]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blame - fs/coredump.c
ubifs: rename_whiteout: Fix double free for whiteout_ui->data
[mirror_ubuntu-focal-kernel.git] / fs / coredump.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
10c28d93
AK
2#include <linux/slab.h>
3#include <linux/file.h>
4#include <linux/fdtable.h>
70d78fe7 5#include <linux/freezer.h>
10c28d93
AK
6#include <linux/mm.h>
7#include <linux/stat.h>
8#include <linux/fcntl.h>
9#include <linux/swap.h>
315c6926 10#include <linux/ctype.h>
10c28d93
AK
11#include <linux/string.h>
12#include <linux/init.h>
13#include <linux/pagemap.h>
14#include <linux/perf_event.h>
15#include <linux/highmem.h>
16#include <linux/spinlock.h>
17#include <linux/key.h>
18#include <linux/personality.h>
19#include <linux/binfmts.h>
179899fd 20#include <linux/coredump.h>
f7ccbae4 21#include <linux/sched/coredump.h>
3f07c014 22#include <linux/sched/signal.h>
68db0cf1 23#include <linux/sched/task_stack.h>
10c28d93
AK
24#include <linux/utsname.h>
25#include <linux/pid_namespace.h>
26#include <linux/module.h>
27#include <linux/namei.h>
28#include <linux/mount.h>
29#include <linux/security.h>
30#include <linux/syscalls.h>
31#include <linux/tsacct_kern.h>
32#include <linux/cn_proc.h>
33#include <linux/audit.h>
34#include <linux/tracehook.h>
35#include <linux/kmod.h>
36#include <linux/fsnotify.h>
37#include <linux/fs_struct.h>
38#include <linux/pipe_fs_i.h>
39#include <linux/oom.h>
40#include <linux/compat.h>
378c6520
JH
41#include <linux/fs.h>
42#include <linux/path.h>
03927c8a 43#include <linux/timekeeping.h>
10c28d93 44
7c0f6ba6 45#include <linux/uaccess.h>
10c28d93
AK
46#include <asm/mmu_context.h>
47#include <asm/tlb.h>
48#include <asm/exec.h>
49
50#include <trace/events/task.h>
51#include "internal.h"
52
53#include <trace/events/sched.h>
54
55int core_uses_pid;
10c28d93 56unsigned int core_pipe_limit;
3ceadcf6
ON
57char core_pattern[CORENAME_MAX_SIZE] = "core";
58static int core_name_size = CORENAME_MAX_SIZE;
10c28d93
AK
59
60struct core_name {
61 char *corename;
62 int used, size;
63};
10c28d93
AK
64
65/* The maximal length of core_pattern is also specified in sysctl.c */
66
3ceadcf6 67static int expand_corename(struct core_name *cn, int size)
10c28d93 68{
e7fd1549 69 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
10c28d93 70
e7fd1549 71 if (!corename)
10c28d93 72 return -ENOMEM;
10c28d93 73
3ceadcf6
ON
74 if (size > core_name_size) /* racy but harmless */
75 core_name_size = size;
76
77 cn->size = ksize(corename);
e7fd1549 78 cn->corename = corename;
10c28d93
AK
79 return 0;
80}
81
b4176b7c
NI
82static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
83 va_list arg)
10c28d93 84{
5fe9d8ca 85 int free, need;
404ca80e 86 va_list arg_copy;
10c28d93 87
5fe9d8ca
ON
88again:
89 free = cn->size - cn->used;
404ca80e
ED
90
91 va_copy(arg_copy, arg);
92 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
93 va_end(arg_copy);
94
5fe9d8ca
ON
95 if (need < free) {
96 cn->used += need;
97 return 0;
98 }
10c28d93 99
3ceadcf6 100 if (!expand_corename(cn, cn->size + need - free + 1))
5fe9d8ca 101 goto again;
10c28d93 102
5fe9d8ca 103 return -ENOMEM;
10c28d93
AK
104}
105
b4176b7c 106static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
bc03c691
ON
107{
108 va_list arg;
109 int ret;
110
111 va_start(arg, fmt);
112 ret = cn_vprintf(cn, fmt, arg);
113 va_end(arg);
114
115 return ret;
116}
117
b4176b7c
NI
118static __printf(2, 3)
119int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
10c28d93 120{
923bed03
ON
121 int cur = cn->used;
122 va_list arg;
123 int ret;
124
125 va_start(arg, fmt);
126 ret = cn_vprintf(cn, fmt, arg);
127 va_end(arg);
128
ac94b6e3
JH
129 if (ret == 0) {
130 /*
131 * Ensure that this coredump name component can't cause the
132 * resulting corefile path to consist of a ".." or ".".
133 */
134 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
135 (cn->used - cur == 2 && cn->corename[cur] == '.'
136 && cn->corename[cur+1] == '.'))
137 cn->corename[cur] = '!';
138
139 /*
140 * Empty names are fishy and could be used to create a "//" in a
141 * corefile name, causing the coredump to happen one directory
142 * level too high. Enforce that all components of the core
143 * pattern are at least one character long.
144 */
145 if (cn->used == cur)
146 ret = cn_printf(cn, "!");
147 }
148
923bed03
ON
149 for (; cur < cn->used; ++cur) {
150 if (cn->corename[cur] == '/')
151 cn->corename[cur] = '!';
152 }
153 return ret;
10c28d93
AK
154}
155
156static int cn_print_exe_file(struct core_name *cn)
157{
158 struct file *exe_file;
159 char *pathbuf, *path;
160 int ret;
161
162 exe_file = get_mm_exe_file(current->mm);
923bed03
ON
163 if (!exe_file)
164 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
10c28d93 165
0ee931c4 166 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
10c28d93
AK
167 if (!pathbuf) {
168 ret = -ENOMEM;
169 goto put_exe_file;
170 }
171
9bf39ab2 172 path = file_path(exe_file, pathbuf, PATH_MAX);
10c28d93
AK
173 if (IS_ERR(path)) {
174 ret = PTR_ERR(path);
175 goto free_buf;
176 }
177
923bed03 178 ret = cn_esc_printf(cn, "%s", path);
10c28d93
AK
179
180free_buf:
181 kfree(pathbuf);
182put_exe_file:
183 fput(exe_file);
184 return ret;
185}
186
187/* format_corename will inspect the pattern parameter, and output a
188 * name into corename, which must have space for at least
189 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
190 */
315c6926
PW
191static int format_corename(struct core_name *cn, struct coredump_params *cprm,
192 size_t **argv, int *argc)
10c28d93
AK
193{
194 const struct cred *cred = current_cred();
195 const char *pat_ptr = core_pattern;
196 int ispipe = (*pat_ptr == '|');
315c6926 197 bool was_space = false;
10c28d93
AK
198 int pid_in_pattern = 0;
199 int err = 0;
200
e7fd1549 201 cn->used = 0;
3ceadcf6
ON
202 cn->corename = NULL;
203 if (expand_corename(cn, core_name_size))
10c28d93 204 return -ENOMEM;
888ffc59
ON
205 cn->corename[0] = '\0';
206
315c6926
PW
207 if (ispipe) {
208 int argvs = sizeof(core_pattern) / 2;
209 (*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
210 if (!(*argv))
211 return -ENOMEM;
212 (*argv)[(*argc)++] = 0;
888ffc59 213 ++pat_ptr;
aa62f813
SM
214 if (!(*pat_ptr))
215 return -ENOMEM;
315c6926 216 }
10c28d93
AK
217
218 /* Repeat as long as we have more pattern to process and more output
219 space */
220 while (*pat_ptr) {
315c6926
PW
221 /*
222 * Split on spaces before doing template expansion so that
223 * %e and %E don't get split if they have spaces in them
224 */
225 if (ispipe) {
226 if (isspace(*pat_ptr)) {
4f6e4758
MD
227 if (cn->used != 0)
228 was_space = true;
315c6926
PW
229 pat_ptr++;
230 continue;
231 } else if (was_space) {
232 was_space = false;
233 err = cn_printf(cn, "%c", '\0');
234 if (err)
235 return err;
236 (*argv)[(*argc)++] = cn->used;
237 }
238 }
10c28d93 239 if (*pat_ptr != '%') {
10c28d93
AK
240 err = cn_printf(cn, "%c", *pat_ptr++);
241 } else {
242 switch (*++pat_ptr) {
243 /* single % at the end, drop that */
244 case 0:
245 goto out;
246 /* Double percent, output one percent */
247 case '%':
248 err = cn_printf(cn, "%c", '%');
249 break;
250 /* pid */
251 case 'p':
252 pid_in_pattern = 1;
253 err = cn_printf(cn, "%d",
254 task_tgid_vnr(current));
255 break;
65aafb1e
SG
256 /* global pid */
257 case 'P':
258 err = cn_printf(cn, "%d",
259 task_tgid_nr(current));
260 break;
b03023ec
ON
261 case 'i':
262 err = cn_printf(cn, "%d",
263 task_pid_vnr(current));
264 break;
265 case 'I':
266 err = cn_printf(cn, "%d",
267 task_pid_nr(current));
268 break;
10c28d93
AK
269 /* uid */
270 case 'u':
5202efe5
NI
271 err = cn_printf(cn, "%u",
272 from_kuid(&init_user_ns,
273 cred->uid));
10c28d93
AK
274 break;
275 /* gid */
276 case 'g':
5202efe5
NI
277 err = cn_printf(cn, "%u",
278 from_kgid(&init_user_ns,
279 cred->gid));
10c28d93 280 break;
12a2b4b2
ON
281 case 'd':
282 err = cn_printf(cn, "%d",
283 __get_dumpable(cprm->mm_flags));
284 break;
10c28d93
AK
285 /* signal that caused the coredump */
286 case 's':
b4176b7c
NI
287 err = cn_printf(cn, "%d",
288 cprm->siginfo->si_signo);
10c28d93
AK
289 break;
290 /* UNIX time of coredump */
291 case 't': {
03927c8a
AB
292 time64_t time;
293
294 time = ktime_get_real_seconds();
295 err = cn_printf(cn, "%lld", time);
10c28d93
AK
296 break;
297 }
298 /* hostname */
923bed03 299 case 'h':
10c28d93 300 down_read(&uts_sem);
923bed03 301 err = cn_esc_printf(cn, "%s",
10c28d93
AK
302 utsname()->nodename);
303 up_read(&uts_sem);
10c28d93 304 break;
10c28d93 305 /* executable */
923bed03
ON
306 case 'e':
307 err = cn_esc_printf(cn, "%s", current->comm);
10c28d93 308 break;
10c28d93
AK
309 case 'E':
310 err = cn_print_exe_file(cn);
311 break;
312 /* core limit size */
313 case 'c':
314 err = cn_printf(cn, "%lu",
315 rlimit(RLIMIT_CORE));
316 break;
317 default:
318 break;
319 }
320 ++pat_ptr;
321 }
322
323 if (err)
324 return err;
325 }
326
888ffc59 327out:
10c28d93
AK
328 /* Backward compatibility with core_uses_pid:
329 *
330 * If core_pattern does not include a %p (as is the default)
331 * and core_uses_pid is set, then .%pid will be appended to
332 * the filename. Do not do this for piped commands. */
333 if (!ispipe && !pid_in_pattern && core_uses_pid) {
334 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
335 if (err)
336 return err;
337 }
10c28d93
AK
338 return ispipe;
339}
340
5fa534c9 341static int zap_process(struct task_struct *start, int exit_code, int flags)
10c28d93
AK
342{
343 struct task_struct *t;
344 int nr = 0;
345
5fa534c9
ON
346 /* ignore all signals except SIGKILL, see prepare_signal() */
347 start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
10c28d93
AK
348 start->signal->group_exit_code = exit_code;
349 start->signal->group_stop_count = 0;
350
d61ba589 351 for_each_thread(start, t) {
10c28d93
AK
352 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
353 if (t != current && t->mm) {
354 sigaddset(&t->pending.signal, SIGKILL);
355 signal_wake_up(t, 1);
356 nr++;
357 }
d61ba589 358 }
10c28d93
AK
359
360 return nr;
361}
362
403bad72
ON
363static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
364 struct core_state *core_state, int exit_code)
10c28d93
AK
365{
366 struct task_struct *g, *p;
367 unsigned long flags;
368 int nr = -EAGAIN;
369
370 spin_lock_irq(&tsk->sighand->siglock);
371 if (!signal_group_exit(tsk->signal)) {
372 mm->core_state = core_state;
6cd8f0ac 373 tsk->signal->group_exit_task = tsk;
5fa534c9 374 nr = zap_process(tsk, exit_code, 0);
403bad72 375 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
10c28d93
AK
376 }
377 spin_unlock_irq(&tsk->sighand->siglock);
378 if (unlikely(nr < 0))
379 return nr;
380
aed8adb7 381 tsk->flags |= PF_DUMPCORE;
10c28d93
AK
382 if (atomic_read(&mm->mm_users) == nr + 1)
383 goto done;
384 /*
385 * We should find and kill all tasks which use this mm, and we should
386 * count them correctly into ->nr_threads. We don't take tasklist
387 * lock, but this is safe wrt:
388 *
389 * fork:
390 * None of sub-threads can fork after zap_process(leader). All
391 * processes which were created before this point should be
392 * visible to zap_threads() because copy_process() adds the new
393 * process to the tail of init_task.tasks list, and lock/unlock
394 * of ->siglock provides a memory barrier.
395 *
396 * do_exit:
397 * The caller holds mm->mmap_sem. This means that the task which
398 * uses this mm can't pass exit_mm(), so it can't exit or clear
399 * its ->mm.
400 *
401 * de_thread:
402 * It does list_replace_rcu(&leader->tasks, &current->tasks),
403 * we must see either old or new leader, this does not matter.
404 * However, it can change p->sighand, so lock_task_sighand(p)
405 * must be used. Since p->mm != NULL and we hold ->mmap_sem
406 * it can't fail.
407 *
408 * Note also that "g" can be the old leader with ->mm == NULL
409 * and already unhashed and thus removed from ->thread_group.
410 * This is OK, __unhash_process()->list_del_rcu() does not
411 * clear the ->next pointer, we will find the new leader via
412 * next_thread().
413 */
414 rcu_read_lock();
415 for_each_process(g) {
416 if (g == tsk->group_leader)
417 continue;
418 if (g->flags & PF_KTHREAD)
419 continue;
d61ba589
ON
420
421 for_each_thread(g, p) {
422 if (unlikely(!p->mm))
423 continue;
424 if (unlikely(p->mm == mm)) {
425 lock_task_sighand(p, &flags);
426 nr += zap_process(p, exit_code,
427 SIGNAL_GROUP_EXIT);
428 unlock_task_sighand(p, &flags);
10c28d93 429 }
d61ba589
ON
430 break;
431 }
10c28d93
AK
432 }
433 rcu_read_unlock();
434done:
435 atomic_set(&core_state->nr_threads, nr);
436 return nr;
437}
438
439static int coredump_wait(int exit_code, struct core_state *core_state)
440{
441 struct task_struct *tsk = current;
442 struct mm_struct *mm = tsk->mm;
443 int core_waiters = -EBUSY;
444
445 init_completion(&core_state->startup);
446 core_state->dumper.task = tsk;
447 core_state->dumper.next = NULL;
448
4136c26b
MH
449 if (down_write_killable(&mm->mmap_sem))
450 return -EINTR;
451
10c28d93
AK
452 if (!mm->core_state)
453 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
454 up_write(&mm->mmap_sem);
455
456 if (core_waiters > 0) {
457 struct core_thread *ptr;
458
70d78fe7 459 freezer_do_not_count();
10c28d93 460 wait_for_completion(&core_state->startup);
70d78fe7 461 freezer_count();
10c28d93
AK
462 /*
463 * Wait for all the threads to become inactive, so that
464 * all the thread context (extended register state, like
465 * fpu etc) gets copied to the memory.
466 */
467 ptr = core_state->dumper.next;
468 while (ptr != NULL) {
469 wait_task_inactive(ptr->task, 0);
470 ptr = ptr->next;
471 }
472 }
473
474 return core_waiters;
475}
476
acdedd99 477static void coredump_finish(struct mm_struct *mm, bool core_dumped)
10c28d93
AK
478{
479 struct core_thread *curr, *next;
480 struct task_struct *task;
481
6cd8f0ac 482 spin_lock_irq(&current->sighand->siglock);
acdedd99
ON
483 if (core_dumped && !__fatal_signal_pending(current))
484 current->signal->group_exit_code |= 0x80;
6cd8f0ac
ON
485 current->signal->group_exit_task = NULL;
486 current->signal->flags = SIGNAL_GROUP_EXIT;
487 spin_unlock_irq(&current->sighand->siglock);
488
10c28d93
AK
489 next = mm->core_state->dumper.next;
490 while ((curr = next) != NULL) {
491 next = curr->next;
492 task = curr->task;
493 /*
494 * see exit_mm(), curr->task must not see
495 * ->task == NULL before we read ->next.
496 */
497 smp_mb();
498 curr->task = NULL;
499 wake_up_process(task);
500 }
501
502 mm->core_state = NULL;
503}
504
528f827e
ON
505static bool dump_interrupted(void)
506{
507 /*
508 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
509 * can do try_to_freeze() and check __fatal_signal_pending(),
510 * but then we need to teach dump_write() to restart and clear
511 * TIF_SIGPENDING.
512 */
513 return signal_pending(current);
514}
515
10c28d93
AK
516static void wait_for_dump_helpers(struct file *file)
517{
de32ec4c 518 struct pipe_inode_info *pipe = file->private_data;
10c28d93
AK
519
520 pipe_lock(pipe);
521 pipe->readers++;
522 pipe->writers--;
dc7ee2aa
ON
523 wake_up_interruptible_sync(&pipe->wait);
524 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
525 pipe_unlock(pipe);
10c28d93 526
dc7ee2aa
ON
527 /*
528 * We actually want wait_event_freezable() but then we need
529 * to clear TIF_SIGPENDING and improve dump_interrupted().
530 */
531 wait_event_interruptible(pipe->wait, pipe->readers == 1);
10c28d93 532
dc7ee2aa 533 pipe_lock(pipe);
10c28d93
AK
534 pipe->readers--;
535 pipe->writers++;
536 pipe_unlock(pipe);
10c28d93
AK
537}
538
539/*
540 * umh_pipe_setup
541 * helper function to customize the process used
542 * to collect the core in userspace. Specifically
543 * it sets up a pipe and installs it as fd 0 (stdin)
544 * for the process. Returns 0 on success, or
545 * PTR_ERR on failure.
546 * Note that it also sets the core limit to 1. This
547 * is a special value that we use to trap recursive
548 * core dumps
549 */
550static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
551{
552 struct file *files[2];
553 struct coredump_params *cp = (struct coredump_params *)info->data;
554 int err = create_pipe_files(files, 0);
555 if (err)
556 return err;
557
558 cp->file = files[1];
559
45525b26
AV
560 err = replace_fd(0, files[0], 0);
561 fput(files[0]);
10c28d93
AK
562 /* and disallow core files too */
563 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
564
45525b26 565 return err;
10c28d93
AK
566}
567
ae7795bc 568void do_coredump(const kernel_siginfo_t *siginfo)
10c28d93
AK
569{
570 struct core_state core_state;
571 struct core_name cn;
572 struct mm_struct *mm = current->mm;
573 struct linux_binfmt * binfmt;
574 const struct cred *old_cred;
575 struct cred *cred;
576 int retval = 0;
10c28d93 577 int ispipe;
315c6926
PW
578 size_t *argv = NULL;
579 int argc = 0;
10c28d93 580 struct files_struct *displaced;
fbb18169
JH
581 /* require nonrelative corefile path and be extra careful */
582 bool need_suid_safe = false;
acdedd99 583 bool core_dumped = false;
10c28d93
AK
584 static atomic_t core_dump_count = ATOMIC_INIT(0);
585 struct coredump_params cprm = {
5ab1c309 586 .siginfo = siginfo,
541880d9 587 .regs = signal_pt_regs(),
10c28d93
AK
588 .limit = rlimit(RLIMIT_CORE),
589 /*
590 * We must use the same mm->flags while dumping core to avoid
591 * inconsistency of bit flags, since this flag is not protected
592 * by any locks.
593 */
594 .mm_flags = mm->flags,
595 };
596
5ab1c309 597 audit_core_dumps(siginfo->si_signo);
10c28d93
AK
598
599 binfmt = mm->binfmt;
600 if (!binfmt || !binfmt->core_dump)
601 goto fail;
602 if (!__get_dumpable(cprm.mm_flags))
603 goto fail;
604
605 cred = prepare_creds();
606 if (!cred)
607 goto fail;
608 /*
609 * We cannot trust fsuid as being the "true" uid of the process
610 * nor do we know its entire history. We only know it was tainted
611 * so we dump it as root in mode 2, and only into a controlled
612 * environment (pipe handler or fully qualified path).
613 */
e579d2c2 614 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
10c28d93 615 /* Setuid core dump mode */
10c28d93 616 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
fbb18169 617 need_suid_safe = true;
10c28d93
AK
618 }
619
5ab1c309 620 retval = coredump_wait(siginfo->si_signo, &core_state);
10c28d93
AK
621 if (retval < 0)
622 goto fail_creds;
623
624 old_cred = override_creds(cred);
625
315c6926 626 ispipe = format_corename(&cn, &cprm, &argv, &argc);
10c28d93 627
fb96c475 628 if (ispipe) {
315c6926 629 int argi;
10c28d93
AK
630 int dump_count;
631 char **helper_argv;
907ed132 632 struct subprocess_info *sub_info;
10c28d93
AK
633
634 if (ispipe < 0) {
635 printk(KERN_WARNING "format_corename failed\n");
636 printk(KERN_WARNING "Aborting core\n");
e7fd1549 637 goto fail_unlock;
10c28d93
AK
638 }
639
640 if (cprm.limit == 1) {
641 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
642 *
643 * Normally core limits are irrelevant to pipes, since
644 * we're not writing to the file system, but we use
fcbc32bc 645 * cprm.limit of 1 here as a special value, this is a
10c28d93
AK
646 * consistent way to catch recursive crashes.
647 * We can still crash if the core_pattern binary sets
648 * RLIM_CORE = !1, but it runs as root, and can do
649 * lots of stupid things.
650 *
651 * Note that we use task_tgid_vnr here to grab the pid
652 * of the process group leader. That way we get the
653 * right pid if a thread in a multi-threaded
654 * core_pattern process dies.
655 */
656 printk(KERN_WARNING
657 "Process %d(%s) has RLIMIT_CORE set to 1\n",
658 task_tgid_vnr(current), current->comm);
659 printk(KERN_WARNING "Aborting core\n");
660 goto fail_unlock;
661 }
662 cprm.limit = RLIM_INFINITY;
663
664 dump_count = atomic_inc_return(&core_dump_count);
665 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
666 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
667 task_tgid_vnr(current), current->comm);
668 printk(KERN_WARNING "Skipping core dump\n");
669 goto fail_dropcount;
670 }
671
315c6926
PW
672 helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
673 GFP_KERNEL);
10c28d93
AK
674 if (!helper_argv) {
675 printk(KERN_WARNING "%s failed to allocate memory\n",
676 __func__);
677 goto fail_dropcount;
678 }
315c6926
PW
679 for (argi = 0; argi < argc; argi++)
680 helper_argv[argi] = cn.corename + argv[argi];
681 helper_argv[argi] = NULL;
10c28d93 682
907ed132
LDM
683 retval = -ENOMEM;
684 sub_info = call_usermodehelper_setup(helper_argv[0],
685 helper_argv, NULL, GFP_KERNEL,
686 umh_pipe_setup, NULL, &cprm);
687 if (sub_info)
688 retval = call_usermodehelper_exec(sub_info,
689 UMH_WAIT_EXEC);
690
315c6926 691 kfree(helper_argv);
10c28d93 692 if (retval) {
888ffc59 693 printk(KERN_INFO "Core dump to |%s pipe failed\n",
10c28d93
AK
694 cn.corename);
695 goto close_fail;
fb96c475 696 }
10c28d93
AK
697 } else {
698 struct inode *inode;
378c6520
JH
699 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
700 O_LARGEFILE | O_EXCL;
10c28d93
AK
701
702 if (cprm.limit < binfmt->min_coredump)
703 goto fail_unlock;
704
fbb18169 705 if (need_suid_safe && cn.corename[0] != '/') {
10c28d93
AK
706 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
707 "to fully qualified path!\n",
708 task_tgid_vnr(current), current->comm);
709 printk(KERN_WARNING "Skipping core dump\n");
710 goto fail_unlock;
711 }
712
fbb18169
JH
713 /*
714 * Unlink the file if it exists unless this is a SUID
715 * binary - in that case, we're running around with root
716 * privs and don't want to unlink another user's coredump.
717 */
718 if (!need_suid_safe) {
fbb18169
JH
719 /*
720 * If it doesn't exist, that's fine. If there's some
721 * other problem, we'll catch it at the filp_open().
722 */
96271654 723 do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
fbb18169
JH
724 }
725
726 /*
727 * There is a race between unlinking and creating the
728 * file, but if that causes an EEXIST here, that's
729 * fine - another process raced with us while creating
730 * the corefile, and the other process won. To userspace,
731 * what matters is that at least one of the two processes
732 * writes its coredump successfully, not which one.
733 */
378c6520
JH
734 if (need_suid_safe) {
735 /*
736 * Using user namespaces, normal user tasks can change
737 * their current->fs->root to point to arbitrary
738 * directories. Since the intention of the "only dump
739 * with a fully qualified path" rule is to control where
740 * coredumps may be placed using root privileges,
741 * current->fs->root must not be used. Instead, use the
742 * root directory of init_task.
743 */
744 struct path root;
745
746 task_lock(&init_task);
747 get_fs_root(init_task.fs, &root);
748 task_unlock(&init_task);
749 cprm.file = file_open_root(root.dentry, root.mnt,
750 cn.corename, open_flags, 0600);
751 path_put(&root);
752 } else {
753 cprm.file = filp_open(cn.corename, open_flags, 0600);
754 }
10c28d93
AK
755 if (IS_ERR(cprm.file))
756 goto fail_unlock;
757
496ad9aa 758 inode = file_inode(cprm.file);
10c28d93
AK
759 if (inode->i_nlink > 1)
760 goto close_fail;
761 if (d_unhashed(cprm.file->f_path.dentry))
762 goto close_fail;
763 /*
764 * AK: actually i see no reason to not allow this for named
765 * pipes etc, but keep the previous behaviour for now.
766 */
767 if (!S_ISREG(inode->i_mode))
768 goto close_fail;
769 /*
40f705a7
JH
770 * Don't dump core if the filesystem changed owner or mode
771 * of the file during file creation. This is an issue when
772 * a process dumps core while its cwd is e.g. on a vfat
773 * filesystem.
10c28d93
AK
774 */
775 if (!uid_eq(inode->i_uid, current_fsuid()))
776 goto close_fail;
40f705a7
JH
777 if ((inode->i_mode & 0677) != 0600)
778 goto close_fail;
86cc0584 779 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
10c28d93
AK
780 goto close_fail;
781 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
782 goto close_fail;
783 }
784
785 /* get us an unshared descriptor table; almost always a no-op */
786 retval = unshare_files(&displaced);
787 if (retval)
788 goto close_fail;
789 if (displaced)
790 put_files_struct(displaced);
e86d35c3 791 if (!dump_interrupted()) {
2f9829cc
LC
792 /*
793 * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would
794 * have this set to NULL.
795 */
796 if (!cprm.file) {
797 pr_info("Core dump to |%s disabled\n", cn.corename);
798 goto close_fail;
799 }
e86d35c3
AV
800 file_start_write(cprm.file);
801 core_dumped = binfmt->core_dump(&cprm);
802 file_end_write(cprm.file);
803 }
10c28d93
AK
804 if (ispipe && core_pipe_limit)
805 wait_for_dump_helpers(cprm.file);
806close_fail:
807 if (cprm.file)
808 filp_close(cprm.file, NULL);
809fail_dropcount:
810 if (ispipe)
811 atomic_dec(&core_dump_count);
812fail_unlock:
315c6926 813 kfree(argv);
10c28d93 814 kfree(cn.corename);
acdedd99 815 coredump_finish(mm, core_dumped);
10c28d93
AK
816 revert_creds(old_cred);
817fail_creds:
818 put_cred(cred);
819fail:
820 return;
821}
822
823/*
824 * Core dumping helper functions. These are the only things you should
825 * do on a core-file: use only these functions to write out all the
826 * necessary info.
827 */
ecc8c772
AV
828int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
829{
830 struct file *file = cprm->file;
2507a4fb
AV
831 loff_t pos = file->f_pos;
832 ssize_t n;
2c4cb043 833 if (cprm->written + nr > cprm->limit)
ecc8c772 834 return 0;
2507a4fb
AV
835 while (nr) {
836 if (dump_interrupted())
837 return 0;
52da40ae 838 n = __kernel_write(file, addr, nr, &pos);
2507a4fb
AV
839 if (n <= 0)
840 return 0;
841 file->f_pos = pos;
2c4cb043 842 cprm->written += n;
1607f09c 843 cprm->pos += n;
2507a4fb
AV
844 nr -= n;
845 }
ecc8c772
AV
846 return 1;
847}
848EXPORT_SYMBOL(dump_emit);
849
9b56d543 850int dump_skip(struct coredump_params *cprm, size_t nr)
10c28d93 851{
9b56d543
AV
852 static char zeroes[PAGE_SIZE];
853 struct file *file = cprm->file;
10c28d93 854 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
528f827e 855 if (dump_interrupted() ||
9b56d543 856 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
10c28d93 857 return 0;
1607f09c 858 cprm->pos += nr;
9b56d543 859 return 1;
10c28d93 860 } else {
9b56d543
AV
861 while (nr > PAGE_SIZE) {
862 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
863 return 0;
864 nr -= PAGE_SIZE;
10c28d93 865 }
9b56d543 866 return dump_emit(cprm, zeroes, nr);
10c28d93 867 }
10c28d93 868}
9b56d543 869EXPORT_SYMBOL(dump_skip);
22a8cb82
AV
870
871int dump_align(struct coredump_params *cprm, int align)
872{
1607f09c 873 unsigned mod = cprm->pos & (align - 1);
22a8cb82 874 if (align & (align - 1))
db51242d
AV
875 return 0;
876 return mod ? dump_skip(cprm, align - mod) : 1;
22a8cb82
AV
877}
878EXPORT_SYMBOL(dump_align);
4d22c75d
DK
879
880/*
881 * Ensures that file size is big enough to contain the current file
882 * postion. This prevents gdb from complaining about a truncated file
883 * if the last "write" to the file was dump_skip.
884 */
885void dump_truncate(struct coredump_params *cprm)
886{
887 struct file *file = cprm->file;
888 loff_t offset;
889
890 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
891 offset = file->f_op->llseek(file, 0, SEEK_CUR);
892 if (i_size_read(file->f_mapping->host) < offset)
893 do_truncate(file->f_path.dentry, offset, 0, file);
894 }
895}
896EXPORT_SYMBOL(dump_truncate);