]>
Commit | Line | Data |
---|---|---|
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> | |
10c28d93 AK |
34 | #include <linux/kmod.h> |
35 | #include <linux/fsnotify.h> | |
36 | #include <linux/fs_struct.h> | |
37 | #include <linux/pipe_fs_i.h> | |
38 | #include <linux/oom.h> | |
39 | #include <linux/compat.h> | |
378c6520 JH |
40 | #include <linux/fs.h> |
41 | #include <linux/path.h> | |
03927c8a | 42 | #include <linux/timekeeping.h> |
f0bc21b2 | 43 | #include <linux/sysctl.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 | ||
f0bc21b2 XN |
55 | static int core_uses_pid; |
56 | static unsigned int core_pipe_limit; | |
57 | static char core_pattern[CORENAME_MAX_SIZE] = "core"; | |
3ceadcf6 | 58 | static int core_name_size = CORENAME_MAX_SIZE; |
10c28d93 AK |
59 | |
60 | struct core_name { | |
61 | char *corename; | |
62 | int used, size; | |
63 | }; | |
10c28d93 | 64 | |
3ceadcf6 | 65 | static int expand_corename(struct core_name *cn, int size) |
10c28d93 | 66 | { |
e7fd1549 | 67 | char *corename = krealloc(cn->corename, size, GFP_KERNEL); |
10c28d93 | 68 | |
e7fd1549 | 69 | if (!corename) |
10c28d93 | 70 | return -ENOMEM; |
10c28d93 | 71 | |
3ceadcf6 ON |
72 | if (size > core_name_size) /* racy but harmless */ |
73 | core_name_size = size; | |
74 | ||
75 | cn->size = ksize(corename); | |
e7fd1549 | 76 | cn->corename = corename; |
10c28d93 AK |
77 | return 0; |
78 | } | |
79 | ||
b4176b7c NI |
80 | static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt, |
81 | va_list arg) | |
10c28d93 | 82 | { |
5fe9d8ca | 83 | int free, need; |
404ca80e | 84 | va_list arg_copy; |
10c28d93 | 85 | |
5fe9d8ca ON |
86 | again: |
87 | free = cn->size - cn->used; | |
404ca80e ED |
88 | |
89 | va_copy(arg_copy, arg); | |
90 | need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy); | |
91 | va_end(arg_copy); | |
92 | ||
5fe9d8ca ON |
93 | if (need < free) { |
94 | cn->used += need; | |
95 | return 0; | |
96 | } | |
10c28d93 | 97 | |
3ceadcf6 | 98 | if (!expand_corename(cn, cn->size + need - free + 1)) |
5fe9d8ca | 99 | goto again; |
10c28d93 | 100 | |
5fe9d8ca | 101 | return -ENOMEM; |
10c28d93 AK |
102 | } |
103 | ||
b4176b7c | 104 | static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...) |
bc03c691 ON |
105 | { |
106 | va_list arg; | |
107 | int ret; | |
108 | ||
109 | va_start(arg, fmt); | |
110 | ret = cn_vprintf(cn, fmt, arg); | |
111 | va_end(arg); | |
112 | ||
113 | return ret; | |
114 | } | |
115 | ||
b4176b7c NI |
116 | static __printf(2, 3) |
117 | int cn_esc_printf(struct core_name *cn, const char *fmt, ...) | |
10c28d93 | 118 | { |
923bed03 ON |
119 | int cur = cn->used; |
120 | va_list arg; | |
121 | int ret; | |
122 | ||
123 | va_start(arg, fmt); | |
124 | ret = cn_vprintf(cn, fmt, arg); | |
125 | va_end(arg); | |
126 | ||
ac94b6e3 JH |
127 | if (ret == 0) { |
128 | /* | |
129 | * Ensure that this coredump name component can't cause the | |
130 | * resulting corefile path to consist of a ".." or ".". | |
131 | */ | |
132 | if ((cn->used - cur == 1 && cn->corename[cur] == '.') || | |
133 | (cn->used - cur == 2 && cn->corename[cur] == '.' | |
134 | && cn->corename[cur+1] == '.')) | |
135 | cn->corename[cur] = '!'; | |
136 | ||
137 | /* | |
138 | * Empty names are fishy and could be used to create a "//" in a | |
139 | * corefile name, causing the coredump to happen one directory | |
140 | * level too high. Enforce that all components of the core | |
141 | * pattern are at least one character long. | |
142 | */ | |
143 | if (cn->used == cur) | |
144 | ret = cn_printf(cn, "!"); | |
145 | } | |
146 | ||
923bed03 ON |
147 | for (; cur < cn->used; ++cur) { |
148 | if (cn->corename[cur] == '/') | |
149 | cn->corename[cur] = '!'; | |
150 | } | |
151 | return ret; | |
10c28d93 AK |
152 | } |
153 | ||
f38c85f1 | 154 | static int cn_print_exe_file(struct core_name *cn, bool name_only) |
10c28d93 AK |
155 | { |
156 | struct file *exe_file; | |
f38c85f1 | 157 | char *pathbuf, *path, *ptr; |
10c28d93 AK |
158 | int ret; |
159 | ||
160 | exe_file = get_mm_exe_file(current->mm); | |
923bed03 ON |
161 | if (!exe_file) |
162 | return cn_esc_printf(cn, "%s (path unknown)", current->comm); | |
10c28d93 | 163 | |
0ee931c4 | 164 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
10c28d93 AK |
165 | if (!pathbuf) { |
166 | ret = -ENOMEM; | |
167 | goto put_exe_file; | |
168 | } | |
169 | ||
9bf39ab2 | 170 | path = file_path(exe_file, pathbuf, PATH_MAX); |
10c28d93 AK |
171 | if (IS_ERR(path)) { |
172 | ret = PTR_ERR(path); | |
173 | goto free_buf; | |
174 | } | |
175 | ||
f38c85f1 LW |
176 | if (name_only) { |
177 | ptr = strrchr(path, '/'); | |
178 | if (ptr) | |
179 | path = ptr + 1; | |
180 | } | |
923bed03 | 181 | ret = cn_esc_printf(cn, "%s", path); |
10c28d93 AK |
182 | |
183 | free_buf: | |
184 | kfree(pathbuf); | |
185 | put_exe_file: | |
186 | fput(exe_file); | |
187 | return ret; | |
188 | } | |
189 | ||
190 | /* format_corename will inspect the pattern parameter, and output a | |
191 | * name into corename, which must have space for at least | |
192 | * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator. | |
193 | */ | |
315c6926 PW |
194 | static int format_corename(struct core_name *cn, struct coredump_params *cprm, |
195 | size_t **argv, int *argc) | |
10c28d93 AK |
196 | { |
197 | const struct cred *cred = current_cred(); | |
198 | const char *pat_ptr = core_pattern; | |
199 | int ispipe = (*pat_ptr == '|'); | |
315c6926 | 200 | bool was_space = false; |
10c28d93 AK |
201 | int pid_in_pattern = 0; |
202 | int err = 0; | |
203 | ||
e7fd1549 | 204 | cn->used = 0; |
3ceadcf6 ON |
205 | cn->corename = NULL; |
206 | if (expand_corename(cn, core_name_size)) | |
10c28d93 | 207 | return -ENOMEM; |
888ffc59 ON |
208 | cn->corename[0] = '\0'; |
209 | ||
315c6926 PW |
210 | if (ispipe) { |
211 | int argvs = sizeof(core_pattern) / 2; | |
212 | (*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL); | |
213 | if (!(*argv)) | |
214 | return -ENOMEM; | |
215 | (*argv)[(*argc)++] = 0; | |
888ffc59 | 216 | ++pat_ptr; |
db973a72 SM |
217 | if (!(*pat_ptr)) |
218 | return -ENOMEM; | |
315c6926 | 219 | } |
10c28d93 AK |
220 | |
221 | /* Repeat as long as we have more pattern to process and more output | |
222 | space */ | |
223 | while (*pat_ptr) { | |
315c6926 PW |
224 | /* |
225 | * Split on spaces before doing template expansion so that | |
226 | * %e and %E don't get split if they have spaces in them | |
227 | */ | |
228 | if (ispipe) { | |
229 | if (isspace(*pat_ptr)) { | |
2bf509d9 MD |
230 | if (cn->used != 0) |
231 | was_space = true; | |
315c6926 PW |
232 | pat_ptr++; |
233 | continue; | |
234 | } else if (was_space) { | |
235 | was_space = false; | |
236 | err = cn_printf(cn, "%c", '\0'); | |
237 | if (err) | |
238 | return err; | |
239 | (*argv)[(*argc)++] = cn->used; | |
240 | } | |
241 | } | |
10c28d93 | 242 | if (*pat_ptr != '%') { |
10c28d93 AK |
243 | err = cn_printf(cn, "%c", *pat_ptr++); |
244 | } else { | |
245 | switch (*++pat_ptr) { | |
246 | /* single % at the end, drop that */ | |
247 | case 0: | |
248 | goto out; | |
249 | /* Double percent, output one percent */ | |
250 | case '%': | |
251 | err = cn_printf(cn, "%c", '%'); | |
252 | break; | |
253 | /* pid */ | |
254 | case 'p': | |
255 | pid_in_pattern = 1; | |
256 | err = cn_printf(cn, "%d", | |
257 | task_tgid_vnr(current)); | |
258 | break; | |
65aafb1e SG |
259 | /* global pid */ |
260 | case 'P': | |
261 | err = cn_printf(cn, "%d", | |
262 | task_tgid_nr(current)); | |
263 | break; | |
b03023ec ON |
264 | case 'i': |
265 | err = cn_printf(cn, "%d", | |
266 | task_pid_vnr(current)); | |
267 | break; | |
268 | case 'I': | |
269 | err = cn_printf(cn, "%d", | |
270 | task_pid_nr(current)); | |
271 | break; | |
10c28d93 AK |
272 | /* uid */ |
273 | case 'u': | |
5202efe5 NI |
274 | err = cn_printf(cn, "%u", |
275 | from_kuid(&init_user_ns, | |
276 | cred->uid)); | |
10c28d93 AK |
277 | break; |
278 | /* gid */ | |
279 | case 'g': | |
5202efe5 NI |
280 | err = cn_printf(cn, "%u", |
281 | from_kgid(&init_user_ns, | |
282 | cred->gid)); | |
10c28d93 | 283 | break; |
12a2b4b2 ON |
284 | case 'd': |
285 | err = cn_printf(cn, "%d", | |
286 | __get_dumpable(cprm->mm_flags)); | |
287 | break; | |
10c28d93 AK |
288 | /* signal that caused the coredump */ |
289 | case 's': | |
b4176b7c NI |
290 | err = cn_printf(cn, "%d", |
291 | cprm->siginfo->si_signo); | |
10c28d93 AK |
292 | break; |
293 | /* UNIX time of coredump */ | |
294 | case 't': { | |
03927c8a AB |
295 | time64_t time; |
296 | ||
297 | time = ktime_get_real_seconds(); | |
298 | err = cn_printf(cn, "%lld", time); | |
10c28d93 AK |
299 | break; |
300 | } | |
301 | /* hostname */ | |
923bed03 | 302 | case 'h': |
10c28d93 | 303 | down_read(&uts_sem); |
923bed03 | 304 | err = cn_esc_printf(cn, "%s", |
10c28d93 AK |
305 | utsname()->nodename); |
306 | up_read(&uts_sem); | |
10c28d93 | 307 | break; |
f38c85f1 | 308 | /* executable, could be changed by prctl PR_SET_NAME etc */ |
923bed03 ON |
309 | case 'e': |
310 | err = cn_esc_printf(cn, "%s", current->comm); | |
10c28d93 | 311 | break; |
f38c85f1 LW |
312 | /* file name of executable */ |
313 | case 'f': | |
314 | err = cn_print_exe_file(cn, true); | |
315 | break; | |
10c28d93 | 316 | case 'E': |
f38c85f1 | 317 | err = cn_print_exe_file(cn, false); |
10c28d93 AK |
318 | break; |
319 | /* core limit size */ | |
320 | case 'c': | |
321 | err = cn_printf(cn, "%lu", | |
322 | rlimit(RLIMIT_CORE)); | |
323 | break; | |
324 | default: | |
325 | break; | |
326 | } | |
327 | ++pat_ptr; | |
328 | } | |
329 | ||
330 | if (err) | |
331 | return err; | |
332 | } | |
333 | ||
888ffc59 | 334 | out: |
10c28d93 AK |
335 | /* Backward compatibility with core_uses_pid: |
336 | * | |
337 | * If core_pattern does not include a %p (as is the default) | |
338 | * and core_uses_pid is set, then .%pid will be appended to | |
339 | * the filename. Do not do this for piped commands. */ | |
340 | if (!ispipe && !pid_in_pattern && core_uses_pid) { | |
341 | err = cn_printf(cn, ".%d", task_tgid_vnr(current)); | |
342 | if (err) | |
343 | return err; | |
344 | } | |
10c28d93 AK |
345 | return ispipe; |
346 | } | |
347 | ||
752dc970 | 348 | static int zap_process(struct task_struct *start, int exit_code) |
10c28d93 AK |
349 | { |
350 | struct task_struct *t; | |
351 | int nr = 0; | |
352 | ||
5fa534c9 | 353 | /* ignore all signals except SIGKILL, see prepare_signal() */ |
2f824d4d | 354 | start->signal->flags = SIGNAL_GROUP_EXIT; |
10c28d93 AK |
355 | start->signal->group_exit_code = exit_code; |
356 | start->signal->group_stop_count = 0; | |
357 | ||
d61ba589 | 358 | for_each_thread(start, t) { |
10c28d93 | 359 | task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK); |
92307383 | 360 | if (t != current && !(t->flags & PF_POSTCOREDUMP)) { |
10c28d93 AK |
361 | sigaddset(&t->pending.signal, SIGKILL); |
362 | signal_wake_up(t, 1); | |
363 | nr++; | |
364 | } | |
d61ba589 | 365 | } |
10c28d93 AK |
366 | |
367 | return nr; | |
368 | } | |
369 | ||
0258b5fd | 370 | static int zap_threads(struct task_struct *tsk, |
403bad72 | 371 | struct core_state *core_state, int exit_code) |
10c28d93 | 372 | { |
49697335 | 373 | struct signal_struct *signal = tsk->signal; |
10c28d93 AK |
374 | int nr = -EAGAIN; |
375 | ||
376 | spin_lock_irq(&tsk->sighand->siglock); | |
49697335 EB |
377 | if (!(signal->flags & SIGNAL_GROUP_EXIT) && !signal->group_exec_task) { |
378 | signal->core_state = core_state; | |
752dc970 | 379 | nr = zap_process(tsk, exit_code); |
403bad72 | 380 | clear_tsk_thread_flag(tsk, TIF_SIGPENDING); |
0258b5fd EB |
381 | tsk->flags |= PF_DUMPCORE; |
382 | atomic_set(&core_state->nr_threads, nr); | |
10c28d93 AK |
383 | } |
384 | spin_unlock_irq(&tsk->sighand->siglock); | |
10c28d93 AK |
385 | return nr; |
386 | } | |
387 | ||
388 | static int coredump_wait(int exit_code, struct core_state *core_state) | |
389 | { | |
390 | struct task_struct *tsk = current; | |
10c28d93 AK |
391 | int core_waiters = -EBUSY; |
392 | ||
393 | init_completion(&core_state->startup); | |
394 | core_state->dumper.task = tsk; | |
395 | core_state->dumper.next = NULL; | |
396 | ||
0258b5fd | 397 | core_waiters = zap_threads(tsk, core_state, exit_code); |
10c28d93 AK |
398 | if (core_waiters > 0) { |
399 | struct core_thread *ptr; | |
400 | ||
70d78fe7 | 401 | freezer_do_not_count(); |
10c28d93 | 402 | wait_for_completion(&core_state->startup); |
70d78fe7 | 403 | freezer_count(); |
10c28d93 AK |
404 | /* |
405 | * Wait for all the threads to become inactive, so that | |
406 | * all the thread context (extended register state, like | |
407 | * fpu etc) gets copied to the memory. | |
408 | */ | |
409 | ptr = core_state->dumper.next; | |
410 | while (ptr != NULL) { | |
411 | wait_task_inactive(ptr->task, 0); | |
412 | ptr = ptr->next; | |
413 | } | |
414 | } | |
415 | ||
416 | return core_waiters; | |
417 | } | |
418 | ||
0258b5fd | 419 | static void coredump_finish(bool core_dumped) |
10c28d93 AK |
420 | { |
421 | struct core_thread *curr, *next; | |
422 | struct task_struct *task; | |
423 | ||
6cd8f0ac | 424 | spin_lock_irq(¤t->sighand->siglock); |
acdedd99 ON |
425 | if (core_dumped && !__fatal_signal_pending(current)) |
426 | current->signal->group_exit_code |= 0x80; | |
0258b5fd EB |
427 | next = current->signal->core_state->dumper.next; |
428 | current->signal->core_state = NULL; | |
6cd8f0ac ON |
429 | spin_unlock_irq(¤t->sighand->siglock); |
430 | ||
10c28d93 AK |
431 | while ((curr = next) != NULL) { |
432 | next = curr->next; | |
433 | task = curr->task; | |
434 | /* | |
92307383 | 435 | * see coredump_task_exit(), curr->task must not see |
10c28d93 AK |
436 | * ->task == NULL before we read ->next. |
437 | */ | |
438 | smp_mb(); | |
439 | curr->task = NULL; | |
440 | wake_up_process(task); | |
441 | } | |
10c28d93 AK |
442 | } |
443 | ||
528f827e ON |
444 | static bool dump_interrupted(void) |
445 | { | |
446 | /* | |
447 | * SIGKILL or freezing() interrupt the coredumping. Perhaps we | |
448 | * can do try_to_freeze() and check __fatal_signal_pending(), | |
449 | * but then we need to teach dump_write() to restart and clear | |
450 | * TIF_SIGPENDING. | |
451 | */ | |
06af8679 | 452 | return fatal_signal_pending(current) || freezing(current); |
528f827e ON |
453 | } |
454 | ||
10c28d93 AK |
455 | static void wait_for_dump_helpers(struct file *file) |
456 | { | |
de32ec4c | 457 | struct pipe_inode_info *pipe = file->private_data; |
10c28d93 AK |
458 | |
459 | pipe_lock(pipe); | |
460 | pipe->readers++; | |
461 | pipe->writers--; | |
0ddad21d | 462 | wake_up_interruptible_sync(&pipe->rd_wait); |
dc7ee2aa ON |
463 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
464 | pipe_unlock(pipe); | |
10c28d93 | 465 | |
dc7ee2aa ON |
466 | /* |
467 | * We actually want wait_event_freezable() but then we need | |
468 | * to clear TIF_SIGPENDING and improve dump_interrupted(). | |
469 | */ | |
0ddad21d | 470 | wait_event_interruptible(pipe->rd_wait, pipe->readers == 1); |
10c28d93 | 471 | |
dc7ee2aa | 472 | pipe_lock(pipe); |
10c28d93 AK |
473 | pipe->readers--; |
474 | pipe->writers++; | |
475 | pipe_unlock(pipe); | |
10c28d93 AK |
476 | } |
477 | ||
478 | /* | |
479 | * umh_pipe_setup | |
480 | * helper function to customize the process used | |
481 | * to collect the core in userspace. Specifically | |
482 | * it sets up a pipe and installs it as fd 0 (stdin) | |
483 | * for the process. Returns 0 on success, or | |
484 | * PTR_ERR on failure. | |
485 | * Note that it also sets the core limit to 1. This | |
486 | * is a special value that we use to trap recursive | |
487 | * core dumps | |
488 | */ | |
489 | static int umh_pipe_setup(struct subprocess_info *info, struct cred *new) | |
490 | { | |
491 | struct file *files[2]; | |
492 | struct coredump_params *cp = (struct coredump_params *)info->data; | |
493 | int err = create_pipe_files(files, 0); | |
494 | if (err) | |
495 | return err; | |
496 | ||
497 | cp->file = files[1]; | |
498 | ||
45525b26 AV |
499 | err = replace_fd(0, files[0], 0); |
500 | fput(files[0]); | |
10c28d93 AK |
501 | /* and disallow core files too */ |
502 | current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1}; | |
503 | ||
45525b26 | 504 | return err; |
10c28d93 AK |
505 | } |
506 | ||
ae7795bc | 507 | void do_coredump(const kernel_siginfo_t *siginfo) |
10c28d93 AK |
508 | { |
509 | struct core_state core_state; | |
510 | struct core_name cn; | |
511 | struct mm_struct *mm = current->mm; | |
512 | struct linux_binfmt * binfmt; | |
513 | const struct cred *old_cred; | |
514 | struct cred *cred; | |
515 | int retval = 0; | |
10c28d93 | 516 | int ispipe; |
315c6926 PW |
517 | size_t *argv = NULL; |
518 | int argc = 0; | |
fbb18169 JH |
519 | /* require nonrelative corefile path and be extra careful */ |
520 | bool need_suid_safe = false; | |
acdedd99 | 521 | bool core_dumped = false; |
10c28d93 AK |
522 | static atomic_t core_dump_count = ATOMIC_INIT(0); |
523 | struct coredump_params cprm = { | |
5ab1c309 | 524 | .siginfo = siginfo, |
541880d9 | 525 | .regs = signal_pt_regs(), |
10c28d93 AK |
526 | .limit = rlimit(RLIMIT_CORE), |
527 | /* | |
528 | * We must use the same mm->flags while dumping core to avoid | |
529 | * inconsistency of bit flags, since this flag is not protected | |
530 | * by any locks. | |
531 | */ | |
532 | .mm_flags = mm->flags, | |
533 | }; | |
534 | ||
5ab1c309 | 535 | audit_core_dumps(siginfo->si_signo); |
10c28d93 AK |
536 | |
537 | binfmt = mm->binfmt; | |
538 | if (!binfmt || !binfmt->core_dump) | |
539 | goto fail; | |
540 | if (!__get_dumpable(cprm.mm_flags)) | |
541 | goto fail; | |
542 | ||
543 | cred = prepare_creds(); | |
544 | if (!cred) | |
545 | goto fail; | |
546 | /* | |
547 | * We cannot trust fsuid as being the "true" uid of the process | |
548 | * nor do we know its entire history. We only know it was tainted | |
549 | * so we dump it as root in mode 2, and only into a controlled | |
550 | * environment (pipe handler or fully qualified path). | |
551 | */ | |
e579d2c2 | 552 | if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) { |
10c28d93 | 553 | /* Setuid core dump mode */ |
10c28d93 | 554 | cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */ |
fbb18169 | 555 | need_suid_safe = true; |
10c28d93 AK |
556 | } |
557 | ||
5ab1c309 | 558 | retval = coredump_wait(siginfo->si_signo, &core_state); |
10c28d93 AK |
559 | if (retval < 0) |
560 | goto fail_creds; | |
561 | ||
562 | old_cred = override_creds(cred); | |
563 | ||
315c6926 | 564 | ispipe = format_corename(&cn, &cprm, &argv, &argc); |
10c28d93 | 565 | |
fb96c475 | 566 | if (ispipe) { |
315c6926 | 567 | int argi; |
10c28d93 AK |
568 | int dump_count; |
569 | char **helper_argv; | |
907ed132 | 570 | struct subprocess_info *sub_info; |
10c28d93 AK |
571 | |
572 | if (ispipe < 0) { | |
573 | printk(KERN_WARNING "format_corename failed\n"); | |
574 | printk(KERN_WARNING "Aborting core\n"); | |
e7fd1549 | 575 | goto fail_unlock; |
10c28d93 AK |
576 | } |
577 | ||
578 | if (cprm.limit == 1) { | |
579 | /* See umh_pipe_setup() which sets RLIMIT_CORE = 1. | |
580 | * | |
581 | * Normally core limits are irrelevant to pipes, since | |
582 | * we're not writing to the file system, but we use | |
fcbc32bc | 583 | * cprm.limit of 1 here as a special value, this is a |
10c28d93 AK |
584 | * consistent way to catch recursive crashes. |
585 | * We can still crash if the core_pattern binary sets | |
586 | * RLIM_CORE = !1, but it runs as root, and can do | |
587 | * lots of stupid things. | |
588 | * | |
589 | * Note that we use task_tgid_vnr here to grab the pid | |
590 | * of the process group leader. That way we get the | |
591 | * right pid if a thread in a multi-threaded | |
592 | * core_pattern process dies. | |
593 | */ | |
594 | printk(KERN_WARNING | |
595 | "Process %d(%s) has RLIMIT_CORE set to 1\n", | |
596 | task_tgid_vnr(current), current->comm); | |
597 | printk(KERN_WARNING "Aborting core\n"); | |
598 | goto fail_unlock; | |
599 | } | |
600 | cprm.limit = RLIM_INFINITY; | |
601 | ||
602 | dump_count = atomic_inc_return(&core_dump_count); | |
603 | if (core_pipe_limit && (core_pipe_limit < dump_count)) { | |
604 | printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n", | |
605 | task_tgid_vnr(current), current->comm); | |
606 | printk(KERN_WARNING "Skipping core dump\n"); | |
607 | goto fail_dropcount; | |
608 | } | |
609 | ||
315c6926 PW |
610 | helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv), |
611 | GFP_KERNEL); | |
10c28d93 AK |
612 | if (!helper_argv) { |
613 | printk(KERN_WARNING "%s failed to allocate memory\n", | |
614 | __func__); | |
615 | goto fail_dropcount; | |
616 | } | |
315c6926 PW |
617 | for (argi = 0; argi < argc; argi++) |
618 | helper_argv[argi] = cn.corename + argv[argi]; | |
619 | helper_argv[argi] = NULL; | |
10c28d93 | 620 | |
907ed132 LDM |
621 | retval = -ENOMEM; |
622 | sub_info = call_usermodehelper_setup(helper_argv[0], | |
623 | helper_argv, NULL, GFP_KERNEL, | |
624 | umh_pipe_setup, NULL, &cprm); | |
625 | if (sub_info) | |
626 | retval = call_usermodehelper_exec(sub_info, | |
627 | UMH_WAIT_EXEC); | |
628 | ||
315c6926 | 629 | kfree(helper_argv); |
10c28d93 | 630 | if (retval) { |
888ffc59 | 631 | printk(KERN_INFO "Core dump to |%s pipe failed\n", |
10c28d93 AK |
632 | cn.corename); |
633 | goto close_fail; | |
fb96c475 | 634 | } |
10c28d93 | 635 | } else { |
643fe55a | 636 | struct user_namespace *mnt_userns; |
10c28d93 | 637 | struct inode *inode; |
378c6520 JH |
638 | int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW | |
639 | O_LARGEFILE | O_EXCL; | |
10c28d93 AK |
640 | |
641 | if (cprm.limit < binfmt->min_coredump) | |
642 | goto fail_unlock; | |
643 | ||
fbb18169 | 644 | if (need_suid_safe && cn.corename[0] != '/') { |
10c28d93 AK |
645 | printk(KERN_WARNING "Pid %d(%s) can only dump core "\ |
646 | "to fully qualified path!\n", | |
647 | task_tgid_vnr(current), current->comm); | |
648 | printk(KERN_WARNING "Skipping core dump\n"); | |
649 | goto fail_unlock; | |
650 | } | |
651 | ||
fbb18169 JH |
652 | /* |
653 | * Unlink the file if it exists unless this is a SUID | |
654 | * binary - in that case, we're running around with root | |
655 | * privs and don't want to unlink another user's coredump. | |
656 | */ | |
657 | if (!need_suid_safe) { | |
fbb18169 JH |
658 | /* |
659 | * If it doesn't exist, that's fine. If there's some | |
660 | * other problem, we'll catch it at the filp_open(). | |
661 | */ | |
96271654 | 662 | do_unlinkat(AT_FDCWD, getname_kernel(cn.corename)); |
fbb18169 JH |
663 | } |
664 | ||
665 | /* | |
666 | * There is a race between unlinking and creating the | |
667 | * file, but if that causes an EEXIST here, that's | |
668 | * fine - another process raced with us while creating | |
669 | * the corefile, and the other process won. To userspace, | |
670 | * what matters is that at least one of the two processes | |
671 | * writes its coredump successfully, not which one. | |
672 | */ | |
378c6520 JH |
673 | if (need_suid_safe) { |
674 | /* | |
675 | * Using user namespaces, normal user tasks can change | |
676 | * their current->fs->root to point to arbitrary | |
677 | * directories. Since the intention of the "only dump | |
678 | * with a fully qualified path" rule is to control where | |
679 | * coredumps may be placed using root privileges, | |
680 | * current->fs->root must not be used. Instead, use the | |
681 | * root directory of init_task. | |
682 | */ | |
683 | struct path root; | |
684 | ||
685 | task_lock(&init_task); | |
686 | get_fs_root(init_task.fs, &root); | |
687 | task_unlock(&init_task); | |
ffb37ca3 AV |
688 | cprm.file = file_open_root(&root, cn.corename, |
689 | open_flags, 0600); | |
378c6520 JH |
690 | path_put(&root); |
691 | } else { | |
692 | cprm.file = filp_open(cn.corename, open_flags, 0600); | |
693 | } | |
10c28d93 AK |
694 | if (IS_ERR(cprm.file)) |
695 | goto fail_unlock; | |
696 | ||
496ad9aa | 697 | inode = file_inode(cprm.file); |
10c28d93 AK |
698 | if (inode->i_nlink > 1) |
699 | goto close_fail; | |
700 | if (d_unhashed(cprm.file->f_path.dentry)) | |
701 | goto close_fail; | |
702 | /* | |
703 | * AK: actually i see no reason to not allow this for named | |
704 | * pipes etc, but keep the previous behaviour for now. | |
705 | */ | |
706 | if (!S_ISREG(inode->i_mode)) | |
707 | goto close_fail; | |
708 | /* | |
40f705a7 JH |
709 | * Don't dump core if the filesystem changed owner or mode |
710 | * of the file during file creation. This is an issue when | |
711 | * a process dumps core while its cwd is e.g. on a vfat | |
712 | * filesystem. | |
10c28d93 | 713 | */ |
643fe55a | 714 | mnt_userns = file_mnt_user_ns(cprm.file); |
dbd9d6f8 DO |
715 | if (!uid_eq(i_uid_into_mnt(mnt_userns, inode), |
716 | current_fsuid())) { | |
717 | pr_info_ratelimited("Core dump to %s aborted: cannot preserve file owner\n", | |
718 | cn.corename); | |
10c28d93 | 719 | goto close_fail; |
dbd9d6f8 DO |
720 | } |
721 | if ((inode->i_mode & 0677) != 0600) { | |
722 | pr_info_ratelimited("Core dump to %s aborted: cannot preserve file permissions\n", | |
723 | cn.corename); | |
40f705a7 | 724 | goto close_fail; |
dbd9d6f8 | 725 | } |
86cc0584 | 726 | if (!(cprm.file->f_mode & FMODE_CAN_WRITE)) |
10c28d93 | 727 | goto close_fail; |
643fe55a CB |
728 | if (do_truncate(mnt_userns, cprm.file->f_path.dentry, |
729 | 0, 0, cprm.file)) | |
10c28d93 AK |
730 | goto close_fail; |
731 | } | |
732 | ||
733 | /* get us an unshared descriptor table; almost always a no-op */ | |
c39ab6de | 734 | /* The cell spufs coredump code reads the file descriptor tables */ |
1f702603 | 735 | retval = unshare_files(); |
10c28d93 AK |
736 | if (retval) |
737 | goto close_fail; | |
e86d35c3 | 738 | if (!dump_interrupted()) { |
3740d93e LC |
739 | /* |
740 | * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would | |
741 | * have this set to NULL. | |
742 | */ | |
743 | if (!cprm.file) { | |
744 | pr_info("Core dump to |%s disabled\n", cn.corename); | |
745 | goto close_fail; | |
746 | } | |
e86d35c3 AV |
747 | file_start_write(cprm.file); |
748 | core_dumped = binfmt->core_dump(&cprm); | |
d0f1088b AV |
749 | /* |
750 | * Ensures that file size is big enough to contain the current | |
751 | * file postion. This prevents gdb from complaining about | |
752 | * a truncated file if the last "write" to the file was | |
753 | * dump_skip. | |
754 | */ | |
755 | if (cprm.to_skip) { | |
756 | cprm.to_skip--; | |
757 | dump_emit(&cprm, "", 1); | |
758 | } | |
e86d35c3 AV |
759 | file_end_write(cprm.file); |
760 | } | |
10c28d93 AK |
761 | if (ispipe && core_pipe_limit) |
762 | wait_for_dump_helpers(cprm.file); | |
763 | close_fail: | |
764 | if (cprm.file) | |
765 | filp_close(cprm.file, NULL); | |
766 | fail_dropcount: | |
767 | if (ispipe) | |
768 | atomic_dec(&core_dump_count); | |
769 | fail_unlock: | |
315c6926 | 770 | kfree(argv); |
10c28d93 | 771 | kfree(cn.corename); |
0258b5fd | 772 | coredump_finish(core_dumped); |
10c28d93 AK |
773 | revert_creds(old_cred); |
774 | fail_creds: | |
775 | put_cred(cred); | |
776 | fail: | |
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 | */ | |
d0f1088b | 785 | static int __dump_emit(struct coredump_params *cprm, const void *addr, int nr) |
ecc8c772 AV |
786 | { |
787 | struct file *file = cprm->file; | |
2507a4fb AV |
788 | loff_t pos = file->f_pos; |
789 | ssize_t n; | |
2c4cb043 | 790 | if (cprm->written + nr > cprm->limit) |
ecc8c772 | 791 | return 0; |
df0c09c0 JH |
792 | |
793 | ||
794 | if (dump_interrupted()) | |
795 | return 0; | |
796 | n = __kernel_write(file, addr, nr, &pos); | |
797 | if (n != nr) | |
798 | return 0; | |
799 | file->f_pos = pos; | |
800 | cprm->written += n; | |
801 | cprm->pos += n; | |
802 | ||
ecc8c772 AV |
803 | return 1; |
804 | } | |
ecc8c772 | 805 | |
d0f1088b | 806 | static int __dump_skip(struct coredump_params *cprm, size_t nr) |
10c28d93 | 807 | { |
9b56d543 AV |
808 | static char zeroes[PAGE_SIZE]; |
809 | struct file *file = cprm->file; | |
10c28d93 | 810 | if (file->f_op->llseek && file->f_op->llseek != no_llseek) { |
528f827e | 811 | if (dump_interrupted() || |
9b56d543 | 812 | file->f_op->llseek(file, nr, SEEK_CUR) < 0) |
10c28d93 | 813 | return 0; |
1607f09c | 814 | cprm->pos += nr; |
9b56d543 | 815 | return 1; |
10c28d93 | 816 | } else { |
9b56d543 | 817 | while (nr > PAGE_SIZE) { |
d0f1088b | 818 | if (!__dump_emit(cprm, zeroes, PAGE_SIZE)) |
9b56d543 AV |
819 | return 0; |
820 | nr -= PAGE_SIZE; | |
10c28d93 | 821 | } |
d0f1088b | 822 | return __dump_emit(cprm, zeroes, nr); |
10c28d93 | 823 | } |
10c28d93 | 824 | } |
d0f1088b AV |
825 | |
826 | int dump_emit(struct coredump_params *cprm, const void *addr, int nr) | |
827 | { | |
828 | if (cprm->to_skip) { | |
829 | if (!__dump_skip(cprm, cprm->to_skip)) | |
830 | return 0; | |
831 | cprm->to_skip = 0; | |
832 | } | |
833 | return __dump_emit(cprm, addr, nr); | |
834 | } | |
835 | EXPORT_SYMBOL(dump_emit); | |
836 | ||
837 | void dump_skip_to(struct coredump_params *cprm, unsigned long pos) | |
838 | { | |
839 | cprm->to_skip = pos - cprm->pos; | |
840 | } | |
841 | EXPORT_SYMBOL(dump_skip_to); | |
842 | ||
843 | void dump_skip(struct coredump_params *cprm, size_t nr) | |
844 | { | |
845 | cprm->to_skip += nr; | |
846 | } | |
9b56d543 | 847 | EXPORT_SYMBOL(dump_skip); |
22a8cb82 | 848 | |
afc63a97 JH |
849 | #ifdef CONFIG_ELF_CORE |
850 | int dump_user_range(struct coredump_params *cprm, unsigned long start, | |
851 | unsigned long len) | |
852 | { | |
853 | unsigned long addr; | |
854 | ||
855 | for (addr = start; addr < start + len; addr += PAGE_SIZE) { | |
856 | struct page *page; | |
857 | int stop; | |
858 | ||
859 | /* | |
860 | * To avoid having to allocate page tables for virtual address | |
861 | * ranges that have never been used yet, and also to make it | |
862 | * easy to generate sparse core files, use a helper that returns | |
863 | * NULL when encountering an empty page table entry that would | |
864 | * otherwise have been filled with the zero page. | |
865 | */ | |
866 | page = get_dump_page(addr); | |
867 | if (page) { | |
3159ed57 | 868 | void *kaddr = kmap_local_page(page); |
afc63a97 JH |
869 | |
870 | stop = !dump_emit(cprm, kaddr, PAGE_SIZE); | |
3159ed57 | 871 | kunmap_local(kaddr); |
afc63a97 | 872 | put_page(page); |
d0f1088b AV |
873 | if (stop) |
874 | return 0; | |
afc63a97 | 875 | } else { |
d0f1088b | 876 | dump_skip(cprm, PAGE_SIZE); |
afc63a97 | 877 | } |
afc63a97 JH |
878 | } |
879 | return 1; | |
880 | } | |
881 | #endif | |
882 | ||
22a8cb82 AV |
883 | int dump_align(struct coredump_params *cprm, int align) |
884 | { | |
d0f1088b | 885 | unsigned mod = (cprm->pos + cprm->to_skip) & (align - 1); |
22a8cb82 | 886 | if (align & (align - 1)) |
db51242d | 887 | return 0; |
d0f1088b AV |
888 | if (mod) |
889 | cprm->to_skip += align - mod; | |
890 | return 1; | |
22a8cb82 AV |
891 | } |
892 | EXPORT_SYMBOL(dump_align); | |
4d22c75d | 893 | |
f0bc21b2 XN |
894 | #ifdef CONFIG_SYSCTL |
895 | ||
896 | void validate_coredump_safety(void) | |
897 | { | |
898 | if (suid_dumpable == SUID_DUMP_ROOT && | |
899 | core_pattern[0] != '/' && core_pattern[0] != '|') { | |
900 | pr_warn( | |
901 | "Unsafe core_pattern used with fs.suid_dumpable=2.\n" | |
902 | "Pipe handler or fully qualified core dump path required.\n" | |
903 | "Set kernel.core_pattern before fs.suid_dumpable.\n" | |
904 | ); | |
905 | } | |
906 | } | |
907 | ||
908 | static int proc_dostring_coredump(struct ctl_table *table, int write, | |
909 | void *buffer, size_t *lenp, loff_t *ppos) | |
910 | { | |
911 | int error = proc_dostring(table, write, buffer, lenp, ppos); | |
912 | ||
913 | if (!error) | |
914 | validate_coredump_safety(); | |
915 | return error; | |
916 | } | |
917 | ||
918 | static struct ctl_table coredump_sysctls[] = { | |
919 | { | |
920 | .procname = "core_uses_pid", | |
921 | .data = &core_uses_pid, | |
922 | .maxlen = sizeof(int), | |
923 | .mode = 0644, | |
924 | .proc_handler = proc_dointvec, | |
925 | }, | |
926 | { | |
927 | .procname = "core_pattern", | |
928 | .data = core_pattern, | |
929 | .maxlen = CORENAME_MAX_SIZE, | |
930 | .mode = 0644, | |
931 | .proc_handler = proc_dostring_coredump, | |
932 | }, | |
933 | { | |
934 | .procname = "core_pipe_limit", | |
935 | .data = &core_pipe_limit, | |
936 | .maxlen = sizeof(unsigned int), | |
937 | .mode = 0644, | |
938 | .proc_handler = proc_dointvec, | |
939 | }, | |
940 | { } | |
941 | }; | |
942 | ||
943 | static int __init init_fs_coredump_sysctls(void) | |
944 | { | |
945 | register_sysctl_init("kernel", coredump_sysctls); | |
946 | return 0; | |
947 | } | |
948 | fs_initcall(init_fs_coredump_sysctls); | |
949 | #endif /* CONFIG_SYSCTL */ | |
950 | ||
429a22e7 JH |
951 | /* |
952 | * The purpose of always_dump_vma() is to make sure that special kernel mappings | |
953 | * that are useful for post-mortem analysis are included in every core dump. | |
954 | * In that way we ensure that the core dump is fully interpretable later | |
955 | * without matching up the same kernel and hardware config to see what PC values | |
956 | * meant. These special mappings include - vDSO, vsyscall, and other | |
957 | * architecture specific mappings | |
958 | */ | |
959 | static bool always_dump_vma(struct vm_area_struct *vma) | |
960 | { | |
961 | /* Any vsyscall mappings? */ | |
962 | if (vma == get_gate_vma(vma->vm_mm)) | |
963 | return true; | |
964 | ||
965 | /* | |
966 | * Assume that all vmas with a .name op should always be dumped. | |
967 | * If this changes, a new vm_ops field can easily be added. | |
968 | */ | |
969 | if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma)) | |
970 | return true; | |
971 | ||
972 | /* | |
973 | * arch_vma_name() returns non-NULL for special architecture mappings, | |
974 | * such as vDSO sections. | |
975 | */ | |
976 | if (arch_vma_name(vma)) | |
977 | return true; | |
978 | ||
979 | return false; | |
980 | } | |
981 | ||
982 | /* | |
983 | * Decide how much of @vma's contents should be included in a core dump. | |
984 | */ | |
a07279c9 JH |
985 | static unsigned long vma_dump_size(struct vm_area_struct *vma, |
986 | unsigned long mm_flags) | |
429a22e7 JH |
987 | { |
988 | #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type)) | |
989 | ||
990 | /* always dump the vdso and vsyscall sections */ | |
991 | if (always_dump_vma(vma)) | |
992 | goto whole; | |
993 | ||
994 | if (vma->vm_flags & VM_DONTDUMP) | |
995 | return 0; | |
996 | ||
997 | /* support for DAX */ | |
998 | if (vma_is_dax(vma)) { | |
999 | if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED)) | |
1000 | goto whole; | |
1001 | if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE)) | |
1002 | goto whole; | |
1003 | return 0; | |
1004 | } | |
1005 | ||
1006 | /* Hugetlb memory check */ | |
1007 | if (is_vm_hugetlb_page(vma)) { | |
1008 | if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED)) | |
1009 | goto whole; | |
1010 | if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE)) | |
1011 | goto whole; | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* Do not dump I/O mapped devices or special mappings */ | |
1016 | if (vma->vm_flags & VM_IO) | |
1017 | return 0; | |
1018 | ||
1019 | /* By default, dump shared memory if mapped from an anonymous file. */ | |
1020 | if (vma->vm_flags & VM_SHARED) { | |
1021 | if (file_inode(vma->vm_file)->i_nlink == 0 ? | |
1022 | FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED)) | |
1023 | goto whole; | |
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | /* Dump segments that have been written to. */ | |
1028 | if ((!IS_ENABLED(CONFIG_MMU) || vma->anon_vma) && FILTER(ANON_PRIVATE)) | |
1029 | goto whole; | |
1030 | if (vma->vm_file == NULL) | |
1031 | return 0; | |
1032 | ||
1033 | if (FILTER(MAPPED_PRIVATE)) | |
1034 | goto whole; | |
1035 | ||
1036 | /* | |
1037 | * If this is the beginning of an executable file mapping, | |
1038 | * dump the first page to aid in determining what was mapped here. | |
1039 | */ | |
1040 | if (FILTER(ELF_HEADERS) && | |
1041 | vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ) && | |
1042 | (READ_ONCE(file_inode(vma->vm_file)->i_mode) & 0111) != 0) | |
1043 | return PAGE_SIZE; | |
1044 | ||
1045 | #undef FILTER | |
1046 | ||
1047 | return 0; | |
1048 | ||
1049 | whole: | |
1050 | return vma->vm_end - vma->vm_start; | |
1051 | } | |
a07279c9 JH |
1052 | |
1053 | static struct vm_area_struct *first_vma(struct task_struct *tsk, | |
1054 | struct vm_area_struct *gate_vma) | |
1055 | { | |
1056 | struct vm_area_struct *ret = tsk->mm->mmap; | |
1057 | ||
1058 | if (ret) | |
1059 | return ret; | |
1060 | return gate_vma; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * Helper function for iterating across a vma list. It ensures that the caller | |
1065 | * will visit `gate_vma' prior to terminating the search. | |
1066 | */ | |
1067 | static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma, | |
1068 | struct vm_area_struct *gate_vma) | |
1069 | { | |
1070 | struct vm_area_struct *ret; | |
1071 | ||
1072 | ret = this_vma->vm_next; | |
1073 | if (ret) | |
1074 | return ret; | |
1075 | if (this_vma == gate_vma) | |
1076 | return NULL; | |
1077 | return gate_vma; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Under the mmap_lock, take a snapshot of relevant information about the task's | |
1082 | * VMAs. | |
1083 | */ | |
1084 | int dump_vma_snapshot(struct coredump_params *cprm, int *vma_count, | |
1085 | struct core_vma_metadata **vma_meta, | |
1086 | size_t *vma_data_size_ptr) | |
1087 | { | |
1088 | struct vm_area_struct *vma, *gate_vma; | |
1089 | struct mm_struct *mm = current->mm; | |
1090 | int i; | |
1091 | size_t vma_data_size = 0; | |
1092 | ||
1093 | /* | |
1094 | * Once the stack expansion code is fixed to not change VMA bounds | |
1095 | * under mmap_lock in read mode, this can be changed to take the | |
1096 | * mmap_lock in read mode. | |
1097 | */ | |
1098 | if (mmap_write_lock_killable(mm)) | |
1099 | return -EINTR; | |
1100 | ||
1101 | gate_vma = get_gate_vma(mm); | |
1102 | *vma_count = mm->map_count + (gate_vma ? 1 : 0); | |
1103 | ||
1104 | *vma_meta = kvmalloc_array(*vma_count, sizeof(**vma_meta), GFP_KERNEL); | |
1105 | if (!*vma_meta) { | |
1106 | mmap_write_unlock(mm); | |
1107 | return -ENOMEM; | |
1108 | } | |
1109 | ||
1110 | for (i = 0, vma = first_vma(current, gate_vma); vma != NULL; | |
1111 | vma = next_vma(vma, gate_vma), i++) { | |
1112 | struct core_vma_metadata *m = (*vma_meta) + i; | |
1113 | ||
1114 | m->start = vma->vm_start; | |
1115 | m->end = vma->vm_end; | |
1116 | m->flags = vma->vm_flags; | |
1117 | m->dump_size = vma_dump_size(vma, cprm->mm_flags); | |
1118 | ||
1119 | vma_data_size += m->dump_size; | |
1120 | } | |
1121 | ||
1122 | mmap_write_unlock(mm); | |
1123 | ||
6fcac87e Q |
1124 | if (WARN_ON(i != *vma_count)) { |
1125 | kvfree(*vma_meta); | |
a07279c9 | 1126 | return -EFAULT; |
6fcac87e | 1127 | } |
a07279c9 JH |
1128 | |
1129 | *vma_data_size_ptr = vma_data_size; | |
1130 | return 0; | |
1131 | } |