blob: 1777331eee767fa323cb864fb95131983eaad588 [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>
35
36#include <asm/uaccess.h>
37#include <asm/mmu_context.h>
38#include <asm/tlb.h>
39#include <asm/exec.h>
40
41#include <trace/events/task.h>
42#include "internal.h"
Alex Kelly10c28d92012-09-26 21:52:08 -040043
44#include <trace/events/sched.h>
45
46int core_uses_pid;
Alex Kelly10c28d92012-09-26 21:52:08 -040047unsigned int core_pipe_limit;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070048char core_pattern[CORENAME_MAX_SIZE] = "core";
49static int core_name_size = CORENAME_MAX_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -040050
51struct core_name {
52 char *corename;
53 int used, size;
54};
Alex Kelly10c28d92012-09-26 21:52:08 -040055
56/* The maximal length of core_pattern is also specified in sysctl.c */
57
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070058static int expand_corename(struct core_name *cn, int size)
Alex Kelly10c28d92012-09-26 21:52:08 -040059{
Oleg Nesterove7fd1542013-07-03 15:08:16 -070060 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
Alex Kelly10c28d92012-09-26 21:52:08 -040061
Oleg Nesterove7fd1542013-07-03 15:08:16 -070062 if (!corename)
Alex Kelly10c28d92012-09-26 21:52:08 -040063 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -040064
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070065 if (size > core_name_size) /* racy but harmless */
66 core_name_size = size;
67
68 cn->size = ksize(corename);
Oleg Nesterove7fd1542013-07-03 15:08:16 -070069 cn->corename = corename;
Alex Kelly10c28d92012-09-26 21:52:08 -040070 return 0;
71}
72
Nicolas Ioossb4176b72015-06-25 15:03:53 -070073static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
74 va_list arg)
Alex Kelly10c28d92012-09-26 21:52:08 -040075{
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070076 int free, need;
Eric Dumazet404ca802014-04-19 10:15:07 -070077 va_list arg_copy;
Alex Kelly10c28d92012-09-26 21:52:08 -040078
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070079again:
80 free = cn->size - cn->used;
Eric Dumazet404ca802014-04-19 10:15:07 -070081
82 va_copy(arg_copy, arg);
83 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
84 va_end(arg_copy);
85
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070086 if (need < free) {
87 cn->used += need;
88 return 0;
89 }
Alex Kelly10c28d92012-09-26 21:52:08 -040090
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -070091 if (!expand_corename(cn, cn->size + need - free + 1))
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070092 goto again;
Alex Kelly10c28d92012-09-26 21:52:08 -040093
Oleg Nesterov5fe9d8c2013-07-03 15:08:19 -070094 return -ENOMEM;
Alex Kelly10c28d92012-09-26 21:52:08 -040095}
96
Nicolas Ioossb4176b72015-06-25 15:03:53 -070097static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
Oleg Nesterovbc03c692013-07-03 15:08:17 -070098{
99 va_list arg;
100 int ret;
101
102 va_start(arg, fmt);
103 ret = cn_vprintf(cn, fmt, arg);
104 va_end(arg);
105
106 return ret;
107}
108
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700109static __printf(2, 3)
110int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
Alex Kelly10c28d92012-09-26 21:52:08 -0400111{
Oleg Nesterov923bed032013-07-03 15:08:20 -0700112 int cur = cn->used;
113 va_list arg;
114 int ret;
115
116 va_start(arg, fmt);
117 ret = cn_vprintf(cn, fmt, arg);
118 va_end(arg);
119
120 for (; cur < cn->used; ++cur) {
121 if (cn->corename[cur] == '/')
122 cn->corename[cur] = '!';
123 }
124 return ret;
Alex Kelly10c28d92012-09-26 21:52:08 -0400125}
126
127static int cn_print_exe_file(struct core_name *cn)
128{
129 struct file *exe_file;
130 char *pathbuf, *path;
131 int ret;
132
133 exe_file = get_mm_exe_file(current->mm);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700134 if (!exe_file)
135 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400136
137 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
138 if (!pathbuf) {
139 ret = -ENOMEM;
140 goto put_exe_file;
141 }
142
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200143 path = file_path(exe_file, pathbuf, PATH_MAX);
Alex Kelly10c28d92012-09-26 21:52:08 -0400144 if (IS_ERR(path)) {
145 ret = PTR_ERR(path);
146 goto free_buf;
147 }
148
Oleg Nesterov923bed032013-07-03 15:08:20 -0700149 ret = cn_esc_printf(cn, "%s", path);
Alex Kelly10c28d92012-09-26 21:52:08 -0400150
151free_buf:
152 kfree(pathbuf);
153put_exe_file:
154 fput(exe_file);
155 return ret;
156}
157
158/* format_corename will inspect the pattern parameter, and output a
159 * name into corename, which must have space for at least
160 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
161 */
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700162static int format_corename(struct core_name *cn, struct coredump_params *cprm)
Alex Kelly10c28d92012-09-26 21:52:08 -0400163{
164 const struct cred *cred = current_cred();
165 const char *pat_ptr = core_pattern;
166 int ispipe = (*pat_ptr == '|');
167 int pid_in_pattern = 0;
168 int err = 0;
169
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700170 cn->used = 0;
Oleg Nesterov3ceadcf2013-07-03 15:08:22 -0700171 cn->corename = NULL;
172 if (expand_corename(cn, core_name_size))
Alex Kelly10c28d92012-09-26 21:52:08 -0400173 return -ENOMEM;
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700174 cn->corename[0] = '\0';
175
176 if (ispipe)
177 ++pat_ptr;
Alex Kelly10c28d92012-09-26 21:52:08 -0400178
179 /* Repeat as long as we have more pattern to process and more output
180 space */
181 while (*pat_ptr) {
182 if (*pat_ptr != '%') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400183 err = cn_printf(cn, "%c", *pat_ptr++);
184 } else {
185 switch (*++pat_ptr) {
186 /* single % at the end, drop that */
187 case 0:
188 goto out;
189 /* Double percent, output one percent */
190 case '%':
191 err = cn_printf(cn, "%c", '%');
192 break;
193 /* pid */
194 case 'p':
195 pid_in_pattern = 1;
196 err = cn_printf(cn, "%d",
197 task_tgid_vnr(current));
198 break;
Stéphane Graber65aafb12013-09-11 14:24:32 -0700199 /* global pid */
200 case 'P':
201 err = cn_printf(cn, "%d",
202 task_tgid_nr(current));
203 break;
Oleg Nesterovb03023e2014-10-13 15:53:35 -0700204 case 'i':
205 err = cn_printf(cn, "%d",
206 task_pid_vnr(current));
207 break;
208 case 'I':
209 err = cn_printf(cn, "%d",
210 task_pid_nr(current));
211 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400212 /* uid */
213 case 'u':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700214 err = cn_printf(cn, "%u",
215 from_kuid(&init_user_ns,
216 cred->uid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400217 break;
218 /* gid */
219 case 'g':
Nicolas Iooss5202efe2015-06-25 15:03:51 -0700220 err = cn_printf(cn, "%u",
221 from_kgid(&init_user_ns,
222 cred->gid));
Alex Kelly10c28d92012-09-26 21:52:08 -0400223 break;
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700224 case 'd':
225 err = cn_printf(cn, "%d",
226 __get_dumpable(cprm->mm_flags));
227 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400228 /* signal that caused the coredump */
229 case 's':
Nicolas Ioossb4176b72015-06-25 15:03:53 -0700230 err = cn_printf(cn, "%d",
231 cprm->siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400232 break;
233 /* UNIX time of coredump */
234 case 't': {
235 struct timeval tv;
236 do_gettimeofday(&tv);
237 err = cn_printf(cn, "%lu", tv.tv_sec);
238 break;
239 }
240 /* hostname */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700241 case 'h':
Alex Kelly10c28d92012-09-26 21:52:08 -0400242 down_read(&uts_sem);
Oleg Nesterov923bed032013-07-03 15:08:20 -0700243 err = cn_esc_printf(cn, "%s",
Alex Kelly10c28d92012-09-26 21:52:08 -0400244 utsname()->nodename);
245 up_read(&uts_sem);
Alex Kelly10c28d92012-09-26 21:52:08 -0400246 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400247 /* executable */
Oleg Nesterov923bed032013-07-03 15:08:20 -0700248 case 'e':
249 err = cn_esc_printf(cn, "%s", current->comm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400250 break;
Alex Kelly10c28d92012-09-26 21:52:08 -0400251 case 'E':
252 err = cn_print_exe_file(cn);
253 break;
254 /* core limit size */
255 case 'c':
256 err = cn_printf(cn, "%lu",
257 rlimit(RLIMIT_CORE));
258 break;
259 default:
260 break;
261 }
262 ++pat_ptr;
263 }
264
265 if (err)
266 return err;
267 }
268
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700269out:
Alex Kelly10c28d92012-09-26 21:52:08 -0400270 /* Backward compatibility with core_uses_pid:
271 *
272 * If core_pattern does not include a %p (as is the default)
273 * and core_uses_pid is set, then .%pid will be appended to
274 * the filename. Do not do this for piped commands. */
275 if (!ispipe && !pid_in_pattern && core_uses_pid) {
276 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
277 if (err)
278 return err;
279 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400280 return ispipe;
281}
282
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800283static int zap_process(struct task_struct *start, int exit_code, int flags)
Alex Kelly10c28d92012-09-26 21:52:08 -0400284{
285 struct task_struct *t;
286 int nr = 0;
287
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800288 /* ignore all signals except SIGKILL, see prepare_signal() */
289 start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
Alex Kelly10c28d92012-09-26 21:52:08 -0400290 start->signal->group_exit_code = exit_code;
291 start->signal->group_stop_count = 0;
292
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800293 for_each_thread(start, t) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400294 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
295 if (t != current && t->mm) {
296 sigaddset(&t->pending.signal, SIGKILL);
297 signal_wake_up(t, 1);
298 nr++;
299 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800300 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400301
302 return nr;
303}
304
Oleg Nesterov403bad72013-04-30 15:28:10 -0700305static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
306 struct core_state *core_state, int exit_code)
Alex Kelly10c28d92012-09-26 21:52:08 -0400307{
308 struct task_struct *g, *p;
309 unsigned long flags;
310 int nr = -EAGAIN;
311
312 spin_lock_irq(&tsk->sighand->siglock);
313 if (!signal_group_exit(tsk->signal)) {
314 mm->core_state = core_state;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700315 tsk->signal->group_exit_task = tsk;
Oleg Nesterov5fa534c2015-11-06 16:32:31 -0800316 nr = zap_process(tsk, exit_code, 0);
Oleg Nesterov403bad72013-04-30 15:28:10 -0700317 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
Alex Kelly10c28d92012-09-26 21:52:08 -0400318 }
319 spin_unlock_irq(&tsk->sighand->siglock);
320 if (unlikely(nr < 0))
321 return nr;
322
Silesh C Vaed8adb2014-07-23 13:59:59 -0700323 tsk->flags |= PF_DUMPCORE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400324 if (atomic_read(&mm->mm_users) == nr + 1)
325 goto done;
326 /*
327 * We should find and kill all tasks which use this mm, and we should
328 * count them correctly into ->nr_threads. We don't take tasklist
329 * lock, but this is safe wrt:
330 *
331 * fork:
332 * None of sub-threads can fork after zap_process(leader). All
333 * processes which were created before this point should be
334 * visible to zap_threads() because copy_process() adds the new
335 * process to the tail of init_task.tasks list, and lock/unlock
336 * of ->siglock provides a memory barrier.
337 *
338 * do_exit:
339 * The caller holds mm->mmap_sem. This means that the task which
340 * uses this mm can't pass exit_mm(), so it can't exit or clear
341 * its ->mm.
342 *
343 * de_thread:
344 * It does list_replace_rcu(&leader->tasks, &current->tasks),
345 * we must see either old or new leader, this does not matter.
346 * However, it can change p->sighand, so lock_task_sighand(p)
347 * must be used. Since p->mm != NULL and we hold ->mmap_sem
348 * it can't fail.
349 *
350 * Note also that "g" can be the old leader with ->mm == NULL
351 * and already unhashed and thus removed from ->thread_group.
352 * This is OK, __unhash_process()->list_del_rcu() does not
353 * clear the ->next pointer, we will find the new leader via
354 * next_thread().
355 */
356 rcu_read_lock();
357 for_each_process(g) {
358 if (g == tsk->group_leader)
359 continue;
360 if (g->flags & PF_KTHREAD)
361 continue;
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800362
363 for_each_thread(g, p) {
364 if (unlikely(!p->mm))
365 continue;
366 if (unlikely(p->mm == mm)) {
367 lock_task_sighand(p, &flags);
368 nr += zap_process(p, exit_code,
369 SIGNAL_GROUP_EXIT);
370 unlock_task_sighand(p, &flags);
Alex Kelly10c28d92012-09-26 21:52:08 -0400371 }
Oleg Nesterovd61ba582015-11-06 16:32:34 -0800372 break;
373 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400374 }
375 rcu_read_unlock();
376done:
377 atomic_set(&core_state->nr_threads, nr);
378 return nr;
379}
380
381static int coredump_wait(int exit_code, struct core_state *core_state)
382{
383 struct task_struct *tsk = current;
384 struct mm_struct *mm = tsk->mm;
385 int core_waiters = -EBUSY;
386
387 init_completion(&core_state->startup);
388 core_state->dumper.task = tsk;
389 core_state->dumper.next = NULL;
390
391 down_write(&mm->mmap_sem);
392 if (!mm->core_state)
393 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
394 up_write(&mm->mmap_sem);
395
396 if (core_waiters > 0) {
397 struct core_thread *ptr;
398
399 wait_for_completion(&core_state->startup);
400 /*
401 * Wait for all the threads to become inactive, so that
402 * all the thread context (extended register state, like
403 * fpu etc) gets copied to the memory.
404 */
405 ptr = core_state->dumper.next;
406 while (ptr != NULL) {
407 wait_task_inactive(ptr->task, 0);
408 ptr = ptr->next;
409 }
410 }
411
412 return core_waiters;
413}
414
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700415static void coredump_finish(struct mm_struct *mm, bool core_dumped)
Alex Kelly10c28d92012-09-26 21:52:08 -0400416{
417 struct core_thread *curr, *next;
418 struct task_struct *task;
419
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700420 spin_lock_irq(&current->sighand->siglock);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700421 if (core_dumped && !__fatal_signal_pending(current))
422 current->signal->group_exit_code |= 0x80;
Oleg Nesterov6cd8f0a2013-04-30 15:28:12 -0700423 current->signal->group_exit_task = NULL;
424 current->signal->flags = SIGNAL_GROUP_EXIT;
425 spin_unlock_irq(&current->sighand->siglock);
426
Alex Kelly10c28d92012-09-26 21:52:08 -0400427 next = mm->core_state->dumper.next;
428 while ((curr = next) != NULL) {
429 next = curr->next;
430 task = curr->task;
431 /*
432 * see exit_mm(), curr->task must not see
433 * ->task == NULL before we read ->next.
434 */
435 smp_mb();
436 curr->task = NULL;
437 wake_up_process(task);
438 }
439
440 mm->core_state = NULL;
441}
442
Oleg Nesterov528f8272013-04-30 15:28:15 -0700443static bool dump_interrupted(void)
444{
445 /*
446 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
447 * can do try_to_freeze() and check __fatal_signal_pending(),
448 * but then we need to teach dump_write() to restart and clear
449 * TIF_SIGPENDING.
450 */
451 return signal_pending(current);
452}
453
Alex Kelly10c28d92012-09-26 21:52:08 -0400454static void wait_for_dump_helpers(struct file *file)
455{
Al Virode32ec42013-03-21 11:16:56 -0400456 struct pipe_inode_info *pipe = file->private_data;
Alex Kelly10c28d92012-09-26 21:52:08 -0400457
458 pipe_lock(pipe);
459 pipe->readers++;
460 pipe->writers--;
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700461 wake_up_interruptible_sync(&pipe->wait);
462 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
463 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400464
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700465 /*
466 * We actually want wait_event_freezable() but then we need
467 * to clear TIF_SIGPENDING and improve dump_interrupted().
468 */
469 wait_event_interruptible(pipe->wait, pipe->readers == 1);
Alex Kelly10c28d92012-09-26 21:52:08 -0400470
Oleg Nesterovdc7ee2a2013-04-30 15:28:17 -0700471 pipe_lock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400472 pipe->readers--;
473 pipe->writers++;
474 pipe_unlock(pipe);
Alex Kelly10c28d92012-09-26 21:52:08 -0400475}
476
477/*
478 * umh_pipe_setup
479 * helper function to customize the process used
480 * to collect the core in userspace. Specifically
481 * it sets up a pipe and installs it as fd 0 (stdin)
482 * for the process. Returns 0 on success, or
483 * PTR_ERR on failure.
484 * Note that it also sets the core limit to 1. This
485 * is a special value that we use to trap recursive
486 * core dumps
487 */
488static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
489{
490 struct file *files[2];
491 struct coredump_params *cp = (struct coredump_params *)info->data;
492 int err = create_pipe_files(files, 0);
493 if (err)
494 return err;
495
496 cp->file = files[1];
497
Al Viro45525b22012-10-16 13:30:07 -0400498 err = replace_fd(0, files[0], 0);
499 fput(files[0]);
Alex Kelly10c28d92012-09-26 21:52:08 -0400500 /* and disallow core files too */
501 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
502
Al Viro45525b22012-10-16 13:30:07 -0400503 return err;
Alex Kelly10c28d92012-09-26 21:52:08 -0400504}
505
Al Viroec579412013-10-13 17:57:29 -0400506void do_coredump(const siginfo_t *siginfo)
Alex Kelly10c28d92012-09-26 21:52:08 -0400507{
508 struct core_state core_state;
509 struct core_name cn;
510 struct mm_struct *mm = current->mm;
511 struct linux_binfmt * binfmt;
512 const struct cred *old_cred;
513 struct cred *cred;
514 int retval = 0;
Alex Kelly10c28d92012-09-26 21:52:08 -0400515 int ispipe;
516 struct files_struct *displaced;
Jann Hornfbb18162015-09-09 15:38:28 -0700517 /* require nonrelative corefile path and be extra careful */
518 bool need_suid_safe = false;
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700519 bool core_dumped = false;
Alex Kelly10c28d92012-09-26 21:52:08 -0400520 static atomic_t core_dump_count = ATOMIC_INIT(0);
521 struct coredump_params cprm = {
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700522 .siginfo = siginfo,
Al Viro541880d2012-11-05 13:11:26 -0500523 .regs = signal_pt_regs(),
Alex Kelly10c28d92012-09-26 21:52:08 -0400524 .limit = rlimit(RLIMIT_CORE),
525 /*
526 * We must use the same mm->flags while dumping core to avoid
527 * inconsistency of bit flags, since this flag is not protected
528 * by any locks.
529 */
530 .mm_flags = mm->flags,
531 };
532
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700533 audit_core_dumps(siginfo->si_signo);
Alex Kelly10c28d92012-09-26 21:52:08 -0400534
535 binfmt = mm->binfmt;
536 if (!binfmt || !binfmt->core_dump)
537 goto fail;
538 if (!__get_dumpable(cprm.mm_flags))
539 goto fail;
540
541 cred = prepare_creds();
542 if (!cred)
543 goto fail;
544 /*
545 * We cannot trust fsuid as being the "true" uid of the process
546 * nor do we know its entire history. We only know it was tainted
547 * so we dump it as root in mode 2, and only into a controlled
548 * environment (pipe handler or fully qualified path).
549 */
Kees Cooke579d2c2013-02-27 17:03:15 -0800550 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400551 /* Setuid core dump mode */
Alex Kelly10c28d92012-09-26 21:52:08 -0400552 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
Jann Hornfbb18162015-09-09 15:38:28 -0700553 need_suid_safe = true;
Alex Kelly10c28d92012-09-26 21:52:08 -0400554 }
555
Denys Vlasenko5ab1c302012-10-04 17:15:29 -0700556 retval = coredump_wait(siginfo->si_signo, &core_state);
Alex Kelly10c28d92012-09-26 21:52:08 -0400557 if (retval < 0)
558 goto fail_creds;
559
560 old_cred = override_creds(cred);
561
Oleg Nesterov12a2b4b2012-10-04 17:15:25 -0700562 ispipe = format_corename(&cn, &cprm);
Alex Kelly10c28d92012-09-26 21:52:08 -0400563
Lucas De Marchifb96c472013-04-30 15:28:06 -0700564 if (ispipe) {
Alex Kelly10c28d92012-09-26 21:52:08 -0400565 int dump_count;
566 char **helper_argv;
Lucas De Marchi907ed132013-04-30 15:28:07 -0700567 struct subprocess_info *sub_info;
Alex Kelly10c28d92012-09-26 21:52:08 -0400568
569 if (ispipe < 0) {
570 printk(KERN_WARNING "format_corename failed\n");
571 printk(KERN_WARNING "Aborting core\n");
Oleg Nesterove7fd1542013-07-03 15:08:16 -0700572 goto fail_unlock;
Alex Kelly10c28d92012-09-26 21:52:08 -0400573 }
574
575 if (cprm.limit == 1) {
576 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
577 *
578 * Normally core limits are irrelevant to pipes, since
579 * we're not writing to the file system, but we use
Bastien Nocerafcbc32b2015-02-05 14:35:05 +0100580 * cprm.limit of 1 here as a special value, this is a
Alex Kelly10c28d92012-09-26 21:52:08 -0400581 * consistent way to catch recursive crashes.
582 * We can still crash if the core_pattern binary sets
583 * RLIM_CORE = !1, but it runs as root, and can do
584 * lots of stupid things.
585 *
586 * Note that we use task_tgid_vnr here to grab the pid
587 * of the process group leader. That way we get the
588 * right pid if a thread in a multi-threaded
589 * core_pattern process dies.
590 */
591 printk(KERN_WARNING
592 "Process %d(%s) has RLIMIT_CORE set to 1\n",
593 task_tgid_vnr(current), current->comm);
594 printk(KERN_WARNING "Aborting core\n");
595 goto fail_unlock;
596 }
597 cprm.limit = RLIM_INFINITY;
598
599 dump_count = atomic_inc_return(&core_dump_count);
600 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
601 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
602 task_tgid_vnr(current), current->comm);
603 printk(KERN_WARNING "Skipping core dump\n");
604 goto fail_dropcount;
605 }
606
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700607 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
Alex Kelly10c28d92012-09-26 21:52:08 -0400608 if (!helper_argv) {
609 printk(KERN_WARNING "%s failed to allocate memory\n",
610 __func__);
611 goto fail_dropcount;
612 }
613
Lucas De Marchi907ed132013-04-30 15:28:07 -0700614 retval = -ENOMEM;
615 sub_info = call_usermodehelper_setup(helper_argv[0],
616 helper_argv, NULL, GFP_KERNEL,
617 umh_pipe_setup, NULL, &cprm);
618 if (sub_info)
619 retval = call_usermodehelper_exec(sub_info,
620 UMH_WAIT_EXEC);
621
Alex Kelly10c28d92012-09-26 21:52:08 -0400622 argv_free(helper_argv);
623 if (retval) {
Oleg Nesterov888ffc52013-07-03 15:08:23 -0700624 printk(KERN_INFO "Core dump to |%s pipe failed\n",
Alex Kelly10c28d92012-09-26 21:52:08 -0400625 cn.corename);
626 goto close_fail;
Lucas De Marchifb96c472013-04-30 15:28:06 -0700627 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400628 } else {
629 struct inode *inode;
630
631 if (cprm.limit < binfmt->min_coredump)
632 goto fail_unlock;
633
Jann Hornfbb18162015-09-09 15:38:28 -0700634 if (need_suid_safe && cn.corename[0] != '/') {
Alex Kelly10c28d92012-09-26 21:52:08 -0400635 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
636 "to fully qualified path!\n",
637 task_tgid_vnr(current), current->comm);
638 printk(KERN_WARNING "Skipping core dump\n");
639 goto fail_unlock;
640 }
641
Jann Hornfbb18162015-09-09 15:38:28 -0700642 /*
643 * Unlink the file if it exists unless this is a SUID
644 * binary - in that case, we're running around with root
645 * privs and don't want to unlink another user's coredump.
646 */
647 if (!need_suid_safe) {
648 mm_segment_t old_fs;
649
650 old_fs = get_fs();
651 set_fs(KERNEL_DS);
652 /*
653 * If it doesn't exist, that's fine. If there's some
654 * other problem, we'll catch it at the filp_open().
655 */
656 (void) sys_unlink((const char __user *)cn.corename);
657 set_fs(old_fs);
658 }
659
660 /*
661 * There is a race between unlinking and creating the
662 * file, but if that causes an EEXIST here, that's
663 * fine - another process raced with us while creating
664 * the corefile, and the other process won. To userspace,
665 * what matters is that at least one of the two processes
666 * writes its coredump successfully, not which one.
667 */
Alex Kelly10c28d92012-09-26 21:52:08 -0400668 cprm.file = filp_open(cn.corename,
Jann Hornfbb18162015-09-09 15:38:28 -0700669 O_CREAT | 2 | O_NOFOLLOW |
670 O_LARGEFILE | O_EXCL,
Alex Kelly10c28d92012-09-26 21:52:08 -0400671 0600);
672 if (IS_ERR(cprm.file))
673 goto fail_unlock;
674
Al Viro496ad9a2013-01-23 17:07:38 -0500675 inode = file_inode(cprm.file);
Alex Kelly10c28d92012-09-26 21:52:08 -0400676 if (inode->i_nlink > 1)
677 goto close_fail;
678 if (d_unhashed(cprm.file->f_path.dentry))
679 goto close_fail;
680 /*
681 * AK: actually i see no reason to not allow this for named
682 * pipes etc, but keep the previous behaviour for now.
683 */
684 if (!S_ISREG(inode->i_mode))
685 goto close_fail;
686 /*
Jann Horn40f705a2015-09-09 15:38:30 -0700687 * Don't dump core if the filesystem changed owner or mode
688 * of the file during file creation. This is an issue when
689 * a process dumps core while its cwd is e.g. on a vfat
690 * filesystem.
Alex Kelly10c28d92012-09-26 21:52:08 -0400691 */
692 if (!uid_eq(inode->i_uid, current_fsuid()))
693 goto close_fail;
Jann Horn40f705a2015-09-09 15:38:30 -0700694 if ((inode->i_mode & 0677) != 0600)
695 goto close_fail;
Al Viro86cc0582015-04-03 15:23:17 -0400696 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
Alex Kelly10c28d92012-09-26 21:52:08 -0400697 goto close_fail;
698 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
699 goto close_fail;
700 }
701
702 /* get us an unshared descriptor table; almost always a no-op */
703 retval = unshare_files(&displaced);
704 if (retval)
705 goto close_fail;
706 if (displaced)
707 put_files_struct(displaced);
Al Viroe86d35c2013-05-04 14:45:54 -0400708 if (!dump_interrupted()) {
709 file_start_write(cprm.file);
710 core_dumped = binfmt->core_dump(&cprm);
711 file_end_write(cprm.file);
712 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400713 if (ispipe && core_pipe_limit)
714 wait_for_dump_helpers(cprm.file);
715close_fail:
716 if (cprm.file)
717 filp_close(cprm.file, NULL);
718fail_dropcount:
719 if (ispipe)
720 atomic_dec(&core_dump_count);
721fail_unlock:
722 kfree(cn.corename);
Oleg Nesterovacdedd92013-04-30 15:28:13 -0700723 coredump_finish(mm, core_dumped);
Alex Kelly10c28d92012-09-26 21:52:08 -0400724 revert_creds(old_cred);
725fail_creds:
726 put_cred(cred);
727fail:
728 return;
729}
730
731/*
732 * Core dumping helper functions. These are the only things you should
733 * do on a core-file: use only these functions to write out all the
734 * necessary info.
735 */
Al Viroecc8c772013-10-05 15:32:35 -0400736int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
737{
738 struct file *file = cprm->file;
Al Viro2507a4f2013-10-08 09:11:48 -0400739 loff_t pos = file->f_pos;
740 ssize_t n;
Al Viroecc8c772013-10-05 15:32:35 -0400741 if (cprm->written + nr > cprm->limit)
742 return 0;
Al Viro2507a4f2013-10-08 09:11:48 -0400743 while (nr) {
744 if (dump_interrupted())
745 return 0;
Al Viro52da40a2013-11-15 21:58:33 -0500746 n = __kernel_write(file, addr, nr, &pos);
Al Viro2507a4f2013-10-08 09:11:48 -0400747 if (n <= 0)
748 return 0;
749 file->f_pos = pos;
750 cprm->written += n;
751 nr -= n;
752 }
Al Viroecc8c772013-10-05 15:32:35 -0400753 return 1;
754}
755EXPORT_SYMBOL(dump_emit);
756
Al Viro9b56d542013-10-08 09:26:08 -0400757int dump_skip(struct coredump_params *cprm, size_t nr)
Alex Kelly10c28d92012-09-26 21:52:08 -0400758{
Al Viro9b56d542013-10-08 09:26:08 -0400759 static char zeroes[PAGE_SIZE];
760 struct file *file = cprm->file;
Alex Kelly10c28d92012-09-26 21:52:08 -0400761 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
Al Viro9b56d542013-10-08 09:26:08 -0400762 if (cprm->written + nr > cprm->limit)
763 return 0;
Oleg Nesterov528f8272013-04-30 15:28:15 -0700764 if (dump_interrupted() ||
Al Viro9b56d542013-10-08 09:26:08 -0400765 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
Alex Kelly10c28d92012-09-26 21:52:08 -0400766 return 0;
Al Viro9b56d542013-10-08 09:26:08 -0400767 cprm->written += nr;
768 return 1;
Alex Kelly10c28d92012-09-26 21:52:08 -0400769 } else {
Al Viro9b56d542013-10-08 09:26:08 -0400770 while (nr > PAGE_SIZE) {
771 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
772 return 0;
773 nr -= PAGE_SIZE;
Alex Kelly10c28d92012-09-26 21:52:08 -0400774 }
Al Viro9b56d542013-10-08 09:26:08 -0400775 return dump_emit(cprm, zeroes, nr);
Alex Kelly10c28d92012-09-26 21:52:08 -0400776 }
Alex Kelly10c28d92012-09-26 21:52:08 -0400777}
Al Viro9b56d542013-10-08 09:26:08 -0400778EXPORT_SYMBOL(dump_skip);
Al Viro22a8cb82013-10-08 11:05:01 -0400779
780int dump_align(struct coredump_params *cprm, int align)
781{
782 unsigned mod = cprm->written & (align - 1);
783 if (align & (align - 1))
Al Virodb512422013-11-15 21:55:52 -0500784 return 0;
785 return mod ? dump_skip(cprm, align - mod) : 1;
Al Viro22a8cb82013-10-08 11:05:01 -0400786}
787EXPORT_SYMBOL(dump_align);