1 #include <linux/slab.h>
2 #include <linux/file.h>
3 #include <linux/fdtable.h>
5 #include <linux/stat.h>
6 #include <linux/fcntl.h>
7 #include <linux/swap.h>
8 #include <linux/string.h>
9 #include <linux/init.h>
10 #include <linux/pagemap.h>
11 #include <linux/perf_event.h>
12 #include <linux/highmem.h>
13 #include <linux/spinlock.h>
14 #include <linux/key.h>
15 #include <linux/personality.h>
16 #include <linux/binfmts.h>
17 #include <linux/coredump.h>
18 #include <linux/utsname.h>
19 #include <linux/pid_namespace.h>
20 #include <linux/module.h>
21 #include <linux/namei.h>
22 #include <linux/mount.h>
23 #include <linux/security.h>
24 #include <linux/syscalls.h>
25 #include <linux/tsacct_kern.h>
26 #include <linux/cn_proc.h>
27 #include <linux/audit.h>
28 #include <linux/tracehook.h>
29 #include <linux/kmod.h>
30 #include <linux/fsnotify.h>
31 #include <linux/fs_struct.h>
32 #include <linux/pipe_fs_i.h>
33 #include <linux/oom.h>
34 #include <linux/compat.h>
35 #include <linux/timekeeping.h>
37 #include <asm/uaccess.h>
38 #include <asm/mmu_context.h>
42 #include <trace/events/task.h>
45 #include <trace/events/sched.h>
48 unsigned int core_pipe_limit
;
49 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
50 static int core_name_size
= CORENAME_MAX_SIZE
;
57 /* The maximal length of core_pattern is also specified in sysctl.c */
59 static int expand_corename(struct core_name
*cn
, int size
)
61 char *corename
= krealloc(cn
->corename
, size
, GFP_KERNEL
);
66 if (size
> core_name_size
) /* racy but harmless */
67 core_name_size
= size
;
69 cn
->size
= ksize(corename
);
70 cn
->corename
= corename
;
74 static __printf(2, 0) int cn_vprintf(struct core_name
*cn
, const char *fmt
,
81 free
= cn
->size
- cn
->used
;
83 va_copy(arg_copy
, arg
);
84 need
= vsnprintf(cn
->corename
+ cn
->used
, free
, fmt
, arg_copy
);
92 if (!expand_corename(cn
, cn
->size
+ need
- free
+ 1))
98 static __printf(2, 3) int cn_printf(struct core_name
*cn
, const char *fmt
, ...)
104 ret
= cn_vprintf(cn
, fmt
, arg
);
110 static __printf(2, 3)
111 int cn_esc_printf(struct core_name
*cn
, const char *fmt
, ...)
118 ret
= cn_vprintf(cn
, fmt
, arg
);
121 for (; cur
< cn
->used
; ++cur
) {
122 if (cn
->corename
[cur
] == '/')
123 cn
->corename
[cur
] = '!';
128 static int cn_print_exe_file(struct core_name
*cn
)
130 struct file
*exe_file
;
131 char *pathbuf
, *path
;
134 exe_file
= get_mm_exe_file(current
->mm
);
136 return cn_esc_printf(cn
, "%s (path unknown)", current
->comm
);
138 pathbuf
= kmalloc(PATH_MAX
, GFP_TEMPORARY
);
144 path
= file_path(exe_file
, pathbuf
, PATH_MAX
);
150 ret
= cn_esc_printf(cn
, "%s", path
);
159 /* format_corename will inspect the pattern parameter, and output a
160 * name into corename, which must have space for at least
161 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
163 static int format_corename(struct core_name
*cn
, struct coredump_params
*cprm
)
165 const struct cred
*cred
= current_cred();
166 const char *pat_ptr
= core_pattern
;
167 int ispipe
= (*pat_ptr
== '|');
168 int pid_in_pattern
= 0;
173 if (expand_corename(cn
, core_name_size
))
175 cn
->corename
[0] = '\0';
180 /* Repeat as long as we have more pattern to process and more output
183 if (*pat_ptr
!= '%') {
184 err
= cn_printf(cn
, "%c", *pat_ptr
++);
186 switch (*++pat_ptr
) {
187 /* single % at the end, drop that */
190 /* Double percent, output one percent */
192 err
= cn_printf(cn
, "%c", '%');
197 err
= cn_printf(cn
, "%d",
198 task_tgid_vnr(current
));
202 err
= cn_printf(cn
, "%d",
203 task_tgid_nr(current
));
206 err
= cn_printf(cn
, "%d",
207 task_pid_vnr(current
));
210 err
= cn_printf(cn
, "%d",
211 task_pid_nr(current
));
215 err
= cn_printf(cn
, "%u",
216 from_kuid(&init_user_ns
,
221 err
= cn_printf(cn
, "%u",
222 from_kgid(&init_user_ns
,
226 err
= cn_printf(cn
, "%d",
227 __get_dumpable(cprm
->mm_flags
));
229 /* signal that caused the coredump */
231 err
= cn_printf(cn
, "%d",
232 cprm
->siginfo
->si_signo
);
234 /* UNIX time of coredump */
238 time
= ktime_get_real_seconds();
239 err
= cn_printf(cn
, "%lld", time
);
245 err
= cn_esc_printf(cn
, "%s",
246 utsname()->nodename
);
251 err
= cn_esc_printf(cn
, "%s", current
->comm
);
254 err
= cn_print_exe_file(cn
);
256 /* core limit size */
258 err
= cn_printf(cn
, "%lu",
259 rlimit(RLIMIT_CORE
));
272 /* Backward compatibility with core_uses_pid:
274 * If core_pattern does not include a %p (as is the default)
275 * and core_uses_pid is set, then .%pid will be appended to
276 * the filename. Do not do this for piped commands. */
277 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
278 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
285 static int zap_process(struct task_struct
*start
, int exit_code
, int flags
)
287 struct task_struct
*t
;
290 /* ignore all signals except SIGKILL, see prepare_signal() */
291 start
->signal
->flags
= SIGNAL_GROUP_COREDUMP
| flags
;
292 start
->signal
->group_exit_code
= exit_code
;
293 start
->signal
->group_stop_count
= 0;
295 for_each_thread(start
, t
) {
296 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
297 if (t
!= current
&& t
->mm
) {
298 sigaddset(&t
->pending
.signal
, SIGKILL
);
299 signal_wake_up(t
, 1);
307 static int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
308 struct core_state
*core_state
, int exit_code
)
310 struct task_struct
*g
, *p
;
314 spin_lock_irq(&tsk
->sighand
->siglock
);
315 if (!signal_group_exit(tsk
->signal
)) {
316 mm
->core_state
= core_state
;
317 tsk
->signal
->group_exit_task
= tsk
;
318 nr
= zap_process(tsk
, exit_code
, 0);
319 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
321 spin_unlock_irq(&tsk
->sighand
->siglock
);
322 if (unlikely(nr
< 0))
325 tsk
->flags
|= PF_DUMPCORE
;
326 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
329 * We should find and kill all tasks which use this mm, and we should
330 * count them correctly into ->nr_threads. We don't take tasklist
331 * lock, but this is safe wrt:
334 * None of sub-threads can fork after zap_process(leader). All
335 * processes which were created before this point should be
336 * visible to zap_threads() because copy_process() adds the new
337 * process to the tail of init_task.tasks list, and lock/unlock
338 * of ->siglock provides a memory barrier.
341 * The caller holds mm->mmap_sem. This means that the task which
342 * uses this mm can't pass exit_mm(), so it can't exit or clear
346 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
347 * we must see either old or new leader, this does not matter.
348 * However, it can change p->sighand, so lock_task_sighand(p)
349 * must be used. Since p->mm != NULL and we hold ->mmap_sem
352 * Note also that "g" can be the old leader with ->mm == NULL
353 * and already unhashed and thus removed from ->thread_group.
354 * This is OK, __unhash_process()->list_del_rcu() does not
355 * clear the ->next pointer, we will find the new leader via
359 for_each_process(g
) {
360 if (g
== tsk
->group_leader
)
362 if (g
->flags
& PF_KTHREAD
)
365 for_each_thread(g
, p
) {
366 if (unlikely(!p
->mm
))
368 if (unlikely(p
->mm
== mm
)) {
369 lock_task_sighand(p
, &flags
);
370 nr
+= zap_process(p
, exit_code
,
372 unlock_task_sighand(p
, &flags
);
379 atomic_set(&core_state
->nr_threads
, nr
);
383 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
385 struct task_struct
*tsk
= current
;
386 struct mm_struct
*mm
= tsk
->mm
;
387 int core_waiters
= -EBUSY
;
389 init_completion(&core_state
->startup
);
390 core_state
->dumper
.task
= tsk
;
391 core_state
->dumper
.next
= NULL
;
393 down_write(&mm
->mmap_sem
);
395 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
396 up_write(&mm
->mmap_sem
);
398 if (core_waiters
> 0) {
399 struct core_thread
*ptr
;
401 wait_for_completion(&core_state
->startup
);
403 * Wait for all the threads to become inactive, so that
404 * all the thread context (extended register state, like
405 * fpu etc) gets copied to the memory.
407 ptr
= core_state
->dumper
.next
;
408 while (ptr
!= NULL
) {
409 wait_task_inactive(ptr
->task
, 0);
417 static void coredump_finish(struct mm_struct
*mm
, bool core_dumped
)
419 struct core_thread
*curr
, *next
;
420 struct task_struct
*task
;
422 spin_lock_irq(¤t
->sighand
->siglock
);
423 if (core_dumped
&& !__fatal_signal_pending(current
))
424 current
->signal
->group_exit_code
|= 0x80;
425 current
->signal
->group_exit_task
= NULL
;
426 current
->signal
->flags
= SIGNAL_GROUP_EXIT
;
427 spin_unlock_irq(¤t
->sighand
->siglock
);
429 next
= mm
->core_state
->dumper
.next
;
430 while ((curr
= next
) != NULL
) {
434 * see exit_mm(), curr->task must not see
435 * ->task == NULL before we read ->next.
439 wake_up_process(task
);
442 mm
->core_state
= NULL
;
445 static bool dump_interrupted(void)
448 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
449 * can do try_to_freeze() and check __fatal_signal_pending(),
450 * but then we need to teach dump_write() to restart and clear
453 return signal_pending(current
);
456 static void wait_for_dump_helpers(struct file
*file
)
458 struct pipe_inode_info
*pipe
= file
->private_data
;
463 wake_up_interruptible_sync(&pipe
->wait
);
464 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
468 * We actually want wait_event_freezable() but then we need
469 * to clear TIF_SIGPENDING and improve dump_interrupted().
471 wait_event_interruptible(pipe
->wait
, pipe
->readers
== 1);
481 * helper function to customize the process used
482 * to collect the core in userspace. Specifically
483 * it sets up a pipe and installs it as fd 0 (stdin)
484 * for the process. Returns 0 on success, or
485 * PTR_ERR on failure.
486 * Note that it also sets the core limit to 1. This
487 * is a special value that we use to trap recursive
490 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
492 struct file
*files
[2];
493 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
494 int err
= create_pipe_files(files
, 0);
500 err
= replace_fd(0, files
[0], 0);
502 /* and disallow core files too */
503 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
508 void do_coredump(const siginfo_t
*siginfo
)
510 struct core_state core_state
;
512 struct mm_struct
*mm
= current
->mm
;
513 struct linux_binfmt
* binfmt
;
514 const struct cred
*old_cred
;
518 struct files_struct
*displaced
;
519 /* require nonrelative corefile path and be extra careful */
520 bool need_suid_safe
= false;
521 bool core_dumped
= false;
522 static atomic_t core_dump_count
= ATOMIC_INIT(0);
523 struct coredump_params cprm
= {
525 .regs
= signal_pt_regs(),
526 .limit
= rlimit(RLIMIT_CORE
),
528 * We must use the same mm->flags while dumping core to avoid
529 * inconsistency of bit flags, since this flag is not protected
532 .mm_flags
= mm
->flags
,
535 audit_core_dumps(siginfo
->si_signo
);
538 if (!binfmt
|| !binfmt
->core_dump
)
540 if (!__get_dumpable(cprm
.mm_flags
))
543 cred
= prepare_creds();
547 * We cannot trust fsuid as being the "true" uid of the process
548 * nor do we know its entire history. We only know it was tainted
549 * so we dump it as root in mode 2, and only into a controlled
550 * environment (pipe handler or fully qualified path).
552 if (__get_dumpable(cprm
.mm_flags
) == SUID_DUMP_ROOT
) {
553 /* Setuid core dump mode */
554 cred
->fsuid
= GLOBAL_ROOT_UID
; /* Dump root private */
555 need_suid_safe
= true;
558 retval
= coredump_wait(siginfo
->si_signo
, &core_state
);
562 old_cred
= override_creds(cred
);
564 ispipe
= format_corename(&cn
, &cprm
);
569 struct subprocess_info
*sub_info
;
572 printk(KERN_WARNING
"format_corename failed\n");
573 printk(KERN_WARNING
"Aborting core\n");
577 if (cprm
.limit
== 1) {
578 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
580 * Normally core limits are irrelevant to pipes, since
581 * we're not writing to the file system, but we use
582 * cprm.limit of 1 here as a special value, this is a
583 * consistent way to catch recursive crashes.
584 * We can still crash if the core_pattern binary sets
585 * RLIM_CORE = !1, but it runs as root, and can do
586 * lots of stupid things.
588 * Note that we use task_tgid_vnr here to grab the pid
589 * of the process group leader. That way we get the
590 * right pid if a thread in a multi-threaded
591 * core_pattern process dies.
594 "Process %d(%s) has RLIMIT_CORE set to 1\n",
595 task_tgid_vnr(current
), current
->comm
);
596 printk(KERN_WARNING
"Aborting core\n");
599 cprm
.limit
= RLIM_INFINITY
;
601 dump_count
= atomic_inc_return(&core_dump_count
);
602 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
603 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
604 task_tgid_vnr(current
), current
->comm
);
605 printk(KERN_WARNING
"Skipping core dump\n");
609 helper_argv
= argv_split(GFP_KERNEL
, cn
.corename
, NULL
);
611 printk(KERN_WARNING
"%s failed to allocate memory\n",
617 sub_info
= call_usermodehelper_setup(helper_argv
[0],
618 helper_argv
, NULL
, GFP_KERNEL
,
619 umh_pipe_setup
, NULL
, &cprm
);
621 retval
= call_usermodehelper_exec(sub_info
,
624 argv_free(helper_argv
);
626 printk(KERN_INFO
"Core dump to |%s pipe failed\n",
633 if (cprm
.limit
< binfmt
->min_coredump
)
636 if (need_suid_safe
&& cn
.corename
[0] != '/') {
637 printk(KERN_WARNING
"Pid %d(%s) can only dump core "\
638 "to fully qualified path!\n",
639 task_tgid_vnr(current
), current
->comm
);
640 printk(KERN_WARNING
"Skipping core dump\n");
645 * Unlink the file if it exists unless this is a SUID
646 * binary - in that case, we're running around with root
647 * privs and don't want to unlink another user's coredump.
649 if (!need_suid_safe
) {
655 * If it doesn't exist, that's fine. If there's some
656 * other problem, we'll catch it at the filp_open().
658 (void) sys_unlink((const char __user
*)cn
.corename
);
663 * There is a race between unlinking and creating the
664 * file, but if that causes an EEXIST here, that's
665 * fine - another process raced with us while creating
666 * the corefile, and the other process won. To userspace,
667 * what matters is that at least one of the two processes
668 * writes its coredump successfully, not which one.
670 cprm
.file
= filp_open(cn
.corename
,
671 O_CREAT
| 2 | O_NOFOLLOW
|
672 O_LARGEFILE
| O_EXCL
,
674 if (IS_ERR(cprm
.file
))
677 inode
= file_inode(cprm
.file
);
678 if (inode
->i_nlink
> 1)
680 if (d_unhashed(cprm
.file
->f_path
.dentry
))
683 * AK: actually i see no reason to not allow this for named
684 * pipes etc, but keep the previous behaviour for now.
686 if (!S_ISREG(inode
->i_mode
))
689 * Don't dump core if the filesystem changed owner or mode
690 * of the file during file creation. This is an issue when
691 * a process dumps core while its cwd is e.g. on a vfat
694 if (!uid_eq(inode
->i_uid
, current_fsuid()))
696 if ((inode
->i_mode
& 0677) != 0600)
698 if (!(cprm
.file
->f_mode
& FMODE_CAN_WRITE
))
700 if (do_truncate(cprm
.file
->f_path
.dentry
, 0, 0, cprm
.file
))
704 /* get us an unshared descriptor table; almost always a no-op */
705 retval
= unshare_files(&displaced
);
709 put_files_struct(displaced
);
710 if (!dump_interrupted()) {
711 file_start_write(cprm
.file
);
712 core_dumped
= binfmt
->core_dump(&cprm
);
713 file_end_write(cprm
.file
);
715 if (ispipe
&& core_pipe_limit
)
716 wait_for_dump_helpers(cprm
.file
);
719 filp_close(cprm
.file
, NULL
);
722 atomic_dec(&core_dump_count
);
725 coredump_finish(mm
, core_dumped
);
726 revert_creds(old_cred
);
734 * Core dumping helper functions. These are the only things you should
735 * do on a core-file: use only these functions to write out all the
738 int dump_emit(struct coredump_params
*cprm
, const void *addr
, int nr
)
740 struct file
*file
= cprm
->file
;
741 loff_t pos
= file
->f_pos
;
743 if (cprm
->written
+ nr
> cprm
->limit
)
746 if (dump_interrupted())
748 n
= __kernel_write(file
, addr
, nr
, &pos
);
757 EXPORT_SYMBOL(dump_emit
);
759 int dump_skip(struct coredump_params
*cprm
, size_t nr
)
761 static char zeroes
[PAGE_SIZE
];
762 struct file
*file
= cprm
->file
;
763 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
764 if (cprm
->written
+ nr
> cprm
->limit
)
766 if (dump_interrupted() ||
767 file
->f_op
->llseek(file
, nr
, SEEK_CUR
) < 0)
772 while (nr
> PAGE_SIZE
) {
773 if (!dump_emit(cprm
, zeroes
, PAGE_SIZE
))
777 return dump_emit(cprm
, zeroes
, nr
);
780 EXPORT_SYMBOL(dump_skip
);
782 int dump_align(struct coredump_params
*cprm
, int align
)
784 unsigned mod
= cprm
->written
& (align
- 1);
785 if (align
& (align
- 1))
787 return mod
? dump_skip(cprm
, align
- mod
) : 1;
789 EXPORT_SYMBOL(dump_align
);
This page took 0.049137 seconds and 6 git commands to generate.