blob: eb9c92c9b20f5de5e325d7a05e5d055d0816989e [file] [log] [blame]
Alex Kelly10c28d92012-09-26 21:52:08 -04001#include <linux/slab.h>
2#include <linux/file.h>
3#include <linux/fdtable.h>
Andrey Ryabinin70d78fe2016-11-10 10:46:38 -08004#include <linux/freezer.h>
Alex Kelly10c28d92012-09-26 21:52:08 -04005#include <linux/mm.h>
6#include <linux/stat.h>
7#include <linux/fcntl.h>
8#include <linux/swap.h>
9#include <linux/string.h>
10#include <linux/init.h>
11#include <linux/pagemap.h>
12#include <linux/perf_event.h>
13#include <linux/highmem.h>
14#include <linux/spinlock.h>
15#include <linux/key.h>
16#include <linux/personality.h>
17#include <linux/binfmts.h>
Alex Kelly179899f2012-10-04 17:15:24 -070018#include <linux/coredump.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040019#include <linux/utsname.h>
20#include <linux/pid_namespace.h>
21#include <linux/module.h>
22#include <linux/namei.h>
23#include <linux/mount.h>
24#include <linux/security.h>
25#include <linux/syscalls.h>
26#include <linux/tsacct_kern.h>
27#include <linux/cn_proc.h>
28#include <linux/audit.h>
29#include <linux/tracehook.h>
30#include <linux/kmod.h>
31#include <linux/fsnotify.h>
32#include <linux/fs_struct.h>
33#include <linux/pipe_fs_i.h>
34#include <linux/oom.h>
35#include <linux/compat.h>
Jann Horn378c6522016-03-22 14:25:36 -070036#include <linux/sched.h>
37#include <linux/fs.h>
38#include <linux/path.h>
Arnd Bergmann03927c82015-11-25 16:22:25 +010039#include <linux/timekeeping.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040040
41#include <asm/uaccess.h>
42#include <asm/mmu_context.h>
43#include <asm/tlb.h>
44#include <asm/exec.h>
45
46#include <trace/events/task.h>
47#include "internal.h"
Alex Kelly10c28d92012-09-26 21:52:08 -040048
49#include <trace/events/sched.h>
50
51int core_uses_pid;
Alex Kelly10c28d92012-09-26 21:52:08 -040052unsigned int core_pipe_limit;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070053char core_pattern[CORENAME_MAX_SIZE] = "core";
54static int core_name_size = CORENAME_MAX_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -040055
56struct core_name {
57 char *corename;
58 int used, size;
59};
Alex Kelly10c28d92012-09-26 21:52:08 -040060
61/* The maximal length of core_pattern is also specified in sysctl.c */
62
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070063static int expand_corename(struct core_name *cn, int size)
Alex Kelly10c28d92012-09-26 21:52:08 -040064{
Oleg Nesterove7fd1542013-07-03 15:08:16 -070065 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
Alex Kelly10c28d92012-09-26 21:52:08 -040066
Oleg Nesterove7fd1542013-07-03 15:08:16 -070067 if (!corename)
Alex Kelly10c28d92012-09-26 21:52:08 -040068 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -040069
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070070 if (size > core_name_size) /* racy but harmless */
71 core_name_size = size;
72
73 cn->size = ksize(corename);
Oleg Nesterove7fd1542013-07-03 15:08:16 -070074 cn->corename = corename;
Alex Kelly10c28d92012-09-26 21:52:08 -040075 return 0;
76}
77
Nicolas Ioossb4176b72015-06-25 15:03:53 -070078static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
79 va_list arg)
Alex Kelly10c28d92012-09-26 21:52:08 -040080{
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070081 int free, need;
Eric Dumazet404ca802014-04-19 10:15:07 -070082 va_list arg_copy;
Alex Kelly10c28d92012-09-26 21:52:08 -040083
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070084again:
85 free = cn->size - cn->used;
Eric Dumazet404ca802014-04-19 10:15:07 -070086
87 va_copy(arg_copy, arg);
88 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
89 va_end(arg_copy);
90
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070091 if (need < free) {
92 cn->used += need;
93 return 0;
94 }
Alex Kelly10c28d92012-09-26 21:52:08 -040095
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070096 if (!expand_corename(cn, cn->size + need - free + 1))
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070097 goto again;
Alex Kelly10c28d92012-09-26 21:52:08 -040098
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070099 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -0400100}
101
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700102static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
Oleg Nesterovbc03c692013-07-03 15:08:17 -0700103{
104 va_list arg;
105 int ret;
106
107 va_start(arg, fmt);
108 ret = cn_vprintf(cn, fmt, arg);
109 va_end(arg);
110
111 return ret;
112}
113
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700114static __printf(2, 3)
115int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
Alex Kelly10c28d92012-09-26 21:52:08 -0400116{
Oleg Nesterov923bed032013-07-03 15:08:20 -0700117 int cur = cn->used;
118 va_list arg;
119 int ret;
120
121 va_start(arg, fmt);
122 ret = cn_vprintf(cn, fmt, arg);
123 va_end(arg);
124
Jann Hornac94b6e2016-01-20 15:00:08 -0800125 if (ret == 0) {
126 /*
127 * Ensure that this coredump name component can't cause the
128 * resulting corefile path to consist of a ".." or ".".
129 */
130 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
131 (cn->used - cur == 2 && cn->corename[cur] == '.'
132 && cn->corename[cur+1] == '.'))
133 cn->corename[cur] = '!';
134
135 /*
136 * Empty names are fishy and could be used to create a "//" in a
137 * corefile name, causing the coredump to happen one directory
138 * level too high. Enforce that all components of the core
139 * pattern are at least one character long.
140 */
141 if (cn->used == cur)
142 ret = cn_printf(cn, "!");
143 }
144
Oleg Nesterov923bed032013-07-03 15:08:20 -0700145 for (; cur < cn->used; ++cur) {
146 if (cn->corename[cur] == '/')
147 cn->corename[cur] = '!';
148 }
149 return ret;
Alex Kelly10c28d92012-09-26 21:52:08 -0400150}
151
152static int cn_print_exe_file(struct core_name *cn)
153{
154 struct file *exe_file;
155 char *pathbuf, *path;
156 int ret;
157
158 exe_file = get_mm_exe_file(current->mm);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700159 if (!exe_file)
160 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400161
162 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
163 if (!pathbuf) {
164 ret = -ENOMEM;
165 goto put_exe_file;
166 }
167
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200168 path = file_path(exe_file, pathbuf, PATH_MAX);
Alex Kelly10c28d92012-09-26 21:52:08 -0400169 if (IS_ERR(path)) {
170 ret = PTR_ERR(path);
171 goto free_buf;
172 }
173
Oleg Nesterov923bed032013-07-03 15:08:20 -0700174 ret = cn_esc_printf(cn, "%s", path);
Alex Kelly10c28d92012-09-26 21:52:08 -0400175
176free_buf:
177 kfree(pathbuf);
178put_exe_file:
179 fput(exe_file);
180 return ret;
181}
182
183/* format_corename will inspect the pattern parameter, and output a
184 * name into corename, which must have space for at least
185 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
186 */
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700187static int format_corename(struct core_name *cn, struct coredump_params *cprm)
Alex Kelly10c28d92012-09-26 21:52:08 -0400188{
189 const struct cred *cred = current_cred();
190 const char *pat_ptr = core_pattern;
191 int ispipe = (*pat_ptr == '|');
192 int pid_in_pattern = 0;
193 int err = 0;
194
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700195 cn->used = 0;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -0700196 cn->corename = NULL;
197 if (expand_corename(cn, core_name_size))
Alex Kelly10c28d92012-09-26 21:52:08 -0400198 return -ENOMEM;
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700199 cn->corename[0] = '\0';
200
201 if (ispipe)
202 ++pat_ptr;
Alex Kelly10c28d92012-09-26 21:52:08 -0400203
204 /* Repeat as long as we have more pattern to process and more output
205 space */
206 while (*pat_ptr) {
207 if (*pat_ptr != '%') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400208 err = cn_printf(cn, "%c", *pat_ptr++);
209 } else {
210 switch (*++pat_ptr) {
211 /* single % at the end, drop that */
212 case 0:
213 goto out;
214 /* Double percent, output one percent */
215 case '%':
216 err = cn_printf(cn, "%c", '%');
217 break;
218 /* pid */
219 case 'p':
220 pid_in_pattern = 1;
221 err = cn_printf(cn, "%d",
222 task_tgid_vnr(current));
223 break;
Stéphane Graber65aafb12013-09-11 14:24:32 -0700224 /* global pid */
225 case 'P':
226 err = cn_printf(cn, "%d",
227 task_tgid_nr(current));
228 break;
Oleg Nesterovb03023e2014-10-13 15:53:35 -0700229 case 'i':
230 err = cn_printf(cn, "%d",
231 task_pid_vnr(current));
232 break;
233 case 'I':
234 err = cn_printf(cn, "%d",
235 task_pid_nr(current));
236 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400237 /* uid */
238 case 'u':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700239 err = cn_printf(cn, "%u",
240 from_kuid(&init_user_ns,
241 cred->uid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400242 break;
243 /* gid */
244 case 'g':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700245 err = cn_printf(cn, "%u",
246 from_kgid(&init_user_ns,
247 cred->gid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400248 break;
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700249 case 'd':
250 err = cn_printf(cn, "%d",
251 __get_dumpable(cprm->mm_flags));
252 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400253 /* signal that caused the coredump */
254 case 's':
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700255 err = cn_printf(cn, "%d",
256 cprm->siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400257 break;
258 /* UNIX time of coredump */
259 case 't': {
Arnd Bergmann03927c82015-11-25 16:22:25 +0100260 time64_t time;
261
262 time = ktime_get_real_seconds();
263 err = cn_printf(cn, "%lld", time);
Alex Kelly10c28d92012-09-26 21:52:08 -0400264 break;
265 }
266 /* hostname */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700267 case 'h':
Alex Kelly10c28d92012-09-26 21:52:08 -0400268 down_read(&uts_sem);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700269 err = cn_esc_printf(cn, "%s",
Alex Kelly10c28d92012-09-26 21:52:08 -0400270 utsname()->nodename);
271 up_read(&uts_sem);
Alex Kelly10c28d92012-09-26 21:52:08 -0400272 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400273 /* executable */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700274 case 'e':
275 err = cn_esc_printf(cn, "%s", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400276 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400277 case 'E':
278 err = cn_print_exe_file(cn);
279 break;
280 /* core limit size */
281 case 'c':
282 err = cn_printf(cn, "%lu",
283 rlimit(RLIMIT_CORE));
284 break;
285 default:
286 break;
287 }
288 ++pat_ptr;
289 }
290
291 if (err)
292 return err;
293 }
294
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700295out:
Alex Kelly10c28d92012-09-26 21:52:08 -0400296 /* Backward compatibility with core_uses_pid:
297 *
298 * If core_pattern does not include a %p (as is the default)
299 * and core_uses_pid is set, then .%pid will be appended to
300 * the filename. Do not do this for piped commands. */
301 if (!ispipe && !pid_in_pattern && core_uses_pid) {
302 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
303 if (err)
304 return err;
305 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400306 return ispipe;
307}
308
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800309static int zap_process(struct task_struct *start, int exit_code, int flags)
Alex Kelly10c28d92012-09-26 21:52:08 -0400310{
311 struct task_struct *t;
312 int nr = 0;
313
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800314 /* ignore all signals except SIGKILL, see prepare_signal() */
315 start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
Alex Kelly10c28d92012-09-26 21:52:08 -0400316 start->signal->group_exit_code = exit_code;
317 start->signal->group_stop_count = 0;
318
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800319 for_each_thread(start, t) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400320 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
321 if (t != current && t->mm) {
322 sigaddset(&t->pending.signal, SIGKILL);
323 signal_wake_up(t, 1);
324 nr++;
325 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800326 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400327
328 return nr;
329}
330
Oleg Nesterov403bad72013-04-30 15:28:10 -0700331static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
332 struct core_state *core_state, int exit_code)
Alex Kelly10c28d92012-09-26 21:52:08 -0400333{
334 struct task_struct *g, *p;
335 unsigned long flags;
336 int nr = -EAGAIN;
337
338 spin_lock_irq(&tsk->sighand->siglock);
339 if (!signal_group_exit(tsk->signal)) {
340 mm->core_state = core_state;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700341 tsk->signal->group_exit_task = tsk;
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800342 nr = zap_process(tsk, exit_code, 0);
Oleg Nesterov403bad72013-04-30 15:28:10 -0700343 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
Alex Kelly10c28d92012-09-26 21:52:08 -0400344 }
345 spin_unlock_irq(&tsk->sighand->siglock);
346 if (unlikely(nr < 0))
347 return nr;
348
Silesh C Vaed8adb2014-07-23 13:59:59 -0700349 tsk->flags |= PF_DUMPCORE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400350 if (atomic_read(&mm->mm_users) == nr + 1)
351 goto done;
352 /*
353 * We should find and kill all tasks which use this mm, and we should
354 * count them correctly into ->nr_threads. We don't take tasklist
355 * lock, but this is safe wrt:
356 *
357 * fork:
358 * None of sub-threads can fork after zap_process(leader). All
359 * processes which were created before this point should be
360 * visible to zap_threads() because copy_process() adds the new
361 * process to the tail of init_task.tasks list, and lock/unlock
362 * of ->siglock provides a memory barrier.
363 *
364 * do_exit:
365 * The caller holds mm->mmap_sem. This means that the task which
366 * uses this mm can't pass exit_mm(), so it can't exit or clear
367 * its ->mm.
368 *
369 * de_thread:
370 * It does list_replace_rcu(&leader->tasks, &current->tasks),
371 * we must see either old or new leader, this does not matter.
372 * However, it can change p->sighand, so lock_task_sighand(p)
373 * must be used. Since p->mm != NULL and we hold ->mmap_sem
374 * it can't fail.
375 *
376 * Note also that "g" can be the old leader with ->mm == NULL
377 * and already unhashed and thus removed from ->thread_group.
378 * This is OK, __unhash_process()->list_del_rcu() does not
379 * clear the ->next pointer, we will find the new leader via
380 * next_thread().
381 */
382 rcu_read_lock();
383 for_each_process(g) {
384 if (g == tsk->group_leader)
385 continue;
386 if (g->flags & PF_KTHREAD)
387 continue;
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800388
389 for_each_thread(g, p) {
390 if (unlikely(!p->mm))
391 continue;
392 if (unlikely(p->mm == mm)) {
393 lock_task_sighand(p, &flags);
394 nr += zap_process(p, exit_code,
395 SIGNAL_GROUP_EXIT);
396 unlock_task_sighand(p, &flags);
Alex Kelly10c28d92012-09-26 21:52:08 -0400397 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800398 break;
399 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400400 }
401 rcu_read_unlock();
402done:
403 atomic_set(&core_state->nr_threads, nr);
404 return nr;
405}
406
407static int coredump_wait(int exit_code, struct core_state *core_state)
408{
409 struct task_struct *tsk = current;
410 struct mm_struct *mm = tsk->mm;
411 int core_waiters = -EBUSY;
412
413 init_completion(&core_state->startup);
414 core_state->dumper.task = tsk;
415 core_state->dumper.next = NULL;
416
Michal Hocko4136c262016-05-23 16:25:57 -0700417 if (down_write_killable(&mm->mmap_sem))
418 return -EINTR;
419
Alex Kelly10c28d92012-09-26 21:52:08 -0400420 if (!mm->core_state)
421 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
422 up_write(&mm->mmap_sem);
423
424 if (core_waiters > 0) {
425 struct core_thread *ptr;
426
Andrey Ryabinin70d78fe2016-11-10 10:46:38 -0800427 freezer_do_not_count();
Alex Kelly10c28d92012-09-26 21:52:08 -0400428 wait_for_completion(&core_state->startup);
Andrey Ryabinin70d78fe2016-11-10 10:46:38 -0800429 freezer_count();
Alex Kelly10c28d92012-09-26 21:52:08 -0400430 /*
431 * Wait for all the threads to become inactive, so that
432 * all the thread context (extended register state, like
433 * fpu etc) gets copied to the memory.
434 */
435 ptr = core_state->dumper.next;
436 while (ptr != NULL) {
437 wait_task_inactive(ptr->task, 0);
438 ptr = ptr->next;
439 }
440 }
441
442 return core_waiters;
443}
444
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700445static void coredump_finish(struct mm_struct *mm, bool core_dumped)
Alex Kelly10c28d92012-09-26 21:52:08 -0400446{
447 struct core_thread *curr, *next;
448 struct task_struct *task;
449
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700450 spin_lock_irq(&current->sighand->siglock);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700451 if (core_dumped && !__fatal_signal_pending(current))
452 current->signal->group_exit_code |= 0x80;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700453 current->signal->group_exit_task = NULL;
454 current->signal->flags = SIGNAL_GROUP_EXIT;
455 spin_unlock_irq(&current->sighand->siglock);
456
Alex Kelly10c28d92012-09-26 21:52:08 -0400457 next = mm->core_state->dumper.next;
458 while ((curr = next) != NULL) {
459 next = curr->next;
460 task = curr->task;
461 /*
462 * see exit_mm(), curr->task must not see
463 * ->task == NULL before we read ->next.
464 */
465 smp_mb();
466 curr->task = NULL;
467 wake_up_process(task);
468 }
469
470 mm->core_state = NULL;
471}
472
Oleg Nesterov528f8272013-04-30 15:28:15 -0700473static bool dump_interrupted(void)
474{
475 /*
476 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
477 * can do try_to_freeze() and check __fatal_signal_pending(),
478 * but then we need to teach dump_write() to restart and clear
479 * TIF_SIGPENDING.
480 */
481 return signal_pending(current);
482}
483
Alex Kelly10c28d92012-09-26 21:52:08 -0400484static void wait_for_dump_helpers(struct file *file)
485{
Al Virode32ec42013-03-21 11:16:56 -0400486 struct pipe_inode_info *pipe = file->private_data;
Alex Kelly10c28d92012-09-26 21:52:08 -0400487
488 pipe_lock(pipe);
489 pipe->readers++;
490 pipe->writers--;
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700491 wake_up_interruptible_sync(&pipe->wait);
492 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
493 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400494
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700495 /*
496 * We actually want wait_event_freezable() but then we need
497 * to clear TIF_SIGPENDING and improve dump_interrupted().
498 */
499 wait_event_interruptible(pipe->wait, pipe->readers == 1);
Alex Kelly10c28d92012-09-26 21:52:08 -0400500
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700501 pipe_lock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400502 pipe->readers--;
503 pipe->writers++;
504 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400505}
506
507/*
508 * umh_pipe_setup
509 * helper function to customize the process used
510 * to collect the core in userspace. Specifically
511 * it sets up a pipe and installs it as fd 0 (stdin)
512 * for the process. Returns 0 on success, or
513 * PTR_ERR on failure.
514 * Note that it also sets the core limit to 1. This
515 * is a special value that we use to trap recursive
516 * core dumps
517 */
518static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
519{
520 struct file *files[2];
521 struct coredump_params *cp = (struct coredump_params *)info->data;
522 int err = create_pipe_files(files, 0);
523 if (err)
524 return err;
525
526 cp->file = files[1];
527
Al Viro45525b22012-10-16 13:30:07 -0400528 err = replace_fd(0, files[0], 0);
529 fput(files[0]);
Alex Kelly10c28d92012-09-26 21:52:08 -0400530 /* and disallow core files too */
531 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
532
Al Viro45525b22012-10-16 13:30:07 -0400533 return err;
Alex Kelly10c28d92012-09-26 21:52:08 -0400534}
535
Al Viroec579412013-10-13 17:57:29 -0400536void do_coredump(const siginfo_t *siginfo)
Alex Kelly10c28d92012-09-26 21:52:08 -0400537{
538 struct core_state core_state;
539 struct core_name cn;
540 struct mm_struct *mm = current->mm;
541 struct linux_binfmt * binfmt;
542 const struct cred *old_cred;
543 struct cred *cred;
544 int retval = 0;
Alex Kelly10c28d92012-09-26 21:52:08 -0400545 int ispipe;
546 struct files_struct *displaced;
Jann Hornfbb18162015-09-09 15:38:28 -0700547 /* require nonrelative corefile path and be extra careful */
548 bool need_suid_safe = false;
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700549 bool core_dumped = false;
Alex Kelly10c28d92012-09-26 21:52:08 -0400550 static atomic_t core_dump_count = ATOMIC_INIT(0);
551 struct coredump_params cprm = {
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700552 .siginfo = siginfo,
Al Viro541880d2012-11-05 13:11:26 -0500553 .regs = signal_pt_regs(),
Alex Kelly10c28d92012-09-26 21:52:08 -0400554 .limit = rlimit(RLIMIT_CORE),
555 /*
556 * We must use the same mm->flags while dumping core to avoid
557 * inconsistency of bit flags, since this flag is not protected
558 * by any locks.
559 */
560 .mm_flags = mm->flags,
561 };
562
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700563 audit_core_dumps(siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400564
565 binfmt = mm->binfmt;
566 if (!binfmt || !binfmt->core_dump)
567 goto fail;
568 if (!__get_dumpable(cprm.mm_flags))
569 goto fail;
570
571 cred = prepare_creds();
572 if (!cred)
573 goto fail;
574 /*
575 * We cannot trust fsuid as being the "true" uid of the process
576 * nor do we know its entire history. We only know it was tainted
577 * so we dump it as root in mode 2, and only into a controlled
578 * environment (pipe handler or fully qualified path).
579 */
Kees Cooke579d2c2013-02-27 17:03:15 -0800580 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400581 /* Setuid core dump mode */
Alex Kelly10c28d92012-09-26 21:52:08 -0400582 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
Jann Hornfbb18162015-09-09 15:38:28 -0700583 need_suid_safe = true;
Alex Kelly10c28d92012-09-26 21:52:08 -0400584 }
585
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700586 retval = coredump_wait(siginfo->si_signo, &core_state);
Alex Kelly10c28d92012-09-26 21:52:08 -0400587 if (retval < 0)
588 goto fail_creds;
589
590 old_cred = override_creds(cred);
591
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700592 ispipe = format_corename(&cn, &cprm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400593
Lucas De Marchifb96c472013-04-30 15:28:06 -0700594 if (ispipe) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400595 int dump_count;
596 char **helper_argv;
Lucas De Marchi907ed132013-04-30 15:28:07 -0700597 struct subprocess_info *sub_info;
Alex Kelly10c28d92012-09-26 21:52:08 -0400598
599 if (ispipe < 0) {
600 printk(KERN_WARNING "format_corename failed\n");
601 printk(KERN_WARNING "Aborting core\n");
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700602 goto fail_unlock;
Alex Kelly10c28d92012-09-26 21:52:08 -0400603 }
604
605 if (cprm.limit == 1) {
606 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
607 *
608 * Normally core limits are irrelevant to pipes, since
609 * we're not writing to the file system, but we use
Bastien Nocerafcbc32b2015-02-05 14:35:05 +0100610 * cprm.limit of 1 here as a special value, this is a
Alex Kelly10c28d92012-09-26 21:52:08 -0400611 * consistent way to catch recursive crashes.
612 * We can still crash if the core_pattern binary sets
613 * RLIM_CORE = !1, but it runs as root, and can do
614 * lots of stupid things.
615 *
616 * Note that we use task_tgid_vnr here to grab the pid
617 * of the process group leader. That way we get the
618 * right pid if a thread in a multi-threaded
619 * core_pattern process dies.
620 */
621 printk(KERN_WARNING
622 "Process %d(%s) has RLIMIT_CORE set to 1\n",
623 task_tgid_vnr(current), current->comm);
624 printk(KERN_WARNING "Aborting core\n");
625 goto fail_unlock;
626 }
627 cprm.limit = RLIM_INFINITY;
628
629 dump_count = atomic_inc_return(&core_dump_count);
630 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
631 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
632 task_tgid_vnr(current), current->comm);
633 printk(KERN_WARNING "Skipping core dump\n");
634 goto fail_dropcount;
635 }
636
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700637 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
Alex Kelly10c28d92012-09-26 21:52:08 -0400638 if (!helper_argv) {
639 printk(KERN_WARNING "%s failed to allocate memory\n",
640 __func__);
641 goto fail_dropcount;
642 }
643
Lucas De Marchi907ed132013-04-30 15:28:07 -0700644 retval = -ENOMEM;
645 sub_info = call_usermodehelper_setup(helper_argv[0],
646 helper_argv, NULL, GFP_KERNEL,
647 umh_pipe_setup, NULL, &cprm);
648 if (sub_info)
649 retval = call_usermodehelper_exec(sub_info,
650 UMH_WAIT_EXEC);
651
Alex Kelly10c28d92012-09-26 21:52:08 -0400652 argv_free(helper_argv);
653 if (retval) {
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700654 printk(KERN_INFO "Core dump to |%s pipe failed\n",
Alex Kelly10c28d92012-09-26 21:52:08 -0400655 cn.corename);
656 goto close_fail;
Lucas De Marchifb96c472013-04-30 15:28:06 -0700657 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400658 } else {
659 struct inode *inode;
Jann Horn378c6522016-03-22 14:25:36 -0700660 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
661 O_LARGEFILE | O_EXCL;
Alex Kelly10c28d92012-09-26 21:52:08 -0400662
663 if (cprm.limit < binfmt->min_coredump)
664 goto fail_unlock;
665
Jann Hornfbb18162015-09-09 15:38:28 -0700666 if (need_suid_safe && cn.corename[0] != '/') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400667 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
668 "to fully qualified path!\n",
669 task_tgid_vnr(current), current->comm);
670 printk(KERN_WARNING "Skipping core dump\n");
671 goto fail_unlock;
672 }
673
Jann Hornfbb18162015-09-09 15:38:28 -0700674 /*
675 * Unlink the file if it exists unless this is a SUID
676 * binary - in that case, we're running around with root
677 * privs and don't want to unlink another user's coredump.
678 */
679 if (!need_suid_safe) {
680 mm_segment_t old_fs;
681
682 old_fs = get_fs();
683 set_fs(KERNEL_DS);
684 /*
685 * If it doesn't exist, that's fine. If there's some
686 * other problem, we'll catch it at the filp_open().
687 */
688 (void) sys_unlink((const char __user *)cn.corename);
689 set_fs(old_fs);
690 }
691
692 /*
693 * There is a race between unlinking and creating the
694 * file, but if that causes an EEXIST here, that's
695 * fine - another process raced with us while creating
696 * the corefile, and the other process won. To userspace,
697 * what matters is that at least one of the two processes
698 * writes its coredump successfully, not which one.
699 */
Jann Horn378c6522016-03-22 14:25:36 -0700700 if (need_suid_safe) {
701 /*
702 * Using user namespaces, normal user tasks can change
703 * their current->fs->root to point to arbitrary
704 * directories. Since the intention of the "only dump
705 * with a fully qualified path" rule is to control where
706 * coredumps may be placed using root privileges,
707 * current->fs->root must not be used. Instead, use the
708 * root directory of init_task.
709 */
710 struct path root;
711
712 task_lock(&init_task);
713 get_fs_root(init_task.fs, &root);
714 task_unlock(&init_task);
715 cprm.file = file_open_root(root.dentry, root.mnt,
716 cn.corename, open_flags, 0600);
717 path_put(&root);
718 } else {
719 cprm.file = filp_open(cn.corename, open_flags, 0600);
720 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400721 if (IS_ERR(cprm.file))
722 goto fail_unlock;
723
Al Viro496ad9a2013-01-23 17:07:38 -0500724 inode = file_inode(cprm.file);
Alex Kelly10c28d92012-09-26 21:52:08 -0400725 if (inode->i_nlink > 1)
726 goto close_fail;
727 if (d_unhashed(cprm.file->f_path.dentry))
728 goto close_fail;
729 /*
730 * AK: actually i see no reason to not allow this for named
731 * pipes etc, but keep the previous behaviour for now.
732 */
733 if (!S_ISREG(inode->i_mode))
734 goto close_fail;
735 /*
Jann Horn40f705a2015-09-09 15:38:30 -0700736 * Don't dump core if the filesystem changed owner or mode
737 * of the file during file creation. This is an issue when
738 * a process dumps core while its cwd is e.g. on a vfat
739 * filesystem.
Alex Kelly10c28d92012-09-26 21:52:08 -0400740 */
741 if (!uid_eq(inode->i_uid, current_fsuid()))
742 goto close_fail;
Jann Horn40f705a2015-09-09 15:38:30 -0700743 if ((inode->i_mode & 0677) != 0600)
744 goto close_fail;
Al Viro86cc0582015-04-03 15:23:17 -0400745 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
Alex Kelly10c28d92012-09-26 21:52:08 -0400746 goto close_fail;
747 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
748 goto close_fail;
749 }
750
751 /* get us an unshared descriptor table; almost always a no-op */
752 retval = unshare_files(&displaced);
753 if (retval)
754 goto close_fail;
755 if (displaced)
756 put_files_struct(displaced);
Al Viroe86d35c2013-05-04 14:45:54 -0400757 if (!dump_interrupted()) {
758 file_start_write(cprm.file);
759 core_dumped = binfmt->core_dump(&cprm);
760 file_end_write(cprm.file);
761 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400762 if (ispipe && core_pipe_limit)
763 wait_for_dump_helpers(cprm.file);
764close_fail:
765 if (cprm.file)
766 filp_close(cprm.file, NULL);
767fail_dropcount:
768 if (ispipe)
769 atomic_dec(&core_dump_count);
770fail_unlock:
771 kfree(cn.corename);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700772 coredump_finish(mm, core_dumped);
Alex Kelly10c28d92012-09-26 21:52:08 -0400773 revert_creds(old_cred);
774fail_creds:
775 put_cred(cred);
776fail:
777 return;
778}
779
780/*
781 * Core dumping helper functions. These are the only things you should
782 * do on a core-file: use only these functions to write out all the
783 * necessary info.
784 */
Al Viroecc8c772013-10-05 15:32:35 -0400785int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
786{
787 struct file *file = cprm->file;
Al Viro2507a4f2013-10-08 09:11:48 -0400788 loff_t pos = file->f_pos;
789 ssize_t n;
Omar Sandoval2c4cb042016-05-11 15:16:37 -0700790 if (cprm->written + nr > cprm->limit)
Al Viroecc8c772013-10-05 15:32:35 -0400791 return 0;
Al Viro2507a4f2013-10-08 09:11:48 -0400792 while (nr) {
793 if (dump_interrupted())
794 return 0;
Al Viro52da40a2013-11-15 21:58:33 -0500795 n = __kernel_write(file, addr, nr, &pos);
Al Viro2507a4f2013-10-08 09:11:48 -0400796 if (n <= 0)
797 return 0;
798 file->f_pos = pos;
Omar Sandoval2c4cb042016-05-11 15:16:37 -0700799 cprm->written += n;
Mateusz Guzik1607f092016-06-05 23:14:14 +0200800 cprm->pos += n;
Al Viro2507a4f2013-10-08 09:11:48 -0400801 nr -= n;
802 }
Al Viroecc8c772013-10-05 15:32:35 -0400803 return 1;
804}
805EXPORT_SYMBOL(dump_emit);
806
Al Viro9b56d542013-10-08 09:26:08 -0400807int dump_skip(struct coredump_params *cprm, size_t nr)
Alex Kelly10c28d92012-09-26 21:52:08 -0400808{
Al Viro9b56d542013-10-08 09:26:08 -0400809 static char zeroes[PAGE_SIZE];
810 struct file *file = cprm->file;
Alex Kelly10c28d92012-09-26 21:52:08 -0400811 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
Oleg Nesterov528f8272013-04-30 15:28:15 -0700812 if (dump_interrupted() ||
Al Viro9b56d542013-10-08 09:26:08 -0400813 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
Alex Kelly10c28d92012-09-26 21:52:08 -0400814 return 0;
Mateusz Guzik1607f092016-06-05 23:14:14 +0200815 cprm->pos += nr;
Al Viro9b56d542013-10-08 09:26:08 -0400816 return 1;
Alex Kelly10c28d92012-09-26 21:52:08 -0400817 } else {
Al Viro9b56d542013-10-08 09:26:08 -0400818 while (nr > PAGE_SIZE) {
819 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
820 return 0;
821 nr -= PAGE_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400822 }
Al Viro9b56d542013-10-08 09:26:08 -0400823 return dump_emit(cprm, zeroes, nr);
Alex Kelly10c28d92012-09-26 21:52:08 -0400824 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400825}
Al Viro9b56d542013-10-08 09:26:08 -0400826EXPORT_SYMBOL(dump_skip);
Al Viro22a8cb82013-10-08 11:05:01 -0400827
828int dump_align(struct coredump_params *cprm, int align)
829{
Mateusz Guzik1607f092016-06-05 23:14:14 +0200830 unsigned mod = cprm->pos & (align - 1);
Al Viro22a8cb82013-10-08 11:05:01 -0400831 if (align & (align - 1))
Al Virodb512422013-11-15 21:55:52 -0500832 return 0;
833 return mod ? dump_skip(cprm, align - mod) : 1;
Al Viro22a8cb82013-10-08 11:05:01 -0400834}
835EXPORT_SYMBOL(dump_align);