Merge tag 'modules-next-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6.git] / fs / coredump.c
blob72f816d6cad99d4d1f81433e928d42e540295188
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"
43 #include "coredump.h"
45 #include <trace/events/sched.h>
47 int core_uses_pid;
48 unsigned int core_pipe_limit;
49 char core_pattern[CORENAME_MAX_SIZE] = "core";
50 static int core_name_size = CORENAME_MAX_SIZE;
52 struct core_name {
53 char *corename;
54 int used, size;
57 /* The maximal length of core_pattern is also specified in sysctl.c */
59 static int expand_corename(struct core_name *cn, int size)
61 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
63 if (!corename)
64 return -ENOMEM;
66 if (size > core_name_size) /* racy but harmless */
67 core_name_size = size;
69 cn->size = ksize(corename);
70 cn->corename = corename;
71 return 0;
74 static int cn_vprintf(struct core_name *cn, const char *fmt, va_list arg)
76 int free, need;
78 again:
79 free = cn->size - cn->used;
80 need = vsnprintf(cn->corename + cn->used, free, fmt, arg);
81 if (need < free) {
82 cn->used += need;
83 return 0;
86 if (!expand_corename(cn, cn->size + need - free + 1))
87 goto again;
89 return -ENOMEM;
92 static int cn_printf(struct core_name *cn, const char *fmt, ...)
94 va_list arg;
95 int ret;
97 va_start(arg, fmt);
98 ret = cn_vprintf(cn, fmt, arg);
99 va_end(arg);
101 return ret;
104 static int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
106 int cur = cn->used;
107 va_list arg;
108 int ret;
110 va_start(arg, fmt);
111 ret = cn_vprintf(cn, fmt, arg);
112 va_end(arg);
114 for (; cur < cn->used; ++cur) {
115 if (cn->corename[cur] == '/')
116 cn->corename[cur] = '!';
118 return ret;
121 static int cn_print_exe_file(struct core_name *cn)
123 struct file *exe_file;
124 char *pathbuf, *path;
125 int ret;
127 exe_file = get_mm_exe_file(current->mm);
128 if (!exe_file)
129 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
131 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
132 if (!pathbuf) {
133 ret = -ENOMEM;
134 goto put_exe_file;
137 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
138 if (IS_ERR(path)) {
139 ret = PTR_ERR(path);
140 goto free_buf;
143 ret = cn_esc_printf(cn, "%s", path);
145 free_buf:
146 kfree(pathbuf);
147 put_exe_file:
148 fput(exe_file);
149 return ret;
152 /* format_corename will inspect the pattern parameter, and output a
153 * name into corename, which must have space for at least
154 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
156 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
158 const struct cred *cred = current_cred();
159 const char *pat_ptr = core_pattern;
160 int ispipe = (*pat_ptr == '|');
161 int pid_in_pattern = 0;
162 int err = 0;
164 cn->used = 0;
165 cn->corename = NULL;
166 if (expand_corename(cn, core_name_size))
167 return -ENOMEM;
168 cn->corename[0] = '\0';
170 if (ispipe)
171 ++pat_ptr;
173 /* Repeat as long as we have more pattern to process and more output
174 space */
175 while (*pat_ptr) {
176 if (*pat_ptr != '%') {
177 err = cn_printf(cn, "%c", *pat_ptr++);
178 } else {
179 switch (*++pat_ptr) {
180 /* single % at the end, drop that */
181 case 0:
182 goto out;
183 /* Double percent, output one percent */
184 case '%':
185 err = cn_printf(cn, "%c", '%');
186 break;
187 /* pid */
188 case 'p':
189 pid_in_pattern = 1;
190 err = cn_printf(cn, "%d",
191 task_tgid_vnr(current));
192 break;
193 /* uid */
194 case 'u':
195 err = cn_printf(cn, "%d", cred->uid);
196 break;
197 /* gid */
198 case 'g':
199 err = cn_printf(cn, "%d", cred->gid);
200 break;
201 case 'd':
202 err = cn_printf(cn, "%d",
203 __get_dumpable(cprm->mm_flags));
204 break;
205 /* signal that caused the coredump */
206 case 's':
207 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
208 break;
209 /* UNIX time of coredump */
210 case 't': {
211 struct timeval tv;
212 do_gettimeofday(&tv);
213 err = cn_printf(cn, "%lu", tv.tv_sec);
214 break;
216 /* hostname */
217 case 'h':
218 down_read(&uts_sem);
219 err = cn_esc_printf(cn, "%s",
220 utsname()->nodename);
221 up_read(&uts_sem);
222 break;
223 /* executable */
224 case 'e':
225 err = cn_esc_printf(cn, "%s", current->comm);
226 break;
227 case 'E':
228 err = cn_print_exe_file(cn);
229 break;
230 /* core limit size */
231 case 'c':
232 err = cn_printf(cn, "%lu",
233 rlimit(RLIMIT_CORE));
234 break;
235 default:
236 break;
238 ++pat_ptr;
241 if (err)
242 return err;
245 out:
246 /* Backward compatibility with core_uses_pid:
248 * If core_pattern does not include a %p (as is the default)
249 * and core_uses_pid is set, then .%pid will be appended to
250 * the filename. Do not do this for piped commands. */
251 if (!ispipe && !pid_in_pattern && core_uses_pid) {
252 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
253 if (err)
254 return err;
256 return ispipe;
259 static int zap_process(struct task_struct *start, int exit_code)
261 struct task_struct *t;
262 int nr = 0;
264 start->signal->group_exit_code = exit_code;
265 start->signal->group_stop_count = 0;
267 t = start;
268 do {
269 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
270 if (t != current && t->mm) {
271 sigaddset(&t->pending.signal, SIGKILL);
272 signal_wake_up(t, 1);
273 nr++;
275 } while_each_thread(start, t);
277 return nr;
280 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
281 struct core_state *core_state, int exit_code)
283 struct task_struct *g, *p;
284 unsigned long flags;
285 int nr = -EAGAIN;
287 spin_lock_irq(&tsk->sighand->siglock);
288 if (!signal_group_exit(tsk->signal)) {
289 mm->core_state = core_state;
290 nr = zap_process(tsk, exit_code);
291 tsk->signal->group_exit_task = tsk;
292 /* ignore all signals except SIGKILL, see prepare_signal() */
293 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
294 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
296 spin_unlock_irq(&tsk->sighand->siglock);
297 if (unlikely(nr < 0))
298 return nr;
300 tsk->flags = PF_DUMPCORE;
301 if (atomic_read(&mm->mm_users) == nr + 1)
302 goto done;
304 * We should find and kill all tasks which use this mm, and we should
305 * count them correctly into ->nr_threads. We don't take tasklist
306 * lock, but this is safe wrt:
308 * fork:
309 * None of sub-threads can fork after zap_process(leader). All
310 * processes which were created before this point should be
311 * visible to zap_threads() because copy_process() adds the new
312 * process to the tail of init_task.tasks list, and lock/unlock
313 * of ->siglock provides a memory barrier.
315 * do_exit:
316 * The caller holds mm->mmap_sem. This means that the task which
317 * uses this mm can't pass exit_mm(), so it can't exit or clear
318 * its ->mm.
320 * de_thread:
321 * It does list_replace_rcu(&leader->tasks, &current->tasks),
322 * we must see either old or new leader, this does not matter.
323 * However, it can change p->sighand, so lock_task_sighand(p)
324 * must be used. Since p->mm != NULL and we hold ->mmap_sem
325 * it can't fail.
327 * Note also that "g" can be the old leader with ->mm == NULL
328 * and already unhashed and thus removed from ->thread_group.
329 * This is OK, __unhash_process()->list_del_rcu() does not
330 * clear the ->next pointer, we will find the new leader via
331 * next_thread().
333 rcu_read_lock();
334 for_each_process(g) {
335 if (g == tsk->group_leader)
336 continue;
337 if (g->flags & PF_KTHREAD)
338 continue;
339 p = g;
340 do {
341 if (p->mm) {
342 if (unlikely(p->mm == mm)) {
343 lock_task_sighand(p, &flags);
344 nr += zap_process(p, exit_code);
345 p->signal->flags = SIGNAL_GROUP_EXIT;
346 unlock_task_sighand(p, &flags);
348 break;
350 } while_each_thread(g, p);
352 rcu_read_unlock();
353 done:
354 atomic_set(&core_state->nr_threads, nr);
355 return nr;
358 static int coredump_wait(int exit_code, struct core_state *core_state)
360 struct task_struct *tsk = current;
361 struct mm_struct *mm = tsk->mm;
362 int core_waiters = -EBUSY;
364 init_completion(&core_state->startup);
365 core_state->dumper.task = tsk;
366 core_state->dumper.next = NULL;
368 down_write(&mm->mmap_sem);
369 if (!mm->core_state)
370 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
371 up_write(&mm->mmap_sem);
373 if (core_waiters > 0) {
374 struct core_thread *ptr;
376 wait_for_completion(&core_state->startup);
378 * Wait for all the threads to become inactive, so that
379 * all the thread context (extended register state, like
380 * fpu etc) gets copied to the memory.
382 ptr = core_state->dumper.next;
383 while (ptr != NULL) {
384 wait_task_inactive(ptr->task, 0);
385 ptr = ptr->next;
389 return core_waiters;
392 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
394 struct core_thread *curr, *next;
395 struct task_struct *task;
397 spin_lock_irq(&current->sighand->siglock);
398 if (core_dumped && !__fatal_signal_pending(current))
399 current->signal->group_exit_code |= 0x80;
400 current->signal->group_exit_task = NULL;
401 current->signal->flags = SIGNAL_GROUP_EXIT;
402 spin_unlock_irq(&current->sighand->siglock);
404 next = mm->core_state->dumper.next;
405 while ((curr = next) != NULL) {
406 next = curr->next;
407 task = curr->task;
409 * see exit_mm(), curr->task must not see
410 * ->task == NULL before we read ->next.
412 smp_mb();
413 curr->task = NULL;
414 wake_up_process(task);
417 mm->core_state = NULL;
420 static bool dump_interrupted(void)
423 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
424 * can do try_to_freeze() and check __fatal_signal_pending(),
425 * but then we need to teach dump_write() to restart and clear
426 * TIF_SIGPENDING.
428 return signal_pending(current);
431 static void wait_for_dump_helpers(struct file *file)
433 struct pipe_inode_info *pipe = file->private_data;
435 pipe_lock(pipe);
436 pipe->readers++;
437 pipe->writers--;
438 wake_up_interruptible_sync(&pipe->wait);
439 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
440 pipe_unlock(pipe);
443 * We actually want wait_event_freezable() but then we need
444 * to clear TIF_SIGPENDING and improve dump_interrupted().
446 wait_event_interruptible(pipe->wait, pipe->readers == 1);
448 pipe_lock(pipe);
449 pipe->readers--;
450 pipe->writers++;
451 pipe_unlock(pipe);
455 * umh_pipe_setup
456 * helper function to customize the process used
457 * to collect the core in userspace. Specifically
458 * it sets up a pipe and installs it as fd 0 (stdin)
459 * for the process. Returns 0 on success, or
460 * PTR_ERR on failure.
461 * Note that it also sets the core limit to 1. This
462 * is a special value that we use to trap recursive
463 * core dumps
465 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
467 struct file *files[2];
468 struct coredump_params *cp = (struct coredump_params *)info->data;
469 int err = create_pipe_files(files, 0);
470 if (err)
471 return err;
473 cp->file = files[1];
475 err = replace_fd(0, files[0], 0);
476 fput(files[0]);
477 /* and disallow core files too */
478 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
480 return err;
483 void do_coredump(siginfo_t *siginfo)
485 struct core_state core_state;
486 struct core_name cn;
487 struct mm_struct *mm = current->mm;
488 struct linux_binfmt * binfmt;
489 const struct cred *old_cred;
490 struct cred *cred;
491 int retval = 0;
492 int flag = 0;
493 int ispipe;
494 struct files_struct *displaced;
495 bool need_nonrelative = false;
496 bool core_dumped = false;
497 static atomic_t core_dump_count = ATOMIC_INIT(0);
498 struct coredump_params cprm = {
499 .siginfo = siginfo,
500 .regs = signal_pt_regs(),
501 .limit = rlimit(RLIMIT_CORE),
503 * We must use the same mm->flags while dumping core to avoid
504 * inconsistency of bit flags, since this flag is not protected
505 * by any locks.
507 .mm_flags = mm->flags,
510 audit_core_dumps(siginfo->si_signo);
512 binfmt = mm->binfmt;
513 if (!binfmt || !binfmt->core_dump)
514 goto fail;
515 if (!__get_dumpable(cprm.mm_flags))
516 goto fail;
518 cred = prepare_creds();
519 if (!cred)
520 goto fail;
522 * We cannot trust fsuid as being the "true" uid of the process
523 * nor do we know its entire history. We only know it was tainted
524 * so we dump it as root in mode 2, and only into a controlled
525 * environment (pipe handler or fully qualified path).
527 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
528 /* Setuid core dump mode */
529 flag = O_EXCL; /* Stop rewrite attacks */
530 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
531 need_nonrelative = true;
534 retval = coredump_wait(siginfo->si_signo, &core_state);
535 if (retval < 0)
536 goto fail_creds;
538 old_cred = override_creds(cred);
540 ispipe = format_corename(&cn, &cprm);
542 if (ispipe) {
543 int dump_count;
544 char **helper_argv;
545 struct subprocess_info *sub_info;
547 if (ispipe < 0) {
548 printk(KERN_WARNING "format_corename failed\n");
549 printk(KERN_WARNING "Aborting core\n");
550 goto fail_unlock;
553 if (cprm.limit == 1) {
554 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
556 * Normally core limits are irrelevant to pipes, since
557 * we're not writing to the file system, but we use
558 * cprm.limit of 1 here as a speacial value, this is a
559 * consistent way to catch recursive crashes.
560 * We can still crash if the core_pattern binary sets
561 * RLIM_CORE = !1, but it runs as root, and can do
562 * lots of stupid things.
564 * Note that we use task_tgid_vnr here to grab the pid
565 * of the process group leader. That way we get the
566 * right pid if a thread in a multi-threaded
567 * core_pattern process dies.
569 printk(KERN_WARNING
570 "Process %d(%s) has RLIMIT_CORE set to 1\n",
571 task_tgid_vnr(current), current->comm);
572 printk(KERN_WARNING "Aborting core\n");
573 goto fail_unlock;
575 cprm.limit = RLIM_INFINITY;
577 dump_count = atomic_inc_return(&core_dump_count);
578 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
579 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
580 task_tgid_vnr(current), current->comm);
581 printk(KERN_WARNING "Skipping core dump\n");
582 goto fail_dropcount;
585 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
586 if (!helper_argv) {
587 printk(KERN_WARNING "%s failed to allocate memory\n",
588 __func__);
589 goto fail_dropcount;
592 retval = -ENOMEM;
593 sub_info = call_usermodehelper_setup(helper_argv[0],
594 helper_argv, NULL, GFP_KERNEL,
595 umh_pipe_setup, NULL, &cprm);
596 if (sub_info)
597 retval = call_usermodehelper_exec(sub_info,
598 UMH_WAIT_EXEC);
600 argv_free(helper_argv);
601 if (retval) {
602 printk(KERN_INFO "Core dump to |%s pipe failed\n",
603 cn.corename);
604 goto close_fail;
606 } else {
607 struct inode *inode;
609 if (cprm.limit < binfmt->min_coredump)
610 goto fail_unlock;
612 if (need_nonrelative && cn.corename[0] != '/') {
613 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
614 "to fully qualified path!\n",
615 task_tgid_vnr(current), current->comm);
616 printk(KERN_WARNING "Skipping core dump\n");
617 goto fail_unlock;
620 cprm.file = filp_open(cn.corename,
621 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
622 0600);
623 if (IS_ERR(cprm.file))
624 goto fail_unlock;
626 inode = file_inode(cprm.file);
627 if (inode->i_nlink > 1)
628 goto close_fail;
629 if (d_unhashed(cprm.file->f_path.dentry))
630 goto close_fail;
632 * AK: actually i see no reason to not allow this for named
633 * pipes etc, but keep the previous behaviour for now.
635 if (!S_ISREG(inode->i_mode))
636 goto close_fail;
638 * Dont allow local users get cute and trick others to coredump
639 * into their pre-created files.
641 if (!uid_eq(inode->i_uid, current_fsuid()))
642 goto close_fail;
643 if (!cprm.file->f_op || !cprm.file->f_op->write)
644 goto close_fail;
645 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
646 goto close_fail;
649 /* get us an unshared descriptor table; almost always a no-op */
650 retval = unshare_files(&displaced);
651 if (retval)
652 goto close_fail;
653 if (displaced)
654 put_files_struct(displaced);
655 if (!dump_interrupted()) {
656 file_start_write(cprm.file);
657 core_dumped = binfmt->core_dump(&cprm);
658 file_end_write(cprm.file);
660 if (ispipe && core_pipe_limit)
661 wait_for_dump_helpers(cprm.file);
662 close_fail:
663 if (cprm.file)
664 filp_close(cprm.file, NULL);
665 fail_dropcount:
666 if (ispipe)
667 atomic_dec(&core_dump_count);
668 fail_unlock:
669 kfree(cn.corename);
670 coredump_finish(mm, core_dumped);
671 revert_creds(old_cred);
672 fail_creds:
673 put_cred(cred);
674 fail:
675 return;
679 * Core dumping helper functions. These are the only things you should
680 * do on a core-file: use only these functions to write out all the
681 * necessary info.
683 int dump_write(struct file *file, const void *addr, int nr)
685 return !dump_interrupted() &&
686 access_ok(VERIFY_READ, addr, nr) &&
687 file->f_op->write(file, addr, nr, &file->f_pos) == nr;
689 EXPORT_SYMBOL(dump_write);
691 int dump_seek(struct file *file, loff_t off)
693 int ret = 1;
695 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
696 if (dump_interrupted() ||
697 file->f_op->llseek(file, off, SEEK_CUR) < 0)
698 return 0;
699 } else {
700 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
702 if (!buf)
703 return 0;
704 while (off > 0) {
705 unsigned long n = off;
707 if (n > PAGE_SIZE)
708 n = PAGE_SIZE;
709 if (!dump_write(file, buf, n)) {
710 ret = 0;
711 break;
713 off -= n;
715 free_page((unsigned long)buf);
717 return ret;
719 EXPORT_SYMBOL(dump_seek);