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/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
57 #include <linux/oom.h>
59 #include <trace/events/fs.h>
61 #include <asm/uaccess.h>
62 #include <asm/mmu_context.h>
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
68 unsigned int core_pipe_limit;
69 int suid_dumpable = 0;
75 static atomic_t call_count = ATOMIC_INIT(1);
77 /* The maximal length of core_pattern is also specified in sysctl.c */
79 static LIST_HEAD(formats);
80 static DEFINE_RWLOCK(binfmt_lock);
82 int __register_binfmt(struct linux_binfmt * fmt, int insert)
86 write_lock(&binfmt_lock);
87 insert ? list_add(&fmt->lh, &formats) :
88 list_add_tail(&fmt->lh, &formats);
89 write_unlock(&binfmt_lock);
93 EXPORT_SYMBOL(__register_binfmt);
95 void unregister_binfmt(struct linux_binfmt * fmt)
97 write_lock(&binfmt_lock);
99 write_unlock(&binfmt_lock);
102 EXPORT_SYMBOL(unregister_binfmt);
104 static inline void put_binfmt(struct linux_binfmt * fmt)
106 module_put(fmt->module);
110 * Note that a shared library must be both readable and executable due to
113 * Also note that we take the address to load from from the file itself.
115 SYSCALL_DEFINE1(uselib, const char __user *, library)
118 char *tmp = getname(library);
119 int error = PTR_ERR(tmp);
124 file = do_filp_open(AT_FDCWD, tmp,
125 O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 0,
126 MAY_READ | MAY_EXEC | MAY_OPEN);
128 error = PTR_ERR(file);
133 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
137 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
144 struct linux_binfmt * fmt;
146 read_lock(&binfmt_lock);
147 list_for_each_entry(fmt, &formats, lh) {
148 if (!fmt->load_shlib)
150 if (!try_module_get(fmt->module))
152 read_unlock(&binfmt_lock);
153 error = fmt->load_shlib(file);
154 read_lock(&binfmt_lock);
156 if (error != -ENOEXEC)
159 read_unlock(&binfmt_lock);
169 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
171 struct mm_struct *mm = current->mm;
172 long diff = (long)(pages - bprm->vma_pages);
177 bprm->vma_pages = pages;
179 #ifdef SPLIT_RSS_COUNTING
180 add_mm_counter(mm, MM_ANONPAGES, diff);
182 spin_lock(&mm->page_table_lock);
183 add_mm_counter(mm, MM_ANONPAGES, diff);
184 spin_unlock(&mm->page_table_lock);
188 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
194 #ifdef CONFIG_STACK_GROWSUP
196 ret = expand_stack_downwards(bprm->vma, pos);
201 ret = get_user_pages(current, bprm->mm, pos,
202 1, write, 1, &page, NULL);
207 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
210 acct_arg_size(bprm, size / PAGE_SIZE);
213 * We've historically supported up to 32 pages (ARG_MAX)
214 * of argument strings even with small stacks
220 * Limit to 1/4-th the stack size for the argv+env strings.
222 * - the remaining binfmt code will not run out of stack space,
223 * - the program will have a reasonable amount of stack left
226 rlim = current->signal->rlim;
227 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
236 static void put_arg_page(struct page *page)
241 static void free_arg_page(struct linux_binprm *bprm, int i)
245 static void free_arg_pages(struct linux_binprm *bprm)
249 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
252 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
255 static int __bprm_mm_init(struct linux_binprm *bprm)
258 struct vm_area_struct *vma = NULL;
259 struct mm_struct *mm = bprm->mm;
261 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
265 down_write(&mm->mmap_sem);
269 * Place the stack at the largest stack address the architecture
270 * supports. Later, we'll move this to an appropriate place. We don't
271 * use STACK_TOP because that can depend on attributes which aren't
274 BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
275 vma->vm_end = STACK_TOP_MAX;
276 vma->vm_start = vma->vm_end - PAGE_SIZE;
277 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
278 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
279 INIT_LIST_HEAD(&vma->anon_vma_chain);
281 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
285 err = insert_vm_struct(mm, vma);
289 mm->stack_vm = mm->total_vm = 1;
290 up_write(&mm->mmap_sem);
291 bprm->p = vma->vm_end - sizeof(void *);
294 up_write(&mm->mmap_sem);
296 kmem_cache_free(vm_area_cachep, vma);
300 static bool valid_arg_len(struct linux_binprm *bprm, long len)
302 return len <= MAX_ARG_STRLEN;
307 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
311 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
316 page = bprm->page[pos / PAGE_SIZE];
317 if (!page && write) {
318 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
321 bprm->page[pos / PAGE_SIZE] = page;
327 static void put_arg_page(struct page *page)
331 static void free_arg_page(struct linux_binprm *bprm, int i)
334 __free_page(bprm->page[i]);
335 bprm->page[i] = NULL;
339 static void free_arg_pages(struct linux_binprm *bprm)
343 for (i = 0; i < MAX_ARG_PAGES; i++)
344 free_arg_page(bprm, i);
347 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
352 static int __bprm_mm_init(struct linux_binprm *bprm)
354 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
358 static bool valid_arg_len(struct linux_binprm *bprm, long len)
360 return len <= bprm->p;
363 #endif /* CONFIG_MMU */
366 * Create a new mm_struct and populate it with a temporary stack
367 * vm_area_struct. We don't have enough context at this point to set the stack
368 * flags, permissions, and offset, so we use temporary values. We'll update
369 * them later in setup_arg_pages().
371 int bprm_mm_init(struct linux_binprm *bprm)
374 struct mm_struct *mm = NULL;
376 bprm->mm = mm = mm_alloc();
381 err = init_new_context(current, mm);
385 err = __bprm_mm_init(bprm);
401 * count() counts the number of strings in array ARGV.
403 static int count(const char __user * const __user * argv, int max)
409 const char __user * p;
411 if (get_user(p, argv))
419 if (fatal_signal_pending(current))
420 return -ERESTARTNOHAND;
428 * 'copy_strings()' copies argument/environment strings from the old
429 * processes's memory to the new process's stack. The call to get_user_pages()
430 * ensures the destination page is created and not swapped out.
432 static int copy_strings(int argc, const char __user *const __user *argv,
433 struct linux_binprm *bprm)
435 struct page *kmapped_page = NULL;
437 unsigned long kpos = 0;
441 const char __user *str;
445 if (get_user(str, argv+argc) ||
446 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
451 if (!valid_arg_len(bprm, len)) {
456 /* We're going to work our way backwords. */
462 int offset, bytes_to_copy;
464 if (fatal_signal_pending(current)) {
465 ret = -ERESTARTNOHAND;
470 offset = pos % PAGE_SIZE;
474 bytes_to_copy = offset;
475 if (bytes_to_copy > len)
478 offset -= bytes_to_copy;
479 pos -= bytes_to_copy;
480 str -= bytes_to_copy;
481 len -= bytes_to_copy;
483 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
486 page = get_arg_page(bprm, pos, 1);
493 flush_kernel_dcache_page(kmapped_page);
494 kunmap(kmapped_page);
495 put_arg_page(kmapped_page);
498 kaddr = kmap(kmapped_page);
499 kpos = pos & PAGE_MASK;
500 flush_arg_page(bprm, kpos, kmapped_page);
502 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
511 flush_kernel_dcache_page(kmapped_page);
512 kunmap(kmapped_page);
513 put_arg_page(kmapped_page);
519 * Like copy_strings, but get argv and its values from kernel memory.
521 int copy_strings_kernel(int argc, const char *const *argv,
522 struct linux_binprm *bprm)
525 mm_segment_t oldfs = get_fs();
527 r = copy_strings(argc, (const char __user *const __user *)argv, bprm);
531 EXPORT_SYMBOL(copy_strings_kernel);
536 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
537 * the binfmt code determines where the new stack should reside, we shift it to
538 * its final location. The process proceeds as follows:
540 * 1) Use shift to calculate the new vma endpoints.
541 * 2) Extend vma to cover both the old and new ranges. This ensures the
542 * arguments passed to subsequent functions are consistent.
543 * 3) Move vma's page tables to the new range.
544 * 4) Free up any cleared pgd range.
545 * 5) Shrink the vma to cover only the new range.
547 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
549 struct mm_struct *mm = vma->vm_mm;
550 unsigned long old_start = vma->vm_start;
551 unsigned long old_end = vma->vm_end;
552 unsigned long length = old_end - old_start;
553 unsigned long new_start = old_start - shift;
554 unsigned long new_end = old_end - shift;
555 struct mmu_gather *tlb;
557 BUG_ON(new_start > new_end);
560 * ensure there are no vmas between where we want to go
563 if (vma != find_vma(mm, new_start))
567 * cover the whole range: [new_start, old_end)
569 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
573 * move the page tables downwards, on failure we rely on
574 * process cleanup to remove whatever mess we made.
576 if (length != move_page_tables(vma, old_start,
577 vma, new_start, length))
581 tlb = tlb_gather_mmu(mm, 0);
582 if (new_end > old_start) {
584 * when the old and new regions overlap clear from new_end.
586 free_pgd_range(tlb, new_end, old_end, new_end,
587 vma->vm_next ? vma->vm_next->vm_start : 0);
590 * otherwise, clean from old_start; this is done to not touch
591 * the address space in [new_end, old_start) some architectures
592 * have constraints on va-space that make this illegal (IA64) -
593 * for the others its just a little faster.
595 free_pgd_range(tlb, old_start, old_end, new_end,
596 vma->vm_next ? vma->vm_next->vm_start : 0);
598 tlb_finish_mmu(tlb, new_end, old_end);
601 * Shrink the vma to just the new range. Always succeeds.
603 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
609 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
610 * the stack is optionally relocated, and some extra space is added.
612 int setup_arg_pages(struct linux_binprm *bprm,
613 unsigned long stack_top,
614 int executable_stack)
617 unsigned long stack_shift;
618 struct mm_struct *mm = current->mm;
619 struct vm_area_struct *vma = bprm->vma;
620 struct vm_area_struct *prev = NULL;
621 unsigned long vm_flags;
622 unsigned long stack_base;
623 unsigned long stack_size;
624 unsigned long stack_expand;
625 unsigned long rlim_stack;
627 #ifdef CONFIG_STACK_GROWSUP
628 /* Limit stack size to 1GB */
629 stack_base = rlimit_max(RLIMIT_STACK);
630 if (stack_base > (1 << 30))
631 stack_base = 1 << 30;
633 /* Make sure we didn't let the argument array grow too large. */
634 if (vma->vm_end - vma->vm_start > stack_base)
637 stack_base = PAGE_ALIGN(stack_top - stack_base);
639 stack_shift = vma->vm_start - stack_base;
640 mm->arg_start = bprm->p - stack_shift;
641 bprm->p = vma->vm_end - stack_shift;
643 stack_top = arch_align_stack(stack_top);
644 stack_top = PAGE_ALIGN(stack_top);
646 if (unlikely(stack_top < mmap_min_addr) ||
647 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
650 stack_shift = vma->vm_end - stack_top;
652 bprm->p -= stack_shift;
653 mm->arg_start = bprm->p;
657 bprm->loader -= stack_shift;
658 bprm->exec -= stack_shift;
660 down_write(&mm->mmap_sem);
661 vm_flags = VM_STACK_FLAGS;
664 * Adjust stack execute permissions; explicitly enable for
665 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
666 * (arch default) otherwise.
668 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
670 else if (executable_stack == EXSTACK_DISABLE_X)
671 vm_flags &= ~VM_EXEC;
672 vm_flags |= mm->def_flags;
673 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
675 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
681 /* Move stack pages down in memory. */
683 ret = shift_arg_pages(vma, stack_shift);
688 /* mprotect_fixup is overkill to remove the temporary stack flags */
689 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
691 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
692 stack_size = vma->vm_end - vma->vm_start;
694 * Align this down to a page boundary as expand_stack
697 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
698 #ifdef CONFIG_STACK_GROWSUP
699 if (stack_size + stack_expand > rlim_stack)
700 stack_base = vma->vm_start + rlim_stack;
702 stack_base = vma->vm_end + stack_expand;
704 if (stack_size + stack_expand > rlim_stack)
705 stack_base = vma->vm_end - rlim_stack;
707 stack_base = vma->vm_start - stack_expand;
709 current->mm->start_stack = bprm->p;
710 ret = expand_stack(vma, stack_base);
715 up_write(&mm->mmap_sem);
718 EXPORT_SYMBOL(setup_arg_pages);
720 #endif /* CONFIG_MMU */
722 struct file *open_exec(const char *name)
727 file = do_filp_open(AT_FDCWD, name,
728 O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 0,
729 MAY_EXEC | MAY_OPEN);
734 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
737 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
742 trace_open_exec(name);
744 err = deny_write_access(file);
755 EXPORT_SYMBOL(open_exec);
757 int kernel_read(struct file *file, loff_t offset,
758 char *addr, unsigned long count)
766 /* The cast to a user pointer is valid due to the set_fs() */
767 result = vfs_read(file, (void __user *)addr, count, &pos);
772 EXPORT_SYMBOL(kernel_read);
774 static int exec_mmap(struct mm_struct *mm)
776 struct task_struct *tsk;
777 struct mm_struct * old_mm, *active_mm;
779 /* Notify parent that we're no longer interested in the old VM */
781 old_mm = current->mm;
782 sync_mm_rss(tsk, old_mm);
783 mm_release(tsk, old_mm);
787 * Make sure that if there is a core dump in progress
788 * for the old mm, we get out and die instead of going
789 * through with the exec. We must hold mmap_sem around
790 * checking core_state and changing tsk->mm.
792 down_read(&old_mm->mmap_sem);
793 if (unlikely(old_mm->core_state)) {
794 up_read(&old_mm->mmap_sem);
799 active_mm = tsk->active_mm;
802 activate_mm(active_mm, mm);
803 if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
804 atomic_dec(&old_mm->oom_disable_count);
805 atomic_inc(&tsk->mm->oom_disable_count);
808 arch_pick_mmap_layout(mm);
810 up_read(&old_mm->mmap_sem);
811 BUG_ON(active_mm != old_mm);
812 mm_update_next_owner(old_mm);
821 * This function makes sure the current process has its own signal table,
822 * so that flush_signal_handlers can later reset the handlers without
823 * disturbing other processes. (Other processes might share the signal
824 * table via the CLONE_SIGHAND option to clone().)
826 static int de_thread(struct task_struct *tsk)
828 struct signal_struct *sig = tsk->signal;
829 struct sighand_struct *oldsighand = tsk->sighand;
830 spinlock_t *lock = &oldsighand->siglock;
832 if (thread_group_empty(tsk))
833 goto no_thread_group;
836 * Kill all other threads in the thread group.
839 if (signal_group_exit(sig)) {
841 * Another group action in progress, just
842 * return so that the signal is processed.
844 spin_unlock_irq(lock);
848 sig->group_exit_task = tsk;
849 sig->notify_count = zap_other_threads(tsk);
850 if (!thread_group_leader(tsk))
853 while (sig->notify_count) {
854 __set_current_state(TASK_UNINTERRUPTIBLE);
855 spin_unlock_irq(lock);
859 spin_unlock_irq(lock);
862 * At this point all other threads have exited, all we have to
863 * do is to wait for the thread group leader to become inactive,
864 * and to assume its PID:
866 if (!thread_group_leader(tsk)) {
867 struct task_struct *leader = tsk->group_leader;
869 sig->notify_count = -1; /* for exit_notify() */
871 write_lock_irq(&tasklist_lock);
872 if (likely(leader->exit_state))
874 __set_current_state(TASK_UNINTERRUPTIBLE);
875 write_unlock_irq(&tasklist_lock);
880 * The only record we have of the real-time age of a
881 * process, regardless of execs it's done, is start_time.
882 * All the past CPU time is accumulated in signal_struct
883 * from sister threads now dead. But in this non-leader
884 * exec, nothing survives from the original leader thread,
885 * whose birth marks the true age of this process now.
886 * When we take on its identity by switching to its PID, we
887 * also take its birthdate (always earlier than our own).
889 tsk->start_time = leader->start_time;
891 BUG_ON(!same_thread_group(leader, tsk));
892 BUG_ON(has_group_leader_pid(tsk));
894 * An exec() starts a new thread group with the
895 * TGID of the previous thread group. Rehash the
896 * two threads with a switched PID, and release
897 * the former thread group leader:
900 /* Become a process group leader with the old leader's pid.
901 * The old leader becomes a thread of the this thread group.
902 * Note: The old leader also uses this pid until release_task
903 * is called. Odd but simple and correct.
905 detach_pid(tsk, PIDTYPE_PID);
906 tsk->pid = leader->pid;
907 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
908 transfer_pid(leader, tsk, PIDTYPE_PGID);
909 transfer_pid(leader, tsk, PIDTYPE_SID);
911 list_replace_rcu(&leader->tasks, &tsk->tasks);
912 list_replace_init(&leader->sibling, &tsk->sibling);
914 tsk->group_leader = tsk;
915 leader->group_leader = tsk;
917 tsk->exit_signal = SIGCHLD;
919 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
920 leader->exit_state = EXIT_DEAD;
921 write_unlock_irq(&tasklist_lock);
923 release_task(leader);
926 sig->group_exit_task = NULL;
927 sig->notify_count = 0;
931 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
934 flush_itimer_signals();
936 if (atomic_read(&oldsighand->count) != 1) {
937 struct sighand_struct *newsighand;
939 * This ->sighand is shared with the CLONE_SIGHAND
940 * but not CLONE_THREAD task, switch to the new one.
942 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
946 atomic_set(&newsighand->count, 1);
947 memcpy(newsighand->action, oldsighand->action,
948 sizeof(newsighand->action));
950 write_lock_irq(&tasklist_lock);
951 spin_lock(&oldsighand->siglock);
952 rcu_assign_pointer(tsk->sighand, newsighand);
953 spin_unlock(&oldsighand->siglock);
954 write_unlock_irq(&tasklist_lock);
956 __cleanup_sighand(oldsighand);
959 BUG_ON(!thread_group_leader(tsk));
964 * These functions flushes out all traces of the currently running executable
965 * so that a new one can be started
967 static void flush_old_files(struct files_struct * files)
972 spin_lock(&files->file_lock);
974 unsigned long set, i;
978 fdt = files_fdtable(files);
979 if (i >= fdt->max_fds)
981 set = fdt->close_on_exec->fds_bits[j];
984 fdt->close_on_exec->fds_bits[j] = 0;
985 spin_unlock(&files->file_lock);
986 for ( ; set ; i++,set >>= 1) {
991 spin_lock(&files->file_lock);
994 spin_unlock(&files->file_lock);
997 char *get_task_comm(char *buf, struct task_struct *tsk)
999 /* buf must be at least sizeof(tsk->comm) in size */
1001 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1006 void set_task_comm(struct task_struct *tsk, char *buf)
1011 * Threads may access current->comm without holding
1012 * the task lock, so write the string carefully.
1013 * Readers without a lock may see incomplete new
1014 * names but are safe from non-terminating string reads.
1016 memset(tsk->comm, 0, TASK_COMM_LEN);
1018 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1020 perf_event_comm(tsk);
1023 int flush_old_exec(struct linux_binprm * bprm)
1028 * Make sure we have a private signal table and that
1029 * we are unassociated from the previous thread group.
1031 retval = de_thread(current);
1035 set_mm_exe_file(bprm->mm, bprm->file);
1038 * Release all of the old mmap stuff
1040 acct_arg_size(bprm, 0);
1041 retval = exec_mmap(bprm->mm);
1045 bprm->mm = NULL; /* We're using it now */
1047 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1049 current->personality &= ~bprm->per_clear;
1056 EXPORT_SYMBOL(flush_old_exec);
1058 void setup_new_exec(struct linux_binprm * bprm)
1062 char tcomm[sizeof(current->comm)];
1064 arch_pick_mmap_layout(current->mm);
1066 /* This is the point of no return */
1067 current->sas_ss_sp = current->sas_ss_size = 0;
1069 if (current_euid() == current_uid() && current_egid() == current_gid())
1070 set_dumpable(current->mm, 1);
1072 set_dumpable(current->mm, suid_dumpable);
1074 name = bprm->filename;
1076 /* Copies the binary name from after last slash */
1077 for (i=0; (ch = *(name++)) != '\0';) {
1079 i = 0; /* overwrite what we wrote */
1081 if (i < (sizeof(tcomm) - 1))
1085 set_task_comm(current, tcomm);
1087 /* Set the new mm task size. We have to do that late because it may
1088 * depend on TIF_32BIT which is only updated in flush_thread() on
1089 * some architectures like powerpc
1091 current->mm->task_size = TASK_SIZE;
1093 /* install the new credentials */
1094 if (bprm->cred->uid != current_euid() ||
1095 bprm->cred->gid != current_egid()) {
1096 current->pdeath_signal = 0;
1097 } else if (file_permission(bprm->file, MAY_READ) ||
1098 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1099 set_dumpable(current->mm, suid_dumpable);
1103 * Flush performance counters when crossing a
1106 if (!get_dumpable(current->mm))
1107 perf_event_exit_task(current);
1109 /* An exec changes our domain. We are no longer part of the thread
1112 current->self_exec_id++;
1114 flush_signal_handlers(current, 0);
1115 flush_old_files(current->files);
1117 EXPORT_SYMBOL(setup_new_exec);
1120 * Prepare credentials and lock ->cred_guard_mutex.
1121 * install_exec_creds() commits the new creds and drops the lock.
1122 * Or, if exec fails before, free_bprm() should release ->cred and
1125 int prepare_bprm_creds(struct linux_binprm *bprm)
1127 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1128 return -ERESTARTNOINTR;
1130 bprm->cred = prepare_exec_creds();
1131 if (likely(bprm->cred))
1134 mutex_unlock(¤t->signal->cred_guard_mutex);
1138 void free_bprm(struct linux_binprm *bprm)
1140 free_arg_pages(bprm);
1142 mutex_unlock(¤t->signal->cred_guard_mutex);
1143 abort_creds(bprm->cred);
1149 * install the new credentials for this executable
1151 void install_exec_creds(struct linux_binprm *bprm)
1153 security_bprm_committing_creds(bprm);
1155 commit_creds(bprm->cred);
1158 * cred_guard_mutex must be held at least to this point to prevent
1159 * ptrace_attach() from altering our determination of the task's
1160 * credentials; any time after this it may be unlocked.
1162 security_bprm_committed_creds(bprm);
1163 mutex_unlock(¤t->signal->cred_guard_mutex);
1165 EXPORT_SYMBOL(install_exec_creds);
1168 * determine how safe it is to execute the proposed program
1169 * - the caller must hold ->cred_guard_mutex to protect against
1172 int check_unsafe_exec(struct linux_binprm *bprm)
1174 struct task_struct *p = current, *t;
1178 bprm->unsafe = tracehook_unsafe_exec(p);
1181 spin_lock(&p->fs->lock);
1183 for (t = next_thread(p); t != p; t = next_thread(t)) {
1189 if (p->fs->users > n_fs) {
1190 bprm->unsafe |= LSM_UNSAFE_SHARE;
1193 if (!p->fs->in_exec) {
1198 spin_unlock(&p->fs->lock);
1204 * Fill the binprm structure from the inode.
1205 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1207 * This may be called multiple times for binary chains (scripts for example).
1209 int prepare_binprm(struct linux_binprm *bprm)
1212 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1215 mode = inode->i_mode;
1216 if (bprm->file->f_op == NULL)
1219 /* clear any previous set[ug]id data from a previous binary */
1220 bprm->cred->euid = current_euid();
1221 bprm->cred->egid = current_egid();
1223 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1225 if (mode & S_ISUID) {
1226 bprm->per_clear |= PER_CLEAR_ON_SETID;
1227 bprm->cred->euid = inode->i_uid;
1232 * If setgid is set but no group execute bit then this
1233 * is a candidate for mandatory locking, not a setgid
1236 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1237 bprm->per_clear |= PER_CLEAR_ON_SETID;
1238 bprm->cred->egid = inode->i_gid;
1242 /* fill in binprm security blob */
1243 retval = security_bprm_set_creds(bprm);
1246 bprm->cred_prepared = 1;
1248 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1249 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1252 EXPORT_SYMBOL(prepare_binprm);
1255 * Arguments are '\0' separated strings found at the location bprm->p
1256 * points to; chop off the first by relocating brpm->p to right after
1257 * the first '\0' encountered.
1259 int remove_arg_zero(struct linux_binprm *bprm)
1262 unsigned long offset;
1270 offset = bprm->p & ~PAGE_MASK;
1271 page = get_arg_page(bprm, bprm->p, 0);
1276 kaddr = kmap_atomic(page, KM_USER0);
1278 for (; offset < PAGE_SIZE && kaddr[offset];
1279 offset++, bprm->p++)
1282 kunmap_atomic(kaddr, KM_USER0);
1285 if (offset == PAGE_SIZE)
1286 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1287 } while (offset == PAGE_SIZE);
1296 EXPORT_SYMBOL(remove_arg_zero);
1299 * cycle the list of binary formats handler, until one recognizes the image
1301 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1303 unsigned int depth = bprm->recursion_depth;
1305 struct linux_binfmt *fmt;
1307 retval = security_bprm_check(bprm);
1311 /* kernel module loader fixup */
1312 /* so we don't try to load run modprobe in kernel space. */
1315 retval = audit_bprm(bprm);
1320 for (try=0; try<2; try++) {
1321 read_lock(&binfmt_lock);
1322 list_for_each_entry(fmt, &formats, lh) {
1323 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1326 if (!try_module_get(fmt->module))
1328 read_unlock(&binfmt_lock);
1329 retval = fn(bprm, regs);
1331 * Restore the depth counter to its starting value
1332 * in this call, so we don't have to rely on every
1333 * load_binary function to restore it on return.
1335 bprm->recursion_depth = depth;
1338 tracehook_report_exec(fmt, bprm, regs);
1340 allow_write_access(bprm->file);
1344 current->did_exec = 1;
1345 proc_exec_connector(current);
1348 read_lock(&binfmt_lock);
1350 if (retval != -ENOEXEC || bprm->mm == NULL)
1353 read_unlock(&binfmt_lock);
1357 read_unlock(&binfmt_lock);
1358 if (retval != -ENOEXEC || bprm->mm == NULL) {
1360 #ifdef CONFIG_MODULES
1362 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1363 if (printable(bprm->buf[0]) &&
1364 printable(bprm->buf[1]) &&
1365 printable(bprm->buf[2]) &&
1366 printable(bprm->buf[3]))
1367 break; /* -ENOEXEC */
1368 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1375 EXPORT_SYMBOL(search_binary_handler);
1378 * sys_execve() executes a new program.
1380 int do_execve(const char * filename,
1381 const char __user *const __user *argv,
1382 const char __user *const __user *envp,
1383 struct pt_regs * regs)
1385 struct linux_binprm *bprm;
1387 struct files_struct *displaced;
1391 retval = unshare_files(&displaced);
1396 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1400 retval = prepare_bprm_creds(bprm);
1404 retval = check_unsafe_exec(bprm);
1407 clear_in_exec = retval;
1408 current->in_execve = 1;
1410 file = open_exec(filename);
1411 retval = PTR_ERR(file);
1418 bprm->filename = filename;
1419 bprm->interp = filename;
1421 retval = bprm_mm_init(bprm);
1425 bprm->argc = count(argv, MAX_ARG_STRINGS);
1426 if ((retval = bprm->argc) < 0)
1429 bprm->envc = count(envp, MAX_ARG_STRINGS);
1430 if ((retval = bprm->envc) < 0)
1433 retval = prepare_binprm(bprm);
1437 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1441 bprm->exec = bprm->p;
1442 retval = copy_strings(bprm->envc, envp, bprm);
1446 retval = copy_strings(bprm->argc, argv, bprm);
1450 retval = search_binary_handler(bprm,regs);
1454 /* execve succeeded */
1455 current->fs->in_exec = 0;
1456 current->in_execve = 0;
1457 acct_update_integrals(current);
1460 put_files_struct(displaced);
1465 acct_arg_size(bprm, 0);
1471 allow_write_access(bprm->file);
1477 current->fs->in_exec = 0;
1478 current->in_execve = 0;
1485 reset_files_struct(displaced);
1490 void set_binfmt(struct linux_binfmt *new)
1492 struct mm_struct *mm = current->mm;
1495 module_put(mm->binfmt->module);
1499 __module_get(new->module);
1502 EXPORT_SYMBOL(set_binfmt);
1504 static int expand_corename(struct core_name *cn)
1506 char *old_corename = cn->corename;
1508 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1509 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1511 if (!cn->corename) {
1512 kfree(old_corename);
1519 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1527 need = vsnprintf(NULL, 0, fmt, arg);
1530 if (likely(need < cn->size - cn->used - 1))
1533 ret = expand_corename(cn);
1538 cur = cn->corename + cn->used;
1540 vsnprintf(cur, need + 1, fmt, arg);
1549 /* format_corename will inspect the pattern parameter, and output a
1550 * name into corename, which must have space for at least
1551 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1553 static int format_corename(struct core_name *cn, long signr)
1555 const struct cred *cred = current_cred();
1556 const char *pat_ptr = core_pattern;
1557 int ispipe = (*pat_ptr == '|');
1558 int pid_in_pattern = 0;
1561 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1562 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1568 /* Repeat as long as we have more pattern to process and more output
1571 if (*pat_ptr != '%') {
1574 err = cn_printf(cn, "%c", *pat_ptr++);
1576 switch (*++pat_ptr) {
1577 /* single % at the end, drop that */
1580 /* Double percent, output one percent */
1582 err = cn_printf(cn, "%c", '%');
1587 err = cn_printf(cn, "%d",
1588 task_tgid_vnr(current));
1592 err = cn_printf(cn, "%d", cred->uid);
1596 err = cn_printf(cn, "%d", cred->gid);
1598 /* signal that caused the coredump */
1600 err = cn_printf(cn, "%ld", signr);
1602 /* UNIX time of coredump */
1605 do_gettimeofday(&tv);
1606 err = cn_printf(cn, "%lu", tv.tv_sec);
1611 down_read(&uts_sem);
1612 err = cn_printf(cn, "%s",
1613 utsname()->nodename);
1618 err = cn_printf(cn, "%s", current->comm);
1620 /* core limit size */
1622 err = cn_printf(cn, "%lu",
1623 rlimit(RLIMIT_CORE));
1635 /* Backward compatibility with core_uses_pid:
1637 * If core_pattern does not include a %p (as is the default)
1638 * and core_uses_pid is set, then .%pid will be appended to
1639 * the filename. Do not do this for piped commands. */
1640 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1641 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1649 static int zap_process(struct task_struct *start, int exit_code)
1651 struct task_struct *t;
1654 start->signal->flags = SIGNAL_GROUP_EXIT;
1655 start->signal->group_exit_code = exit_code;
1656 start->signal->group_stop_count = 0;
1660 if (t != current && t->mm) {
1661 sigaddset(&t->pending.signal, SIGKILL);
1662 signal_wake_up(t, 1);
1665 } while_each_thread(start, t);
1670 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1671 struct core_state *core_state, int exit_code)
1673 struct task_struct *g, *p;
1674 unsigned long flags;
1677 spin_lock_irq(&tsk->sighand->siglock);
1678 if (!signal_group_exit(tsk->signal)) {
1679 mm->core_state = core_state;
1680 nr = zap_process(tsk, exit_code);
1682 spin_unlock_irq(&tsk->sighand->siglock);
1683 if (unlikely(nr < 0))
1686 if (atomic_read(&mm->mm_users) == nr + 1)
1689 * We should find and kill all tasks which use this mm, and we should
1690 * count them correctly into ->nr_threads. We don't take tasklist
1691 * lock, but this is safe wrt:
1694 * None of sub-threads can fork after zap_process(leader). All
1695 * processes which were created before this point should be
1696 * visible to zap_threads() because copy_process() adds the new
1697 * process to the tail of init_task.tasks list, and lock/unlock
1698 * of ->siglock provides a memory barrier.
1701 * The caller holds mm->mmap_sem. This means that the task which
1702 * uses this mm can't pass exit_mm(), so it can't exit or clear
1706 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1707 * we must see either old or new leader, this does not matter.
1708 * However, it can change p->sighand, so lock_task_sighand(p)
1709 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1712 * Note also that "g" can be the old leader with ->mm == NULL
1713 * and already unhashed and thus removed from ->thread_group.
1714 * This is OK, __unhash_process()->list_del_rcu() does not
1715 * clear the ->next pointer, we will find the new leader via
1719 for_each_process(g) {
1720 if (g == tsk->group_leader)
1722 if (g->flags & PF_KTHREAD)
1727 if (unlikely(p->mm == mm)) {
1728 lock_task_sighand(p, &flags);
1729 nr += zap_process(p, exit_code);
1730 unlock_task_sighand(p, &flags);
1734 } while_each_thread(g, p);
1738 atomic_set(&core_state->nr_threads, nr);
1742 static int coredump_wait(int exit_code, struct core_state *core_state)
1744 struct task_struct *tsk = current;
1745 struct mm_struct *mm = tsk->mm;
1746 struct completion *vfork_done;
1747 int core_waiters = -EBUSY;
1749 init_completion(&core_state->startup);
1750 core_state->dumper.task = tsk;
1751 core_state->dumper.next = NULL;
1753 down_write(&mm->mmap_sem);
1754 if (!mm->core_state)
1755 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1756 up_write(&mm->mmap_sem);
1758 if (unlikely(core_waiters < 0))
1762 * Make sure nobody is waiting for us to release the VM,
1763 * otherwise we can deadlock when we wait on each other
1765 vfork_done = tsk->vfork_done;
1767 tsk->vfork_done = NULL;
1768 complete(vfork_done);
1772 wait_for_completion(&core_state->startup);
1774 return core_waiters;
1777 static void coredump_finish(struct mm_struct *mm)
1779 struct core_thread *curr, *next;
1780 struct task_struct *task;
1782 next = mm->core_state->dumper.next;
1783 while ((curr = next) != NULL) {
1787 * see exit_mm(), curr->task must not see
1788 * ->task == NULL before we read ->next.
1792 wake_up_process(task);
1795 mm->core_state = NULL;
1799 * set_dumpable converts traditional three-value dumpable to two flags and
1800 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1801 * these bits are not changed atomically. So get_dumpable can observe the
1802 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1803 * return either old dumpable or new one by paying attention to the order of
1804 * modifying the bits.
1806 * dumpable | mm->flags (binary)
1807 * old new | initial interim final
1808 * ---------+-----------------------
1816 * (*) get_dumpable regards interim value of 10 as 11.
1818 void set_dumpable(struct mm_struct *mm, int value)
1822 clear_bit(MMF_DUMPABLE, &mm->flags);
1824 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1827 set_bit(MMF_DUMPABLE, &mm->flags);
1829 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1832 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1834 set_bit(MMF_DUMPABLE, &mm->flags);
1839 static int __get_dumpable(unsigned long mm_flags)
1843 ret = mm_flags & MMF_DUMPABLE_MASK;
1844 return (ret >= 2) ? 2 : ret;
1847 int get_dumpable(struct mm_struct *mm)
1849 return __get_dumpable(mm->flags);
1852 static void wait_for_dump_helpers(struct file *file)
1854 struct pipe_inode_info *pipe;
1856 pipe = file->f_path.dentry->d_inode->i_pipe;
1862 while ((pipe->readers > 1) && (!signal_pending(current))) {
1863 wake_up_interruptible_sync(&pipe->wait);
1864 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1877 * helper function to customize the process used
1878 * to collect the core in userspace. Specifically
1879 * it sets up a pipe and installs it as fd 0 (stdin)
1880 * for the process. Returns 0 on success, or
1881 * PTR_ERR on failure.
1882 * Note that it also sets the core limit to 1. This
1883 * is a special value that we use to trap recursive
1886 static int umh_pipe_setup(struct subprocess_info *info)
1888 struct file *rp, *wp;
1889 struct fdtable *fdt;
1890 struct coredump_params *cp = (struct coredump_params *)info->data;
1891 struct files_struct *cf = current->files;
1893 wp = create_write_pipe(0);
1897 rp = create_read_pipe(wp, 0);
1899 free_write_pipe(wp);
1907 spin_lock(&cf->file_lock);
1908 fdt = files_fdtable(cf);
1909 FD_SET(0, fdt->open_fds);
1910 FD_CLR(0, fdt->close_on_exec);
1911 spin_unlock(&cf->file_lock);
1913 /* and disallow core files too */
1914 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1919 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1921 struct core_state core_state;
1922 struct core_name cn;
1923 struct mm_struct *mm = current->mm;
1924 struct linux_binfmt * binfmt;
1925 const struct cred *old_cred;
1930 static atomic_t core_dump_count = ATOMIC_INIT(0);
1931 struct coredump_params cprm = {
1934 .limit = rlimit(RLIMIT_CORE),
1936 * We must use the same mm->flags while dumping core to avoid
1937 * inconsistency of bit flags, since this flag is not protected
1940 .mm_flags = mm->flags,
1943 audit_core_dumps(signr);
1945 binfmt = mm->binfmt;
1946 if (!binfmt || !binfmt->core_dump)
1948 if (!__get_dumpable(cprm.mm_flags))
1951 cred = prepare_creds();
1955 * We cannot trust fsuid as being the "true" uid of the
1956 * process nor do we know its entire history. We only know it
1957 * was tainted so we dump it as root in mode 2.
1959 if (__get_dumpable(cprm.mm_flags) == 2) {
1960 /* Setuid core dump mode */
1961 flag = O_EXCL; /* Stop rewrite attacks */
1962 cred->fsuid = 0; /* Dump root private */
1965 retval = coredump_wait(exit_code, &core_state);
1969 old_cred = override_creds(cred);
1972 * Clear any false indication of pending signals that might
1973 * be seen by the filesystem code called to write the core file.
1975 clear_thread_flag(TIF_SIGPENDING);
1977 ispipe = format_corename(&cn, signr);
1979 if (ispipe == -ENOMEM) {
1980 printk(KERN_WARNING "format_corename failed\n");
1981 printk(KERN_WARNING "Aborting core\n");
1989 if (cprm.limit == 1) {
1991 * Normally core limits are irrelevant to pipes, since
1992 * we're not writing to the file system, but we use
1993 * cprm.limit of 1 here as a speacial value. Any
1994 * non-1 limit gets set to RLIM_INFINITY below, but
1995 * a limit of 0 skips the dump. This is a consistent
1996 * way to catch recursive crashes. We can still crash
1997 * if the core_pattern binary sets RLIM_CORE = !1
1998 * but it runs as root, and can do lots of stupid things
1999 * Note that we use task_tgid_vnr here to grab the pid
2000 * of the process group leader. That way we get the
2001 * right pid if a thread in a multi-threaded
2002 * core_pattern process dies.
2005 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2006 task_tgid_vnr(current), current->comm);
2007 printk(KERN_WARNING "Aborting core\n");
2010 cprm.limit = RLIM_INFINITY;
2012 dump_count = atomic_inc_return(&core_dump_count);
2013 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2014 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2015 task_tgid_vnr(current), current->comm);
2016 printk(KERN_WARNING "Skipping core dump\n");
2017 goto fail_dropcount;
2020 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2022 printk(KERN_WARNING "%s failed to allocate memory\n",
2024 goto fail_dropcount;
2027 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2028 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2030 argv_free(helper_argv);
2032 printk(KERN_INFO "Core dump to %s pipe failed\n",
2037 struct inode *inode;
2039 if (cprm.limit < binfmt->min_coredump)
2042 cprm.file = filp_open(cn.corename,
2043 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2045 if (IS_ERR(cprm.file))
2048 inode = cprm.file->f_path.dentry->d_inode;
2049 if (inode->i_nlink > 1)
2051 if (d_unhashed(cprm.file->f_path.dentry))
2054 * AK: actually i see no reason to not allow this for named
2055 * pipes etc, but keep the previous behaviour for now.
2057 if (!S_ISREG(inode->i_mode))
2060 * Dont allow local users get cute and trick others to coredump
2061 * into their pre-created files.
2063 if (inode->i_uid != current_fsuid())
2065 if (!cprm.file->f_op || !cprm.file->f_op->write)
2067 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2071 retval = binfmt->core_dump(&cprm);
2073 current->signal->group_exit_code |= 0x80;
2075 if (ispipe && core_pipe_limit)
2076 wait_for_dump_helpers(cprm.file);
2079 filp_close(cprm.file, NULL);
2082 atomic_dec(&core_dump_count);
2086 coredump_finish(mm);
2087 revert_creds(old_cred);
2095 * Core dumping helper functions. These are the only things you should
2096 * do on a core-file: use only these functions to write out all the
2099 int dump_write(struct file *file, const void *addr, int nr)
2101 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2103 EXPORT_SYMBOL(dump_write);
2105 int dump_seek(struct file *file, loff_t off)
2109 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2110 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2113 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2118 unsigned long n = off;
2122 if (!dump_write(file, buf, n)) {
2128 free_page((unsigned long)buf);
2132 EXPORT_SYMBOL(dump_seek);