]>
git.proxmox.com Git - mirror_ubuntu-eoan-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/string.h>
11 #include <linux/init.h>
12 #include <linux/pagemap.h>
13 #include <linux/perf_event.h>
14 #include <linux/highmem.h>
15 #include <linux/spinlock.h>
16 #include <linux/key.h>
17 #include <linux/personality.h>
18 #include <linux/binfmts.h>
19 #include <linux/coredump.h>
20 #include <linux/sched/coredump.h>
21 #include <linux/sched/signal.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/utsname.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/module.h>
26 #include <linux/namei.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
30 #include <linux/tsacct_kern.h>
31 #include <linux/cn_proc.h>
32 #include <linux/audit.h>
33 #include <linux/tracehook.h>
34 #include <linux/kmod.h>
35 #include <linux/fsnotify.h>
36 #include <linux/fs_struct.h>
37 #include <linux/pipe_fs_i.h>
38 #include <linux/oom.h>
39 #include <linux/compat.h>
41 #include <linux/path.h>
42 #include <linux/timekeeping.h>
44 #include <linux/uaccess.h>
45 #include <asm/mmu_context.h>
49 #include <trace/events/task.h>
52 #include <trace/events/sched.h>
55 unsigned int core_pipe_limit
;
56 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
57 static int core_name_size
= CORENAME_MAX_SIZE
;
64 /* The maximal length of core_pattern is also specified in sysctl.c */
66 static int expand_corename(struct core_name
*cn
, int size
)
68 char *corename
= krealloc(cn
->corename
, size
, GFP_KERNEL
);
73 if (size
> core_name_size
) /* racy but harmless */
74 core_name_size
= size
;
76 cn
->size
= ksize(corename
);
77 cn
->corename
= corename
;
81 static __printf(2, 0) int cn_vprintf(struct core_name
*cn
, const char *fmt
,
88 free
= cn
->size
- cn
->used
;
90 va_copy(arg_copy
, arg
);
91 need
= vsnprintf(cn
->corename
+ cn
->used
, free
, fmt
, arg_copy
);
99 if (!expand_corename(cn
, cn
->size
+ need
- free
+ 1))
105 static __printf(2, 3) int cn_printf(struct core_name
*cn
, const char *fmt
, ...)
111 ret
= cn_vprintf(cn
, fmt
, arg
);
117 static __printf(2, 3)
118 int cn_esc_printf(struct core_name
*cn
, const char *fmt
, ...)
125 ret
= cn_vprintf(cn
, fmt
, arg
);
130 * Ensure that this coredump name component can't cause the
131 * resulting corefile path to consist of a ".." or ".".
133 if ((cn
->used
- cur
== 1 && cn
->corename
[cur
] == '.') ||
134 (cn
->used
- cur
== 2 && cn
->corename
[cur
] == '.'
135 && cn
->corename
[cur
+1] == '.'))
136 cn
->corename
[cur
] = '!';
139 * Empty names are fishy and could be used to create a "//" in a
140 * corefile name, causing the coredump to happen one directory
141 * level too high. Enforce that all components of the core
142 * pattern are at least one character long.
145 ret
= cn_printf(cn
, "!");
148 for (; cur
< cn
->used
; ++cur
) {
149 if (cn
->corename
[cur
] == '/')
150 cn
->corename
[cur
] = '!';
155 static int cn_print_exe_file(struct core_name
*cn
)
157 struct file
*exe_file
;
158 char *pathbuf
, *path
;
161 exe_file
= get_mm_exe_file(current
->mm
);
163 return cn_esc_printf(cn
, "%s (path unknown)", current
->comm
);
165 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
171 path
= file_path(exe_file
, pathbuf
, PATH_MAX
);
177 ret
= cn_esc_printf(cn
, "%s", path
);
186 /* format_corename will inspect the pattern parameter, and output a
187 * name into corename, which must have space for at least
188 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
190 static int format_corename(struct core_name
*cn
, struct coredump_params
*cprm
)
192 const struct cred
*cred
= current_cred();
193 const char *pat_ptr
= core_pattern
;
194 int ispipe
= (*pat_ptr
== '|');
195 int pid_in_pattern
= 0;
200 if (expand_corename(cn
, core_name_size
))
202 cn
->corename
[0] = '\0';
207 /* Repeat as long as we have more pattern to process and more output
210 if (*pat_ptr
!= '%') {
211 err
= cn_printf(cn
, "%c", *pat_ptr
++);
213 switch (*++pat_ptr
) {
214 /* single % at the end, drop that */
217 /* Double percent, output one percent */
219 err
= cn_printf(cn
, "%c", '%');
224 err
= cn_printf(cn
, "%d",
225 task_tgid_vnr(current
));
229 err
= cn_printf(cn
, "%d",
230 task_tgid_nr(current
));
233 err
= cn_printf(cn
, "%d",
234 task_pid_vnr(current
));
237 err
= cn_printf(cn
, "%d",
238 task_pid_nr(current
));
242 err
= cn_printf(cn
, "%u",
243 from_kuid(&init_user_ns
,
248 err
= cn_printf(cn
, "%u",
249 from_kgid(&init_user_ns
,
253 err
= cn_printf(cn
, "%d",
254 __get_dumpable(cprm
->mm_flags
));
256 /* signal that caused the coredump */
258 err
= cn_printf(cn
, "%d",
259 cprm
->siginfo
->si_signo
);
261 /* UNIX time of coredump */
265 time
= ktime_get_real_seconds();
266 err
= cn_printf(cn
, "%lld", time
);
272 err
= cn_esc_printf(cn
, "%s",
273 utsname()->nodename
);
278 err
= cn_esc_printf(cn
, "%s", current
->comm
);
281 err
= cn_print_exe_file(cn
);
283 /* core limit size */
285 err
= cn_printf(cn
, "%lu",
286 rlimit(RLIMIT_CORE
));
299 /* Backward compatibility with core_uses_pid:
301 * If core_pattern does not include a %p (as is the default)
302 * and core_uses_pid is set, then .%pid will be appended to
303 * the filename. Do not do this for piped commands. */
304 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
305 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
312 static int zap_process(struct task_struct
*start
, int exit_code
, int flags
)
314 struct task_struct
*t
;
317 /* ignore all signals except SIGKILL, see prepare_signal() */
318 start
->signal
->flags
= SIGNAL_GROUP_COREDUMP
| flags
;
319 start
->signal
->group_exit_code
= exit_code
;
320 start
->signal
->group_stop_count
= 0;
322 for_each_thread(start
, t
) {
323 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
324 if (t
!= current
&& t
->mm
) {
325 sigaddset(&t
->pending
.signal
, SIGKILL
);
326 signal_wake_up(t
, 1);
334 static int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
335 struct core_state
*core_state
, int exit_code
)
337 struct task_struct
*g
, *p
;
341 spin_lock_irq(&tsk
->sighand
->siglock
);
342 if (!signal_group_exit(tsk
->signal
)) {
343 mm
->core_state
= core_state
;
344 tsk
->signal
->group_exit_task
= tsk
;
345 nr
= zap_process(tsk
, exit_code
, 0);
346 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
348 spin_unlock_irq(&tsk
->sighand
->siglock
);
349 if (unlikely(nr
< 0))
352 tsk
->flags
|= PF_DUMPCORE
;
353 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
356 * We should find and kill all tasks which use this mm, and we should
357 * count them correctly into ->nr_threads. We don't take tasklist
358 * lock, but this is safe wrt:
361 * None of sub-threads can fork after zap_process(leader). All
362 * processes which were created before this point should be
363 * visible to zap_threads() because copy_process() adds the new
364 * process to the tail of init_task.tasks list, and lock/unlock
365 * of ->siglock provides a memory barrier.
368 * The caller holds mm->mmap_sem. This means that the task which
369 * uses this mm can't pass exit_mm(), so it can't exit or clear
373 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
374 * we must see either old or new leader, this does not matter.
375 * However, it can change p->sighand, so lock_task_sighand(p)
376 * must be used. Since p->mm != NULL and we hold ->mmap_sem
379 * Note also that "g" can be the old leader with ->mm == NULL
380 * and already unhashed and thus removed from ->thread_group.
381 * This is OK, __unhash_process()->list_del_rcu() does not
382 * clear the ->next pointer, we will find the new leader via
386 for_each_process(g
) {
387 if (g
== tsk
->group_leader
)
389 if (g
->flags
& PF_KTHREAD
)
392 for_each_thread(g
, p
) {
393 if (unlikely(!p
->mm
))
395 if (unlikely(p
->mm
== mm
)) {
396 lock_task_sighand(p
, &flags
);
397 nr
+= zap_process(p
, exit_code
,
399 unlock_task_sighand(p
, &flags
);
406 atomic_set(&core_state
->nr_threads
, nr
);
410 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
412 struct task_struct
*tsk
= current
;
413 struct mm_struct
*mm
= tsk
->mm
;
414 int core_waiters
= -EBUSY
;
416 init_completion(&core_state
->startup
);
417 core_state
->dumper
.task
= tsk
;
418 core_state
->dumper
.next
= NULL
;
420 if (down_write_killable(&mm
->mmap_sem
))
424 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
425 up_write(&mm
->mmap_sem
);
427 if (core_waiters
> 0) {
428 struct core_thread
*ptr
;
430 freezer_do_not_count();
431 wait_for_completion(&core_state
->startup
);
434 * Wait for all the threads to become inactive, so that
435 * all the thread context (extended register state, like
436 * fpu etc) gets copied to the memory.
438 ptr
= core_state
->dumper
.next
;
439 while (ptr
!= NULL
) {
440 wait_task_inactive(ptr
->task
, 0);
448 static void coredump_finish(struct mm_struct
*mm
, bool core_dumped
)
450 struct core_thread
*curr
, *next
;
451 struct task_struct
*task
;
453 spin_lock_irq(¤t
->sighand
->siglock
);
454 if (core_dumped
&& !__fatal_signal_pending(current
))
455 current
->signal
->group_exit_code
|= 0x80;
456 current
->signal
->group_exit_task
= NULL
;
457 current
->signal
->flags
= SIGNAL_GROUP_EXIT
;
458 spin_unlock_irq(¤t
->sighand
->siglock
);
460 next
= mm
->core_state
->dumper
.next
;
461 while ((curr
= next
) != NULL
) {
465 * see exit_mm(), curr->task must not see
466 * ->task == NULL before we read ->next.
470 wake_up_process(task
);
473 mm
->core_state
= NULL
;
476 static bool dump_interrupted(void)
479 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
480 * can do try_to_freeze() and check __fatal_signal_pending(),
481 * but then we need to teach dump_write() to restart and clear
484 return signal_pending(current
);
487 static void wait_for_dump_helpers(struct file
*file
)
489 struct pipe_inode_info
*pipe
= file
->private_data
;
494 wake_up_interruptible_sync(&pipe
->wait
);
495 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
499 * We actually want wait_event_freezable() but then we need
500 * to clear TIF_SIGPENDING and improve dump_interrupted().
502 wait_event_interruptible(pipe
->wait
, pipe
->readers
== 1);
512 * helper function to customize the process used
513 * to collect the core in userspace. Specifically
514 * it sets up a pipe and installs it as fd 0 (stdin)
515 * for the process. Returns 0 on success, or
516 * PTR_ERR on failure.
517 * Note that it also sets the core limit to 1. This
518 * is a special value that we use to trap recursive
521 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
523 struct file
*files
[2];
524 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
525 int err
= create_pipe_files(files
, 0);
531 err
= replace_fd(0, files
[0], 0);
533 /* and disallow core files too */
534 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
539 void do_coredump(const kernel_siginfo_t
*siginfo
)
541 struct core_state core_state
;
543 struct mm_struct
*mm
= current
->mm
;
544 struct linux_binfmt
* binfmt
;
545 const struct cred
*old_cred
;
549 struct files_struct
*displaced
;
550 /* require nonrelative corefile path and be extra careful */
551 bool need_suid_safe
= false;
552 bool core_dumped
= false;
553 static atomic_t core_dump_count
= ATOMIC_INIT(0);
554 struct coredump_params cprm
= {
556 .regs
= signal_pt_regs(),
557 .limit
= rlimit(RLIMIT_CORE
),
559 * We must use the same mm->flags while dumping core to avoid
560 * inconsistency of bit flags, since this flag is not protected
563 .mm_flags
= mm
->flags
,
566 audit_core_dumps(siginfo
->si_signo
);
569 if (!binfmt
|| !binfmt
->core_dump
)
571 if (!__get_dumpable(cprm
.mm_flags
))
574 cred
= prepare_creds();
578 * We cannot trust fsuid as being the "true" uid of the process
579 * nor do we know its entire history. We only know it was tainted
580 * so we dump it as root in mode 2, and only into a controlled
581 * environment (pipe handler or fully qualified path).
583 if (__get_dumpable(cprm
.mm_flags
) == SUID_DUMP_ROOT
) {
584 /* Setuid core dump mode */
585 cred
->fsuid
= GLOBAL_ROOT_UID
; /* Dump root private */
586 need_suid_safe
= true;
589 retval
= coredump_wait(siginfo
->si_signo
, &core_state
);
593 old_cred
= override_creds(cred
);
595 ispipe
= format_corename(&cn
, &cprm
);
600 struct subprocess_info
*sub_info
;
603 printk(KERN_WARNING
"format_corename failed\n");
604 printk(KERN_WARNING
"Aborting core\n");
608 if (cprm
.limit
== 1) {
609 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
611 * Normally core limits are irrelevant to pipes, since
612 * we're not writing to the file system, but we use
613 * cprm.limit of 1 here as a special value, this is a
614 * consistent way to catch recursive crashes.
615 * We can still crash if the core_pattern binary sets
616 * RLIM_CORE = !1, but it runs as root, and can do
617 * lots of stupid things.
619 * Note that we use task_tgid_vnr here to grab the pid
620 * of the process group leader. That way we get the
621 * right pid if a thread in a multi-threaded
622 * core_pattern process dies.
625 "Process %d(%s) has RLIMIT_CORE set to 1\n",
626 task_tgid_vnr(current
), current
->comm
);
627 printk(KERN_WARNING
"Aborting core\n");
630 cprm
.limit
= RLIM_INFINITY
;
632 dump_count
= atomic_inc_return(&core_dump_count
);
633 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
634 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
635 task_tgid_vnr(current
), current
->comm
);
636 printk(KERN_WARNING
"Skipping core dump\n");
640 helper_argv
= argv_split(GFP_KERNEL
, cn
.corename
, NULL
);
642 printk(KERN_WARNING
"%s failed to allocate memory\n",
648 sub_info
= call_usermodehelper_setup(helper_argv
[0],
649 helper_argv
, NULL
, GFP_KERNEL
,
650 umh_pipe_setup
, NULL
, &cprm
);
652 retval
= call_usermodehelper_exec(sub_info
,
655 argv_free(helper_argv
);
657 printk(KERN_INFO
"Core dump to |%s pipe failed\n",
663 int open_flags
= O_CREAT
| O_RDWR
| O_NOFOLLOW
|
664 O_LARGEFILE
| O_EXCL
;
666 if (cprm
.limit
< binfmt
->min_coredump
)
669 if (need_suid_safe
&& cn
.corename
[0] != '/') {
670 printk(KERN_WARNING
"Pid %d(%s) can only dump core "\
671 "to fully qualified path!\n",
672 task_tgid_vnr(current
), current
->comm
);
673 printk(KERN_WARNING
"Skipping core dump\n");
678 * Unlink the file if it exists unless this is a SUID
679 * binary - in that case, we're running around with root
680 * privs and don't want to unlink another user's coredump.
682 if (!need_suid_safe
) {
684 * If it doesn't exist, that's fine. If there's some
685 * other problem, we'll catch it at the filp_open().
687 do_unlinkat(AT_FDCWD
, getname_kernel(cn
.corename
));
691 * There is a race between unlinking and creating the
692 * file, but if that causes an EEXIST here, that's
693 * fine - another process raced with us while creating
694 * the corefile, and the other process won. To userspace,
695 * what matters is that at least one of the two processes
696 * writes its coredump successfully, not which one.
698 if (need_suid_safe
) {
700 * Using user namespaces, normal user tasks can change
701 * their current->fs->root to point to arbitrary
702 * directories. Since the intention of the "only dump
703 * with a fully qualified path" rule is to control where
704 * coredumps may be placed using root privileges,
705 * current->fs->root must not be used. Instead, use the
706 * root directory of init_task.
710 task_lock(&init_task
);
711 get_fs_root(init_task
.fs
, &root
);
712 task_unlock(&init_task
);
713 cprm
.file
= file_open_root(root
.dentry
, root
.mnt
,
714 cn
.corename
, open_flags
, 0600);
717 cprm
.file
= filp_open(cn
.corename
, open_flags
, 0600);
719 if (IS_ERR(cprm
.file
))
722 inode
= file_inode(cprm
.file
);
723 if (inode
->i_nlink
> 1)
725 if (d_unhashed(cprm
.file
->f_path
.dentry
))
728 * AK: actually i see no reason to not allow this for named
729 * pipes etc, but keep the previous behaviour for now.
731 if (!S_ISREG(inode
->i_mode
))
734 * Don't dump core if the filesystem changed owner or mode
735 * of the file during file creation. This is an issue when
736 * a process dumps core while its cwd is e.g. on a vfat
739 if (!uid_eq(inode
->i_uid
, current_fsuid()))
741 if ((inode
->i_mode
& 0677) != 0600)
743 if (!(cprm
.file
->f_mode
& FMODE_CAN_WRITE
))
745 if (do_truncate(cprm
.file
->f_path
.dentry
, 0, 0, cprm
.file
))
749 /* get us an unshared descriptor table; almost always a no-op */
750 retval
= unshare_files(&displaced
);
754 put_files_struct(displaced
);
755 if (!dump_interrupted()) {
756 file_start_write(cprm
.file
);
757 core_dumped
= binfmt
->core_dump(&cprm
);
758 file_end_write(cprm
.file
);
760 if (ispipe
&& core_pipe_limit
)
761 wait_for_dump_helpers(cprm
.file
);
764 filp_close(cprm
.file
, NULL
);
767 atomic_dec(&core_dump_count
);
770 coredump_finish(mm
, core_dumped
);
771 revert_creds(old_cred
);
779 * Core dumping helper functions. These are the only things you should
780 * do on a core-file: use only these functions to write out all the
783 int dump_emit(struct coredump_params
*cprm
, const void *addr
, int nr
)
785 struct file
*file
= cprm
->file
;
786 loff_t pos
= file
->f_pos
;
788 if (cprm
->written
+ nr
> cprm
->limit
)
791 if (dump_interrupted())
793 n
= __kernel_write(file
, addr
, nr
, &pos
);
803 EXPORT_SYMBOL(dump_emit
);
805 int dump_skip(struct coredump_params
*cprm
, size_t nr
)
807 static char zeroes
[PAGE_SIZE
];
808 struct file
*file
= cprm
->file
;
809 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
810 if (dump_interrupted() ||
811 file
->f_op
->llseek(file
, nr
, SEEK_CUR
) < 0)
816 while (nr
> PAGE_SIZE
) {
817 if (!dump_emit(cprm
, zeroes
, PAGE_SIZE
))
821 return dump_emit(cprm
, zeroes
, nr
);
824 EXPORT_SYMBOL(dump_skip
);
826 int dump_align(struct coredump_params
*cprm
, int align
)
828 unsigned mod
= cprm
->pos
& (align
- 1);
829 if (align
& (align
- 1))
831 return mod
? dump_skip(cprm
, align
- mod
) : 1;
833 EXPORT_SYMBOL(dump_align
);
836 * Ensures that file size is big enough to contain the current file
837 * postion. This prevents gdb from complaining about a truncated file
838 * if the last "write" to the file was dump_skip.
840 void dump_truncate(struct coredump_params
*cprm
)
842 struct file
*file
= cprm
->file
;
845 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
846 offset
= file
->f_op
->llseek(file
, 0, SEEK_CUR
);
847 if (i_size_read(file
->f_mapping
->host
) < offset
)
848 do_truncate(file
->f_path
.dentry
, offset
, 0, file
);
851 EXPORT_SYMBOL(dump_truncate
);