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>
55 #include <linux/fsnotify.h>
57 #include <asm/uaccess.h>
58 #include <asm/mmu_context.h>
63 char core_pattern[CORENAME_MAX_SIZE] = "core";
64 int suid_dumpable = 0;
66 /* The maximal length of core_pattern is also specified in sysctl.c */
68 static LIST_HEAD(formats);
69 static DEFINE_RWLOCK(binfmt_lock);
71 int register_binfmt(struct linux_binfmt * fmt)
75 write_lock(&binfmt_lock);
76 list_add(&fmt->lh, &formats);
77 write_unlock(&binfmt_lock);
81 EXPORT_SYMBOL(register_binfmt);
83 void unregister_binfmt(struct linux_binfmt * fmt)
85 write_lock(&binfmt_lock);
87 write_unlock(&binfmt_lock);
90 EXPORT_SYMBOL(unregister_binfmt);
92 static inline void put_binfmt(struct linux_binfmt * fmt)
94 module_put(fmt->module);
98 * Note that a shared library must be both readable and executable due to
101 * Also note that we take the address to load from from the file itself.
103 asmlinkage long sys_uselib(const char __user * library)
107 char *tmp = getname(library);
108 int error = PTR_ERR(tmp);
111 error = path_lookup_open(AT_FDCWD, tmp,
113 FMODE_READ|FMODE_EXEC);
120 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
124 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
127 error = inode_permission(nd.path.dentry->d_inode,
128 MAY_READ | MAY_EXEC | MAY_OPEN);
132 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
133 error = PTR_ERR(file);
137 fsnotify_open(file->f_path.dentry);
141 struct linux_binfmt * fmt;
143 read_lock(&binfmt_lock);
144 list_for_each_entry(fmt, &formats, lh) {
145 if (!fmt->load_shlib)
147 if (!try_module_get(fmt->module))
149 read_unlock(&binfmt_lock);
150 error = fmt->load_shlib(file);
151 read_lock(&binfmt_lock);
153 if (error != -ENOEXEC)
156 read_unlock(&binfmt_lock);
162 release_open_intent(&nd);
169 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
175 #ifdef CONFIG_STACK_GROWSUP
177 ret = expand_stack_downwards(bprm->vma, pos);
182 ret = get_user_pages(current, bprm->mm, pos,
183 1, write, 1, &page, NULL);
188 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
192 * We've historically supported up to 32 pages (ARG_MAX)
193 * of argument strings even with small stacks
199 * Limit to 1/4-th the stack size for the argv+env strings.
201 * - the remaining binfmt code will not run out of stack space,
202 * - the program will have a reasonable amount of stack left
205 rlim = current->signal->rlim;
206 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
215 static void put_arg_page(struct page *page)
220 static void free_arg_page(struct linux_binprm *bprm, int i)
224 static void free_arg_pages(struct linux_binprm *bprm)
228 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
231 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
234 static int __bprm_mm_init(struct linux_binprm *bprm)
237 struct vm_area_struct *vma = NULL;
238 struct mm_struct *mm = bprm->mm;
240 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
244 down_write(&mm->mmap_sem);
248 * Place the stack at the largest stack address the architecture
249 * supports. Later, we'll move this to an appropriate place. We don't
250 * use STACK_TOP because that can depend on attributes which aren't
253 vma->vm_end = STACK_TOP_MAX;
254 vma->vm_start = vma->vm_end - PAGE_SIZE;
255 vma->vm_flags = VM_STACK_FLAGS;
256 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
257 err = insert_vm_struct(mm, vma);
261 mm->stack_vm = mm->total_vm = 1;
262 up_write(&mm->mmap_sem);
263 bprm->p = vma->vm_end - sizeof(void *);
266 up_write(&mm->mmap_sem);
268 kmem_cache_free(vm_area_cachep, vma);
272 static bool valid_arg_len(struct linux_binprm *bprm, long len)
274 return len <= MAX_ARG_STRLEN;
279 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
284 page = bprm->page[pos / PAGE_SIZE];
285 if (!page && write) {
286 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
289 bprm->page[pos / PAGE_SIZE] = page;
295 static void put_arg_page(struct page *page)
299 static void free_arg_page(struct linux_binprm *bprm, int i)
302 __free_page(bprm->page[i]);
303 bprm->page[i] = NULL;
307 static void free_arg_pages(struct linux_binprm *bprm)
311 for (i = 0; i < MAX_ARG_PAGES; i++)
312 free_arg_page(bprm, i);
315 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
320 static int __bprm_mm_init(struct linux_binprm *bprm)
322 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
326 static bool valid_arg_len(struct linux_binprm *bprm, long len)
328 return len <= bprm->p;
331 #endif /* CONFIG_MMU */
334 * Create a new mm_struct and populate it with a temporary stack
335 * vm_area_struct. We don't have enough context at this point to set the stack
336 * flags, permissions, and offset, so we use temporary values. We'll update
337 * them later in setup_arg_pages().
339 int bprm_mm_init(struct linux_binprm *bprm)
342 struct mm_struct *mm = NULL;
344 bprm->mm = mm = mm_alloc();
349 err = init_new_context(current, mm);
353 err = __bprm_mm_init(bprm);
369 * count() counts the number of strings in array ARGV.
371 static int count(char __user * __user * argv, int max)
379 if (get_user(p, argv))
393 * 'copy_strings()' copies argument/environment strings from the old
394 * processes's memory to the new process's stack. The call to get_user_pages()
395 * ensures the destination page is created and not swapped out.
397 static int copy_strings(int argc, char __user * __user * argv,
398 struct linux_binprm *bprm)
400 struct page *kmapped_page = NULL;
402 unsigned long kpos = 0;
410 if (get_user(str, argv+argc) ||
411 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
416 if (!valid_arg_len(bprm, len)) {
421 /* We're going to work our way backwords. */
427 int offset, bytes_to_copy;
429 offset = pos % PAGE_SIZE;
433 bytes_to_copy = offset;
434 if (bytes_to_copy > len)
437 offset -= bytes_to_copy;
438 pos -= bytes_to_copy;
439 str -= bytes_to_copy;
440 len -= bytes_to_copy;
442 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
445 page = get_arg_page(bprm, pos, 1);
452 flush_kernel_dcache_page(kmapped_page);
453 kunmap(kmapped_page);
454 put_arg_page(kmapped_page);
457 kaddr = kmap(kmapped_page);
458 kpos = pos & PAGE_MASK;
459 flush_arg_page(bprm, kpos, kmapped_page);
461 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
470 flush_kernel_dcache_page(kmapped_page);
471 kunmap(kmapped_page);
472 put_arg_page(kmapped_page);
478 * Like copy_strings, but get argv and its values from kernel memory.
480 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
483 mm_segment_t oldfs = get_fs();
485 r = copy_strings(argc, (char __user * __user *)argv, bprm);
489 EXPORT_SYMBOL(copy_strings_kernel);
494 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
495 * the binfmt code determines where the new stack should reside, we shift it to
496 * its final location. The process proceeds as follows:
498 * 1) Use shift to calculate the new vma endpoints.
499 * 2) Extend vma to cover both the old and new ranges. This ensures the
500 * arguments passed to subsequent functions are consistent.
501 * 3) Move vma's page tables to the new range.
502 * 4) Free up any cleared pgd range.
503 * 5) Shrink the vma to cover only the new range.
505 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
507 struct mm_struct *mm = vma->vm_mm;
508 unsigned long old_start = vma->vm_start;
509 unsigned long old_end = vma->vm_end;
510 unsigned long length = old_end - old_start;
511 unsigned long new_start = old_start - shift;
512 unsigned long new_end = old_end - shift;
513 struct mmu_gather *tlb;
515 BUG_ON(new_start > new_end);
518 * ensure there are no vmas between where we want to go
521 if (vma != find_vma(mm, new_start))
525 * cover the whole range: [new_start, old_end)
527 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
530 * move the page tables downwards, on failure we rely on
531 * process cleanup to remove whatever mess we made.
533 if (length != move_page_tables(vma, old_start,
534 vma, new_start, length))
538 tlb = tlb_gather_mmu(mm, 0);
539 if (new_end > old_start) {
541 * when the old and new regions overlap clear from new_end.
543 free_pgd_range(tlb, new_end, old_end, new_end,
544 vma->vm_next ? vma->vm_next->vm_start : 0);
547 * otherwise, clean from old_start; this is done to not touch
548 * the address space in [new_end, old_start) some architectures
549 * have constraints on va-space that make this illegal (IA64) -
550 * for the others its just a little faster.
552 free_pgd_range(tlb, old_start, old_end, new_end,
553 vma->vm_next ? vma->vm_next->vm_start : 0);
555 tlb_finish_mmu(tlb, new_end, old_end);
558 * shrink the vma to just the new range.
560 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
565 #define EXTRA_STACK_VM_PAGES 20 /* random */
568 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
569 * the stack is optionally relocated, and some extra space is added.
571 int setup_arg_pages(struct linux_binprm *bprm,
572 unsigned long stack_top,
573 int executable_stack)
576 unsigned long stack_shift;
577 struct mm_struct *mm = current->mm;
578 struct vm_area_struct *vma = bprm->vma;
579 struct vm_area_struct *prev = NULL;
580 unsigned long vm_flags;
581 unsigned long stack_base;
583 #ifdef CONFIG_STACK_GROWSUP
584 /* Limit stack size to 1GB */
585 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
586 if (stack_base > (1 << 30))
587 stack_base = 1 << 30;
589 /* Make sure we didn't let the argument array grow too large. */
590 if (vma->vm_end - vma->vm_start > stack_base)
593 stack_base = PAGE_ALIGN(stack_top - stack_base);
595 stack_shift = vma->vm_start - stack_base;
596 mm->arg_start = bprm->p - stack_shift;
597 bprm->p = vma->vm_end - stack_shift;
599 stack_top = arch_align_stack(stack_top);
600 stack_top = PAGE_ALIGN(stack_top);
601 stack_shift = vma->vm_end - stack_top;
603 bprm->p -= stack_shift;
604 mm->arg_start = bprm->p;
608 bprm->loader -= stack_shift;
609 bprm->exec -= stack_shift;
611 down_write(&mm->mmap_sem);
612 vm_flags = VM_STACK_FLAGS;
615 * Adjust stack execute permissions; explicitly enable for
616 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
617 * (arch default) otherwise.
619 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
621 else if (executable_stack == EXSTACK_DISABLE_X)
622 vm_flags &= ~VM_EXEC;
623 vm_flags |= mm->def_flags;
625 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
631 /* Move stack pages down in memory. */
633 ret = shift_arg_pages(vma, stack_shift);
635 up_write(&mm->mmap_sem);
640 #ifdef CONFIG_STACK_GROWSUP
641 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
643 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
645 ret = expand_stack(vma, stack_base);
650 up_write(&mm->mmap_sem);
653 EXPORT_SYMBOL(setup_arg_pages);
655 #endif /* CONFIG_MMU */
657 struct file *open_exec(const char *name)
663 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
664 FMODE_READ|FMODE_EXEC);
669 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
672 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
675 err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
679 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
683 fsnotify_open(file->f_path.dentry);
685 err = deny_write_access(file);
694 release_open_intent(&nd);
699 EXPORT_SYMBOL(open_exec);
701 int kernel_read(struct file *file, unsigned long offset,
702 char *addr, unsigned long count)
710 /* The cast to a user pointer is valid due to the set_fs() */
711 result = vfs_read(file, (void __user *)addr, count, &pos);
716 EXPORT_SYMBOL(kernel_read);
718 static int exec_mmap(struct mm_struct *mm)
720 struct task_struct *tsk;
721 struct mm_struct * old_mm, *active_mm;
723 /* Notify parent that we're no longer interested in the old VM */
725 old_mm = current->mm;
726 mm_release(tsk, old_mm);
730 * Make sure that if there is a core dump in progress
731 * for the old mm, we get out and die instead of going
732 * through with the exec. We must hold mmap_sem around
733 * checking core_state and changing tsk->mm.
735 down_read(&old_mm->mmap_sem);
736 if (unlikely(old_mm->core_state)) {
737 up_read(&old_mm->mmap_sem);
742 active_mm = tsk->active_mm;
745 activate_mm(active_mm, mm);
747 arch_pick_mmap_layout(mm);
749 up_read(&old_mm->mmap_sem);
750 BUG_ON(active_mm != old_mm);
751 mm_update_next_owner(old_mm);
760 * This function makes sure the current process has its own signal table,
761 * so that flush_signal_handlers can later reset the handlers without
762 * disturbing other processes. (Other processes might share the signal
763 * table via the CLONE_SIGHAND option to clone().)
765 static int de_thread(struct task_struct *tsk)
767 struct signal_struct *sig = tsk->signal;
768 struct sighand_struct *oldsighand = tsk->sighand;
769 spinlock_t *lock = &oldsighand->siglock;
772 if (thread_group_empty(tsk))
773 goto no_thread_group;
776 * Kill all other threads in the thread group.
779 if (signal_group_exit(sig)) {
781 * Another group action in progress, just
782 * return so that the signal is processed.
784 spin_unlock_irq(lock);
787 sig->group_exit_task = tsk;
788 zap_other_threads(tsk);
790 /* Account for the thread group leader hanging around: */
791 count = thread_group_leader(tsk) ? 1 : 2;
792 sig->notify_count = count;
793 while (atomic_read(&sig->count) > count) {
794 __set_current_state(TASK_UNINTERRUPTIBLE);
795 spin_unlock_irq(lock);
799 spin_unlock_irq(lock);
802 * At this point all other threads have exited, all we have to
803 * do is to wait for the thread group leader to become inactive,
804 * and to assume its PID:
806 if (!thread_group_leader(tsk)) {
807 struct task_struct *leader = tsk->group_leader;
809 sig->notify_count = -1; /* for exit_notify() */
811 write_lock_irq(&tasklist_lock);
812 if (likely(leader->exit_state))
814 __set_current_state(TASK_UNINTERRUPTIBLE);
815 write_unlock_irq(&tasklist_lock);
820 * The only record we have of the real-time age of a
821 * process, regardless of execs it's done, is start_time.
822 * All the past CPU time is accumulated in signal_struct
823 * from sister threads now dead. But in this non-leader
824 * exec, nothing survives from the original leader thread,
825 * whose birth marks the true age of this process now.
826 * When we take on its identity by switching to its PID, we
827 * also take its birthdate (always earlier than our own).
829 tsk->start_time = leader->start_time;
831 BUG_ON(!same_thread_group(leader, tsk));
832 BUG_ON(has_group_leader_pid(tsk));
834 * An exec() starts a new thread group with the
835 * TGID of the previous thread group. Rehash the
836 * two threads with a switched PID, and release
837 * the former thread group leader:
840 /* Become a process group leader with the old leader's pid.
841 * The old leader becomes a thread of the this thread group.
842 * Note: The old leader also uses this pid until release_task
843 * is called. Odd but simple and correct.
845 detach_pid(tsk, PIDTYPE_PID);
846 tsk->pid = leader->pid;
847 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
848 transfer_pid(leader, tsk, PIDTYPE_PGID);
849 transfer_pid(leader, tsk, PIDTYPE_SID);
850 list_replace_rcu(&leader->tasks, &tsk->tasks);
852 tsk->group_leader = tsk;
853 leader->group_leader = tsk;
855 tsk->exit_signal = SIGCHLD;
857 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
858 leader->exit_state = EXIT_DEAD;
859 write_unlock_irq(&tasklist_lock);
861 release_task(leader);
864 sig->group_exit_task = NULL;
865 sig->notify_count = 0;
869 flush_itimer_signals();
871 if (atomic_read(&oldsighand->count) != 1) {
872 struct sighand_struct *newsighand;
874 * This ->sighand is shared with the CLONE_SIGHAND
875 * but not CLONE_THREAD task, switch to the new one.
877 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
881 atomic_set(&newsighand->count, 1);
882 memcpy(newsighand->action, oldsighand->action,
883 sizeof(newsighand->action));
885 write_lock_irq(&tasklist_lock);
886 spin_lock(&oldsighand->siglock);
887 rcu_assign_pointer(tsk->sighand, newsighand);
888 spin_unlock(&oldsighand->siglock);
889 write_unlock_irq(&tasklist_lock);
891 __cleanup_sighand(oldsighand);
894 BUG_ON(!thread_group_leader(tsk));
899 * These functions flushes out all traces of the currently running executable
900 * so that a new one can be started
902 static void flush_old_files(struct files_struct * files)
907 spin_lock(&files->file_lock);
909 unsigned long set, i;
913 fdt = files_fdtable(files);
914 if (i >= fdt->max_fds)
916 set = fdt->close_on_exec->fds_bits[j];
919 fdt->close_on_exec->fds_bits[j] = 0;
920 spin_unlock(&files->file_lock);
921 for ( ; set ; i++,set >>= 1) {
926 spin_lock(&files->file_lock);
929 spin_unlock(&files->file_lock);
932 char *get_task_comm(char *buf, struct task_struct *tsk)
934 /* buf must be at least sizeof(tsk->comm) in size */
936 strncpy(buf, tsk->comm, sizeof(tsk->comm));
941 void set_task_comm(struct task_struct *tsk, char *buf)
944 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
948 int flush_old_exec(struct linux_binprm * bprm)
952 char tcomm[sizeof(current->comm)];
955 * Make sure we have a private signal table and that
956 * we are unassociated from the previous thread group.
958 retval = de_thread(current);
962 set_mm_exe_file(bprm->mm, bprm->file);
965 * Release all of the old mmap stuff
967 retval = exec_mmap(bprm->mm);
971 bprm->mm = NULL; /* We're using it now */
973 /* This is the point of no return */
974 current->sas_ss_sp = current->sas_ss_size = 0;
976 if (current_euid() == current_uid() && current_egid() == current_gid())
977 set_dumpable(current->mm, 1);
979 set_dumpable(current->mm, suid_dumpable);
981 name = bprm->filename;
983 /* Copies the binary name from after last slash */
984 for (i=0; (ch = *(name++)) != '\0';) {
986 i = 0; /* overwrite what we wrote */
988 if (i < (sizeof(tcomm) - 1))
992 set_task_comm(current, tcomm);
994 current->flags &= ~PF_RANDOMIZE;
997 /* Set the new mm task size. We have to do that late because it may
998 * depend on TIF_32BIT which is only updated in flush_thread() on
999 * some architectures like powerpc
1001 current->mm->task_size = TASK_SIZE;
1003 /* install the new credentials */
1004 if (bprm->cred->uid != current_euid() ||
1005 bprm->cred->gid != current_egid()) {
1006 current->pdeath_signal = 0;
1007 } else if (file_permission(bprm->file, MAY_READ) ||
1008 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1009 set_dumpable(current->mm, suid_dumpable);
1012 current->personality &= ~bprm->per_clear;
1015 * Flush performance counters when crossing a
1018 if (!get_dumpable(current->mm))
1019 perf_counter_exit_task(current);
1021 /* An exec changes our domain. We are no longer part of the thread
1024 current->self_exec_id++;
1026 flush_signal_handlers(current, 0);
1027 flush_old_files(current->files);
1035 EXPORT_SYMBOL(flush_old_exec);
1038 * install the new credentials for this executable
1040 void install_exec_creds(struct linux_binprm *bprm)
1042 security_bprm_committing_creds(bprm);
1044 commit_creds(bprm->cred);
1047 /* cred_exec_mutex must be held at least to this point to prevent
1048 * ptrace_attach() from altering our determination of the task's
1049 * credentials; any time after this it may be unlocked */
1051 security_bprm_committed_creds(bprm);
1053 EXPORT_SYMBOL(install_exec_creds);
1056 * determine how safe it is to execute the proposed program
1057 * - the caller must hold current->cred_exec_mutex to protect against
1060 void check_unsafe_exec(struct linux_binprm *bprm)
1062 struct task_struct *p = current;
1064 bprm->unsafe = tracehook_unsafe_exec(p);
1066 if (atomic_read(&p->fs->count) > 1 ||
1067 atomic_read(&p->files->count) > 1 ||
1068 atomic_read(&p->sighand->count) > 1)
1069 bprm->unsafe |= LSM_UNSAFE_SHARE;
1073 * Fill the binprm structure from the inode.
1074 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1076 * This may be called multiple times for binary chains (scripts for example).
1078 int prepare_binprm(struct linux_binprm *bprm)
1081 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1084 mode = inode->i_mode;
1085 if (bprm->file->f_op == NULL)
1088 /* clear any previous set[ug]id data from a previous binary */
1089 bprm->cred->euid = current_euid();
1090 bprm->cred->egid = current_egid();
1092 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1094 if (mode & S_ISUID) {
1095 bprm->per_clear |= PER_CLEAR_ON_SETID;
1096 bprm->cred->euid = inode->i_uid;
1101 * If setgid is set but no group execute bit then this
1102 * is a candidate for mandatory locking, not a setgid
1105 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1106 bprm->per_clear |= PER_CLEAR_ON_SETID;
1107 bprm->cred->egid = inode->i_gid;
1111 /* fill in binprm security blob */
1112 retval = security_bprm_set_creds(bprm);
1115 bprm->cred_prepared = 1;
1117 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1118 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1121 EXPORT_SYMBOL(prepare_binprm);
1124 * Arguments are '\0' separated strings found at the location bprm->p
1125 * points to; chop off the first by relocating brpm->p to right after
1126 * the first '\0' encountered.
1128 int remove_arg_zero(struct linux_binprm *bprm)
1131 unsigned long offset;
1139 offset = bprm->p & ~PAGE_MASK;
1140 page = get_arg_page(bprm, bprm->p, 0);
1145 kaddr = kmap_atomic(page, KM_USER0);
1147 for (; offset < PAGE_SIZE && kaddr[offset];
1148 offset++, bprm->p++)
1151 kunmap_atomic(kaddr, KM_USER0);
1154 if (offset == PAGE_SIZE)
1155 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1156 } while (offset == PAGE_SIZE);
1165 EXPORT_SYMBOL(remove_arg_zero);
1168 * cycle the list of binary formats handler, until one recognizes the image
1170 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1172 unsigned int depth = bprm->recursion_depth;
1174 struct linux_binfmt *fmt;
1176 retval = security_bprm_check(bprm);
1180 /* kernel module loader fixup */
1181 /* so we don't try to load run modprobe in kernel space. */
1184 retval = audit_bprm(bprm);
1189 for (try=0; try<2; try++) {
1190 read_lock(&binfmt_lock);
1191 list_for_each_entry(fmt, &formats, lh) {
1192 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1195 if (!try_module_get(fmt->module))
1197 read_unlock(&binfmt_lock);
1198 retval = fn(bprm, regs);
1200 * Restore the depth counter to its starting value
1201 * in this call, so we don't have to rely on every
1202 * load_binary function to restore it on return.
1204 bprm->recursion_depth = depth;
1207 tracehook_report_exec(fmt, bprm, regs);
1209 allow_write_access(bprm->file);
1213 current->did_exec = 1;
1214 proc_exec_connector(current);
1217 read_lock(&binfmt_lock);
1219 if (retval != -ENOEXEC || bprm->mm == NULL)
1222 read_unlock(&binfmt_lock);
1226 read_unlock(&binfmt_lock);
1227 if (retval != -ENOEXEC || bprm->mm == NULL) {
1229 #ifdef CONFIG_MODULES
1231 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1232 if (printable(bprm->buf[0]) &&
1233 printable(bprm->buf[1]) &&
1234 printable(bprm->buf[2]) &&
1235 printable(bprm->buf[3]))
1236 break; /* -ENOEXEC */
1237 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1244 EXPORT_SYMBOL(search_binary_handler);
1246 void free_bprm(struct linux_binprm *bprm)
1248 free_arg_pages(bprm);
1250 abort_creds(bprm->cred);
1255 * sys_execve() executes a new program.
1257 int do_execve(char * filename,
1258 char __user *__user *argv,
1259 char __user *__user *envp,
1260 struct pt_regs * regs)
1262 struct linux_binprm *bprm;
1264 struct files_struct *displaced;
1267 retval = unshare_files(&displaced);
1272 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1276 retval = mutex_lock_interruptible(¤t->cred_exec_mutex);
1281 bprm->cred = prepare_exec_creds();
1284 check_unsafe_exec(bprm);
1286 file = open_exec(filename);
1287 retval = PTR_ERR(file);
1294 bprm->filename = filename;
1295 bprm->interp = filename;
1297 retval = bprm_mm_init(bprm);
1301 bprm->argc = count(argv, MAX_ARG_STRINGS);
1302 if ((retval = bprm->argc) < 0)
1305 bprm->envc = count(envp, MAX_ARG_STRINGS);
1306 if ((retval = bprm->envc) < 0)
1309 retval = prepare_binprm(bprm);
1313 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1317 bprm->exec = bprm->p;
1318 retval = copy_strings(bprm->envc, envp, bprm);
1322 retval = copy_strings(bprm->argc, argv, bprm);
1326 current->flags &= ~PF_KTHREAD;
1327 retval = search_binary_handler(bprm,regs);
1331 /* execve succeeded */
1332 mutex_unlock(¤t->cred_exec_mutex);
1333 acct_update_integrals(current);
1336 put_files_struct(displaced);
1345 allow_write_access(bprm->file);
1350 mutex_unlock(¤t->cred_exec_mutex);
1357 reset_files_struct(displaced);
1362 int set_binfmt(struct linux_binfmt *new)
1364 struct linux_binfmt *old = current->binfmt;
1367 if (!try_module_get(new->module))
1370 current->binfmt = new;
1372 module_put(old->module);
1376 EXPORT_SYMBOL(set_binfmt);
1378 /* format_corename will inspect the pattern parameter, and output a
1379 * name into corename, which must have space for at least
1380 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1382 static int format_corename(char *corename, long signr)
1384 const struct cred *cred = current_cred();
1385 const char *pat_ptr = core_pattern;
1386 int ispipe = (*pat_ptr == '|');
1387 char *out_ptr = corename;
1388 char *const out_end = corename + CORENAME_MAX_SIZE;
1390 int pid_in_pattern = 0;
1392 /* Repeat as long as we have more pattern to process and more output
1395 if (*pat_ptr != '%') {
1396 if (out_ptr == out_end)
1398 *out_ptr++ = *pat_ptr++;
1400 switch (*++pat_ptr) {
1403 /* Double percent, output one percent */
1405 if (out_ptr == out_end)
1412 rc = snprintf(out_ptr, out_end - out_ptr,
1413 "%d", task_tgid_vnr(current));
1414 if (rc > out_end - out_ptr)
1420 rc = snprintf(out_ptr, out_end - out_ptr,
1422 if (rc > out_end - out_ptr)
1428 rc = snprintf(out_ptr, out_end - out_ptr,
1430 if (rc > out_end - out_ptr)
1434 /* signal that caused the coredump */
1436 rc = snprintf(out_ptr, out_end - out_ptr,
1438 if (rc > out_end - out_ptr)
1442 /* UNIX time of coredump */
1445 do_gettimeofday(&tv);
1446 rc = snprintf(out_ptr, out_end - out_ptr,
1448 if (rc > out_end - out_ptr)
1455 down_read(&uts_sem);
1456 rc = snprintf(out_ptr, out_end - out_ptr,
1457 "%s", utsname()->nodename);
1459 if (rc > out_end - out_ptr)
1465 rc = snprintf(out_ptr, out_end - out_ptr,
1466 "%s", current->comm);
1467 if (rc > out_end - out_ptr)
1471 /* core limit size */
1473 rc = snprintf(out_ptr, out_end - out_ptr,
1474 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1475 if (rc > out_end - out_ptr)
1485 /* Backward compatibility with core_uses_pid:
1487 * If core_pattern does not include a %p (as is the default)
1488 * and core_uses_pid is set, then .%pid will be appended to
1489 * the filename. Do not do this for piped commands. */
1490 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1491 rc = snprintf(out_ptr, out_end - out_ptr,
1492 ".%d", task_tgid_vnr(current));
1493 if (rc > out_end - out_ptr)
1502 static int zap_process(struct task_struct *start)
1504 struct task_struct *t;
1507 start->signal->flags = SIGNAL_GROUP_EXIT;
1508 start->signal->group_stop_count = 0;
1512 if (t != current && t->mm) {
1513 sigaddset(&t->pending.signal, SIGKILL);
1514 signal_wake_up(t, 1);
1517 } while_each_thread(start, t);
1522 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1523 struct core_state *core_state, int exit_code)
1525 struct task_struct *g, *p;
1526 unsigned long flags;
1529 spin_lock_irq(&tsk->sighand->siglock);
1530 if (!signal_group_exit(tsk->signal)) {
1531 mm->core_state = core_state;
1532 tsk->signal->group_exit_code = exit_code;
1533 nr = zap_process(tsk);
1535 spin_unlock_irq(&tsk->sighand->siglock);
1536 if (unlikely(nr < 0))
1539 if (atomic_read(&mm->mm_users) == nr + 1)
1542 * We should find and kill all tasks which use this mm, and we should
1543 * count them correctly into ->nr_threads. We don't take tasklist
1544 * lock, but this is safe wrt:
1547 * None of sub-threads can fork after zap_process(leader). All
1548 * processes which were created before this point should be
1549 * visible to zap_threads() because copy_process() adds the new
1550 * process to the tail of init_task.tasks list, and lock/unlock
1551 * of ->siglock provides a memory barrier.
1554 * The caller holds mm->mmap_sem. This means that the task which
1555 * uses this mm can't pass exit_mm(), so it can't exit or clear
1559 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1560 * we must see either old or new leader, this does not matter.
1561 * However, it can change p->sighand, so lock_task_sighand(p)
1562 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1565 * Note also that "g" can be the old leader with ->mm == NULL
1566 * and already unhashed and thus removed from ->thread_group.
1567 * This is OK, __unhash_process()->list_del_rcu() does not
1568 * clear the ->next pointer, we will find the new leader via
1572 for_each_process(g) {
1573 if (g == tsk->group_leader)
1575 if (g->flags & PF_KTHREAD)
1580 if (unlikely(p->mm == mm)) {
1581 lock_task_sighand(p, &flags);
1582 nr += zap_process(p);
1583 unlock_task_sighand(p, &flags);
1587 } while_each_thread(g, p);
1591 atomic_set(&core_state->nr_threads, nr);
1595 static int coredump_wait(int exit_code, struct core_state *core_state)
1597 struct task_struct *tsk = current;
1598 struct mm_struct *mm = tsk->mm;
1599 struct completion *vfork_done;
1602 init_completion(&core_state->startup);
1603 core_state->dumper.task = tsk;
1604 core_state->dumper.next = NULL;
1605 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1606 up_write(&mm->mmap_sem);
1608 if (unlikely(core_waiters < 0))
1612 * Make sure nobody is waiting for us to release the VM,
1613 * otherwise we can deadlock when we wait on each other
1615 vfork_done = tsk->vfork_done;
1617 tsk->vfork_done = NULL;
1618 complete(vfork_done);
1622 wait_for_completion(&core_state->startup);
1624 return core_waiters;
1627 static void coredump_finish(struct mm_struct *mm)
1629 struct core_thread *curr, *next;
1630 struct task_struct *task;
1632 next = mm->core_state->dumper.next;
1633 while ((curr = next) != NULL) {
1637 * see exit_mm(), curr->task must not see
1638 * ->task == NULL before we read ->next.
1642 wake_up_process(task);
1645 mm->core_state = NULL;
1649 * set_dumpable converts traditional three-value dumpable to two flags and
1650 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1651 * these bits are not changed atomically. So get_dumpable can observe the
1652 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1653 * return either old dumpable or new one by paying attention to the order of
1654 * modifying the bits.
1656 * dumpable | mm->flags (binary)
1657 * old new | initial interim final
1658 * ---------+-----------------------
1666 * (*) get_dumpable regards interim value of 10 as 11.
1668 void set_dumpable(struct mm_struct *mm, int value)
1672 clear_bit(MMF_DUMPABLE, &mm->flags);
1674 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1677 set_bit(MMF_DUMPABLE, &mm->flags);
1679 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1682 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1684 set_bit(MMF_DUMPABLE, &mm->flags);
1689 int get_dumpable(struct mm_struct *mm)
1693 ret = mm->flags & 0x3;
1694 return (ret >= 2) ? 2 : ret;
1697 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1699 struct core_state core_state;
1700 char corename[CORENAME_MAX_SIZE + 1];
1701 struct mm_struct *mm = current->mm;
1702 struct linux_binfmt * binfmt;
1703 struct inode * inode;
1705 const struct cred *old_cred;
1710 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1711 char **helper_argv = NULL;
1712 int helper_argc = 0;
1715 audit_core_dumps(signr);
1717 binfmt = current->binfmt;
1718 if (!binfmt || !binfmt->core_dump)
1721 cred = prepare_creds();
1727 down_write(&mm->mmap_sem);
1729 * If another thread got here first, or we are not dumpable, bail out.
1731 if (mm->core_state || !get_dumpable(mm)) {
1732 up_write(&mm->mmap_sem);
1738 * We cannot trust fsuid as being the "true" uid of the
1739 * process nor do we know its entire history. We only know it
1740 * was tainted so we dump it as root in mode 2.
1742 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1743 flag = O_EXCL; /* Stop rewrite attacks */
1744 cred->fsuid = 0; /* Dump root private */
1747 retval = coredump_wait(exit_code, &core_state);
1753 old_cred = override_creds(cred);
1756 * Clear any false indication of pending signals that might
1757 * be seen by the filesystem code called to write the core file.
1759 clear_thread_flag(TIF_SIGPENDING);
1762 * lock_kernel() because format_corename() is controlled by sysctl, which
1763 * uses lock_kernel()
1766 ispipe = format_corename(corename, signr);
1769 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1770 * to a pipe. Since we're not writing directly to the filesystem
1771 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1772 * created unless the pipe reader choses to write out the core file
1773 * at which point file size limits and permissions will be imposed
1774 * as it does with any other process
1776 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1780 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1782 printk(KERN_WARNING "%s failed to allocate memory\n",
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 revert_creds(old_cred);
1851 coredump_finish(mm);