]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/coredump.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/slab.h>
3 #include <linux/file.h>
4 #include <linux/fdtable.h>
5 #include <linux/freezer.h>
7 #include <linux/stat.h>
8 #include <linux/fcntl.h>
9 #include <linux/swap.h>
10 #include <linux/ctype.h>
11 #include <linux/string.h>
12 #include <linux/init.h>
13 #include <linux/pagemap.h>
14 #include <linux/perf_event.h>
15 #include <linux/highmem.h>
16 #include <linux/spinlock.h>
17 #include <linux/key.h>
18 #include <linux/personality.h>
19 #include <linux/binfmts.h>
20 #include <linux/coredump.h>
21 #include <linux/sched/coredump.h>
22 #include <linux/sched/signal.h>
23 #include <linux/sched/task_stack.h>
24 #include <linux/utsname.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/module.h>
27 #include <linux/namei.h>
28 #include <linux/mount.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/tsacct_kern.h>
32 #include <linux/cn_proc.h>
33 #include <linux/audit.h>
34 #include <linux/tracehook.h>
35 #include <linux/kmod.h>
36 #include <linux/fsnotify.h>
37 #include <linux/fs_struct.h>
38 #include <linux/pipe_fs_i.h>
39 #include <linux/oom.h>
40 #include <linux/compat.h>
42 #include <linux/path.h>
43 #include <linux/timekeeping.h>
45 #include <linux/uaccess.h>
46 #include <asm/mmu_context.h>
50 #include <trace/events/task.h>
53 #include <trace/events/sched.h>
56 unsigned int core_pipe_limit
;
57 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
58 static int core_name_size
= CORENAME_MAX_SIZE
;
65 /* The maximal length of core_pattern is also specified in sysctl.c */
67 static int expand_corename(struct core_name
*cn
, int size
)
69 char *corename
= krealloc(cn
->corename
, size
, GFP_KERNEL
);
74 if (size
> core_name_size
) /* racy but harmless */
75 core_name_size
= size
;
77 cn
->size
= ksize(corename
);
78 cn
->corename
= corename
;
82 static __printf(2, 0) int cn_vprintf(struct core_name
*cn
, const char *fmt
,
89 free
= cn
->size
- cn
->used
;
91 va_copy(arg_copy
, arg
);
92 need
= vsnprintf(cn
->corename
+ cn
->used
, free
, fmt
, arg_copy
);
100 if (!expand_corename(cn
, cn
->size
+ need
- free
+ 1))
106 static __printf(2, 3) int cn_printf(struct core_name
*cn
, const char *fmt
, ...)
112 ret
= cn_vprintf(cn
, fmt
, arg
);
118 static __printf(2, 3)
119 int cn_esc_printf(struct core_name
*cn
, const char *fmt
, ...)
126 ret
= cn_vprintf(cn
, fmt
, arg
);
131 * Ensure that this coredump name component can't cause the
132 * resulting corefile path to consist of a ".." or ".".
134 if ((cn
->used
- cur
== 1 && cn
->corename
[cur
] == '.') ||
135 (cn
->used
- cur
== 2 && cn
->corename
[cur
] == '.'
136 && cn
->corename
[cur
+1] == '.'))
137 cn
->corename
[cur
] = '!';
140 * Empty names are fishy and could be used to create a "//" in a
141 * corefile name, causing the coredump to happen one directory
142 * level too high. Enforce that all components of the core
143 * pattern are at least one character long.
146 ret
= cn_printf(cn
, "!");
149 for (; cur
< cn
->used
; ++cur
) {
150 if (cn
->corename
[cur
] == '/')
151 cn
->corename
[cur
] = '!';
156 static int cn_print_exe_file(struct core_name
*cn
)
158 struct file
*exe_file
;
159 char *pathbuf
, *path
;
162 exe_file
= get_mm_exe_file(current
->mm
);
164 return cn_esc_printf(cn
, "%s (path unknown)", current
->comm
);
166 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
172 path
= file_path(exe_file
, pathbuf
, PATH_MAX
);
178 ret
= cn_esc_printf(cn
, "%s", path
);
187 /* format_corename will inspect the pattern parameter, and output a
188 * name into corename, which must have space for at least
189 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
191 static int format_corename(struct core_name
*cn
, struct coredump_params
*cprm
,
192 size_t **argv
, int *argc
)
194 const struct cred
*cred
= current_cred();
195 const char *pat_ptr
= core_pattern
;
196 int ispipe
= (*pat_ptr
== '|');
197 bool was_space
= false;
198 int pid_in_pattern
= 0;
203 if (expand_corename(cn
, core_name_size
))
205 cn
->corename
[0] = '\0';
208 int argvs
= sizeof(core_pattern
) / 2;
209 (*argv
) = kmalloc_array(argvs
, sizeof(**argv
), GFP_KERNEL
);
212 (*argv
)[(*argc
)++] = 0;
216 /* Repeat as long as we have more pattern to process and more output
220 * Split on spaces before doing template expansion so that
221 * %e and %E don't get split if they have spaces in them
224 if (isspace(*pat_ptr
)) {
228 } else if (was_space
) {
230 err
= cn_printf(cn
, "%c", '\0');
233 (*argv
)[(*argc
)++] = cn
->used
;
236 if (*pat_ptr
!= '%') {
237 err
= cn_printf(cn
, "%c", *pat_ptr
++);
239 switch (*++pat_ptr
) {
240 /* single % at the end, drop that */
243 /* Double percent, output one percent */
245 err
= cn_printf(cn
, "%c", '%');
250 err
= cn_printf(cn
, "%d",
251 task_tgid_vnr(current
));
255 err
= cn_printf(cn
, "%d",
256 task_tgid_nr(current
));
259 err
= cn_printf(cn
, "%d",
260 task_pid_vnr(current
));
263 err
= cn_printf(cn
, "%d",
264 task_pid_nr(current
));
268 err
= cn_printf(cn
, "%u",
269 from_kuid(&init_user_ns
,
274 err
= cn_printf(cn
, "%u",
275 from_kgid(&init_user_ns
,
279 err
= cn_printf(cn
, "%d",
280 __get_dumpable(cprm
->mm_flags
));
282 /* signal that caused the coredump */
284 err
= cn_printf(cn
, "%d",
285 cprm
->siginfo
->si_signo
);
287 /* UNIX time of coredump */
291 time
= ktime_get_real_seconds();
292 err
= cn_printf(cn
, "%lld", time
);
298 err
= cn_esc_printf(cn
, "%s",
299 utsname()->nodename
);
304 err
= cn_esc_printf(cn
, "%s", current
->comm
);
307 err
= cn_print_exe_file(cn
);
309 /* core limit size */
311 err
= cn_printf(cn
, "%lu",
312 rlimit(RLIMIT_CORE
));
325 /* Backward compatibility with core_uses_pid:
327 * If core_pattern does not include a %p (as is the default)
328 * and core_uses_pid is set, then .%pid will be appended to
329 * the filename. Do not do this for piped commands. */
330 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
331 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
338 static int zap_process(struct task_struct
*start
, int exit_code
, int flags
)
340 struct task_struct
*t
;
343 /* ignore all signals except SIGKILL, see prepare_signal() */
344 start
->signal
->flags
= SIGNAL_GROUP_COREDUMP
| flags
;
345 start
->signal
->group_exit_code
= exit_code
;
346 start
->signal
->group_stop_count
= 0;
348 for_each_thread(start
, t
) {
349 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
350 if (t
!= current
&& t
->mm
) {
351 sigaddset(&t
->pending
.signal
, SIGKILL
);
352 signal_wake_up(t
, 1);
360 static int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
361 struct core_state
*core_state
, int exit_code
)
363 struct task_struct
*g
, *p
;
367 spin_lock_irq(&tsk
->sighand
->siglock
);
368 if (!signal_group_exit(tsk
->signal
)) {
369 mm
->core_state
= core_state
;
370 tsk
->signal
->group_exit_task
= tsk
;
371 nr
= zap_process(tsk
, exit_code
, 0);
372 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
374 spin_unlock_irq(&tsk
->sighand
->siglock
);
375 if (unlikely(nr
< 0))
378 tsk
->flags
|= PF_DUMPCORE
;
379 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
382 * We should find and kill all tasks which use this mm, and we should
383 * count them correctly into ->nr_threads. We don't take tasklist
384 * lock, but this is safe wrt:
387 * None of sub-threads can fork after zap_process(leader). All
388 * processes which were created before this point should be
389 * visible to zap_threads() because copy_process() adds the new
390 * process to the tail of init_task.tasks list, and lock/unlock
391 * of ->siglock provides a memory barrier.
394 * The caller holds mm->mmap_sem. This means that the task which
395 * uses this mm can't pass exit_mm(), so it can't exit or clear
399 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
400 * we must see either old or new leader, this does not matter.
401 * However, it can change p->sighand, so lock_task_sighand(p)
402 * must be used. Since p->mm != NULL and we hold ->mmap_sem
405 * Note also that "g" can be the old leader with ->mm == NULL
406 * and already unhashed and thus removed from ->thread_group.
407 * This is OK, __unhash_process()->list_del_rcu() does not
408 * clear the ->next pointer, we will find the new leader via
412 for_each_process(g
) {
413 if (g
== tsk
->group_leader
)
415 if (g
->flags
& PF_KTHREAD
)
418 for_each_thread(g
, p
) {
419 if (unlikely(!p
->mm
))
421 if (unlikely(p
->mm
== mm
)) {
422 lock_task_sighand(p
, &flags
);
423 nr
+= zap_process(p
, exit_code
,
425 unlock_task_sighand(p
, &flags
);
432 atomic_set(&core_state
->nr_threads
, nr
);
436 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
438 struct task_struct
*tsk
= current
;
439 struct mm_struct
*mm
= tsk
->mm
;
440 int core_waiters
= -EBUSY
;
442 init_completion(&core_state
->startup
);
443 core_state
->dumper
.task
= tsk
;
444 core_state
->dumper
.next
= NULL
;
446 if (down_write_killable(&mm
->mmap_sem
))
450 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
451 up_write(&mm
->mmap_sem
);
453 if (core_waiters
> 0) {
454 struct core_thread
*ptr
;
456 freezer_do_not_count();
457 wait_for_completion(&core_state
->startup
);
460 * Wait for all the threads to become inactive, so that
461 * all the thread context (extended register state, like
462 * fpu etc) gets copied to the memory.
464 ptr
= core_state
->dumper
.next
;
465 while (ptr
!= NULL
) {
466 wait_task_inactive(ptr
->task
, 0);
474 static void coredump_finish(struct mm_struct
*mm
, bool core_dumped
)
476 struct core_thread
*curr
, *next
;
477 struct task_struct
*task
;
479 spin_lock_irq(¤t
->sighand
->siglock
);
480 if (core_dumped
&& !__fatal_signal_pending(current
))
481 current
->signal
->group_exit_code
|= 0x80;
482 current
->signal
->group_exit_task
= NULL
;
483 current
->signal
->flags
= SIGNAL_GROUP_EXIT
;
484 spin_unlock_irq(¤t
->sighand
->siglock
);
486 next
= mm
->core_state
->dumper
.next
;
487 while ((curr
= next
) != NULL
) {
491 * see exit_mm(), curr->task must not see
492 * ->task == NULL before we read ->next.
496 wake_up_process(task
);
499 mm
->core_state
= NULL
;
502 static bool dump_interrupted(void)
505 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
506 * can do try_to_freeze() and check __fatal_signal_pending(),
507 * but then we need to teach dump_write() to restart and clear
510 return signal_pending(current
);
513 static void wait_for_dump_helpers(struct file
*file
)
515 struct pipe_inode_info
*pipe
= file
->private_data
;
520 wake_up_interruptible_sync(&pipe
->wait
);
521 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
525 * We actually want wait_event_freezable() but then we need
526 * to clear TIF_SIGPENDING and improve dump_interrupted().
528 wait_event_interruptible(pipe
->wait
, pipe
->readers
== 1);
538 * helper function to customize the process used
539 * to collect the core in userspace. Specifically
540 * it sets up a pipe and installs it as fd 0 (stdin)
541 * for the process. Returns 0 on success, or
542 * PTR_ERR on failure.
543 * Note that it also sets the core limit to 1. This
544 * is a special value that we use to trap recursive
547 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
549 struct file
*files
[2];
550 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
551 int err
= create_pipe_files(files
, 0);
557 err
= replace_fd(0, files
[0], 0);
559 /* and disallow core files too */
560 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
565 void do_coredump(const kernel_siginfo_t
*siginfo
)
567 struct core_state core_state
;
569 struct mm_struct
*mm
= current
->mm
;
570 struct linux_binfmt
* binfmt
;
571 const struct cred
*old_cred
;
577 struct files_struct
*displaced
;
578 /* require nonrelative corefile path and be extra careful */
579 bool need_suid_safe
= false;
580 bool core_dumped
= false;
581 static atomic_t core_dump_count
= ATOMIC_INIT(0);
582 struct coredump_params cprm
= {
584 .regs
= signal_pt_regs(),
585 .limit
= rlimit(RLIMIT_CORE
),
587 * We must use the same mm->flags while dumping core to avoid
588 * inconsistency of bit flags, since this flag is not protected
591 .mm_flags
= mm
->flags
,
594 audit_core_dumps(siginfo
->si_signo
);
597 if (!binfmt
|| !binfmt
->core_dump
)
599 if (!__get_dumpable(cprm
.mm_flags
))
602 cred
= prepare_creds();
606 * We cannot trust fsuid as being the "true" uid of the process
607 * nor do we know its entire history. We only know it was tainted
608 * so we dump it as root in mode 2, and only into a controlled
609 * environment (pipe handler or fully qualified path).
611 if (__get_dumpable(cprm
.mm_flags
) == SUID_DUMP_ROOT
) {
612 /* Setuid core dump mode */
613 cred
->fsuid
= GLOBAL_ROOT_UID
; /* Dump root private */
614 need_suid_safe
= true;
617 retval
= coredump_wait(siginfo
->si_signo
, &core_state
);
621 old_cred
= override_creds(cred
);
623 ispipe
= format_corename(&cn
, &cprm
, &argv
, &argc
);
629 struct subprocess_info
*sub_info
;
632 printk(KERN_WARNING
"format_corename failed\n");
633 printk(KERN_WARNING
"Aborting core\n");
637 if (cprm
.limit
== 1) {
638 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
640 * Normally core limits are irrelevant to pipes, since
641 * we're not writing to the file system, but we use
642 * cprm.limit of 1 here as a special value, this is a
643 * consistent way to catch recursive crashes.
644 * We can still crash if the core_pattern binary sets
645 * RLIM_CORE = !1, but it runs as root, and can do
646 * lots of stupid things.
648 * Note that we use task_tgid_vnr here to grab the pid
649 * of the process group leader. That way we get the
650 * right pid if a thread in a multi-threaded
651 * core_pattern process dies.
654 "Process %d(%s) has RLIMIT_CORE set to 1\n",
655 task_tgid_vnr(current
), current
->comm
);
656 printk(KERN_WARNING
"Aborting core\n");
659 cprm
.limit
= RLIM_INFINITY
;
661 dump_count
= atomic_inc_return(&core_dump_count
);
662 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
663 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
664 task_tgid_vnr(current
), current
->comm
);
665 printk(KERN_WARNING
"Skipping core dump\n");
669 helper_argv
= kmalloc_array(argc
+ 1, sizeof(*helper_argv
),
672 printk(KERN_WARNING
"%s failed to allocate memory\n",
676 for (argi
= 0; argi
< argc
; argi
++)
677 helper_argv
[argi
] = cn
.corename
+ argv
[argi
];
678 helper_argv
[argi
] = NULL
;
681 sub_info
= call_usermodehelper_setup(helper_argv
[0],
682 helper_argv
, NULL
, GFP_KERNEL
,
683 umh_pipe_setup
, NULL
, &cprm
);
685 retval
= call_usermodehelper_exec(sub_info
,
690 printk(KERN_INFO
"Core dump to |%s pipe failed\n",
696 int open_flags
= O_CREAT
| O_RDWR
| O_NOFOLLOW
|
697 O_LARGEFILE
| O_EXCL
;
699 if (cprm
.limit
< binfmt
->min_coredump
)
702 if (need_suid_safe
&& cn
.corename
[0] != '/') {
703 printk(KERN_WARNING
"Pid %d(%s) can only dump core "\
704 "to fully qualified path!\n",
705 task_tgid_vnr(current
), current
->comm
);
706 printk(KERN_WARNING
"Skipping core dump\n");
711 * Unlink the file if it exists unless this is a SUID
712 * binary - in that case, we're running around with root
713 * privs and don't want to unlink another user's coredump.
715 if (!need_suid_safe
) {
717 * If it doesn't exist, that's fine. If there's some
718 * other problem, we'll catch it at the filp_open().
720 do_unlinkat(AT_FDCWD
, getname_kernel(cn
.corename
));
724 * There is a race between unlinking and creating the
725 * file, but if that causes an EEXIST here, that's
726 * fine - another process raced with us while creating
727 * the corefile, and the other process won. To userspace,
728 * what matters is that at least one of the two processes
729 * writes its coredump successfully, not which one.
731 if (need_suid_safe
) {
733 * Using user namespaces, normal user tasks can change
734 * their current->fs->root to point to arbitrary
735 * directories. Since the intention of the "only dump
736 * with a fully qualified path" rule is to control where
737 * coredumps may be placed using root privileges,
738 * current->fs->root must not be used. Instead, use the
739 * root directory of init_task.
743 task_lock(&init_task
);
744 get_fs_root(init_task
.fs
, &root
);
745 task_unlock(&init_task
);
746 cprm
.file
= file_open_root(root
.dentry
, root
.mnt
,
747 cn
.corename
, open_flags
, 0600);
750 cprm
.file
= filp_open(cn
.corename
, open_flags
, 0600);
752 if (IS_ERR(cprm
.file
))
755 inode
= file_inode(cprm
.file
);
756 if (inode
->i_nlink
> 1)
758 if (d_unhashed(cprm
.file
->f_path
.dentry
))
761 * AK: actually i see no reason to not allow this for named
762 * pipes etc, but keep the previous behaviour for now.
764 if (!S_ISREG(inode
->i_mode
))
767 * Don't dump core if the filesystem changed owner or mode
768 * of the file during file creation. This is an issue when
769 * a process dumps core while its cwd is e.g. on a vfat
772 if (!uid_eq(inode
->i_uid
, current_fsuid()))
774 if ((inode
->i_mode
& 0677) != 0600)
776 if (!(cprm
.file
->f_mode
& FMODE_CAN_WRITE
))
778 if (do_truncate(cprm
.file
->f_path
.dentry
, 0, 0, cprm
.file
))
782 /* get us an unshared descriptor table; almost always a no-op */
783 retval
= unshare_files(&displaced
);
787 put_files_struct(displaced
);
788 if (!dump_interrupted()) {
789 file_start_write(cprm
.file
);
790 core_dumped
= binfmt
->core_dump(&cprm
);
791 file_end_write(cprm
.file
);
793 if (ispipe
&& core_pipe_limit
)
794 wait_for_dump_helpers(cprm
.file
);
797 filp_close(cprm
.file
, NULL
);
800 atomic_dec(&core_dump_count
);
804 coredump_finish(mm
, core_dumped
);
805 revert_creds(old_cred
);
813 * Core dumping helper functions. These are the only things you should
814 * do on a core-file: use only these functions to write out all the
817 int dump_emit(struct coredump_params
*cprm
, const void *addr
, int nr
)
819 struct file
*file
= cprm
->file
;
820 loff_t pos
= file
->f_pos
;
822 if (cprm
->written
+ nr
> cprm
->limit
)
825 if (dump_interrupted())
827 n
= __kernel_write(file
, addr
, nr
, &pos
);
837 EXPORT_SYMBOL(dump_emit
);
839 int dump_skip(struct coredump_params
*cprm
, size_t nr
)
841 static char zeroes
[PAGE_SIZE
];
842 struct file
*file
= cprm
->file
;
843 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
844 if (dump_interrupted() ||
845 file
->f_op
->llseek(file
, nr
, SEEK_CUR
) < 0)
850 while (nr
> PAGE_SIZE
) {
851 if (!dump_emit(cprm
, zeroes
, PAGE_SIZE
))
855 return dump_emit(cprm
, zeroes
, nr
);
858 EXPORT_SYMBOL(dump_skip
);
860 int dump_align(struct coredump_params
*cprm
, int align
)
862 unsigned mod
= cprm
->pos
& (align
- 1);
863 if (align
& (align
- 1))
865 return mod
? dump_skip(cprm
, align
- mod
) : 1;
867 EXPORT_SYMBOL(dump_align
);
870 * Ensures that file size is big enough to contain the current file
871 * postion. This prevents gdb from complaining about a truncated file
872 * if the last "write" to the file was dump_skip.
874 void dump_truncate(struct coredump_params
*cprm
)
876 struct file
*file
= cprm
->file
;
879 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
880 offset
= file
->f_op
->llseek(file
, 0, SEEK_CUR
);
881 if (i_size_read(file
->f_mapping
->host
) < offset
)
882 do_truncate(file
->f_path
.dentry
, offset
, 0, file
);
885 EXPORT_SYMBOL(dump_truncate
);