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>
44 #include <linux/elf.h>
46 #include <linux/uaccess.h>
47 #include <asm/mmu_context.h>
51 #include <trace/events/task.h>
54 #include <trace/events/sched.h>
56 static bool dump_vma_snapshot(struct coredump_params
*cprm
);
57 static void free_vma_snapshot(struct coredump_params
*cprm
);
60 unsigned int core_pipe_limit
;
61 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
62 static int core_name_size
= CORENAME_MAX_SIZE
;
69 /* The maximal length of core_pattern is also specified in sysctl.c */
71 static int expand_corename(struct core_name
*cn
, int size
)
73 char *corename
= krealloc(cn
->corename
, size
, GFP_KERNEL
);
78 if (size
> core_name_size
) /* racy but harmless */
79 core_name_size
= size
;
81 cn
->size
= ksize(corename
);
82 cn
->corename
= corename
;
86 static __printf(2, 0) int cn_vprintf(struct core_name
*cn
, const char *fmt
,
93 free
= cn
->size
- cn
->used
;
95 va_copy(arg_copy
, arg
);
96 need
= vsnprintf(cn
->corename
+ cn
->used
, free
, fmt
, arg_copy
);
104 if (!expand_corename(cn
, cn
->size
+ need
- free
+ 1))
110 static __printf(2, 3) int cn_printf(struct core_name
*cn
, const char *fmt
, ...)
116 ret
= cn_vprintf(cn
, fmt
, arg
);
122 static __printf(2, 3)
123 int cn_esc_printf(struct core_name
*cn
, const char *fmt
, ...)
130 ret
= cn_vprintf(cn
, fmt
, arg
);
135 * Ensure that this coredump name component can't cause the
136 * resulting corefile path to consist of a ".." or ".".
138 if ((cn
->used
- cur
== 1 && cn
->corename
[cur
] == '.') ||
139 (cn
->used
- cur
== 2 && cn
->corename
[cur
] == '.'
140 && cn
->corename
[cur
+1] == '.'))
141 cn
->corename
[cur
] = '!';
144 * Empty names are fishy and could be used to create a "//" in a
145 * corefile name, causing the coredump to happen one directory
146 * level too high. Enforce that all components of the core
147 * pattern are at least one character long.
150 ret
= cn_printf(cn
, "!");
153 for (; cur
< cn
->used
; ++cur
) {
154 if (cn
->corename
[cur
] == '/')
155 cn
->corename
[cur
] = '!';
160 static int cn_print_exe_file(struct core_name
*cn
, bool name_only
)
162 struct file
*exe_file
;
163 char *pathbuf
, *path
, *ptr
;
166 exe_file
= get_mm_exe_file(current
->mm
);
168 return cn_esc_printf(cn
, "%s (path unknown)", current
->comm
);
170 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
176 path
= file_path(exe_file
, pathbuf
, PATH_MAX
);
183 ptr
= strrchr(path
, '/');
187 ret
= cn_esc_printf(cn
, "%s", path
);
196 /* format_corename will inspect the pattern parameter, and output a
197 * name into corename, which must have space for at least
198 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
200 static int format_corename(struct core_name
*cn
, struct coredump_params
*cprm
,
201 size_t **argv
, int *argc
)
203 const struct cred
*cred
= current_cred();
204 const char *pat_ptr
= core_pattern
;
205 int ispipe
= (*pat_ptr
== '|');
206 bool was_space
= false;
207 int pid_in_pattern
= 0;
212 if (expand_corename(cn
, core_name_size
))
214 cn
->corename
[0] = '\0';
217 int argvs
= sizeof(core_pattern
) / 2;
218 (*argv
) = kmalloc_array(argvs
, sizeof(**argv
), GFP_KERNEL
);
221 (*argv
)[(*argc
)++] = 0;
227 /* Repeat as long as we have more pattern to process and more output
231 * Split on spaces before doing template expansion so that
232 * %e and %E don't get split if they have spaces in them
235 if (isspace(*pat_ptr
)) {
240 } else if (was_space
) {
242 err
= cn_printf(cn
, "%c", '\0');
245 (*argv
)[(*argc
)++] = cn
->used
;
248 if (*pat_ptr
!= '%') {
249 err
= cn_printf(cn
, "%c", *pat_ptr
++);
251 switch (*++pat_ptr
) {
252 /* single % at the end, drop that */
255 /* Double percent, output one percent */
257 err
= cn_printf(cn
, "%c", '%');
262 err
= cn_printf(cn
, "%d",
263 task_tgid_vnr(current
));
267 err
= cn_printf(cn
, "%d",
268 task_tgid_nr(current
));
271 err
= cn_printf(cn
, "%d",
272 task_pid_vnr(current
));
275 err
= cn_printf(cn
, "%d",
276 task_pid_nr(current
));
280 err
= cn_printf(cn
, "%u",
281 from_kuid(&init_user_ns
,
286 err
= cn_printf(cn
, "%u",
287 from_kgid(&init_user_ns
,
291 err
= cn_printf(cn
, "%d",
292 __get_dumpable(cprm
->mm_flags
));
294 /* signal that caused the coredump */
296 err
= cn_printf(cn
, "%d",
297 cprm
->siginfo
->si_signo
);
299 /* UNIX time of coredump */
303 time
= ktime_get_real_seconds();
304 err
= cn_printf(cn
, "%lld", time
);
310 err
= cn_esc_printf(cn
, "%s",
311 utsname()->nodename
);
314 /* executable, could be changed by prctl PR_SET_NAME etc */
316 err
= cn_esc_printf(cn
, "%s", current
->comm
);
318 /* file name of executable */
320 err
= cn_print_exe_file(cn
, true);
323 err
= cn_print_exe_file(cn
, false);
325 /* core limit size */
327 err
= cn_printf(cn
, "%lu",
328 rlimit(RLIMIT_CORE
));
341 /* Backward compatibility with core_uses_pid:
343 * If core_pattern does not include a %p (as is the default)
344 * and core_uses_pid is set, then .%pid will be appended to
345 * the filename. Do not do this for piped commands. */
346 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
347 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
354 static int zap_process(struct task_struct
*start
, int exit_code
, int flags
)
356 struct task_struct
*t
;
359 /* ignore all signals except SIGKILL, see prepare_signal() */
360 start
->signal
->flags
= SIGNAL_GROUP_COREDUMP
| flags
;
361 start
->signal
->group_exit_code
= exit_code
;
362 start
->signal
->group_stop_count
= 0;
364 for_each_thread(start
, t
) {
365 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
366 if (t
!= current
&& t
->mm
) {
367 sigaddset(&t
->pending
.signal
, SIGKILL
);
368 signal_wake_up(t
, 1);
376 static int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
377 struct core_state
*core_state
, int exit_code
)
379 struct task_struct
*g
, *p
;
383 spin_lock_irq(&tsk
->sighand
->siglock
);
384 if (!signal_group_exit(tsk
->signal
)) {
385 mm
->core_state
= core_state
;
386 tsk
->signal
->group_exit_task
= tsk
;
387 nr
= zap_process(tsk
, exit_code
, 0);
388 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
390 spin_unlock_irq(&tsk
->sighand
->siglock
);
391 if (unlikely(nr
< 0))
394 tsk
->flags
|= PF_DUMPCORE
;
395 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
398 * We should find and kill all tasks which use this mm, and we should
399 * count them correctly into ->nr_threads. We don't take tasklist
400 * lock, but this is safe wrt:
403 * None of sub-threads can fork after zap_process(leader). All
404 * processes which were created before this point should be
405 * visible to zap_threads() because copy_process() adds the new
406 * process to the tail of init_task.tasks list, and lock/unlock
407 * of ->siglock provides a memory barrier.
410 * The caller holds mm->mmap_lock. This means that the task which
411 * uses this mm can't pass exit_mm(), so it can't exit or clear
415 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
416 * we must see either old or new leader, this does not matter.
417 * However, it can change p->sighand, so lock_task_sighand(p)
418 * must be used. Since p->mm != NULL and we hold ->mmap_lock
421 * Note also that "g" can be the old leader with ->mm == NULL
422 * and already unhashed and thus removed from ->thread_group.
423 * This is OK, __unhash_process()->list_del_rcu() does not
424 * clear the ->next pointer, we will find the new leader via
428 for_each_process(g
) {
429 if (g
== tsk
->group_leader
)
431 if (g
->flags
& PF_KTHREAD
)
434 for_each_thread(g
, p
) {
435 if (unlikely(!p
->mm
))
437 if (unlikely(p
->mm
== mm
)) {
438 lock_task_sighand(p
, &flags
);
439 nr
+= zap_process(p
, exit_code
,
441 unlock_task_sighand(p
, &flags
);
448 atomic_set(&core_state
->nr_threads
, nr
);
452 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
454 struct task_struct
*tsk
= current
;
455 struct mm_struct
*mm
= tsk
->mm
;
456 int core_waiters
= -EBUSY
;
458 init_completion(&core_state
->startup
);
459 core_state
->dumper
.task
= tsk
;
460 core_state
->dumper
.next
= NULL
;
462 if (mmap_write_lock_killable(mm
))
466 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
467 mmap_write_unlock(mm
);
469 if (core_waiters
> 0) {
470 struct core_thread
*ptr
;
472 freezer_do_not_count();
473 wait_for_completion(&core_state
->startup
);
476 * Wait for all the threads to become inactive, so that
477 * all the thread context (extended register state, like
478 * fpu etc) gets copied to the memory.
480 ptr
= core_state
->dumper
.next
;
481 while (ptr
!= NULL
) {
482 wait_task_inactive(ptr
->task
, 0);
490 static void coredump_finish(struct mm_struct
*mm
, bool core_dumped
)
492 struct core_thread
*curr
, *next
;
493 struct task_struct
*task
;
495 spin_lock_irq(¤t
->sighand
->siglock
);
496 if (core_dumped
&& !__fatal_signal_pending(current
))
497 current
->signal
->group_exit_code
|= 0x80;
498 current
->signal
->group_exit_task
= NULL
;
499 current
->signal
->flags
= SIGNAL_GROUP_EXIT
;
500 spin_unlock_irq(¤t
->sighand
->siglock
);
502 next
= mm
->core_state
->dumper
.next
;
503 while ((curr
= next
) != NULL
) {
507 * see exit_mm(), curr->task must not see
508 * ->task == NULL before we read ->next.
512 wake_up_process(task
);
515 mm
->core_state
= NULL
;
518 static bool dump_interrupted(void)
521 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
522 * can do try_to_freeze() and check __fatal_signal_pending(),
523 * but then we need to teach dump_write() to restart and clear
526 return fatal_signal_pending(current
) || freezing(current
);
529 static void wait_for_dump_helpers(struct file
*file
)
531 struct pipe_inode_info
*pipe
= file
->private_data
;
536 wake_up_interruptible_sync(&pipe
->rd_wait
);
537 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
541 * We actually want wait_event_freezable() but then we need
542 * to clear TIF_SIGPENDING and improve dump_interrupted().
544 wait_event_interruptible(pipe
->rd_wait
, pipe
->readers
== 1);
554 * helper function to customize the process used
555 * to collect the core in userspace. Specifically
556 * it sets up a pipe and installs it as fd 0 (stdin)
557 * for the process. Returns 0 on success, or
558 * PTR_ERR on failure.
559 * Note that it also sets the core limit to 1. This
560 * is a special value that we use to trap recursive
563 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
565 struct file
*files
[2];
566 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
567 int err
= create_pipe_files(files
, 0);
573 err
= replace_fd(0, files
[0], 0);
575 /* and disallow core files too */
576 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
581 void do_coredump(const kernel_siginfo_t
*siginfo
)
583 struct core_state core_state
;
585 struct mm_struct
*mm
= current
->mm
;
586 struct linux_binfmt
* binfmt
;
587 const struct cred
*old_cred
;
593 /* require nonrelative corefile path and be extra careful */
594 bool need_suid_safe
= false;
595 bool core_dumped
= false;
596 static atomic_t core_dump_count
= ATOMIC_INIT(0);
597 struct coredump_params cprm
= {
599 .regs
= signal_pt_regs(),
600 .limit
= rlimit(RLIMIT_CORE
),
602 * We must use the same mm->flags while dumping core to avoid
603 * inconsistency of bit flags, since this flag is not protected
606 .mm_flags
= mm
->flags
,
610 audit_core_dumps(siginfo
->si_signo
);
613 if (!binfmt
|| !binfmt
->core_dump
)
615 if (!__get_dumpable(cprm
.mm_flags
))
618 cred
= prepare_creds();
622 * We cannot trust fsuid as being the "true" uid of the process
623 * nor do we know its entire history. We only know it was tainted
624 * so we dump it as root in mode 2, and only into a controlled
625 * environment (pipe handler or fully qualified path).
627 if (__get_dumpable(cprm
.mm_flags
) == SUID_DUMP_ROOT
) {
628 /* Setuid core dump mode */
629 cred
->fsuid
= GLOBAL_ROOT_UID
; /* Dump root private */
630 need_suid_safe
= true;
633 retval
= coredump_wait(siginfo
->si_signo
, &core_state
);
637 old_cred
= override_creds(cred
);
639 ispipe
= format_corename(&cn
, &cprm
, &argv
, &argc
);
645 struct subprocess_info
*sub_info
;
648 printk(KERN_WARNING
"format_corename failed\n");
649 printk(KERN_WARNING
"Aborting core\n");
653 if (cprm
.limit
== 1) {
654 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
656 * Normally core limits are irrelevant to pipes, since
657 * we're not writing to the file system, but we use
658 * cprm.limit of 1 here as a special value, this is a
659 * consistent way to catch recursive crashes.
660 * We can still crash if the core_pattern binary sets
661 * RLIM_CORE = !1, but it runs as root, and can do
662 * lots of stupid things.
664 * Note that we use task_tgid_vnr here to grab the pid
665 * of the process group leader. That way we get the
666 * right pid if a thread in a multi-threaded
667 * core_pattern process dies.
670 "Process %d(%s) has RLIMIT_CORE set to 1\n",
671 task_tgid_vnr(current
), current
->comm
);
672 printk(KERN_WARNING
"Aborting core\n");
675 cprm
.limit
= RLIM_INFINITY
;
677 dump_count
= atomic_inc_return(&core_dump_count
);
678 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
679 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
680 task_tgid_vnr(current
), current
->comm
);
681 printk(KERN_WARNING
"Skipping core dump\n");
685 helper_argv
= kmalloc_array(argc
+ 1, sizeof(*helper_argv
),
688 printk(KERN_WARNING
"%s failed to allocate memory\n",
692 for (argi
= 0; argi
< argc
; argi
++)
693 helper_argv
[argi
] = cn
.corename
+ argv
[argi
];
694 helper_argv
[argi
] = NULL
;
697 sub_info
= call_usermodehelper_setup(helper_argv
[0],
698 helper_argv
, NULL
, GFP_KERNEL
,
699 umh_pipe_setup
, NULL
, &cprm
);
701 retval
= call_usermodehelper_exec(sub_info
,
706 printk(KERN_INFO
"Core dump to |%s pipe failed\n",
711 struct user_namespace
*mnt_userns
;
713 int open_flags
= O_CREAT
| O_RDWR
| O_NOFOLLOW
|
714 O_LARGEFILE
| O_EXCL
;
716 if (cprm
.limit
< binfmt
->min_coredump
)
719 if (need_suid_safe
&& cn
.corename
[0] != '/') {
720 printk(KERN_WARNING
"Pid %d(%s) can only dump core "\
721 "to fully qualified path!\n",
722 task_tgid_vnr(current
), current
->comm
);
723 printk(KERN_WARNING
"Skipping core dump\n");
728 * Unlink the file if it exists unless this is a SUID
729 * binary - in that case, we're running around with root
730 * privs and don't want to unlink another user's coredump.
732 if (!need_suid_safe
) {
734 * If it doesn't exist, that's fine. If there's some
735 * other problem, we'll catch it at the filp_open().
737 do_unlinkat(AT_FDCWD
, getname_kernel(cn
.corename
));
741 * There is a race between unlinking and creating the
742 * file, but if that causes an EEXIST here, that's
743 * fine - another process raced with us while creating
744 * the corefile, and the other process won. To userspace,
745 * what matters is that at least one of the two processes
746 * writes its coredump successfully, not which one.
748 if (need_suid_safe
) {
750 * Using user namespaces, normal user tasks can change
751 * their current->fs->root to point to arbitrary
752 * directories. Since the intention of the "only dump
753 * with a fully qualified path" rule is to control where
754 * coredumps may be placed using root privileges,
755 * current->fs->root must not be used. Instead, use the
756 * root directory of init_task.
760 task_lock(&init_task
);
761 get_fs_root(init_task
.fs
, &root
);
762 task_unlock(&init_task
);
763 cprm
.file
= file_open_root(&root
, cn
.corename
,
767 cprm
.file
= filp_open(cn
.corename
, open_flags
, 0600);
769 if (IS_ERR(cprm
.file
))
772 inode
= file_inode(cprm
.file
);
773 if (inode
->i_nlink
> 1)
775 if (d_unhashed(cprm
.file
->f_path
.dentry
))
778 * AK: actually i see no reason to not allow this for named
779 * pipes etc, but keep the previous behaviour for now.
781 if (!S_ISREG(inode
->i_mode
))
784 * Don't dump core if the filesystem changed owner or mode
785 * of the file during file creation. This is an issue when
786 * a process dumps core while its cwd is e.g. on a vfat
789 mnt_userns
= file_mnt_user_ns(cprm
.file
);
790 if (!uid_eq(i_uid_into_mnt(mnt_userns
, inode
),
792 pr_info_ratelimited("Core dump to %s aborted: cannot preserve file owner\n",
796 if ((inode
->i_mode
& 0677) != 0600) {
797 pr_info_ratelimited("Core dump to %s aborted: cannot preserve file permissions\n",
801 if (!(cprm
.file
->f_mode
& FMODE_CAN_WRITE
))
803 if (do_truncate(mnt_userns
, cprm
.file
->f_path
.dentry
,
808 /* get us an unshared descriptor table; almost always a no-op */
809 /* The cell spufs coredump code reads the file descriptor tables */
810 retval
= unshare_files();
813 if (!dump_interrupted()) {
815 * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would
816 * have this set to NULL.
819 pr_info("Core dump to |%s disabled\n", cn
.corename
);
822 if (!dump_vma_snapshot(&cprm
))
825 file_start_write(cprm
.file
);
826 core_dumped
= binfmt
->core_dump(&cprm
);
828 * Ensures that file size is big enough to contain the current
829 * file postion. This prevents gdb from complaining about
830 * a truncated file if the last "write" to the file was
835 dump_emit(&cprm
, "", 1);
837 file_end_write(cprm
.file
);
838 free_vma_snapshot(&cprm
);
840 if (ispipe
&& core_pipe_limit
)
841 wait_for_dump_helpers(cprm
.file
);
844 filp_close(cprm
.file
, NULL
);
847 atomic_dec(&core_dump_count
);
851 coredump_finish(mm
, core_dumped
);
852 revert_creds(old_cred
);
860 * Core dumping helper functions. These are the only things you should
861 * do on a core-file: use only these functions to write out all the
864 static int __dump_emit(struct coredump_params
*cprm
, const void *addr
, int nr
)
866 struct file
*file
= cprm
->file
;
867 loff_t pos
= file
->f_pos
;
869 if (cprm
->written
+ nr
> cprm
->limit
)
873 if (dump_interrupted())
875 n
= __kernel_write(file
, addr
, nr
, &pos
);
885 static int __dump_skip(struct coredump_params
*cprm
, size_t nr
)
887 static char zeroes
[PAGE_SIZE
];
888 struct file
*file
= cprm
->file
;
889 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
890 if (dump_interrupted() ||
891 file
->f_op
->llseek(file
, nr
, SEEK_CUR
) < 0)
896 while (nr
> PAGE_SIZE
) {
897 if (!__dump_emit(cprm
, zeroes
, PAGE_SIZE
))
901 return __dump_emit(cprm
, zeroes
, nr
);
905 int dump_emit(struct coredump_params
*cprm
, const void *addr
, int nr
)
908 if (!__dump_skip(cprm
, cprm
->to_skip
))
912 return __dump_emit(cprm
, addr
, nr
);
914 EXPORT_SYMBOL(dump_emit
);
916 void dump_skip_to(struct coredump_params
*cprm
, unsigned long pos
)
918 cprm
->to_skip
= pos
- cprm
->pos
;
920 EXPORT_SYMBOL(dump_skip_to
);
922 void dump_skip(struct coredump_params
*cprm
, size_t nr
)
926 EXPORT_SYMBOL(dump_skip
);
928 #ifdef CONFIG_ELF_CORE
929 int dump_user_range(struct coredump_params
*cprm
, unsigned long start
,
934 for (addr
= start
; addr
< start
+ len
; addr
+= PAGE_SIZE
) {
939 * To avoid having to allocate page tables for virtual address
940 * ranges that have never been used yet, and also to make it
941 * easy to generate sparse core files, use a helper that returns
942 * NULL when encountering an empty page table entry that would
943 * otherwise have been filled with the zero page.
945 page
= get_dump_page(addr
);
947 void *kaddr
= kmap_local_page(page
);
949 stop
= !dump_emit(cprm
, kaddr
, PAGE_SIZE
);
955 dump_skip(cprm
, PAGE_SIZE
);
962 int dump_align(struct coredump_params
*cprm
, int align
)
964 unsigned mod
= (cprm
->pos
+ cprm
->to_skip
) & (align
- 1);
965 if (align
& (align
- 1))
968 cprm
->to_skip
+= align
- mod
;
971 EXPORT_SYMBOL(dump_align
);
974 * The purpose of always_dump_vma() is to make sure that special kernel mappings
975 * that are useful for post-mortem analysis are included in every core dump.
976 * In that way we ensure that the core dump is fully interpretable later
977 * without matching up the same kernel and hardware config to see what PC values
978 * meant. These special mappings include - vDSO, vsyscall, and other
979 * architecture specific mappings
981 static bool always_dump_vma(struct vm_area_struct
*vma
)
983 /* Any vsyscall mappings? */
984 if (vma
== get_gate_vma(vma
->vm_mm
))
988 * Assume that all vmas with a .name op should always be dumped.
989 * If this changes, a new vm_ops field can easily be added.
991 if (vma
->vm_ops
&& vma
->vm_ops
->name
&& vma
->vm_ops
->name(vma
))
995 * arch_vma_name() returns non-NULL for special architecture mappings,
996 * such as vDSO sections.
998 if (arch_vma_name(vma
))
1004 #define DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER 1
1007 * Decide how much of @vma's contents should be included in a core dump.
1009 static unsigned long vma_dump_size(struct vm_area_struct
*vma
,
1010 unsigned long mm_flags
)
1012 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1014 /* always dump the vdso and vsyscall sections */
1015 if (always_dump_vma(vma
))
1018 if (vma
->vm_flags
& VM_DONTDUMP
)
1021 /* support for DAX */
1022 if (vma_is_dax(vma
)) {
1023 if ((vma
->vm_flags
& VM_SHARED
) && FILTER(DAX_SHARED
))
1025 if (!(vma
->vm_flags
& VM_SHARED
) && FILTER(DAX_PRIVATE
))
1030 /* Hugetlb memory check */
1031 if (is_vm_hugetlb_page(vma
)) {
1032 if ((vma
->vm_flags
& VM_SHARED
) && FILTER(HUGETLB_SHARED
))
1034 if (!(vma
->vm_flags
& VM_SHARED
) && FILTER(HUGETLB_PRIVATE
))
1039 /* Do not dump I/O mapped devices or special mappings */
1040 if (vma
->vm_flags
& VM_IO
)
1043 /* By default, dump shared memory if mapped from an anonymous file. */
1044 if (vma
->vm_flags
& VM_SHARED
) {
1045 if (file_inode(vma
->vm_file
)->i_nlink
== 0 ?
1046 FILTER(ANON_SHARED
) : FILTER(MAPPED_SHARED
))
1051 /* Dump segments that have been written to. */
1052 if ((!IS_ENABLED(CONFIG_MMU
) || vma
->anon_vma
) && FILTER(ANON_PRIVATE
))
1054 if (vma
->vm_file
== NULL
)
1057 if (FILTER(MAPPED_PRIVATE
))
1061 * If this is the beginning of an executable file mapping,
1062 * dump the first page to aid in determining what was mapped here.
1064 if (FILTER(ELF_HEADERS
) &&
1065 vma
->vm_pgoff
== 0 && (vma
->vm_flags
& VM_READ
)) {
1066 if ((READ_ONCE(file_inode(vma
->vm_file
)->i_mode
) & 0111) != 0)
1070 * ELF libraries aren't always executable.
1071 * We'll want to check whether the mapping starts with the ELF
1072 * magic, but not now - we're holding the mmap lock,
1073 * so copy_from_user() doesn't work here.
1074 * Use a placeholder instead, and fix it up later in
1075 * dump_vma_snapshot().
1077 return DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER
;
1085 return vma
->vm_end
- vma
->vm_start
;
1088 static struct vm_area_struct
*first_vma(struct task_struct
*tsk
,
1089 struct vm_area_struct
*gate_vma
)
1091 struct vm_area_struct
*ret
= tsk
->mm
->mmap
;
1099 * Helper function for iterating across a vma list. It ensures that the caller
1100 * will visit `gate_vma' prior to terminating the search.
1102 static struct vm_area_struct
*next_vma(struct vm_area_struct
*this_vma
,
1103 struct vm_area_struct
*gate_vma
)
1105 struct vm_area_struct
*ret
;
1107 ret
= this_vma
->vm_next
;
1110 if (this_vma
== gate_vma
)
1115 static void free_vma_snapshot(struct coredump_params
*cprm
)
1117 if (cprm
->vma_meta
) {
1119 for (i
= 0; i
< cprm
->vma_count
; i
++) {
1120 struct file
*file
= cprm
->vma_meta
[i
].file
;
1124 kvfree(cprm
->vma_meta
);
1125 cprm
->vma_meta
= NULL
;
1130 * Under the mmap_lock, take a snapshot of relevant information about the task's
1133 static bool dump_vma_snapshot(struct coredump_params
*cprm
)
1135 struct vm_area_struct
*vma
, *gate_vma
;
1136 struct mm_struct
*mm
= current
->mm
;
1140 * Once the stack expansion code is fixed to not change VMA bounds
1141 * under mmap_lock in read mode, this can be changed to take the
1142 * mmap_lock in read mode.
1144 if (mmap_write_lock_killable(mm
))
1147 cprm
->vma_data_size
= 0;
1148 gate_vma
= get_gate_vma(mm
);
1149 cprm
->vma_count
= mm
->map_count
+ (gate_vma
? 1 : 0);
1151 cprm
->vma_meta
= kvmalloc_array(cprm
->vma_count
, sizeof(*cprm
->vma_meta
), GFP_KERNEL
);
1152 if (!cprm
->vma_meta
) {
1153 mmap_write_unlock(mm
);
1157 for (i
= 0, vma
= first_vma(current
, gate_vma
); vma
!= NULL
;
1158 vma
= next_vma(vma
, gate_vma
), i
++) {
1159 struct core_vma_metadata
*m
= cprm
->vma_meta
+ i
;
1161 m
->start
= vma
->vm_start
;
1162 m
->end
= vma
->vm_end
;
1163 m
->flags
= vma
->vm_flags
;
1164 m
->dump_size
= vma_dump_size(vma
, cprm
->mm_flags
);
1165 m
->pgoff
= vma
->vm_pgoff
;
1167 m
->file
= vma
->vm_file
;
1172 mmap_write_unlock(mm
);
1174 for (i
= 0; i
< cprm
->vma_count
; i
++) {
1175 struct core_vma_metadata
*m
= cprm
->vma_meta
+ i
;
1177 if (m
->dump_size
== DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER
) {
1178 char elfmag
[SELFMAG
];
1180 if (copy_from_user(elfmag
, (void __user
*)m
->start
, SELFMAG
) ||
1181 memcmp(elfmag
, ELFMAG
, SELFMAG
) != 0) {
1184 m
->dump_size
= PAGE_SIZE
;
1188 cprm
->vma_data_size
+= m
->dump_size
;