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>
6 #include <linux/mm.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/sort.h>
22 #include <linux/sched/coredump.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/utsname.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/module.h>
28 #include <linux/namei.h>
29 #include <linux/mount.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/tsacct_kern.h>
33 #include <linux/cn_proc.h>
34 #include <linux/audit.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>
41 #include <linux/fs.h>
42 #include <linux/path.h>
43 #include <linux/timekeeping.h>
44 #include <linux/sysctl.h>
45 #include <linux/elf.h>
46
47 #include <linux/uaccess.h>
48 #include <asm/mmu_context.h>
49 #include <asm/tlb.h>
50 #include <asm/exec.h>
51
52 #include <trace/events/task.h>
53 #include "internal.h"
54
55 #include <trace/events/sched.h>
56
57 static bool dump_vma_snapshot(struct coredump_params *cprm);
58 static void free_vma_snapshot(struct coredump_params *cprm);
59
60 #define CORE_FILE_NOTE_SIZE_DEFAULT (4*1024*1024)
61 /* Define a reasonable max cap */
62 #define CORE_FILE_NOTE_SIZE_MAX (16*1024*1024)
63
64 static int core_uses_pid;
65 static unsigned int core_pipe_limit;
66 static unsigned int core_sort_vma;
67 static char core_pattern[CORENAME_MAX_SIZE] = "core";
68 static int core_name_size = CORENAME_MAX_SIZE;
69 unsigned int core_file_note_size_limit = CORE_FILE_NOTE_SIZE_DEFAULT;
70
71 struct core_name {
72 char *corename;
73 int used, size;
74 };
75
expand_corename(struct core_name * cn,int size)76 static int expand_corename(struct core_name *cn, int size)
77 {
78 char *corename;
79
80 size = kmalloc_size_roundup(size);
81 corename = krealloc(cn->corename, size, GFP_KERNEL);
82
83 if (!corename)
84 return -ENOMEM;
85
86 if (size > core_name_size) /* racy but harmless */
87 core_name_size = size;
88
89 cn->size = size;
90 cn->corename = corename;
91 return 0;
92 }
93
cn_vprintf(struct core_name * cn,const char * fmt,va_list arg)94 static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
95 va_list arg)
96 {
97 int free, need;
98 va_list arg_copy;
99
100 again:
101 free = cn->size - cn->used;
102
103 va_copy(arg_copy, arg);
104 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
105 va_end(arg_copy);
106
107 if (need < free) {
108 cn->used += need;
109 return 0;
110 }
111
112 if (!expand_corename(cn, cn->size + need - free + 1))
113 goto again;
114
115 return -ENOMEM;
116 }
117
cn_printf(struct core_name * cn,const char * fmt,...)118 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
119 {
120 va_list arg;
121 int ret;
122
123 va_start(arg, fmt);
124 ret = cn_vprintf(cn, fmt, arg);
125 va_end(arg);
126
127 return ret;
128 }
129
130 static __printf(2, 3)
cn_esc_printf(struct core_name * cn,const char * fmt,...)131 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
132 {
133 int cur = cn->used;
134 va_list arg;
135 int ret;
136
137 va_start(arg, fmt);
138 ret = cn_vprintf(cn, fmt, arg);
139 va_end(arg);
140
141 if (ret == 0) {
142 /*
143 * Ensure that this coredump name component can't cause the
144 * resulting corefile path to consist of a ".." or ".".
145 */
146 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
147 (cn->used - cur == 2 && cn->corename[cur] == '.'
148 && cn->corename[cur+1] == '.'))
149 cn->corename[cur] = '!';
150
151 /*
152 * Empty names are fishy and could be used to create a "//" in a
153 * corefile name, causing the coredump to happen one directory
154 * level too high. Enforce that all components of the core
155 * pattern are at least one character long.
156 */
157 if (cn->used == cur)
158 ret = cn_printf(cn, "!");
159 }
160
161 for (; cur < cn->used; ++cur) {
162 if (cn->corename[cur] == '/')
163 cn->corename[cur] = '!';
164 }
165 return ret;
166 }
167
cn_print_exe_file(struct core_name * cn,bool name_only)168 static int cn_print_exe_file(struct core_name *cn, bool name_only)
169 {
170 struct file *exe_file;
171 char *pathbuf, *path, *ptr;
172 int ret;
173
174 exe_file = get_mm_exe_file(current->mm);
175 if (!exe_file)
176 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
177
178 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
179 if (!pathbuf) {
180 ret = -ENOMEM;
181 goto put_exe_file;
182 }
183
184 path = file_path(exe_file, pathbuf, PATH_MAX);
185 if (IS_ERR(path)) {
186 ret = PTR_ERR(path);
187 goto free_buf;
188 }
189
190 if (name_only) {
191 ptr = strrchr(path, '/');
192 if (ptr)
193 path = ptr + 1;
194 }
195 ret = cn_esc_printf(cn, "%s", path);
196
197 free_buf:
198 kfree(pathbuf);
199 put_exe_file:
200 fput(exe_file);
201 return ret;
202 }
203
204 /* format_corename will inspect the pattern parameter, and output a
205 * name into corename, which must have space for at least
206 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
207 */
format_corename(struct core_name * cn,struct coredump_params * cprm,size_t ** argv,int * argc)208 static int format_corename(struct core_name *cn, struct coredump_params *cprm,
209 size_t **argv, int *argc)
210 {
211 const struct cred *cred = current_cred();
212 const char *pat_ptr = core_pattern;
213 int ispipe = (*pat_ptr == '|');
214 bool was_space = false;
215 int pid_in_pattern = 0;
216 int err = 0;
217
218 cn->used = 0;
219 cn->corename = NULL;
220 if (expand_corename(cn, core_name_size))
221 return -ENOMEM;
222 cn->corename[0] = '\0';
223
224 if (ispipe) {
225 int argvs = sizeof(core_pattern) / 2;
226 (*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
227 if (!(*argv))
228 return -ENOMEM;
229 (*argv)[(*argc)++] = 0;
230 ++pat_ptr;
231 if (!(*pat_ptr))
232 return -ENOMEM;
233 }
234
235 /* Repeat as long as we have more pattern to process and more output
236 space */
237 while (*pat_ptr) {
238 /*
239 * Split on spaces before doing template expansion so that
240 * %e and %E don't get split if they have spaces in them
241 */
242 if (ispipe) {
243 if (isspace(*pat_ptr)) {
244 if (cn->used != 0)
245 was_space = true;
246 pat_ptr++;
247 continue;
248 } else if (was_space) {
249 was_space = false;
250 err = cn_printf(cn, "%c", '\0');
251 if (err)
252 return err;
253 (*argv)[(*argc)++] = cn->used;
254 }
255 }
256 if (*pat_ptr != '%') {
257 err = cn_printf(cn, "%c", *pat_ptr++);
258 } else {
259 switch (*++pat_ptr) {
260 /* single % at the end, drop that */
261 case 0:
262 goto out;
263 /* Double percent, output one percent */
264 case '%':
265 err = cn_printf(cn, "%c", '%');
266 break;
267 /* pid */
268 case 'p':
269 pid_in_pattern = 1;
270 err = cn_printf(cn, "%d",
271 task_tgid_vnr(current));
272 break;
273 /* global pid */
274 case 'P':
275 err = cn_printf(cn, "%d",
276 task_tgid_nr(current));
277 break;
278 case 'i':
279 err = cn_printf(cn, "%d",
280 task_pid_vnr(current));
281 break;
282 case 'I':
283 err = cn_printf(cn, "%d",
284 task_pid_nr(current));
285 break;
286 /* uid */
287 case 'u':
288 err = cn_printf(cn, "%u",
289 from_kuid(&init_user_ns,
290 cred->uid));
291 break;
292 /* gid */
293 case 'g':
294 err = cn_printf(cn, "%u",
295 from_kgid(&init_user_ns,
296 cred->gid));
297 break;
298 case 'd':
299 err = cn_printf(cn, "%d",
300 __get_dumpable(cprm->mm_flags));
301 break;
302 /* signal that caused the coredump */
303 case 's':
304 err = cn_printf(cn, "%d",
305 cprm->siginfo->si_signo);
306 break;
307 /* UNIX time of coredump */
308 case 't': {
309 time64_t time;
310
311 time = ktime_get_real_seconds();
312 err = cn_printf(cn, "%lld", time);
313 break;
314 }
315 /* hostname */
316 case 'h':
317 down_read(&uts_sem);
318 err = cn_esc_printf(cn, "%s",
319 utsname()->nodename);
320 up_read(&uts_sem);
321 break;
322 /* executable, could be changed by prctl PR_SET_NAME etc */
323 case 'e':
324 err = cn_esc_printf(cn, "%s", current->comm);
325 break;
326 /* file name of executable */
327 case 'f':
328 err = cn_print_exe_file(cn, true);
329 break;
330 case 'E':
331 err = cn_print_exe_file(cn, false);
332 break;
333 /* core limit size */
334 case 'c':
335 err = cn_printf(cn, "%lu",
336 rlimit(RLIMIT_CORE));
337 break;
338 /* CPU the task ran on */
339 case 'C':
340 err = cn_printf(cn, "%d", cprm->cpu);
341 break;
342 default:
343 break;
344 }
345 ++pat_ptr;
346 }
347
348 if (err)
349 return err;
350 }
351
352 out:
353 /* Backward compatibility with core_uses_pid:
354 *
355 * If core_pattern does not include a %p (as is the default)
356 * and core_uses_pid is set, then .%pid will be appended to
357 * the filename. Do not do this for piped commands. */
358 if (!ispipe && !pid_in_pattern && core_uses_pid) {
359 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
360 if (err)
361 return err;
362 }
363 return ispipe;
364 }
365
zap_process(struct signal_struct * signal,int exit_code)366 static int zap_process(struct signal_struct *signal, int exit_code)
367 {
368 struct task_struct *t;
369 int nr = 0;
370
371 signal->flags = SIGNAL_GROUP_EXIT;
372 signal->group_exit_code = exit_code;
373 signal->group_stop_count = 0;
374
375 __for_each_thread(signal, t) {
376 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
377 if (t != current && !(t->flags & PF_POSTCOREDUMP)) {
378 sigaddset(&t->pending.signal, SIGKILL);
379 signal_wake_up(t, 1);
380 nr++;
381 }
382 }
383
384 return nr;
385 }
386
zap_threads(struct task_struct * tsk,struct core_state * core_state,int exit_code)387 static int zap_threads(struct task_struct *tsk,
388 struct core_state *core_state, int exit_code)
389 {
390 struct signal_struct *signal = tsk->signal;
391 int nr = -EAGAIN;
392
393 spin_lock_irq(&tsk->sighand->siglock);
394 if (!(signal->flags & SIGNAL_GROUP_EXIT) && !signal->group_exec_task) {
395 /* Allow SIGKILL, see prepare_signal() */
396 signal->core_state = core_state;
397 nr = zap_process(signal, exit_code);
398 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
399 tsk->flags |= PF_DUMPCORE;
400 atomic_set(&core_state->nr_threads, nr);
401 }
402 spin_unlock_irq(&tsk->sighand->siglock);
403 return nr;
404 }
405
coredump_wait(int exit_code,struct core_state * core_state)406 static int coredump_wait(int exit_code, struct core_state *core_state)
407 {
408 struct task_struct *tsk = current;
409 int core_waiters = -EBUSY;
410
411 init_completion(&core_state->startup);
412 core_state->dumper.task = tsk;
413 core_state->dumper.next = NULL;
414
415 core_waiters = zap_threads(tsk, core_state, exit_code);
416 if (core_waiters > 0) {
417 struct core_thread *ptr;
418
419 wait_for_completion_state(&core_state->startup,
420 TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
421 /*
422 * Wait for all the threads to become inactive, so that
423 * all the thread context (extended register state, like
424 * fpu etc) gets copied to the memory.
425 */
426 ptr = core_state->dumper.next;
427 while (ptr != NULL) {
428 wait_task_inactive(ptr->task, TASK_ANY);
429 ptr = ptr->next;
430 }
431 }
432
433 return core_waiters;
434 }
435
coredump_finish(bool core_dumped)436 static void coredump_finish(bool core_dumped)
437 {
438 struct core_thread *curr, *next;
439 struct task_struct *task;
440
441 spin_lock_irq(¤t->sighand->siglock);
442 if (core_dumped && !__fatal_signal_pending(current))
443 current->signal->group_exit_code |= 0x80;
444 next = current->signal->core_state->dumper.next;
445 current->signal->core_state = NULL;
446 spin_unlock_irq(¤t->sighand->siglock);
447
448 while ((curr = next) != NULL) {
449 next = curr->next;
450 task = curr->task;
451 /*
452 * see coredump_task_exit(), curr->task must not see
453 * ->task == NULL before we read ->next.
454 */
455 smp_mb();
456 curr->task = NULL;
457 wake_up_process(task);
458 }
459 }
460
dump_interrupted(void)461 static bool dump_interrupted(void)
462 {
463 /*
464 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
465 * can do try_to_freeze() and check __fatal_signal_pending(),
466 * but then we need to teach dump_write() to restart and clear
467 * TIF_SIGPENDING.
468 */
469 return fatal_signal_pending(current) || freezing(current);
470 }
471
wait_for_dump_helpers(struct file * file)472 static void wait_for_dump_helpers(struct file *file)
473 {
474 struct pipe_inode_info *pipe = file->private_data;
475
476 pipe_lock(pipe);
477 pipe->readers++;
478 pipe->writers--;
479 wake_up_interruptible_sync(&pipe->rd_wait);
480 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
481 pipe_unlock(pipe);
482
483 /*
484 * We actually want wait_event_freezable() but then we need
485 * to clear TIF_SIGPENDING and improve dump_interrupted().
486 */
487 wait_event_interruptible(pipe->rd_wait, pipe->readers == 1);
488
489 pipe_lock(pipe);
490 pipe->readers--;
491 pipe->writers++;
492 pipe_unlock(pipe);
493 }
494
495 /*
496 * umh_pipe_setup
497 * helper function to customize the process used
498 * to collect the core in userspace. Specifically
499 * it sets up a pipe and installs it as fd 0 (stdin)
500 * for the process. Returns 0 on success, or
501 * PTR_ERR on failure.
502 * Note that it also sets the core limit to 1. This
503 * is a special value that we use to trap recursive
504 * core dumps
505 */
umh_pipe_setup(struct subprocess_info * info,struct cred * new)506 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
507 {
508 struct file *files[2];
509 struct coredump_params *cp = (struct coredump_params *)info->data;
510 int err;
511
512 err = create_pipe_files(files, 0);
513 if (err)
514 return err;
515
516 cp->file = files[1];
517
518 err = replace_fd(0, files[0], 0);
519 fput(files[0]);
520 if (err < 0)
521 return err;
522
523 /* and disallow core files too */
524 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
525
526 return 0;
527 }
528
do_coredump(const kernel_siginfo_t * siginfo)529 void do_coredump(const kernel_siginfo_t *siginfo)
530 {
531 struct core_state core_state;
532 struct core_name cn;
533 struct mm_struct *mm = current->mm;
534 struct linux_binfmt * binfmt;
535 const struct cred *old_cred;
536 struct cred *cred;
537 int retval = 0;
538 int ispipe;
539 size_t *argv = NULL;
540 int argc = 0;
541 /* require nonrelative corefile path and be extra careful */
542 bool need_suid_safe = false;
543 bool core_dumped = false;
544 static atomic_t core_dump_count = ATOMIC_INIT(0);
545 struct coredump_params cprm = {
546 .siginfo = siginfo,
547 .limit = rlimit(RLIMIT_CORE),
548 /*
549 * We must use the same mm->flags while dumping core to avoid
550 * inconsistency of bit flags, since this flag is not protected
551 * by any locks.
552 */
553 .mm_flags = mm->flags,
554 .vma_meta = NULL,
555 .cpu = raw_smp_processor_id(),
556 };
557
558 audit_core_dumps(siginfo->si_signo);
559
560 binfmt = mm->binfmt;
561 if (!binfmt || !binfmt->core_dump)
562 goto fail;
563 if (!__get_dumpable(cprm.mm_flags))
564 goto fail;
565
566 cred = prepare_creds();
567 if (!cred)
568 goto fail;
569 /*
570 * We cannot trust fsuid as being the "true" uid of the process
571 * nor do we know its entire history. We only know it was tainted
572 * so we dump it as root in mode 2, and only into a controlled
573 * environment (pipe handler or fully qualified path).
574 */
575 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
576 /* Setuid core dump mode */
577 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
578 need_suid_safe = true;
579 }
580
581 retval = coredump_wait(siginfo->si_signo, &core_state);
582 if (retval < 0)
583 goto fail_creds;
584
585 old_cred = override_creds(cred);
586
587 ispipe = format_corename(&cn, &cprm, &argv, &argc);
588
589 if (ispipe) {
590 int argi;
591 int dump_count;
592 char **helper_argv;
593 struct subprocess_info *sub_info;
594
595 if (ispipe < 0) {
596 coredump_report_failure("format_corename failed, aborting core");
597 goto fail_unlock;
598 }
599
600 if (cprm.limit == 1) {
601 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
602 *
603 * Normally core limits are irrelevant to pipes, since
604 * we're not writing to the file system, but we use
605 * cprm.limit of 1 here as a special value, this is a
606 * consistent way to catch recursive crashes.
607 * We can still crash if the core_pattern binary sets
608 * RLIM_CORE = !1, but it runs as root, and can do
609 * lots of stupid things.
610 *
611 * Note that we use task_tgid_vnr here to grab the pid
612 * of the process group leader. That way we get the
613 * right pid if a thread in a multi-threaded
614 * core_pattern process dies.
615 */
616 coredump_report_failure("RLIMIT_CORE is set to 1, aborting core");
617 goto fail_unlock;
618 }
619 cprm.limit = RLIM_INFINITY;
620
621 dump_count = atomic_inc_return(&core_dump_count);
622 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
623 coredump_report_failure("over core_pipe_limit, skipping core dump");
624 goto fail_dropcount;
625 }
626
627 helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
628 GFP_KERNEL);
629 if (!helper_argv) {
630 coredump_report_failure("%s failed to allocate memory", __func__);
631 goto fail_dropcount;
632 }
633 for (argi = 0; argi < argc; argi++)
634 helper_argv[argi] = cn.corename + argv[argi];
635 helper_argv[argi] = NULL;
636
637 retval = -ENOMEM;
638 sub_info = call_usermodehelper_setup(helper_argv[0],
639 helper_argv, NULL, GFP_KERNEL,
640 umh_pipe_setup, NULL, &cprm);
641 if (sub_info)
642 retval = call_usermodehelper_exec(sub_info,
643 UMH_WAIT_EXEC);
644
645 kfree(helper_argv);
646 if (retval) {
647 coredump_report_failure("|%s pipe failed", cn.corename);
648 goto close_fail;
649 }
650 } else {
651 struct mnt_idmap *idmap;
652 struct inode *inode;
653 int open_flags = O_CREAT | O_WRONLY | O_NOFOLLOW |
654 O_LARGEFILE | O_EXCL;
655
656 if (cprm.limit < binfmt->min_coredump)
657 goto fail_unlock;
658
659 if (need_suid_safe && cn.corename[0] != '/') {
660 coredump_report_failure(
661 "this process can only dump core to a fully qualified path, skipping core dump");
662 goto fail_unlock;
663 }
664
665 /*
666 * Unlink the file if it exists unless this is a SUID
667 * binary - in that case, we're running around with root
668 * privs and don't want to unlink another user's coredump.
669 */
670 if (!need_suid_safe) {
671 /*
672 * If it doesn't exist, that's fine. If there's some
673 * other problem, we'll catch it at the filp_open().
674 */
675 do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
676 }
677
678 /*
679 * There is a race between unlinking and creating the
680 * file, but if that causes an EEXIST here, that's
681 * fine - another process raced with us while creating
682 * the corefile, and the other process won. To userspace,
683 * what matters is that at least one of the two processes
684 * writes its coredump successfully, not which one.
685 */
686 if (need_suid_safe) {
687 /*
688 * Using user namespaces, normal user tasks can change
689 * their current->fs->root to point to arbitrary
690 * directories. Since the intention of the "only dump
691 * with a fully qualified path" rule is to control where
692 * coredumps may be placed using root privileges,
693 * current->fs->root must not be used. Instead, use the
694 * root directory of init_task.
695 */
696 struct path root;
697
698 task_lock(&init_task);
699 get_fs_root(init_task.fs, &root);
700 task_unlock(&init_task);
701 cprm.file = file_open_root(&root, cn.corename,
702 open_flags, 0600);
703 path_put(&root);
704 } else {
705 cprm.file = filp_open(cn.corename, open_flags, 0600);
706 }
707 if (IS_ERR(cprm.file))
708 goto fail_unlock;
709
710 inode = file_inode(cprm.file);
711 if (inode->i_nlink > 1)
712 goto close_fail;
713 if (d_unhashed(cprm.file->f_path.dentry))
714 goto close_fail;
715 /*
716 * AK: actually i see no reason to not allow this for named
717 * pipes etc, but keep the previous behaviour for now.
718 */
719 if (!S_ISREG(inode->i_mode))
720 goto close_fail;
721 /*
722 * Don't dump core if the filesystem changed owner or mode
723 * of the file during file creation. This is an issue when
724 * a process dumps core while its cwd is e.g. on a vfat
725 * filesystem.
726 */
727 idmap = file_mnt_idmap(cprm.file);
728 if (!vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode),
729 current_fsuid())) {
730 coredump_report_failure("Core dump to %s aborted: "
731 "cannot preserve file owner", cn.corename);
732 goto close_fail;
733 }
734 if ((inode->i_mode & 0677) != 0600) {
735 coredump_report_failure("Core dump to %s aborted: "
736 "cannot preserve file permissions", cn.corename);
737 goto close_fail;
738 }
739 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
740 goto close_fail;
741 if (do_truncate(idmap, cprm.file->f_path.dentry,
742 0, 0, cprm.file))
743 goto close_fail;
744 }
745
746 /* get us an unshared descriptor table; almost always a no-op */
747 /* The cell spufs coredump code reads the file descriptor tables */
748 retval = unshare_files();
749 if (retval)
750 goto close_fail;
751 if (!dump_interrupted()) {
752 /*
753 * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would
754 * have this set to NULL.
755 */
756 if (!cprm.file) {
757 coredump_report_failure("Core dump to |%s disabled", cn.corename);
758 goto close_fail;
759 }
760 if (!dump_vma_snapshot(&cprm))
761 goto close_fail;
762
763 file_start_write(cprm.file);
764 core_dumped = binfmt->core_dump(&cprm);
765 /*
766 * Ensures that file size is big enough to contain the current
767 * file postion. This prevents gdb from complaining about
768 * a truncated file if the last "write" to the file was
769 * dump_skip.
770 */
771 if (cprm.to_skip) {
772 cprm.to_skip--;
773 dump_emit(&cprm, "", 1);
774 }
775 file_end_write(cprm.file);
776 free_vma_snapshot(&cprm);
777 }
778 if (ispipe && core_pipe_limit)
779 wait_for_dump_helpers(cprm.file);
780 close_fail:
781 if (cprm.file)
782 filp_close(cprm.file, NULL);
783 fail_dropcount:
784 if (ispipe)
785 atomic_dec(&core_dump_count);
786 fail_unlock:
787 kfree(argv);
788 kfree(cn.corename);
789 coredump_finish(core_dumped);
790 revert_creds(old_cred);
791 fail_creds:
792 put_cred(cred);
793 fail:
794 return;
795 }
796
797 /*
798 * Core dumping helper functions. These are the only things you should
799 * do on a core-file: use only these functions to write out all the
800 * necessary info.
801 */
__dump_emit(struct coredump_params * cprm,const void * addr,int nr)802 static int __dump_emit(struct coredump_params *cprm, const void *addr, int nr)
803 {
804 struct file *file = cprm->file;
805 loff_t pos = file->f_pos;
806 ssize_t n;
807 if (cprm->written + nr > cprm->limit)
808 return 0;
809
810
811 if (dump_interrupted())
812 return 0;
813 n = __kernel_write(file, addr, nr, &pos);
814 if (n != nr)
815 return 0;
816 file->f_pos = pos;
817 cprm->written += n;
818 cprm->pos += n;
819
820 return 1;
821 }
822
__dump_skip(struct coredump_params * cprm,size_t nr)823 static int __dump_skip(struct coredump_params *cprm, size_t nr)
824 {
825 static char zeroes[PAGE_SIZE];
826 struct file *file = cprm->file;
827 if (file->f_mode & FMODE_LSEEK) {
828 if (dump_interrupted() ||
829 vfs_llseek(file, nr, SEEK_CUR) < 0)
830 return 0;
831 cprm->pos += nr;
832 return 1;
833 } else {
834 while (nr > PAGE_SIZE) {
835 if (!__dump_emit(cprm, zeroes, PAGE_SIZE))
836 return 0;
837 nr -= PAGE_SIZE;
838 }
839 return __dump_emit(cprm, zeroes, nr);
840 }
841 }
842
dump_emit(struct coredump_params * cprm,const void * addr,int nr)843 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
844 {
845 if (cprm->to_skip) {
846 if (!__dump_skip(cprm, cprm->to_skip))
847 return 0;
848 cprm->to_skip = 0;
849 }
850 return __dump_emit(cprm, addr, nr);
851 }
852 EXPORT_SYMBOL(dump_emit);
853
dump_skip_to(struct coredump_params * cprm,unsigned long pos)854 void dump_skip_to(struct coredump_params *cprm, unsigned long pos)
855 {
856 cprm->to_skip = pos - cprm->pos;
857 }
858 EXPORT_SYMBOL(dump_skip_to);
859
dump_skip(struct coredump_params * cprm,size_t nr)860 void dump_skip(struct coredump_params *cprm, size_t nr)
861 {
862 cprm->to_skip += nr;
863 }
864 EXPORT_SYMBOL(dump_skip);
865
866 #ifdef CONFIG_ELF_CORE
dump_emit_page(struct coredump_params * cprm,struct page * page)867 static int dump_emit_page(struct coredump_params *cprm, struct page *page)
868 {
869 struct bio_vec bvec;
870 struct iov_iter iter;
871 struct file *file = cprm->file;
872 loff_t pos;
873 ssize_t n;
874
875 if (!page)
876 return 0;
877
878 if (cprm->to_skip) {
879 if (!__dump_skip(cprm, cprm->to_skip))
880 return 0;
881 cprm->to_skip = 0;
882 }
883 if (cprm->written + PAGE_SIZE > cprm->limit)
884 return 0;
885 if (dump_interrupted())
886 return 0;
887 pos = file->f_pos;
888 bvec_set_page(&bvec, page, PAGE_SIZE, 0);
889 iov_iter_bvec(&iter, ITER_SOURCE, &bvec, 1, PAGE_SIZE);
890 n = __kernel_write_iter(cprm->file, &iter, &pos);
891 if (n != PAGE_SIZE)
892 return 0;
893 file->f_pos = pos;
894 cprm->written += PAGE_SIZE;
895 cprm->pos += PAGE_SIZE;
896
897 return 1;
898 }
899
900 /*
901 * If we might get machine checks from kernel accesses during the
902 * core dump, let's get those errors early rather than during the
903 * IO. This is not performance-critical enough to warrant having
904 * all the machine check logic in the iovec paths.
905 */
906 #ifdef copy_mc_to_kernel
907
908 #define dump_page_alloc() alloc_page(GFP_KERNEL)
909 #define dump_page_free(x) __free_page(x)
dump_page_copy(struct page * src,struct page * dst)910 static struct page *dump_page_copy(struct page *src, struct page *dst)
911 {
912 void *buf = kmap_local_page(src);
913 size_t left = copy_mc_to_kernel(page_address(dst), buf, PAGE_SIZE);
914 kunmap_local(buf);
915 return left ? NULL : dst;
916 }
917
918 #else
919
920 /* We just want to return non-NULL; it's never used. */
921 #define dump_page_alloc() ERR_PTR(-EINVAL)
922 #define dump_page_free(x) ((void)(x))
dump_page_copy(struct page * src,struct page * dst)923 static inline struct page *dump_page_copy(struct page *src, struct page *dst)
924 {
925 return src;
926 }
927 #endif
928
dump_user_range(struct coredump_params * cprm,unsigned long start,unsigned long len)929 int dump_user_range(struct coredump_params *cprm, unsigned long start,
930 unsigned long len)
931 {
932 unsigned long addr;
933 struct page *dump_page;
934
935 dump_page = dump_page_alloc();
936 if (!dump_page)
937 return 0;
938
939 for (addr = start; addr < start + len; addr += PAGE_SIZE) {
940 struct page *page;
941
942 /*
943 * To avoid having to allocate page tables for virtual address
944 * ranges that have never been used yet, and also to make it
945 * easy to generate sparse core files, use a helper that returns
946 * NULL when encountering an empty page table entry that would
947 * otherwise have been filled with the zero page.
948 */
949 page = get_dump_page(addr);
950 if (page) {
951 int stop = !dump_emit_page(cprm, dump_page_copy(page, dump_page));
952 put_page(page);
953 if (stop) {
954 dump_page_free(dump_page);
955 return 0;
956 }
957 } else {
958 dump_skip(cprm, PAGE_SIZE);
959 }
960 }
961 dump_page_free(dump_page);
962 return 1;
963 }
964 #endif
965
dump_align(struct coredump_params * cprm,int align)966 int dump_align(struct coredump_params *cprm, int align)
967 {
968 unsigned mod = (cprm->pos + cprm->to_skip) & (align - 1);
969 if (align & (align - 1))
970 return 0;
971 if (mod)
972 cprm->to_skip += align - mod;
973 return 1;
974 }
975 EXPORT_SYMBOL(dump_align);
976
977 #ifdef CONFIG_SYSCTL
978
validate_coredump_safety(void)979 void validate_coredump_safety(void)
980 {
981 if (suid_dumpable == SUID_DUMP_ROOT &&
982 core_pattern[0] != '/' && core_pattern[0] != '|') {
983
984 coredump_report_failure("Unsafe core_pattern used with fs.suid_dumpable=2: "
985 "pipe handler or fully qualified core dump path required. "
986 "Set kernel.core_pattern before fs.suid_dumpable.");
987 }
988 }
989
proc_dostring_coredump(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)990 static int proc_dostring_coredump(const struct ctl_table *table, int write,
991 void *buffer, size_t *lenp, loff_t *ppos)
992 {
993 int error = proc_dostring(table, write, buffer, lenp, ppos);
994
995 if (!error)
996 validate_coredump_safety();
997 return error;
998 }
999
1000 static const unsigned int core_file_note_size_min = CORE_FILE_NOTE_SIZE_DEFAULT;
1001 static const unsigned int core_file_note_size_max = CORE_FILE_NOTE_SIZE_MAX;
1002
1003 static struct ctl_table coredump_sysctls[] = {
1004 {
1005 .procname = "core_uses_pid",
1006 .data = &core_uses_pid,
1007 .maxlen = sizeof(int),
1008 .mode = 0644,
1009 .proc_handler = proc_dointvec,
1010 },
1011 {
1012 .procname = "core_pattern",
1013 .data = core_pattern,
1014 .maxlen = CORENAME_MAX_SIZE,
1015 .mode = 0644,
1016 .proc_handler = proc_dostring_coredump,
1017 },
1018 {
1019 .procname = "core_pipe_limit",
1020 .data = &core_pipe_limit,
1021 .maxlen = sizeof(unsigned int),
1022 .mode = 0644,
1023 .proc_handler = proc_dointvec,
1024 },
1025 {
1026 .procname = "core_file_note_size_limit",
1027 .data = &core_file_note_size_limit,
1028 .maxlen = sizeof(unsigned int),
1029 .mode = 0644,
1030 .proc_handler = proc_douintvec_minmax,
1031 .extra1 = (unsigned int *)&core_file_note_size_min,
1032 .extra2 = (unsigned int *)&core_file_note_size_max,
1033 },
1034 {
1035 .procname = "core_sort_vma",
1036 .data = &core_sort_vma,
1037 .maxlen = sizeof(int),
1038 .mode = 0644,
1039 .proc_handler = proc_douintvec_minmax,
1040 .extra1 = SYSCTL_ZERO,
1041 .extra2 = SYSCTL_ONE,
1042 },
1043 };
1044
init_fs_coredump_sysctls(void)1045 static int __init init_fs_coredump_sysctls(void)
1046 {
1047 register_sysctl_init("kernel", coredump_sysctls);
1048 return 0;
1049 }
1050 fs_initcall(init_fs_coredump_sysctls);
1051 #endif /* CONFIG_SYSCTL */
1052
1053 /*
1054 * The purpose of always_dump_vma() is to make sure that special kernel mappings
1055 * that are useful for post-mortem analysis are included in every core dump.
1056 * In that way we ensure that the core dump is fully interpretable later
1057 * without matching up the same kernel and hardware config to see what PC values
1058 * meant. These special mappings include - vDSO, vsyscall, and other
1059 * architecture specific mappings
1060 */
always_dump_vma(struct vm_area_struct * vma)1061 static bool always_dump_vma(struct vm_area_struct *vma)
1062 {
1063 /* Any vsyscall mappings? */
1064 if (vma == get_gate_vma(vma->vm_mm))
1065 return true;
1066
1067 /*
1068 * Assume that all vmas with a .name op should always be dumped.
1069 * If this changes, a new vm_ops field can easily be added.
1070 */
1071 if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1072 return true;
1073
1074 /*
1075 * arch_vma_name() returns non-NULL for special architecture mappings,
1076 * such as vDSO sections.
1077 */
1078 if (arch_vma_name(vma))
1079 return true;
1080
1081 return false;
1082 }
1083
1084 #define DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER 1
1085
1086 /*
1087 * Decide how much of @vma's contents should be included in a core dump.
1088 */
vma_dump_size(struct vm_area_struct * vma,unsigned long mm_flags)1089 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1090 unsigned long mm_flags)
1091 {
1092 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1093
1094 /* always dump the vdso and vsyscall sections */
1095 if (always_dump_vma(vma))
1096 goto whole;
1097
1098 if (vma->vm_flags & VM_DONTDUMP)
1099 return 0;
1100
1101 /* support for DAX */
1102 if (vma_is_dax(vma)) {
1103 if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1104 goto whole;
1105 if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1106 goto whole;
1107 return 0;
1108 }
1109
1110 /* Hugetlb memory check */
1111 if (is_vm_hugetlb_page(vma)) {
1112 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1113 goto whole;
1114 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1115 goto whole;
1116 return 0;
1117 }
1118
1119 /* Do not dump I/O mapped devices or special mappings */
1120 if (vma->vm_flags & VM_IO)
1121 return 0;
1122
1123 /* By default, dump shared memory if mapped from an anonymous file. */
1124 if (vma->vm_flags & VM_SHARED) {
1125 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1126 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1127 goto whole;
1128 return 0;
1129 }
1130
1131 /* Dump segments that have been written to. */
1132 if ((!IS_ENABLED(CONFIG_MMU) || vma->anon_vma) && FILTER(ANON_PRIVATE))
1133 goto whole;
1134 if (vma->vm_file == NULL)
1135 return 0;
1136
1137 if (FILTER(MAPPED_PRIVATE))
1138 goto whole;
1139
1140 /*
1141 * If this is the beginning of an executable file mapping,
1142 * dump the first page to aid in determining what was mapped here.
1143 */
1144 if (FILTER(ELF_HEADERS) &&
1145 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1146 if ((READ_ONCE(file_inode(vma->vm_file)->i_mode) & 0111) != 0)
1147 return PAGE_SIZE;
1148
1149 /*
1150 * ELF libraries aren't always executable.
1151 * We'll want to check whether the mapping starts with the ELF
1152 * magic, but not now - we're holding the mmap lock,
1153 * so copy_from_user() doesn't work here.
1154 * Use a placeholder instead, and fix it up later in
1155 * dump_vma_snapshot().
1156 */
1157 return DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER;
1158 }
1159
1160 #undef FILTER
1161
1162 return 0;
1163
1164 whole:
1165 return vma->vm_end - vma->vm_start;
1166 }
1167
1168 /*
1169 * Helper function for iterating across a vma list. It ensures that the caller
1170 * will visit `gate_vma' prior to terminating the search.
1171 */
coredump_next_vma(struct vma_iterator * vmi,struct vm_area_struct * vma,struct vm_area_struct * gate_vma)1172 static struct vm_area_struct *coredump_next_vma(struct vma_iterator *vmi,
1173 struct vm_area_struct *vma,
1174 struct vm_area_struct *gate_vma)
1175 {
1176 if (gate_vma && (vma == gate_vma))
1177 return NULL;
1178
1179 vma = vma_next(vmi);
1180 if (vma)
1181 return vma;
1182 return gate_vma;
1183 }
1184
free_vma_snapshot(struct coredump_params * cprm)1185 static void free_vma_snapshot(struct coredump_params *cprm)
1186 {
1187 if (cprm->vma_meta) {
1188 int i;
1189 for (i = 0; i < cprm->vma_count; i++) {
1190 struct file *file = cprm->vma_meta[i].file;
1191 if (file)
1192 fput(file);
1193 }
1194 kvfree(cprm->vma_meta);
1195 cprm->vma_meta = NULL;
1196 }
1197 }
1198
cmp_vma_size(const void * vma_meta_lhs_ptr,const void * vma_meta_rhs_ptr)1199 static int cmp_vma_size(const void *vma_meta_lhs_ptr, const void *vma_meta_rhs_ptr)
1200 {
1201 const struct core_vma_metadata *vma_meta_lhs = vma_meta_lhs_ptr;
1202 const struct core_vma_metadata *vma_meta_rhs = vma_meta_rhs_ptr;
1203
1204 if (vma_meta_lhs->dump_size < vma_meta_rhs->dump_size)
1205 return -1;
1206 if (vma_meta_lhs->dump_size > vma_meta_rhs->dump_size)
1207 return 1;
1208 return 0;
1209 }
1210
1211 /*
1212 * Under the mmap_lock, take a snapshot of relevant information about the task's
1213 * VMAs.
1214 */
dump_vma_snapshot(struct coredump_params * cprm)1215 static bool dump_vma_snapshot(struct coredump_params *cprm)
1216 {
1217 struct vm_area_struct *gate_vma, *vma = NULL;
1218 struct mm_struct *mm = current->mm;
1219 VMA_ITERATOR(vmi, mm, 0);
1220 int i = 0;
1221
1222 /*
1223 * Once the stack expansion code is fixed to not change VMA bounds
1224 * under mmap_lock in read mode, this can be changed to take the
1225 * mmap_lock in read mode.
1226 */
1227 if (mmap_write_lock_killable(mm))
1228 return false;
1229
1230 cprm->vma_data_size = 0;
1231 gate_vma = get_gate_vma(mm);
1232 cprm->vma_count = mm->map_count + (gate_vma ? 1 : 0);
1233
1234 cprm->vma_meta = kvmalloc_array(cprm->vma_count, sizeof(*cprm->vma_meta), GFP_KERNEL);
1235 if (!cprm->vma_meta) {
1236 mmap_write_unlock(mm);
1237 return false;
1238 }
1239
1240 while ((vma = coredump_next_vma(&vmi, vma, gate_vma)) != NULL) {
1241 struct core_vma_metadata *m = cprm->vma_meta + i;
1242
1243 m->start = vma->vm_start;
1244 m->end = vma->vm_end;
1245 m->flags = vma->vm_flags;
1246 m->dump_size = vma_dump_size(vma, cprm->mm_flags);
1247 m->pgoff = vma->vm_pgoff;
1248 m->file = vma->vm_file;
1249 if (m->file)
1250 get_file(m->file);
1251 i++;
1252 }
1253
1254 mmap_write_unlock(mm);
1255
1256 for (i = 0; i < cprm->vma_count; i++) {
1257 struct core_vma_metadata *m = cprm->vma_meta + i;
1258
1259 if (m->dump_size == DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER) {
1260 char elfmag[SELFMAG];
1261
1262 if (copy_from_user(elfmag, (void __user *)m->start, SELFMAG) ||
1263 memcmp(elfmag, ELFMAG, SELFMAG) != 0) {
1264 m->dump_size = 0;
1265 } else {
1266 m->dump_size = PAGE_SIZE;
1267 }
1268 }
1269
1270 cprm->vma_data_size += m->dump_size;
1271 }
1272
1273 if (core_sort_vma)
1274 sort(cprm->vma_meta, cprm->vma_count, sizeof(*cprm->vma_meta),
1275 cmp_vma_size, NULL);
1276
1277 return true;
1278 }
1279