]>
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
, bool name_only
)
158 struct file
*exe_file
;
159 char *pathbuf
, *path
, *ptr
;
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
);
179 ptr
= strrchr(path
, '/');
183 ret
= cn_esc_printf(cn
, "%s", path
);
192 /* format_corename will inspect the pattern parameter, and output a
193 * name into corename, which must have space for at least
194 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
196 static int format_corename(struct core_name
*cn
, struct coredump_params
*cprm
,
197 size_t **argv
, int *argc
)
199 const struct cred
*cred
= current_cred();
200 const char *pat_ptr
= core_pattern
;
201 int ispipe
= (*pat_ptr
== '|');
202 bool was_space
= false;
203 int pid_in_pattern
= 0;
208 if (expand_corename(cn
, core_name_size
))
210 cn
->corename
[0] = '\0';
213 int argvs
= sizeof(core_pattern
) / 2;
214 (*argv
) = kmalloc_array(argvs
, sizeof(**argv
), GFP_KERNEL
);
217 (*argv
)[(*argc
)++] = 0;
223 /* Repeat as long as we have more pattern to process and more output
227 * Split on spaces before doing template expansion so that
228 * %e and %E don't get split if they have spaces in them
231 if (isspace(*pat_ptr
)) {
236 } else if (was_space
) {
238 err
= cn_printf(cn
, "%c", '\0');
241 (*argv
)[(*argc
)++] = cn
->used
;
244 if (*pat_ptr
!= '%') {
245 err
= cn_printf(cn
, "%c", *pat_ptr
++);
247 switch (*++pat_ptr
) {
248 /* single % at the end, drop that */
251 /* Double percent, output one percent */
253 err
= cn_printf(cn
, "%c", '%');
258 err
= cn_printf(cn
, "%d",
259 task_tgid_vnr(current
));
263 err
= cn_printf(cn
, "%d",
264 task_tgid_nr(current
));
267 err
= cn_printf(cn
, "%d",
268 task_pid_vnr(current
));
271 err
= cn_printf(cn
, "%d",
272 task_pid_nr(current
));
276 err
= cn_printf(cn
, "%u",
277 from_kuid(&init_user_ns
,
282 err
= cn_printf(cn
, "%u",
283 from_kgid(&init_user_ns
,
287 err
= cn_printf(cn
, "%d",
288 __get_dumpable(cprm
->mm_flags
));
290 /* signal that caused the coredump */
292 err
= cn_printf(cn
, "%d",
293 cprm
->siginfo
->si_signo
);
295 /* UNIX time of coredump */
299 time
= ktime_get_real_seconds();
300 err
= cn_printf(cn
, "%lld", time
);
306 err
= cn_esc_printf(cn
, "%s",
307 utsname()->nodename
);
310 /* executable, could be changed by prctl PR_SET_NAME etc */
312 err
= cn_esc_printf(cn
, "%s", current
->comm
);
314 /* file name of executable */
316 err
= cn_print_exe_file(cn
, true);
319 err
= cn_print_exe_file(cn
, false);
321 /* core limit size */
323 err
= cn_printf(cn
, "%lu",
324 rlimit(RLIMIT_CORE
));
337 /* Backward compatibility with core_uses_pid:
339 * If core_pattern does not include a %p (as is the default)
340 * and core_uses_pid is set, then .%pid will be appended to
341 * the filename. Do not do this for piped commands. */
342 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
343 err
= cn_printf(cn
, ".%d", task_tgid_vnr(current
));
350 static int zap_process(struct task_struct
*start
, int exit_code
, int flags
)
352 struct task_struct
*t
;
355 /* ignore all signals except SIGKILL, see prepare_signal() */
356 start
->signal
->flags
= SIGNAL_GROUP_COREDUMP
| flags
;
357 start
->signal
->group_exit_code
= exit_code
;
358 start
->signal
->group_stop_count
= 0;
360 for_each_thread(start
, t
) {
361 task_clear_jobctl_pending(t
, JOBCTL_PENDING_MASK
);
362 if (t
!= current
&& t
->mm
) {
363 sigaddset(&t
->pending
.signal
, SIGKILL
);
364 signal_wake_up(t
, 1);
372 static int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
373 struct core_state
*core_state
, int exit_code
)
375 struct task_struct
*g
, *p
;
379 spin_lock_irq(&tsk
->sighand
->siglock
);
380 if (!signal_group_exit(tsk
->signal
)) {
381 mm
->core_state
= core_state
;
382 tsk
->signal
->group_exit_task
= tsk
;
383 nr
= zap_process(tsk
, exit_code
, 0);
384 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
386 spin_unlock_irq(&tsk
->sighand
->siglock
);
387 if (unlikely(nr
< 0))
390 tsk
->flags
|= PF_DUMPCORE
;
391 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
394 * We should find and kill all tasks which use this mm, and we should
395 * count them correctly into ->nr_threads. We don't take tasklist
396 * lock, but this is safe wrt:
399 * None of sub-threads can fork after zap_process(leader). All
400 * processes which were created before this point should be
401 * visible to zap_threads() because copy_process() adds the new
402 * process to the tail of init_task.tasks list, and lock/unlock
403 * of ->siglock provides a memory barrier.
406 * The caller holds mm->mmap_lock. This means that the task which
407 * uses this mm can't pass exit_mm(), so it can't exit or clear
411 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
412 * we must see either old or new leader, this does not matter.
413 * However, it can change p->sighand, so lock_task_sighand(p)
414 * must be used. Since p->mm != NULL and we hold ->mmap_lock
417 * Note also that "g" can be the old leader with ->mm == NULL
418 * and already unhashed and thus removed from ->thread_group.
419 * This is OK, __unhash_process()->list_del_rcu() does not
420 * clear the ->next pointer, we will find the new leader via
424 for_each_process(g
) {
425 if (g
== tsk
->group_leader
)
427 if (g
->flags
& PF_KTHREAD
)
430 for_each_thread(g
, p
) {
431 if (unlikely(!p
->mm
))
433 if (unlikely(p
->mm
== mm
)) {
434 lock_task_sighand(p
, &flags
);
435 nr
+= zap_process(p
, exit_code
,
437 unlock_task_sighand(p
, &flags
);
444 atomic_set(&core_state
->nr_threads
, nr
);
448 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
450 struct task_struct
*tsk
= current
;
451 struct mm_struct
*mm
= tsk
->mm
;
452 int core_waiters
= -EBUSY
;
454 init_completion(&core_state
->startup
);
455 core_state
->dumper
.task
= tsk
;
456 core_state
->dumper
.next
= NULL
;
458 if (mmap_write_lock_killable(mm
))
462 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
463 mmap_write_unlock(mm
);
465 if (core_waiters
> 0) {
466 struct core_thread
*ptr
;
468 freezer_do_not_count();
469 wait_for_completion(&core_state
->startup
);
472 * Wait for all the threads to become inactive, so that
473 * all the thread context (extended register state, like
474 * fpu etc) gets copied to the memory.
476 ptr
= core_state
->dumper
.next
;
477 while (ptr
!= NULL
) {
478 wait_task_inactive(ptr
->task
, 0);
486 static void coredump_finish(struct mm_struct
*mm
, bool core_dumped
)
488 struct core_thread
*curr
, *next
;
489 struct task_struct
*task
;
491 spin_lock_irq(¤t
->sighand
->siglock
);
492 if (core_dumped
&& !__fatal_signal_pending(current
))
493 current
->signal
->group_exit_code
|= 0x80;
494 current
->signal
->group_exit_task
= NULL
;
495 current
->signal
->flags
= SIGNAL_GROUP_EXIT
;
496 spin_unlock_irq(¤t
->sighand
->siglock
);
498 next
= mm
->core_state
->dumper
.next
;
499 while ((curr
= next
) != NULL
) {
503 * see exit_mm(), curr->task must not see
504 * ->task == NULL before we read ->next.
508 wake_up_process(task
);
511 mm
->core_state
= NULL
;
514 static bool dump_interrupted(void)
517 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
518 * can do try_to_freeze() and check __fatal_signal_pending(),
519 * but then we need to teach dump_write() to restart and clear
522 return signal_pending(current
);
525 static void wait_for_dump_helpers(struct file
*file
)
527 struct pipe_inode_info
*pipe
= file
->private_data
;
532 wake_up_interruptible_sync(&pipe
->rd_wait
);
533 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
537 * We actually want wait_event_freezable() but then we need
538 * to clear TIF_SIGPENDING and improve dump_interrupted().
540 wait_event_interruptible(pipe
->rd_wait
, pipe
->readers
== 1);
550 * helper function to customize the process used
551 * to collect the core in userspace. Specifically
552 * it sets up a pipe and installs it as fd 0 (stdin)
553 * for the process. Returns 0 on success, or
554 * PTR_ERR on failure.
555 * Note that it also sets the core limit to 1. This
556 * is a special value that we use to trap recursive
559 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
561 struct file
*files
[2];
562 struct coredump_params
*cp
= (struct coredump_params
*)info
->data
;
563 int err
= create_pipe_files(files
, 0);
569 err
= replace_fd(0, files
[0], 0);
571 /* and disallow core files too */
572 current
->signal
->rlim
[RLIMIT_CORE
] = (struct rlimit
){1, 1};
577 void do_coredump(const kernel_siginfo_t
*siginfo
)
579 struct core_state core_state
;
581 struct mm_struct
*mm
= current
->mm
;
582 struct linux_binfmt
* binfmt
;
583 const struct cred
*old_cred
;
589 /* require nonrelative corefile path and be extra careful */
590 bool need_suid_safe
= false;
591 bool core_dumped
= false;
592 static atomic_t core_dump_count
= ATOMIC_INIT(0);
593 struct coredump_params cprm
= {
595 .regs
= signal_pt_regs(),
596 .limit
= rlimit(RLIMIT_CORE
),
598 * We must use the same mm->flags while dumping core to avoid
599 * inconsistency of bit flags, since this flag is not protected
602 .mm_flags
= mm
->flags
,
605 audit_core_dumps(siginfo
->si_signo
);
608 if (!binfmt
|| !binfmt
->core_dump
)
610 if (!__get_dumpable(cprm
.mm_flags
))
613 cred
= prepare_creds();
617 * We cannot trust fsuid as being the "true" uid of the process
618 * nor do we know its entire history. We only know it was tainted
619 * so we dump it as root in mode 2, and only into a controlled
620 * environment (pipe handler or fully qualified path).
622 if (__get_dumpable(cprm
.mm_flags
) == SUID_DUMP_ROOT
) {
623 /* Setuid core dump mode */
624 cred
->fsuid
= GLOBAL_ROOT_UID
; /* Dump root private */
625 need_suid_safe
= true;
628 retval
= coredump_wait(siginfo
->si_signo
, &core_state
);
632 old_cred
= override_creds(cred
);
634 ispipe
= format_corename(&cn
, &cprm
, &argv
, &argc
);
640 struct subprocess_info
*sub_info
;
643 printk(KERN_WARNING
"format_corename failed\n");
644 printk(KERN_WARNING
"Aborting core\n");
648 if (cprm
.limit
== 1) {
649 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
651 * Normally core limits are irrelevant to pipes, since
652 * we're not writing to the file system, but we use
653 * cprm.limit of 1 here as a special value, this is a
654 * consistent way to catch recursive crashes.
655 * We can still crash if the core_pattern binary sets
656 * RLIM_CORE = !1, but it runs as root, and can do
657 * lots of stupid things.
659 * Note that we use task_tgid_vnr here to grab the pid
660 * of the process group leader. That way we get the
661 * right pid if a thread in a multi-threaded
662 * core_pattern process dies.
665 "Process %d(%s) has RLIMIT_CORE set to 1\n",
666 task_tgid_vnr(current
), current
->comm
);
667 printk(KERN_WARNING
"Aborting core\n");
670 cprm
.limit
= RLIM_INFINITY
;
672 dump_count
= atomic_inc_return(&core_dump_count
);
673 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
674 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
675 task_tgid_vnr(current
), current
->comm
);
676 printk(KERN_WARNING
"Skipping core dump\n");
680 helper_argv
= kmalloc_array(argc
+ 1, sizeof(*helper_argv
),
683 printk(KERN_WARNING
"%s failed to allocate memory\n",
687 for (argi
= 0; argi
< argc
; argi
++)
688 helper_argv
[argi
] = cn
.corename
+ argv
[argi
];
689 helper_argv
[argi
] = NULL
;
692 sub_info
= call_usermodehelper_setup(helper_argv
[0],
693 helper_argv
, NULL
, GFP_KERNEL
,
694 umh_pipe_setup
, NULL
, &cprm
);
696 retval
= call_usermodehelper_exec(sub_info
,
701 printk(KERN_INFO
"Core dump to |%s pipe failed\n",
707 int open_flags
= O_CREAT
| O_RDWR
| O_NOFOLLOW
|
708 O_LARGEFILE
| O_EXCL
;
710 if (cprm
.limit
< binfmt
->min_coredump
)
713 if (need_suid_safe
&& cn
.corename
[0] != '/') {
714 printk(KERN_WARNING
"Pid %d(%s) can only dump core "\
715 "to fully qualified path!\n",
716 task_tgid_vnr(current
), current
->comm
);
717 printk(KERN_WARNING
"Skipping core dump\n");
722 * Unlink the file if it exists unless this is a SUID
723 * binary - in that case, we're running around with root
724 * privs and don't want to unlink another user's coredump.
726 if (!need_suid_safe
) {
728 * If it doesn't exist, that's fine. If there's some
729 * other problem, we'll catch it at the filp_open().
731 do_unlinkat(AT_FDCWD
, getname_kernel(cn
.corename
));
735 * There is a race between unlinking and creating the
736 * file, but if that causes an EEXIST here, that's
737 * fine - another process raced with us while creating
738 * the corefile, and the other process won. To userspace,
739 * what matters is that at least one of the two processes
740 * writes its coredump successfully, not which one.
742 if (need_suid_safe
) {
744 * Using user namespaces, normal user tasks can change
745 * their current->fs->root to point to arbitrary
746 * directories. Since the intention of the "only dump
747 * with a fully qualified path" rule is to control where
748 * coredumps may be placed using root privileges,
749 * current->fs->root must not be used. Instead, use the
750 * root directory of init_task.
754 task_lock(&init_task
);
755 get_fs_root(init_task
.fs
, &root
);
756 task_unlock(&init_task
);
757 cprm
.file
= file_open_root(root
.dentry
, root
.mnt
,
758 cn
.corename
, open_flags
, 0600);
761 cprm
.file
= filp_open(cn
.corename
, open_flags
, 0600);
763 if (IS_ERR(cprm
.file
))
766 inode
= file_inode(cprm
.file
);
767 if (inode
->i_nlink
> 1)
769 if (d_unhashed(cprm
.file
->f_path
.dentry
))
772 * AK: actually i see no reason to not allow this for named
773 * pipes etc, but keep the previous behaviour for now.
775 if (!S_ISREG(inode
->i_mode
))
778 * Don't dump core if the filesystem changed owner or mode
779 * of the file during file creation. This is an issue when
780 * a process dumps core while its cwd is e.g. on a vfat
783 if (!uid_eq(inode
->i_uid
, current_fsuid()))
785 if ((inode
->i_mode
& 0677) != 0600)
787 if (!(cprm
.file
->f_mode
& FMODE_CAN_WRITE
))
789 if (do_truncate(cprm
.file
->f_path
.dentry
, 0, 0, cprm
.file
))
793 /* get us an unshared descriptor table; almost always a no-op */
794 /* The cell spufs coredump code reads the file descriptor tables */
795 retval
= unshare_files();
798 if (!dump_interrupted()) {
800 * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would
801 * have this set to NULL.
804 pr_info("Core dump to |%s disabled\n", cn
.corename
);
807 file_start_write(cprm
.file
);
808 core_dumped
= binfmt
->core_dump(&cprm
);
809 file_end_write(cprm
.file
);
811 if (ispipe
&& core_pipe_limit
)
812 wait_for_dump_helpers(cprm
.file
);
815 filp_close(cprm
.file
, NULL
);
818 atomic_dec(&core_dump_count
);
822 coredump_finish(mm
, core_dumped
);
823 revert_creds(old_cred
);
831 * Core dumping helper functions. These are the only things you should
832 * do on a core-file: use only these functions to write out all the
835 int dump_emit(struct coredump_params
*cprm
, const void *addr
, int nr
)
837 struct file
*file
= cprm
->file
;
838 loff_t pos
= file
->f_pos
;
840 if (cprm
->written
+ nr
> cprm
->limit
)
844 if (dump_interrupted())
846 n
= __kernel_write(file
, addr
, nr
, &pos
);
855 EXPORT_SYMBOL(dump_emit
);
857 int dump_skip(struct coredump_params
*cprm
, size_t nr
)
859 static char zeroes
[PAGE_SIZE
];
860 struct file
*file
= cprm
->file
;
861 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
862 if (dump_interrupted() ||
863 file
->f_op
->llseek(file
, nr
, SEEK_CUR
) < 0)
868 while (nr
> PAGE_SIZE
) {
869 if (!dump_emit(cprm
, zeroes
, PAGE_SIZE
))
873 return dump_emit(cprm
, zeroes
, nr
);
876 EXPORT_SYMBOL(dump_skip
);
878 #ifdef CONFIG_ELF_CORE
879 int dump_user_range(struct coredump_params
*cprm
, unsigned long start
,
884 for (addr
= start
; addr
< start
+ len
; addr
+= PAGE_SIZE
) {
889 * To avoid having to allocate page tables for virtual address
890 * ranges that have never been used yet, and also to make it
891 * easy to generate sparse core files, use a helper that returns
892 * NULL when encountering an empty page table entry that would
893 * otherwise have been filled with the zero page.
895 page
= get_dump_page(addr
);
897 void *kaddr
= kmap(page
);
899 stop
= !dump_emit(cprm
, kaddr
, PAGE_SIZE
);
903 stop
= !dump_skip(cprm
, PAGE_SIZE
);
912 int dump_align(struct coredump_params
*cprm
, int align
)
914 unsigned mod
= cprm
->pos
& (align
- 1);
915 if (align
& (align
- 1))
917 return mod
? dump_skip(cprm
, align
- mod
) : 1;
919 EXPORT_SYMBOL(dump_align
);
922 * Ensures that file size is big enough to contain the current file
923 * postion. This prevents gdb from complaining about a truncated file
924 * if the last "write" to the file was dump_skip.
926 void dump_truncate(struct coredump_params
*cprm
)
928 struct file
*file
= cprm
->file
;
931 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
932 offset
= file
->f_op
->llseek(file
, 0, SEEK_CUR
);
933 if (i_size_read(file
->f_mapping
->host
) < offset
)
934 do_truncate(file
->f_path
.dentry
, offset
, 0, file
);
937 EXPORT_SYMBOL(dump_truncate
);
940 * The purpose of always_dump_vma() is to make sure that special kernel mappings
941 * that are useful for post-mortem analysis are included in every core dump.
942 * In that way we ensure that the core dump is fully interpretable later
943 * without matching up the same kernel and hardware config to see what PC values
944 * meant. These special mappings include - vDSO, vsyscall, and other
945 * architecture specific mappings
947 static bool always_dump_vma(struct vm_area_struct
*vma
)
949 /* Any vsyscall mappings? */
950 if (vma
== get_gate_vma(vma
->vm_mm
))
954 * Assume that all vmas with a .name op should always be dumped.
955 * If this changes, a new vm_ops field can easily be added.
957 if (vma
->vm_ops
&& vma
->vm_ops
->name
&& vma
->vm_ops
->name(vma
))
961 * arch_vma_name() returns non-NULL for special architecture mappings,
962 * such as vDSO sections.
964 if (arch_vma_name(vma
))
971 * Decide how much of @vma's contents should be included in a core dump.
973 static unsigned long vma_dump_size(struct vm_area_struct
*vma
,
974 unsigned long mm_flags
)
976 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
978 /* always dump the vdso and vsyscall sections */
979 if (always_dump_vma(vma
))
982 if (vma
->vm_flags
& VM_DONTDUMP
)
985 /* support for DAX */
986 if (vma_is_dax(vma
)) {
987 if ((vma
->vm_flags
& VM_SHARED
) && FILTER(DAX_SHARED
))
989 if (!(vma
->vm_flags
& VM_SHARED
) && FILTER(DAX_PRIVATE
))
994 /* Hugetlb memory check */
995 if (is_vm_hugetlb_page(vma
)) {
996 if ((vma
->vm_flags
& VM_SHARED
) && FILTER(HUGETLB_SHARED
))
998 if (!(vma
->vm_flags
& VM_SHARED
) && FILTER(HUGETLB_PRIVATE
))
1003 /* Do not dump I/O mapped devices or special mappings */
1004 if (vma
->vm_flags
& VM_IO
)
1007 /* By default, dump shared memory if mapped from an anonymous file. */
1008 if (vma
->vm_flags
& VM_SHARED
) {
1009 if (file_inode(vma
->vm_file
)->i_nlink
== 0 ?
1010 FILTER(ANON_SHARED
) : FILTER(MAPPED_SHARED
))
1015 /* Dump segments that have been written to. */
1016 if ((!IS_ENABLED(CONFIG_MMU
) || vma
->anon_vma
) && FILTER(ANON_PRIVATE
))
1018 if (vma
->vm_file
== NULL
)
1021 if (FILTER(MAPPED_PRIVATE
))
1025 * If this is the beginning of an executable file mapping,
1026 * dump the first page to aid in determining what was mapped here.
1028 if (FILTER(ELF_HEADERS
) &&
1029 vma
->vm_pgoff
== 0 && (vma
->vm_flags
& VM_READ
) &&
1030 (READ_ONCE(file_inode(vma
->vm_file
)->i_mode
) & 0111) != 0)
1038 return vma
->vm_end
- vma
->vm_start
;
1041 static struct vm_area_struct
*first_vma(struct task_struct
*tsk
,
1042 struct vm_area_struct
*gate_vma
)
1044 struct vm_area_struct
*ret
= tsk
->mm
->mmap
;
1052 * Helper function for iterating across a vma list. It ensures that the caller
1053 * will visit `gate_vma' prior to terminating the search.
1055 static struct vm_area_struct
*next_vma(struct vm_area_struct
*this_vma
,
1056 struct vm_area_struct
*gate_vma
)
1058 struct vm_area_struct
*ret
;
1060 ret
= this_vma
->vm_next
;
1063 if (this_vma
== gate_vma
)
1069 * Under the mmap_lock, take a snapshot of relevant information about the task's
1072 int dump_vma_snapshot(struct coredump_params
*cprm
, int *vma_count
,
1073 struct core_vma_metadata
**vma_meta
,
1074 size_t *vma_data_size_ptr
)
1076 struct vm_area_struct
*vma
, *gate_vma
;
1077 struct mm_struct
*mm
= current
->mm
;
1079 size_t vma_data_size
= 0;
1082 * Once the stack expansion code is fixed to not change VMA bounds
1083 * under mmap_lock in read mode, this can be changed to take the
1084 * mmap_lock in read mode.
1086 if (mmap_write_lock_killable(mm
))
1089 gate_vma
= get_gate_vma(mm
);
1090 *vma_count
= mm
->map_count
+ (gate_vma
? 1 : 0);
1092 *vma_meta
= kvmalloc_array(*vma_count
, sizeof(**vma_meta
), GFP_KERNEL
);
1094 mmap_write_unlock(mm
);
1098 for (i
= 0, vma
= first_vma(current
, gate_vma
); vma
!= NULL
;
1099 vma
= next_vma(vma
, gate_vma
), i
++) {
1100 struct core_vma_metadata
*m
= (*vma_meta
) + i
;
1102 m
->start
= vma
->vm_start
;
1103 m
->end
= vma
->vm_end
;
1104 m
->flags
= vma
->vm_flags
;
1105 m
->dump_size
= vma_dump_size(vma
, cprm
->mm_flags
);
1107 vma_data_size
+= m
->dump_size
;
1110 mmap_write_unlock(mm
);
1112 if (WARN_ON(i
!= *vma_count
))
1115 *vma_data_size_ptr
= vma_data_size
;