coredump: factor out the setting of PF_DUMPCORE
[linux-2.6.git] / fs / coredump.c
blob7300e312fb3a2b3882dc1deb8339e3feb453c358
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 char core_pattern[CORENAME_MAX_SIZE] = "core";
49 unsigned int core_pipe_limit;
51 struct core_name {
52 char *corename;
53 int used, size;
55 static atomic_t call_count = ATOMIC_INIT(1);
57 /* The maximal length of core_pattern is also specified in sysctl.c */
59 static int expand_corename(struct core_name *cn)
61 char *old_corename = cn->corename;
63 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
64 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
66 if (!cn->corename) {
67 kfree(old_corename);
68 return -ENOMEM;
71 return 0;
74 static int cn_printf(struct core_name *cn, const char *fmt, ...)
76 char *cur;
77 int need;
78 int ret;
79 va_list arg;
81 va_start(arg, fmt);
82 need = vsnprintf(NULL, 0, fmt, arg);
83 va_end(arg);
85 if (likely(need < cn->size - cn->used - 1))
86 goto out_printf;
88 ret = expand_corename(cn);
89 if (ret)
90 goto expand_fail;
92 out_printf:
93 cur = cn->corename + cn->used;
94 va_start(arg, fmt);
95 vsnprintf(cur, need + 1, fmt, arg);
96 va_end(arg);
97 cn->used += need;
98 return 0;
100 expand_fail:
101 return ret;
104 static void cn_escape(char *str)
106 for (; *str; str++)
107 if (*str == '/')
108 *str = '!';
111 static int cn_print_exe_file(struct core_name *cn)
113 struct file *exe_file;
114 char *pathbuf, *path;
115 int ret;
117 exe_file = get_mm_exe_file(current->mm);
118 if (!exe_file) {
119 char *commstart = cn->corename + cn->used;
120 ret = cn_printf(cn, "%s (path unknown)", current->comm);
121 cn_escape(commstart);
122 return ret;
125 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
126 if (!pathbuf) {
127 ret = -ENOMEM;
128 goto put_exe_file;
131 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
132 if (IS_ERR(path)) {
133 ret = PTR_ERR(path);
134 goto free_buf;
137 cn_escape(path);
139 ret = cn_printf(cn, "%s", path);
141 free_buf:
142 kfree(pathbuf);
143 put_exe_file:
144 fput(exe_file);
145 return ret;
148 /* format_corename will inspect the pattern parameter, and output a
149 * name into corename, which must have space for at least
150 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
152 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
154 const struct cred *cred = current_cred();
155 const char *pat_ptr = core_pattern;
156 int ispipe = (*pat_ptr == '|');
157 int pid_in_pattern = 0;
158 int err = 0;
160 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
161 cn->corename = kmalloc(cn->size, GFP_KERNEL);
162 cn->used = 0;
164 if (!cn->corename)
165 return -ENOMEM;
167 /* Repeat as long as we have more pattern to process and more output
168 space */
169 while (*pat_ptr) {
170 if (*pat_ptr != '%') {
171 if (*pat_ptr == 0)
172 goto out;
173 err = cn_printf(cn, "%c", *pat_ptr++);
174 } else {
175 switch (*++pat_ptr) {
176 /* single % at the end, drop that */
177 case 0:
178 goto out;
179 /* Double percent, output one percent */
180 case '%':
181 err = cn_printf(cn, "%c", '%');
182 break;
183 /* pid */
184 case 'p':
185 pid_in_pattern = 1;
186 err = cn_printf(cn, "%d",
187 task_tgid_vnr(current));
188 break;
189 /* uid */
190 case 'u':
191 err = cn_printf(cn, "%d", cred->uid);
192 break;
193 /* gid */
194 case 'g':
195 err = cn_printf(cn, "%d", cred->gid);
196 break;
197 case 'd':
198 err = cn_printf(cn, "%d",
199 __get_dumpable(cprm->mm_flags));
200 break;
201 /* signal that caused the coredump */
202 case 's':
203 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
204 break;
205 /* UNIX time of coredump */
206 case 't': {
207 struct timeval tv;
208 do_gettimeofday(&tv);
209 err = cn_printf(cn, "%lu", tv.tv_sec);
210 break;
212 /* hostname */
213 case 'h': {
214 char *namestart = cn->corename + cn->used;
215 down_read(&uts_sem);
216 err = cn_printf(cn, "%s",
217 utsname()->nodename);
218 up_read(&uts_sem);
219 cn_escape(namestart);
220 break;
222 /* executable */
223 case 'e': {
224 char *commstart = cn->corename + cn->used;
225 err = cn_printf(cn, "%s", current->comm);
226 cn_escape(commstart);
227 break;
229 case 'E':
230 err = cn_print_exe_file(cn);
231 break;
232 /* core limit size */
233 case 'c':
234 err = cn_printf(cn, "%lu",
235 rlimit(RLIMIT_CORE));
236 break;
237 default:
238 break;
240 ++pat_ptr;
243 if (err)
244 return err;
247 /* Backward compatibility with core_uses_pid:
249 * If core_pattern does not include a %p (as is the default)
250 * and core_uses_pid is set, then .%pid will be appended to
251 * the filename. Do not do this for piped commands. */
252 if (!ispipe && !pid_in_pattern && core_uses_pid) {
253 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
254 if (err)
255 return err;
257 out:
258 return ispipe;
261 static int zap_process(struct task_struct *start, int exit_code)
263 struct task_struct *t;
264 int nr = 0;
266 start->signal->group_exit_code = exit_code;
267 start->signal->group_stop_count = 0;
269 t = start;
270 do {
271 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
272 if (t != current && t->mm) {
273 sigaddset(&t->pending.signal, SIGKILL);
274 signal_wake_up(t, 1);
275 nr++;
277 } while_each_thread(start, t);
279 return nr;
282 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
283 struct core_state *core_state, int exit_code)
285 struct task_struct *g, *p;
286 unsigned long flags;
287 int nr = -EAGAIN;
289 spin_lock_irq(&tsk->sighand->siglock);
290 if (!signal_group_exit(tsk->signal)) {
291 mm->core_state = core_state;
292 nr = zap_process(tsk, exit_code);
293 tsk->signal->group_exit_task = tsk;
294 /* ignore all signals except SIGKILL, see prepare_signal() */
295 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
296 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
298 spin_unlock_irq(&tsk->sighand->siglock);
299 if (unlikely(nr < 0))
300 return nr;
302 tsk->flags = PF_DUMPCORE;
303 if (atomic_read(&mm->mm_users) == nr + 1)
304 goto done;
306 * We should find and kill all tasks which use this mm, and we should
307 * count them correctly into ->nr_threads. We don't take tasklist
308 * lock, but this is safe wrt:
310 * fork:
311 * None of sub-threads can fork after zap_process(leader). All
312 * processes which were created before this point should be
313 * visible to zap_threads() because copy_process() adds the new
314 * process to the tail of init_task.tasks list, and lock/unlock
315 * of ->siglock provides a memory barrier.
317 * do_exit:
318 * The caller holds mm->mmap_sem. This means that the task which
319 * uses this mm can't pass exit_mm(), so it can't exit or clear
320 * its ->mm.
322 * de_thread:
323 * It does list_replace_rcu(&leader->tasks, &current->tasks),
324 * we must see either old or new leader, this does not matter.
325 * However, it can change p->sighand, so lock_task_sighand(p)
326 * must be used. Since p->mm != NULL and we hold ->mmap_sem
327 * it can't fail.
329 * Note also that "g" can be the old leader with ->mm == NULL
330 * and already unhashed and thus removed from ->thread_group.
331 * This is OK, __unhash_process()->list_del_rcu() does not
332 * clear the ->next pointer, we will find the new leader via
333 * next_thread().
335 rcu_read_lock();
336 for_each_process(g) {
337 if (g == tsk->group_leader)
338 continue;
339 if (g->flags & PF_KTHREAD)
340 continue;
341 p = g;
342 do {
343 if (p->mm) {
344 if (unlikely(p->mm == mm)) {
345 lock_task_sighand(p, &flags);
346 nr += zap_process(p, exit_code);
347 p->signal->flags = SIGNAL_GROUP_EXIT;
348 unlock_task_sighand(p, &flags);
350 break;
352 } while_each_thread(g, p);
354 rcu_read_unlock();
355 done:
356 atomic_set(&core_state->nr_threads, nr);
357 return nr;
360 static int coredump_wait(int exit_code, struct core_state *core_state)
362 struct task_struct *tsk = current;
363 struct mm_struct *mm = tsk->mm;
364 int core_waiters = -EBUSY;
366 init_completion(&core_state->startup);
367 core_state->dumper.task = tsk;
368 core_state->dumper.next = NULL;
370 down_write(&mm->mmap_sem);
371 if (!mm->core_state)
372 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
373 up_write(&mm->mmap_sem);
375 if (core_waiters > 0) {
376 struct core_thread *ptr;
378 wait_for_completion(&core_state->startup);
380 * Wait for all the threads to become inactive, so that
381 * all the thread context (extended register state, like
382 * fpu etc) gets copied to the memory.
384 ptr = core_state->dumper.next;
385 while (ptr != NULL) {
386 wait_task_inactive(ptr->task, 0);
387 ptr = ptr->next;
391 return core_waiters;
394 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
396 struct core_thread *curr, *next;
397 struct task_struct *task;
399 spin_lock_irq(&current->sighand->siglock);
400 if (core_dumped && !__fatal_signal_pending(current))
401 current->signal->group_exit_code |= 0x80;
402 current->signal->group_exit_task = NULL;
403 current->signal->flags = SIGNAL_GROUP_EXIT;
404 spin_unlock_irq(&current->sighand->siglock);
406 next = mm->core_state->dumper.next;
407 while ((curr = next) != NULL) {
408 next = curr->next;
409 task = curr->task;
411 * see exit_mm(), curr->task must not see
412 * ->task == NULL before we read ->next.
414 smp_mb();
415 curr->task = NULL;
416 wake_up_process(task);
419 mm->core_state = NULL;
422 static bool dump_interrupted(void)
425 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
426 * can do try_to_freeze() and check __fatal_signal_pending(),
427 * but then we need to teach dump_write() to restart and clear
428 * TIF_SIGPENDING.
430 return signal_pending(current);
433 static void wait_for_dump_helpers(struct file *file)
435 struct pipe_inode_info *pipe;
437 pipe = file_inode(file)->i_pipe;
439 pipe_lock(pipe);
440 pipe->readers++;
441 pipe->writers--;
443 while ((pipe->readers > 1) && (!signal_pending(current))) {
444 wake_up_interruptible_sync(&pipe->wait);
445 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
446 pipe_wait(pipe);
449 pipe->readers--;
450 pipe->writers++;
451 pipe_unlock(pipe);
456 * umh_pipe_setup
457 * helper function to customize the process used
458 * to collect the core in userspace. Specifically
459 * it sets up a pipe and installs it as fd 0 (stdin)
460 * for the process. Returns 0 on success, or
461 * PTR_ERR on failure.
462 * Note that it also sets the core limit to 1. This
463 * is a special value that we use to trap recursive
464 * core dumps
466 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
468 struct file *files[2];
469 struct coredump_params *cp = (struct coredump_params *)info->data;
470 int err = create_pipe_files(files, 0);
471 if (err)
472 return err;
474 cp->file = files[1];
476 err = replace_fd(0, files[0], 0);
477 fput(files[0]);
478 /* and disallow core files too */
479 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
481 return err;
484 void do_coredump(siginfo_t *siginfo)
486 struct core_state core_state;
487 struct core_name cn;
488 struct mm_struct *mm = current->mm;
489 struct linux_binfmt * binfmt;
490 const struct cred *old_cred;
491 struct cred *cred;
492 int retval = 0;
493 int flag = 0;
494 int ispipe;
495 struct files_struct *displaced;
496 bool need_nonrelative = false;
497 bool core_dumped = false;
498 static atomic_t core_dump_count = ATOMIC_INIT(0);
499 struct coredump_params cprm = {
500 .siginfo = siginfo,
501 .regs = signal_pt_regs(),
502 .limit = rlimit(RLIMIT_CORE),
504 * We must use the same mm->flags while dumping core to avoid
505 * inconsistency of bit flags, since this flag is not protected
506 * by any locks.
508 .mm_flags = mm->flags,
511 audit_core_dumps(siginfo->si_signo);
513 binfmt = mm->binfmt;
514 if (!binfmt || !binfmt->core_dump)
515 goto fail;
516 if (!__get_dumpable(cprm.mm_flags))
517 goto fail;
519 cred = prepare_creds();
520 if (!cred)
521 goto fail;
523 * We cannot trust fsuid as being the "true" uid of the process
524 * nor do we know its entire history. We only know it was tainted
525 * so we dump it as root in mode 2, and only into a controlled
526 * environment (pipe handler or fully qualified path).
528 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
529 /* Setuid core dump mode */
530 flag = O_EXCL; /* Stop rewrite attacks */
531 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
532 need_nonrelative = true;
535 retval = coredump_wait(siginfo->si_signo, &core_state);
536 if (retval < 0)
537 goto fail_creds;
539 old_cred = override_creds(cred);
541 ispipe = format_corename(&cn, &cprm);
543 if (ispipe) {
544 int dump_count;
545 char **helper_argv;
546 struct subprocess_info *sub_info;
548 if (ispipe < 0) {
549 printk(KERN_WARNING "format_corename failed\n");
550 printk(KERN_WARNING "Aborting core\n");
551 goto fail_corename;
554 if (cprm.limit == 1) {
555 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
557 * Normally core limits are irrelevant to pipes, since
558 * we're not writing to the file system, but we use
559 * cprm.limit of 1 here as a speacial value, this is a
560 * consistent way to catch recursive crashes.
561 * We can still crash if the core_pattern binary sets
562 * RLIM_CORE = !1, but it runs as root, and can do
563 * lots of stupid things.
565 * Note that we use task_tgid_vnr here to grab the pid
566 * of the process group leader. That way we get the
567 * right pid if a thread in a multi-threaded
568 * core_pattern process dies.
570 printk(KERN_WARNING
571 "Process %d(%s) has RLIMIT_CORE set to 1\n",
572 task_tgid_vnr(current), current->comm);
573 printk(KERN_WARNING "Aborting core\n");
574 goto fail_unlock;
576 cprm.limit = RLIM_INFINITY;
578 dump_count = atomic_inc_return(&core_dump_count);
579 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
580 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
581 task_tgid_vnr(current), current->comm);
582 printk(KERN_WARNING "Skipping core dump\n");
583 goto fail_dropcount;
586 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
587 if (!helper_argv) {
588 printk(KERN_WARNING "%s failed to allocate memory\n",
589 __func__);
590 goto fail_dropcount;
593 retval = -ENOMEM;
594 sub_info = call_usermodehelper_setup(helper_argv[0],
595 helper_argv, NULL, GFP_KERNEL,
596 umh_pipe_setup, NULL, &cprm);
597 if (sub_info)
598 retval = call_usermodehelper_exec(sub_info,
599 UMH_WAIT_EXEC);
601 argv_free(helper_argv);
602 if (retval) {
603 printk(KERN_INFO "Core dump to %s pipe failed\n",
604 cn.corename);
605 goto close_fail;
607 } else {
608 struct inode *inode;
610 if (cprm.limit < binfmt->min_coredump)
611 goto fail_unlock;
613 if (need_nonrelative && cn.corename[0] != '/') {
614 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
615 "to fully qualified path!\n",
616 task_tgid_vnr(current), current->comm);
617 printk(KERN_WARNING "Skipping core dump\n");
618 goto fail_unlock;
621 cprm.file = filp_open(cn.corename,
622 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
623 0600);
624 if (IS_ERR(cprm.file))
625 goto fail_unlock;
627 inode = file_inode(cprm.file);
628 if (inode->i_nlink > 1)
629 goto close_fail;
630 if (d_unhashed(cprm.file->f_path.dentry))
631 goto close_fail;
633 * AK: actually i see no reason to not allow this for named
634 * pipes etc, but keep the previous behaviour for now.
636 if (!S_ISREG(inode->i_mode))
637 goto close_fail;
639 * Dont allow local users get cute and trick others to coredump
640 * into their pre-created files.
642 if (!uid_eq(inode->i_uid, current_fsuid()))
643 goto close_fail;
644 if (!cprm.file->f_op || !cprm.file->f_op->write)
645 goto close_fail;
646 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
647 goto close_fail;
650 /* get us an unshared descriptor table; almost always a no-op */
651 retval = unshare_files(&displaced);
652 if (retval)
653 goto close_fail;
654 if (displaced)
655 put_files_struct(displaced);
656 core_dumped = !dump_interrupted() && binfmt->core_dump(&cprm);
658 if (ispipe && core_pipe_limit)
659 wait_for_dump_helpers(cprm.file);
660 close_fail:
661 if (cprm.file)
662 filp_close(cprm.file, NULL);
663 fail_dropcount:
664 if (ispipe)
665 atomic_dec(&core_dump_count);
666 fail_unlock:
667 kfree(cn.corename);
668 fail_corename:
669 coredump_finish(mm, core_dumped);
670 revert_creds(old_cred);
671 fail_creds:
672 put_cred(cred);
673 fail:
674 return;
678 * Core dumping helper functions. These are the only things you should
679 * do on a core-file: use only these functions to write out all the
680 * necessary info.
682 int dump_write(struct file *file, const void *addr, int nr)
684 return !dump_interrupted() &&
685 access_ok(VERIFY_READ, addr, nr) &&
686 file->f_op->write(file, addr, nr, &file->f_pos) == nr;
688 EXPORT_SYMBOL(dump_write);
690 int dump_seek(struct file *file, loff_t off)
692 int ret = 1;
694 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
695 if (dump_interrupted() ||
696 file->f_op->llseek(file, off, SEEK_CUR) < 0)
697 return 0;
698 } else {
699 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
701 if (!buf)
702 return 0;
703 while (off > 0) {
704 unsigned long n = off;
706 if (n > PAGE_SIZE)
707 n = PAGE_SIZE;
708 if (!dump_write(file, buf, n)) {
709 ret = 0;
710 break;
712 off -= n;
714 free_page((unsigned long)buf);
716 return ret;
718 EXPORT_SYMBOL(dump_seek);