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>
36 #include <asm/uaccess.h>
37 #include <asm/mmu_context.h>
41 #include <trace/events/task.h>
45 #include <trace/events/sched.h>
48 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
49 unsigned int core_pipe_limit
;
55 static atomic_t call_count
= ATOMIC_INIT(1);
57 /* The maximal length of core_pattern is also specified in sysctl.c */
59 static int expand_corename(struct core_name
*cn
)
61 char *old_corename
= cn
->corename
;
63 cn
->size
= CORENAME_MAX_SIZE
* atomic_inc_return(&call_count
);
64 cn
->corename
= krealloc(old_corename
, cn
->size
, GFP_KERNEL
);
74 static int cn_printf(struct core_name
*cn
, const char *fmt
, ...)
82 need
= vsnprintf(NULL
, 0, fmt
, arg
);
85 if (likely(need
< cn
->size
- cn
->used
- 1))
88 ret
= expand_corename(cn
);
93 cur
= cn
->corename
+ cn
->used
;
95 vsnprintf(cur
, need
+ 1, fmt
, arg
);
104 static void cn_escape(char *str
)
111 static int cn_print_exe_file(struct core_name
*cn
)
113 struct file
*exe_file
;
114 char *pathbuf
, *path
;
117 exe_file
= get_mm_exe_file(current
->mm
);
119 char *commstart
= cn
->corename
+ cn
->used
;
120 ret
= cn_printf(cn
, "%s (path unknown)", current
->comm
);
121 cn_escape(commstart
);
125 pathbuf
= kmalloc(PATH_MAX
, GFP_TEMPORARY
);
131 path
= d_path(&exe_file
->f_path
, pathbuf
, PATH_MAX
);
139 ret
= cn_printf(cn
, "%s", path
);
148 /* format_corename will inspect the pattern parameter, and output a
149 * name into corename, which must have space for at least
150 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
152 static int format_corename(struct core_name
*cn
, struct coredump_params
*cprm
)
154 const struct cred
*cred
= current_cred();
155 const char *pat_ptr
= core_pattern
;
156 int ispipe
= (*pat_ptr
== '|');
157 int pid_in_pattern
= 0;
160 cn
->size
= CORENAME_MAX_SIZE
* atomic_read(&call_count
);
161 cn
->corename
= kmalloc(cn
->size
, GFP_KERNEL
);
167 /* Repeat as long as we have more pattern to process and more output
170 if (*pat_ptr
!= '%') {
173 err
= cn_printf(cn
, "%c", *pat_ptr
++);
175 switch (*++pat_ptr
) {
176 /* single % at the end, drop that */
179 /* Double percent, output one percent */
181 err
= cn_printf(cn
, "%c", '%');
186 err
= cn_printf(cn
, "%d",
187 task_tgid_vnr(current
));
191 err
= cn_printf(cn
, "%d", cred
->uid
);
195 err
= cn_printf(cn
, "%d", cred
->gid
);
198 err
= cn_printf(cn
, "%d",
199 __get_dumpable(cprm
->mm_flags
));
201 /* signal that caused the coredump */
203 err
= cn_printf(cn
, "%ld", cprm
->siginfo
->si_signo
);
205 /* UNIX time of coredump */
208 do_gettimeofday(&tv
);
209 err
= cn_printf(cn
, "%lu", tv
.tv_sec
);
214 char *namestart
= cn
->corename
+ cn
->used
;
216 err
= cn_printf(cn
, "%s",
217 utsname()->nodename
);
219 cn_escape(namestart
);
224 char *commstart
= cn
->corename
+ cn
->used
;
225 err
= cn_printf(cn
, "%s", current
->comm
);
226 cn_escape(commstart
);
230 err
= cn_print_exe_file(cn
);
232 /* core limit size */
234 err
= cn_printf(cn
, "%lu",
235 rlimit(RLIMIT_CORE
));
247 /* Backward compatibility with core_uses_pid:
249 * If core_pattern does not include a %p (as is the default)
250 * and core_uses_pid is set, then .%pid will be appended to
251 * the filename. Do not do this for piped commands. */
252 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
253 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
261 static int zap_process(struct task_struct
*start
, int exit_code
)
263 struct task_struct
*t
;
266 start
->signal
->group_exit_code
= exit_code
;
267 start
->signal
->group_stop_count
= 0;
271 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
272 if (t
!= current
&& t
->mm
) {
273 sigaddset(&t
->pending
.signal
, SIGKILL
);
274 signal_wake_up(t
, 1);
277 } while_each_thread(start
, t
);
282 static int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
283 struct core_state
*core_state
, int exit_code
)
285 struct task_struct
*g
, *p
;
289 spin_lock_irq(&tsk
->sighand
->siglock
);
290 if (!signal_group_exit(tsk
->signal
)) {
291 mm
->core_state
= core_state
;
292 nr
= zap_process(tsk
, exit_code
);
293 tsk
->signal
->group_exit_task
= tsk
;
294 /* ignore all signals except SIGKILL, see prepare_signal() */
295 tsk
->signal
->flags
= SIGNAL_GROUP_COREDUMP
;
296 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
298 spin_unlock_irq(&tsk
->sighand
->siglock
);
299 if (unlikely(nr
< 0))
302 tsk
->flags
= PF_DUMPCORE
;
303 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
306 * We should find and kill all tasks which use this mm, and we should
307 * count them correctly into ->nr_threads. We don't take tasklist
308 * lock, but this is safe wrt:
311 * None of sub-threads can fork after zap_process(leader). All
312 * processes which were created before this point should be
313 * visible to zap_threads() because copy_process() adds the new
314 * process to the tail of init_task.tasks list, and lock/unlock
315 * of ->siglock provides a memory barrier.
318 * The caller holds mm->mmap_sem. This means that the task which
319 * uses this mm can't pass exit_mm(), so it can't exit or clear
323 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
324 * we must see either old or new leader, this does not matter.
325 * However, it can change p->sighand, so lock_task_sighand(p)
326 * must be used. Since p->mm != NULL and we hold ->mmap_sem
329 * Note also that "g" can be the old leader with ->mm == NULL
330 * and already unhashed and thus removed from ->thread_group.
331 * This is OK, __unhash_process()->list_del_rcu() does not
332 * clear the ->next pointer, we will find the new leader via
336 for_each_process(g
) {
337 if (g
== tsk
->group_leader
)
339 if (g
->flags
& PF_KTHREAD
)
344 if (unlikely(p
->mm
== mm
)) {
345 lock_task_sighand(p
, &flags
);
346 nr
+= zap_process(p
, exit_code
);
347 p
->signal
->flags
= SIGNAL_GROUP_EXIT
;
348 unlock_task_sighand(p
, &flags
);
352 } while_each_thread(g
, p
);
356 atomic_set(&core_state
->nr_threads
, nr
);
360 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
362 struct task_struct
*tsk
= current
;
363 struct mm_struct
*mm
= tsk
->mm
;
364 int core_waiters
= -EBUSY
;
366 init_completion(&core_state
->startup
);
367 core_state
->dumper
.task
= tsk
;
368 core_state
->dumper
.next
= NULL
;
370 down_write(&mm
->mmap_sem
);
372 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
373 up_write(&mm
->mmap_sem
);
375 if (core_waiters
> 0) {
376 struct core_thread
*ptr
;
378 wait_for_completion(&core_state
->startup
);
380 * Wait for all the threads to become inactive, so that
381 * all the thread context (extended register state, like
382 * fpu etc) gets copied to the memory.
384 ptr
= core_state
->dumper
.next
;
385 while (ptr
!= NULL
) {
386 wait_task_inactive(ptr
->task
, 0);
394 static void coredump_finish(struct mm_struct
*mm
, bool core_dumped
)
396 struct core_thread
*curr
, *next
;
397 struct task_struct
*task
;
399 spin_lock_irq(¤t
->sighand
->siglock
);
400 if (core_dumped
&& !__fatal_signal_pending(current
))
401 current
->signal
->group_exit_code
|= 0x80;
402 current
->signal
->group_exit_task
= NULL
;
403 current
->signal
->flags
= SIGNAL_GROUP_EXIT
;
404 spin_unlock_irq(¤t
->sighand
->siglock
);
406 next
= mm
->core_state
->dumper
.next
;
407 while ((curr
= next
) != NULL
) {
411 * see exit_mm(), curr->task must not see
412 * ->task == NULL before we read ->next.
416 wake_up_process(task
);
419 mm
->core_state
= NULL
;
422 static bool dump_interrupted(void)
425 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
426 * can do try_to_freeze() and check __fatal_signal_pending(),
427 * but then we need to teach dump_write() to restart and clear
430 return signal_pending(current
);
433 static void wait_for_dump_helpers(struct file
*file
)
435 struct pipe_inode_info
*pipe
= file
->private_data
;
440 wake_up_interruptible_sync(&pipe
->wait
);
441 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
445 * We actually want wait_event_freezable() but then we need
446 * to clear TIF_SIGPENDING and improve dump_interrupted().
448 wait_event_interruptible(pipe
->wait
, pipe
->readers
== 1);
458 * helper function to customize the process used
459 * to collect the core in userspace. Specifically
460 * it sets up a pipe and installs it as fd 0 (stdin)
461 * for the process. Returns 0 on success, or
462 * PTR_ERR on failure.
463 * Note that it also sets the core limit to 1. This
464 * is a special value that we use to trap recursive
467 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
469 struct file
*files
[2];
470 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
471 int err
= create_pipe_files(files
, 0);
477 err
= replace_fd(0, files
[0], 0);
479 /* and disallow core files too */
480 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
485 void do_coredump(siginfo_t
*siginfo
)
487 struct core_state core_state
;
489 struct mm_struct
*mm
= current
->mm
;
490 struct linux_binfmt
* binfmt
;
491 const struct cred
*old_cred
;
496 struct files_struct
*displaced
;
497 bool need_nonrelative
= false;
498 bool core_dumped
= false;
499 static atomic_t core_dump_count
= ATOMIC_INIT(0);
500 struct coredump_params cprm
= {
502 .regs
= signal_pt_regs(),
503 .limit
= rlimit(RLIMIT_CORE
),
505 * We must use the same mm->flags while dumping core to avoid
506 * inconsistency of bit flags, since this flag is not protected
509 .mm_flags
= mm
->flags
,
512 audit_core_dumps(siginfo
->si_signo
);
515 if (!binfmt
|| !binfmt
->core_dump
)
517 if (!__get_dumpable(cprm
.mm_flags
))
520 cred
= prepare_creds();
524 * We cannot trust fsuid as being the "true" uid of the process
525 * nor do we know its entire history. We only know it was tainted
526 * so we dump it as root in mode 2, and only into a controlled
527 * environment (pipe handler or fully qualified path).
529 if (__get_dumpable(cprm
.mm_flags
) == SUID_DUMP_ROOT
) {
530 /* Setuid core dump mode */
531 flag
= O_EXCL
; /* Stop rewrite attacks */
532 cred
->fsuid
= GLOBAL_ROOT_UID
; /* Dump root private */
533 need_nonrelative
= true;
536 retval
= coredump_wait(siginfo
->si_signo
, &core_state
);
540 old_cred
= override_creds(cred
);
542 ispipe
= format_corename(&cn
, &cprm
);
547 struct subprocess_info
*sub_info
;
550 printk(KERN_WARNING
"format_corename failed\n");
551 printk(KERN_WARNING
"Aborting core\n");
555 if (cprm
.limit
== 1) {
556 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
558 * Normally core limits are irrelevant to pipes, since
559 * we're not writing to the file system, but we use
560 * cprm.limit of 1 here as a speacial value, this is a
561 * consistent way to catch recursive crashes.
562 * We can still crash if the core_pattern binary sets
563 * RLIM_CORE = !1, but it runs as root, and can do
564 * lots of stupid things.
566 * Note that we use task_tgid_vnr here to grab the pid
567 * of the process group leader. That way we get the
568 * right pid if a thread in a multi-threaded
569 * core_pattern process dies.
572 "Process %d(%s) has RLIMIT_CORE set to 1\n",
573 task_tgid_vnr(current
), current
->comm
);
574 printk(KERN_WARNING
"Aborting core\n");
577 cprm
.limit
= RLIM_INFINITY
;
579 dump_count
= atomic_inc_return(&core_dump_count
);
580 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
581 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
582 task_tgid_vnr(current
), current
->comm
);
583 printk(KERN_WARNING
"Skipping core dump\n");
587 helper_argv
= argv_split(GFP_KERNEL
, cn
.corename
+1, NULL
);
589 printk(KERN_WARNING
"%s failed to allocate memory\n",
595 sub_info
= call_usermodehelper_setup(helper_argv
[0],
596 helper_argv
, NULL
, GFP_KERNEL
,
597 umh_pipe_setup
, NULL
, &cprm
);
599 retval
= call_usermodehelper_exec(sub_info
,
602 argv_free(helper_argv
);
604 printk(KERN_INFO
"Core dump to %s pipe failed\n",
611 if (cprm
.limit
< binfmt
->min_coredump
)
614 if (need_nonrelative
&& cn
.corename
[0] != '/') {
615 printk(KERN_WARNING
"Pid %d(%s) can only dump core "\
616 "to fully qualified path!\n",
617 task_tgid_vnr(current
), current
->comm
);
618 printk(KERN_WARNING
"Skipping core dump\n");
622 cprm
.file
= filp_open(cn
.corename
,
623 O_CREAT
| 2 | O_NOFOLLOW
| O_LARGEFILE
| flag
,
625 if (IS_ERR(cprm
.file
))
628 inode
= file_inode(cprm
.file
);
629 if (inode
->i_nlink
> 1)
631 if (d_unhashed(cprm
.file
->f_path
.dentry
))
634 * AK: actually i see no reason to not allow this for named
635 * pipes etc, but keep the previous behaviour for now.
637 if (!S_ISREG(inode
->i_mode
))
640 * Dont allow local users get cute and trick others to coredump
641 * into their pre-created files.
643 if (!uid_eq(inode
->i_uid
, current_fsuid()))
645 if (!cprm
.file
->f_op
|| !cprm
.file
->f_op
->write
)
647 if (do_truncate(cprm
.file
->f_path
.dentry
, 0, 0, cprm
.file
))
651 /* get us an unshared descriptor table; almost always a no-op */
652 retval
= unshare_files(&displaced
);
656 put_files_struct(displaced
);
657 if (!dump_interrupted()) {
658 file_start_write(cprm
.file
);
659 core_dumped
= binfmt
->core_dump(&cprm
);
660 file_end_write(cprm
.file
);
662 if (ispipe
&& core_pipe_limit
)
663 wait_for_dump_helpers(cprm
.file
);
666 filp_close(cprm
.file
, NULL
);
669 atomic_dec(&core_dump_count
);
673 coredump_finish(mm
, core_dumped
);
674 revert_creds(old_cred
);
682 * Core dumping helper functions. These are the only things you should
683 * do on a core-file: use only these functions to write out all the
686 int dump_write(struct file
*file
, const void *addr
, int nr
)
688 return !dump_interrupted() &&
689 access_ok(VERIFY_READ
, addr
, nr
) &&
690 file
->f_op
->write(file
, addr
, nr
, &file
->f_pos
) == nr
;
692 EXPORT_SYMBOL(dump_write
);
694 int dump_seek(struct file
*file
, loff_t off
)
698 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
699 if (dump_interrupted() ||
700 file
->f_op
->llseek(file
, off
, SEEK_CUR
) < 0)
703 char *buf
= (char *)get_zeroed_page(GFP_KERNEL
);
708 unsigned long n
= off
;
712 if (!dump_write(file
, buf
, n
)) {
718 free_page((unsigned long)buf
);
722 EXPORT_SYMBOL(dump_seek
);