Root/fs/exec.c

1/*
2 * linux/fs/exec.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * #!-checking implemented by tytso.
9 */
10/*
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.
14 *
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.
17 *
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
22 * formats.
23 */
24
25#include <linux/slab.h>
26#include <linux/file.h>
27#include <linux/fdtable.h>
28#include <linux/mm.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>
58
59#include <asm/uaccess.h>
60#include <asm/mmu_context.h>
61#include <asm/tlb.h>
62#include "internal.h"
63
64int core_uses_pid;
65char core_pattern[CORENAME_MAX_SIZE] = "core";
66unsigned int core_pipe_limit;
67int suid_dumpable = 0;
68
69struct core_name {
70    char *corename;
71    int used, size;
72};
73static atomic_t call_count = ATOMIC_INIT(1);
74
75/* The maximal length of core_pattern is also specified in sysctl.c */
76
77static LIST_HEAD(formats);
78static DEFINE_RWLOCK(binfmt_lock);
79
80int __register_binfmt(struct linux_binfmt * fmt, int insert)
81{
82    if (!fmt)
83        return -EINVAL;
84    write_lock(&binfmt_lock);
85    insert ? list_add(&fmt->lh, &formats) :
86         list_add_tail(&fmt->lh, &formats);
87    write_unlock(&binfmt_lock);
88    return 0;
89}
90
91EXPORT_SYMBOL(__register_binfmt);
92
93void unregister_binfmt(struct linux_binfmt * fmt)
94{
95    write_lock(&binfmt_lock);
96    list_del(&fmt->lh);
97    write_unlock(&binfmt_lock);
98}
99
100EXPORT_SYMBOL(unregister_binfmt);
101
102static inline void put_binfmt(struct linux_binfmt * fmt)
103{
104    module_put(fmt->module);
105}
106
107/*
108 * Note that a shared library must be both readable and executable due to
109 * security reasons.
110 *
111 * Also note that we take the address to load from from the file itself.
112 */
113SYSCALL_DEFINE1(uselib, const char __user *, library)
114{
115    struct file *file;
116    char *tmp = getname(library);
117    int error = PTR_ERR(tmp);
118    static const struct open_flags uselib_flags = {
119        .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
120        .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
121        .intent = LOOKUP_OPEN
122    };
123
124    if (IS_ERR(tmp))
125        goto out;
126
127    file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
128    putname(tmp);
129    error = PTR_ERR(file);
130    if (IS_ERR(file))
131        goto out;
132
133    error = -EINVAL;
134    if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
135        goto exit;
136
137    error = -EACCES;
138    if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
139        goto exit;
140
141    fsnotify_open(file);
142
143    error = -ENOEXEC;
144    if(file->f_op) {
145        struct linux_binfmt * fmt;
146
147        read_lock(&binfmt_lock);
148        list_for_each_entry(fmt, &formats, lh) {
149            if (!fmt->load_shlib)
150                continue;
151            if (!try_module_get(fmt->module))
152                continue;
153            read_unlock(&binfmt_lock);
154            error = fmt->load_shlib(file);
155            read_lock(&binfmt_lock);
156            put_binfmt(fmt);
157            if (error != -ENOEXEC)
158                break;
159        }
160        read_unlock(&binfmt_lock);
161    }
162exit:
163    fput(file);
164out:
165      return error;
166}
167
168#ifdef CONFIG_MMU
169/*
170 * The nascent bprm->mm is not visible until exec_mmap() but it can
171 * use a lot of memory, account these pages in current->mm temporary
172 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
173 * change the counter back via acct_arg_size(0).
174 */
175static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
176{
177    struct mm_struct *mm = current->mm;
178    long diff = (long)(pages - bprm->vma_pages);
179
180    if (!mm || !diff)
181        return;
182
183    bprm->vma_pages = pages;
184
185#ifdef SPLIT_RSS_COUNTING
186    add_mm_counter(mm, MM_ANONPAGES, diff);
187#else
188    spin_lock(&mm->page_table_lock);
189    add_mm_counter(mm, MM_ANONPAGES, diff);
190    spin_unlock(&mm->page_table_lock);
191#endif
192}
193
194static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
195        int write)
196{
197    struct page *page;
198    int ret;
199
200#ifdef CONFIG_STACK_GROWSUP
201    if (write) {
202        ret = expand_downwards(bprm->vma, pos);
203        if (ret < 0)
204            return NULL;
205    }
206#endif
207    ret = get_user_pages(current, bprm->mm, pos,
208            1, write, 1, &page, NULL);
209    if (ret <= 0)
210        return NULL;
211
212    if (write) {
213        unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
214        struct rlimit *rlim;
215
216        acct_arg_size(bprm, size / PAGE_SIZE);
217
218        /*
219         * We've historically supported up to 32 pages (ARG_MAX)
220         * of argument strings even with small stacks
221         */
222        if (size <= ARG_MAX)
223            return page;
224
225        /*
226         * Limit to 1/4-th the stack size for the argv+env strings.
227         * This ensures that:
228         * - the remaining binfmt code will not run out of stack space,
229         * - the program will have a reasonable amount of stack left
230         * to work from.
231         */
232        rlim = current->signal->rlim;
233        if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
234            put_page(page);
235            return NULL;
236        }
237    }
238
239    return page;
240}
241
242static void put_arg_page(struct page *page)
243{
244    put_page(page);
245}
246
247static void free_arg_page(struct linux_binprm *bprm, int i)
248{
249}
250
251static void free_arg_pages(struct linux_binprm *bprm)
252{
253}
254
255static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
256        struct page *page)
257{
258    flush_cache_page(bprm->vma, pos, page_to_pfn(page));
259}
260
261static int __bprm_mm_init(struct linux_binprm *bprm)
262{
263    int err;
264    struct vm_area_struct *vma = NULL;
265    struct mm_struct *mm = bprm->mm;
266
267    bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
268    if (!vma)
269        return -ENOMEM;
270
271    down_write(&mm->mmap_sem);
272    vma->vm_mm = mm;
273
274    /*
275     * Place the stack at the largest stack address the architecture
276     * supports. Later, we'll move this to an appropriate place. We don't
277     * use STACK_TOP because that can depend on attributes which aren't
278     * configured yet.
279     */
280    BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
281    vma->vm_end = STACK_TOP_MAX;
282    vma->vm_start = vma->vm_end - PAGE_SIZE;
283    vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
284    vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
285    INIT_LIST_HEAD(&vma->anon_vma_chain);
286
287    err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
288    if (err)
289        goto err;
290
291    err = insert_vm_struct(mm, vma);
292    if (err)
293        goto err;
294
295    mm->stack_vm = mm->total_vm = 1;
296    up_write(&mm->mmap_sem);
297    bprm->p = vma->vm_end - sizeof(void *);
298    return 0;
299err:
300    up_write(&mm->mmap_sem);
301    bprm->vma = NULL;
302    kmem_cache_free(vm_area_cachep, vma);
303    return err;
304}
305
306static bool valid_arg_len(struct linux_binprm *bprm, long len)
307{
308    return len <= MAX_ARG_STRLEN;
309}
310
311#else
312
313static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
314{
315}
316
317static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
318        int write)
319{
320    struct page *page;
321
322    page = bprm->page[pos / PAGE_SIZE];
323    if (!page && write) {
324        page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
325        if (!page)
326            return NULL;
327        bprm->page[pos / PAGE_SIZE] = page;
328    }
329
330    return page;
331}
332
333static void put_arg_page(struct page *page)
334{
335}
336
337static void free_arg_page(struct linux_binprm *bprm, int i)
338{
339    if (bprm->page[i]) {
340        __free_page(bprm->page[i]);
341        bprm->page[i] = NULL;
342    }
343}
344
345static void free_arg_pages(struct linux_binprm *bprm)
346{
347    int i;
348
349    for (i = 0; i < MAX_ARG_PAGES; i++)
350        free_arg_page(bprm, i);
351}
352
353static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
354        struct page *page)
355{
356}
357
358static int __bprm_mm_init(struct linux_binprm *bprm)
359{
360    bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
361    return 0;
362}
363
364static bool valid_arg_len(struct linux_binprm *bprm, long len)
365{
366    return len <= bprm->p;
367}
368
369#endif /* CONFIG_MMU */
370
371/*
372 * Create a new mm_struct and populate it with a temporary stack
373 * vm_area_struct. We don't have enough context at this point to set the stack
374 * flags, permissions, and offset, so we use temporary values. We'll update
375 * them later in setup_arg_pages().
376 */
377int bprm_mm_init(struct linux_binprm *bprm)
378{
379    int err;
380    struct mm_struct *mm = NULL;
381
382    bprm->mm = mm = mm_alloc();
383    err = -ENOMEM;
384    if (!mm)
385        goto err;
386
387    err = init_new_context(current, mm);
388    if (err)
389        goto err;
390
391    err = __bprm_mm_init(bprm);
392    if (err)
393        goto err;
394
395    return 0;
396
397err:
398    if (mm) {
399        bprm->mm = NULL;
400        mmdrop(mm);
401    }
402
403    return err;
404}
405
406struct user_arg_ptr {
407#ifdef CONFIG_COMPAT
408    bool is_compat;
409#endif
410    union {
411        const char __user *const __user *native;
412#ifdef CONFIG_COMPAT
413        compat_uptr_t __user *compat;
414#endif
415    } ptr;
416};
417
418static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
419{
420    const char __user *native;
421
422#ifdef CONFIG_COMPAT
423    if (unlikely(argv.is_compat)) {
424        compat_uptr_t compat;
425
426        if (get_user(compat, argv.ptr.compat + nr))
427            return ERR_PTR(-EFAULT);
428
429        return compat_ptr(compat);
430    }
431#endif
432
433    if (get_user(native, argv.ptr.native + nr))
434        return ERR_PTR(-EFAULT);
435
436    return native;
437}
438
439/*
440 * count() counts the number of strings in array ARGV.
441 */
442static int count(struct user_arg_ptr argv, int max)
443{
444    int i = 0;
445
446    if (argv.ptr.native != NULL) {
447        for (;;) {
448            const char __user *p = get_user_arg_ptr(argv, i);
449
450            if (!p)
451                break;
452
453            if (IS_ERR(p))
454                return -EFAULT;
455
456            if (i++ >= max)
457                return -E2BIG;
458
459            if (fatal_signal_pending(current))
460                return -ERESTARTNOHAND;
461            cond_resched();
462        }
463    }
464    return i;
465}
466
467/*
468 * 'copy_strings()' copies argument/environment strings from the old
469 * processes's memory to the new process's stack. The call to get_user_pages()
470 * ensures the destination page is created and not swapped out.
471 */
472static int copy_strings(int argc, struct user_arg_ptr argv,
473            struct linux_binprm *bprm)
474{
475    struct page *kmapped_page = NULL;
476    char *kaddr = NULL;
477    unsigned long kpos = 0;
478    int ret;
479
480    while (argc-- > 0) {
481        const char __user *str;
482        int len;
483        unsigned long pos;
484
485        ret = -EFAULT;
486        str = get_user_arg_ptr(argv, argc);
487        if (IS_ERR(str))
488            goto out;
489
490        len = strnlen_user(str, MAX_ARG_STRLEN);
491        if (!len)
492            goto out;
493
494        ret = -E2BIG;
495        if (!valid_arg_len(bprm, len))
496            goto out;
497
498        /* We're going to work our way backwords. */
499        pos = bprm->p;
500        str += len;
501        bprm->p -= len;
502
503        while (len > 0) {
504            int offset, bytes_to_copy;
505
506            if (fatal_signal_pending(current)) {
507                ret = -ERESTARTNOHAND;
508                goto out;
509            }
510            cond_resched();
511
512            offset = pos % PAGE_SIZE;
513            if (offset == 0)
514                offset = PAGE_SIZE;
515
516            bytes_to_copy = offset;
517            if (bytes_to_copy > len)
518                bytes_to_copy = len;
519
520            offset -= bytes_to_copy;
521            pos -= bytes_to_copy;
522            str -= bytes_to_copy;
523            len -= bytes_to_copy;
524
525            if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
526                struct page *page;
527
528                page = get_arg_page(bprm, pos, 1);
529                if (!page) {
530                    ret = -E2BIG;
531                    goto out;
532                }
533
534                if (kmapped_page) {
535                    flush_kernel_dcache_page(kmapped_page);
536                    kunmap(kmapped_page);
537                    put_arg_page(kmapped_page);
538                }
539                kmapped_page = page;
540                kaddr = kmap(kmapped_page);
541                kpos = pos & PAGE_MASK;
542                flush_arg_page(bprm, kpos, kmapped_page);
543            }
544            if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
545                ret = -EFAULT;
546                goto out;
547            }
548        }
549    }
550    ret = 0;
551out:
552    if (kmapped_page) {
553        flush_kernel_dcache_page(kmapped_page);
554        kunmap(kmapped_page);
555        put_arg_page(kmapped_page);
556    }
557    return ret;
558}
559
560/*
561 * Like copy_strings, but get argv and its values from kernel memory.
562 */
563int copy_strings_kernel(int argc, const char *const *__argv,
564            struct linux_binprm *bprm)
565{
566    int r;
567    mm_segment_t oldfs = get_fs();
568    struct user_arg_ptr argv = {
569        .ptr.native = (const char __user *const __user *)__argv,
570    };
571
572    set_fs(KERNEL_DS);
573    r = copy_strings(argc, argv, bprm);
574    set_fs(oldfs);
575
576    return r;
577}
578EXPORT_SYMBOL(copy_strings_kernel);
579
580#ifdef CONFIG_MMU
581
582/*
583 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
584 * the binfmt code determines where the new stack should reside, we shift it to
585 * its final location. The process proceeds as follows:
586 *
587 * 1) Use shift to calculate the new vma endpoints.
588 * 2) Extend vma to cover both the old and new ranges. This ensures the
589 * arguments passed to subsequent functions are consistent.
590 * 3) Move vma's page tables to the new range.
591 * 4) Free up any cleared pgd range.
592 * 5) Shrink the vma to cover only the new range.
593 */
594static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
595{
596    struct mm_struct *mm = vma->vm_mm;
597    unsigned long old_start = vma->vm_start;
598    unsigned long old_end = vma->vm_end;
599    unsigned long length = old_end - old_start;
600    unsigned long new_start = old_start - shift;
601    unsigned long new_end = old_end - shift;
602    struct mmu_gather tlb;
603
604    BUG_ON(new_start > new_end);
605
606    /*
607     * ensure there are no vmas between where we want to go
608     * and where we are
609     */
610    if (vma != find_vma(mm, new_start))
611        return -EFAULT;
612
613    /*
614     * cover the whole range: [new_start, old_end)
615     */
616    if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
617        return -ENOMEM;
618
619    /*
620     * move the page tables downwards, on failure we rely on
621     * process cleanup to remove whatever mess we made.
622     */
623    if (length != move_page_tables(vma, old_start,
624                       vma, new_start, length))
625        return -ENOMEM;
626
627    lru_add_drain();
628    tlb_gather_mmu(&tlb, mm, 0);
629    if (new_end > old_start) {
630        /*
631         * when the old and new regions overlap clear from new_end.
632         */
633        free_pgd_range(&tlb, new_end, old_end, new_end,
634            vma->vm_next ? vma->vm_next->vm_start : 0);
635    } else {
636        /*
637         * otherwise, clean from old_start; this is done to not touch
638         * the address space in [new_end, old_start) some architectures
639         * have constraints on va-space that make this illegal (IA64) -
640         * for the others its just a little faster.
641         */
642        free_pgd_range(&tlb, old_start, old_end, new_end,
643            vma->vm_next ? vma->vm_next->vm_start : 0);
644    }
645    tlb_finish_mmu(&tlb, new_end, old_end);
646
647    /*
648     * Shrink the vma to just the new range. Always succeeds.
649     */
650    vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
651
652    return 0;
653}
654
655/*
656 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
657 * the stack is optionally relocated, and some extra space is added.
658 */
659int setup_arg_pages(struct linux_binprm *bprm,
660            unsigned long stack_top,
661            int executable_stack)
662{
663    unsigned long ret;
664    unsigned long stack_shift;
665    struct mm_struct *mm = current->mm;
666    struct vm_area_struct *vma = bprm->vma;
667    struct vm_area_struct *prev = NULL;
668    unsigned long vm_flags;
669    unsigned long stack_base;
670    unsigned long stack_size;
671    unsigned long stack_expand;
672    unsigned long rlim_stack;
673
674#ifdef CONFIG_STACK_GROWSUP
675    /* Limit stack size to 1GB */
676    stack_base = rlimit_max(RLIMIT_STACK);
677    if (stack_base > (1 << 30))
678        stack_base = 1 << 30;
679
680    /* Make sure we didn't let the argument array grow too large. */
681    if (vma->vm_end - vma->vm_start > stack_base)
682        return -ENOMEM;
683
684    stack_base = PAGE_ALIGN(stack_top - stack_base);
685
686    stack_shift = vma->vm_start - stack_base;
687    mm->arg_start = bprm->p - stack_shift;
688    bprm->p = vma->vm_end - stack_shift;
689#else
690    stack_top = arch_align_stack(stack_top);
691    stack_top = PAGE_ALIGN(stack_top);
692
693    if (unlikely(stack_top < mmap_min_addr) ||
694        unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
695        return -ENOMEM;
696
697    stack_shift = vma->vm_end - stack_top;
698
699    bprm->p -= stack_shift;
700    mm->arg_start = bprm->p;
701#endif
702
703    if (bprm->loader)
704        bprm->loader -= stack_shift;
705    bprm->exec -= stack_shift;
706
707    down_write(&mm->mmap_sem);
708    vm_flags = VM_STACK_FLAGS;
709
710    /*
711     * Adjust stack execute permissions; explicitly enable for
712     * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
713     * (arch default) otherwise.
714     */
715    if (unlikely(executable_stack == EXSTACK_ENABLE_X))
716        vm_flags |= VM_EXEC;
717    else if (executable_stack == EXSTACK_DISABLE_X)
718        vm_flags &= ~VM_EXEC;
719    vm_flags |= mm->def_flags;
720    vm_flags |= VM_STACK_INCOMPLETE_SETUP;
721
722    ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
723            vm_flags);
724    if (ret)
725        goto out_unlock;
726    BUG_ON(prev != vma);
727
728    /* Move stack pages down in memory. */
729    if (stack_shift) {
730        ret = shift_arg_pages(vma, stack_shift);
731        if (ret)
732            goto out_unlock;
733    }
734
735    /* mprotect_fixup is overkill to remove the temporary stack flags */
736    vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
737
738    stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
739    stack_size = vma->vm_end - vma->vm_start;
740    /*
741     * Align this down to a page boundary as expand_stack
742     * will align it up.
743     */
744    rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
745#ifdef CONFIG_STACK_GROWSUP
746    if (stack_size + stack_expand > rlim_stack)
747        stack_base = vma->vm_start + rlim_stack;
748    else
749        stack_base = vma->vm_end + stack_expand;
750#else
751    if (stack_size + stack_expand > rlim_stack)
752        stack_base = vma->vm_end - rlim_stack;
753    else
754        stack_base = vma->vm_start - stack_expand;
755#endif
756    current->mm->start_stack = bprm->p;
757    ret = expand_stack(vma, stack_base);
758    if (ret)
759        ret = -EFAULT;
760
761out_unlock:
762    up_write(&mm->mmap_sem);
763    return ret;
764}
765EXPORT_SYMBOL(setup_arg_pages);
766
767#endif /* CONFIG_MMU */
768
769struct file *open_exec(const char *name)
770{
771    struct file *file;
772    int err;
773    static const struct open_flags open_exec_flags = {
774        .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
775        .acc_mode = MAY_EXEC | MAY_OPEN,
776        .intent = LOOKUP_OPEN
777    };
778
779    file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
780    if (IS_ERR(file))
781        goto out;
782
783    err = -EACCES;
784    if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
785        goto exit;
786
787    if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
788        goto exit;
789
790    fsnotify_open(file);
791
792    err = deny_write_access(file);
793    if (err)
794        goto exit;
795
796out:
797    return file;
798
799exit:
800    fput(file);
801    return ERR_PTR(err);
802}
803EXPORT_SYMBOL(open_exec);
804
805int kernel_read(struct file *file, loff_t offset,
806        char *addr, unsigned long count)
807{
808    mm_segment_t old_fs;
809    loff_t pos = offset;
810    int result;
811
812    old_fs = get_fs();
813    set_fs(get_ds());
814    /* The cast to a user pointer is valid due to the set_fs() */
815    result = vfs_read(file, (void __user *)addr, count, &pos);
816    set_fs(old_fs);
817    return result;
818}
819
820EXPORT_SYMBOL(kernel_read);
821
822static int exec_mmap(struct mm_struct *mm)
823{
824    struct task_struct *tsk;
825    struct mm_struct * old_mm, *active_mm;
826
827    /* Notify parent that we're no longer interested in the old VM */
828    tsk = current;
829    old_mm = current->mm;
830    sync_mm_rss(tsk, old_mm);
831    mm_release(tsk, old_mm);
832
833    if (old_mm) {
834        /*
835         * Make sure that if there is a core dump in progress
836         * for the old mm, we get out and die instead of going
837         * through with the exec. We must hold mmap_sem around
838         * checking core_state and changing tsk->mm.
839         */
840        down_read(&old_mm->mmap_sem);
841        if (unlikely(old_mm->core_state)) {
842            up_read(&old_mm->mmap_sem);
843            return -EINTR;
844        }
845    }
846    task_lock(tsk);
847    active_mm = tsk->active_mm;
848    tsk->mm = mm;
849    tsk->active_mm = mm;
850    activate_mm(active_mm, mm);
851    if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
852        atomic_dec(&old_mm->oom_disable_count);
853        atomic_inc(&tsk->mm->oom_disable_count);
854    }
855    task_unlock(tsk);
856    arch_pick_mmap_layout(mm);
857    if (old_mm) {
858        up_read(&old_mm->mmap_sem);
859        BUG_ON(active_mm != old_mm);
860        mm_update_next_owner(old_mm);
861        mmput(old_mm);
862        return 0;
863    }
864    mmdrop(active_mm);
865    return 0;
866}
867
868/*
869 * This function makes sure the current process has its own signal table,
870 * so that flush_signal_handlers can later reset the handlers without
871 * disturbing other processes. (Other processes might share the signal
872 * table via the CLONE_SIGHAND option to clone().)
873 */
874static int de_thread(struct task_struct *tsk)
875{
876    struct signal_struct *sig = tsk->signal;
877    struct sighand_struct *oldsighand = tsk->sighand;
878    spinlock_t *lock = &oldsighand->siglock;
879
880    if (thread_group_empty(tsk))
881        goto no_thread_group;
882
883    /*
884     * Kill all other threads in the thread group.
885     */
886    spin_lock_irq(lock);
887    if (signal_group_exit(sig)) {
888        /*
889         * Another group action in progress, just
890         * return so that the signal is processed.
891         */
892        spin_unlock_irq(lock);
893        return -EAGAIN;
894    }
895
896    sig->group_exit_task = tsk;
897    sig->notify_count = zap_other_threads(tsk);
898    if (!thread_group_leader(tsk))
899        sig->notify_count--;
900
901    while (sig->notify_count) {
902        __set_current_state(TASK_UNINTERRUPTIBLE);
903        spin_unlock_irq(lock);
904        schedule();
905        spin_lock_irq(lock);
906    }
907    spin_unlock_irq(lock);
908
909    /*
910     * At this point all other threads have exited, all we have to
911     * do is to wait for the thread group leader to become inactive,
912     * and to assume its PID:
913     */
914    if (!thread_group_leader(tsk)) {
915        struct task_struct *leader = tsk->group_leader;
916
917        sig->notify_count = -1; /* for exit_notify() */
918        for (;;) {
919            write_lock_irq(&tasklist_lock);
920            if (likely(leader->exit_state))
921                break;
922            __set_current_state(TASK_UNINTERRUPTIBLE);
923            write_unlock_irq(&tasklist_lock);
924            schedule();
925        }
926
927        /*
928         * The only record we have of the real-time age of a
929         * process, regardless of execs it's done, is start_time.
930         * All the past CPU time is accumulated in signal_struct
931         * from sister threads now dead. But in this non-leader
932         * exec, nothing survives from the original leader thread,
933         * whose birth marks the true age of this process now.
934         * When we take on its identity by switching to its PID, we
935         * also take its birthdate (always earlier than our own).
936         */
937        tsk->start_time = leader->start_time;
938
939        BUG_ON(!same_thread_group(leader, tsk));
940        BUG_ON(has_group_leader_pid(tsk));
941        /*
942         * An exec() starts a new thread group with the
943         * TGID of the previous thread group. Rehash the
944         * two threads with a switched PID, and release
945         * the former thread group leader:
946         */
947
948        /* Become a process group leader with the old leader's pid.
949         * The old leader becomes a thread of the this thread group.
950         * Note: The old leader also uses this pid until release_task
951         * is called. Odd but simple and correct.
952         */
953        detach_pid(tsk, PIDTYPE_PID);
954        tsk->pid = leader->pid;
955        attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
956        transfer_pid(leader, tsk, PIDTYPE_PGID);
957        transfer_pid(leader, tsk, PIDTYPE_SID);
958
959        list_replace_rcu(&leader->tasks, &tsk->tasks);
960        list_replace_init(&leader->sibling, &tsk->sibling);
961
962        tsk->group_leader = tsk;
963        leader->group_leader = tsk;
964
965        tsk->exit_signal = SIGCHLD;
966
967        BUG_ON(leader->exit_state != EXIT_ZOMBIE);
968        leader->exit_state = EXIT_DEAD;
969        write_unlock_irq(&tasklist_lock);
970
971        release_task(leader);
972    }
973
974    sig->group_exit_task = NULL;
975    sig->notify_count = 0;
976
977no_thread_group:
978    if (current->mm)
979        setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
980
981    exit_itimers(sig);
982    flush_itimer_signals();
983
984    if (atomic_read(&oldsighand->count) != 1) {
985        struct sighand_struct *newsighand;
986        /*
987         * This ->sighand is shared with the CLONE_SIGHAND
988         * but not CLONE_THREAD task, switch to the new one.
989         */
990        newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
991        if (!newsighand)
992            return -ENOMEM;
993
994        atomic_set(&newsighand->count, 1);
995        memcpy(newsighand->action, oldsighand->action,
996               sizeof(newsighand->action));
997
998        write_lock_irq(&tasklist_lock);
999        spin_lock(&oldsighand->siglock);
1000        rcu_assign_pointer(tsk->sighand, newsighand);
1001        spin_unlock(&oldsighand->siglock);
1002        write_unlock_irq(&tasklist_lock);
1003
1004        __cleanup_sighand(oldsighand);
1005    }
1006
1007    BUG_ON(!thread_group_leader(tsk));
1008    return 0;
1009}
1010
1011/*
1012 * These functions flushes out all traces of the currently running executable
1013 * so that a new one can be started
1014 */
1015static void flush_old_files(struct files_struct * files)
1016{
1017    long j = -1;
1018    struct fdtable *fdt;
1019
1020    spin_lock(&files->file_lock);
1021    for (;;) {
1022        unsigned long set, i;
1023
1024        j++;
1025        i = j * __NFDBITS;
1026        fdt = files_fdtable(files);
1027        if (i >= fdt->max_fds)
1028            break;
1029        set = fdt->close_on_exec->fds_bits[j];
1030        if (!set)
1031            continue;
1032        fdt->close_on_exec->fds_bits[j] = 0;
1033        spin_unlock(&files->file_lock);
1034        for ( ; set ; i++,set >>= 1) {
1035            if (set & 1) {
1036                sys_close(i);
1037            }
1038        }
1039        spin_lock(&files->file_lock);
1040
1041    }
1042    spin_unlock(&files->file_lock);
1043}
1044
1045char *get_task_comm(char *buf, struct task_struct *tsk)
1046{
1047    /* buf must be at least sizeof(tsk->comm) in size */
1048    task_lock(tsk);
1049    strncpy(buf, tsk->comm, sizeof(tsk->comm));
1050    task_unlock(tsk);
1051    return buf;
1052}
1053EXPORT_SYMBOL_GPL(get_task_comm);
1054
1055void set_task_comm(struct task_struct *tsk, char *buf)
1056{
1057    task_lock(tsk);
1058
1059    /*
1060     * Threads may access current->comm without holding
1061     * the task lock, so write the string carefully.
1062     * Readers without a lock may see incomplete new
1063     * names but are safe from non-terminating string reads.
1064     */
1065    memset(tsk->comm, 0, TASK_COMM_LEN);
1066    wmb();
1067    strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1068    task_unlock(tsk);
1069    perf_event_comm(tsk);
1070}
1071
1072int flush_old_exec(struct linux_binprm * bprm)
1073{
1074    int retval;
1075
1076    /*
1077     * Make sure we have a private signal table and that
1078     * we are unassociated from the previous thread group.
1079     */
1080    retval = de_thread(current);
1081    if (retval)
1082        goto out;
1083
1084    set_mm_exe_file(bprm->mm, bprm->file);
1085
1086    /*
1087     * Release all of the old mmap stuff
1088     */
1089    acct_arg_size(bprm, 0);
1090    retval = exec_mmap(bprm->mm);
1091    if (retval)
1092        goto out;
1093
1094    bprm->mm = NULL; /* We're using it now */
1095
1096    set_fs(USER_DS);
1097    current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1098    flush_thread();
1099    current->personality &= ~bprm->per_clear;
1100
1101    return 0;
1102
1103out:
1104    return retval;
1105}
1106EXPORT_SYMBOL(flush_old_exec);
1107
1108void setup_new_exec(struct linux_binprm * bprm)
1109{
1110    int i, ch;
1111    const char *name;
1112    char tcomm[sizeof(current->comm)];
1113
1114    arch_pick_mmap_layout(current->mm);
1115
1116    /* This is the point of no return */
1117    current->sas_ss_sp = current->sas_ss_size = 0;
1118
1119    if (current_euid() == current_uid() && current_egid() == current_gid())
1120        set_dumpable(current->mm, 1);
1121    else
1122        set_dumpable(current->mm, suid_dumpable);
1123
1124    name = bprm->filename;
1125
1126    /* Copies the binary name from after last slash */
1127    for (i=0; (ch = *(name++)) != '\0';) {
1128        if (ch == '/')
1129            i = 0; /* overwrite what we wrote */
1130        else
1131            if (i < (sizeof(tcomm) - 1))
1132                tcomm[i++] = ch;
1133    }
1134    tcomm[i] = '\0';
1135    set_task_comm(current, tcomm);
1136
1137    /* Set the new mm task size. We have to do that late because it may
1138     * depend on TIF_32BIT which is only updated in flush_thread() on
1139     * some architectures like powerpc
1140     */
1141    current->mm->task_size = TASK_SIZE;
1142
1143    /* install the new credentials */
1144    if (bprm->cred->uid != current_euid() ||
1145        bprm->cred->gid != current_egid()) {
1146        current->pdeath_signal = 0;
1147    } else if (file_permission(bprm->file, MAY_READ) ||
1148           bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1149        set_dumpable(current->mm, suid_dumpable);
1150    }
1151
1152    /*
1153     * Flush performance counters when crossing a
1154     * security domain:
1155     */
1156    if (!get_dumpable(current->mm))
1157        perf_event_exit_task(current);
1158
1159    /* An exec changes our domain. We are no longer part of the thread
1160       group */
1161
1162    current->self_exec_id++;
1163            
1164    flush_signal_handlers(current, 0);
1165    flush_old_files(current->files);
1166}
1167EXPORT_SYMBOL(setup_new_exec);
1168
1169/*
1170 * Prepare credentials and lock ->cred_guard_mutex.
1171 * install_exec_creds() commits the new creds and drops the lock.
1172 * Or, if exec fails before, free_bprm() should release ->cred and
1173 * and unlock.
1174 */
1175int prepare_bprm_creds(struct linux_binprm *bprm)
1176{
1177    if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1178        return -ERESTARTNOINTR;
1179
1180    bprm->cred = prepare_exec_creds();
1181    if (likely(bprm->cred))
1182        return 0;
1183
1184    mutex_unlock(&current->signal->cred_guard_mutex);
1185    return -ENOMEM;
1186}
1187
1188void free_bprm(struct linux_binprm *bprm)
1189{
1190    free_arg_pages(bprm);
1191    if (bprm->cred) {
1192        mutex_unlock(&current->signal->cred_guard_mutex);
1193        abort_creds(bprm->cred);
1194    }
1195    kfree(bprm);
1196}
1197
1198/*
1199 * install the new credentials for this executable
1200 */
1201void install_exec_creds(struct linux_binprm *bprm)
1202{
1203    security_bprm_committing_creds(bprm);
1204
1205    commit_creds(bprm->cred);
1206    bprm->cred = NULL;
1207    /*
1208     * cred_guard_mutex must be held at least to this point to prevent
1209     * ptrace_attach() from altering our determination of the task's
1210     * credentials; any time after this it may be unlocked.
1211     */
1212    security_bprm_committed_creds(bprm);
1213    mutex_unlock(&current->signal->cred_guard_mutex);
1214}
1215EXPORT_SYMBOL(install_exec_creds);
1216
1217/*
1218 * determine how safe it is to execute the proposed program
1219 * - the caller must hold ->cred_guard_mutex to protect against
1220 * PTRACE_ATTACH
1221 */
1222int check_unsafe_exec(struct linux_binprm *bprm)
1223{
1224    struct task_struct *p = current, *t;
1225    unsigned n_fs;
1226    int res = 0;
1227
1228    bprm->unsafe = tracehook_unsafe_exec(p);
1229
1230    n_fs = 1;
1231    spin_lock(&p->fs->lock);
1232    rcu_read_lock();
1233    for (t = next_thread(p); t != p; t = next_thread(t)) {
1234        if (t->fs == p->fs)
1235            n_fs++;
1236    }
1237    rcu_read_unlock();
1238
1239    if (p->fs->users > n_fs) {
1240        bprm->unsafe |= LSM_UNSAFE_SHARE;
1241    } else {
1242        res = -EAGAIN;
1243        if (!p->fs->in_exec) {
1244            p->fs->in_exec = 1;
1245            res = 1;
1246        }
1247    }
1248    spin_unlock(&p->fs->lock);
1249
1250    return res;
1251}
1252
1253/*
1254 * Fill the binprm structure from the inode.
1255 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1256 *
1257 * This may be called multiple times for binary chains (scripts for example).
1258 */
1259int prepare_binprm(struct linux_binprm *bprm)
1260{
1261    umode_t mode;
1262    struct inode * inode = bprm->file->f_path.dentry->d_inode;
1263    int retval;
1264
1265    mode = inode->i_mode;
1266    if (bprm->file->f_op == NULL)
1267        return -EACCES;
1268
1269    /* clear any previous set[ug]id data from a previous binary */
1270    bprm->cred->euid = current_euid();
1271    bprm->cred->egid = current_egid();
1272
1273    if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1274        /* Set-uid? */
1275        if (mode & S_ISUID) {
1276            bprm->per_clear |= PER_CLEAR_ON_SETID;
1277            bprm->cred->euid = inode->i_uid;
1278        }
1279
1280        /* Set-gid? */
1281        /*
1282         * If setgid is set but no group execute bit then this
1283         * is a candidate for mandatory locking, not a setgid
1284         * executable.
1285         */
1286        if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1287            bprm->per_clear |= PER_CLEAR_ON_SETID;
1288            bprm->cred->egid = inode->i_gid;
1289        }
1290    }
1291
1292    /* fill in binprm security blob */
1293    retval = security_bprm_set_creds(bprm);
1294    if (retval)
1295        return retval;
1296    bprm->cred_prepared = 1;
1297
1298    memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1299    return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1300}
1301
1302EXPORT_SYMBOL(prepare_binprm);
1303
1304/*
1305 * Arguments are '\0' separated strings found at the location bprm->p
1306 * points to; chop off the first by relocating brpm->p to right after
1307 * the first '\0' encountered.
1308 */
1309int remove_arg_zero(struct linux_binprm *bprm)
1310{
1311    int ret = 0;
1312    unsigned long offset;
1313    char *kaddr;
1314    struct page *page;
1315
1316    if (!bprm->argc)
1317        return 0;
1318
1319    do {
1320        offset = bprm->p & ~PAGE_MASK;
1321        page = get_arg_page(bprm, bprm->p, 0);
1322        if (!page) {
1323            ret = -EFAULT;
1324            goto out;
1325        }
1326        kaddr = kmap_atomic(page, KM_USER0);
1327
1328        for (; offset < PAGE_SIZE && kaddr[offset];
1329                offset++, bprm->p++)
1330            ;
1331
1332        kunmap_atomic(kaddr, KM_USER0);
1333        put_arg_page(page);
1334
1335        if (offset == PAGE_SIZE)
1336            free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1337    } while (offset == PAGE_SIZE);
1338
1339    bprm->p++;
1340    bprm->argc--;
1341    ret = 0;
1342
1343out:
1344    return ret;
1345}
1346EXPORT_SYMBOL(remove_arg_zero);
1347
1348/*
1349 * cycle the list of binary formats handler, until one recognizes the image
1350 */
1351int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1352{
1353    unsigned int depth = bprm->recursion_depth;
1354    int try,retval;
1355    struct linux_binfmt *fmt;
1356
1357    retval = security_bprm_check(bprm);
1358    if (retval)
1359        return retval;
1360
1361    retval = audit_bprm(bprm);
1362    if (retval)
1363        return retval;
1364
1365    retval = -ENOENT;
1366    for (try=0; try<2; try++) {
1367        read_lock(&binfmt_lock);
1368        list_for_each_entry(fmt, &formats, lh) {
1369            int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1370            if (!fn)
1371                continue;
1372            if (!try_module_get(fmt->module))
1373                continue;
1374            read_unlock(&binfmt_lock);
1375            retval = fn(bprm, regs);
1376            /*
1377             * Restore the depth counter to its starting value
1378             * in this call, so we don't have to rely on every
1379             * load_binary function to restore it on return.
1380             */
1381            bprm->recursion_depth = depth;
1382            if (retval >= 0) {
1383                if (depth == 0)
1384                    tracehook_report_exec(fmt, bprm, regs);
1385                put_binfmt(fmt);
1386                allow_write_access(bprm->file);
1387                if (bprm->file)
1388                    fput(bprm->file);
1389                bprm->file = NULL;
1390                current->did_exec = 1;
1391                proc_exec_connector(current);
1392                return retval;
1393            }
1394            read_lock(&binfmt_lock);
1395            put_binfmt(fmt);
1396            if (retval != -ENOEXEC || bprm->mm == NULL)
1397                break;
1398            if (!bprm->file) {
1399                read_unlock(&binfmt_lock);
1400                return retval;
1401            }
1402        }
1403        read_unlock(&binfmt_lock);
1404        if (retval != -ENOEXEC || bprm->mm == NULL) {
1405            break;
1406#ifdef CONFIG_MODULES
1407        } else {
1408#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1409            if (printable(bprm->buf[0]) &&
1410                printable(bprm->buf[1]) &&
1411                printable(bprm->buf[2]) &&
1412                printable(bprm->buf[3]))
1413                break; /* -ENOEXEC */
1414            request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1415#endif
1416        }
1417    }
1418    return retval;
1419}
1420
1421EXPORT_SYMBOL(search_binary_handler);
1422
1423/*
1424 * sys_execve() executes a new program.
1425 */
1426static int do_execve_common(const char *filename,
1427                struct user_arg_ptr argv,
1428                struct user_arg_ptr envp,
1429                struct pt_regs *regs)
1430{
1431    struct linux_binprm *bprm;
1432    struct file *file;
1433    struct files_struct *displaced;
1434    bool clear_in_exec;
1435    int retval;
1436
1437    retval = unshare_files(&displaced);
1438    if (retval)
1439        goto out_ret;
1440
1441    retval = -ENOMEM;
1442    bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1443    if (!bprm)
1444        goto out_files;
1445
1446    retval = prepare_bprm_creds(bprm);
1447    if (retval)
1448        goto out_free;
1449
1450    retval = check_unsafe_exec(bprm);
1451    if (retval < 0)
1452        goto out_free;
1453    clear_in_exec = retval;
1454    current->in_execve = 1;
1455
1456    file = open_exec(filename);
1457    retval = PTR_ERR(file);
1458    if (IS_ERR(file))
1459        goto out_unmark;
1460
1461    sched_exec();
1462
1463    bprm->file = file;
1464    bprm->filename = filename;
1465    bprm->interp = filename;
1466
1467    retval = bprm_mm_init(bprm);
1468    if (retval)
1469        goto out_file;
1470
1471    bprm->argc = count(argv, MAX_ARG_STRINGS);
1472    if ((retval = bprm->argc) < 0)
1473        goto out;
1474
1475    bprm->envc = count(envp, MAX_ARG_STRINGS);
1476    if ((retval = bprm->envc) < 0)
1477        goto out;
1478
1479    retval = prepare_binprm(bprm);
1480    if (retval < 0)
1481        goto out;
1482
1483    retval = copy_strings_kernel(1, &bprm->filename, bprm);
1484    if (retval < 0)
1485        goto out;
1486
1487    bprm->exec = bprm->p;
1488    retval = copy_strings(bprm->envc, envp, bprm);
1489    if (retval < 0)
1490        goto out;
1491
1492    retval = copy_strings(bprm->argc, argv, bprm);
1493    if (retval < 0)
1494        goto out;
1495
1496    retval = search_binary_handler(bprm,regs);
1497    if (retval < 0)
1498        goto out;
1499
1500    /* execve succeeded */
1501    current->fs->in_exec = 0;
1502    current->in_execve = 0;
1503    acct_update_integrals(current);
1504    free_bprm(bprm);
1505    if (displaced)
1506        put_files_struct(displaced);
1507    return retval;
1508
1509out:
1510    if (bprm->mm) {
1511        acct_arg_size(bprm, 0);
1512        mmput(bprm->mm);
1513    }
1514
1515out_file:
1516    if (bprm->file) {
1517        allow_write_access(bprm->file);
1518        fput(bprm->file);
1519    }
1520
1521out_unmark:
1522    if (clear_in_exec)
1523        current->fs->in_exec = 0;
1524    current->in_execve = 0;
1525
1526out_free:
1527    free_bprm(bprm);
1528
1529out_files:
1530    if (displaced)
1531        reset_files_struct(displaced);
1532out_ret:
1533    return retval;
1534}
1535
1536int do_execve(const char *filename,
1537    const char __user *const __user *__argv,
1538    const char __user *const __user *__envp,
1539    struct pt_regs *regs)
1540{
1541    struct user_arg_ptr argv = { .ptr.native = __argv };
1542    struct user_arg_ptr envp = { .ptr.native = __envp };
1543    return do_execve_common(filename, argv, envp, regs);
1544}
1545
1546#ifdef CONFIG_COMPAT
1547int compat_do_execve(char *filename,
1548    compat_uptr_t __user *__argv,
1549    compat_uptr_t __user *__envp,
1550    struct pt_regs *regs)
1551{
1552    struct user_arg_ptr argv = {
1553        .is_compat = true,
1554        .ptr.compat = __argv,
1555    };
1556    struct user_arg_ptr envp = {
1557        .is_compat = true,
1558        .ptr.compat = __envp,
1559    };
1560    return do_execve_common(filename, argv, envp, regs);
1561}
1562#endif
1563
1564void set_binfmt(struct linux_binfmt *new)
1565{
1566    struct mm_struct *mm = current->mm;
1567
1568    if (mm->binfmt)
1569        module_put(mm->binfmt->module);
1570
1571    mm->binfmt = new;
1572    if (new)
1573        __module_get(new->module);
1574}
1575
1576EXPORT_SYMBOL(set_binfmt);
1577
1578static int expand_corename(struct core_name *cn)
1579{
1580    char *old_corename = cn->corename;
1581
1582    cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1583    cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1584
1585    if (!cn->corename) {
1586        kfree(old_corename);
1587        return -ENOMEM;
1588    }
1589
1590    return 0;
1591}
1592
1593static int cn_printf(struct core_name *cn, const char *fmt, ...)
1594{
1595    char *cur;
1596    int need;
1597    int ret;
1598    va_list arg;
1599
1600    va_start(arg, fmt);
1601    need = vsnprintf(NULL, 0, fmt, arg);
1602    va_end(arg);
1603
1604    if (likely(need < cn->size - cn->used - 1))
1605        goto out_printf;
1606
1607    ret = expand_corename(cn);
1608    if (ret)
1609        goto expand_fail;
1610
1611out_printf:
1612    cur = cn->corename + cn->used;
1613    va_start(arg, fmt);
1614    vsnprintf(cur, need + 1, fmt, arg);
1615    va_end(arg);
1616    cn->used += need;
1617    return 0;
1618
1619expand_fail:
1620    return ret;
1621}
1622
1623static int cn_print_exe_file(struct core_name *cn)
1624{
1625    struct file *exe_file;
1626    char *pathbuf, *path, *p;
1627    int ret;
1628
1629    exe_file = get_mm_exe_file(current->mm);
1630    if (!exe_file)
1631        return cn_printf(cn, "(unknown)");
1632
1633    pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1634    if (!pathbuf) {
1635        ret = -ENOMEM;
1636        goto put_exe_file;
1637    }
1638
1639    path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1640    if (IS_ERR(path)) {
1641        ret = PTR_ERR(path);
1642        goto free_buf;
1643    }
1644
1645    for (p = path; *p; p++)
1646        if (*p == '/')
1647            *p = '!';
1648
1649    ret = cn_printf(cn, "%s", path);
1650
1651free_buf:
1652    kfree(pathbuf);
1653put_exe_file:
1654    fput(exe_file);
1655    return ret;
1656}
1657
1658/* format_corename will inspect the pattern parameter, and output a
1659 * name into corename, which must have space for at least
1660 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1661 */
1662static int format_corename(struct core_name *cn, long signr)
1663{
1664    const struct cred *cred = current_cred();
1665    const char *pat_ptr = core_pattern;
1666    int ispipe = (*pat_ptr == '|');
1667    int pid_in_pattern = 0;
1668    int err = 0;
1669
1670    cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1671    cn->corename = kmalloc(cn->size, GFP_KERNEL);
1672    cn->used = 0;
1673
1674    if (!cn->corename)
1675        return -ENOMEM;
1676
1677    /* Repeat as long as we have more pattern to process and more output
1678       space */
1679    while (*pat_ptr) {
1680        if (*pat_ptr != '%') {
1681            if (*pat_ptr == 0)
1682                goto out;
1683            err = cn_printf(cn, "%c", *pat_ptr++);
1684        } else {
1685            switch (*++pat_ptr) {
1686            /* single % at the end, drop that */
1687            case 0:
1688                goto out;
1689            /* Double percent, output one percent */
1690            case '%':
1691                err = cn_printf(cn, "%c", '%');
1692                break;
1693            /* pid */
1694            case 'p':
1695                pid_in_pattern = 1;
1696                err = cn_printf(cn, "%d",
1697                          task_tgid_vnr(current));
1698                break;
1699            /* uid */
1700            case 'u':
1701                err = cn_printf(cn, "%d", cred->uid);
1702                break;
1703            /* gid */
1704            case 'g':
1705                err = cn_printf(cn, "%d", cred->gid);
1706                break;
1707            /* signal that caused the coredump */
1708            case 's':
1709                err = cn_printf(cn, "%ld", signr);
1710                break;
1711            /* UNIX time of coredump */
1712            case 't': {
1713                struct timeval tv;
1714                do_gettimeofday(&tv);
1715                err = cn_printf(cn, "%lu", tv.tv_sec);
1716                break;
1717            }
1718            /* hostname */
1719            case 'h':
1720                down_read(&uts_sem);
1721                err = cn_printf(cn, "%s",
1722                          utsname()->nodename);
1723                up_read(&uts_sem);
1724                break;
1725            /* executable */
1726            case 'e':
1727                err = cn_printf(cn, "%s", current->comm);
1728                break;
1729            case 'E':
1730                err = cn_print_exe_file(cn);
1731                break;
1732            /* core limit size */
1733            case 'c':
1734                err = cn_printf(cn, "%lu",
1735                          rlimit(RLIMIT_CORE));
1736                break;
1737            default:
1738                break;
1739            }
1740            ++pat_ptr;
1741        }
1742
1743        if (err)
1744            return err;
1745    }
1746
1747    /* Backward compatibility with core_uses_pid:
1748     *
1749     * If core_pattern does not include a %p (as is the default)
1750     * and core_uses_pid is set, then .%pid will be appended to
1751     * the filename. Do not do this for piped commands. */
1752    if (!ispipe && !pid_in_pattern && core_uses_pid) {
1753        err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1754        if (err)
1755            return err;
1756    }
1757out:
1758    return ispipe;
1759}
1760
1761static int zap_process(struct task_struct *start, int exit_code)
1762{
1763    struct task_struct *t;
1764    int nr = 0;
1765
1766    start->signal->flags = SIGNAL_GROUP_EXIT;
1767    start->signal->group_exit_code = exit_code;
1768    start->signal->group_stop_count = 0;
1769
1770    t = start;
1771    do {
1772        task_clear_group_stop_pending(t);
1773        if (t != current && t->mm) {
1774            sigaddset(&t->pending.signal, SIGKILL);
1775            signal_wake_up(t, 1);
1776            nr++;
1777        }
1778    } while_each_thread(start, t);
1779
1780    return nr;
1781}
1782
1783static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1784                struct core_state *core_state, int exit_code)
1785{
1786    struct task_struct *g, *p;
1787    unsigned long flags;
1788    int nr = -EAGAIN;
1789
1790    spin_lock_irq(&tsk->sighand->siglock);
1791    if (!signal_group_exit(tsk->signal)) {
1792        mm->core_state = core_state;
1793        nr = zap_process(tsk, exit_code);
1794    }
1795    spin_unlock_irq(&tsk->sighand->siglock);
1796    if (unlikely(nr < 0))
1797        return nr;
1798
1799    if (atomic_read(&mm->mm_users) == nr + 1)
1800        goto done;
1801    /*
1802     * We should find and kill all tasks which use this mm, and we should
1803     * count them correctly into ->nr_threads. We don't take tasklist
1804     * lock, but this is safe wrt:
1805     *
1806     * fork:
1807     * None of sub-threads can fork after zap_process(leader). All
1808     * processes which were created before this point should be
1809     * visible to zap_threads() because copy_process() adds the new
1810     * process to the tail of init_task.tasks list, and lock/unlock
1811     * of ->siglock provides a memory barrier.
1812     *
1813     * do_exit:
1814     * The caller holds mm->mmap_sem. This means that the task which
1815     * uses this mm can't pass exit_mm(), so it can't exit or clear
1816     * its ->mm.
1817     *
1818     * de_thread:
1819     * It does list_replace_rcu(&leader->tasks, &current->tasks),
1820     * we must see either old or new leader, this does not matter.
1821     * However, it can change p->sighand, so lock_task_sighand(p)
1822     * must be used. Since p->mm != NULL and we hold ->mmap_sem
1823     * it can't fail.
1824     *
1825     * Note also that "g" can be the old leader with ->mm == NULL
1826     * and already unhashed and thus removed from ->thread_group.
1827     * This is OK, __unhash_process()->list_del_rcu() does not
1828     * clear the ->next pointer, we will find the new leader via
1829     * next_thread().
1830     */
1831    rcu_read_lock();
1832    for_each_process(g) {
1833        if (g == tsk->group_leader)
1834            continue;
1835        if (g->flags & PF_KTHREAD)
1836            continue;
1837        p = g;
1838        do {
1839            if (p->mm) {
1840                if (unlikely(p->mm == mm)) {
1841                    lock_task_sighand(p, &flags);
1842                    nr += zap_process(p, exit_code);
1843                    unlock_task_sighand(p, &flags);
1844                }
1845                break;
1846            }
1847        } while_each_thread(g, p);
1848    }
1849    rcu_read_unlock();
1850done:
1851    atomic_set(&core_state->nr_threads, nr);
1852    return nr;
1853}
1854
1855static int coredump_wait(int exit_code, struct core_state *core_state)
1856{
1857    struct task_struct *tsk = current;
1858    struct mm_struct *mm = tsk->mm;
1859    struct completion *vfork_done;
1860    int core_waiters = -EBUSY;
1861
1862    init_completion(&core_state->startup);
1863    core_state->dumper.task = tsk;
1864    core_state->dumper.next = NULL;
1865
1866    down_write(&mm->mmap_sem);
1867    if (!mm->core_state)
1868        core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1869    up_write(&mm->mmap_sem);
1870
1871    if (unlikely(core_waiters < 0))
1872        goto fail;
1873
1874    /*
1875     * Make sure nobody is waiting for us to release the VM,
1876     * otherwise we can deadlock when we wait on each other
1877     */
1878    vfork_done = tsk->vfork_done;
1879    if (vfork_done) {
1880        tsk->vfork_done = NULL;
1881        complete(vfork_done);
1882    }
1883
1884    if (core_waiters)
1885        wait_for_completion(&core_state->startup);
1886fail:
1887    return core_waiters;
1888}
1889
1890static void coredump_finish(struct mm_struct *mm)
1891{
1892    struct core_thread *curr, *next;
1893    struct task_struct *task;
1894
1895    next = mm->core_state->dumper.next;
1896    while ((curr = next) != NULL) {
1897        next = curr->next;
1898        task = curr->task;
1899        /*
1900         * see exit_mm(), curr->task must not see
1901         * ->task == NULL before we read ->next.
1902         */
1903        smp_mb();
1904        curr->task = NULL;
1905        wake_up_process(task);
1906    }
1907
1908    mm->core_state = NULL;
1909}
1910
1911/*
1912 * set_dumpable converts traditional three-value dumpable to two flags and
1913 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1914 * these bits are not changed atomically. So get_dumpable can observe the
1915 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1916 * return either old dumpable or new one by paying attention to the order of
1917 * modifying the bits.
1918 *
1919 * dumpable | mm->flags (binary)
1920 * old new | initial interim final
1921 * ---------+-----------------------
1922 * 0 1 | 00 01 01
1923 * 0 2 | 00 10(*) 11
1924 * 1 0 | 01 00 00
1925 * 1 2 | 01 11 11
1926 * 2 0 | 11 10(*) 00
1927 * 2 1 | 11 11 01
1928 *
1929 * (*) get_dumpable regards interim value of 10 as 11.
1930 */
1931void set_dumpable(struct mm_struct *mm, int value)
1932{
1933    switch (value) {
1934    case 0:
1935        clear_bit(MMF_DUMPABLE, &mm->flags);
1936        smp_wmb();
1937        clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1938        break;
1939    case 1:
1940        set_bit(MMF_DUMPABLE, &mm->flags);
1941        smp_wmb();
1942        clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1943        break;
1944    case 2:
1945        set_bit(MMF_DUMP_SECURELY, &mm->flags);
1946        smp_wmb();
1947        set_bit(MMF_DUMPABLE, &mm->flags);
1948        break;
1949    }
1950}
1951
1952static int __get_dumpable(unsigned long mm_flags)
1953{
1954    int ret;
1955
1956    ret = mm_flags & MMF_DUMPABLE_MASK;
1957    return (ret >= 2) ? 2 : ret;
1958}
1959
1960int get_dumpable(struct mm_struct *mm)
1961{
1962    return __get_dumpable(mm->flags);
1963}
1964
1965static void wait_for_dump_helpers(struct file *file)
1966{
1967    struct pipe_inode_info *pipe;
1968
1969    pipe = file->f_path.dentry->d_inode->i_pipe;
1970
1971    pipe_lock(pipe);
1972    pipe->readers++;
1973    pipe->writers--;
1974
1975    while ((pipe->readers > 1) && (!signal_pending(current))) {
1976        wake_up_interruptible_sync(&pipe->wait);
1977        kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1978        pipe_wait(pipe);
1979    }
1980
1981    pipe->readers--;
1982    pipe->writers++;
1983    pipe_unlock(pipe);
1984
1985}
1986
1987
1988/*
1989 * umh_pipe_setup
1990 * helper function to customize the process used
1991 * to collect the core in userspace. Specifically
1992 * it sets up a pipe and installs it as fd 0 (stdin)
1993 * for the process. Returns 0 on success, or
1994 * PTR_ERR on failure.
1995 * Note that it also sets the core limit to 1. This
1996 * is a special value that we use to trap recursive
1997 * core dumps
1998 */
1999static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2000{
2001    struct file *rp, *wp;
2002    struct fdtable *fdt;
2003    struct coredump_params *cp = (struct coredump_params *)info->data;
2004    struct files_struct *cf = current->files;
2005
2006    wp = create_write_pipe(0);
2007    if (IS_ERR(wp))
2008        return PTR_ERR(wp);
2009
2010    rp = create_read_pipe(wp, 0);
2011    if (IS_ERR(rp)) {
2012        free_write_pipe(wp);
2013        return PTR_ERR(rp);
2014    }
2015
2016    cp->file = wp;
2017
2018    sys_close(0);
2019    fd_install(0, rp);
2020    spin_lock(&cf->file_lock);
2021    fdt = files_fdtable(cf);
2022    FD_SET(0, fdt->open_fds);
2023    FD_CLR(0, fdt->close_on_exec);
2024    spin_unlock(&cf->file_lock);
2025
2026    /* and disallow core files too */
2027    current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2028
2029    return 0;
2030}
2031
2032void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2033{
2034    struct core_state core_state;
2035    struct core_name cn;
2036    struct mm_struct *mm = current->mm;
2037    struct linux_binfmt * binfmt;
2038    const struct cred *old_cred;
2039    struct cred *cred;
2040    int retval = 0;
2041    int flag = 0;
2042    int ispipe;
2043    static atomic_t core_dump_count = ATOMIC_INIT(0);
2044    struct coredump_params cprm = {
2045        .signr = signr,
2046        .regs = regs,
2047        .limit = rlimit(RLIMIT_CORE),
2048        /*
2049         * We must use the same mm->flags while dumping core to avoid
2050         * inconsistency of bit flags, since this flag is not protected
2051         * by any locks.
2052         */
2053        .mm_flags = mm->flags,
2054    };
2055
2056    audit_core_dumps(signr);
2057
2058    binfmt = mm->binfmt;
2059    if (!binfmt || !binfmt->core_dump)
2060        goto fail;
2061    if (!__get_dumpable(cprm.mm_flags))
2062        goto fail;
2063
2064    cred = prepare_creds();
2065    if (!cred)
2066        goto fail;
2067    /*
2068     * We cannot trust fsuid as being the "true" uid of the
2069     * process nor do we know its entire history. We only know it
2070     * was tainted so we dump it as root in mode 2.
2071     */
2072    if (__get_dumpable(cprm.mm_flags) == 2) {
2073        /* Setuid core dump mode */
2074        flag = O_EXCL; /* Stop rewrite attacks */
2075        cred->fsuid = 0; /* Dump root private */
2076    }
2077
2078    retval = coredump_wait(exit_code, &core_state);
2079    if (retval < 0)
2080        goto fail_creds;
2081
2082    old_cred = override_creds(cred);
2083
2084    /*
2085     * Clear any false indication of pending signals that might
2086     * be seen by the filesystem code called to write the core file.
2087     */
2088    clear_thread_flag(TIF_SIGPENDING);
2089
2090    ispipe = format_corename(&cn, signr);
2091
2092    if (ispipe == -ENOMEM) {
2093        printk(KERN_WARNING "format_corename failed\n");
2094        printk(KERN_WARNING "Aborting core\n");
2095        goto fail_corename;
2096    }
2097
2098     if (ispipe) {
2099        int dump_count;
2100        char **helper_argv;
2101
2102        if (cprm.limit == 1) {
2103            /*
2104             * Normally core limits are irrelevant to pipes, since
2105             * we're not writing to the file system, but we use
2106             * cprm.limit of 1 here as a speacial value. Any
2107             * non-1 limit gets set to RLIM_INFINITY below, but
2108             * a limit of 0 skips the dump. This is a consistent
2109             * way to catch recursive crashes. We can still crash
2110             * if the core_pattern binary sets RLIM_CORE = !1
2111             * but it runs as root, and can do lots of stupid things
2112             * Note that we use task_tgid_vnr here to grab the pid
2113             * of the process group leader. That way we get the
2114             * right pid if a thread in a multi-threaded
2115             * core_pattern process dies.
2116             */
2117            printk(KERN_WARNING
2118                "Process %d(%s) has RLIMIT_CORE set to 1\n",
2119                task_tgid_vnr(current), current->comm);
2120            printk(KERN_WARNING "Aborting core\n");
2121            goto fail_unlock;
2122        }
2123        cprm.limit = RLIM_INFINITY;
2124
2125        dump_count = atomic_inc_return(&core_dump_count);
2126        if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2127            printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2128                   task_tgid_vnr(current), current->comm);
2129            printk(KERN_WARNING "Skipping core dump\n");
2130            goto fail_dropcount;
2131        }
2132
2133        helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2134        if (!helper_argv) {
2135            printk(KERN_WARNING "%s failed to allocate memory\n",
2136                   __func__);
2137            goto fail_dropcount;
2138        }
2139
2140        retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2141                    NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2142                    NULL, &cprm);
2143        argv_free(helper_argv);
2144        if (retval) {
2145             printk(KERN_INFO "Core dump to %s pipe failed\n",
2146                   cn.corename);
2147            goto close_fail;
2148         }
2149    } else {
2150        struct inode *inode;
2151
2152        if (cprm.limit < binfmt->min_coredump)
2153            goto fail_unlock;
2154
2155        cprm.file = filp_open(cn.corename,
2156                 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2157                 0600);
2158        if (IS_ERR(cprm.file))
2159            goto fail_unlock;
2160
2161        inode = cprm.file->f_path.dentry->d_inode;
2162        if (inode->i_nlink > 1)
2163            goto close_fail;
2164        if (d_unhashed(cprm.file->f_path.dentry))
2165            goto close_fail;
2166        /*
2167         * AK: actually i see no reason to not allow this for named
2168         * pipes etc, but keep the previous behaviour for now.
2169         */
2170        if (!S_ISREG(inode->i_mode))
2171            goto close_fail;
2172        /*
2173         * Dont allow local users get cute and trick others to coredump
2174         * into their pre-created files.
2175         */
2176        if (inode->i_uid != current_fsuid())
2177            goto close_fail;
2178        if (!cprm.file->f_op || !cprm.file->f_op->write)
2179            goto close_fail;
2180        if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2181            goto close_fail;
2182    }
2183
2184    retval = binfmt->core_dump(&cprm);
2185    if (retval)
2186        current->signal->group_exit_code |= 0x80;
2187
2188    if (ispipe && core_pipe_limit)
2189        wait_for_dump_helpers(cprm.file);
2190close_fail:
2191    if (cprm.file)
2192        filp_close(cprm.file, NULL);
2193fail_dropcount:
2194    if (ispipe)
2195        atomic_dec(&core_dump_count);
2196fail_unlock:
2197    kfree(cn.corename);
2198fail_corename:
2199    coredump_finish(mm);
2200    revert_creds(old_cred);
2201fail_creds:
2202    put_cred(cred);
2203fail:
2204    return;
2205}
2206
2207/*
2208 * Core dumping helper functions. These are the only things you should
2209 * do on a core-file: use only these functions to write out all the
2210 * necessary info.
2211 */
2212int dump_write(struct file *file, const void *addr, int nr)
2213{
2214    return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2215}
2216EXPORT_SYMBOL(dump_write);
2217
2218int dump_seek(struct file *file, loff_t off)
2219{
2220    int ret = 1;
2221
2222    if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2223        if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2224            return 0;
2225    } else {
2226        char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2227
2228        if (!buf)
2229            return 0;
2230        while (off > 0) {
2231            unsigned long n = off;
2232
2233            if (n > PAGE_SIZE)
2234                n = PAGE_SIZE;
2235            if (!dump_write(file, buf, n)) {
2236                ret = 0;
2237                break;
2238            }
2239            off -= n;
2240        }
2241        free_page((unsigned long)buf);
2242    }
2243    return ret;
2244}
2245EXPORT_SYMBOL(dump_seek);
2246

Archive Download this file



interactive