ARM: kvm: rename cpu_reset to avoid name clash
[linux-2.6.git] / fs / coredump.c
blob9bdeca12ae0e388ecda010964c5209313cb363a9
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 /* global pid */
194 case 'P':
195 err = cn_printf(cn, "%d",
196 task_tgid_nr(current));
197 break;
198 /* uid */
199 case 'u':
200 err = cn_printf(cn, "%d", cred->uid);
201 break;
202 /* gid */
203 case 'g':
204 err = cn_printf(cn, "%d", cred->gid);
205 break;
206 case 'd':
207 err = cn_printf(cn, "%d",
208 __get_dumpable(cprm->mm_flags));
209 break;
210 /* signal that caused the coredump */
211 case 's':
212 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
213 break;
214 /* UNIX time of coredump */
215 case 't': {
216 struct timeval tv;
217 do_gettimeofday(&tv);
218 err = cn_printf(cn, "%lu", tv.tv_sec);
219 break;
221 /* hostname */
222 case 'h':
223 down_read(&uts_sem);
224 err = cn_esc_printf(cn, "%s",
225 utsname()->nodename);
226 up_read(&uts_sem);
227 break;
228 /* executable */
229 case 'e':
230 err = cn_esc_printf(cn, "%s", current->comm);
231 break;
232 case 'E':
233 err = cn_print_exe_file(cn);
234 break;
235 /* core limit size */
236 case 'c':
237 err = cn_printf(cn, "%lu",
238 rlimit(RLIMIT_CORE));
239 break;
240 default:
241 break;
243 ++pat_ptr;
246 if (err)
247 return err;
250 out:
251 /* Backward compatibility with core_uses_pid:
253 * If core_pattern does not include a %p (as is the default)
254 * and core_uses_pid is set, then .%pid will be appended to
255 * the filename. Do not do this for piped commands. */
256 if (!ispipe && !pid_in_pattern && core_uses_pid) {
257 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
258 if (err)
259 return err;
261 return ispipe;
264 static int zap_process(struct task_struct *start, int exit_code)
266 struct task_struct *t;
267 int nr = 0;
269 start->signal->group_exit_code = exit_code;
270 start->signal->group_stop_count = 0;
272 t = start;
273 do {
274 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
275 if (t != current && t->mm) {
276 sigaddset(&t->pending.signal, SIGKILL);
277 signal_wake_up(t, 1);
278 nr++;
280 } while_each_thread(start, t);
282 return nr;
285 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
286 struct core_state *core_state, int exit_code)
288 struct task_struct *g, *p;
289 unsigned long flags;
290 int nr = -EAGAIN;
292 spin_lock_irq(&tsk->sighand->siglock);
293 if (!signal_group_exit(tsk->signal)) {
294 mm->core_state = core_state;
295 nr = zap_process(tsk, exit_code);
296 tsk->signal->group_exit_task = tsk;
297 /* ignore all signals except SIGKILL, see prepare_signal() */
298 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
299 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
301 spin_unlock_irq(&tsk->sighand->siglock);
302 if (unlikely(nr < 0))
303 return nr;
305 tsk->flags = PF_DUMPCORE;
306 if (atomic_read(&mm->mm_users) == nr + 1)
307 goto done;
309 * We should find and kill all tasks which use this mm, and we should
310 * count them correctly into ->nr_threads. We don't take tasklist
311 * lock, but this is safe wrt:
313 * fork:
314 * None of sub-threads can fork after zap_process(leader). All
315 * processes which were created before this point should be
316 * visible to zap_threads() because copy_process() adds the new
317 * process to the tail of init_task.tasks list, and lock/unlock
318 * of ->siglock provides a memory barrier.
320 * do_exit:
321 * The caller holds mm->mmap_sem. This means that the task which
322 * uses this mm can't pass exit_mm(), so it can't exit or clear
323 * its ->mm.
325 * de_thread:
326 * It does list_replace_rcu(&leader->tasks, &current->tasks),
327 * we must see either old or new leader, this does not matter.
328 * However, it can change p->sighand, so lock_task_sighand(p)
329 * must be used. Since p->mm != NULL and we hold ->mmap_sem
330 * it can't fail.
332 * Note also that "g" can be the old leader with ->mm == NULL
333 * and already unhashed and thus removed from ->thread_group.
334 * This is OK, __unhash_process()->list_del_rcu() does not
335 * clear the ->next pointer, we will find the new leader via
336 * next_thread().
338 rcu_read_lock();
339 for_each_process(g) {
340 if (g == tsk->group_leader)
341 continue;
342 if (g->flags & PF_KTHREAD)
343 continue;
344 p = g;
345 do {
346 if (p->mm) {
347 if (unlikely(p->mm == mm)) {
348 lock_task_sighand(p, &flags);
349 nr += zap_process(p, exit_code);
350 p->signal->flags = SIGNAL_GROUP_EXIT;
351 unlock_task_sighand(p, &flags);
353 break;
355 } while_each_thread(g, p);
357 rcu_read_unlock();
358 done:
359 atomic_set(&core_state->nr_threads, nr);
360 return nr;
363 static int coredump_wait(int exit_code, struct core_state *core_state)
365 struct task_struct *tsk = current;
366 struct mm_struct *mm = tsk->mm;
367 int core_waiters = -EBUSY;
369 init_completion(&core_state->startup);
370 core_state->dumper.task = tsk;
371 core_state->dumper.next = NULL;
373 down_write(&mm->mmap_sem);
374 if (!mm->core_state)
375 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
376 up_write(&mm->mmap_sem);
378 if (core_waiters > 0) {
379 struct core_thread *ptr;
381 wait_for_completion(&core_state->startup);
383 * Wait for all the threads to become inactive, so that
384 * all the thread context (extended register state, like
385 * fpu etc) gets copied to the memory.
387 ptr = core_state->dumper.next;
388 while (ptr != NULL) {
389 wait_task_inactive(ptr->task, 0);
390 ptr = ptr->next;
394 return core_waiters;
397 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
399 struct core_thread *curr, *next;
400 struct task_struct *task;
402 spin_lock_irq(&current->sighand->siglock);
403 if (core_dumped && !__fatal_signal_pending(current))
404 current->signal->group_exit_code |= 0x80;
405 current->signal->group_exit_task = NULL;
406 current->signal->flags = SIGNAL_GROUP_EXIT;
407 spin_unlock_irq(&current->sighand->siglock);
409 next = mm->core_state->dumper.next;
410 while ((curr = next) != NULL) {
411 next = curr->next;
412 task = curr->task;
414 * see exit_mm(), curr->task must not see
415 * ->task == NULL before we read ->next.
417 smp_mb();
418 curr->task = NULL;
419 wake_up_process(task);
422 mm->core_state = NULL;
425 static bool dump_interrupted(void)
428 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
429 * can do try_to_freeze() and check __fatal_signal_pending(),
430 * but then we need to teach dump_write() to restart and clear
431 * TIF_SIGPENDING.
433 return signal_pending(current);
436 static void wait_for_dump_helpers(struct file *file)
438 struct pipe_inode_info *pipe = file->private_data;
440 pipe_lock(pipe);
441 pipe->readers++;
442 pipe->writers--;
443 wake_up_interruptible_sync(&pipe->wait);
444 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
445 pipe_unlock(pipe);
448 * We actually want wait_event_freezable() but then we need
449 * to clear TIF_SIGPENDING and improve dump_interrupted().
451 wait_event_interruptible(pipe->wait, pipe->readers == 1);
453 pipe_lock(pipe);
454 pipe->readers--;
455 pipe->writers++;
456 pipe_unlock(pipe);
460 * umh_pipe_setup
461 * helper function to customize the process used
462 * to collect the core in userspace. Specifically
463 * it sets up a pipe and installs it as fd 0 (stdin)
464 * for the process. Returns 0 on success, or
465 * PTR_ERR on failure.
466 * Note that it also sets the core limit to 1. This
467 * is a special value that we use to trap recursive
468 * core dumps
470 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
472 struct file *files[2];
473 struct coredump_params *cp = (struct coredump_params *)info->data;
474 int err = create_pipe_files(files, 0);
475 if (err)
476 return err;
478 cp->file = files[1];
480 err = replace_fd(0, files[0], 0);
481 fput(files[0]);
482 /* and disallow core files too */
483 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
485 return err;
488 void do_coredump(siginfo_t *siginfo)
490 struct core_state core_state;
491 struct core_name cn;
492 struct mm_struct *mm = current->mm;
493 struct linux_binfmt * binfmt;
494 const struct cred *old_cred;
495 struct cred *cred;
496 int retval = 0;
497 int flag = 0;
498 int ispipe;
499 struct files_struct *displaced;
500 bool need_nonrelative = false;
501 bool core_dumped = false;
502 static atomic_t core_dump_count = ATOMIC_INIT(0);
503 struct coredump_params cprm = {
504 .siginfo = siginfo,
505 .regs = signal_pt_regs(),
506 .limit = rlimit(RLIMIT_CORE),
508 * We must use the same mm->flags while dumping core to avoid
509 * inconsistency of bit flags, since this flag is not protected
510 * by any locks.
512 .mm_flags = mm->flags,
515 audit_core_dumps(siginfo->si_signo);
517 binfmt = mm->binfmt;
518 if (!binfmt || !binfmt->core_dump)
519 goto fail;
520 if (!__get_dumpable(cprm.mm_flags))
521 goto fail;
523 cred = prepare_creds();
524 if (!cred)
525 goto fail;
527 * We cannot trust fsuid as being the "true" uid of the process
528 * nor do we know its entire history. We only know it was tainted
529 * so we dump it as root in mode 2, and only into a controlled
530 * environment (pipe handler or fully qualified path).
532 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
533 /* Setuid core dump mode */
534 flag = O_EXCL; /* Stop rewrite attacks */
535 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
536 need_nonrelative = true;
539 retval = coredump_wait(siginfo->si_signo, &core_state);
540 if (retval < 0)
541 goto fail_creds;
543 old_cred = override_creds(cred);
545 ispipe = format_corename(&cn, &cprm);
547 if (ispipe) {
548 int dump_count;
549 char **helper_argv;
550 struct subprocess_info *sub_info;
552 if (ispipe < 0) {
553 printk(KERN_WARNING "format_corename failed\n");
554 printk(KERN_WARNING "Aborting core\n");
555 goto fail_unlock;
558 if (cprm.limit == 1) {
559 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
561 * Normally core limits are irrelevant to pipes, since
562 * we're not writing to the file system, but we use
563 * cprm.limit of 1 here as a speacial value, this is a
564 * consistent way to catch recursive crashes.
565 * We can still crash if the core_pattern binary sets
566 * RLIM_CORE = !1, but it runs as root, and can do
567 * lots of stupid things.
569 * Note that we use task_tgid_vnr here to grab the pid
570 * of the process group leader. That way we get the
571 * right pid if a thread in a multi-threaded
572 * core_pattern process dies.
574 printk(KERN_WARNING
575 "Process %d(%s) has RLIMIT_CORE set to 1\n",
576 task_tgid_vnr(current), current->comm);
577 printk(KERN_WARNING "Aborting core\n");
578 goto fail_unlock;
580 cprm.limit = RLIM_INFINITY;
582 dump_count = atomic_inc_return(&core_dump_count);
583 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
584 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
585 task_tgid_vnr(current), current->comm);
586 printk(KERN_WARNING "Skipping core dump\n");
587 goto fail_dropcount;
590 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
591 if (!helper_argv) {
592 printk(KERN_WARNING "%s failed to allocate memory\n",
593 __func__);
594 goto fail_dropcount;
597 retval = -ENOMEM;
598 sub_info = call_usermodehelper_setup(helper_argv[0],
599 helper_argv, NULL, GFP_KERNEL,
600 umh_pipe_setup, NULL, &cprm);
601 if (sub_info)
602 retval = call_usermodehelper_exec(sub_info,
603 UMH_WAIT_EXEC);
605 argv_free(helper_argv);
606 if (retval) {
607 printk(KERN_INFO "Core dump to |%s pipe failed\n",
608 cn.corename);
609 goto close_fail;
611 } else {
612 struct inode *inode;
614 if (cprm.limit < binfmt->min_coredump)
615 goto fail_unlock;
617 if (need_nonrelative && cn.corename[0] != '/') {
618 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
619 "to fully qualified path!\n",
620 task_tgid_vnr(current), current->comm);
621 printk(KERN_WARNING "Skipping core dump\n");
622 goto fail_unlock;
625 cprm.file = filp_open(cn.corename,
626 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
627 0600);
628 if (IS_ERR(cprm.file))
629 goto fail_unlock;
631 inode = file_inode(cprm.file);
632 if (inode->i_nlink > 1)
633 goto close_fail;
634 if (d_unhashed(cprm.file->f_path.dentry))
635 goto close_fail;
637 * AK: actually i see no reason to not allow this for named
638 * pipes etc, but keep the previous behaviour for now.
640 if (!S_ISREG(inode->i_mode))
641 goto close_fail;
643 * Dont allow local users get cute and trick others to coredump
644 * into their pre-created files.
646 if (!uid_eq(inode->i_uid, current_fsuid()))
647 goto close_fail;
648 if (!cprm.file->f_op || !cprm.file->f_op->write)
649 goto close_fail;
650 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
651 goto close_fail;
654 /* get us an unshared descriptor table; almost always a no-op */
655 retval = unshare_files(&displaced);
656 if (retval)
657 goto close_fail;
658 if (displaced)
659 put_files_struct(displaced);
660 if (!dump_interrupted()) {
661 file_start_write(cprm.file);
662 core_dumped = binfmt->core_dump(&cprm);
663 file_end_write(cprm.file);
665 if (ispipe && core_pipe_limit)
666 wait_for_dump_helpers(cprm.file);
667 close_fail:
668 if (cprm.file)
669 filp_close(cprm.file, NULL);
670 fail_dropcount:
671 if (ispipe)
672 atomic_dec(&core_dump_count);
673 fail_unlock:
674 kfree(cn.corename);
675 coredump_finish(mm, core_dumped);
676 revert_creds(old_cred);
677 fail_creds:
678 put_cred(cred);
679 fail:
680 return;
684 * Core dumping helper functions. These are the only things you should
685 * do on a core-file: use only these functions to write out all the
686 * necessary info.
688 int dump_write(struct file *file, const void *addr, int nr)
690 return !dump_interrupted() &&
691 access_ok(VERIFY_READ, addr, nr) &&
692 file->f_op->write(file, addr, nr, &file->f_pos) == nr;
694 EXPORT_SYMBOL(dump_write);
696 int dump_seek(struct file *file, loff_t off)
698 int ret = 1;
700 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
701 if (dump_interrupted() ||
702 file->f_op->llseek(file, off, SEEK_CUR) < 0)
703 return 0;
704 } else {
705 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
707 if (!buf)
708 return 0;
709 while (off > 0) {
710 unsigned long n = off;
712 if (n > PAGE_SIZE)
713 n = PAGE_SIZE;
714 if (!dump_write(file, buf, n)) {
715 ret = 0;
716 break;
718 off -= n;
720 free_page((unsigned long)buf);
722 return ret;
724 EXPORT_SYMBOL(dump_seek);