Merge back earlier 'pm-sleep' material for v3.18.
[linux/fpc-iii.git] / fs / coredump.c
bloba93f7e6ea4cf935aea64e8266b221a2c53477628
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;
76 va_list arg_copy;
78 again:
79 free = cn->size - cn->used;
81 va_copy(arg_copy, arg);
82 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
83 va_end(arg_copy);
85 if (need < free) {
86 cn->used += need;
87 return 0;
90 if (!expand_corename(cn, cn->size + need - free + 1))
91 goto again;
93 return -ENOMEM;
96 static int cn_printf(struct core_name *cn, const char *fmt, ...)
98 va_list arg;
99 int ret;
101 va_start(arg, fmt);
102 ret = cn_vprintf(cn, fmt, arg);
103 va_end(arg);
105 return ret;
108 static int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
110 int cur = cn->used;
111 va_list arg;
112 int ret;
114 va_start(arg, fmt);
115 ret = cn_vprintf(cn, fmt, arg);
116 va_end(arg);
118 for (; cur < cn->used; ++cur) {
119 if (cn->corename[cur] == '/')
120 cn->corename[cur] = '!';
122 return ret;
125 static int cn_print_exe_file(struct core_name *cn)
127 struct file *exe_file;
128 char *pathbuf, *path;
129 int ret;
131 exe_file = get_mm_exe_file(current->mm);
132 if (!exe_file)
133 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
135 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
136 if (!pathbuf) {
137 ret = -ENOMEM;
138 goto put_exe_file;
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;
147 ret = cn_esc_printf(cn, "%s", path);
149 free_buf:
150 kfree(pathbuf);
151 put_exe_file:
152 fput(exe_file);
153 return ret;
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.
160 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
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;
168 cn->used = 0;
169 cn->corename = NULL;
170 if (expand_corename(cn, core_name_size))
171 return -ENOMEM;
172 cn->corename[0] = '\0';
174 if (ispipe)
175 ++pat_ptr;
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 /* uid */
203 case 'u':
204 err = cn_printf(cn, "%d", cred->uid);
205 break;
206 /* gid */
207 case 'g':
208 err = cn_printf(cn, "%d", cred->gid);
209 break;
210 case 'd':
211 err = cn_printf(cn, "%d",
212 __get_dumpable(cprm->mm_flags));
213 break;
214 /* signal that caused the coredump */
215 case 's':
216 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
217 break;
218 /* UNIX time of coredump */
219 case 't': {
220 struct timeval tv;
221 do_gettimeofday(&tv);
222 err = cn_printf(cn, "%lu", tv.tv_sec);
223 break;
225 /* hostname */
226 case 'h':
227 down_read(&uts_sem);
228 err = cn_esc_printf(cn, "%s",
229 utsname()->nodename);
230 up_read(&uts_sem);
231 break;
232 /* executable */
233 case 'e':
234 err = cn_esc_printf(cn, "%s", current->comm);
235 break;
236 case 'E':
237 err = cn_print_exe_file(cn);
238 break;
239 /* core limit size */
240 case 'c':
241 err = cn_printf(cn, "%lu",
242 rlimit(RLIMIT_CORE));
243 break;
244 default:
245 break;
247 ++pat_ptr;
250 if (err)
251 return err;
254 out:
255 /* Backward compatibility with core_uses_pid:
257 * If core_pattern does not include a %p (as is the default)
258 * and core_uses_pid is set, then .%pid will be appended to
259 * the filename. Do not do this for piped commands. */
260 if (!ispipe && !pid_in_pattern && core_uses_pid) {
261 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
262 if (err)
263 return err;
265 return ispipe;
268 static int zap_process(struct task_struct *start, int exit_code)
270 struct task_struct *t;
271 int nr = 0;
273 start->signal->group_exit_code = exit_code;
274 start->signal->group_stop_count = 0;
276 t = start;
277 do {
278 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
279 if (t != current && t->mm) {
280 sigaddset(&t->pending.signal, SIGKILL);
281 signal_wake_up(t, 1);
282 nr++;
284 } while_each_thread(start, t);
286 return nr;
289 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
290 struct core_state *core_state, int exit_code)
292 struct task_struct *g, *p;
293 unsigned long flags;
294 int nr = -EAGAIN;
296 spin_lock_irq(&tsk->sighand->siglock);
297 if (!signal_group_exit(tsk->signal)) {
298 mm->core_state = core_state;
299 nr = zap_process(tsk, exit_code);
300 tsk->signal->group_exit_task = tsk;
301 /* ignore all signals except SIGKILL, see prepare_signal() */
302 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
303 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
305 spin_unlock_irq(&tsk->sighand->siglock);
306 if (unlikely(nr < 0))
307 return nr;
309 tsk->flags |= PF_DUMPCORE;
310 if (atomic_read(&mm->mm_users) == nr + 1)
311 goto done;
313 * We should find and kill all tasks which use this mm, and we should
314 * count them correctly into ->nr_threads. We don't take tasklist
315 * lock, but this is safe wrt:
317 * fork:
318 * None of sub-threads can fork after zap_process(leader). All
319 * processes which were created before this point should be
320 * visible to zap_threads() because copy_process() adds the new
321 * process to the tail of init_task.tasks list, and lock/unlock
322 * of ->siglock provides a memory barrier.
324 * do_exit:
325 * The caller holds mm->mmap_sem. This means that the task which
326 * uses this mm can't pass exit_mm(), so it can't exit or clear
327 * its ->mm.
329 * de_thread:
330 * It does list_replace_rcu(&leader->tasks, &current->tasks),
331 * we must see either old or new leader, this does not matter.
332 * However, it can change p->sighand, so lock_task_sighand(p)
333 * must be used. Since p->mm != NULL and we hold ->mmap_sem
334 * it can't fail.
336 * Note also that "g" can be the old leader with ->mm == NULL
337 * and already unhashed and thus removed from ->thread_group.
338 * This is OK, __unhash_process()->list_del_rcu() does not
339 * clear the ->next pointer, we will find the new leader via
340 * next_thread().
342 rcu_read_lock();
343 for_each_process(g) {
344 if (g == tsk->group_leader)
345 continue;
346 if (g->flags & PF_KTHREAD)
347 continue;
348 p = g;
349 do {
350 if (p->mm) {
351 if (unlikely(p->mm == mm)) {
352 lock_task_sighand(p, &flags);
353 nr += zap_process(p, exit_code);
354 p->signal->flags = SIGNAL_GROUP_EXIT;
355 unlock_task_sighand(p, &flags);
357 break;
359 } while_each_thread(g, p);
361 rcu_read_unlock();
362 done:
363 atomic_set(&core_state->nr_threads, nr);
364 return nr;
367 static int coredump_wait(int exit_code, struct core_state *core_state)
369 struct task_struct *tsk = current;
370 struct mm_struct *mm = tsk->mm;
371 int core_waiters = -EBUSY;
373 init_completion(&core_state->startup);
374 core_state->dumper.task = tsk;
375 core_state->dumper.next = NULL;
377 down_write(&mm->mmap_sem);
378 if (!mm->core_state)
379 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
380 up_write(&mm->mmap_sem);
382 if (core_waiters > 0) {
383 struct core_thread *ptr;
385 wait_for_completion(&core_state->startup);
387 * Wait for all the threads to become inactive, so that
388 * all the thread context (extended register state, like
389 * fpu etc) gets copied to the memory.
391 ptr = core_state->dumper.next;
392 while (ptr != NULL) {
393 wait_task_inactive(ptr->task, 0);
394 ptr = ptr->next;
398 return core_waiters;
401 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
403 struct core_thread *curr, *next;
404 struct task_struct *task;
406 spin_lock_irq(&current->sighand->siglock);
407 if (core_dumped && !__fatal_signal_pending(current))
408 current->signal->group_exit_code |= 0x80;
409 current->signal->group_exit_task = NULL;
410 current->signal->flags = SIGNAL_GROUP_EXIT;
411 spin_unlock_irq(&current->sighand->siglock);
413 next = mm->core_state->dumper.next;
414 while ((curr = next) != NULL) {
415 next = curr->next;
416 task = curr->task;
418 * see exit_mm(), curr->task must not see
419 * ->task == NULL before we read ->next.
421 smp_mb();
422 curr->task = NULL;
423 wake_up_process(task);
426 mm->core_state = NULL;
429 static bool dump_interrupted(void)
432 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
433 * can do try_to_freeze() and check __fatal_signal_pending(),
434 * but then we need to teach dump_write() to restart and clear
435 * TIF_SIGPENDING.
437 return signal_pending(current);
440 static void wait_for_dump_helpers(struct file *file)
442 struct pipe_inode_info *pipe = file->private_data;
444 pipe_lock(pipe);
445 pipe->readers++;
446 pipe->writers--;
447 wake_up_interruptible_sync(&pipe->wait);
448 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
449 pipe_unlock(pipe);
452 * We actually want wait_event_freezable() but then we need
453 * to clear TIF_SIGPENDING and improve dump_interrupted().
455 wait_event_interruptible(pipe->wait, pipe->readers == 1);
457 pipe_lock(pipe);
458 pipe->readers--;
459 pipe->writers++;
460 pipe_unlock(pipe);
464 * umh_pipe_setup
465 * helper function to customize the process used
466 * to collect the core in userspace. Specifically
467 * it sets up a pipe and installs it as fd 0 (stdin)
468 * for the process. Returns 0 on success, or
469 * PTR_ERR on failure.
470 * Note that it also sets the core limit to 1. This
471 * is a special value that we use to trap recursive
472 * core dumps
474 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
476 struct file *files[2];
477 struct coredump_params *cp = (struct coredump_params *)info->data;
478 int err = create_pipe_files(files, 0);
479 if (err)
480 return err;
482 cp->file = files[1];
484 err = replace_fd(0, files[0], 0);
485 fput(files[0]);
486 /* and disallow core files too */
487 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
489 return err;
492 void do_coredump(const siginfo_t *siginfo)
494 struct core_state core_state;
495 struct core_name cn;
496 struct mm_struct *mm = current->mm;
497 struct linux_binfmt * binfmt;
498 const struct cred *old_cred;
499 struct cred *cred;
500 int retval = 0;
501 int flag = 0;
502 int ispipe;
503 struct files_struct *displaced;
504 bool need_nonrelative = false;
505 bool core_dumped = false;
506 static atomic_t core_dump_count = ATOMIC_INIT(0);
507 struct coredump_params cprm = {
508 .siginfo = siginfo,
509 .regs = signal_pt_regs(),
510 .limit = rlimit(RLIMIT_CORE),
512 * We must use the same mm->flags while dumping core to avoid
513 * inconsistency of bit flags, since this flag is not protected
514 * by any locks.
516 .mm_flags = mm->flags,
519 audit_core_dumps(siginfo->si_signo);
521 binfmt = mm->binfmt;
522 if (!binfmt || !binfmt->core_dump)
523 goto fail;
524 if (!__get_dumpable(cprm.mm_flags))
525 goto fail;
527 cred = prepare_creds();
528 if (!cred)
529 goto fail;
531 * We cannot trust fsuid as being the "true" uid of the process
532 * nor do we know its entire history. We only know it was tainted
533 * so we dump it as root in mode 2, and only into a controlled
534 * environment (pipe handler or fully qualified path).
536 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
537 /* Setuid core dump mode */
538 flag = O_EXCL; /* Stop rewrite attacks */
539 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
540 need_nonrelative = true;
543 retval = coredump_wait(siginfo->si_signo, &core_state);
544 if (retval < 0)
545 goto fail_creds;
547 old_cred = override_creds(cred);
549 ispipe = format_corename(&cn, &cprm);
551 if (ispipe) {
552 int dump_count;
553 char **helper_argv;
554 struct subprocess_info *sub_info;
556 if (ispipe < 0) {
557 printk(KERN_WARNING "format_corename failed\n");
558 printk(KERN_WARNING "Aborting core\n");
559 goto fail_unlock;
562 if (cprm.limit == 1) {
563 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
565 * Normally core limits are irrelevant to pipes, since
566 * we're not writing to the file system, but we use
567 * cprm.limit of 1 here as a speacial value, this is a
568 * consistent way to catch recursive crashes.
569 * We can still crash if the core_pattern binary sets
570 * RLIM_CORE = !1, but it runs as root, and can do
571 * lots of stupid things.
573 * Note that we use task_tgid_vnr here to grab the pid
574 * of the process group leader. That way we get the
575 * right pid if a thread in a multi-threaded
576 * core_pattern process dies.
578 printk(KERN_WARNING
579 "Process %d(%s) has RLIMIT_CORE set to 1\n",
580 task_tgid_vnr(current), current->comm);
581 printk(KERN_WARNING "Aborting core\n");
582 goto fail_unlock;
584 cprm.limit = RLIM_INFINITY;
586 dump_count = atomic_inc_return(&core_dump_count);
587 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
588 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
589 task_tgid_vnr(current), current->comm);
590 printk(KERN_WARNING "Skipping core dump\n");
591 goto fail_dropcount;
594 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
595 if (!helper_argv) {
596 printk(KERN_WARNING "%s failed to allocate memory\n",
597 __func__);
598 goto fail_dropcount;
601 retval = -ENOMEM;
602 sub_info = call_usermodehelper_setup(helper_argv[0],
603 helper_argv, NULL, GFP_KERNEL,
604 umh_pipe_setup, NULL, &cprm);
605 if (sub_info)
606 retval = call_usermodehelper_exec(sub_info,
607 UMH_WAIT_EXEC);
609 argv_free(helper_argv);
610 if (retval) {
611 printk(KERN_INFO "Core dump to |%s pipe failed\n",
612 cn.corename);
613 goto close_fail;
615 } else {
616 struct inode *inode;
618 if (cprm.limit < binfmt->min_coredump)
619 goto fail_unlock;
621 if (need_nonrelative && cn.corename[0] != '/') {
622 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
623 "to fully qualified path!\n",
624 task_tgid_vnr(current), current->comm);
625 printk(KERN_WARNING "Skipping core dump\n");
626 goto fail_unlock;
629 cprm.file = filp_open(cn.corename,
630 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
631 0600);
632 if (IS_ERR(cprm.file))
633 goto fail_unlock;
635 inode = file_inode(cprm.file);
636 if (inode->i_nlink > 1)
637 goto close_fail;
638 if (d_unhashed(cprm.file->f_path.dentry))
639 goto close_fail;
641 * AK: actually i see no reason to not allow this for named
642 * pipes etc, but keep the previous behaviour for now.
644 if (!S_ISREG(inode->i_mode))
645 goto close_fail;
647 * Dont allow local users get cute and trick others to coredump
648 * into their pre-created files.
650 if (!uid_eq(inode->i_uid, current_fsuid()))
651 goto close_fail;
652 if (!cprm.file->f_op->write)
653 goto close_fail;
654 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
655 goto close_fail;
658 /* get us an unshared descriptor table; almost always a no-op */
659 retval = unshare_files(&displaced);
660 if (retval)
661 goto close_fail;
662 if (displaced)
663 put_files_struct(displaced);
664 if (!dump_interrupted()) {
665 file_start_write(cprm.file);
666 core_dumped = binfmt->core_dump(&cprm);
667 file_end_write(cprm.file);
669 if (ispipe && core_pipe_limit)
670 wait_for_dump_helpers(cprm.file);
671 close_fail:
672 if (cprm.file)
673 filp_close(cprm.file, NULL);
674 fail_dropcount:
675 if (ispipe)
676 atomic_dec(&core_dump_count);
677 fail_unlock:
678 kfree(cn.corename);
679 coredump_finish(mm, core_dumped);
680 revert_creds(old_cred);
681 fail_creds:
682 put_cred(cred);
683 fail:
684 return;
688 * Core dumping helper functions. These are the only things you should
689 * do on a core-file: use only these functions to write out all the
690 * necessary info.
692 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
694 struct file *file = cprm->file;
695 loff_t pos = file->f_pos;
696 ssize_t n;
697 if (cprm->written + nr > cprm->limit)
698 return 0;
699 while (nr) {
700 if (dump_interrupted())
701 return 0;
702 n = __kernel_write(file, addr, nr, &pos);
703 if (n <= 0)
704 return 0;
705 file->f_pos = pos;
706 cprm->written += n;
707 nr -= n;
709 return 1;
711 EXPORT_SYMBOL(dump_emit);
713 int dump_skip(struct coredump_params *cprm, size_t nr)
715 static char zeroes[PAGE_SIZE];
716 struct file *file = cprm->file;
717 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
718 if (cprm->written + nr > cprm->limit)
719 return 0;
720 if (dump_interrupted() ||
721 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
722 return 0;
723 cprm->written += nr;
724 return 1;
725 } else {
726 while (nr > PAGE_SIZE) {
727 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
728 return 0;
729 nr -= PAGE_SIZE;
731 return dump_emit(cprm, zeroes, nr);
734 EXPORT_SYMBOL(dump_skip);
736 int dump_align(struct coredump_params *cprm, int align)
738 unsigned mod = cprm->written & (align - 1);
739 if (align & (align - 1))
740 return 0;
741 return mod ? dump_skip(cprm, align - mod) : 1;
743 EXPORT_SYMBOL(dump_align);