]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/coredump.c
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/sched.h>
37 #include <linux/path.h>
38 #include <linux/timekeeping.h>
40 #include <asm/uaccess.h>
41 #include <asm/mmu_context.h>
45 #include <trace/events/task.h>
48 #include <trace/events/sched.h>
51 unsigned int core_pipe_limit
;
52 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
53 static int core_name_size
= CORENAME_MAX_SIZE
;
60 /* The maximal length of core_pattern is also specified in sysctl.c */
62 static int expand_corename(struct core_name
*cn
, int size
)
64 char *corename
= krealloc(cn
->corename
, size
, GFP_KERNEL
);
69 if (size
> core_name_size
) /* racy but harmless */
70 core_name_size
= size
;
72 cn
->size
= ksize(corename
);
73 cn
->corename
= corename
;
77 static __printf(2, 0) int cn_vprintf(struct core_name
*cn
, const char *fmt
,
84 free
= cn
->size
- cn
->used
;
86 va_copy(arg_copy
, arg
);
87 need
= vsnprintf(cn
->corename
+ cn
->used
, free
, fmt
, arg_copy
);
95 if (!expand_corename(cn
, cn
->size
+ need
- free
+ 1))
101 static __printf(2, 3) int cn_printf(struct core_name
*cn
, const char *fmt
, ...)
107 ret
= cn_vprintf(cn
, fmt
, arg
);
113 static __printf(2, 3)
114 int cn_esc_printf(struct core_name
*cn
, const char *fmt
, ...)
121 ret
= cn_vprintf(cn
, fmt
, arg
);
126 * Ensure that this coredump name component can't cause the
127 * resulting corefile path to consist of a ".." or ".".
129 if ((cn
->used
- cur
== 1 && cn
->corename
[cur
] == '.') ||
130 (cn
->used
- cur
== 2 && cn
->corename
[cur
] == '.'
131 && cn
->corename
[cur
+1] == '.'))
132 cn
->corename
[cur
] = '!';
135 * Empty names are fishy and could be used to create a "//" in a
136 * corefile name, causing the coredump to happen one directory
137 * level too high. Enforce that all components of the core
138 * pattern are at least one character long.
141 ret
= cn_printf(cn
, "!");
144 for (; cur
< cn
->used
; ++cur
) {
145 if (cn
->corename
[cur
] == '/')
146 cn
->corename
[cur
] = '!';
151 static int cn_print_exe_file(struct core_name
*cn
)
153 struct file
*exe_file
;
154 char *pathbuf
, *path
;
157 exe_file
= get_mm_exe_file(current
->mm
);
159 return cn_esc_printf(cn
, "%s (path unknown)", current
->comm
);
161 pathbuf
= kmalloc(PATH_MAX
, GFP_TEMPORARY
);
167 path
= file_path(exe_file
, pathbuf
, PATH_MAX
);
173 ret
= cn_esc_printf(cn
, "%s", path
);
182 /* format_corename will inspect the pattern parameter, and output a
183 * name into corename, which must have space for at least
184 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
186 static int format_corename(struct core_name
*cn
, struct coredump_params
*cprm
)
188 const struct cred
*cred
= current_cred();
189 const char *pat_ptr
= core_pattern
;
190 int ispipe
= (*pat_ptr
== '|');
191 int pid_in_pattern
= 0;
196 if (expand_corename(cn
, core_name_size
))
198 cn
->corename
[0] = '\0';
203 /* Repeat as long as we have more pattern to process and more output
206 if (*pat_ptr
!= '%') {
207 err
= cn_printf(cn
, "%c", *pat_ptr
++);
209 switch (*++pat_ptr
) {
210 /* single % at the end, drop that */
213 /* Double percent, output one percent */
215 err
= cn_printf(cn
, "%c", '%');
220 err
= cn_printf(cn
, "%d",
221 task_tgid_vnr(current
));
225 err
= cn_printf(cn
, "%d",
226 task_tgid_nr(current
));
229 err
= cn_printf(cn
, "%d",
230 task_pid_vnr(current
));
233 err
= cn_printf(cn
, "%d",
234 task_pid_nr(current
));
238 err
= cn_printf(cn
, "%u",
239 from_kuid(&init_user_ns
,
244 err
= cn_printf(cn
, "%u",
245 from_kgid(&init_user_ns
,
249 err
= cn_printf(cn
, "%d",
250 __get_dumpable(cprm
->mm_flags
));
252 /* signal that caused the coredump */
254 err
= cn_printf(cn
, "%d",
255 cprm
->siginfo
->si_signo
);
257 /* UNIX time of coredump */
261 time
= ktime_get_real_seconds();
262 err
= cn_printf(cn
, "%lld", time
);
268 err
= cn_esc_printf(cn
, "%s",
269 utsname()->nodename
);
274 err
= cn_esc_printf(cn
, "%s", current
->comm
);
277 err
= cn_print_exe_file(cn
);
279 /* core limit size */
281 err
= cn_printf(cn
, "%lu",
282 rlimit(RLIMIT_CORE
));
295 /* Backward compatibility with core_uses_pid:
297 * If core_pattern does not include a %p (as is the default)
298 * and core_uses_pid is set, then .%pid will be appended to
299 * the filename. Do not do this for piped commands. */
300 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
301 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
308 static int zap_process(struct task_struct
*start
, int exit_code
, int flags
)
310 struct task_struct
*t
;
313 /* ignore all signals except SIGKILL, see prepare_signal() */
314 start
->signal
->flags
= SIGNAL_GROUP_COREDUMP
| flags
;
315 start
->signal
->group_exit_code
= exit_code
;
316 start
->signal
->group_stop_count
= 0;
318 for_each_thread(start
, t
) {
319 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
320 if (t
!= current
&& t
->mm
) {
321 sigaddset(&t
->pending
.signal
, SIGKILL
);
322 signal_wake_up(t
, 1);
330 static int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
331 struct core_state
*core_state
, int exit_code
)
333 struct task_struct
*g
, *p
;
337 spin_lock_irq(&tsk
->sighand
->siglock
);
338 if (!signal_group_exit(tsk
->signal
)) {
339 mm
->core_state
= core_state
;
340 tsk
->signal
->group_exit_task
= tsk
;
341 nr
= zap_process(tsk
, exit_code
, 0);
342 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
344 spin_unlock_irq(&tsk
->sighand
->siglock
);
345 if (unlikely(nr
< 0))
348 tsk
->flags
|= PF_DUMPCORE
;
349 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
352 * We should find and kill all tasks which use this mm, and we should
353 * count them correctly into ->nr_threads. We don't take tasklist
354 * lock, but this is safe wrt:
357 * None of sub-threads can fork after zap_process(leader). All
358 * processes which were created before this point should be
359 * visible to zap_threads() because copy_process() adds the new
360 * process to the tail of init_task.tasks list, and lock/unlock
361 * of ->siglock provides a memory barrier.
364 * The caller holds mm->mmap_sem. This means that the task which
365 * uses this mm can't pass exit_mm(), so it can't exit or clear
369 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
370 * we must see either old or new leader, this does not matter.
371 * However, it can change p->sighand, so lock_task_sighand(p)
372 * must be used. Since p->mm != NULL and we hold ->mmap_sem
375 * Note also that "g" can be the old leader with ->mm == NULL
376 * and already unhashed and thus removed from ->thread_group.
377 * This is OK, __unhash_process()->list_del_rcu() does not
378 * clear the ->next pointer, we will find the new leader via
382 for_each_process(g
) {
383 if (g
== tsk
->group_leader
)
385 if (g
->flags
& PF_KTHREAD
)
388 for_each_thread(g
, p
) {
389 if (unlikely(!p
->mm
))
391 if (unlikely(p
->mm
== mm
)) {
392 lock_task_sighand(p
, &flags
);
393 nr
+= zap_process(p
, exit_code
,
395 unlock_task_sighand(p
, &flags
);
402 atomic_set(&core_state
->nr_threads
, nr
);
406 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
408 struct task_struct
*tsk
= current
;
409 struct mm_struct
*mm
= tsk
->mm
;
410 int core_waiters
= -EBUSY
;
412 init_completion(&core_state
->startup
);
413 core_state
->dumper
.task
= tsk
;
414 core_state
->dumper
.next
= NULL
;
416 if (down_write_killable(&mm
->mmap_sem
))
420 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
421 up_write(&mm
->mmap_sem
);
423 if (core_waiters
> 0) {
424 struct core_thread
*ptr
;
426 wait_for_completion(&core_state
->startup
);
428 * Wait for all the threads to become inactive, so that
429 * all the thread context (extended register state, like
430 * fpu etc) gets copied to the memory.
432 ptr
= core_state
->dumper
.next
;
433 while (ptr
!= NULL
) {
434 wait_task_inactive(ptr
->task
, 0);
442 static void coredump_finish(struct mm_struct
*mm
, bool core_dumped
)
444 struct core_thread
*curr
, *next
;
445 struct task_struct
*task
;
447 spin_lock_irq(¤t
->sighand
->siglock
);
448 if (core_dumped
&& !__fatal_signal_pending(current
))
449 current
->signal
->group_exit_code
|= 0x80;
450 current
->signal
->group_exit_task
= NULL
;
451 current
->signal
->flags
= SIGNAL_GROUP_EXIT
;
452 spin_unlock_irq(¤t
->sighand
->siglock
);
454 next
= mm
->core_state
->dumper
.next
;
455 while ((curr
= next
) != NULL
) {
459 * see exit_mm(), curr->task must not see
460 * ->task == NULL before we read ->next.
464 wake_up_process(task
);
467 mm
->core_state
= NULL
;
470 static bool dump_interrupted(void)
473 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
474 * can do try_to_freeze() and check __fatal_signal_pending(),
475 * but then we need to teach dump_write() to restart and clear
478 return signal_pending(current
);
481 static void wait_for_dump_helpers(struct file
*file
)
483 struct pipe_inode_info
*pipe
= file
->private_data
;
488 wake_up_interruptible_sync(&pipe
->wait
);
489 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
493 * We actually want wait_event_freezable() but then we need
494 * to clear TIF_SIGPENDING and improve dump_interrupted().
496 wait_event_interruptible(pipe
->wait
, pipe
->readers
== 1);
506 * helper function to customize the process used
507 * to collect the core in userspace. Specifically
508 * it sets up a pipe and installs it as fd 0 (stdin)
509 * for the process. Returns 0 on success, or
510 * PTR_ERR on failure.
511 * Note that it also sets the core limit to 1. This
512 * is a special value that we use to trap recursive
515 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
517 struct file
*files
[2];
518 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
519 int err
= create_pipe_files(files
, 0);
525 err
= replace_fd(0, files
[0], 0);
527 /* and disallow core files too */
528 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
533 void do_coredump(const siginfo_t
*siginfo
)
535 struct core_state core_state
;
537 struct mm_struct
*mm
= current
->mm
;
538 struct linux_binfmt
* binfmt
;
539 const struct cred
*old_cred
;
543 struct files_struct
*displaced
;
544 /* require nonrelative corefile path and be extra careful */
545 bool need_suid_safe
= false;
546 bool core_dumped
= false;
547 static atomic_t core_dump_count
= ATOMIC_INIT(0);
548 struct coredump_params cprm
= {
550 .regs
= signal_pt_regs(),
551 .limit
= rlimit(RLIMIT_CORE
),
553 * We must use the same mm->flags while dumping core to avoid
554 * inconsistency of bit flags, since this flag is not protected
557 .mm_flags
= mm
->flags
,
560 audit_core_dumps(siginfo
->si_signo
);
563 if (!binfmt
|| !binfmt
->core_dump
)
565 if (!__get_dumpable(cprm
.mm_flags
))
568 cred
= prepare_creds();
572 * We cannot trust fsuid as being the "true" uid of the process
573 * nor do we know its entire history. We only know it was tainted
574 * so we dump it as root in mode 2, and only into a controlled
575 * environment (pipe handler or fully qualified path).
577 if (__get_dumpable(cprm
.mm_flags
) == SUID_DUMP_ROOT
) {
578 /* Setuid core dump mode */
579 cred
->fsuid
= GLOBAL_ROOT_UID
; /* Dump root private */
580 need_suid_safe
= true;
583 retval
= coredump_wait(siginfo
->si_signo
, &core_state
);
587 old_cred
= override_creds(cred
);
589 ispipe
= format_corename(&cn
, &cprm
);
594 struct subprocess_info
*sub_info
;
597 printk(KERN_WARNING
"format_corename failed\n");
598 printk(KERN_WARNING
"Aborting core\n");
602 if (cprm
.limit
== 1) {
603 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
605 * Normally core limits are irrelevant to pipes, since
606 * we're not writing to the file system, but we use
607 * cprm.limit of 1 here as a special value, this is a
608 * consistent way to catch recursive crashes.
609 * We can still crash if the core_pattern binary sets
610 * RLIM_CORE = !1, but it runs as root, and can do
611 * lots of stupid things.
613 * Note that we use task_tgid_vnr here to grab the pid
614 * of the process group leader. That way we get the
615 * right pid if a thread in a multi-threaded
616 * core_pattern process dies.
619 "Process %d(%s) has RLIMIT_CORE set to 1\n",
620 task_tgid_vnr(current
), current
->comm
);
621 printk(KERN_WARNING
"Aborting core\n");
624 cprm
.limit
= RLIM_INFINITY
;
626 dump_count
= atomic_inc_return(&core_dump_count
);
627 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
628 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
629 task_tgid_vnr(current
), current
->comm
);
630 printk(KERN_WARNING
"Skipping core dump\n");
634 helper_argv
= argv_split(GFP_KERNEL
, cn
.corename
, NULL
);
636 printk(KERN_WARNING
"%s failed to allocate memory\n",
642 sub_info
= call_usermodehelper_setup(helper_argv
[0],
643 helper_argv
, NULL
, GFP_KERNEL
,
644 umh_pipe_setup
, NULL
, &cprm
);
646 retval
= call_usermodehelper_exec(sub_info
,
649 argv_free(helper_argv
);
651 printk(KERN_INFO
"Core dump to |%s pipe failed\n",
657 int open_flags
= O_CREAT
| O_RDWR
| O_NOFOLLOW
|
658 O_LARGEFILE
| O_EXCL
;
660 if (cprm
.limit
< binfmt
->min_coredump
)
663 if (need_suid_safe
&& cn
.corename
[0] != '/') {
664 printk(KERN_WARNING
"Pid %d(%s) can only dump core "\
665 "to fully qualified path!\n",
666 task_tgid_vnr(current
), current
->comm
);
667 printk(KERN_WARNING
"Skipping core dump\n");
672 * Unlink the file if it exists unless this is a SUID
673 * binary - in that case, we're running around with root
674 * privs and don't want to unlink another user's coredump.
676 if (!need_suid_safe
) {
682 * If it doesn't exist, that's fine. If there's some
683 * other problem, we'll catch it at the filp_open().
685 (void) sys_unlink((const char __user
*)cn
.corename
);
690 * There is a race between unlinking and creating the
691 * file, but if that causes an EEXIST here, that's
692 * fine - another process raced with us while creating
693 * the corefile, and the other process won. To userspace,
694 * what matters is that at least one of the two processes
695 * writes its coredump successfully, not which one.
697 if (need_suid_safe
) {
699 * Using user namespaces, normal user tasks can change
700 * their current->fs->root to point to arbitrary
701 * directories. Since the intention of the "only dump
702 * with a fully qualified path" rule is to control where
703 * coredumps may be placed using root privileges,
704 * current->fs->root must not be used. Instead, use the
705 * root directory of init_task.
709 task_lock(&init_task
);
710 get_fs_root(init_task
.fs
, &root
);
711 task_unlock(&init_task
);
712 cprm
.file
= file_open_root(root
.dentry
, root
.mnt
,
713 cn
.corename
, open_flags
, 0600);
716 cprm
.file
= filp_open(cn
.corename
, open_flags
, 0600);
718 if (IS_ERR(cprm
.file
))
721 inode
= file_inode(cprm
.file
);
722 if (inode
->i_nlink
> 1)
724 if (d_unhashed(cprm
.file
->f_path
.dentry
))
727 * AK: actually i see no reason to not allow this for named
728 * pipes etc, but keep the previous behaviour for now.
730 if (!S_ISREG(inode
->i_mode
))
733 * Don't dump core if the filesystem changed owner or mode
734 * of the file during file creation. This is an issue when
735 * a process dumps core while its cwd is e.g. on a vfat
738 if (!uid_eq(inode
->i_uid
, current_fsuid()))
740 if ((inode
->i_mode
& 0677) != 0600)
742 if (!(cprm
.file
->f_mode
& FMODE_CAN_WRITE
))
744 if (do_truncate(cprm
.file
->f_path
.dentry
, 0, 0, cprm
.file
))
748 /* get us an unshared descriptor table; almost always a no-op */
749 retval
= unshare_files(&displaced
);
753 put_files_struct(displaced
);
754 if (!dump_interrupted()) {
755 file_start_write(cprm
.file
);
756 core_dumped
= binfmt
->core_dump(&cprm
);
757 file_end_write(cprm
.file
);
759 if (ispipe
&& core_pipe_limit
)
760 wait_for_dump_helpers(cprm
.file
);
763 filp_close(cprm
.file
, NULL
);
766 atomic_dec(&core_dump_count
);
769 coredump_finish(mm
, core_dumped
);
770 revert_creds(old_cred
);
778 * Core dumping helper functions. These are the only things you should
779 * do on a core-file: use only these functions to write out all the
782 int dump_emit(struct coredump_params
*cprm
, const void *addr
, int nr
)
784 struct file
*file
= cprm
->file
;
785 loff_t pos
= file
->f_pos
;
787 if (cprm
->written
+ nr
> cprm
->limit
)
790 if (dump_interrupted())
792 n
= __kernel_write(file
, addr
, nr
, &pos
);
802 EXPORT_SYMBOL(dump_emit
);
804 int dump_skip(struct coredump_params
*cprm
, size_t nr
)
806 static char zeroes
[PAGE_SIZE
];
807 struct file
*file
= cprm
->file
;
808 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
809 if (dump_interrupted() ||
810 file
->f_op
->llseek(file
, nr
, SEEK_CUR
) < 0)
815 while (nr
> PAGE_SIZE
) {
816 if (!dump_emit(cprm
, zeroes
, PAGE_SIZE
))
820 return dump_emit(cprm
, zeroes
, nr
);
823 EXPORT_SYMBOL(dump_skip
);
825 int dump_align(struct coredump_params
*cprm
, int align
)
827 unsigned mod
= cprm
->pos
& (align
- 1);
828 if (align
& (align
- 1))
830 return mod
? dump_skip(cprm
, align
- mod
) : 1;
832 EXPORT_SYMBOL(dump_align
);