4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_counter.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
56 #include <asm/uaccess.h>
57 #include <asm/mmu_context.h>
62 /* for /sbin/loader handling in search_binary_handler() */
63 #include <linux/a.out.h>
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
68 int suid_dumpable = 0;
70 /* The maximal length of core_pattern is also specified in sysctl.c */
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
75 int register_binfmt(struct linux_binfmt * fmt)
79 write_lock(&binfmt_lock);
80 list_add(&fmt->lh, &formats);
81 write_unlock(&binfmt_lock);
85 EXPORT_SYMBOL(register_binfmt);
87 void unregister_binfmt(struct linux_binfmt * fmt)
89 write_lock(&binfmt_lock);
91 write_unlock(&binfmt_lock);
94 EXPORT_SYMBOL(unregister_binfmt);
96 static inline void put_binfmt(struct linux_binfmt * fmt)
98 module_put(fmt->module);
102 * Note that a shared library must be both readable and executable due to
105 * Also note that we take the address to load from from the file itself.
107 asmlinkage long sys_uselib(const char __user * library)
111 char *tmp = getname(library);
112 int error = PTR_ERR(tmp);
115 error = path_lookup_open(AT_FDCWD, tmp,
117 FMODE_READ|FMODE_EXEC);
124 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
128 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
131 error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
135 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
136 error = PTR_ERR(file);
142 struct linux_binfmt * fmt;
144 read_lock(&binfmt_lock);
145 list_for_each_entry(fmt, &formats, lh) {
146 if (!fmt->load_shlib)
148 if (!try_module_get(fmt->module))
150 read_unlock(&binfmt_lock);
151 error = fmt->load_shlib(file);
152 read_lock(&binfmt_lock);
154 if (error != -ENOEXEC)
157 read_unlock(&binfmt_lock);
163 release_open_intent(&nd);
170 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
176 #ifdef CONFIG_STACK_GROWSUP
178 ret = expand_stack_downwards(bprm->vma, pos);
183 ret = get_user_pages(current, bprm->mm, pos,
184 1, write, 1, &page, NULL);
189 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
193 * We've historically supported up to 32 pages (ARG_MAX)
194 * of argument strings even with small stacks
200 * Limit to 1/4-th the stack size for the argv+env strings.
202 * - the remaining binfmt code will not run out of stack space,
203 * - the program will have a reasonable amount of stack left
206 rlim = current->signal->rlim;
207 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
216 static void put_arg_page(struct page *page)
221 static void free_arg_page(struct linux_binprm *bprm, int i)
225 static void free_arg_pages(struct linux_binprm *bprm)
229 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
232 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
235 static int __bprm_mm_init(struct linux_binprm *bprm)
238 struct vm_area_struct *vma = NULL;
239 struct mm_struct *mm = bprm->mm;
241 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
245 down_write(&mm->mmap_sem);
249 * Place the stack at the largest stack address the architecture
250 * supports. Later, we'll move this to an appropriate place. We don't
251 * use STACK_TOP because that can depend on attributes which aren't
254 vma->vm_end = STACK_TOP_MAX;
255 vma->vm_start = vma->vm_end - PAGE_SIZE;
257 vma->vm_flags = VM_STACK_FLAGS;
258 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
259 err = insert_vm_struct(mm, vma);
261 up_write(&mm->mmap_sem);
265 mm->stack_vm = mm->total_vm = 1;
266 up_write(&mm->mmap_sem);
268 bprm->p = vma->vm_end - sizeof(void *);
275 kmem_cache_free(vm_area_cachep, vma);
281 static bool valid_arg_len(struct linux_binprm *bprm, long len)
283 return len <= MAX_ARG_STRLEN;
288 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
293 page = bprm->page[pos / PAGE_SIZE];
294 if (!page && write) {
295 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
298 bprm->page[pos / PAGE_SIZE] = page;
304 static void put_arg_page(struct page *page)
308 static void free_arg_page(struct linux_binprm *bprm, int i)
311 __free_page(bprm->page[i]);
312 bprm->page[i] = NULL;
316 static void free_arg_pages(struct linux_binprm *bprm)
320 for (i = 0; i < MAX_ARG_PAGES; i++)
321 free_arg_page(bprm, i);
324 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
329 static int __bprm_mm_init(struct linux_binprm *bprm)
331 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
335 static bool valid_arg_len(struct linux_binprm *bprm, long len)
337 return len <= bprm->p;
340 #endif /* CONFIG_MMU */
343 * Create a new mm_struct and populate it with a temporary stack
344 * vm_area_struct. We don't have enough context at this point to set the stack
345 * flags, permissions, and offset, so we use temporary values. We'll update
346 * them later in setup_arg_pages().
348 int bprm_mm_init(struct linux_binprm *bprm)
351 struct mm_struct *mm = NULL;
353 bprm->mm = mm = mm_alloc();
358 err = init_new_context(current, mm);
362 err = __bprm_mm_init(bprm);
378 * count() counts the number of strings in array ARGV.
380 static int count(char __user * __user * argv, int max)
388 if (get_user(p, argv))
402 * 'copy_strings()' copies argument/environment strings from the old
403 * processes's memory to the new process's stack. The call to get_user_pages()
404 * ensures the destination page is created and not swapped out.
406 static int copy_strings(int argc, char __user * __user * argv,
407 struct linux_binprm *bprm)
409 struct page *kmapped_page = NULL;
411 unsigned long kpos = 0;
419 if (get_user(str, argv+argc) ||
420 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
425 if (!valid_arg_len(bprm, len)) {
430 /* We're going to work our way backwords. */
436 int offset, bytes_to_copy;
438 offset = pos % PAGE_SIZE;
442 bytes_to_copy = offset;
443 if (bytes_to_copy > len)
446 offset -= bytes_to_copy;
447 pos -= bytes_to_copy;
448 str -= bytes_to_copy;
449 len -= bytes_to_copy;
451 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
454 page = get_arg_page(bprm, pos, 1);
461 flush_kernel_dcache_page(kmapped_page);
462 kunmap(kmapped_page);
463 put_arg_page(kmapped_page);
466 kaddr = kmap(kmapped_page);
467 kpos = pos & PAGE_MASK;
468 flush_arg_page(bprm, kpos, kmapped_page);
470 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
479 flush_kernel_dcache_page(kmapped_page);
480 kunmap(kmapped_page);
481 put_arg_page(kmapped_page);
487 * Like copy_strings, but get argv and its values from kernel memory.
489 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
492 mm_segment_t oldfs = get_fs();
494 r = copy_strings(argc, (char __user * __user *)argv, bprm);
498 EXPORT_SYMBOL(copy_strings_kernel);
503 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
504 * the binfmt code determines where the new stack should reside, we shift it to
505 * its final location. The process proceeds as follows:
507 * 1) Use shift to calculate the new vma endpoints.
508 * 2) Extend vma to cover both the old and new ranges. This ensures the
509 * arguments passed to subsequent functions are consistent.
510 * 3) Move vma's page tables to the new range.
511 * 4) Free up any cleared pgd range.
512 * 5) Shrink the vma to cover only the new range.
514 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
516 struct mm_struct *mm = vma->vm_mm;
517 unsigned long old_start = vma->vm_start;
518 unsigned long old_end = vma->vm_end;
519 unsigned long length = old_end - old_start;
520 unsigned long new_start = old_start - shift;
521 unsigned long new_end = old_end - shift;
522 struct mmu_gather *tlb;
524 BUG_ON(new_start > new_end);
527 * ensure there are no vmas between where we want to go
530 if (vma != find_vma(mm, new_start))
534 * cover the whole range: [new_start, old_end)
536 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
539 * move the page tables downwards, on failure we rely on
540 * process cleanup to remove whatever mess we made.
542 if (length != move_page_tables(vma, old_start,
543 vma, new_start, length))
547 tlb = tlb_gather_mmu(mm, 0);
548 if (new_end > old_start) {
550 * when the old and new regions overlap clear from new_end.
552 free_pgd_range(tlb, new_end, old_end, new_end,
553 vma->vm_next ? vma->vm_next->vm_start : 0);
556 * otherwise, clean from old_start; this is done to not touch
557 * the address space in [new_end, old_start) some architectures
558 * have constraints on va-space that make this illegal (IA64) -
559 * for the others its just a little faster.
561 free_pgd_range(tlb, old_start, old_end, new_end,
562 vma->vm_next ? vma->vm_next->vm_start : 0);
564 tlb_finish_mmu(tlb, new_end, old_end);
567 * shrink the vma to just the new range.
569 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
574 #define EXTRA_STACK_VM_PAGES 20 /* random */
577 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
578 * the stack is optionally relocated, and some extra space is added.
580 int setup_arg_pages(struct linux_binprm *bprm,
581 unsigned long stack_top,
582 int executable_stack)
585 unsigned long stack_shift;
586 struct mm_struct *mm = current->mm;
587 struct vm_area_struct *vma = bprm->vma;
588 struct vm_area_struct *prev = NULL;
589 unsigned long vm_flags;
590 unsigned long stack_base;
592 #ifdef CONFIG_STACK_GROWSUP
593 /* Limit stack size to 1GB */
594 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
595 if (stack_base > (1 << 30))
596 stack_base = 1 << 30;
598 /* Make sure we didn't let the argument array grow too large. */
599 if (vma->vm_end - vma->vm_start > stack_base)
602 stack_base = PAGE_ALIGN(stack_top - stack_base);
604 stack_shift = vma->vm_start - stack_base;
605 mm->arg_start = bprm->p - stack_shift;
606 bprm->p = vma->vm_end - stack_shift;
608 stack_top = arch_align_stack(stack_top);
609 stack_top = PAGE_ALIGN(stack_top);
610 stack_shift = vma->vm_end - stack_top;
612 bprm->p -= stack_shift;
613 mm->arg_start = bprm->p;
617 bprm->loader -= stack_shift;
618 bprm->exec -= stack_shift;
620 down_write(&mm->mmap_sem);
621 vm_flags = VM_STACK_FLAGS;
624 * Adjust stack execute permissions; explicitly enable for
625 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
626 * (arch default) otherwise.
628 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
630 else if (executable_stack == EXSTACK_DISABLE_X)
631 vm_flags &= ~VM_EXEC;
632 vm_flags |= mm->def_flags;
634 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
640 /* Move stack pages down in memory. */
642 ret = shift_arg_pages(vma, stack_shift);
644 up_write(&mm->mmap_sem);
649 #ifdef CONFIG_STACK_GROWSUP
650 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
652 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
654 ret = expand_stack(vma, stack_base);
659 up_write(&mm->mmap_sem);
662 EXPORT_SYMBOL(setup_arg_pages);
664 #endif /* CONFIG_MMU */
666 struct file *open_exec(const char *name)
672 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
673 FMODE_READ|FMODE_EXEC);
678 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
681 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
684 err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
688 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
692 err = deny_write_access(file);
701 release_open_intent(&nd);
706 EXPORT_SYMBOL(open_exec);
708 int kernel_read(struct file *file, unsigned long offset,
709 char *addr, unsigned long count)
717 /* The cast to a user pointer is valid due to the set_fs() */
718 result = vfs_read(file, (void __user *)addr, count, &pos);
723 EXPORT_SYMBOL(kernel_read);
725 static int exec_mmap(struct mm_struct *mm)
727 struct task_struct *tsk;
728 struct mm_struct * old_mm, *active_mm;
730 /* Notify parent that we're no longer interested in the old VM */
732 old_mm = current->mm;
733 mm_release(tsk, old_mm);
737 * Make sure that if there is a core dump in progress
738 * for the old mm, we get out and die instead of going
739 * through with the exec. We must hold mmap_sem around
740 * checking core_state and changing tsk->mm.
742 down_read(&old_mm->mmap_sem);
743 if (unlikely(old_mm->core_state)) {
744 up_read(&old_mm->mmap_sem);
749 active_mm = tsk->active_mm;
752 activate_mm(active_mm, mm);
754 arch_pick_mmap_layout(mm);
756 up_read(&old_mm->mmap_sem);
757 BUG_ON(active_mm != old_mm);
758 mm_update_next_owner(old_mm);
767 * This function makes sure the current process has its own signal table,
768 * so that flush_signal_handlers can later reset the handlers without
769 * disturbing other processes. (Other processes might share the signal
770 * table via the CLONE_SIGHAND option to clone().)
772 static int de_thread(struct task_struct *tsk)
774 struct signal_struct *sig = tsk->signal;
775 struct sighand_struct *oldsighand = tsk->sighand;
776 spinlock_t *lock = &oldsighand->siglock;
777 struct task_struct *leader = NULL;
780 if (thread_group_empty(tsk))
781 goto no_thread_group;
784 * Kill all other threads in the thread group.
787 if (signal_group_exit(sig)) {
789 * Another group action in progress, just
790 * return so that the signal is processed.
792 spin_unlock_irq(lock);
795 sig->group_exit_task = tsk;
796 zap_other_threads(tsk);
798 /* Account for the thread group leader hanging around: */
799 count = thread_group_leader(tsk) ? 1 : 2;
800 sig->notify_count = count;
801 while (atomic_read(&sig->count) > count) {
802 __set_current_state(TASK_UNINTERRUPTIBLE);
803 spin_unlock_irq(lock);
807 spin_unlock_irq(lock);
810 * At this point all other threads have exited, all we have to
811 * do is to wait for the thread group leader to become inactive,
812 * and to assume its PID:
814 if (!thread_group_leader(tsk)) {
815 leader = tsk->group_leader;
817 sig->notify_count = -1; /* for exit_notify() */
819 write_lock_irq(&tasklist_lock);
820 if (likely(leader->exit_state))
822 __set_current_state(TASK_UNINTERRUPTIBLE);
823 write_unlock_irq(&tasklist_lock);
828 * The only record we have of the real-time age of a
829 * process, regardless of execs it's done, is start_time.
830 * All the past CPU time is accumulated in signal_struct
831 * from sister threads now dead. But in this non-leader
832 * exec, nothing survives from the original leader thread,
833 * whose birth marks the true age of this process now.
834 * When we take on its identity by switching to its PID, we
835 * also take its birthdate (always earlier than our own).
837 tsk->start_time = leader->start_time;
839 BUG_ON(!same_thread_group(leader, tsk));
840 BUG_ON(has_group_leader_pid(tsk));
842 * An exec() starts a new thread group with the
843 * TGID of the previous thread group. Rehash the
844 * two threads with a switched PID, and release
845 * the former thread group leader:
848 /* Become a process group leader with the old leader's pid.
849 * The old leader becomes a thread of the this thread group.
850 * Note: The old leader also uses this pid until release_task
851 * is called. Odd but simple and correct.
853 detach_pid(tsk, PIDTYPE_PID);
854 tsk->pid = leader->pid;
855 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
856 transfer_pid(leader, tsk, PIDTYPE_PGID);
857 transfer_pid(leader, tsk, PIDTYPE_SID);
858 list_replace_rcu(&leader->tasks, &tsk->tasks);
860 tsk->group_leader = tsk;
861 leader->group_leader = tsk;
863 tsk->exit_signal = SIGCHLD;
865 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
866 leader->exit_state = EXIT_DEAD;
868 write_unlock_irq(&tasklist_lock);
871 sig->group_exit_task = NULL;
872 sig->notify_count = 0;
876 flush_itimer_signals();
878 release_task(leader);
880 if (atomic_read(&oldsighand->count) != 1) {
881 struct sighand_struct *newsighand;
883 * This ->sighand is shared with the CLONE_SIGHAND
884 * but not CLONE_THREAD task, switch to the new one.
886 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
890 atomic_set(&newsighand->count, 1);
891 memcpy(newsighand->action, oldsighand->action,
892 sizeof(newsighand->action));
894 write_lock_irq(&tasklist_lock);
895 spin_lock(&oldsighand->siglock);
896 rcu_assign_pointer(tsk->sighand, newsighand);
897 spin_unlock(&oldsighand->siglock);
898 write_unlock_irq(&tasklist_lock);
900 __cleanup_sighand(oldsighand);
903 BUG_ON(!thread_group_leader(tsk));
908 * These functions flushes out all traces of the currently running executable
909 * so that a new one can be started
911 static void flush_old_files(struct files_struct * files)
916 spin_lock(&files->file_lock);
918 unsigned long set, i;
922 fdt = files_fdtable(files);
923 if (i >= fdt->max_fds)
925 set = fdt->close_on_exec->fds_bits[j];
928 fdt->close_on_exec->fds_bits[j] = 0;
929 spin_unlock(&files->file_lock);
930 for ( ; set ; i++,set >>= 1) {
935 spin_lock(&files->file_lock);
938 spin_unlock(&files->file_lock);
941 char *get_task_comm(char *buf, struct task_struct *tsk)
943 /* buf must be at least sizeof(tsk->comm) in size */
945 strncpy(buf, tsk->comm, sizeof(tsk->comm));
950 void set_task_comm(struct task_struct *tsk, char *buf)
953 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
957 int flush_old_exec(struct linux_binprm * bprm)
961 char tcomm[sizeof(current->comm)];
964 * Make sure we have a private signal table and that
965 * we are unassociated from the previous thread group.
967 retval = de_thread(current);
971 set_mm_exe_file(bprm->mm, bprm->file);
974 * Release all of the old mmap stuff
976 retval = exec_mmap(bprm->mm);
980 bprm->mm = NULL; /* We're using it now */
982 /* This is the point of no return */
983 current->sas_ss_sp = current->sas_ss_size = 0;
985 if (current_euid() == current_uid() && current_egid() == current_gid())
986 set_dumpable(current->mm, 1);
988 set_dumpable(current->mm, suid_dumpable);
990 name = bprm->filename;
992 /* Copies the binary name from after last slash */
993 for (i=0; (ch = *(name++)) != '\0';) {
995 i = 0; /* overwrite what we wrote */
997 if (i < (sizeof(tcomm) - 1))
1001 set_task_comm(current, tcomm);
1003 current->flags &= ~PF_RANDOMIZE;
1006 /* Set the new mm task size. We have to do that late because it may
1007 * depend on TIF_32BIT which is only updated in flush_thread() on
1008 * some architectures like powerpc
1010 current->mm->task_size = TASK_SIZE;
1012 /* install the new credentials */
1013 if (bprm->cred->uid != current_euid() ||
1014 bprm->cred->gid != current_egid()) {
1015 current->pdeath_signal = 0;
1016 } else if (file_permission(bprm->file, MAY_READ) ||
1017 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1018 set_dumpable(current->mm, suid_dumpable);
1021 current->personality &= ~bprm->per_clear;
1024 * Flush performance counters when crossing a
1027 if (!get_dumpable(current->mm))
1028 perf_counter_exit_task(current);
1030 /* An exec changes our domain. We are no longer part of the thread
1033 current->self_exec_id++;
1035 flush_signal_handlers(current, 0);
1036 flush_old_files(current->files);
1044 EXPORT_SYMBOL(flush_old_exec);
1047 * install the new credentials for this executable
1049 void install_exec_creds(struct linux_binprm *bprm)
1051 security_bprm_committing_creds(bprm);
1053 commit_creds(bprm->cred);
1056 /* cred_exec_mutex must be held at least to this point to prevent
1057 * ptrace_attach() from altering our determination of the task's
1058 * credentials; any time after this it may be unlocked */
1060 security_bprm_committed_creds(bprm);
1062 EXPORT_SYMBOL(install_exec_creds);
1065 * determine how safe it is to execute the proposed program
1066 * - the caller must hold current->cred_exec_mutex to protect against
1069 void check_unsafe_exec(struct linux_binprm *bprm)
1071 struct task_struct *p = current;
1073 bprm->unsafe = tracehook_unsafe_exec(p);
1075 if (atomic_read(&p->fs->count) > 1 ||
1076 atomic_read(&p->files->count) > 1 ||
1077 atomic_read(&p->sighand->count) > 1)
1078 bprm->unsafe |= LSM_UNSAFE_SHARE;
1082 * Fill the binprm structure from the inode.
1083 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1085 * This may be called multiple times for binary chains (scripts for example).
1087 int prepare_binprm(struct linux_binprm *bprm)
1090 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1093 mode = inode->i_mode;
1094 if (bprm->file->f_op == NULL)
1097 /* clear any previous set[ug]id data from a previous binary */
1098 bprm->cred->euid = current_euid();
1099 bprm->cred->egid = current_egid();
1101 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1103 if (mode & S_ISUID) {
1104 bprm->per_clear |= PER_CLEAR_ON_SETID;
1105 bprm->cred->euid = inode->i_uid;
1110 * If setgid is set but no group execute bit then this
1111 * is a candidate for mandatory locking, not a setgid
1114 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1115 bprm->per_clear |= PER_CLEAR_ON_SETID;
1116 bprm->cred->egid = inode->i_gid;
1120 /* fill in binprm security blob */
1121 retval = security_bprm_set_creds(bprm);
1124 bprm->cred_prepared = 1;
1126 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1127 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1130 EXPORT_SYMBOL(prepare_binprm);
1133 * Arguments are '\0' separated strings found at the location bprm->p
1134 * points to; chop off the first by relocating brpm->p to right after
1135 * the first '\0' encountered.
1137 int remove_arg_zero(struct linux_binprm *bprm)
1140 unsigned long offset;
1148 offset = bprm->p & ~PAGE_MASK;
1149 page = get_arg_page(bprm, bprm->p, 0);
1154 kaddr = kmap_atomic(page, KM_USER0);
1156 for (; offset < PAGE_SIZE && kaddr[offset];
1157 offset++, bprm->p++)
1160 kunmap_atomic(kaddr, KM_USER0);
1163 if (offset == PAGE_SIZE)
1164 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1165 } while (offset == PAGE_SIZE);
1174 EXPORT_SYMBOL(remove_arg_zero);
1177 * cycle the list of binary formats handler, until one recognizes the image
1179 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1181 unsigned int depth = bprm->recursion_depth;
1183 struct linux_binfmt *fmt;
1185 /* handle /sbin/loader.. */
1187 struct exec * eh = (struct exec *) bprm->buf;
1189 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1190 (eh->fh.f_flags & 0x3000) == 0x3000)
1193 unsigned long loader;
1195 allow_write_access(bprm->file);
1199 loader = bprm->vma->vm_end - sizeof(void *);
1201 file = open_exec("/sbin/loader");
1202 retval = PTR_ERR(file);
1206 /* Remember if the application is TASO. */
1207 bprm->taso = eh->ah.entry < 0x100000000UL;
1210 bprm->loader = loader;
1211 retval = prepare_binprm(bprm);
1214 /* should call search_binary_handler recursively here,
1215 but it does not matter */
1219 retval = security_bprm_check(bprm);
1223 /* kernel module loader fixup */
1224 /* so we don't try to load run modprobe in kernel space. */
1227 retval = audit_bprm(bprm);
1232 for (try=0; try<2; try++) {
1233 read_lock(&binfmt_lock);
1234 list_for_each_entry(fmt, &formats, lh) {
1235 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1238 if (!try_module_get(fmt->module))
1240 read_unlock(&binfmt_lock);
1241 retval = fn(bprm, regs);
1243 * Restore the depth counter to its starting value
1244 * in this call, so we don't have to rely on every
1245 * load_binary function to restore it on return.
1247 bprm->recursion_depth = depth;
1250 tracehook_report_exec(fmt, bprm, regs);
1252 allow_write_access(bprm->file);
1256 current->did_exec = 1;
1257 proc_exec_connector(current);
1260 read_lock(&binfmt_lock);
1262 if (retval != -ENOEXEC || bprm->mm == NULL)
1265 read_unlock(&binfmt_lock);
1269 read_unlock(&binfmt_lock);
1270 if (retval != -ENOEXEC || bprm->mm == NULL) {
1272 #ifdef CONFIG_MODULES
1274 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1275 if (printable(bprm->buf[0]) &&
1276 printable(bprm->buf[1]) &&
1277 printable(bprm->buf[2]) &&
1278 printable(bprm->buf[3]))
1279 break; /* -ENOEXEC */
1280 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1287 EXPORT_SYMBOL(search_binary_handler);
1289 void free_bprm(struct linux_binprm *bprm)
1291 free_arg_pages(bprm);
1293 abort_creds(bprm->cred);
1298 * sys_execve() executes a new program.
1300 int do_execve(char * filename,
1301 char __user *__user *argv,
1302 char __user *__user *envp,
1303 struct pt_regs * regs)
1305 struct linux_binprm *bprm;
1307 struct files_struct *displaced;
1310 retval = unshare_files(&displaced);
1315 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1319 retval = mutex_lock_interruptible(¤t->cred_exec_mutex);
1324 bprm->cred = prepare_exec_creds();
1327 check_unsafe_exec(bprm);
1329 file = open_exec(filename);
1330 retval = PTR_ERR(file);
1337 bprm->filename = filename;
1338 bprm->interp = filename;
1340 retval = bprm_mm_init(bprm);
1344 bprm->argc = count(argv, MAX_ARG_STRINGS);
1345 if ((retval = bprm->argc) < 0)
1348 bprm->envc = count(envp, MAX_ARG_STRINGS);
1349 if ((retval = bprm->envc) < 0)
1352 retval = prepare_binprm(bprm);
1356 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1360 bprm->exec = bprm->p;
1361 retval = copy_strings(bprm->envc, envp, bprm);
1365 retval = copy_strings(bprm->argc, argv, bprm);
1369 current->flags &= ~PF_KTHREAD;
1370 retval = search_binary_handler(bprm,regs);
1374 /* execve succeeded */
1375 mutex_unlock(¤t->cred_exec_mutex);
1376 acct_update_integrals(current);
1379 put_files_struct(displaced);
1388 allow_write_access(bprm->file);
1393 mutex_unlock(¤t->cred_exec_mutex);
1400 reset_files_struct(displaced);
1405 int set_binfmt(struct linux_binfmt *new)
1407 struct linux_binfmt *old = current->binfmt;
1410 if (!try_module_get(new->module))
1413 current->binfmt = new;
1415 module_put(old->module);
1419 EXPORT_SYMBOL(set_binfmt);
1421 /* format_corename will inspect the pattern parameter, and output a
1422 * name into corename, which must have space for at least
1423 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1425 static int format_corename(char *corename, long signr)
1427 const struct cred *cred = current_cred();
1428 const char *pat_ptr = core_pattern;
1429 int ispipe = (*pat_ptr == '|');
1430 char *out_ptr = corename;
1431 char *const out_end = corename + CORENAME_MAX_SIZE;
1433 int pid_in_pattern = 0;
1435 /* Repeat as long as we have more pattern to process and more output
1438 if (*pat_ptr != '%') {
1439 if (out_ptr == out_end)
1441 *out_ptr++ = *pat_ptr++;
1443 switch (*++pat_ptr) {
1446 /* Double percent, output one percent */
1448 if (out_ptr == out_end)
1455 rc = snprintf(out_ptr, out_end - out_ptr,
1456 "%d", task_tgid_vnr(current));
1457 if (rc > out_end - out_ptr)
1463 rc = snprintf(out_ptr, out_end - out_ptr,
1465 if (rc > out_end - out_ptr)
1471 rc = snprintf(out_ptr, out_end - out_ptr,
1473 if (rc > out_end - out_ptr)
1477 /* signal that caused the coredump */
1479 rc = snprintf(out_ptr, out_end - out_ptr,
1481 if (rc > out_end - out_ptr)
1485 /* UNIX time of coredump */
1488 do_gettimeofday(&tv);
1489 rc = snprintf(out_ptr, out_end - out_ptr,
1491 if (rc > out_end - out_ptr)
1498 down_read(&uts_sem);
1499 rc = snprintf(out_ptr, out_end - out_ptr,
1500 "%s", utsname()->nodename);
1502 if (rc > out_end - out_ptr)
1508 rc = snprintf(out_ptr, out_end - out_ptr,
1509 "%s", current->comm);
1510 if (rc > out_end - out_ptr)
1514 /* core limit size */
1516 rc = snprintf(out_ptr, out_end - out_ptr,
1517 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1518 if (rc > out_end - out_ptr)
1528 /* Backward compatibility with core_uses_pid:
1530 * If core_pattern does not include a %p (as is the default)
1531 * and core_uses_pid is set, then .%pid will be appended to
1532 * the filename. Do not do this for piped commands. */
1533 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1534 rc = snprintf(out_ptr, out_end - out_ptr,
1535 ".%d", task_tgid_vnr(current));
1536 if (rc > out_end - out_ptr)
1545 static int zap_process(struct task_struct *start)
1547 struct task_struct *t;
1550 start->signal->flags = SIGNAL_GROUP_EXIT;
1551 start->signal->group_stop_count = 0;
1555 if (t != current && t->mm) {
1556 sigaddset(&t->pending.signal, SIGKILL);
1557 signal_wake_up(t, 1);
1560 } while_each_thread(start, t);
1565 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1566 struct core_state *core_state, int exit_code)
1568 struct task_struct *g, *p;
1569 unsigned long flags;
1572 spin_lock_irq(&tsk->sighand->siglock);
1573 if (!signal_group_exit(tsk->signal)) {
1574 mm->core_state = core_state;
1575 tsk->signal->group_exit_code = exit_code;
1576 nr = zap_process(tsk);
1578 spin_unlock_irq(&tsk->sighand->siglock);
1579 if (unlikely(nr < 0))
1582 if (atomic_read(&mm->mm_users) == nr + 1)
1585 * We should find and kill all tasks which use this mm, and we should
1586 * count them correctly into ->nr_threads. We don't take tasklist
1587 * lock, but this is safe wrt:
1590 * None of sub-threads can fork after zap_process(leader). All
1591 * processes which were created before this point should be
1592 * visible to zap_threads() because copy_process() adds the new
1593 * process to the tail of init_task.tasks list, and lock/unlock
1594 * of ->siglock provides a memory barrier.
1597 * The caller holds mm->mmap_sem. This means that the task which
1598 * uses this mm can't pass exit_mm(), so it can't exit or clear
1602 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1603 * we must see either old or new leader, this does not matter.
1604 * However, it can change p->sighand, so lock_task_sighand(p)
1605 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1608 * Note also that "g" can be the old leader with ->mm == NULL
1609 * and already unhashed and thus removed from ->thread_group.
1610 * This is OK, __unhash_process()->list_del_rcu() does not
1611 * clear the ->next pointer, we will find the new leader via
1615 for_each_process(g) {
1616 if (g == tsk->group_leader)
1618 if (g->flags & PF_KTHREAD)
1623 if (unlikely(p->mm == mm)) {
1624 lock_task_sighand(p, &flags);
1625 nr += zap_process(p);
1626 unlock_task_sighand(p, &flags);
1630 } while_each_thread(g, p);
1634 atomic_set(&core_state->nr_threads, nr);
1638 static int coredump_wait(int exit_code, struct core_state *core_state)
1640 struct task_struct *tsk = current;
1641 struct mm_struct *mm = tsk->mm;
1642 struct completion *vfork_done;
1645 init_completion(&core_state->startup);
1646 core_state->dumper.task = tsk;
1647 core_state->dumper.next = NULL;
1648 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1649 up_write(&mm->mmap_sem);
1651 if (unlikely(core_waiters < 0))
1655 * Make sure nobody is waiting for us to release the VM,
1656 * otherwise we can deadlock when we wait on each other
1658 vfork_done = tsk->vfork_done;
1660 tsk->vfork_done = NULL;
1661 complete(vfork_done);
1665 wait_for_completion(&core_state->startup);
1667 return core_waiters;
1670 static void coredump_finish(struct mm_struct *mm)
1672 struct core_thread *curr, *next;
1673 struct task_struct *task;
1675 next = mm->core_state->dumper.next;
1676 while ((curr = next) != NULL) {
1680 * see exit_mm(), curr->task must not see
1681 * ->task == NULL before we read ->next.
1685 wake_up_process(task);
1688 mm->core_state = NULL;
1692 * set_dumpable converts traditional three-value dumpable to two flags and
1693 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1694 * these bits are not changed atomically. So get_dumpable can observe the
1695 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1696 * return either old dumpable or new one by paying attention to the order of
1697 * modifying the bits.
1699 * dumpable | mm->flags (binary)
1700 * old new | initial interim final
1701 * ---------+-----------------------
1709 * (*) get_dumpable regards interim value of 10 as 11.
1711 void set_dumpable(struct mm_struct *mm, int value)
1715 clear_bit(MMF_DUMPABLE, &mm->flags);
1717 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1720 set_bit(MMF_DUMPABLE, &mm->flags);
1722 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1725 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1727 set_bit(MMF_DUMPABLE, &mm->flags);
1732 int get_dumpable(struct mm_struct *mm)
1736 ret = mm->flags & 0x3;
1737 return (ret >= 2) ? 2 : ret;
1740 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1742 struct core_state core_state;
1743 char corename[CORENAME_MAX_SIZE + 1];
1744 struct mm_struct *mm = current->mm;
1745 struct linux_binfmt * binfmt;
1746 struct inode * inode;
1748 const struct cred *old_cred;
1753 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1754 char **helper_argv = NULL;
1755 int helper_argc = 0;
1758 audit_core_dumps(signr);
1760 binfmt = current->binfmt;
1761 if (!binfmt || !binfmt->core_dump)
1764 cred = prepare_creds();
1770 down_write(&mm->mmap_sem);
1772 * If another thread got here first, or we are not dumpable, bail out.
1774 if (mm->core_state || !get_dumpable(mm)) {
1775 up_write(&mm->mmap_sem);
1781 * We cannot trust fsuid as being the "true" uid of the
1782 * process nor do we know its entire history. We only know it
1783 * was tainted so we dump it as root in mode 2.
1785 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1786 flag = O_EXCL; /* Stop rewrite attacks */
1787 cred->fsuid = 0; /* Dump root private */
1790 retval = coredump_wait(exit_code, &core_state);
1796 old_cred = override_creds(cred);
1799 * Clear any false indication of pending signals that might
1800 * be seen by the filesystem code called to write the core file.
1802 clear_thread_flag(TIF_SIGPENDING);
1805 * lock_kernel() because format_corename() is controlled by sysctl, which
1806 * uses lock_kernel()
1809 ispipe = format_corename(corename, signr);
1812 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1813 * to a pipe. Since we're not writing directly to the filesystem
1814 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1815 * created unless the pipe reader choses to write out the core file
1816 * at which point file size limits and permissions will be imposed
1817 * as it does with any other process
1819 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1823 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1824 /* Terminate the string before the first option */
1825 delimit = strchr(corename, ' ');
1828 delimit = strrchr(helper_argv[0], '/');
1832 delimit = helper_argv[0];
1833 if (!strcmp(delimit, current->comm)) {
1834 printk(KERN_NOTICE "Recursive core dump detected, "
1839 core_limit = RLIM_INFINITY;
1841 /* SIGPIPE can happen, but it's just never processed */
1842 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1844 printk(KERN_INFO "Core dump to %s pipe failed\n",
1849 file = filp_open(corename,
1850 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1854 inode = file->f_path.dentry->d_inode;
1855 if (inode->i_nlink > 1)
1856 goto close_fail; /* multiple links - don't dump */
1857 if (!ispipe && d_unhashed(file->f_path.dentry))
1860 /* AK: actually i see no reason to not allow this for named pipes etc.,
1861 but keep the previous behaviour for now. */
1862 if (!ispipe && !S_ISREG(inode->i_mode))
1865 * Dont allow local users get cute and trick others to coredump
1866 * into their pre-created files:
1868 if (inode->i_uid != current_fsuid())
1872 if (!file->f_op->write)
1874 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1877 retval = binfmt->core_dump(signr, regs, file, core_limit);
1880 current->signal->group_exit_code |= 0x80;
1882 filp_close(file, NULL);
1885 argv_free(helper_argv);
1887 revert_creds(old_cred);
1889 coredump_finish(mm);