blob: 281b768000e664e4d4ef9092d4bb567d003623a0 [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>
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>
Alex Kelly179899f2012-10-04 17:15:24 -070017#include <linux/coredump.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040018#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>
Jann Horn378c6522016-03-22 14:25:36 -070035#include <linux/sched.h>
36#include <linux/fs.h>
37#include <linux/path.h>
Arnd Bergmann03927c82015-11-25 16:22:25 +010038#include <linux/timekeeping.h>
Alex Kelly10c28d92012-09-26 21:52:08 -040039
40#include <asm/uaccess.h>
41#include <asm/mmu_context.h>
42#include <asm/tlb.h>
43#include <asm/exec.h>
44
45#include <trace/events/task.h>
46#include "internal.h"
Alex Kelly10c28d92012-09-26 21:52:08 -040047
48#include <trace/events/sched.h>
49
50int core_uses_pid;
Alex Kelly10c28d92012-09-26 21:52:08 -040051unsigned int core_pipe_limit;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070052char core_pattern[CORENAME_MAX_SIZE] = "core";
53static int core_name_size = CORENAME_MAX_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -040054
55struct core_name {
56 char *corename;
57 int used, size;
58};
Alex Kelly10c28d92012-09-26 21:52:08 -040059
60/* The maximal length of core_pattern is also specified in sysctl.c */
61
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070062static int expand_corename(struct core_name *cn, int size)
Alex Kelly10c28d92012-09-26 21:52:08 -040063{
Oleg Nesterove7fd1542013-07-03 15:08:16 -070064 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
Alex Kelly10c28d92012-09-26 21:52:08 -040065
Oleg Nesterove7fd1542013-07-03 15:08:16 -070066 if (!corename)
Alex Kelly10c28d92012-09-26 21:52:08 -040067 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -040068
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070069 if (size > core_name_size) /* racy but harmless */
70 core_name_size = size;
71
72 cn->size = ksize(corename);
Oleg Nesterove7fd1542013-07-03 15:08:16 -070073 cn->corename = corename;
Alex Kelly10c28d92012-09-26 21:52:08 -040074 return 0;
75}
76
Nicolas Ioossb4176b72015-06-25 15:03:53 -070077static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
78 va_list arg)
Alex Kelly10c28d92012-09-26 21:52:08 -040079{
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070080 int free, need;
Eric Dumazet404ca802014-04-19 10:15:07 -070081 va_list arg_copy;
Alex Kelly10c28d92012-09-26 21:52:08 -040082
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070083again:
84 free = cn->size - cn->used;
Eric Dumazet404ca802014-04-19 10:15:07 -070085
86 va_copy(arg_copy, arg);
87 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
88 va_end(arg_copy);
89
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070090 if (need < free) {
91 cn->used += need;
92 return 0;
93 }
Alex Kelly10c28d92012-09-26 21:52:08 -040094
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070095 if (!expand_corename(cn, cn->size + need - free + 1))
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070096 goto again;
Alex Kelly10c28d92012-09-26 21:52:08 -040097
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070098 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -040099}
100
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700101static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
Oleg Nesterovbc03c692013-07-03 15:08:17 -0700102{
103 va_list arg;
104 int ret;
105
106 va_start(arg, fmt);
107 ret = cn_vprintf(cn, fmt, arg);
108 va_end(arg);
109
110 return ret;
111}
112
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700113static __printf(2, 3)
114int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
Alex Kelly10c28d92012-09-26 21:52:08 -0400115{
Oleg Nesterov923bed032013-07-03 15:08:20 -0700116 int cur = cn->used;
117 va_list arg;
118 int ret;
119
120 va_start(arg, fmt);
121 ret = cn_vprintf(cn, fmt, arg);
122 va_end(arg);
123
Jann Hornac94b6e2016-01-20 15:00:08 -0800124 if (ret == 0) {
125 /*
126 * Ensure that this coredump name component can't cause the
127 * resulting corefile path to consist of a ".." or ".".
128 */
129 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
130 (cn->used - cur == 2 && cn->corename[cur] == '.'
131 && cn->corename[cur+1] == '.'))
132 cn->corename[cur] = '!';
133
134 /*
135 * Empty names are fishy and could be used to create a "//" in a
136 * corefile name, causing the coredump to happen one directory
137 * level too high. Enforce that all components of the core
138 * pattern are at least one character long.
139 */
140 if (cn->used == cur)
141 ret = cn_printf(cn, "!");
142 }
143
Oleg Nesterov923bed032013-07-03 15:08:20 -0700144 for (; cur < cn->used; ++cur) {
145 if (cn->corename[cur] == '/')
146 cn->corename[cur] = '!';
147 }
148 return ret;
Alex Kelly10c28d92012-09-26 21:52:08 -0400149}
150
151static int cn_print_exe_file(struct core_name *cn)
152{
153 struct file *exe_file;
154 char *pathbuf, *path;
155 int ret;
156
157 exe_file = get_mm_exe_file(current->mm);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700158 if (!exe_file)
159 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400160
161 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
162 if (!pathbuf) {
163 ret = -ENOMEM;
164 goto put_exe_file;
165 }
166
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200167 path = file_path(exe_file, pathbuf, PATH_MAX);
Alex Kelly10c28d92012-09-26 21:52:08 -0400168 if (IS_ERR(path)) {
169 ret = PTR_ERR(path);
170 goto free_buf;
171 }
172
Oleg Nesterov923bed032013-07-03 15:08:20 -0700173 ret = cn_esc_printf(cn, "%s", path);
Alex Kelly10c28d92012-09-26 21:52:08 -0400174
175free_buf:
176 kfree(pathbuf);
177put_exe_file:
178 fput(exe_file);
179 return ret;
180}
181
182/* format_corename will inspect the pattern parameter, and output a
183 * name into corename, which must have space for at least
184 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
185 */
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700186static int format_corename(struct core_name *cn, struct coredump_params *cprm)
Alex Kelly10c28d92012-09-26 21:52:08 -0400187{
188 const struct cred *cred = current_cred();
189 const char *pat_ptr = core_pattern;
190 int ispipe = (*pat_ptr == '|');
191 int pid_in_pattern = 0;
192 int err = 0;
193
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700194 cn->used = 0;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -0700195 cn->corename = NULL;
196 if (expand_corename(cn, core_name_size))
Alex Kelly10c28d92012-09-26 21:52:08 -0400197 return -ENOMEM;
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700198 cn->corename[0] = '\0';
199
200 if (ispipe)
201 ++pat_ptr;
Alex Kelly10c28d92012-09-26 21:52:08 -0400202
203 /* Repeat as long as we have more pattern to process and more output
204 space */
205 while (*pat_ptr) {
206 if (*pat_ptr != '%') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400207 err = cn_printf(cn, "%c", *pat_ptr++);
208 } else {
209 switch (*++pat_ptr) {
210 /* single % at the end, drop that */
211 case 0:
212 goto out;
213 /* Double percent, output one percent */
214 case '%':
215 err = cn_printf(cn, "%c", '%');
216 break;
217 /* pid */
218 case 'p':
219 pid_in_pattern = 1;
220 err = cn_printf(cn, "%d",
221 task_tgid_vnr(current));
222 break;
Stéphane Graber65aafb12013-09-11 14:24:32 -0700223 /* global pid */
224 case 'P':
225 err = cn_printf(cn, "%d",
226 task_tgid_nr(current));
227 break;
Oleg Nesterovb03023e2014-10-13 15:53:35 -0700228 case 'i':
229 err = cn_printf(cn, "%d",
230 task_pid_vnr(current));
231 break;
232 case 'I':
233 err = cn_printf(cn, "%d",
234 task_pid_nr(current));
235 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400236 /* uid */
237 case 'u':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700238 err = cn_printf(cn, "%u",
239 from_kuid(&init_user_ns,
240 cred->uid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400241 break;
242 /* gid */
243 case 'g':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700244 err = cn_printf(cn, "%u",
245 from_kgid(&init_user_ns,
246 cred->gid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400247 break;
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700248 case 'd':
249 err = cn_printf(cn, "%d",
250 __get_dumpable(cprm->mm_flags));
251 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400252 /* signal that caused the coredump */
253 case 's':
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700254 err = cn_printf(cn, "%d",
255 cprm->siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400256 break;
257 /* UNIX time of coredump */
258 case 't': {
Arnd Bergmann03927c82015-11-25 16:22:25 +0100259 time64_t time;
260
261 time = ktime_get_real_seconds();
262 err = cn_printf(cn, "%lld", time);
Alex Kelly10c28d92012-09-26 21:52:08 -0400263 break;
264 }
265 /* hostname */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700266 case 'h':
Alex Kelly10c28d92012-09-26 21:52:08 -0400267 down_read(&uts_sem);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700268 err = cn_esc_printf(cn, "%s",
Alex Kelly10c28d92012-09-26 21:52:08 -0400269 utsname()->nodename);
270 up_read(&uts_sem);
Alex Kelly10c28d92012-09-26 21:52:08 -0400271 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400272 /* executable */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700273 case 'e':
274 err = cn_esc_printf(cn, "%s", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400275 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400276 case 'E':
277 err = cn_print_exe_file(cn);
278 break;
279 /* core limit size */
280 case 'c':
281 err = cn_printf(cn, "%lu",
282 rlimit(RLIMIT_CORE));
283 break;
284 default:
285 break;
286 }
287 ++pat_ptr;
288 }
289
290 if (err)
291 return err;
292 }
293
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700294out:
Alex Kelly10c28d92012-09-26 21:52:08 -0400295 /* Backward compatibility with core_uses_pid:
296 *
297 * If core_pattern does not include a %p (as is the default)
298 * and core_uses_pid is set, then .%pid will be appended to
299 * the filename. Do not do this for piped commands. */
300 if (!ispipe && !pid_in_pattern && core_uses_pid) {
301 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
302 if (err)
303 return err;
304 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400305 return ispipe;
306}
307
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800308static int zap_process(struct task_struct *start, int exit_code, int flags)
Alex Kelly10c28d92012-09-26 21:52:08 -0400309{
310 struct task_struct *t;
311 int nr = 0;
312
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800313 /* ignore all signals except SIGKILL, see prepare_signal() */
314 start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
Alex Kelly10c28d92012-09-26 21:52:08 -0400315 start->signal->group_exit_code = exit_code;
316 start->signal->group_stop_count = 0;
317
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800318 for_each_thread(start, t) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400319 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
320 if (t != current && t->mm) {
321 sigaddset(&t->pending.signal, SIGKILL);
322 signal_wake_up(t, 1);
323 nr++;
324 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800325 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400326
327 return nr;
328}
329
Oleg Nesterov403bad72013-04-30 15:28:10 -0700330static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
331 struct core_state *core_state, int exit_code)
Alex Kelly10c28d92012-09-26 21:52:08 -0400332{
333 struct task_struct *g, *p;
334 unsigned long flags;
335 int nr = -EAGAIN;
336
337 spin_lock_irq(&tsk->sighand->siglock);
338 if (!signal_group_exit(tsk->signal)) {
339 mm->core_state = core_state;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700340 tsk->signal->group_exit_task = tsk;
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800341 nr = zap_process(tsk, exit_code, 0);
Oleg Nesterov403bad72013-04-30 15:28:10 -0700342 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
Alex Kelly10c28d92012-09-26 21:52:08 -0400343 }
344 spin_unlock_irq(&tsk->sighand->siglock);
345 if (unlikely(nr < 0))
346 return nr;
347
Silesh C Vaed8adb2014-07-23 13:59:59 -0700348 tsk->flags |= PF_DUMPCORE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400349 if (atomic_read(&mm->mm_users) == nr + 1)
350 goto done;
351 /*
352 * We should find and kill all tasks which use this mm, and we should
353 * count them correctly into ->nr_threads. We don't take tasklist
354 * lock, but this is safe wrt:
355 *
356 * fork:
357 * None of sub-threads can fork after zap_process(leader). All
358 * processes which were created before this point should be
359 * visible to zap_threads() because copy_process() adds the new
360 * process to the tail of init_task.tasks list, and lock/unlock
361 * of ->siglock provides a memory barrier.
362 *
363 * do_exit:
364 * The caller holds mm->mmap_sem. This means that the task which
365 * uses this mm can't pass exit_mm(), so it can't exit or clear
366 * its ->mm.
367 *
368 * de_thread:
369 * It does list_replace_rcu(&leader->tasks, &current->tasks),
370 * we must see either old or new leader, this does not matter.
371 * However, it can change p->sighand, so lock_task_sighand(p)
372 * must be used. Since p->mm != NULL and we hold ->mmap_sem
373 * it can't fail.
374 *
375 * Note also that "g" can be the old leader with ->mm == NULL
376 * and already unhashed and thus removed from ->thread_group.
377 * This is OK, __unhash_process()->list_del_rcu() does not
378 * clear the ->next pointer, we will find the new leader via
379 * next_thread().
380 */
381 rcu_read_lock();
382 for_each_process(g) {
383 if (g == tsk->group_leader)
384 continue;
385 if (g->flags & PF_KTHREAD)
386 continue;
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800387
388 for_each_thread(g, p) {
389 if (unlikely(!p->mm))
390 continue;
391 if (unlikely(p->mm == mm)) {
392 lock_task_sighand(p, &flags);
393 nr += zap_process(p, exit_code,
394 SIGNAL_GROUP_EXIT);
395 unlock_task_sighand(p, &flags);
Alex Kelly10c28d92012-09-26 21:52:08 -0400396 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800397 break;
398 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400399 }
400 rcu_read_unlock();
401done:
402 atomic_set(&core_state->nr_threads, nr);
403 return nr;
404}
405
406static int coredump_wait(int exit_code, struct core_state *core_state)
407{
408 struct task_struct *tsk = current;
409 struct mm_struct *mm = tsk->mm;
410 int core_waiters = -EBUSY;
411
412 init_completion(&core_state->startup);
413 core_state->dumper.task = tsk;
414 core_state->dumper.next = NULL;
415
Michal Hocko4136c262016-05-23 16:25:57 -0700416 if (down_write_killable(&mm->mmap_sem))
417 return -EINTR;
418
Alex Kelly10c28d92012-09-26 21:52:08 -0400419 if (!mm->core_state)
420 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
421 up_write(&mm->mmap_sem);
422
423 if (core_waiters > 0) {
424 struct core_thread *ptr;
425
426 wait_for_completion(&core_state->startup);
427 /*
428 * Wait for all the threads to become inactive, so that
429 * all the thread context (extended register state, like
430 * fpu etc) gets copied to the memory.
431 */
432 ptr = core_state->dumper.next;
433 while (ptr != NULL) {
434 wait_task_inactive(ptr->task, 0);
435 ptr = ptr->next;
436 }
437 }
438
439 return core_waiters;
440}
441
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700442static void coredump_finish(struct mm_struct *mm, bool core_dumped)
Alex Kelly10c28d92012-09-26 21:52:08 -0400443{
444 struct core_thread *curr, *next;
445 struct task_struct *task;
446
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700447 spin_lock_irq(&current->sighand->siglock);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700448 if (core_dumped && !__fatal_signal_pending(current))
449 current->signal->group_exit_code |= 0x80;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700450 current->signal->group_exit_task = NULL;
451 current->signal->flags = SIGNAL_GROUP_EXIT;
452 spin_unlock_irq(&current->sighand->siglock);
453
Alex Kelly10c28d92012-09-26 21:52:08 -0400454 next = mm->core_state->dumper.next;
455 while ((curr = next) != NULL) {
456 next = curr->next;
457 task = curr->task;
458 /*
459 * see exit_mm(), curr->task must not see
460 * ->task == NULL before we read ->next.
461 */
462 smp_mb();
463 curr->task = NULL;
464 wake_up_process(task);
465 }
466
467 mm->core_state = NULL;
468}
469
Oleg Nesterov528f8272013-04-30 15:28:15 -0700470static bool dump_interrupted(void)
471{
472 /*
473 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
474 * can do try_to_freeze() and check __fatal_signal_pending(),
475 * but then we need to teach dump_write() to restart and clear
476 * TIF_SIGPENDING.
477 */
478 return signal_pending(current);
479}
480
Alex Kelly10c28d92012-09-26 21:52:08 -0400481static void wait_for_dump_helpers(struct file *file)
482{
Al Virode32ec42013-03-21 11:16:56 -0400483 struct pipe_inode_info *pipe = file->private_data;
Alex Kelly10c28d92012-09-26 21:52:08 -0400484
485 pipe_lock(pipe);
486 pipe->readers++;
487 pipe->writers--;
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700488 wake_up_interruptible_sync(&pipe->wait);
489 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
490 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400491
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700492 /*
493 * We actually want wait_event_freezable() but then we need
494 * to clear TIF_SIGPENDING and improve dump_interrupted().
495 */
496 wait_event_interruptible(pipe->wait, pipe->readers == 1);
Alex Kelly10c28d92012-09-26 21:52:08 -0400497
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700498 pipe_lock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400499 pipe->readers--;
500 pipe->writers++;
501 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400502}
503
504/*
505 * umh_pipe_setup
506 * helper function to customize the process used
507 * to collect the core in userspace. Specifically
508 * it sets up a pipe and installs it as fd 0 (stdin)
509 * for the process. Returns 0 on success, or
510 * PTR_ERR on failure.
511 * Note that it also sets the core limit to 1. This
512 * is a special value that we use to trap recursive
513 * core dumps
514 */
515static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
516{
517 struct file *files[2];
518 struct coredump_params *cp = (struct coredump_params *)info->data;
519 int err = create_pipe_files(files, 0);
520 if (err)
521 return err;
522
523 cp->file = files[1];
524
Al Viro45525b22012-10-16 13:30:07 -0400525 err = replace_fd(0, files[0], 0);
526 fput(files[0]);
Alex Kelly10c28d92012-09-26 21:52:08 -0400527 /* and disallow core files too */
528 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
529
Al Viro45525b22012-10-16 13:30:07 -0400530 return err;
Alex Kelly10c28d92012-09-26 21:52:08 -0400531}
532
Al Viroec579412013-10-13 17:57:29 -0400533void do_coredump(const siginfo_t *siginfo)
Alex Kelly10c28d92012-09-26 21:52:08 -0400534{
535 struct core_state core_state;
536 struct core_name cn;
537 struct mm_struct *mm = current->mm;
538 struct linux_binfmt * binfmt;
539 const struct cred *old_cred;
540 struct cred *cred;
541 int retval = 0;
Alex Kelly10c28d92012-09-26 21:52:08 -0400542 int ispipe;
543 struct files_struct *displaced;
Jann Hornfbb18162015-09-09 15:38:28 -0700544 /* require nonrelative corefile path and be extra careful */
545 bool need_suid_safe = false;
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700546 bool core_dumped = false;
Alex Kelly10c28d92012-09-26 21:52:08 -0400547 static atomic_t core_dump_count = ATOMIC_INIT(0);
548 struct coredump_params cprm = {
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700549 .siginfo = siginfo,
Al Viro541880d2012-11-05 13:11:26 -0500550 .regs = signal_pt_regs(),
Alex Kelly10c28d92012-09-26 21:52:08 -0400551 .limit = rlimit(RLIMIT_CORE),
552 /*
553 * We must use the same mm->flags while dumping core to avoid
554 * inconsistency of bit flags, since this flag is not protected
555 * by any locks.
556 */
557 .mm_flags = mm->flags,
558 };
559
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700560 audit_core_dumps(siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400561
562 binfmt = mm->binfmt;
563 if (!binfmt || !binfmt->core_dump)
564 goto fail;
565 if (!__get_dumpable(cprm.mm_flags))
566 goto fail;
567
568 cred = prepare_creds();
569 if (!cred)
570 goto fail;
571 /*
572 * We cannot trust fsuid as being the "true" uid of the process
573 * nor do we know its entire history. We only know it was tainted
574 * so we dump it as root in mode 2, and only into a controlled
575 * environment (pipe handler or fully qualified path).
576 */
Kees Cooke579d2c2013-02-27 17:03:15 -0800577 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400578 /* Setuid core dump mode */
Alex Kelly10c28d92012-09-26 21:52:08 -0400579 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
Jann Hornfbb18162015-09-09 15:38:28 -0700580 need_suid_safe = true;
Alex Kelly10c28d92012-09-26 21:52:08 -0400581 }
582
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700583 retval = coredump_wait(siginfo->si_signo, &core_state);
Alex Kelly10c28d92012-09-26 21:52:08 -0400584 if (retval < 0)
585 goto fail_creds;
586
587 old_cred = override_creds(cred);
588
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700589 ispipe = format_corename(&cn, &cprm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400590
Lucas De Marchifb96c472013-04-30 15:28:06 -0700591 if (ispipe) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400592 int dump_count;
593 char **helper_argv;
Lucas De Marchi907ed132013-04-30 15:28:07 -0700594 struct subprocess_info *sub_info;
Alex Kelly10c28d92012-09-26 21:52:08 -0400595
596 if (ispipe < 0) {
597 printk(KERN_WARNING "format_corename failed\n");
598 printk(KERN_WARNING "Aborting core\n");
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700599 goto fail_unlock;
Alex Kelly10c28d92012-09-26 21:52:08 -0400600 }
601
602 if (cprm.limit == 1) {
603 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
604 *
605 * Normally core limits are irrelevant to pipes, since
606 * we're not writing to the file system, but we use
Bastien Nocerafcbc32b2015-02-05 14:35:05 +0100607 * cprm.limit of 1 here as a special value, this is a
Alex Kelly10c28d92012-09-26 21:52:08 -0400608 * consistent way to catch recursive crashes.
609 * We can still crash if the core_pattern binary sets
610 * RLIM_CORE = !1, but it runs as root, and can do
611 * lots of stupid things.
612 *
613 * Note that we use task_tgid_vnr here to grab the pid
614 * of the process group leader. That way we get the
615 * right pid if a thread in a multi-threaded
616 * core_pattern process dies.
617 */
618 printk(KERN_WARNING
619 "Process %d(%s) has RLIMIT_CORE set to 1\n",
620 task_tgid_vnr(current), current->comm);
621 printk(KERN_WARNING "Aborting core\n");
622 goto fail_unlock;
623 }
624 cprm.limit = RLIM_INFINITY;
625
626 dump_count = atomic_inc_return(&core_dump_count);
627 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
628 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
629 task_tgid_vnr(current), current->comm);
630 printk(KERN_WARNING "Skipping core dump\n");
631 goto fail_dropcount;
632 }
633
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700634 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
Alex Kelly10c28d92012-09-26 21:52:08 -0400635 if (!helper_argv) {
636 printk(KERN_WARNING "%s failed to allocate memory\n",
637 __func__);
638 goto fail_dropcount;
639 }
640
Lucas De Marchi907ed132013-04-30 15:28:07 -0700641 retval = -ENOMEM;
642 sub_info = call_usermodehelper_setup(helper_argv[0],
643 helper_argv, NULL, GFP_KERNEL,
644 umh_pipe_setup, NULL, &cprm);
645 if (sub_info)
646 retval = call_usermodehelper_exec(sub_info,
647 UMH_WAIT_EXEC);
648
Alex Kelly10c28d92012-09-26 21:52:08 -0400649 argv_free(helper_argv);
650 if (retval) {
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700651 printk(KERN_INFO "Core dump to |%s pipe failed\n",
Alex Kelly10c28d92012-09-26 21:52:08 -0400652 cn.corename);
653 goto close_fail;
Lucas De Marchifb96c472013-04-30 15:28:06 -0700654 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400655 } else {
656 struct inode *inode;
Jann Horn378c6522016-03-22 14:25:36 -0700657 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
658 O_LARGEFILE | O_EXCL;
Alex Kelly10c28d92012-09-26 21:52:08 -0400659
660 if (cprm.limit < binfmt->min_coredump)
661 goto fail_unlock;
662
Jann Hornfbb18162015-09-09 15:38:28 -0700663 if (need_suid_safe && cn.corename[0] != '/') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400664 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
665 "to fully qualified path!\n",
666 task_tgid_vnr(current), current->comm);
667 printk(KERN_WARNING "Skipping core dump\n");
668 goto fail_unlock;
669 }
670
Jann Hornfbb18162015-09-09 15:38:28 -0700671 /*
672 * Unlink the file if it exists unless this is a SUID
673 * binary - in that case, we're running around with root
674 * privs and don't want to unlink another user's coredump.
675 */
676 if (!need_suid_safe) {
677 mm_segment_t old_fs;
678
679 old_fs = get_fs();
680 set_fs(KERNEL_DS);
681 /*
682 * If it doesn't exist, that's fine. If there's some
683 * other problem, we'll catch it at the filp_open().
684 */
685 (void) sys_unlink((const char __user *)cn.corename);
686 set_fs(old_fs);
687 }
688
689 /*
690 * There is a race between unlinking and creating the
691 * file, but if that causes an EEXIST here, that's
692 * fine - another process raced with us while creating
693 * the corefile, and the other process won. To userspace,
694 * what matters is that at least one of the two processes
695 * writes its coredump successfully, not which one.
696 */
Jann Horn378c6522016-03-22 14:25:36 -0700697 if (need_suid_safe) {
698 /*
699 * Using user namespaces, normal user tasks can change
700 * their current->fs->root to point to arbitrary
701 * directories. Since the intention of the "only dump
702 * with a fully qualified path" rule is to control where
703 * coredumps may be placed using root privileges,
704 * current->fs->root must not be used. Instead, use the
705 * root directory of init_task.
706 */
707 struct path root;
708
709 task_lock(&init_task);
710 get_fs_root(init_task.fs, &root);
711 task_unlock(&init_task);
712 cprm.file = file_open_root(root.dentry, root.mnt,
713 cn.corename, open_flags, 0600);
714 path_put(&root);
715 } else {
716 cprm.file = filp_open(cn.corename, open_flags, 0600);
717 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400718 if (IS_ERR(cprm.file))
719 goto fail_unlock;
720
Al Viro496ad9a2013-01-23 17:07:38 -0500721 inode = file_inode(cprm.file);
Alex Kelly10c28d92012-09-26 21:52:08 -0400722 if (inode->i_nlink > 1)
723 goto close_fail;
724 if (d_unhashed(cprm.file->f_path.dentry))
725 goto close_fail;
726 /*
727 * AK: actually i see no reason to not allow this for named
728 * pipes etc, but keep the previous behaviour for now.
729 */
730 if (!S_ISREG(inode->i_mode))
731 goto close_fail;
732 /*
Jann Horn40f705a2015-09-09 15:38:30 -0700733 * Don't dump core if the filesystem changed owner or mode
734 * of the file during file creation. This is an issue when
735 * a process dumps core while its cwd is e.g. on a vfat
736 * filesystem.
Alex Kelly10c28d92012-09-26 21:52:08 -0400737 */
738 if (!uid_eq(inode->i_uid, current_fsuid()))
739 goto close_fail;
Jann Horn40f705a2015-09-09 15:38:30 -0700740 if ((inode->i_mode & 0677) != 0600)
741 goto close_fail;
Al Viro86cc0582015-04-03 15:23:17 -0400742 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
Alex Kelly10c28d92012-09-26 21:52:08 -0400743 goto close_fail;
744 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
745 goto close_fail;
746 }
747
748 /* get us an unshared descriptor table; almost always a no-op */
749 retval = unshare_files(&displaced);
750 if (retval)
751 goto close_fail;
752 if (displaced)
753 put_files_struct(displaced);
Al Viroe86d35c2013-05-04 14:45:54 -0400754 if (!dump_interrupted()) {
755 file_start_write(cprm.file);
756 core_dumped = binfmt->core_dump(&cprm);
757 file_end_write(cprm.file);
758 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400759 if (ispipe && core_pipe_limit)
760 wait_for_dump_helpers(cprm.file);
761close_fail:
762 if (cprm.file)
763 filp_close(cprm.file, NULL);
764fail_dropcount:
765 if (ispipe)
766 atomic_dec(&core_dump_count);
767fail_unlock:
768 kfree(cn.corename);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700769 coredump_finish(mm, core_dumped);
Alex Kelly10c28d92012-09-26 21:52:08 -0400770 revert_creds(old_cred);
771fail_creds:
772 put_cred(cred);
773fail:
774 return;
775}
776
777/*
778 * Core dumping helper functions. These are the only things you should
779 * do on a core-file: use only these functions to write out all the
780 * necessary info.
781 */
Al Viroecc8c772013-10-05 15:32:35 -0400782int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
783{
784 struct file *file = cprm->file;
Al Viro2507a4f2013-10-08 09:11:48 -0400785 loff_t pos = file->f_pos;
786 ssize_t n;
Omar Sandoval2c4cb042016-05-11 15:16:37 -0700787 if (cprm->written + nr > cprm->limit)
Al Viroecc8c772013-10-05 15:32:35 -0400788 return 0;
Al Viro2507a4f2013-10-08 09:11:48 -0400789 while (nr) {
790 if (dump_interrupted())
791 return 0;
Al Viro52da40a2013-11-15 21:58:33 -0500792 n = __kernel_write(file, addr, nr, &pos);
Al Viro2507a4f2013-10-08 09:11:48 -0400793 if (n <= 0)
794 return 0;
795 file->f_pos = pos;
Omar Sandoval2c4cb042016-05-11 15:16:37 -0700796 cprm->written += n;
Mateusz Guzik1607f092016-06-05 23:14:14 +0200797 cprm->pos += n;
Al Viro2507a4f2013-10-08 09:11:48 -0400798 nr -= n;
799 }
Al Viroecc8c772013-10-05 15:32:35 -0400800 return 1;
801}
802EXPORT_SYMBOL(dump_emit);
803
Al Viro9b56d542013-10-08 09:26:08 -0400804int dump_skip(struct coredump_params *cprm, size_t nr)
Alex Kelly10c28d92012-09-26 21:52:08 -0400805{
Al Viro9b56d542013-10-08 09:26:08 -0400806 static char zeroes[PAGE_SIZE];
807 struct file *file = cprm->file;
Alex Kelly10c28d92012-09-26 21:52:08 -0400808 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
Oleg Nesterov528f8272013-04-30 15:28:15 -0700809 if (dump_interrupted() ||
Al Viro9b56d542013-10-08 09:26:08 -0400810 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
Alex Kelly10c28d92012-09-26 21:52:08 -0400811 return 0;
Mateusz Guzik1607f092016-06-05 23:14:14 +0200812 cprm->pos += nr;
Al Viro9b56d542013-10-08 09:26:08 -0400813 return 1;
Alex Kelly10c28d92012-09-26 21:52:08 -0400814 } else {
Al Viro9b56d542013-10-08 09:26:08 -0400815 while (nr > PAGE_SIZE) {
816 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
817 return 0;
818 nr -= PAGE_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400819 }
Al Viro9b56d542013-10-08 09:26:08 -0400820 return dump_emit(cprm, zeroes, nr);
Alex Kelly10c28d92012-09-26 21:52:08 -0400821 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400822}
Al Viro9b56d542013-10-08 09:26:08 -0400823EXPORT_SYMBOL(dump_skip);
Al Viro22a8cb82013-10-08 11:05:01 -0400824
825int dump_align(struct coredump_params *cprm, int align)
826{
Mateusz Guzik1607f092016-06-05 23:14:14 +0200827 unsigned mod = cprm->pos & (align - 1);
Al Viro22a8cb82013-10-08 11:05:01 -0400828 if (align & (align - 1))
Al Virodb512422013-11-15 21:55:52 -0500829 return 0;
830 return mod ? dump_skip(cprm, align - mod) : 1;
Al Viro22a8cb82013-10-08 11:05:01 -0400831}
832EXPORT_SYMBOL(dump_align);