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/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52 #include <linux/kmod.h>
53 #include <linux/fsnotify.h>
54 #include <linux/fs_struct.h>
55 #include <linux/pipe_fs_i.h>
56 #include <linux/oom.h>
57 #include <linux/compat.h>
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
64 #include <trace/events/task.h>
67 #include <trace/events/sched.h>
70 char core_pattern[CORENAME_MAX_SIZE] = "core";
71 unsigned int core_pipe_limit;
72 int suid_dumpable = 0;
78 static atomic_t call_count = ATOMIC_INIT(1);
80 /* The maximal length of core_pattern is also specified in sysctl.c */
82 static LIST_HEAD(formats);
83 static DEFINE_RWLOCK(binfmt_lock);
85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
88 write_lock(&binfmt_lock);
89 insert ? list_add(&fmt->lh, &formats) :
90 list_add_tail(&fmt->lh, &formats);
91 write_unlock(&binfmt_lock);
94 EXPORT_SYMBOL(__register_binfmt);
96 void unregister_binfmt(struct linux_binfmt * fmt)
98 write_lock(&binfmt_lock);
100 write_unlock(&binfmt_lock);
103 EXPORT_SYMBOL(unregister_binfmt);
105 static inline void put_binfmt(struct linux_binfmt * fmt)
107 module_put(fmt->module);
111 * Note that a shared library must be both readable and executable due to
114 * Also note that we take the address to load from from the file itself.
116 SYSCALL_DEFINE1(uselib, const char __user *, library)
119 char *tmp = getname(library);
120 int error = PTR_ERR(tmp);
121 static const struct open_flags uselib_flags = {
122 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
123 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
124 .intent = LOOKUP_OPEN
130 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
132 error = PTR_ERR(file);
137 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
141 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
148 struct linux_binfmt * fmt;
150 read_lock(&binfmt_lock);
151 list_for_each_entry(fmt, &formats, lh) {
152 if (!fmt->load_shlib)
154 if (!try_module_get(fmt->module))
156 read_unlock(&binfmt_lock);
157 error = fmt->load_shlib(file);
158 read_lock(&binfmt_lock);
160 if (error != -ENOEXEC)
163 read_unlock(&binfmt_lock);
173 * The nascent bprm->mm is not visible until exec_mmap() but it can
174 * use a lot of memory, account these pages in current->mm temporary
175 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
176 * change the counter back via acct_arg_size(0).
178 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
180 struct mm_struct *mm = current->mm;
181 long diff = (long)(pages - bprm->vma_pages);
186 bprm->vma_pages = pages;
187 add_mm_counter(mm, MM_ANONPAGES, diff);
190 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
196 #ifdef CONFIG_STACK_GROWSUP
198 ret = expand_downwards(bprm->vma, pos);
203 ret = get_user_pages(current, bprm->mm, pos,
204 1, write, 1, &page, NULL);
209 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
212 acct_arg_size(bprm, size / PAGE_SIZE);
215 * We've historically supported up to 32 pages (ARG_MAX)
216 * of argument strings even with small stacks
222 * Limit to 1/4-th the stack size for the argv+env strings.
224 * - the remaining binfmt code will not run out of stack space,
225 * - the program will have a reasonable amount of stack left
228 rlim = current->signal->rlim;
229 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
238 static void put_arg_page(struct page *page)
243 static void free_arg_page(struct linux_binprm *bprm, int i)
247 static void free_arg_pages(struct linux_binprm *bprm)
251 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
254 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
257 static int __bprm_mm_init(struct linux_binprm *bprm)
260 struct vm_area_struct *vma = NULL;
261 struct mm_struct *mm = bprm->mm;
263 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
267 down_write(&mm->mmap_sem);
271 * Place the stack at the largest stack address the architecture
272 * supports. Later, we'll move this to an appropriate place. We don't
273 * use STACK_TOP because that can depend on attributes which aren't
276 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
277 vma->vm_end = STACK_TOP_MAX;
278 vma->vm_start = vma->vm_end - PAGE_SIZE;
279 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
280 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
281 INIT_LIST_HEAD(&vma->anon_vma_chain);
283 err = insert_vm_struct(mm, vma);
287 mm->stack_vm = mm->total_vm = 1;
288 up_write(&mm->mmap_sem);
289 bprm->p = vma->vm_end - sizeof(void *);
292 up_write(&mm->mmap_sem);
294 kmem_cache_free(vm_area_cachep, vma);
298 static bool valid_arg_len(struct linux_binprm *bprm, long len)
300 return len <= MAX_ARG_STRLEN;
305 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
309 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
314 page = bprm->page[pos / PAGE_SIZE];
315 if (!page && write) {
316 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
319 bprm->page[pos / PAGE_SIZE] = page;
325 static void put_arg_page(struct page *page)
329 static void free_arg_page(struct linux_binprm *bprm, int i)
332 __free_page(bprm->page[i]);
333 bprm->page[i] = NULL;
337 static void free_arg_pages(struct linux_binprm *bprm)
341 for (i = 0; i < MAX_ARG_PAGES; i++)
342 free_arg_page(bprm, i);
345 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
350 static int __bprm_mm_init(struct linux_binprm *bprm)
352 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
356 static bool valid_arg_len(struct linux_binprm *bprm, long len)
358 return len <= bprm->p;
361 #endif /* CONFIG_MMU */
364 * Create a new mm_struct and populate it with a temporary stack
365 * vm_area_struct. We don't have enough context at this point to set the stack
366 * flags, permissions, and offset, so we use temporary values. We'll update
367 * them later in setup_arg_pages().
369 int bprm_mm_init(struct linux_binprm *bprm)
372 struct mm_struct *mm = NULL;
374 bprm->mm = mm = mm_alloc();
379 err = init_new_context(current, mm);
383 err = __bprm_mm_init(bprm);
398 struct user_arg_ptr {
403 const char __user *const __user *native;
405 compat_uptr_t __user *compat;
410 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
412 const char __user *native;
415 if (unlikely(argv.is_compat)) {
416 compat_uptr_t compat;
418 if (get_user(compat, argv.ptr.compat + nr))
419 return ERR_PTR(-EFAULT);
421 return compat_ptr(compat);
425 if (get_user(native, argv.ptr.native + nr))
426 return ERR_PTR(-EFAULT);
432 * count() counts the number of strings in array ARGV.
434 static int count(struct user_arg_ptr argv, int max)
438 if (argv.ptr.native != NULL) {
440 const char __user *p = get_user_arg_ptr(argv, i);
451 if (fatal_signal_pending(current))
452 return -ERESTARTNOHAND;
460 * 'copy_strings()' copies argument/environment strings from the old
461 * processes's memory to the new process's stack. The call to get_user_pages()
462 * ensures the destination page is created and not swapped out.
464 static int copy_strings(int argc, struct user_arg_ptr argv,
465 struct linux_binprm *bprm)
467 struct page *kmapped_page = NULL;
469 unsigned long kpos = 0;
473 const char __user *str;
478 str = get_user_arg_ptr(argv, argc);
482 len = strnlen_user(str, MAX_ARG_STRLEN);
487 if (!valid_arg_len(bprm, len))
490 /* We're going to work our way backwords. */
496 int offset, bytes_to_copy;
498 if (fatal_signal_pending(current)) {
499 ret = -ERESTARTNOHAND;
504 offset = pos % PAGE_SIZE;
508 bytes_to_copy = offset;
509 if (bytes_to_copy > len)
512 offset -= bytes_to_copy;
513 pos -= bytes_to_copy;
514 str -= bytes_to_copy;
515 len -= bytes_to_copy;
517 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
520 page = get_arg_page(bprm, pos, 1);
527 flush_kernel_dcache_page(kmapped_page);
528 kunmap(kmapped_page);
529 put_arg_page(kmapped_page);
532 kaddr = kmap(kmapped_page);
533 kpos = pos & PAGE_MASK;
534 flush_arg_page(bprm, kpos, kmapped_page);
536 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
545 flush_kernel_dcache_page(kmapped_page);
546 kunmap(kmapped_page);
547 put_arg_page(kmapped_page);
553 * Like copy_strings, but get argv and its values from kernel memory.
555 int copy_strings_kernel(int argc, const char *const *__argv,
556 struct linux_binprm *bprm)
559 mm_segment_t oldfs = get_fs();
560 struct user_arg_ptr argv = {
561 .ptr.native = (const char __user *const __user *)__argv,
565 r = copy_strings(argc, argv, bprm);
570 EXPORT_SYMBOL(copy_strings_kernel);
575 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
576 * the binfmt code determines where the new stack should reside, we shift it to
577 * its final location. The process proceeds as follows:
579 * 1) Use shift to calculate the new vma endpoints.
580 * 2) Extend vma to cover both the old and new ranges. This ensures the
581 * arguments passed to subsequent functions are consistent.
582 * 3) Move vma's page tables to the new range.
583 * 4) Free up any cleared pgd range.
584 * 5) Shrink the vma to cover only the new range.
586 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
588 struct mm_struct *mm = vma->vm_mm;
589 unsigned long old_start = vma->vm_start;
590 unsigned long old_end = vma->vm_end;
591 unsigned long length = old_end - old_start;
592 unsigned long new_start = old_start - shift;
593 unsigned long new_end = old_end - shift;
594 struct mmu_gather tlb;
596 BUG_ON(new_start > new_end);
599 * ensure there are no vmas between where we want to go
602 if (vma != find_vma(mm, new_start))
606 * cover the whole range: [new_start, old_end)
608 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
612 * move the page tables downwards, on failure we rely on
613 * process cleanup to remove whatever mess we made.
615 if (length != move_page_tables(vma, old_start,
616 vma, new_start, length))
620 tlb_gather_mmu(&tlb, mm, 0);
621 if (new_end > old_start) {
623 * when the old and new regions overlap clear from new_end.
625 free_pgd_range(&tlb, new_end, old_end, new_end,
626 vma->vm_next ? vma->vm_next->vm_start : 0);
629 * otherwise, clean from old_start; this is done to not touch
630 * the address space in [new_end, old_start) some architectures
631 * have constraints on va-space that make this illegal (IA64) -
632 * for the others its just a little faster.
634 free_pgd_range(&tlb, old_start, old_end, new_end,
635 vma->vm_next ? vma->vm_next->vm_start : 0);
637 tlb_finish_mmu(&tlb, new_end, old_end);
640 * Shrink the vma to just the new range. Always succeeds.
642 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
648 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
649 * the stack is optionally relocated, and some extra space is added.
651 int setup_arg_pages(struct linux_binprm *bprm,
652 unsigned long stack_top,
653 int executable_stack)
656 unsigned long stack_shift;
657 struct mm_struct *mm = current->mm;
658 struct vm_area_struct *vma = bprm->vma;
659 struct vm_area_struct *prev = NULL;
660 unsigned long vm_flags;
661 unsigned long stack_base;
662 unsigned long stack_size;
663 unsigned long stack_expand;
664 unsigned long rlim_stack;
666 #ifdef CONFIG_STACK_GROWSUP
667 /* Limit stack size to 1GB */
668 stack_base = rlimit_max(RLIMIT_STACK);
669 if (stack_base > (1 << 30))
670 stack_base = 1 << 30;
672 /* Make sure we didn't let the argument array grow too large. */
673 if (vma->vm_end - vma->vm_start > stack_base)
676 stack_base = PAGE_ALIGN(stack_top - stack_base);
678 stack_shift = vma->vm_start - stack_base;
679 mm->arg_start = bprm->p - stack_shift;
680 bprm->p = vma->vm_end - stack_shift;
682 stack_top = arch_align_stack(stack_top);
683 stack_top = PAGE_ALIGN(stack_top);
685 if (unlikely(stack_top < mmap_min_addr) ||
686 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
689 stack_shift = vma->vm_end - stack_top;
691 bprm->p -= stack_shift;
692 mm->arg_start = bprm->p;
696 bprm->loader -= stack_shift;
697 bprm->exec -= stack_shift;
699 down_write(&mm->mmap_sem);
700 vm_flags = VM_STACK_FLAGS;
703 * Adjust stack execute permissions; explicitly enable for
704 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
705 * (arch default) otherwise.
707 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
709 else if (executable_stack == EXSTACK_DISABLE_X)
710 vm_flags &= ~VM_EXEC;
711 vm_flags |= mm->def_flags;
712 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
714 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
720 /* Move stack pages down in memory. */
722 ret = shift_arg_pages(vma, stack_shift);
727 /* mprotect_fixup is overkill to remove the temporary stack flags */
728 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
730 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
731 stack_size = vma->vm_end - vma->vm_start;
733 * Align this down to a page boundary as expand_stack
736 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
737 #ifdef CONFIG_STACK_GROWSUP
738 if (stack_size + stack_expand > rlim_stack)
739 stack_base = vma->vm_start + rlim_stack;
741 stack_base = vma->vm_end + stack_expand;
743 if (stack_size + stack_expand > rlim_stack)
744 stack_base = vma->vm_end - rlim_stack;
746 stack_base = vma->vm_start - stack_expand;
748 current->mm->start_stack = bprm->p;
749 ret = expand_stack(vma, stack_base);
754 up_write(&mm->mmap_sem);
757 EXPORT_SYMBOL(setup_arg_pages);
759 #endif /* CONFIG_MMU */
761 struct file *open_exec(const char *name)
765 static const struct open_flags open_exec_flags = {
766 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
767 .acc_mode = MAY_EXEC | MAY_OPEN,
768 .intent = LOOKUP_OPEN
771 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
776 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
779 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
784 err = deny_write_access(file);
795 EXPORT_SYMBOL(open_exec);
797 int kernel_read(struct file *file, loff_t offset,
798 char *addr, unsigned long count)
806 /* The cast to a user pointer is valid due to the set_fs() */
807 result = vfs_read(file, (void __user *)addr, count, &pos);
812 EXPORT_SYMBOL(kernel_read);
814 static int exec_mmap(struct mm_struct *mm)
816 struct task_struct *tsk;
817 struct mm_struct * old_mm, *active_mm;
819 /* Notify parent that we're no longer interested in the old VM */
821 old_mm = current->mm;
822 mm_release(tsk, old_mm);
827 * Make sure that if there is a core dump in progress
828 * for the old mm, we get out and die instead of going
829 * through with the exec. We must hold mmap_sem around
830 * checking core_state and changing tsk->mm.
832 down_read(&old_mm->mmap_sem);
833 if (unlikely(old_mm->core_state)) {
834 up_read(&old_mm->mmap_sem);
839 active_mm = tsk->active_mm;
842 activate_mm(active_mm, mm);
844 arch_pick_mmap_layout(mm);
846 up_read(&old_mm->mmap_sem);
847 BUG_ON(active_mm != old_mm);
848 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
849 mm_update_next_owner(old_mm);
858 * This function makes sure the current process has its own signal table,
859 * so that flush_signal_handlers can later reset the handlers without
860 * disturbing other processes. (Other processes might share the signal
861 * table via the CLONE_SIGHAND option to clone().)
863 static int de_thread(struct task_struct *tsk)
865 struct signal_struct *sig = tsk->signal;
866 struct sighand_struct *oldsighand = tsk->sighand;
867 spinlock_t *lock = &oldsighand->siglock;
869 if (thread_group_empty(tsk))
870 goto no_thread_group;
873 * Kill all other threads in the thread group.
876 if (signal_group_exit(sig)) {
878 * Another group action in progress, just
879 * return so that the signal is processed.
881 spin_unlock_irq(lock);
885 sig->group_exit_task = tsk;
886 sig->notify_count = zap_other_threads(tsk);
887 if (!thread_group_leader(tsk))
890 while (sig->notify_count) {
891 __set_current_state(TASK_UNINTERRUPTIBLE);
892 spin_unlock_irq(lock);
896 spin_unlock_irq(lock);
899 * At this point all other threads have exited, all we have to
900 * do is to wait for the thread group leader to become inactive,
901 * and to assume its PID:
903 if (!thread_group_leader(tsk)) {
904 struct task_struct *leader = tsk->group_leader;
906 sig->notify_count = -1; /* for exit_notify() */
908 write_lock_irq(&tasklist_lock);
909 if (likely(leader->exit_state))
911 __set_current_state(TASK_UNINTERRUPTIBLE);
912 write_unlock_irq(&tasklist_lock);
917 * The only record we have of the real-time age of a
918 * process, regardless of execs it's done, is start_time.
919 * All the past CPU time is accumulated in signal_struct
920 * from sister threads now dead. But in this non-leader
921 * exec, nothing survives from the original leader thread,
922 * whose birth marks the true age of this process now.
923 * When we take on its identity by switching to its PID, we
924 * also take its birthdate (always earlier than our own).
926 tsk->start_time = leader->start_time;
928 BUG_ON(!same_thread_group(leader, tsk));
929 BUG_ON(has_group_leader_pid(tsk));
931 * An exec() starts a new thread group with the
932 * TGID of the previous thread group. Rehash the
933 * two threads with a switched PID, and release
934 * the former thread group leader:
937 /* Become a process group leader with the old leader's pid.
938 * The old leader becomes a thread of the this thread group.
939 * Note: The old leader also uses this pid until release_task
940 * is called. Odd but simple and correct.
942 detach_pid(tsk, PIDTYPE_PID);
943 tsk->pid = leader->pid;
944 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
945 transfer_pid(leader, tsk, PIDTYPE_PGID);
946 transfer_pid(leader, tsk, PIDTYPE_SID);
948 list_replace_rcu(&leader->tasks, &tsk->tasks);
949 list_replace_init(&leader->sibling, &tsk->sibling);
951 tsk->group_leader = tsk;
952 leader->group_leader = tsk;
954 tsk->exit_signal = SIGCHLD;
955 leader->exit_signal = -1;
957 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
958 leader->exit_state = EXIT_DEAD;
961 * We are going to release_task()->ptrace_unlink() silently,
962 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
963 * the tracer wont't block again waiting for this thread.
965 if (unlikely(leader->ptrace))
966 __wake_up_parent(leader, leader->parent);
967 write_unlock_irq(&tasklist_lock);
969 release_task(leader);
972 sig->group_exit_task = NULL;
973 sig->notify_count = 0;
976 /* we have changed execution domain */
977 tsk->exit_signal = SIGCHLD;
980 flush_itimer_signals();
982 if (atomic_read(&oldsighand->count) != 1) {
983 struct sighand_struct *newsighand;
985 * This ->sighand is shared with the CLONE_SIGHAND
986 * but not CLONE_THREAD task, switch to the new one.
988 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
992 atomic_set(&newsighand->count, 1);
993 memcpy(newsighand->action, oldsighand->action,
994 sizeof(newsighand->action));
996 write_lock_irq(&tasklist_lock);
997 spin_lock(&oldsighand->siglock);
998 rcu_assign_pointer(tsk->sighand, newsighand);
999 spin_unlock(&oldsighand->siglock);
1000 write_unlock_irq(&tasklist_lock);
1002 __cleanup_sighand(oldsighand);
1005 BUG_ON(!thread_group_leader(tsk));
1009 char *get_task_comm(char *buf, struct task_struct *tsk)
1011 /* buf must be at least sizeof(tsk->comm) in size */
1013 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1017 EXPORT_SYMBOL_GPL(get_task_comm);
1020 * These functions flushes out all traces of the currently running executable
1021 * so that a new one can be started
1024 void set_task_comm(struct task_struct *tsk, char *buf)
1028 trace_task_rename(tsk, buf);
1031 * Threads may access current->comm without holding
1032 * the task lock, so write the string carefully.
1033 * Readers without a lock may see incomplete new
1034 * names but are safe from non-terminating string reads.
1036 memset(tsk->comm, 0, TASK_COMM_LEN);
1038 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1040 perf_event_comm(tsk);
1043 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1047 /* Copies the binary name from after last slash */
1048 for (i = 0; (ch = *(fn++)) != '\0';) {
1050 i = 0; /* overwrite what we wrote */
1058 int flush_old_exec(struct linux_binprm * bprm)
1063 * Make sure we have a private signal table and that
1064 * we are unassociated from the previous thread group.
1066 retval = de_thread(current);
1070 set_mm_exe_file(bprm->mm, bprm->file);
1072 filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1074 * Release all of the old mmap stuff
1076 acct_arg_size(bprm, 0);
1077 retval = exec_mmap(bprm->mm);
1081 bprm->mm = NULL; /* We're using it now */
1084 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1086 current->personality &= ~bprm->per_clear;
1093 EXPORT_SYMBOL(flush_old_exec);
1095 void would_dump(struct linux_binprm *bprm, struct file *file)
1097 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1098 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1100 EXPORT_SYMBOL(would_dump);
1102 void setup_new_exec(struct linux_binprm * bprm)
1104 arch_pick_mmap_layout(current->mm);
1106 /* This is the point of no return */
1107 current->sas_ss_sp = current->sas_ss_size = 0;
1109 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1110 set_dumpable(current->mm, 1);
1112 set_dumpable(current->mm, suid_dumpable);
1114 set_task_comm(current, bprm->tcomm);
1116 /* Set the new mm task size. We have to do that late because it may
1117 * depend on TIF_32BIT which is only updated in flush_thread() on
1118 * some architectures like powerpc
1120 current->mm->task_size = TASK_SIZE;
1122 /* install the new credentials */
1123 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1124 !gid_eq(bprm->cred->gid, current_egid())) {
1125 current->pdeath_signal = 0;
1127 would_dump(bprm, bprm->file);
1128 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1129 set_dumpable(current->mm, suid_dumpable);
1133 * Flush performance counters when crossing a
1136 if (!get_dumpable(current->mm))
1137 perf_event_exit_task(current);
1139 /* An exec changes our domain. We are no longer part of the thread
1142 current->self_exec_id++;
1144 flush_signal_handlers(current, 0);
1145 do_close_on_exec(current->files);
1147 EXPORT_SYMBOL(setup_new_exec);
1150 * Prepare credentials and lock ->cred_guard_mutex.
1151 * install_exec_creds() commits the new creds and drops the lock.
1152 * Or, if exec fails before, free_bprm() should release ->cred and
1155 int prepare_bprm_creds(struct linux_binprm *bprm)
1157 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1158 return -ERESTARTNOINTR;
1160 bprm->cred = prepare_exec_creds();
1161 if (likely(bprm->cred))
1164 mutex_unlock(¤t->signal->cred_guard_mutex);
1168 void free_bprm(struct linux_binprm *bprm)
1170 free_arg_pages(bprm);
1172 mutex_unlock(¤t->signal->cred_guard_mutex);
1173 abort_creds(bprm->cred);
1179 * install the new credentials for this executable
1181 void install_exec_creds(struct linux_binprm *bprm)
1183 security_bprm_committing_creds(bprm);
1185 commit_creds(bprm->cred);
1188 * cred_guard_mutex must be held at least to this point to prevent
1189 * ptrace_attach() from altering our determination of the task's
1190 * credentials; any time after this it may be unlocked.
1192 security_bprm_committed_creds(bprm);
1193 mutex_unlock(¤t->signal->cred_guard_mutex);
1195 EXPORT_SYMBOL(install_exec_creds);
1198 * determine how safe it is to execute the proposed program
1199 * - the caller must hold ->cred_guard_mutex to protect against
1202 static int check_unsafe_exec(struct linux_binprm *bprm)
1204 struct task_struct *p = current, *t;
1209 if (p->ptrace & PT_PTRACE_CAP)
1210 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1212 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1216 * This isn't strictly necessary, but it makes it harder for LSMs to
1219 if (current->no_new_privs)
1220 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1223 spin_lock(&p->fs->lock);
1225 for (t = next_thread(p); t != p; t = next_thread(t)) {
1231 if (p->fs->users > n_fs) {
1232 bprm->unsafe |= LSM_UNSAFE_SHARE;
1235 if (!p->fs->in_exec) {
1240 spin_unlock(&p->fs->lock);
1246 * Fill the binprm structure from the inode.
1247 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1249 * This may be called multiple times for binary chains (scripts for example).
1251 int prepare_binprm(struct linux_binprm *bprm)
1254 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1257 mode = inode->i_mode;
1258 if (bprm->file->f_op == NULL)
1261 /* clear any previous set[ug]id data from a previous binary */
1262 bprm->cred->euid = current_euid();
1263 bprm->cred->egid = current_egid();
1265 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1266 !current->no_new_privs) {
1268 if (mode & S_ISUID) {
1269 if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1271 bprm->per_clear |= PER_CLEAR_ON_SETID;
1272 bprm->cred->euid = inode->i_uid;
1278 * If setgid is set but no group execute bit then this
1279 * is a candidate for mandatory locking, not a setgid
1282 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1283 if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1285 bprm->per_clear |= PER_CLEAR_ON_SETID;
1286 bprm->cred->egid = inode->i_gid;
1290 /* fill in binprm security blob */
1291 retval = security_bprm_set_creds(bprm);
1294 bprm->cred_prepared = 1;
1296 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1297 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1300 EXPORT_SYMBOL(prepare_binprm);
1303 * Arguments are '\0' separated strings found at the location bprm->p
1304 * points to; chop off the first by relocating brpm->p to right after
1305 * the first '\0' encountered.
1307 int remove_arg_zero(struct linux_binprm *bprm)
1310 unsigned long offset;
1318 offset = bprm->p & ~PAGE_MASK;
1319 page = get_arg_page(bprm, bprm->p, 0);
1324 kaddr = kmap_atomic(page);
1326 for (; offset < PAGE_SIZE && kaddr[offset];
1327 offset++, bprm->p++)
1330 kunmap_atomic(kaddr);
1333 if (offset == PAGE_SIZE)
1334 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1335 } while (offset == PAGE_SIZE);
1344 EXPORT_SYMBOL(remove_arg_zero);
1347 * cycle the list of binary formats handler, until one recognizes the image
1349 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1351 unsigned int depth = bprm->recursion_depth;
1353 struct linux_binfmt *fmt;
1354 pid_t old_pid, old_vpid;
1356 retval = security_bprm_check(bprm);
1360 retval = audit_bprm(bprm);
1364 /* Need to fetch pid before load_binary changes it */
1365 old_pid = current->pid;
1367 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1371 for (try=0; try<2; try++) {
1372 read_lock(&binfmt_lock);
1373 list_for_each_entry(fmt, &formats, lh) {
1374 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1377 if (!try_module_get(fmt->module))
1379 read_unlock(&binfmt_lock);
1380 retval = fn(bprm, regs);
1382 * Restore the depth counter to its starting value
1383 * in this call, so we don't have to rely on every
1384 * load_binary function to restore it on return.
1386 bprm->recursion_depth = depth;
1389 trace_sched_process_exec(current, old_pid, bprm);
1390 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1393 allow_write_access(bprm->file);
1397 current->did_exec = 1;
1398 proc_exec_connector(current);
1401 read_lock(&binfmt_lock);
1403 if (retval != -ENOEXEC || bprm->mm == NULL)
1406 read_unlock(&binfmt_lock);
1410 read_unlock(&binfmt_lock);
1411 #ifdef CONFIG_MODULES
1412 if (retval != -ENOEXEC || bprm->mm == NULL) {
1415 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1416 if (printable(bprm->buf[0]) &&
1417 printable(bprm->buf[1]) &&
1418 printable(bprm->buf[2]) &&
1419 printable(bprm->buf[3]))
1420 break; /* -ENOEXEC */
1422 break; /* -ENOEXEC */
1423 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1432 EXPORT_SYMBOL(search_binary_handler);
1435 * sys_execve() executes a new program.
1437 static int do_execve_common(const char *filename,
1438 struct user_arg_ptr argv,
1439 struct user_arg_ptr envp,
1440 struct pt_regs *regs)
1442 struct linux_binprm *bprm;
1444 struct files_struct *displaced;
1447 const struct cred *cred = current_cred();
1450 * We move the actual failure in case of RLIMIT_NPROC excess from
1451 * set*uid() to execve() because too many poorly written programs
1452 * don't check setuid() return code. Here we additionally recheck
1453 * whether NPROC limit is still exceeded.
1455 if ((current->flags & PF_NPROC_EXCEEDED) &&
1456 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1461 /* We're below the limit (still or again), so we don't want to make
1462 * further execve() calls fail. */
1463 current->flags &= ~PF_NPROC_EXCEEDED;
1465 retval = unshare_files(&displaced);
1470 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1474 retval = prepare_bprm_creds(bprm);
1478 retval = check_unsafe_exec(bprm);
1481 clear_in_exec = retval;
1482 current->in_execve = 1;
1484 file = open_exec(filename);
1485 retval = PTR_ERR(file);
1492 bprm->filename = filename;
1493 bprm->interp = filename;
1495 retval = bprm_mm_init(bprm);
1499 bprm->argc = count(argv, MAX_ARG_STRINGS);
1500 if ((retval = bprm->argc) < 0)
1503 bprm->envc = count(envp, MAX_ARG_STRINGS);
1504 if ((retval = bprm->envc) < 0)
1507 retval = prepare_binprm(bprm);
1511 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1515 bprm->exec = bprm->p;
1516 retval = copy_strings(bprm->envc, envp, bprm);
1520 retval = copy_strings(bprm->argc, argv, bprm);
1524 retval = search_binary_handler(bprm,regs);
1528 /* execve succeeded */
1529 current->fs->in_exec = 0;
1530 current->in_execve = 0;
1531 acct_update_integrals(current);
1534 put_files_struct(displaced);
1539 acct_arg_size(bprm, 0);
1545 allow_write_access(bprm->file);
1551 current->fs->in_exec = 0;
1552 current->in_execve = 0;
1559 reset_files_struct(displaced);
1564 int do_execve(const char *filename,
1565 const char __user *const __user *__argv,
1566 const char __user *const __user *__envp,
1567 struct pt_regs *regs)
1569 struct user_arg_ptr argv = { .ptr.native = __argv };
1570 struct user_arg_ptr envp = { .ptr.native = __envp };
1571 return do_execve_common(filename, argv, envp, regs);
1574 #ifdef CONFIG_COMPAT
1575 int compat_do_execve(char *filename,
1576 compat_uptr_t __user *__argv,
1577 compat_uptr_t __user *__envp,
1578 struct pt_regs *regs)
1580 struct user_arg_ptr argv = {
1582 .ptr.compat = __argv,
1584 struct user_arg_ptr envp = {
1586 .ptr.compat = __envp,
1588 return do_execve_common(filename, argv, envp, regs);
1592 void set_binfmt(struct linux_binfmt *new)
1594 struct mm_struct *mm = current->mm;
1597 module_put(mm->binfmt->module);
1601 __module_get(new->module);
1604 EXPORT_SYMBOL(set_binfmt);
1606 static int expand_corename(struct core_name *cn)
1608 char *old_corename = cn->corename;
1610 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1611 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1613 if (!cn->corename) {
1614 kfree(old_corename);
1621 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1629 need = vsnprintf(NULL, 0, fmt, arg);
1632 if (likely(need < cn->size - cn->used - 1))
1635 ret = expand_corename(cn);
1640 cur = cn->corename + cn->used;
1642 vsnprintf(cur, need + 1, fmt, arg);
1651 static void cn_escape(char *str)
1658 static int cn_print_exe_file(struct core_name *cn)
1660 struct file *exe_file;
1661 char *pathbuf, *path;
1664 exe_file = get_mm_exe_file(current->mm);
1666 char *commstart = cn->corename + cn->used;
1667 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1668 cn_escape(commstart);
1672 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1678 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1680 ret = PTR_ERR(path);
1686 ret = cn_printf(cn, "%s", path);
1695 /* format_corename will inspect the pattern parameter, and output a
1696 * name into corename, which must have space for at least
1697 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1699 static int format_corename(struct core_name *cn, long signr)
1701 const struct cred *cred = current_cred();
1702 const char *pat_ptr = core_pattern;
1703 int ispipe = (*pat_ptr == '|');
1704 int pid_in_pattern = 0;
1707 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1708 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1714 /* Repeat as long as we have more pattern to process and more output
1717 if (*pat_ptr != '%') {
1720 err = cn_printf(cn, "%c", *pat_ptr++);
1722 switch (*++pat_ptr) {
1723 /* single % at the end, drop that */
1726 /* Double percent, output one percent */
1728 err = cn_printf(cn, "%c", '%');
1733 err = cn_printf(cn, "%d",
1734 task_tgid_vnr(current));
1738 err = cn_printf(cn, "%d", cred->uid);
1742 err = cn_printf(cn, "%d", cred->gid);
1744 /* signal that caused the coredump */
1746 err = cn_printf(cn, "%ld", signr);
1748 /* UNIX time of coredump */
1751 do_gettimeofday(&tv);
1752 err = cn_printf(cn, "%lu", tv.tv_sec);
1757 char *namestart = cn->corename + cn->used;
1758 down_read(&uts_sem);
1759 err = cn_printf(cn, "%s",
1760 utsname()->nodename);
1762 cn_escape(namestart);
1767 char *commstart = cn->corename + cn->used;
1768 err = cn_printf(cn, "%s", current->comm);
1769 cn_escape(commstart);
1773 err = cn_print_exe_file(cn);
1775 /* core limit size */
1777 err = cn_printf(cn, "%lu",
1778 rlimit(RLIMIT_CORE));
1790 /* Backward compatibility with core_uses_pid:
1792 * If core_pattern does not include a %p (as is the default)
1793 * and core_uses_pid is set, then .%pid will be appended to
1794 * the filename. Do not do this for piped commands. */
1795 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1796 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1804 static int zap_process(struct task_struct *start, int exit_code)
1806 struct task_struct *t;
1809 start->signal->flags = SIGNAL_GROUP_EXIT;
1810 start->signal->group_exit_code = exit_code;
1811 start->signal->group_stop_count = 0;
1815 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1816 if (t != current && t->mm) {
1817 sigaddset(&t->pending.signal, SIGKILL);
1818 signal_wake_up(t, 1);
1821 } while_each_thread(start, t);
1826 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1827 struct core_state *core_state, int exit_code)
1829 struct task_struct *g, *p;
1830 unsigned long flags;
1833 spin_lock_irq(&tsk->sighand->siglock);
1834 if (!signal_group_exit(tsk->signal)) {
1835 mm->core_state = core_state;
1836 nr = zap_process(tsk, exit_code);
1838 spin_unlock_irq(&tsk->sighand->siglock);
1839 if (unlikely(nr < 0))
1842 if (atomic_read(&mm->mm_users) == nr + 1)
1845 * We should find and kill all tasks which use this mm, and we should
1846 * count them correctly into ->nr_threads. We don't take tasklist
1847 * lock, but this is safe wrt:
1850 * None of sub-threads can fork after zap_process(leader). All
1851 * processes which were created before this point should be
1852 * visible to zap_threads() because copy_process() adds the new
1853 * process to the tail of init_task.tasks list, and lock/unlock
1854 * of ->siglock provides a memory barrier.
1857 * The caller holds mm->mmap_sem. This means that the task which
1858 * uses this mm can't pass exit_mm(), so it can't exit or clear
1862 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1863 * we must see either old or new leader, this does not matter.
1864 * However, it can change p->sighand, so lock_task_sighand(p)
1865 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1868 * Note also that "g" can be the old leader with ->mm == NULL
1869 * and already unhashed and thus removed from ->thread_group.
1870 * This is OK, __unhash_process()->list_del_rcu() does not
1871 * clear the ->next pointer, we will find the new leader via
1875 for_each_process(g) {
1876 if (g == tsk->group_leader)
1878 if (g->flags & PF_KTHREAD)
1883 if (unlikely(p->mm == mm)) {
1884 lock_task_sighand(p, &flags);
1885 nr += zap_process(p, exit_code);
1886 unlock_task_sighand(p, &flags);
1890 } while_each_thread(g, p);
1894 atomic_set(&core_state->nr_threads, nr);
1898 static int coredump_wait(int exit_code, struct core_state *core_state)
1900 struct task_struct *tsk = current;
1901 struct mm_struct *mm = tsk->mm;
1902 int core_waiters = -EBUSY;
1904 init_completion(&core_state->startup);
1905 core_state->dumper.task = tsk;
1906 core_state->dumper.next = NULL;
1908 down_write(&mm->mmap_sem);
1909 if (!mm->core_state)
1910 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1911 up_write(&mm->mmap_sem);
1913 if (core_waiters > 0) {
1914 struct core_thread *ptr;
1916 wait_for_completion(&core_state->startup);
1918 * Wait for all the threads to become inactive, so that
1919 * all the thread context (extended register state, like
1920 * fpu etc) gets copied to the memory.
1922 ptr = core_state->dumper.next;
1923 while (ptr != NULL) {
1924 wait_task_inactive(ptr->task, 0);
1929 return core_waiters;
1932 static void coredump_finish(struct mm_struct *mm)
1934 struct core_thread *curr, *next;
1935 struct task_struct *task;
1937 next = mm->core_state->dumper.next;
1938 while ((curr = next) != NULL) {
1942 * see exit_mm(), curr->task must not see
1943 * ->task == NULL before we read ->next.
1947 wake_up_process(task);
1950 mm->core_state = NULL;
1954 * set_dumpable converts traditional three-value dumpable to two flags and
1955 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1956 * these bits are not changed atomically. So get_dumpable can observe the
1957 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1958 * return either old dumpable or new one by paying attention to the order of
1959 * modifying the bits.
1961 * dumpable | mm->flags (binary)
1962 * old new | initial interim final
1963 * ---------+-----------------------
1971 * (*) get_dumpable regards interim value of 10 as 11.
1973 void set_dumpable(struct mm_struct *mm, int value)
1976 case SUID_DUMPABLE_DISABLED:
1977 clear_bit(MMF_DUMPABLE, &mm->flags);
1979 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1981 case SUID_DUMPABLE_ENABLED:
1982 set_bit(MMF_DUMPABLE, &mm->flags);
1984 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1986 case SUID_DUMPABLE_SAFE:
1987 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1989 set_bit(MMF_DUMPABLE, &mm->flags);
1994 static int __get_dumpable(unsigned long mm_flags)
1998 ret = mm_flags & MMF_DUMPABLE_MASK;
1999 return (ret > SUID_DUMPABLE_ENABLED) ? SUID_DUMPABLE_SAFE : ret;
2002 int get_dumpable(struct mm_struct *mm)
2004 return __get_dumpable(mm->flags);
2007 static void wait_for_dump_helpers(struct file *file)
2009 struct pipe_inode_info *pipe;
2011 pipe = file->f_path.dentry->d_inode->i_pipe;
2017 while ((pipe->readers > 1) && (!signal_pending(current))) {
2018 wake_up_interruptible_sync(&pipe->wait);
2019 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2032 * helper function to customize the process used
2033 * to collect the core in userspace. Specifically
2034 * it sets up a pipe and installs it as fd 0 (stdin)
2035 * for the process. Returns 0 on success, or
2036 * PTR_ERR on failure.
2037 * Note that it also sets the core limit to 1. This
2038 * is a special value that we use to trap recursive
2041 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2043 struct file *files[2];
2044 struct coredump_params *cp = (struct coredump_params *)info->data;
2045 int err = create_pipe_files(files, 0);
2049 cp->file = files[1];
2051 replace_fd(0, files[0], 0);
2052 /* and disallow core files too */
2053 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2058 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2060 struct core_state core_state;
2061 struct core_name cn;
2062 struct mm_struct *mm = current->mm;
2063 struct linux_binfmt * binfmt;
2064 const struct cred *old_cred;
2069 bool need_nonrelative = false;
2070 static atomic_t core_dump_count = ATOMIC_INIT(0);
2071 struct coredump_params cprm = {
2074 .limit = rlimit(RLIMIT_CORE),
2076 * We must use the same mm->flags while dumping core to avoid
2077 * inconsistency of bit flags, since this flag is not protected
2080 .mm_flags = mm->flags,
2083 audit_core_dumps(signr);
2085 binfmt = mm->binfmt;
2086 if (!binfmt || !binfmt->core_dump)
2088 if (!__get_dumpable(cprm.mm_flags))
2091 cred = prepare_creds();
2095 * We cannot trust fsuid as being the "true" uid of the process
2096 * nor do we know its entire history. We only know it was tainted
2097 * so we dump it as root in mode 2, and only into a controlled
2098 * environment (pipe handler or fully qualified path).
2100 if (__get_dumpable(cprm.mm_flags) == SUID_DUMPABLE_SAFE) {
2101 /* Setuid core dump mode */
2102 flag = O_EXCL; /* Stop rewrite attacks */
2103 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
2104 need_nonrelative = true;
2107 retval = coredump_wait(exit_code, &core_state);
2111 old_cred = override_creds(cred);
2114 * Clear any false indication of pending signals that might
2115 * be seen by the filesystem code called to write the core file.
2117 clear_thread_flag(TIF_SIGPENDING);
2119 ispipe = format_corename(&cn, signr);
2126 printk(KERN_WARNING "format_corename failed\n");
2127 printk(KERN_WARNING "Aborting core\n");
2131 if (cprm.limit == 1) {
2132 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
2134 * Normally core limits are irrelevant to pipes, since
2135 * we're not writing to the file system, but we use
2136 * cprm.limit of 1 here as a speacial value, this is a
2137 * consistent way to catch recursive crashes.
2138 * We can still crash if the core_pattern binary sets
2139 * RLIM_CORE = !1, but it runs as root, and can do
2140 * lots of stupid things.
2142 * Note that we use task_tgid_vnr here to grab the pid
2143 * of the process group leader. That way we get the
2144 * right pid if a thread in a multi-threaded
2145 * core_pattern process dies.
2148 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2149 task_tgid_vnr(current), current->comm);
2150 printk(KERN_WARNING "Aborting core\n");
2153 cprm.limit = RLIM_INFINITY;
2155 dump_count = atomic_inc_return(&core_dump_count);
2156 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2157 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2158 task_tgid_vnr(current), current->comm);
2159 printk(KERN_WARNING "Skipping core dump\n");
2160 goto fail_dropcount;
2163 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2165 printk(KERN_WARNING "%s failed to allocate memory\n",
2167 goto fail_dropcount;
2170 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2171 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2173 argv_free(helper_argv);
2175 printk(KERN_INFO "Core dump to %s pipe failed\n",
2180 struct inode *inode;
2182 if (cprm.limit < binfmt->min_coredump)
2185 if (need_nonrelative && cn.corename[0] != '/') {
2186 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
2187 "to fully qualified path!\n",
2188 task_tgid_vnr(current), current->comm);
2189 printk(KERN_WARNING "Skipping core dump\n");
2193 cprm.file = filp_open(cn.corename,
2194 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2196 if (IS_ERR(cprm.file))
2199 inode = cprm.file->f_path.dentry->d_inode;
2200 if (inode->i_nlink > 1)
2202 if (d_unhashed(cprm.file->f_path.dentry))
2205 * AK: actually i see no reason to not allow this for named
2206 * pipes etc, but keep the previous behaviour for now.
2208 if (!S_ISREG(inode->i_mode))
2211 * Dont allow local users get cute and trick others to coredump
2212 * into their pre-created files.
2214 if (!uid_eq(inode->i_uid, current_fsuid()))
2216 if (!cprm.file->f_op || !cprm.file->f_op->write)
2218 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2222 retval = binfmt->core_dump(&cprm);
2224 current->signal->group_exit_code |= 0x80;
2226 if (ispipe && core_pipe_limit)
2227 wait_for_dump_helpers(cprm.file);
2230 filp_close(cprm.file, NULL);
2233 atomic_dec(&core_dump_count);
2237 coredump_finish(mm);
2238 revert_creds(old_cred);
2246 * Core dumping helper functions. These are the only things you should
2247 * do on a core-file: use only these functions to write out all the
2250 int dump_write(struct file *file, const void *addr, int nr)
2252 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2254 EXPORT_SYMBOL(dump_write);
2256 int dump_seek(struct file *file, loff_t off)
2260 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2261 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2264 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2269 unsigned long n = off;
2273 if (!dump_write(file, buf, n)) {
2279 free_page((unsigned long)buf);
2283 EXPORT_SYMBOL(dump_seek);