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>
63 #include <trace/events/task.h>
66 #include <trace/events/sched.h>
69 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
70 unsigned int core_pipe_limit
;
71 int suid_dumpable
= 0;
77 static atomic_t call_count
= ATOMIC_INIT(1);
79 /* The maximal length of core_pattern is also specified in sysctl.c */
81 static LIST_HEAD(formats
);
82 static DEFINE_RWLOCK(binfmt_lock
);
84 int __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
);
95 EXPORT_SYMBOL(__register_binfmt
);
97 void unregister_binfmt(struct linux_binfmt
* fmt
)
99 write_lock(&binfmt_lock
);
101 write_unlock(&binfmt_lock
);
104 EXPORT_SYMBOL(unregister_binfmt
);
106 static inline void put_binfmt(struct linux_binfmt
* fmt
)
108 module_put(fmt
->module
);
112 * Note that a shared library must be both readable and executable due to
115 * Also note that we take the address to load from from the file itself.
117 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
120 char *tmp
= getname(library
);
121 int error
= PTR_ERR(tmp
);
122 static const struct open_flags uselib_flags
= {
123 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
124 .acc_mode
= MAY_READ
| MAY_EXEC
| MAY_OPEN
,
125 .intent
= LOOKUP_OPEN
131 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
, LOOKUP_FOLLOW
);
133 error
= PTR_ERR(file
);
138 if (!S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
142 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
149 struct linux_binfmt
* fmt
;
151 read_lock(&binfmt_lock
);
152 list_for_each_entry(fmt
, &formats
, lh
) {
153 if (!fmt
->load_shlib
)
155 if (!try_module_get(fmt
->module
))
157 read_unlock(&binfmt_lock
);
158 error
= fmt
->load_shlib(file
);
159 read_lock(&binfmt_lock
);
161 if (error
!= -ENOEXEC
)
164 read_unlock(&binfmt_lock
);
174 * The nascent bprm->mm is not visible until exec_mmap() but it can
175 * use a lot of memory, account these pages in current->mm temporary
176 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
177 * change the counter back via acct_arg_size(0).
179 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
181 struct mm_struct
*mm
= current
->mm
;
182 long diff
= (long)(pages
- bprm
->vma_pages
);
187 bprm
->vma_pages
= pages
;
188 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
191 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
197 #ifdef CONFIG_STACK_GROWSUP
199 ret
= expand_downwards(bprm
->vma
, pos
);
204 ret
= get_user_pages(current
, bprm
->mm
, pos
,
205 1, write
, 1, &page
, NULL
);
210 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
213 acct_arg_size(bprm
, size
/ PAGE_SIZE
);
216 * We've historically supported up to 32 pages (ARG_MAX)
217 * of argument strings even with small stacks
223 * Limit to 1/4-th the stack size for the argv+env strings.
225 * - the remaining binfmt code will not run out of stack space,
226 * - the program will have a reasonable amount of stack left
229 rlim
= current
->signal
->rlim
;
230 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
) / 4) {
239 static void put_arg_page(struct page
*page
)
244 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
248 static void free_arg_pages(struct linux_binprm
*bprm
)
252 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
255 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
258 static int __bprm_mm_init(struct linux_binprm
*bprm
)
261 struct vm_area_struct
*vma
= NULL
;
262 struct mm_struct
*mm
= bprm
->mm
;
264 bprm
->vma
= vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
268 down_write(&mm
->mmap_sem
);
272 * Place the stack at the largest stack address the architecture
273 * supports. Later, we'll move this to an appropriate place. We don't
274 * use STACK_TOP because that can depend on attributes which aren't
277 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
278 vma
->vm_end
= STACK_TOP_MAX
;
279 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
280 vma
->vm_flags
= VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
281 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
282 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
284 err
= security_file_mmap(NULL
, 0, 0, 0, vma
->vm_start
, 1);
288 err
= insert_vm_struct(mm
, vma
);
292 mm
->stack_vm
= mm
->total_vm
= 1;
293 up_write(&mm
->mmap_sem
);
294 bprm
->p
= vma
->vm_end
- sizeof(void *);
297 up_write(&mm
->mmap_sem
);
299 kmem_cache_free(vm_area_cachep
, vma
);
303 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
305 return len
<= MAX_ARG_STRLEN
;
310 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
314 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
319 page
= bprm
->page
[pos
/ PAGE_SIZE
];
320 if (!page
&& write
) {
321 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
324 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
330 static void put_arg_page(struct page
*page
)
334 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
337 __free_page(bprm
->page
[i
]);
338 bprm
->page
[i
] = NULL
;
342 static void free_arg_pages(struct linux_binprm
*bprm
)
346 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
347 free_arg_page(bprm
, i
);
350 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
355 static int __bprm_mm_init(struct linux_binprm
*bprm
)
357 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
361 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
363 return len
<= bprm
->p
;
366 #endif /* CONFIG_MMU */
369 * Create a new mm_struct and populate it with a temporary stack
370 * vm_area_struct. We don't have enough context at this point to set the stack
371 * flags, permissions, and offset, so we use temporary values. We'll update
372 * them later in setup_arg_pages().
374 int bprm_mm_init(struct linux_binprm
*bprm
)
377 struct mm_struct
*mm
= NULL
;
379 bprm
->mm
= mm
= mm_alloc();
384 err
= init_new_context(current
, mm
);
388 err
= __bprm_mm_init(bprm
);
403 struct user_arg_ptr
{
408 const char __user
*const __user
*native
;
410 compat_uptr_t __user
*compat
;
415 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
417 const char __user
*native
;
420 if (unlikely(argv
.is_compat
)) {
421 compat_uptr_t compat
;
423 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
424 return ERR_PTR(-EFAULT
);
426 return compat_ptr(compat
);
430 if (get_user(native
, argv
.ptr
.native
+ nr
))
431 return ERR_PTR(-EFAULT
);
437 * count() counts the number of strings in array ARGV.
439 static int count(struct user_arg_ptr argv
, int max
)
443 if (argv
.ptr
.native
!= NULL
) {
445 const char __user
*p
= get_user_arg_ptr(argv
, i
);
456 if (fatal_signal_pending(current
))
457 return -ERESTARTNOHAND
;
465 * 'copy_strings()' copies argument/environment strings from the old
466 * processes's memory to the new process's stack. The call to get_user_pages()
467 * ensures the destination page is created and not swapped out.
469 static int copy_strings(int argc
, struct user_arg_ptr argv
,
470 struct linux_binprm
*bprm
)
472 struct page
*kmapped_page
= NULL
;
474 unsigned long kpos
= 0;
478 const char __user
*str
;
483 str
= get_user_arg_ptr(argv
, argc
);
487 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
492 if (!valid_arg_len(bprm
, len
))
495 /* We're going to work our way backwords. */
501 int offset
, bytes_to_copy
;
503 if (fatal_signal_pending(current
)) {
504 ret
= -ERESTARTNOHAND
;
509 offset
= pos
% PAGE_SIZE
;
513 bytes_to_copy
= offset
;
514 if (bytes_to_copy
> len
)
517 offset
-= bytes_to_copy
;
518 pos
-= bytes_to_copy
;
519 str
-= bytes_to_copy
;
520 len
-= bytes_to_copy
;
522 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
525 page
= get_arg_page(bprm
, pos
, 1);
532 flush_kernel_dcache_page(kmapped_page
);
533 kunmap(kmapped_page
);
534 put_arg_page(kmapped_page
);
537 kaddr
= kmap(kmapped_page
);
538 kpos
= pos
& PAGE_MASK
;
539 flush_arg_page(bprm
, kpos
, kmapped_page
);
541 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
550 flush_kernel_dcache_page(kmapped_page
);
551 kunmap(kmapped_page
);
552 put_arg_page(kmapped_page
);
558 * Like copy_strings, but get argv and its values from kernel memory.
560 int copy_strings_kernel(int argc
, const char *const *__argv
,
561 struct linux_binprm
*bprm
)
564 mm_segment_t oldfs
= get_fs();
565 struct user_arg_ptr argv
= {
566 .ptr
.native
= (const char __user
*const __user
*)__argv
,
570 r
= copy_strings(argc
, argv
, bprm
);
575 EXPORT_SYMBOL(copy_strings_kernel
);
580 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
581 * the binfmt code determines where the new stack should reside, we shift it to
582 * its final location. The process proceeds as follows:
584 * 1) Use shift to calculate the new vma endpoints.
585 * 2) Extend vma to cover both the old and new ranges. This ensures the
586 * arguments passed to subsequent functions are consistent.
587 * 3) Move vma's page tables to the new range.
588 * 4) Free up any cleared pgd range.
589 * 5) Shrink the vma to cover only the new range.
591 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
593 struct mm_struct
*mm
= vma
->vm_mm
;
594 unsigned long old_start
= vma
->vm_start
;
595 unsigned long old_end
= vma
->vm_end
;
596 unsigned long length
= old_end
- old_start
;
597 unsigned long new_start
= old_start
- shift
;
598 unsigned long new_end
= old_end
- shift
;
599 struct mmu_gather tlb
;
601 BUG_ON(new_start
> new_end
);
604 * ensure there are no vmas between where we want to go
607 if (vma
!= find_vma(mm
, new_start
))
611 * cover the whole range: [new_start, old_end)
613 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
617 * move the page tables downwards, on failure we rely on
618 * process cleanup to remove whatever mess we made.
620 if (length
!= move_page_tables(vma
, old_start
,
621 vma
, new_start
, length
))
625 tlb_gather_mmu(&tlb
, mm
, 0);
626 if (new_end
> old_start
) {
628 * when the old and new regions overlap clear from new_end.
630 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
631 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
634 * otherwise, clean from old_start; this is done to not touch
635 * the address space in [new_end, old_start) some architectures
636 * have constraints on va-space that make this illegal (IA64) -
637 * for the others its just a little faster.
639 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
640 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
642 tlb_finish_mmu(&tlb
, new_end
, old_end
);
645 * Shrink the vma to just the new range. Always succeeds.
647 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
653 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
654 * the stack is optionally relocated, and some extra space is added.
656 int setup_arg_pages(struct linux_binprm
*bprm
,
657 unsigned long stack_top
,
658 int executable_stack
)
661 unsigned long stack_shift
;
662 struct mm_struct
*mm
= current
->mm
;
663 struct vm_area_struct
*vma
= bprm
->vma
;
664 struct vm_area_struct
*prev
= NULL
;
665 unsigned long vm_flags
;
666 unsigned long stack_base
;
667 unsigned long stack_size
;
668 unsigned long stack_expand
;
669 unsigned long rlim_stack
;
671 #ifdef CONFIG_STACK_GROWSUP
672 /* Limit stack size to 1GB */
673 stack_base
= rlimit_max(RLIMIT_STACK
);
674 if (stack_base
> (1 << 30))
675 stack_base
= 1 << 30;
677 /* Make sure we didn't let the argument array grow too large. */
678 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
681 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
683 stack_shift
= vma
->vm_start
- stack_base
;
684 mm
->arg_start
= bprm
->p
- stack_shift
;
685 bprm
->p
= vma
->vm_end
- stack_shift
;
687 stack_top
= arch_align_stack(stack_top
);
688 stack_top
= PAGE_ALIGN(stack_top
);
690 if (unlikely(stack_top
< mmap_min_addr
) ||
691 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
694 stack_shift
= vma
->vm_end
- stack_top
;
696 bprm
->p
-= stack_shift
;
697 mm
->arg_start
= bprm
->p
;
701 bprm
->loader
-= stack_shift
;
702 bprm
->exec
-= stack_shift
;
704 down_write(&mm
->mmap_sem
);
705 vm_flags
= VM_STACK_FLAGS
;
708 * Adjust stack execute permissions; explicitly enable for
709 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
710 * (arch default) otherwise.
712 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
714 else if (executable_stack
== EXSTACK_DISABLE_X
)
715 vm_flags
&= ~VM_EXEC
;
716 vm_flags
|= mm
->def_flags
;
717 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
719 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
725 /* Move stack pages down in memory. */
727 ret
= shift_arg_pages(vma
, stack_shift
);
732 /* mprotect_fixup is overkill to remove the temporary stack flags */
733 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
735 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
736 stack_size
= vma
->vm_end
- vma
->vm_start
;
738 * Align this down to a page boundary as expand_stack
741 rlim_stack
= rlimit(RLIMIT_STACK
) & PAGE_MASK
;
742 #ifdef CONFIG_STACK_GROWSUP
743 if (stack_size
+ stack_expand
> rlim_stack
)
744 stack_base
= vma
->vm_start
+ rlim_stack
;
746 stack_base
= vma
->vm_end
+ stack_expand
;
748 if (stack_size
+ stack_expand
> rlim_stack
)
749 stack_base
= vma
->vm_end
- rlim_stack
;
751 stack_base
= vma
->vm_start
- stack_expand
;
753 current
->mm
->start_stack
= bprm
->p
;
754 ret
= expand_stack(vma
, stack_base
);
759 up_write(&mm
->mmap_sem
);
762 EXPORT_SYMBOL(setup_arg_pages
);
764 #endif /* CONFIG_MMU */
766 struct file
*open_exec(const char *name
)
770 static const struct open_flags open_exec_flags
= {
771 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
772 .acc_mode
= MAY_EXEC
| MAY_OPEN
,
773 .intent
= LOOKUP_OPEN
776 file
= do_filp_open(AT_FDCWD
, name
, &open_exec_flags
, LOOKUP_FOLLOW
);
781 if (!S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
784 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
789 err
= deny_write_access(file
);
800 EXPORT_SYMBOL(open_exec
);
802 int kernel_read(struct file
*file
, loff_t offset
,
803 char *addr
, unsigned long count
)
811 /* The cast to a user pointer is valid due to the set_fs() */
812 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
817 EXPORT_SYMBOL(kernel_read
);
819 static int exec_mmap(struct mm_struct
*mm
)
821 struct task_struct
*tsk
;
822 struct mm_struct
* old_mm
, *active_mm
;
824 /* Notify parent that we're no longer interested in the old VM */
826 old_mm
= current
->mm
;
827 sync_mm_rss(tsk
, old_mm
);
828 mm_release(tsk
, old_mm
);
832 * Make sure that if there is a core dump in progress
833 * for the old mm, we get out and die instead of going
834 * through with the exec. We must hold mmap_sem around
835 * checking core_state and changing tsk->mm.
837 down_read(&old_mm
->mmap_sem
);
838 if (unlikely(old_mm
->core_state
)) {
839 up_read(&old_mm
->mmap_sem
);
844 active_mm
= tsk
->active_mm
;
847 activate_mm(active_mm
, mm
);
849 arch_pick_mmap_layout(mm
);
851 up_read(&old_mm
->mmap_sem
);
852 BUG_ON(active_mm
!= old_mm
);
853 mm_update_next_owner(old_mm
);
862 * This function makes sure the current process has its own signal table,
863 * so that flush_signal_handlers can later reset the handlers without
864 * disturbing other processes. (Other processes might share the signal
865 * table via the CLONE_SIGHAND option to clone().)
867 static int de_thread(struct task_struct
*tsk
)
869 struct signal_struct
*sig
= tsk
->signal
;
870 struct sighand_struct
*oldsighand
= tsk
->sighand
;
871 spinlock_t
*lock
= &oldsighand
->siglock
;
873 if (thread_group_empty(tsk
))
874 goto no_thread_group
;
877 * Kill all other threads in the thread group.
880 if (signal_group_exit(sig
)) {
882 * Another group action in progress, just
883 * return so that the signal is processed.
885 spin_unlock_irq(lock
);
889 sig
->group_exit_task
= tsk
;
890 sig
->notify_count
= zap_other_threads(tsk
);
891 if (!thread_group_leader(tsk
))
894 while (sig
->notify_count
) {
895 __set_current_state(TASK_UNINTERRUPTIBLE
);
896 spin_unlock_irq(lock
);
900 spin_unlock_irq(lock
);
903 * At this point all other threads have exited, all we have to
904 * do is to wait for the thread group leader to become inactive,
905 * and to assume its PID:
907 if (!thread_group_leader(tsk
)) {
908 struct task_struct
*leader
= tsk
->group_leader
;
910 sig
->notify_count
= -1; /* for exit_notify() */
912 write_lock_irq(&tasklist_lock
);
913 if (likely(leader
->exit_state
))
915 __set_current_state(TASK_UNINTERRUPTIBLE
);
916 write_unlock_irq(&tasklist_lock
);
921 * The only record we have of the real-time age of a
922 * process, regardless of execs it's done, is start_time.
923 * All the past CPU time is accumulated in signal_struct
924 * from sister threads now dead. But in this non-leader
925 * exec, nothing survives from the original leader thread,
926 * whose birth marks the true age of this process now.
927 * When we take on its identity by switching to its PID, we
928 * also take its birthdate (always earlier than our own).
930 tsk
->start_time
= leader
->start_time
;
932 BUG_ON(!same_thread_group(leader
, tsk
));
933 BUG_ON(has_group_leader_pid(tsk
));
935 * An exec() starts a new thread group with the
936 * TGID of the previous thread group. Rehash the
937 * two threads with a switched PID, and release
938 * the former thread group leader:
941 /* Become a process group leader with the old leader's pid.
942 * The old leader becomes a thread of the this thread group.
943 * Note: The old leader also uses this pid until release_task
944 * is called. Odd but simple and correct.
946 detach_pid(tsk
, PIDTYPE_PID
);
947 tsk
->pid
= leader
->pid
;
948 attach_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
949 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
950 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
952 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
953 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
955 tsk
->group_leader
= tsk
;
956 leader
->group_leader
= tsk
;
958 tsk
->exit_signal
= SIGCHLD
;
959 leader
->exit_signal
= -1;
961 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
962 leader
->exit_state
= EXIT_DEAD
;
965 * We are going to release_task()->ptrace_unlink() silently,
966 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
967 * the tracer wont't block again waiting for this thread.
969 if (unlikely(leader
->ptrace
))
970 __wake_up_parent(leader
, leader
->parent
);
971 write_unlock_irq(&tasklist_lock
);
973 release_task(leader
);
976 sig
->group_exit_task
= NULL
;
977 sig
->notify_count
= 0;
980 /* we have changed execution domain */
981 tsk
->exit_signal
= SIGCHLD
;
984 setmax_mm_hiwater_rss(&sig
->maxrss
, current
->mm
);
987 flush_itimer_signals();
989 if (atomic_read(&oldsighand
->count
) != 1) {
990 struct sighand_struct
*newsighand
;
992 * This ->sighand is shared with the CLONE_SIGHAND
993 * but not CLONE_THREAD task, switch to the new one.
995 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
999 atomic_set(&newsighand
->count
, 1);
1000 memcpy(newsighand
->action
, oldsighand
->action
,
1001 sizeof(newsighand
->action
));
1003 write_lock_irq(&tasklist_lock
);
1004 spin_lock(&oldsighand
->siglock
);
1005 rcu_assign_pointer(tsk
->sighand
, newsighand
);
1006 spin_unlock(&oldsighand
->siglock
);
1007 write_unlock_irq(&tasklist_lock
);
1009 __cleanup_sighand(oldsighand
);
1012 BUG_ON(!thread_group_leader(tsk
));
1017 * These functions flushes out all traces of the currently running executable
1018 * so that a new one can be started
1020 static void flush_old_files(struct files_struct
* files
)
1023 struct fdtable
*fdt
;
1025 spin_lock(&files
->file_lock
);
1027 unsigned long set
, i
;
1031 fdt
= files_fdtable(files
);
1032 if (i
>= fdt
->max_fds
)
1034 set
= fdt
->close_on_exec
->fds_bits
[j
];
1037 fdt
->close_on_exec
->fds_bits
[j
] = 0;
1038 spin_unlock(&files
->file_lock
);
1039 for ( ; set
; i
++,set
>>= 1) {
1044 spin_lock(&files
->file_lock
);
1047 spin_unlock(&files
->file_lock
);
1050 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
1052 /* buf must be at least sizeof(tsk->comm) in size */
1054 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
1058 EXPORT_SYMBOL_GPL(get_task_comm
);
1060 void set_task_comm(struct task_struct
*tsk
, char *buf
)
1064 trace_task_rename(tsk
, buf
);
1067 * Threads may access current->comm without holding
1068 * the task lock, so write the string carefully.
1069 * Readers without a lock may see incomplete new
1070 * names but are safe from non-terminating string reads.
1072 memset(tsk
->comm
, 0, TASK_COMM_LEN
);
1074 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1076 perf_event_comm(tsk
);
1079 static void filename_to_taskname(char *tcomm
, const char *fn
, unsigned int len
)
1083 /* Copies the binary name from after last slash */
1084 for (i
= 0; (ch
= *(fn
++)) != '\0';) {
1086 i
= 0; /* overwrite what we wrote */
1094 int flush_old_exec(struct linux_binprm
* bprm
)
1099 * Make sure we have a private signal table and that
1100 * we are unassociated from the previous thread group.
1102 retval
= de_thread(current
);
1106 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1108 filename_to_taskname(bprm
->tcomm
, bprm
->filename
, sizeof(bprm
->tcomm
));
1110 * Release all of the old mmap stuff
1112 acct_arg_size(bprm
, 0);
1113 retval
= exec_mmap(bprm
->mm
);
1117 bprm
->mm
= NULL
; /* We're using it now */
1120 current
->flags
&= ~(PF_RANDOMIZE
| PF_KTHREAD
);
1122 current
->personality
&= ~bprm
->per_clear
;
1129 EXPORT_SYMBOL(flush_old_exec
);
1131 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1133 if (inode_permission(file
->f_path
.dentry
->d_inode
, MAY_READ
) < 0)
1134 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1136 EXPORT_SYMBOL(would_dump
);
1138 void setup_new_exec(struct linux_binprm
* bprm
)
1140 arch_pick_mmap_layout(current
->mm
);
1142 /* This is the point of no return */
1143 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
1145 if (current_euid() == current_uid() && current_egid() == current_gid())
1146 set_dumpable(current
->mm
, 1);
1148 set_dumpable(current
->mm
, suid_dumpable
);
1150 set_task_comm(current
, bprm
->tcomm
);
1152 /* Set the new mm task size. We have to do that late because it may
1153 * depend on TIF_32BIT which is only updated in flush_thread() on
1154 * some architectures like powerpc
1156 current
->mm
->task_size
= TASK_SIZE
;
1158 /* install the new credentials */
1159 if (bprm
->cred
->uid
!= current_euid() ||
1160 bprm
->cred
->gid
!= current_egid()) {
1161 current
->pdeath_signal
= 0;
1163 would_dump(bprm
, bprm
->file
);
1164 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
)
1165 set_dumpable(current
->mm
, suid_dumpable
);
1169 * Flush performance counters when crossing a
1172 if (!get_dumpable(current
->mm
))
1173 perf_event_exit_task(current
);
1175 /* An exec changes our domain. We are no longer part of the thread
1178 current
->self_exec_id
++;
1180 flush_signal_handlers(current
, 0);
1181 flush_old_files(current
->files
);
1183 EXPORT_SYMBOL(setup_new_exec
);
1186 * Prepare credentials and lock ->cred_guard_mutex.
1187 * install_exec_creds() commits the new creds and drops the lock.
1188 * Or, if exec fails before, free_bprm() should release ->cred and
1191 int prepare_bprm_creds(struct linux_binprm
*bprm
)
1193 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1194 return -ERESTARTNOINTR
;
1196 bprm
->cred
= prepare_exec_creds();
1197 if (likely(bprm
->cred
))
1200 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1204 void free_bprm(struct linux_binprm
*bprm
)
1206 free_arg_pages(bprm
);
1208 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1209 abort_creds(bprm
->cred
);
1215 * install the new credentials for this executable
1217 void install_exec_creds(struct linux_binprm
*bprm
)
1219 security_bprm_committing_creds(bprm
);
1221 commit_creds(bprm
->cred
);
1224 * cred_guard_mutex must be held at least to this point to prevent
1225 * ptrace_attach() from altering our determination of the task's
1226 * credentials; any time after this it may be unlocked.
1228 security_bprm_committed_creds(bprm
);
1229 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1231 EXPORT_SYMBOL(install_exec_creds
);
1234 * determine how safe it is to execute the proposed program
1235 * - the caller must hold ->cred_guard_mutex to protect against
1238 static int check_unsafe_exec(struct linux_binprm
*bprm
)
1240 struct task_struct
*p
= current
, *t
;
1245 if (p
->ptrace
& PT_PTRACE_CAP
)
1246 bprm
->unsafe
|= LSM_UNSAFE_PTRACE_CAP
;
1248 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1252 spin_lock(&p
->fs
->lock
);
1254 for (t
= next_thread(p
); t
!= p
; t
= next_thread(t
)) {
1260 if (p
->fs
->users
> n_fs
) {
1261 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1264 if (!p
->fs
->in_exec
) {
1269 spin_unlock(&p
->fs
->lock
);
1275 * Fill the binprm structure from the inode.
1276 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1278 * This may be called multiple times for binary chains (scripts for example).
1280 int prepare_binprm(struct linux_binprm
*bprm
)
1283 struct inode
* inode
= bprm
->file
->f_path
.dentry
->d_inode
;
1286 mode
= inode
->i_mode
;
1287 if (bprm
->file
->f_op
== NULL
)
1290 /* clear any previous set[ug]id data from a previous binary */
1291 bprm
->cred
->euid
= current_euid();
1292 bprm
->cred
->egid
= current_egid();
1294 if (!(bprm
->file
->f_path
.mnt
->mnt_flags
& MNT_NOSUID
)) {
1296 if (mode
& S_ISUID
) {
1297 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1298 bprm
->cred
->euid
= inode
->i_uid
;
1303 * If setgid is set but no group execute bit then this
1304 * is a candidate for mandatory locking, not a setgid
1307 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1308 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1309 bprm
->cred
->egid
= inode
->i_gid
;
1313 /* fill in binprm security blob */
1314 retval
= security_bprm_set_creds(bprm
);
1317 bprm
->cred_prepared
= 1;
1319 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1320 return kernel_read(bprm
->file
, 0, bprm
->buf
, BINPRM_BUF_SIZE
);
1323 EXPORT_SYMBOL(prepare_binprm
);
1326 * Arguments are '\0' separated strings found at the location bprm->p
1327 * points to; chop off the first by relocating brpm->p to right after
1328 * the first '\0' encountered.
1330 int remove_arg_zero(struct linux_binprm
*bprm
)
1333 unsigned long offset
;
1341 offset
= bprm
->p
& ~PAGE_MASK
;
1342 page
= get_arg_page(bprm
, bprm
->p
, 0);
1347 kaddr
= kmap_atomic(page
, KM_USER0
);
1349 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1350 offset
++, bprm
->p
++)
1353 kunmap_atomic(kaddr
, KM_USER0
);
1356 if (offset
== PAGE_SIZE
)
1357 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1358 } while (offset
== PAGE_SIZE
);
1367 EXPORT_SYMBOL(remove_arg_zero
);
1370 * cycle the list of binary formats handler, until one recognizes the image
1372 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
1374 unsigned int depth
= bprm
->recursion_depth
;
1376 struct linux_binfmt
*fmt
;
1379 retval
= security_bprm_check(bprm
);
1383 retval
= audit_bprm(bprm
);
1387 /* Need to fetch pid before load_binary changes it */
1389 old_pid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1393 for (try=0; try<2; try++) {
1394 read_lock(&binfmt_lock
);
1395 list_for_each_entry(fmt
, &formats
, lh
) {
1396 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
1399 if (!try_module_get(fmt
->module
))
1401 read_unlock(&binfmt_lock
);
1402 retval
= fn(bprm
, regs
);
1404 * Restore the depth counter to its starting value
1405 * in this call, so we don't have to rely on every
1406 * load_binary function to restore it on return.
1408 bprm
->recursion_depth
= depth
;
1411 trace_sched_process_exec(current
, old_pid
, bprm
);
1412 ptrace_event(PTRACE_EVENT_EXEC
, old_pid
);
1415 allow_write_access(bprm
->file
);
1419 current
->did_exec
= 1;
1420 proc_exec_connector(current
);
1423 read_lock(&binfmt_lock
);
1425 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
)
1428 read_unlock(&binfmt_lock
);
1432 read_unlock(&binfmt_lock
);
1433 #ifdef CONFIG_MODULES
1434 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
) {
1437 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1438 if (printable(bprm
->buf
[0]) &&
1439 printable(bprm
->buf
[1]) &&
1440 printable(bprm
->buf
[2]) &&
1441 printable(bprm
->buf
[3]))
1442 break; /* -ENOEXEC */
1444 break; /* -ENOEXEC */
1445 request_module("binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
1454 EXPORT_SYMBOL(search_binary_handler
);
1457 * sys_execve() executes a new program.
1459 static int do_execve_common(const char *filename
,
1460 struct user_arg_ptr argv
,
1461 struct user_arg_ptr envp
,
1462 struct pt_regs
*regs
)
1464 struct linux_binprm
*bprm
;
1466 struct files_struct
*displaced
;
1469 const struct cred
*cred
= current_cred();
1472 * We move the actual failure in case of RLIMIT_NPROC excess from
1473 * set*uid() to execve() because too many poorly written programs
1474 * don't check setuid() return code. Here we additionally recheck
1475 * whether NPROC limit is still exceeded.
1477 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1478 atomic_read(&cred
->user
->processes
) > rlimit(RLIMIT_NPROC
)) {
1483 /* We're below the limit (still or again), so we don't want to make
1484 * further execve() calls fail. */
1485 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1487 retval
= unshare_files(&displaced
);
1492 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1496 retval
= prepare_bprm_creds(bprm
);
1500 retval
= check_unsafe_exec(bprm
);
1503 clear_in_exec
= retval
;
1504 current
->in_execve
= 1;
1506 file
= open_exec(filename
);
1507 retval
= PTR_ERR(file
);
1514 bprm
->filename
= filename
;
1515 bprm
->interp
= filename
;
1517 retval
= bprm_mm_init(bprm
);
1521 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1522 if ((retval
= bprm
->argc
) < 0)
1525 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1526 if ((retval
= bprm
->envc
) < 0)
1529 retval
= prepare_binprm(bprm
);
1533 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1537 bprm
->exec
= bprm
->p
;
1538 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1542 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1546 retval
= search_binary_handler(bprm
,regs
);
1550 /* execve succeeded */
1551 current
->fs
->in_exec
= 0;
1552 current
->in_execve
= 0;
1553 acct_update_integrals(current
);
1556 put_files_struct(displaced
);
1561 acct_arg_size(bprm
, 0);
1567 allow_write_access(bprm
->file
);
1573 current
->fs
->in_exec
= 0;
1574 current
->in_execve
= 0;
1581 reset_files_struct(displaced
);
1586 int do_execve(const char *filename
,
1587 const char __user
*const __user
*__argv
,
1588 const char __user
*const __user
*__envp
,
1589 struct pt_regs
*regs
)
1591 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1592 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1593 return do_execve_common(filename
, argv
, envp
, regs
);
1596 #ifdef CONFIG_COMPAT
1597 int compat_do_execve(char *filename
,
1598 compat_uptr_t __user
*__argv
,
1599 compat_uptr_t __user
*__envp
,
1600 struct pt_regs
*regs
)
1602 struct user_arg_ptr argv
= {
1604 .ptr
.compat
= __argv
,
1606 struct user_arg_ptr envp
= {
1608 .ptr
.compat
= __envp
,
1610 return do_execve_common(filename
, argv
, envp
, regs
);
1614 void set_binfmt(struct linux_binfmt
*new)
1616 struct mm_struct
*mm
= current
->mm
;
1619 module_put(mm
->binfmt
->module
);
1623 __module_get(new->module
);
1626 EXPORT_SYMBOL(set_binfmt
);
1628 static int expand_corename(struct core_name
*cn
)
1630 char *old_corename
= cn
->corename
;
1632 cn
->size
= CORENAME_MAX_SIZE
* atomic_inc_return(&call_count
);
1633 cn
->corename
= krealloc(old_corename
, cn
->size
, GFP_KERNEL
);
1635 if (!cn
->corename
) {
1636 kfree(old_corename
);
1643 static int cn_printf(struct core_name
*cn
, const char *fmt
, ...)
1651 need
= vsnprintf(NULL
, 0, fmt
, arg
);
1654 if (likely(need
< cn
->size
- cn
->used
- 1))
1657 ret
= expand_corename(cn
);
1662 cur
= cn
->corename
+ cn
->used
;
1664 vsnprintf(cur
, need
+ 1, fmt
, arg
);
1673 static void cn_escape(char *str
)
1680 static int cn_print_exe_file(struct core_name
*cn
)
1682 struct file
*exe_file
;
1683 char *pathbuf
, *path
;
1686 exe_file
= get_mm_exe_file(current
->mm
);
1688 char *commstart
= cn
->corename
+ cn
->used
;
1689 ret
= cn_printf(cn
, "%s (path unknown)", current
->comm
);
1690 cn_escape(commstart
);
1694 pathbuf
= kmalloc(PATH_MAX
, GFP_TEMPORARY
);
1700 path
= d_path(&exe_file
->f_path
, pathbuf
, PATH_MAX
);
1702 ret
= PTR_ERR(path
);
1708 ret
= cn_printf(cn
, "%s", path
);
1717 /* format_corename will inspect the pattern parameter, and output a
1718 * name into corename, which must have space for at least
1719 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1721 static int format_corename(struct core_name
*cn
, long signr
)
1723 const struct cred
*cred
= current_cred();
1724 const char *pat_ptr
= core_pattern
;
1725 int ispipe
= (*pat_ptr
== '|');
1726 int pid_in_pattern
= 0;
1729 cn
->size
= CORENAME_MAX_SIZE
* atomic_read(&call_count
);
1730 cn
->corename
= kmalloc(cn
->size
, GFP_KERNEL
);
1736 /* Repeat as long as we have more pattern to process and more output
1739 if (*pat_ptr
!= '%') {
1742 err
= cn_printf(cn
, "%c", *pat_ptr
++);
1744 switch (*++pat_ptr
) {
1745 /* single % at the end, drop that */
1748 /* Double percent, output one percent */
1750 err
= cn_printf(cn
, "%c", '%');
1755 err
= cn_printf(cn
, "%d",
1756 task_tgid_vnr(current
));
1760 err
= cn_printf(cn
, "%d", cred
->uid
);
1764 err
= cn_printf(cn
, "%d", cred
->gid
);
1766 /* signal that caused the coredump */
1768 err
= cn_printf(cn
, "%ld", signr
);
1770 /* UNIX time of coredump */
1773 do_gettimeofday(&tv
);
1774 err
= cn_printf(cn
, "%lu", tv
.tv_sec
);
1779 char *namestart
= cn
->corename
+ cn
->used
;
1780 down_read(&uts_sem
);
1781 err
= cn_printf(cn
, "%s",
1782 utsname()->nodename
);
1784 cn_escape(namestart
);
1789 char *commstart
= cn
->corename
+ cn
->used
;
1790 err
= cn_printf(cn
, "%s", current
->comm
);
1791 cn_escape(commstart
);
1795 err
= cn_print_exe_file(cn
);
1797 /* core limit size */
1799 err
= cn_printf(cn
, "%lu",
1800 rlimit(RLIMIT_CORE
));
1812 /* Backward compatibility with core_uses_pid:
1814 * If core_pattern does not include a %p (as is the default)
1815 * and core_uses_pid is set, then .%pid will be appended to
1816 * the filename. Do not do this for piped commands. */
1817 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
1818 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
1826 static int zap_process(struct task_struct
*start
, int exit_code
)
1828 struct task_struct
*t
;
1831 start
->signal
->flags
= SIGNAL_GROUP_EXIT
;
1832 start
->signal
->group_exit_code
= exit_code
;
1833 start
->signal
->group_stop_count
= 0;
1837 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
1838 if (t
!= current
&& t
->mm
) {
1839 sigaddset(&t
->pending
.signal
, SIGKILL
);
1840 signal_wake_up(t
, 1);
1843 } while_each_thread(start
, t
);
1848 static inline int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
1849 struct core_state
*core_state
, int exit_code
)
1851 struct task_struct
*g
, *p
;
1852 unsigned long flags
;
1855 spin_lock_irq(&tsk
->sighand
->siglock
);
1856 if (!signal_group_exit(tsk
->signal
)) {
1857 mm
->core_state
= core_state
;
1858 nr
= zap_process(tsk
, exit_code
);
1860 spin_unlock_irq(&tsk
->sighand
->siglock
);
1861 if (unlikely(nr
< 0))
1864 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
1867 * We should find and kill all tasks which use this mm, and we should
1868 * count them correctly into ->nr_threads. We don't take tasklist
1869 * lock, but this is safe wrt:
1872 * None of sub-threads can fork after zap_process(leader). All
1873 * processes which were created before this point should be
1874 * visible to zap_threads() because copy_process() adds the new
1875 * process to the tail of init_task.tasks list, and lock/unlock
1876 * of ->siglock provides a memory barrier.
1879 * The caller holds mm->mmap_sem. This means that the task which
1880 * uses this mm can't pass exit_mm(), so it can't exit or clear
1884 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1885 * we must see either old or new leader, this does not matter.
1886 * However, it can change p->sighand, so lock_task_sighand(p)
1887 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1890 * Note also that "g" can be the old leader with ->mm == NULL
1891 * and already unhashed and thus removed from ->thread_group.
1892 * This is OK, __unhash_process()->list_del_rcu() does not
1893 * clear the ->next pointer, we will find the new leader via
1897 for_each_process(g
) {
1898 if (g
== tsk
->group_leader
)
1900 if (g
->flags
& PF_KTHREAD
)
1905 if (unlikely(p
->mm
== mm
)) {
1906 lock_task_sighand(p
, &flags
);
1907 nr
+= zap_process(p
, exit_code
);
1908 unlock_task_sighand(p
, &flags
);
1912 } while_each_thread(g
, p
);
1916 atomic_set(&core_state
->nr_threads
, nr
);
1920 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
1922 struct task_struct
*tsk
= current
;
1923 struct mm_struct
*mm
= tsk
->mm
;
1924 int core_waiters
= -EBUSY
;
1926 init_completion(&core_state
->startup
);
1927 core_state
->dumper
.task
= tsk
;
1928 core_state
->dumper
.next
= NULL
;
1930 down_write(&mm
->mmap_sem
);
1931 if (!mm
->core_state
)
1932 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
1933 up_write(&mm
->mmap_sem
);
1935 if (core_waiters
> 0)
1936 wait_for_completion(&core_state
->startup
);
1938 return core_waiters
;
1941 static void coredump_finish(struct mm_struct
*mm
)
1943 struct core_thread
*curr
, *next
;
1944 struct task_struct
*task
;
1946 next
= mm
->core_state
->dumper
.next
;
1947 while ((curr
= next
) != NULL
) {
1951 * see exit_mm(), curr->task must not see
1952 * ->task == NULL before we read ->next.
1956 wake_up_process(task
);
1959 mm
->core_state
= NULL
;
1963 * set_dumpable converts traditional three-value dumpable to two flags and
1964 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1965 * these bits are not changed atomically. So get_dumpable can observe the
1966 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1967 * return either old dumpable or new one by paying attention to the order of
1968 * modifying the bits.
1970 * dumpable | mm->flags (binary)
1971 * old new | initial interim final
1972 * ---------+-----------------------
1980 * (*) get_dumpable regards interim value of 10 as 11.
1982 void set_dumpable(struct mm_struct
*mm
, int value
)
1986 clear_bit(MMF_DUMPABLE
, &mm
->flags
);
1988 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1991 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1993 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1996 set_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1998 set_bit(MMF_DUMPABLE
, &mm
->flags
);
2003 static int __get_dumpable(unsigned long mm_flags
)
2007 ret
= mm_flags
& MMF_DUMPABLE_MASK
;
2008 return (ret
>= 2) ? 2 : ret
;
2011 int get_dumpable(struct mm_struct
*mm
)
2013 return __get_dumpable(mm
->flags
);
2016 static void wait_for_dump_helpers(struct file
*file
)
2018 struct pipe_inode_info
*pipe
;
2020 pipe
= file
->f_path
.dentry
->d_inode
->i_pipe
;
2026 while ((pipe
->readers
> 1) && (!signal_pending(current
))) {
2027 wake_up_interruptible_sync(&pipe
->wait
);
2028 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
2041 * helper function to customize the process used
2042 * to collect the core in userspace. Specifically
2043 * it sets up a pipe and installs it as fd 0 (stdin)
2044 * for the process. Returns 0 on success, or
2045 * PTR_ERR on failure.
2046 * Note that it also sets the core limit to 1. This
2047 * is a special value that we use to trap recursive
2050 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
2052 struct file
*rp
, *wp
;
2053 struct fdtable
*fdt
;
2054 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
2055 struct files_struct
*cf
= current
->files
;
2057 wp
= create_write_pipe(0);
2061 rp
= create_read_pipe(wp
, 0);
2063 free_write_pipe(wp
);
2071 spin_lock(&cf
->file_lock
);
2072 fdt
= files_fdtable(cf
);
2073 FD_SET(0, fdt
->open_fds
);
2074 FD_CLR(0, fdt
->close_on_exec
);
2075 spin_unlock(&cf
->file_lock
);
2077 /* and disallow core files too */
2078 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
2083 void do_coredump(long signr
, int exit_code
, struct pt_regs
*regs
)
2085 struct core_state core_state
;
2086 struct core_name cn
;
2087 struct mm_struct
*mm
= current
->mm
;
2088 struct linux_binfmt
* binfmt
;
2089 const struct cred
*old_cred
;
2094 static atomic_t core_dump_count
= ATOMIC_INIT(0);
2095 struct coredump_params cprm
= {
2098 .limit
= rlimit(RLIMIT_CORE
),
2100 * We must use the same mm->flags while dumping core to avoid
2101 * inconsistency of bit flags, since this flag is not protected
2104 .mm_flags
= mm
->flags
,
2107 audit_core_dumps(signr
);
2109 binfmt
= mm
->binfmt
;
2110 if (!binfmt
|| !binfmt
->core_dump
)
2112 if (!__get_dumpable(cprm
.mm_flags
))
2115 cred
= prepare_creds();
2119 * We cannot trust fsuid as being the "true" uid of the
2120 * process nor do we know its entire history. We only know it
2121 * was tainted so we dump it as root in mode 2.
2123 if (__get_dumpable(cprm
.mm_flags
) == 2) {
2124 /* Setuid core dump mode */
2125 flag
= O_EXCL
; /* Stop rewrite attacks */
2126 cred
->fsuid
= 0; /* Dump root private */
2129 retval
= coredump_wait(exit_code
, &core_state
);
2133 old_cred
= override_creds(cred
);
2136 * Clear any false indication of pending signals that might
2137 * be seen by the filesystem code called to write the core file.
2139 clear_thread_flag(TIF_SIGPENDING
);
2141 ispipe
= format_corename(&cn
, signr
);
2148 printk(KERN_WARNING
"format_corename failed\n");
2149 printk(KERN_WARNING
"Aborting core\n");
2153 if (cprm
.limit
== 1) {
2155 * Normally core limits are irrelevant to pipes, since
2156 * we're not writing to the file system, but we use
2157 * cprm.limit of 1 here as a speacial value. Any
2158 * non-1 limit gets set to RLIM_INFINITY below, but
2159 * a limit of 0 skips the dump. This is a consistent
2160 * way to catch recursive crashes. We can still crash
2161 * if the core_pattern binary sets RLIM_CORE = !1
2162 * but it runs as root, and can do lots of stupid things
2163 * Note that we use task_tgid_vnr here to grab the pid
2164 * of the process group leader. That way we get the
2165 * right pid if a thread in a multi-threaded
2166 * core_pattern process dies.
2169 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2170 task_tgid_vnr(current
), current
->comm
);
2171 printk(KERN_WARNING
"Aborting core\n");
2174 cprm
.limit
= RLIM_INFINITY
;
2176 dump_count
= atomic_inc_return(&core_dump_count
);
2177 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
2178 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
2179 task_tgid_vnr(current
), current
->comm
);
2180 printk(KERN_WARNING
"Skipping core dump\n");
2181 goto fail_dropcount
;
2184 helper_argv
= argv_split(GFP_KERNEL
, cn
.corename
+1, NULL
);
2186 printk(KERN_WARNING
"%s failed to allocate memory\n",
2188 goto fail_dropcount
;
2191 retval
= call_usermodehelper_fns(helper_argv
[0], helper_argv
,
2192 NULL
, UMH_WAIT_EXEC
, umh_pipe_setup
,
2194 argv_free(helper_argv
);
2196 printk(KERN_INFO
"Core dump to %s pipe failed\n",
2201 struct inode
*inode
;
2203 if (cprm
.limit
< binfmt
->min_coredump
)
2206 cprm
.file
= filp_open(cn
.corename
,
2207 O_CREAT
| 2 | O_NOFOLLOW
| O_LARGEFILE
| flag
,
2209 if (IS_ERR(cprm
.file
))
2212 inode
= cprm
.file
->f_path
.dentry
->d_inode
;
2213 if (inode
->i_nlink
> 1)
2215 if (d_unhashed(cprm
.file
->f_path
.dentry
))
2218 * AK: actually i see no reason to not allow this for named
2219 * pipes etc, but keep the previous behaviour for now.
2221 if (!S_ISREG(inode
->i_mode
))
2224 * Dont allow local users get cute and trick others to coredump
2225 * into their pre-created files.
2227 if (inode
->i_uid
!= current_fsuid())
2229 if (!cprm
.file
->f_op
|| !cprm
.file
->f_op
->write
)
2231 if (do_truncate(cprm
.file
->f_path
.dentry
, 0, 0, cprm
.file
))
2235 retval
= binfmt
->core_dump(&cprm
);
2237 current
->signal
->group_exit_code
|= 0x80;
2239 if (ispipe
&& core_pipe_limit
)
2240 wait_for_dump_helpers(cprm
.file
);
2243 filp_close(cprm
.file
, NULL
);
2246 atomic_dec(&core_dump_count
);
2250 coredump_finish(mm
);
2251 revert_creds(old_cred
);
2259 * Core dumping helper functions. These are the only things you should
2260 * do on a core-file: use only these functions to write out all the
2263 int dump_write(struct file
*file
, const void *addr
, int nr
)
2265 return access_ok(VERIFY_READ
, addr
, nr
) && file
->f_op
->write(file
, addr
, nr
, &file
->f_pos
) == nr
;
2267 EXPORT_SYMBOL(dump_write
);
2269 int dump_seek(struct file
*file
, loff_t off
)
2273 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
2274 if (file
->f_op
->llseek(file
, off
, SEEK_CUR
) < 0)
2277 char *buf
= (char *)get_zeroed_page(GFP_KERNEL
);
2282 unsigned long n
= off
;
2286 if (!dump_write(file
, buf
, n
)) {
2292 free_page((unsigned long)buf
);
2296 EXPORT_SYMBOL(dump_seek
);