Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / fs / coredump.c
blobe3ad709a4232f414b91fe1ed18989a07d0f7aaa3
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>
36 #include <asm/uaccess.h>
37 #include <asm/mmu_context.h>
38 #include <asm/tlb.h>
39 #include <asm/exec.h>
41 #include <trace/events/task.h>
42 #include "internal.h"
44 #include <trace/events/sched.h>
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;
51 struct core_name {
52 char *corename;
53 int used, size;
56 /* The maximal length of core_pattern is also specified in sysctl.c */
58 static int expand_corename(struct core_name *cn, int size)
60 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
62 if (!corename)
63 return -ENOMEM;
65 if (size > core_name_size) /* racy but harmless */
66 core_name_size = size;
68 cn->size = ksize(corename);
69 cn->corename = corename;
70 return 0;
73 static int cn_vprintf(struct core_name *cn, const char *fmt, va_list arg)
75 int free, need;
77 again:
78 free = cn->size - cn->used;
79 need = vsnprintf(cn->corename + cn->used, free, fmt, arg);
80 if (need < free) {
81 cn->used += need;
82 return 0;
85 if (!expand_corename(cn, cn->size + need - free + 1))
86 goto again;
88 return -ENOMEM;
91 static int cn_printf(struct core_name *cn, const char *fmt, ...)
93 va_list arg;
94 int ret;
96 va_start(arg, fmt);
97 ret = cn_vprintf(cn, fmt, arg);
98 va_end(arg);
100 return ret;
103 static int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
105 int cur = cn->used;
106 va_list arg;
107 int ret;
109 va_start(arg, fmt);
110 ret = cn_vprintf(cn, fmt, arg);
111 va_end(arg);
113 for (; cur < cn->used; ++cur) {
114 if (cn->corename[cur] == '/')
115 cn->corename[cur] = '!';
117 return ret;
120 static int cn_print_exe_file(struct core_name *cn)
122 struct file *exe_file;
123 char *pathbuf, *path;
124 int ret;
126 exe_file = get_mm_exe_file(current->mm);
127 if (!exe_file)
128 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
130 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
131 if (!pathbuf) {
132 ret = -ENOMEM;
133 goto put_exe_file;
136 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
137 if (IS_ERR(path)) {
138 ret = PTR_ERR(path);
139 goto free_buf;
142 ret = cn_esc_printf(cn, "%s", path);
144 free_buf:
145 kfree(pathbuf);
146 put_exe_file:
147 fput(exe_file);
148 return ret;
151 /* format_corename will inspect the pattern parameter, and output a
152 * name into corename, which must have space for at least
153 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
155 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
157 const struct cred *cred = current_cred();
158 const char *pat_ptr = core_pattern;
159 int ispipe = (*pat_ptr == '|');
160 int pid_in_pattern = 0;
161 int err = 0;
163 cn->used = 0;
164 cn->corename = NULL;
165 if (expand_corename(cn, core_name_size))
166 return -ENOMEM;
167 cn->corename[0] = '\0';
169 if (ispipe)
170 ++pat_ptr;
172 /* Repeat as long as we have more pattern to process and more output
173 space */
174 while (*pat_ptr) {
175 if (*pat_ptr != '%') {
176 err = cn_printf(cn, "%c", *pat_ptr++);
177 } else {
178 switch (*++pat_ptr) {
179 /* single % at the end, drop that */
180 case 0:
181 goto out;
182 /* Double percent, output one percent */
183 case '%':
184 err = cn_printf(cn, "%c", '%');
185 break;
186 /* pid */
187 case 'p':
188 pid_in_pattern = 1;
189 err = cn_printf(cn, "%d",
190 task_tgid_vnr(current));
191 break;
192 /* global pid */
193 case 'P':
194 err = cn_printf(cn, "%d",
195 task_tgid_nr(current));
196 break;
197 /* uid */
198 case 'u':
199 err = cn_printf(cn, "%d", cred->uid);
200 break;
201 /* gid */
202 case 'g':
203 err = cn_printf(cn, "%d", cred->gid);
204 break;
205 case 'd':
206 err = cn_printf(cn, "%d",
207 __get_dumpable(cprm->mm_flags));
208 break;
209 /* signal that caused the coredump */
210 case 's':
211 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
212 break;
213 /* UNIX time of coredump */
214 case 't': {
215 struct timeval tv;
216 do_gettimeofday(&tv);
217 err = cn_printf(cn, "%lu", tv.tv_sec);
218 break;
220 /* hostname */
221 case 'h':
222 down_read(&uts_sem);
223 err = cn_esc_printf(cn, "%s",
224 utsname()->nodename);
225 up_read(&uts_sem);
226 break;
227 /* executable */
228 case 'e':
229 err = cn_esc_printf(cn, "%s", current->comm);
230 break;
231 case 'E':
232 err = cn_print_exe_file(cn);
233 break;
234 /* core limit size */
235 case 'c':
236 err = cn_printf(cn, "%lu",
237 rlimit(RLIMIT_CORE));
238 break;
239 default:
240 break;
242 ++pat_ptr;
245 if (err)
246 return err;
249 out:
250 /* Backward compatibility with core_uses_pid:
252 * If core_pattern does not include a %p (as is the default)
253 * and core_uses_pid is set, then .%pid will be appended to
254 * the filename. Do not do this for piped commands. */
255 if (!ispipe && !pid_in_pattern && core_uses_pid) {
256 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
257 if (err)
258 return err;
260 return ispipe;
263 static int zap_process(struct task_struct *start, int exit_code)
265 struct task_struct *t;
266 int nr = 0;
268 start->signal->group_exit_code = exit_code;
269 start->signal->group_stop_count = 0;
271 t = start;
272 do {
273 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
274 if (t != current && t->mm) {
275 sigaddset(&t->pending.signal, SIGKILL);
276 signal_wake_up(t, 1);
277 nr++;
279 } while_each_thread(start, t);
281 return nr;
284 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
285 struct core_state *core_state, int exit_code)
287 struct task_struct *g, *p;
288 unsigned long flags;
289 int nr = -EAGAIN;
291 spin_lock_irq(&tsk->sighand->siglock);
292 if (!signal_group_exit(tsk->signal)) {
293 mm->core_state = core_state;
294 nr = zap_process(tsk, exit_code);
295 tsk->signal->group_exit_task = tsk;
296 /* ignore all signals except SIGKILL, see prepare_signal() */
297 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
298 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
300 spin_unlock_irq(&tsk->sighand->siglock);
301 if (unlikely(nr < 0))
302 return nr;
304 tsk->flags = PF_DUMPCORE;
305 if (atomic_read(&mm->mm_users) == nr + 1)
306 goto done;
308 * We should find and kill all tasks which use this mm, and we should
309 * count them correctly into ->nr_threads. We don't take tasklist
310 * lock, but this is safe wrt:
312 * fork:
313 * None of sub-threads can fork after zap_process(leader). All
314 * processes which were created before this point should be
315 * visible to zap_threads() because copy_process() adds the new
316 * process to the tail of init_task.tasks list, and lock/unlock
317 * of ->siglock provides a memory barrier.
319 * do_exit:
320 * The caller holds mm->mmap_sem. This means that the task which
321 * uses this mm can't pass exit_mm(), so it can't exit or clear
322 * its ->mm.
324 * de_thread:
325 * It does list_replace_rcu(&leader->tasks, &current->tasks),
326 * we must see either old or new leader, this does not matter.
327 * However, it can change p->sighand, so lock_task_sighand(p)
328 * must be used. Since p->mm != NULL and we hold ->mmap_sem
329 * it can't fail.
331 * Note also that "g" can be the old leader with ->mm == NULL
332 * and already unhashed and thus removed from ->thread_group.
333 * This is OK, __unhash_process()->list_del_rcu() does not
334 * clear the ->next pointer, we will find the new leader via
335 * next_thread().
337 rcu_read_lock();
338 for_each_process(g) {
339 if (g == tsk->group_leader)
340 continue;
341 if (g->flags & PF_KTHREAD)
342 continue;
343 p = g;
344 do {
345 if (p->mm) {
346 if (unlikely(p->mm == mm)) {
347 lock_task_sighand(p, &flags);
348 nr += zap_process(p, exit_code);
349 p->signal->flags = SIGNAL_GROUP_EXIT;
350 unlock_task_sighand(p, &flags);
352 break;
354 } while_each_thread(g, p);
356 rcu_read_unlock();
357 done:
358 atomic_set(&core_state->nr_threads, nr);
359 return nr;
362 static int coredump_wait(int exit_code, struct core_state *core_state)
364 struct task_struct *tsk = current;
365 struct mm_struct *mm = tsk->mm;
366 int core_waiters = -EBUSY;
368 init_completion(&core_state->startup);
369 core_state->dumper.task = tsk;
370 core_state->dumper.next = NULL;
372 down_write(&mm->mmap_sem);
373 if (!mm->core_state)
374 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
375 up_write(&mm->mmap_sem);
377 if (core_waiters > 0) {
378 struct core_thread *ptr;
380 wait_for_completion(&core_state->startup);
382 * Wait for all the threads to become inactive, so that
383 * all the thread context (extended register state, like
384 * fpu etc) gets copied to the memory.
386 ptr = core_state->dumper.next;
387 while (ptr != NULL) {
388 wait_task_inactive(ptr->task, 0);
389 ptr = ptr->next;
393 return core_waiters;
396 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
398 struct core_thread *curr, *next;
399 struct task_struct *task;
401 spin_lock_irq(&current->sighand->siglock);
402 if (core_dumped && !__fatal_signal_pending(current))
403 current->signal->group_exit_code |= 0x80;
404 current->signal->group_exit_task = NULL;
405 current->signal->flags = SIGNAL_GROUP_EXIT;
406 spin_unlock_irq(&current->sighand->siglock);
408 next = mm->core_state->dumper.next;
409 while ((curr = next) != NULL) {
410 next = curr->next;
411 task = curr->task;
413 * see exit_mm(), curr->task must not see
414 * ->task == NULL before we read ->next.
416 smp_mb();
417 curr->task = NULL;
418 wake_up_process(task);
421 mm->core_state = NULL;
424 static bool dump_interrupted(void)
427 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
428 * can do try_to_freeze() and check __fatal_signal_pending(),
429 * but then we need to teach dump_write() to restart and clear
430 * TIF_SIGPENDING.
432 return signal_pending(current);
435 static void wait_for_dump_helpers(struct file *file)
437 struct pipe_inode_info *pipe = file->private_data;
439 pipe_lock(pipe);
440 pipe->readers++;
441 pipe->writers--;
442 wake_up_interruptible_sync(&pipe->wait);
443 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
444 pipe_unlock(pipe);
447 * We actually want wait_event_freezable() but then we need
448 * to clear TIF_SIGPENDING and improve dump_interrupted().
450 wait_event_interruptible(pipe->wait, pipe->readers == 1);
452 pipe_lock(pipe);
453 pipe->readers--;
454 pipe->writers++;
455 pipe_unlock(pipe);
459 * umh_pipe_setup
460 * helper function to customize the process used
461 * to collect the core in userspace. Specifically
462 * it sets up a pipe and installs it as fd 0 (stdin)
463 * for the process. Returns 0 on success, or
464 * PTR_ERR on failure.
465 * Note that it also sets the core limit to 1. This
466 * is a special value that we use to trap recursive
467 * core dumps
469 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
471 struct file *files[2];
472 struct coredump_params *cp = (struct coredump_params *)info->data;
473 int err = create_pipe_files(files, 0);
474 if (err)
475 return err;
477 cp->file = files[1];
479 err = replace_fd(0, files[0], 0);
480 fput(files[0]);
481 /* and disallow core files too */
482 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
484 return err;
487 void do_coredump(const siginfo_t *siginfo)
489 struct core_state core_state;
490 struct core_name cn;
491 struct mm_struct *mm = current->mm;
492 struct linux_binfmt * binfmt;
493 const struct cred *old_cred;
494 struct cred *cred;
495 int retval = 0;
496 int flag = 0;
497 int ispipe;
498 struct files_struct *displaced;
499 bool need_nonrelative = false;
500 bool core_dumped = false;
501 static atomic_t core_dump_count = ATOMIC_INIT(0);
502 struct coredump_params cprm = {
503 .siginfo = siginfo,
504 .regs = signal_pt_regs(),
505 .limit = rlimit(RLIMIT_CORE),
507 * We must use the same mm->flags while dumping core to avoid
508 * inconsistency of bit flags, since this flag is not protected
509 * by any locks.
511 .mm_flags = mm->flags,
514 audit_core_dumps(siginfo->si_signo);
516 binfmt = mm->binfmt;
517 if (!binfmt || !binfmt->core_dump)
518 goto fail;
519 if (!__get_dumpable(cprm.mm_flags))
520 goto fail;
522 cred = prepare_creds();
523 if (!cred)
524 goto fail;
526 * We cannot trust fsuid as being the "true" uid of the process
527 * nor do we know its entire history. We only know it was tainted
528 * so we dump it as root in mode 2, and only into a controlled
529 * environment (pipe handler or fully qualified path).
531 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
532 /* Setuid core dump mode */
533 flag = O_EXCL; /* Stop rewrite attacks */
534 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
535 need_nonrelative = true;
538 retval = coredump_wait(siginfo->si_signo, &core_state);
539 if (retval < 0)
540 goto fail_creds;
542 old_cred = override_creds(cred);
544 ispipe = format_corename(&cn, &cprm);
546 if (ispipe) {
547 int dump_count;
548 char **helper_argv;
549 struct subprocess_info *sub_info;
551 if (ispipe < 0) {
552 printk(KERN_WARNING "format_corename failed\n");
553 printk(KERN_WARNING "Aborting core\n");
554 goto fail_unlock;
557 if (cprm.limit == 1) {
558 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
560 * Normally core limits are irrelevant to pipes, since
561 * we're not writing to the file system, but we use
562 * cprm.limit of 1 here as a speacial value, this is a
563 * consistent way to catch recursive crashes.
564 * We can still crash if the core_pattern binary sets
565 * RLIM_CORE = !1, but it runs as root, and can do
566 * lots of stupid things.
568 * Note that we use task_tgid_vnr here to grab the pid
569 * of the process group leader. That way we get the
570 * right pid if a thread in a multi-threaded
571 * core_pattern process dies.
573 printk(KERN_WARNING
574 "Process %d(%s) has RLIMIT_CORE set to 1\n",
575 task_tgid_vnr(current), current->comm);
576 printk(KERN_WARNING "Aborting core\n");
577 goto fail_unlock;
579 cprm.limit = RLIM_INFINITY;
581 dump_count = atomic_inc_return(&core_dump_count);
582 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
583 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
584 task_tgid_vnr(current), current->comm);
585 printk(KERN_WARNING "Skipping core dump\n");
586 goto fail_dropcount;
589 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
590 if (!helper_argv) {
591 printk(KERN_WARNING "%s failed to allocate memory\n",
592 __func__);
593 goto fail_dropcount;
596 retval = -ENOMEM;
597 sub_info = call_usermodehelper_setup(helper_argv[0],
598 helper_argv, NULL, GFP_KERNEL,
599 umh_pipe_setup, NULL, &cprm);
600 if (sub_info)
601 retval = call_usermodehelper_exec(sub_info,
602 UMH_WAIT_EXEC);
604 argv_free(helper_argv);
605 if (retval) {
606 printk(KERN_INFO "Core dump to |%s pipe failed\n",
607 cn.corename);
608 goto close_fail;
610 } else {
611 struct inode *inode;
613 if (cprm.limit < binfmt->min_coredump)
614 goto fail_unlock;
616 if (need_nonrelative && cn.corename[0] != '/') {
617 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
618 "to fully qualified path!\n",
619 task_tgid_vnr(current), current->comm);
620 printk(KERN_WARNING "Skipping core dump\n");
621 goto fail_unlock;
624 cprm.file = filp_open(cn.corename,
625 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
626 0600);
627 if (IS_ERR(cprm.file))
628 goto fail_unlock;
630 inode = file_inode(cprm.file);
631 if (inode->i_nlink > 1)
632 goto close_fail;
633 if (d_unhashed(cprm.file->f_path.dentry))
634 goto close_fail;
636 * AK: actually i see no reason to not allow this for named
637 * pipes etc, but keep the previous behaviour for now.
639 if (!S_ISREG(inode->i_mode))
640 goto close_fail;
642 * Dont allow local users get cute and trick others to coredump
643 * into their pre-created files.
645 if (!uid_eq(inode->i_uid, current_fsuid()))
646 goto close_fail;
647 if (!cprm.file->f_op->write)
648 goto close_fail;
649 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
650 goto close_fail;
653 /* get us an unshared descriptor table; almost always a no-op */
654 retval = unshare_files(&displaced);
655 if (retval)
656 goto close_fail;
657 if (displaced)
658 put_files_struct(displaced);
659 if (!dump_interrupted()) {
660 file_start_write(cprm.file);
661 core_dumped = binfmt->core_dump(&cprm);
662 file_end_write(cprm.file);
664 if (ispipe && core_pipe_limit)
665 wait_for_dump_helpers(cprm.file);
666 close_fail:
667 if (cprm.file)
668 filp_close(cprm.file, NULL);
669 fail_dropcount:
670 if (ispipe)
671 atomic_dec(&core_dump_count);
672 fail_unlock:
673 kfree(cn.corename);
674 coredump_finish(mm, core_dumped);
675 revert_creds(old_cred);
676 fail_creds:
677 put_cred(cred);
678 fail:
679 return;
683 * Core dumping helper functions. These are the only things you should
684 * do on a core-file: use only these functions to write out all the
685 * necessary info.
687 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
689 struct file *file = cprm->file;
690 loff_t pos = file->f_pos;
691 ssize_t n;
692 if (cprm->written + nr > cprm->limit)
693 return 0;
694 while (nr) {
695 if (dump_interrupted())
696 return 0;
697 n = __kernel_write(file, addr, nr, &pos);
698 if (n <= 0)
699 return 0;
700 file->f_pos = pos;
701 cprm->written += n;
702 nr -= n;
704 return 1;
706 EXPORT_SYMBOL(dump_emit);
708 int dump_skip(struct coredump_params *cprm, size_t nr)
710 static char zeroes[PAGE_SIZE];
711 struct file *file = cprm->file;
712 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
713 if (cprm->written + nr > cprm->limit)
714 return 0;
715 if (dump_interrupted() ||
716 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
717 return 0;
718 cprm->written += nr;
719 return 1;
720 } else {
721 while (nr > PAGE_SIZE) {
722 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
723 return 0;
724 nr -= PAGE_SIZE;
726 return dump_emit(cprm, zeroes, nr);
729 EXPORT_SYMBOL(dump_skip);
731 int dump_align(struct coredump_params *cprm, int align)
733 unsigned mod = cprm->written & (align - 1);
734 if (align & (align - 1))
735 return 0;
736 return mod ? dump_skip(cprm, align - mod) : 1;
738 EXPORT_SYMBOL(dump_align);