coredump: use from_kuid/kgid when formatting corename
[deliverable/linux.git] / fs / coredump.c
1 #include <linux/slab.h>
2 #include <linux/file.h>
3 #include <linux/fdtable.h>
4 #include <linux/mm.h>
5 #include <linux/stat.h>
6 #include <linux/fcntl.h>
7 #include <linux/swap.h>
8 #include <linux/string.h>
9 #include <linux/init.h>
10 #include <linux/pagemap.h>
11 #include <linux/perf_event.h>
12 #include <linux/highmem.h>
13 #include <linux/spinlock.h>
14 #include <linux/key.h>
15 #include <linux/personality.h>
16 #include <linux/binfmts.h>
17 #include <linux/coredump.h>
18 #include <linux/utsname.h>
19 #include <linux/pid_namespace.h>
20 #include <linux/module.h>
21 #include <linux/namei.h>
22 #include <linux/mount.h>
23 #include <linux/security.h>
24 #include <linux/syscalls.h>
25 #include <linux/tsacct_kern.h>
26 #include <linux/cn_proc.h>
27 #include <linux/audit.h>
28 #include <linux/tracehook.h>
29 #include <linux/kmod.h>
30 #include <linux/fsnotify.h>
31 #include <linux/fs_struct.h>
32 #include <linux/pipe_fs_i.h>
33 #include <linux/oom.h>
34 #include <linux/compat.h>
35
36 #include <asm/uaccess.h>
37 #include <asm/mmu_context.h>
38 #include <asm/tlb.h>
39 #include <asm/exec.h>
40
41 #include <trace/events/task.h>
42 #include "internal.h"
43
44 #include <trace/events/sched.h>
45
46 int core_uses_pid;
47 unsigned int core_pipe_limit;
48 char core_pattern[CORENAME_MAX_SIZE] = "core";
49 static int core_name_size = CORENAME_MAX_SIZE;
50
51 struct core_name {
52 char *corename;
53 int used, size;
54 };
55
56 /* The maximal length of core_pattern is also specified in sysctl.c */
57
58 static int expand_corename(struct core_name *cn, int size)
59 {
60 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
61
62 if (!corename)
63 return -ENOMEM;
64
65 if (size > core_name_size) /* racy but harmless */
66 core_name_size = size;
67
68 cn->size = ksize(corename);
69 cn->corename = corename;
70 return 0;
71 }
72
73 static int cn_vprintf(struct core_name *cn, const char *fmt, va_list arg)
74 {
75 int free, need;
76 va_list arg_copy;
77
78 again:
79 free = cn->size - cn->used;
80
81 va_copy(arg_copy, arg);
82 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
83 va_end(arg_copy);
84
85 if (need < free) {
86 cn->used += need;
87 return 0;
88 }
89
90 if (!expand_corename(cn, cn->size + need - free + 1))
91 goto again;
92
93 return -ENOMEM;
94 }
95
96 static int cn_printf(struct core_name *cn, const char *fmt, ...)
97 {
98 va_list arg;
99 int ret;
100
101 va_start(arg, fmt);
102 ret = cn_vprintf(cn, fmt, arg);
103 va_end(arg);
104
105 return ret;
106 }
107
108 static int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
109 {
110 int cur = cn->used;
111 va_list arg;
112 int ret;
113
114 va_start(arg, fmt);
115 ret = cn_vprintf(cn, fmt, arg);
116 va_end(arg);
117
118 for (; cur < cn->used; ++cur) {
119 if (cn->corename[cur] == '/')
120 cn->corename[cur] = '!';
121 }
122 return ret;
123 }
124
125 static int cn_print_exe_file(struct core_name *cn)
126 {
127 struct file *exe_file;
128 char *pathbuf, *path;
129 int ret;
130
131 exe_file = get_mm_exe_file(current->mm);
132 if (!exe_file)
133 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
134
135 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
136 if (!pathbuf) {
137 ret = -ENOMEM;
138 goto put_exe_file;
139 }
140
141 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
142 if (IS_ERR(path)) {
143 ret = PTR_ERR(path);
144 goto free_buf;
145 }
146
147 ret = cn_esc_printf(cn, "%s", path);
148
149 free_buf:
150 kfree(pathbuf);
151 put_exe_file:
152 fput(exe_file);
153 return ret;
154 }
155
156 /* format_corename will inspect the pattern parameter, and output a
157 * name into corename, which must have space for at least
158 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
159 */
160 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
161 {
162 const struct cred *cred = current_cred();
163 const char *pat_ptr = core_pattern;
164 int ispipe = (*pat_ptr == '|');
165 int pid_in_pattern = 0;
166 int err = 0;
167
168 cn->used = 0;
169 cn->corename = NULL;
170 if (expand_corename(cn, core_name_size))
171 return -ENOMEM;
172 cn->corename[0] = '\0';
173
174 if (ispipe)
175 ++pat_ptr;
176
177 /* Repeat as long as we have more pattern to process and more output
178 space */
179 while (*pat_ptr) {
180 if (*pat_ptr != '%') {
181 err = cn_printf(cn, "%c", *pat_ptr++);
182 } else {
183 switch (*++pat_ptr) {
184 /* single % at the end, drop that */
185 case 0:
186 goto out;
187 /* Double percent, output one percent */
188 case '%':
189 err = cn_printf(cn, "%c", '%');
190 break;
191 /* pid */
192 case 'p':
193 pid_in_pattern = 1;
194 err = cn_printf(cn, "%d",
195 task_tgid_vnr(current));
196 break;
197 /* global pid */
198 case 'P':
199 err = cn_printf(cn, "%d",
200 task_tgid_nr(current));
201 break;
202 case 'i':
203 err = cn_printf(cn, "%d",
204 task_pid_vnr(current));
205 break;
206 case 'I':
207 err = cn_printf(cn, "%d",
208 task_pid_nr(current));
209 break;
210 /* uid */
211 case 'u':
212 err = cn_printf(cn, "%u",
213 from_kuid(&init_user_ns,
214 cred->uid));
215 break;
216 /* gid */
217 case 'g':
218 err = cn_printf(cn, "%u",
219 from_kgid(&init_user_ns,
220 cred->gid));
221 break;
222 case 'd':
223 err = cn_printf(cn, "%d",
224 __get_dumpable(cprm->mm_flags));
225 break;
226 /* signal that caused the coredump */
227 case 's':
228 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
229 break;
230 /* UNIX time of coredump */
231 case 't': {
232 struct timeval tv;
233 do_gettimeofday(&tv);
234 err = cn_printf(cn, "%lu", tv.tv_sec);
235 break;
236 }
237 /* hostname */
238 case 'h':
239 down_read(&uts_sem);
240 err = cn_esc_printf(cn, "%s",
241 utsname()->nodename);
242 up_read(&uts_sem);
243 break;
244 /* executable */
245 case 'e':
246 err = cn_esc_printf(cn, "%s", current->comm);
247 break;
248 case 'E':
249 err = cn_print_exe_file(cn);
250 break;
251 /* core limit size */
252 case 'c':
253 err = cn_printf(cn, "%lu",
254 rlimit(RLIMIT_CORE));
255 break;
256 default:
257 break;
258 }
259 ++pat_ptr;
260 }
261
262 if (err)
263 return err;
264 }
265
266 out:
267 /* Backward compatibility with core_uses_pid:
268 *
269 * If core_pattern does not include a %p (as is the default)
270 * and core_uses_pid is set, then .%pid will be appended to
271 * the filename. Do not do this for piped commands. */
272 if (!ispipe && !pid_in_pattern && core_uses_pid) {
273 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
274 if (err)
275 return err;
276 }
277 return ispipe;
278 }
279
280 static int zap_process(struct task_struct *start, int exit_code)
281 {
282 struct task_struct *t;
283 int nr = 0;
284
285 start->signal->group_exit_code = exit_code;
286 start->signal->group_stop_count = 0;
287
288 t = start;
289 do {
290 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
291 if (t != current && t->mm) {
292 sigaddset(&t->pending.signal, SIGKILL);
293 signal_wake_up(t, 1);
294 nr++;
295 }
296 } while_each_thread(start, t);
297
298 return nr;
299 }
300
301 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
302 struct core_state *core_state, int exit_code)
303 {
304 struct task_struct *g, *p;
305 unsigned long flags;
306 int nr = -EAGAIN;
307
308 spin_lock_irq(&tsk->sighand->siglock);
309 if (!signal_group_exit(tsk->signal)) {
310 mm->core_state = core_state;
311 nr = zap_process(tsk, exit_code);
312 tsk->signal->group_exit_task = tsk;
313 /* ignore all signals except SIGKILL, see prepare_signal() */
314 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
315 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
316 }
317 spin_unlock_irq(&tsk->sighand->siglock);
318 if (unlikely(nr < 0))
319 return nr;
320
321 tsk->flags |= PF_DUMPCORE;
322 if (atomic_read(&mm->mm_users) == nr + 1)
323 goto done;
324 /*
325 * We should find and kill all tasks which use this mm, and we should
326 * count them correctly into ->nr_threads. We don't take tasklist
327 * lock, but this is safe wrt:
328 *
329 * fork:
330 * None of sub-threads can fork after zap_process(leader). All
331 * processes which were created before this point should be
332 * visible to zap_threads() because copy_process() adds the new
333 * process to the tail of init_task.tasks list, and lock/unlock
334 * of ->siglock provides a memory barrier.
335 *
336 * do_exit:
337 * The caller holds mm->mmap_sem. This means that the task which
338 * uses this mm can't pass exit_mm(), so it can't exit or clear
339 * its ->mm.
340 *
341 * de_thread:
342 * It does list_replace_rcu(&leader->tasks, &current->tasks),
343 * we must see either old or new leader, this does not matter.
344 * However, it can change p->sighand, so lock_task_sighand(p)
345 * must be used. Since p->mm != NULL and we hold ->mmap_sem
346 * it can't fail.
347 *
348 * Note also that "g" can be the old leader with ->mm == NULL
349 * and already unhashed and thus removed from ->thread_group.
350 * This is OK, __unhash_process()->list_del_rcu() does not
351 * clear the ->next pointer, we will find the new leader via
352 * next_thread().
353 */
354 rcu_read_lock();
355 for_each_process(g) {
356 if (g == tsk->group_leader)
357 continue;
358 if (g->flags & PF_KTHREAD)
359 continue;
360 p = g;
361 do {
362 if (p->mm) {
363 if (unlikely(p->mm == mm)) {
364 lock_task_sighand(p, &flags);
365 nr += zap_process(p, exit_code);
366 p->signal->flags = SIGNAL_GROUP_EXIT;
367 unlock_task_sighand(p, &flags);
368 }
369 break;
370 }
371 } while_each_thread(g, p);
372 }
373 rcu_read_unlock();
374 done:
375 atomic_set(&core_state->nr_threads, nr);
376 return nr;
377 }
378
379 static int coredump_wait(int exit_code, struct core_state *core_state)
380 {
381 struct task_struct *tsk = current;
382 struct mm_struct *mm = tsk->mm;
383 int core_waiters = -EBUSY;
384
385 init_completion(&core_state->startup);
386 core_state->dumper.task = tsk;
387 core_state->dumper.next = NULL;
388
389 down_write(&mm->mmap_sem);
390 if (!mm->core_state)
391 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
392 up_write(&mm->mmap_sem);
393
394 if (core_waiters > 0) {
395 struct core_thread *ptr;
396
397 wait_for_completion(&core_state->startup);
398 /*
399 * Wait for all the threads to become inactive, so that
400 * all the thread context (extended register state, like
401 * fpu etc) gets copied to the memory.
402 */
403 ptr = core_state->dumper.next;
404 while (ptr != NULL) {
405 wait_task_inactive(ptr->task, 0);
406 ptr = ptr->next;
407 }
408 }
409
410 return core_waiters;
411 }
412
413 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
414 {
415 struct core_thread *curr, *next;
416 struct task_struct *task;
417
418 spin_lock_irq(&current->sighand->siglock);
419 if (core_dumped && !__fatal_signal_pending(current))
420 current->signal->group_exit_code |= 0x80;
421 current->signal->group_exit_task = NULL;
422 current->signal->flags = SIGNAL_GROUP_EXIT;
423 spin_unlock_irq(&current->sighand->siglock);
424
425 next = mm->core_state->dumper.next;
426 while ((curr = next) != NULL) {
427 next = curr->next;
428 task = curr->task;
429 /*
430 * see exit_mm(), curr->task must not see
431 * ->task == NULL before we read ->next.
432 */
433 smp_mb();
434 curr->task = NULL;
435 wake_up_process(task);
436 }
437
438 mm->core_state = NULL;
439 }
440
441 static bool dump_interrupted(void)
442 {
443 /*
444 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
445 * can do try_to_freeze() and check __fatal_signal_pending(),
446 * but then we need to teach dump_write() to restart and clear
447 * TIF_SIGPENDING.
448 */
449 return signal_pending(current);
450 }
451
452 static void wait_for_dump_helpers(struct file *file)
453 {
454 struct pipe_inode_info *pipe = file->private_data;
455
456 pipe_lock(pipe);
457 pipe->readers++;
458 pipe->writers--;
459 wake_up_interruptible_sync(&pipe->wait);
460 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
461 pipe_unlock(pipe);
462
463 /*
464 * We actually want wait_event_freezable() but then we need
465 * to clear TIF_SIGPENDING and improve dump_interrupted().
466 */
467 wait_event_interruptible(pipe->wait, pipe->readers == 1);
468
469 pipe_lock(pipe);
470 pipe->readers--;
471 pipe->writers++;
472 pipe_unlock(pipe);
473 }
474
475 /*
476 * umh_pipe_setup
477 * helper function to customize the process used
478 * to collect the core in userspace. Specifically
479 * it sets up a pipe and installs it as fd 0 (stdin)
480 * for the process. Returns 0 on success, or
481 * PTR_ERR on failure.
482 * Note that it also sets the core limit to 1. This
483 * is a special value that we use to trap recursive
484 * core dumps
485 */
486 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
487 {
488 struct file *files[2];
489 struct coredump_params *cp = (struct coredump_params *)info->data;
490 int err = create_pipe_files(files, 0);
491 if (err)
492 return err;
493
494 cp->file = files[1];
495
496 err = replace_fd(0, files[0], 0);
497 fput(files[0]);
498 /* and disallow core files too */
499 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
500
501 return err;
502 }
503
504 void do_coredump(const siginfo_t *siginfo)
505 {
506 struct core_state core_state;
507 struct core_name cn;
508 struct mm_struct *mm = current->mm;
509 struct linux_binfmt * binfmt;
510 const struct cred *old_cred;
511 struct cred *cred;
512 int retval = 0;
513 int flag = 0;
514 int ispipe;
515 struct files_struct *displaced;
516 bool need_nonrelative = false;
517 bool core_dumped = false;
518 static atomic_t core_dump_count = ATOMIC_INIT(0);
519 struct coredump_params cprm = {
520 .siginfo = siginfo,
521 .regs = signal_pt_regs(),
522 .limit = rlimit(RLIMIT_CORE),
523 /*
524 * We must use the same mm->flags while dumping core to avoid
525 * inconsistency of bit flags, since this flag is not protected
526 * by any locks.
527 */
528 .mm_flags = mm->flags,
529 };
530
531 audit_core_dumps(siginfo->si_signo);
532
533 binfmt = mm->binfmt;
534 if (!binfmt || !binfmt->core_dump)
535 goto fail;
536 if (!__get_dumpable(cprm.mm_flags))
537 goto fail;
538
539 cred = prepare_creds();
540 if (!cred)
541 goto fail;
542 /*
543 * We cannot trust fsuid as being the "true" uid of the process
544 * nor do we know its entire history. We only know it was tainted
545 * so we dump it as root in mode 2, and only into a controlled
546 * environment (pipe handler or fully qualified path).
547 */
548 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
549 /* Setuid core dump mode */
550 flag = O_EXCL; /* Stop rewrite attacks */
551 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
552 need_nonrelative = true;
553 }
554
555 retval = coredump_wait(siginfo->si_signo, &core_state);
556 if (retval < 0)
557 goto fail_creds;
558
559 old_cred = override_creds(cred);
560
561 ispipe = format_corename(&cn, &cprm);
562
563 if (ispipe) {
564 int dump_count;
565 char **helper_argv;
566 struct subprocess_info *sub_info;
567
568 if (ispipe < 0) {
569 printk(KERN_WARNING "format_corename failed\n");
570 printk(KERN_WARNING "Aborting core\n");
571 goto fail_unlock;
572 }
573
574 if (cprm.limit == 1) {
575 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
576 *
577 * Normally core limits are irrelevant to pipes, since
578 * we're not writing to the file system, but we use
579 * cprm.limit of 1 here as a special value, this is a
580 * consistent way to catch recursive crashes.
581 * We can still crash if the core_pattern binary sets
582 * RLIM_CORE = !1, but it runs as root, and can do
583 * lots of stupid things.
584 *
585 * Note that we use task_tgid_vnr here to grab the pid
586 * of the process group leader. That way we get the
587 * right pid if a thread in a multi-threaded
588 * core_pattern process dies.
589 */
590 printk(KERN_WARNING
591 "Process %d(%s) has RLIMIT_CORE set to 1\n",
592 task_tgid_vnr(current), current->comm);
593 printk(KERN_WARNING "Aborting core\n");
594 goto fail_unlock;
595 }
596 cprm.limit = RLIM_INFINITY;
597
598 dump_count = atomic_inc_return(&core_dump_count);
599 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
600 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
601 task_tgid_vnr(current), current->comm);
602 printk(KERN_WARNING "Skipping core dump\n");
603 goto fail_dropcount;
604 }
605
606 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
607 if (!helper_argv) {
608 printk(KERN_WARNING "%s failed to allocate memory\n",
609 __func__);
610 goto fail_dropcount;
611 }
612
613 retval = -ENOMEM;
614 sub_info = call_usermodehelper_setup(helper_argv[0],
615 helper_argv, NULL, GFP_KERNEL,
616 umh_pipe_setup, NULL, &cprm);
617 if (sub_info)
618 retval = call_usermodehelper_exec(sub_info,
619 UMH_WAIT_EXEC);
620
621 argv_free(helper_argv);
622 if (retval) {
623 printk(KERN_INFO "Core dump to |%s pipe failed\n",
624 cn.corename);
625 goto close_fail;
626 }
627 } else {
628 struct inode *inode;
629
630 if (cprm.limit < binfmt->min_coredump)
631 goto fail_unlock;
632
633 if (need_nonrelative && cn.corename[0] != '/') {
634 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
635 "to fully qualified path!\n",
636 task_tgid_vnr(current), current->comm);
637 printk(KERN_WARNING "Skipping core dump\n");
638 goto fail_unlock;
639 }
640
641 cprm.file = filp_open(cn.corename,
642 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
643 0600);
644 if (IS_ERR(cprm.file))
645 goto fail_unlock;
646
647 inode = file_inode(cprm.file);
648 if (inode->i_nlink > 1)
649 goto close_fail;
650 if (d_unhashed(cprm.file->f_path.dentry))
651 goto close_fail;
652 /*
653 * AK: actually i see no reason to not allow this for named
654 * pipes etc, but keep the previous behaviour for now.
655 */
656 if (!S_ISREG(inode->i_mode))
657 goto close_fail;
658 /*
659 * Dont allow local users get cute and trick others to coredump
660 * into their pre-created files.
661 */
662 if (!uid_eq(inode->i_uid, current_fsuid()))
663 goto close_fail;
664 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
665 goto close_fail;
666 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
667 goto close_fail;
668 }
669
670 /* get us an unshared descriptor table; almost always a no-op */
671 retval = unshare_files(&displaced);
672 if (retval)
673 goto close_fail;
674 if (displaced)
675 put_files_struct(displaced);
676 if (!dump_interrupted()) {
677 file_start_write(cprm.file);
678 core_dumped = binfmt->core_dump(&cprm);
679 file_end_write(cprm.file);
680 }
681 if (ispipe && core_pipe_limit)
682 wait_for_dump_helpers(cprm.file);
683 close_fail:
684 if (cprm.file)
685 filp_close(cprm.file, NULL);
686 fail_dropcount:
687 if (ispipe)
688 atomic_dec(&core_dump_count);
689 fail_unlock:
690 kfree(cn.corename);
691 coredump_finish(mm, core_dumped);
692 revert_creds(old_cred);
693 fail_creds:
694 put_cred(cred);
695 fail:
696 return;
697 }
698
699 /*
700 * Core dumping helper functions. These are the only things you should
701 * do on a core-file: use only these functions to write out all the
702 * necessary info.
703 */
704 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
705 {
706 struct file *file = cprm->file;
707 loff_t pos = file->f_pos;
708 ssize_t n;
709 if (cprm->written + nr > cprm->limit)
710 return 0;
711 while (nr) {
712 if (dump_interrupted())
713 return 0;
714 n = __kernel_write(file, addr, nr, &pos);
715 if (n <= 0)
716 return 0;
717 file->f_pos = pos;
718 cprm->written += n;
719 nr -= n;
720 }
721 return 1;
722 }
723 EXPORT_SYMBOL(dump_emit);
724
725 int dump_skip(struct coredump_params *cprm, size_t nr)
726 {
727 static char zeroes[PAGE_SIZE];
728 struct file *file = cprm->file;
729 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
730 if (cprm->written + nr > cprm->limit)
731 return 0;
732 if (dump_interrupted() ||
733 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
734 return 0;
735 cprm->written += nr;
736 return 1;
737 } else {
738 while (nr > PAGE_SIZE) {
739 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
740 return 0;
741 nr -= PAGE_SIZE;
742 }
743 return dump_emit(cprm, zeroes, nr);
744 }
745 }
746 EXPORT_SYMBOL(dump_skip);
747
748 int dump_align(struct coredump_params *cprm, int align)
749 {
750 unsigned mod = cprm->written & (align - 1);
751 if (align & (align - 1))
752 return 0;
753 return mod ? dump_skip(cprm, align - mod) : 1;
754 }
755 EXPORT_SYMBOL(dump_align);
This page took 0.056948 seconds and 6 git commands to generate.