(MMF_DUMPABLE, &mm->flags);
1646 int get_dumpable(struct mm_struct *mm)
1650 ret = mm->flags & 0x3;
1651 return (ret >= 2) ? 2 : ret;
1654 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1656 char corename[CORENAME_MAX_SIZE + 1];
1657 struct mm_struct *mm = current->mm;
1658 struct linux_binfmt * binfmt;
1659 struct inode * inode;
1662 int fsuid = current->fsuid;
1665 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1666 char **helper_argv = NULL;
1667 int helper_argc = 0;
1670 audit_core_dumps(signr);
1672 binfmt = current->binfmt;
1673 if (!binfmt || !binfmt->core_dump)
1675 down_write(&mm->mmap_sem);
1677 * If another thread got here first, or we are not dumpable, bail out.
1679 if (mm->core_waiters || !get_dumpable(mm)) {
1680 up_write(&mm->mmap_sem);
1685 * We cannot trust fsuid as being the "true" uid of the
1686 * process nor do we know its entire history. We only know it
1687 * was tainted so we dump it as root in mode 2.
1689 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1690 flag = O_EXCL; /* Stop rewrite attacks */
1691 current->fsuid = 0; /* Dump root private */
1694 retval = coredump_wait(exit_code);
1699 * Clear any false indication of pending signals that might
1700 * be seen by the filesystem code called to write the core file.
1702 clear_thread_flag(TIF_SIGPENDING);
1705 * lock_kernel() because format_corename() is controlled by sysctl, which
1706 * uses lock_kernel()
1709 ispipe = format_corename(corename, core_pattern, signr);
1712 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1713 * to a pipe. Since we're not writing directly to the filesystem
1714 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1715 * created unless the pipe reader choses to write out the core file
1716 * at which point file size limits and permissions will be imposed
1717 * as it does with any other process
1719 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1723 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1724 /* Terminate the string before the first option */
1725 delimit = strchr(corename, ' ');
1728 delimit = strrchr(helper_argv[0], '/');
1732 delimit = helper_argv[0];
1733 if (!strcmp(delimit, current->comm)) {
1734 printk(KERN_NOTICE "Recursive core dump detected, "
1739 core_limit = RLIM_INFINITY;
1741 /* SIGPIPE can happen, but it's just never processed */
1742 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1744 printk(KERN_INFO "Core dump to %s pipe failed\n",
1749 file = filp_open(corename,
1750 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1754 inode = file->f_path.dentry->d_inode;
1755 if (inode->i_nlink > 1)
1756 goto close_fail; /* multiple links - don't dump */
1757 if (!ispipe && d_unhashed(file->f_path.dentry))
1760 /* AK: actually i see no reason to not allow this for named pipes etc.,
1761 but keep the previous behaviour for now. */
1762 if (!ispipe && !S_ISREG(inode->i_mode))
1765 * Dont allow local users get cute and trick others to coredump
1766 * into their pre-created files:
1768 if (inode->i_uid != current->fsuid)
1772 if (!file->f_op->write)
1774 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1777 retval = binfmt->core_dump(signr, regs, file, core_limit);
1780 current->signal->group_exit_code |= 0x80;
1782 filp_close(file, NULL);
1785 argv_free(helper_argv);
1787 current->fsuid = fsuid;
1788 complete_all(&mm->core_done);