Commit | Line | Data |
---|---|---|
4b8b5e72 GB |
1 | /* Linux namespaces(7) support. |
2 | ||
42a4f53d | 3 | Copyright (C) 2015-2019 Free Software Foundation, Inc. |
4b8b5e72 GB |
4 | |
5 | This file is part of GDB. | |
6 | ||
7 | This program is free software; you can redistribute it and/or modify | |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 3 of the License, or | |
10 | (at your option) any later version. | |
11 | ||
12 | This program is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | GNU General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with this program. If not, see <http://www.gnu.org/licenses/>. */ | |
19 | ||
268a13a5 | 20 | #include "gdbsupport/common-defs.h" |
4b8b5e72 | 21 | #include "nat/linux-namespaces.h" |
268a13a5 | 22 | #include "gdbsupport/filestuff.h" |
4b8b5e72 GB |
23 | #include <fcntl.h> |
24 | #include <sys/syscall.h> | |
25 | #include <sys/types.h> | |
26 | #include <sys/stat.h> | |
27 | #include <sys/socket.h> | |
268a13a5 | 28 | #include "gdbsupport/gdb_wait.h" |
4b8b5e72 GB |
29 | #include <signal.h> |
30 | #include <sched.h> | |
268a13a5 | 31 | #include "gdbsupport/scope-exit.h" |
4b8b5e72 GB |
32 | |
33 | /* See nat/linux-namespaces.h. */ | |
491144b5 | 34 | bool debug_linux_namespaces; |
4b8b5e72 | 35 | |
eb0edac8 GB |
36 | /* Handle systems without fork. */ |
37 | ||
38 | static inline pid_t | |
39 | do_fork (void) | |
40 | { | |
41 | #ifdef HAVE_FORK | |
42 | return fork (); | |
43 | #else | |
44 | errno = ENOSYS; | |
45 | return -1; | |
46 | #endif | |
47 | } | |
48 | ||
4b8b5e72 GB |
49 | /* Handle systems without setns. */ |
50 | ||
99fe86f7 PB |
51 | static inline int |
52 | do_setns (int fd, int nstype) | |
4b8b5e72 | 53 | { |
99fe86f7 PB |
54 | #ifdef HAVE_SETNS |
55 | return setns (fd, nstype); | |
56 | #elif defined __NR_setns | |
4b8b5e72 GB |
57 | return syscall (__NR_setns, fd, nstype); |
58 | #else | |
59 | errno = ENOSYS; | |
60 | return -1; | |
61 | #endif | |
62 | } | |
4b8b5e72 | 63 | |
4da680ad ME |
64 | /* Handle systems without MSG_CMSG_CLOEXEC. */ |
65 | ||
66 | #ifndef MSG_CMSG_CLOEXEC | |
67 | #define MSG_CMSG_CLOEXEC 0 | |
68 | #endif | |
69 | ||
4b8b5e72 GB |
70 | /* A Linux namespace. */ |
71 | ||
72 | struct linux_ns | |
73 | { | |
74 | /* Filename of this namespace's entries in /proc/PID/ns. */ | |
75 | const char *filename; | |
76 | ||
77 | /* Nonzero if this object has been initialized. */ | |
78 | int initialized; | |
79 | ||
80 | /* Nonzero if this namespace is supported on this system. */ | |
81 | int supported; | |
82 | ||
83 | /* ID of the namespace the calling process is in, used to | |
84 | see if other processes share the namespace. The code in | |
85 | this file assumes that the calling process never changes | |
86 | namespace. */ | |
87 | ino_t id; | |
88 | }; | |
89 | ||
90 | /* Return the absolute filename of process PID's /proc/PID/ns | |
91 | entry for namespace NS. The returned value persists until | |
92 | this function is next called. */ | |
93 | ||
94 | static const char * | |
95 | linux_ns_filename (struct linux_ns *ns, int pid) | |
96 | { | |
97 | static char filename[PATH_MAX]; | |
98 | ||
99 | gdb_assert (pid > 0); | |
100 | xsnprintf (filename, sizeof (filename), "/proc/%d/ns/%s", pid, | |
101 | ns->filename); | |
102 | ||
103 | return filename; | |
104 | } | |
105 | ||
106 | /* Return a representation of the caller's TYPE namespace, or | |
107 | NULL if TYPE namespaces are not supported on this system. */ | |
108 | ||
109 | static struct linux_ns * | |
110 | linux_ns_get_namespace (enum linux_ns_type type) | |
111 | { | |
112 | static struct linux_ns namespaces[NUM_LINUX_NS_TYPES] = | |
113 | { | |
114 | { "ipc" }, | |
115 | { "mnt" }, | |
116 | { "net" }, | |
117 | { "pid" }, | |
118 | { "user" }, | |
119 | { "uts" }, | |
120 | }; | |
121 | struct linux_ns *ns; | |
122 | ||
123 | gdb_assert (type >= 0 && type < NUM_LINUX_NS_TYPES); | |
124 | ns = &namespaces[type]; | |
125 | ||
126 | if (!ns->initialized) | |
127 | { | |
128 | struct stat sb; | |
129 | ||
130 | if (stat (linux_ns_filename (ns, getpid ()), &sb) == 0) | |
131 | { | |
132 | ns->id = sb.st_ino; | |
133 | ||
134 | ns->supported = 1; | |
135 | } | |
136 | ||
137 | ns->initialized = 1; | |
138 | } | |
139 | ||
140 | return ns->supported ? ns : NULL; | |
141 | } | |
142 | ||
143 | /* See nat/linux-namespaces.h. */ | |
144 | ||
145 | int | |
146 | linux_ns_same (pid_t pid, enum linux_ns_type type) | |
147 | { | |
148 | struct linux_ns *ns = linux_ns_get_namespace (type); | |
149 | const char *filename; | |
150 | struct stat sb; | |
151 | ||
152 | /* If the kernel does not support TYPE namespaces then there's | |
153 | effectively only one TYPE namespace that all processes on | |
154 | the system share. */ | |
155 | if (ns == NULL) | |
156 | return 1; | |
157 | ||
158 | /* Stat PID's TYPE namespace entry to get the namespace ID. This | |
159 | might fail if the process died, or if we don't have the right | |
160 | permissions (though we should be attached by this time so this | |
161 | seems unlikely). In any event, we can't make any decisions and | |
162 | must throw. */ | |
163 | filename = linux_ns_filename (ns, pid); | |
164 | if (stat (filename, &sb) != 0) | |
165 | perror_with_name (filename); | |
166 | ||
167 | return sb.st_ino == ns->id; | |
168 | } | |
169 | ||
170 | /* We need to use setns(2) to handle filesystem access in mount | |
171 | namespaces other than our own, but this isn't permitted for | |
172 | multithreaded processes. GDB is multithreaded when compiled | |
173 | with Guile support, and may become multithreaded if compiled | |
174 | with Python support. We deal with this by spawning a single- | |
175 | threaded helper process to access mount namespaces other than | |
176 | our own. | |
177 | ||
178 | The helper process is started the first time a call to setns | |
179 | is required. The main process (GDB or gdbserver) communicates | |
180 | with the helper via sockets, passing file descriptors where | |
181 | necessary using SCM_RIGHTS. Once started the helper process | |
182 | runs until the main process terminates; when this happens the | |
183 | helper will receive socket errors, notice that its parent died, | |
184 | and exit accordingly (see mnsh_maybe_mourn_peer). | |
185 | ||
186 | The protocol is that the main process sends a request in a | |
187 | single message, and the helper replies to every message it | |
188 | receives with a single-message response. If the helper | |
189 | receives a message it does not understand it will reply with | |
190 | a MNSH_MSG_ERROR message. The main process checks all | |
191 | responses it receives with gdb_assert, so if the main process | |
192 | receives something unexpected (which includes MNSH_MSG_ERROR) | |
193 | the main process will call internal_error. | |
194 | ||
195 | For avoidance of doubt, if the helper process receives a | |
196 | message it doesn't handle it will reply with MNSH_MSG_ERROR. | |
197 | If the main process receives MNSH_MSG_ERROR at any time then | |
198 | it will call internal_error. If internal_error causes the | |
199 | main process to exit, the helper will notice this and also | |
200 | exit. The helper will not exit until the main process | |
201 | terminates, so if the user continues through internal_error | |
202 | the helper will still be there awaiting requests from the | |
203 | main process. | |
204 | ||
205 | Messages in both directions have the following payload: | |
206 | ||
207 | - TYPE (enum mnsh_msg_type, always sent) - the message type. | |
208 | - INT1 and | |
209 | - INT2 (int, always sent, though not always used) - two | |
210 | values whose meaning is message-type-dependent. | |
211 | See enum mnsh_msg_type documentation below. | |
212 | - FD (int, optional, sent using SCM_RIGHTS) - an open file | |
213 | descriptor. | |
214 | - BUF (unstructured data, optional) - some data with message- | |
215 | type-dependent meaning. | |
216 | ||
217 | Note that the helper process is the child of a call to fork, | |
218 | so all code in the helper must be async-signal-safe. */ | |
219 | ||
220 | /* Mount namespace helper message types. */ | |
221 | ||
222 | enum mnsh_msg_type | |
223 | { | |
224 | /* A communication error occurred. Receipt of this message | |
225 | by either end will cause an assertion failure in the main | |
226 | process. */ | |
227 | MNSH_MSG_ERROR, | |
228 | ||
229 | /* Requests, sent from the main process to the helper. */ | |
230 | ||
231 | /* A request that the helper call setns. Arguments should | |
232 | be passed in FD and INT1. Helper should respond with a | |
233 | MNSH_RET_INT. */ | |
234 | MNSH_REQ_SETNS, | |
235 | ||
236 | /* A request that the helper call open. Arguments should | |
237 | be passed in BUF, INT1 and INT2. The filename (in BUF) | |
238 | should include a terminating NUL character. The helper | |
239 | should respond with a MNSH_RET_FD. */ | |
240 | MNSH_REQ_OPEN, | |
241 | ||
242 | /* A request that the helper call unlink. The single | |
243 | argument (the filename) should be passed in BUF, and | |
244 | should include a terminating NUL character. The helper | |
245 | should respond with a MNSH_RET_INT. */ | |
246 | MNSH_REQ_UNLINK, | |
247 | ||
248 | /* A request that the helper call readlink. The single | |
249 | argument (the filename) should be passed in BUF, and | |
250 | should include a terminating NUL character. The helper | |
251 | should respond with a MNSH_RET_INTSTR. */ | |
252 | MNSH_REQ_READLINK, | |
253 | ||
254 | /* Responses, sent to the main process from the helper. */ | |
255 | ||
256 | /* Return an integer in INT1 and errno in INT2. */ | |
257 | MNSH_RET_INT, | |
258 | ||
259 | /* Return a file descriptor in FD if one was opened or an | |
260 | integer in INT1 otherwise. Return errno in INT2. */ | |
261 | MNSH_RET_FD, | |
262 | ||
263 | /* Return an integer in INT1, errno in INT2, and optionally | |
264 | some data in BUF. */ | |
265 | MNSH_RET_INTSTR, | |
266 | }; | |
267 | ||
268 | /* Print a string representation of a message using debug_printf. | |
269 | This function is not async-signal-safe so should never be | |
270 | called from the helper. */ | |
271 | ||
272 | static void | |
273 | mnsh_debug_print_message (enum mnsh_msg_type type, | |
274 | int fd, int int1, int int2, | |
275 | const void *buf, int bufsiz) | |
276 | { | |
277 | gdb_byte *c = (gdb_byte *) buf; | |
278 | gdb_byte *cl = c + bufsiz; | |
279 | ||
280 | switch (type) | |
281 | { | |
282 | case MNSH_MSG_ERROR: | |
283 | debug_printf ("ERROR"); | |
284 | break; | |
285 | ||
286 | case MNSH_REQ_SETNS: | |
287 | debug_printf ("SETNS"); | |
288 | break; | |
289 | ||
290 | case MNSH_REQ_OPEN: | |
291 | debug_printf ("OPEN"); | |
292 | break; | |
293 | ||
294 | case MNSH_REQ_UNLINK: | |
295 | debug_printf ("UNLINK"); | |
296 | break; | |
297 | ||
298 | case MNSH_REQ_READLINK: | |
299 | debug_printf ("READLINK"); | |
300 | break; | |
301 | ||
302 | case MNSH_RET_INT: | |
303 | debug_printf ("INT"); | |
304 | break; | |
305 | ||
306 | case MNSH_RET_FD: | |
307 | debug_printf ("FD"); | |
308 | break; | |
309 | ||
310 | case MNSH_RET_INTSTR: | |
311 | debug_printf ("INTSTR"); | |
312 | break; | |
313 | ||
314 | default: | |
315 | debug_printf ("unknown-packet-%d", type); | |
316 | } | |
317 | ||
318 | debug_printf (" %d %d %d \"", fd, int1, int2); | |
319 | ||
320 | for (; c < cl; c++) | |
321 | debug_printf (*c >= ' ' && *c <= '~' ? "%c" : "\\%o", *c); | |
322 | ||
323 | debug_printf ("\""); | |
324 | } | |
325 | ||
326 | /* Forward declaration. */ | |
327 | ||
328 | static void mnsh_maybe_mourn_peer (void); | |
329 | ||
330 | /* Send a message. The argument SOCK is the file descriptor of the | |
331 | sending socket, the other arguments are the payload to send. | |
332 | Return the number of bytes sent on success. Return -1 on failure | |
333 | and set errno appropriately. This function is called by both the | |
334 | main process and the helper so must be async-signal-safe. */ | |
335 | ||
336 | static ssize_t | |
337 | mnsh_send_message (int sock, enum mnsh_msg_type type, | |
338 | int fd, int int1, int int2, | |
339 | const void *buf, int bufsiz) | |
340 | { | |
341 | struct msghdr msg; | |
342 | struct iovec iov[4]; | |
343 | char fdbuf[CMSG_SPACE (sizeof (fd))]; | |
344 | ssize_t size; | |
345 | ||
346 | /* Build the basic TYPE, INT1, INT2 message. */ | |
347 | memset (&msg, 0, sizeof (msg)); | |
348 | msg.msg_iov = iov; | |
349 | ||
350 | iov[0].iov_base = &type; | |
351 | iov[0].iov_len = sizeof (type); | |
352 | iov[1].iov_base = &int1; | |
353 | iov[1].iov_len = sizeof (int1); | |
354 | iov[2].iov_base = &int2; | |
355 | iov[2].iov_len = sizeof (int2); | |
356 | ||
357 | msg.msg_iovlen = 3; | |
358 | ||
359 | /* Append BUF if supplied. */ | |
360 | if (buf != NULL && bufsiz > 0) | |
361 | { | |
362 | iov[3].iov_base = alloca (bufsiz); | |
363 | memcpy (iov[3].iov_base, buf, bufsiz); | |
364 | iov[3].iov_len = bufsiz; | |
365 | ||
366 | msg.msg_iovlen ++; | |
367 | } | |
368 | ||
369 | /* Attach FD if supplied. */ | |
370 | if (fd >= 0) | |
371 | { | |
372 | struct cmsghdr *cmsg; | |
373 | ||
374 | msg.msg_control = fdbuf; | |
375 | msg.msg_controllen = sizeof (fdbuf); | |
376 | ||
377 | cmsg = CMSG_FIRSTHDR (&msg); | |
378 | cmsg->cmsg_level = SOL_SOCKET; | |
379 | cmsg->cmsg_type = SCM_RIGHTS; | |
380 | cmsg->cmsg_len = CMSG_LEN (sizeof (int)); | |
381 | ||
382 | memcpy (CMSG_DATA (cmsg), &fd, sizeof (int)); | |
383 | ||
384 | msg.msg_controllen = cmsg->cmsg_len; | |
385 | } | |
386 | ||
387 | /* Send the message. */ | |
388 | size = sendmsg (sock, &msg, 0); | |
389 | ||
390 | if (size < 0) | |
391 | mnsh_maybe_mourn_peer (); | |
392 | ||
393 | if (debug_linux_namespaces) | |
394 | { | |
395 | debug_printf ("mnsh: send: "); | |
396 | mnsh_debug_print_message (type, fd, int1, int2, buf, bufsiz); | |
6449ed0d | 397 | debug_printf (" -> %s\n", pulongest (size)); |
4b8b5e72 GB |
398 | } |
399 | ||
400 | return size; | |
401 | } | |
402 | ||
403 | /* Receive a message. The argument SOCK is the file descriptor of | |
404 | the receiving socket, the other arguments point to storage for | |
405 | the received payload. Returns the number of bytes stored into | |
406 | BUF on success, which may be zero in the event no BUF was sent. | |
407 | Return -1 on failure and set errno appropriately. This function | |
408 | is called from both the main process and the helper and must be | |
409 | async-signal-safe. */ | |
410 | ||
411 | static ssize_t | |
412 | mnsh_recv_message (int sock, enum mnsh_msg_type *type, | |
413 | int *fd, int *int1, int *int2, | |
414 | void *buf, int bufsiz) | |
415 | { | |
416 | struct msghdr msg; | |
417 | struct iovec iov[4]; | |
418 | char fdbuf[CMSG_SPACE (sizeof (*fd))]; | |
419 | struct cmsghdr *cmsg; | |
420 | ssize_t size, fixed_size; | |
421 | int i; | |
422 | ||
423 | /* Build the message to receive data into. */ | |
424 | memset (&msg, 0, sizeof (msg)); | |
425 | msg.msg_iov = iov; | |
426 | ||
427 | iov[0].iov_base = type; | |
428 | iov[0].iov_len = sizeof (*type); | |
429 | iov[1].iov_base = int1; | |
430 | iov[1].iov_len = sizeof (*int1); | |
431 | iov[2].iov_base = int2; | |
432 | iov[2].iov_len = sizeof (*int2); | |
433 | iov[3].iov_base = buf; | |
434 | iov[3].iov_len = bufsiz; | |
435 | ||
436 | msg.msg_iovlen = 4; | |
437 | ||
438 | for (fixed_size = i = 0; i < msg.msg_iovlen - 1; i++) | |
439 | fixed_size += iov[i].iov_len; | |
440 | ||
441 | msg.msg_control = fdbuf; | |
442 | msg.msg_controllen = sizeof (fdbuf); | |
443 | ||
444 | /* Receive the message. */ | |
445 | size = recvmsg (sock, &msg, MSG_CMSG_CLOEXEC); | |
446 | if (size < 0) | |
447 | { | |
448 | if (debug_linux_namespaces) | |
6449ed0d GB |
449 | debug_printf ("namespace-helper: recv failed (%s)\n", |
450 | pulongest (size)); | |
4b8b5e72 GB |
451 | |
452 | mnsh_maybe_mourn_peer (); | |
453 | ||
454 | return size; | |
455 | } | |
456 | ||
457 | /* Check for truncation. */ | |
458 | if (size < fixed_size || (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) | |
459 | { | |
460 | if (debug_linux_namespaces) | |
6449ed0d GB |
461 | debug_printf ("namespace-helper: recv truncated (%s 0x%x)\n", |
462 | pulongest (size), msg.msg_flags); | |
4b8b5e72 GB |
463 | |
464 | mnsh_maybe_mourn_peer (); | |
465 | ||
466 | errno = EBADMSG; | |
467 | return -1; | |
468 | } | |
469 | ||
470 | /* Unpack the file descriptor if supplied. */ | |
471 | cmsg = CMSG_FIRSTHDR (&msg); | |
472 | if (cmsg != NULL | |
473 | && cmsg->cmsg_len == CMSG_LEN (sizeof (int)) | |
474 | && cmsg->cmsg_level == SOL_SOCKET | |
475 | && cmsg->cmsg_type == SCM_RIGHTS) | |
476 | memcpy (fd, CMSG_DATA (cmsg), sizeof (int)); | |
477 | else | |
478 | *fd = -1; | |
479 | ||
480 | if (debug_linux_namespaces) | |
481 | { | |
482 | debug_printf ("mnsh: recv: "); | |
483 | mnsh_debug_print_message (*type, *fd, *int1, *int2, buf, | |
484 | size - fixed_size); | |
485 | debug_printf ("\n"); | |
486 | } | |
487 | ||
488 | /* Return the number of bytes of data in BUF. */ | |
489 | return size - fixed_size; | |
490 | } | |
491 | ||
492 | /* Shortcuts for returning results from the helper. */ | |
493 | ||
494 | #define mnsh_return_int(sock, result, error) \ | |
495 | mnsh_send_message (sock, MNSH_RET_INT, -1, result, error, NULL, 0) | |
496 | ||
497 | #define mnsh_return_fd(sock, fd, error) \ | |
498 | mnsh_send_message (sock, MNSH_RET_FD, \ | |
499 | (fd) < 0 ? -1 : (fd), \ | |
500 | (fd) < 0 ? (fd) : 0, \ | |
501 | error, NULL, 0) | |
502 | ||
503 | #define mnsh_return_intstr(sock, result, buf, bufsiz, error) \ | |
504 | mnsh_send_message (sock, MNSH_RET_INTSTR, -1, result, error, \ | |
505 | buf, bufsiz) | |
506 | ||
507 | /* Handle a MNSH_REQ_SETNS message. Must be async-signal-safe. */ | |
508 | ||
509 | static ssize_t | |
510 | mnsh_handle_setns (int sock, int fd, int nstype) | |
511 | { | |
99fe86f7 | 512 | int result = do_setns (fd, nstype); |
4b8b5e72 GB |
513 | |
514 | return mnsh_return_int (sock, result, errno); | |
515 | } | |
516 | ||
517 | /* Handle a MNSH_REQ_OPEN message. Must be async-signal-safe. */ | |
518 | ||
519 | static ssize_t | |
520 | mnsh_handle_open (int sock, const char *filename, | |
521 | int flags, mode_t mode) | |
522 | { | |
523 | int fd = gdb_open_cloexec (filename, flags, mode); | |
524 | ssize_t result = mnsh_return_fd (sock, fd, errno); | |
525 | ||
526 | if (fd >= 0) | |
527 | close (fd); | |
528 | ||
529 | return result; | |
530 | } | |
531 | ||
532 | /* Handle a MNSH_REQ_UNLINK message. Must be async-signal-safe. */ | |
533 | ||
534 | static ssize_t | |
535 | mnsh_handle_unlink (int sock, const char *filename) | |
536 | { | |
537 | int result = unlink (filename); | |
538 | ||
539 | return mnsh_return_int (sock, result, errno); | |
540 | } | |
541 | ||
542 | /* Handle a MNSH_REQ_READLINK message. Must be async-signal-safe. */ | |
543 | ||
544 | static ssize_t | |
545 | mnsh_handle_readlink (int sock, const char *filename) | |
546 | { | |
547 | char buf[PATH_MAX]; | |
548 | int len = readlink (filename, buf, sizeof (buf)); | |
549 | ||
550 | return mnsh_return_intstr (sock, len, | |
551 | buf, len < 0 ? 0 : len, | |
552 | errno); | |
553 | } | |
554 | ||
555 | /* The helper process. Never returns. Must be async-signal-safe. */ | |
556 | ||
557 | static void mnsh_main (int sock) ATTRIBUTE_NORETURN; | |
558 | ||
559 | static void | |
560 | mnsh_main (int sock) | |
561 | { | |
562 | while (1) | |
563 | { | |
564 | enum mnsh_msg_type type; | |
550105b7 | 565 | int fd = -1, int1, int2; |
4b8b5e72 GB |
566 | char buf[PATH_MAX]; |
567 | ssize_t size, response = -1; | |
568 | ||
569 | size = mnsh_recv_message (sock, &type, | |
570 | &fd, &int1, &int2, | |
571 | buf, sizeof (buf)); | |
572 | ||
573 | if (size >= 0 && size < sizeof (buf)) | |
574 | { | |
575 | switch (type) | |
576 | { | |
577 | case MNSH_REQ_SETNS: | |
578 | if (fd > 0) | |
579 | response = mnsh_handle_setns (sock, fd, int1); | |
580 | break; | |
581 | ||
582 | case MNSH_REQ_OPEN: | |
583 | if (size > 0 && buf[size - 1] == '\0') | |
584 | response = mnsh_handle_open (sock, buf, int1, int2); | |
585 | break; | |
586 | ||
587 | case MNSH_REQ_UNLINK: | |
588 | if (size > 0 && buf[size - 1] == '\0') | |
589 | response = mnsh_handle_unlink (sock, buf); | |
590 | break; | |
591 | ||
592 | case MNSH_REQ_READLINK: | |
593 | if (size > 0 && buf[size - 1] == '\0') | |
594 | response = mnsh_handle_readlink (sock, buf); | |
595 | break; | |
596 | ||
597 | default: | |
598 | break; /* Handled below. */ | |
599 | } | |
600 | } | |
601 | ||
602 | /* Close any file descriptors we were passed. */ | |
603 | if (fd >= 0) | |
604 | close (fd); | |
605 | ||
606 | /* Can't handle this message, bounce it back. */ | |
607 | if (response < 0) | |
608 | { | |
609 | if (size < 0) | |
610 | size = 0; | |
611 | ||
612 | mnsh_send_message (sock, MNSH_MSG_ERROR, | |
613 | -1, int1, int2, buf, size); | |
614 | } | |
615 | } | |
616 | } | |
617 | ||
618 | /* The mount namespace helper process. */ | |
619 | ||
620 | struct linux_mnsh | |
621 | { | |
622 | /* PID of helper. */ | |
623 | pid_t pid; | |
624 | ||
625 | /* Socket for communication. */ | |
626 | int sock; | |
627 | ||
628 | /* ID of the mount namespace the helper is currently in. */ | |
629 | ino_t nsid; | |
630 | }; | |
631 | ||
632 | /* In the helper process this is set to the PID of the process that | |
633 | created the helper (i.e. GDB or gdbserver). In the main process | |
634 | this is set to zero. Used by mnsh_maybe_mourn_peer. */ | |
635 | static int mnsh_creator_pid = 0; | |
636 | ||
637 | /* Return an object representing the mount namespace helper process. | |
638 | If no mount namespace helper process has been started then start | |
639 | one. Return NULL if no mount namespace helper process could be | |
640 | started. */ | |
641 | ||
642 | static struct linux_mnsh * | |
643 | linux_mntns_get_helper (void) | |
644 | { | |
645 | static struct linux_mnsh *helper = NULL; | |
646 | ||
647 | if (helper == NULL) | |
648 | { | |
649 | static struct linux_mnsh h; | |
650 | struct linux_ns *ns; | |
651 | pid_t helper_creator = getpid (); | |
652 | int sv[2]; | |
653 | ||
654 | ns = linux_ns_get_namespace (LINUX_NS_MNT); | |
655 | if (ns == NULL) | |
656 | return NULL; | |
657 | ||
658 | if (gdb_socketpair_cloexec (AF_UNIX, SOCK_STREAM, 0, sv) < 0) | |
659 | return NULL; | |
660 | ||
eb0edac8 | 661 | h.pid = do_fork (); |
4b8b5e72 GB |
662 | if (h.pid < 0) |
663 | { | |
664 | int saved_errno = errno; | |
665 | ||
666 | close (sv[0]); | |
667 | close (sv[1]); | |
668 | ||
669 | errno = saved_errno; | |
670 | return NULL; | |
671 | } | |
672 | ||
673 | if (h.pid == 0) | |
674 | { | |
675 | /* Child process. */ | |
676 | close (sv[0]); | |
677 | ||
678 | mnsh_creator_pid = helper_creator; | |
679 | ||
680 | /* Debug printing isn't async-signal-safe. */ | |
681 | debug_linux_namespaces = 0; | |
682 | ||
683 | mnsh_main (sv[1]); | |
684 | } | |
685 | ||
686 | /* Parent process. */ | |
687 | close (sv[1]); | |
688 | ||
689 | helper = &h; | |
690 | helper->sock = sv[0]; | |
691 | helper->nsid = ns->id; | |
692 | ||
693 | if (debug_linux_namespaces) | |
694 | debug_printf ("Started mount namespace helper process %d\n", | |
695 | helper->pid); | |
696 | } | |
697 | ||
698 | return helper; | |
699 | } | |
700 | ||
701 | /* Check whether the other process died and act accordingly. Called | |
702 | whenever a socket error occurs, from both the main process and the | |
703 | helper. Must be async-signal-safe when called from the helper. */ | |
704 | ||
705 | static void | |
706 | mnsh_maybe_mourn_peer (void) | |
707 | { | |
708 | if (mnsh_creator_pid != 0) | |
709 | { | |
710 | /* We're in the helper. Check if our current parent is the | |
711 | process that started us. If it isn't, then our original | |
712 | parent died and we've been reparented. Exit immediately | |
713 | if that's the case. */ | |
714 | if (getppid () != mnsh_creator_pid) | |
715 | _exit (0); | |
716 | } | |
717 | else | |
718 | { | |
719 | /* We're in the main process. */ | |
720 | ||
721 | struct linux_mnsh *helper = linux_mntns_get_helper (); | |
722 | int status; | |
723 | pid_t pid; | |
724 | ||
725 | if (helper->pid < 0) | |
726 | { | |
727 | /* We already mourned it. */ | |
728 | return; | |
729 | } | |
730 | ||
731 | pid = waitpid (helper->pid, &status, WNOHANG); | |
732 | if (pid == 0) | |
733 | { | |
734 | /* The helper is still alive. */ | |
735 | return; | |
736 | } | |
737 | else if (pid == -1) | |
738 | { | |
739 | if (errno == ECHILD) | |
740 | warning (_("mount namespace helper vanished?")); | |
741 | else | |
742 | internal_warning (__FILE__, __LINE__, | |
743 | _("unhandled error %d"), errno); | |
744 | } | |
745 | else if (pid == helper->pid) | |
746 | { | |
747 | if (WIFEXITED (status)) | |
748 | warning (_("mount namespace helper exited with status %d"), | |
749 | WEXITSTATUS (status)); | |
750 | else if (WIFSIGNALED (status)) | |
751 | warning (_("mount namespace helper killed by signal %d"), | |
752 | WTERMSIG (status)); | |
753 | else | |
754 | internal_warning (__FILE__, __LINE__, | |
755 | _("unhandled status %d"), status); | |
756 | } | |
757 | else | |
758 | internal_warning (__FILE__, __LINE__, | |
759 | _("unknown pid %d"), pid); | |
760 | ||
761 | /* Something unrecoverable happened. */ | |
762 | helper->pid = -1; | |
763 | } | |
764 | } | |
765 | ||
766 | /* Shortcuts for sending messages to the helper. */ | |
767 | ||
768 | #define mnsh_send_setns(helper, fd, nstype) \ | |
769 | mnsh_send_message (helper->sock, MNSH_REQ_SETNS, fd, nstype, 0, \ | |
770 | NULL, 0) | |
771 | ||
772 | #define mnsh_send_open(helper, filename, flags, mode) \ | |
773 | mnsh_send_message (helper->sock, MNSH_REQ_OPEN, -1, flags, mode, \ | |
774 | filename, strlen (filename) + 1) | |
775 | ||
776 | #define mnsh_send_unlink(helper, filename) \ | |
777 | mnsh_send_message (helper->sock, MNSH_REQ_UNLINK, -1, 0, 0, \ | |
778 | filename, strlen (filename) + 1) | |
779 | ||
780 | #define mnsh_send_readlink(helper, filename) \ | |
781 | mnsh_send_message (helper->sock, MNSH_REQ_READLINK, -1, 0, 0, \ | |
782 | filename, strlen (filename) + 1) | |
783 | ||
784 | /* Receive a message from the helper. Issue an assertion failure if | |
785 | the message isn't a correctly-formatted MNSH_RET_INT. Set RESULT | |
786 | and ERROR and return 0 on success. Set errno and return -1 on | |
787 | failure. */ | |
788 | ||
789 | static int | |
790 | mnsh_recv_int (struct linux_mnsh *helper, int *result, int *error) | |
791 | { | |
792 | enum mnsh_msg_type type; | |
793 | char buf[PATH_MAX]; | |
794 | ssize_t size; | |
795 | int fd; | |
796 | ||
797 | size = mnsh_recv_message (helper->sock, &type, &fd, | |
798 | result, error, | |
799 | buf, sizeof (buf)); | |
800 | if (size < 0) | |
801 | return -1; | |
802 | ||
803 | gdb_assert (type == MNSH_RET_INT); | |
804 | gdb_assert (fd == -1); | |
805 | gdb_assert (size == 0); | |
806 | ||
807 | return 0; | |
808 | } | |
809 | ||
810 | /* Receive a message from the helper. Issue an assertion failure if | |
811 | the message isn't a correctly-formatted MNSH_RET_FD. Set FD and | |
812 | ERROR and return 0 on success. Set errno and return -1 on | |
813 | failure. */ | |
814 | ||
815 | static int | |
816 | mnsh_recv_fd (struct linux_mnsh *helper, int *fd, int *error) | |
817 | { | |
818 | enum mnsh_msg_type type; | |
819 | char buf[PATH_MAX]; | |
820 | ssize_t size; | |
821 | int result; | |
822 | ||
823 | size = mnsh_recv_message (helper->sock, &type, fd, | |
824 | &result, error, | |
825 | buf, sizeof (buf)); | |
826 | if (size < 0) | |
827 | return -1; | |
828 | ||
829 | gdb_assert (type == MNSH_RET_FD); | |
830 | gdb_assert (size == 0); | |
831 | ||
832 | if (*fd < 0) | |
833 | { | |
834 | gdb_assert (result < 0); | |
835 | *fd = result; | |
836 | } | |
837 | ||
838 | return 0; | |
839 | } | |
840 | ||
841 | /* Receive a message from the helper. Issue an assertion failure if | |
842 | the message isn't a correctly-formatted MNSH_RET_INTSTR. Set | |
843 | RESULT and ERROR and optionally store data in BUF, then return | |
844 | the number of bytes stored in BUF on success (this may be zero). | |
845 | Set errno and return -1 on error. */ | |
846 | ||
847 | static ssize_t | |
848 | mnsh_recv_intstr (struct linux_mnsh *helper, | |
849 | int *result, int *error, | |
850 | void *buf, int bufsiz) | |
851 | { | |
852 | enum mnsh_msg_type type; | |
853 | ssize_t size; | |
854 | int fd; | |
855 | ||
856 | size = mnsh_recv_message (helper->sock, &type, &fd, | |
857 | result, error, | |
858 | buf, bufsiz); | |
859 | ||
860 | if (size < 0) | |
861 | return -1; | |
862 | ||
863 | gdb_assert (type == MNSH_RET_INTSTR); | |
864 | gdb_assert (fd == -1); | |
865 | ||
866 | return size; | |
867 | } | |
868 | ||
869 | /* Return values for linux_mntns_access_fs. */ | |
870 | ||
871 | enum mnsh_fs_code | |
872 | { | |
873 | /* Something went wrong, errno is set. */ | |
874 | MNSH_FS_ERROR = -1, | |
875 | ||
876 | /* The main process is in the correct mount namespace. | |
877 | The caller should access the filesystem directly. */ | |
878 | MNSH_FS_DIRECT, | |
879 | ||
880 | /* The helper is in the correct mount namespace. | |
881 | The caller should access the filesystem via the helper. */ | |
882 | MNSH_FS_HELPER | |
883 | }; | |
884 | ||
885 | /* Return a value indicating how the caller should access the | |
886 | mount namespace of process PID. */ | |
887 | ||
888 | static enum mnsh_fs_code | |
889 | linux_mntns_access_fs (pid_t pid) | |
890 | { | |
4b8b5e72 GB |
891 | struct linux_ns *ns; |
892 | struct stat sb; | |
893 | struct linux_mnsh *helper; | |
894 | ssize_t size; | |
6cceac94 | 895 | int fd; |
4b8b5e72 GB |
896 | |
897 | if (pid == getpid ()) | |
898 | return MNSH_FS_DIRECT; | |
899 | ||
900 | ns = linux_ns_get_namespace (LINUX_NS_MNT); | |
901 | if (ns == NULL) | |
902 | return MNSH_FS_DIRECT; | |
903 | ||
4b8b5e72 GB |
904 | fd = gdb_open_cloexec (linux_ns_filename (ns, pid), O_RDONLY, 0); |
905 | if (fd < 0) | |
6cceac94 | 906 | return MNSH_FS_ERROR; |
4b8b5e72 | 907 | |
6cceac94 TT |
908 | SCOPE_EXIT |
909 | { | |
910 | int save_errno = errno; | |
911 | close (fd); | |
912 | errno = save_errno; | |
913 | }; | |
4b8b5e72 GB |
914 | |
915 | if (fstat (fd, &sb) != 0) | |
6cceac94 | 916 | return MNSH_FS_ERROR; |
4b8b5e72 GB |
917 | |
918 | if (sb.st_ino == ns->id) | |
6cceac94 | 919 | return MNSH_FS_DIRECT; |
4b8b5e72 GB |
920 | |
921 | helper = linux_mntns_get_helper (); | |
922 | if (helper == NULL) | |
6cceac94 | 923 | return MNSH_FS_ERROR; |
4b8b5e72 GB |
924 | |
925 | if (sb.st_ino != helper->nsid) | |
926 | { | |
927 | int result, error; | |
928 | ||
929 | size = mnsh_send_setns (helper, fd, 0); | |
930 | if (size < 0) | |
6cceac94 | 931 | return MNSH_FS_ERROR; |
4b8b5e72 GB |
932 | |
933 | if (mnsh_recv_int (helper, &result, &error) != 0) | |
6cceac94 | 934 | return MNSH_FS_ERROR; |
4b8b5e72 GB |
935 | |
936 | if (result != 0) | |
937 | { | |
938 | /* ENOSYS indicates that an entire function is unsupported | |
939 | (it's not appropriate for our versions of open/unlink/ | |
940 | readlink to sometimes return with ENOSYS depending on how | |
941 | they're called) so we convert ENOSYS to ENOTSUP if setns | |
942 | fails. */ | |
943 | if (error == ENOSYS) | |
944 | error = ENOTSUP; | |
945 | ||
946 | errno = error; | |
6cceac94 | 947 | return MNSH_FS_ERROR; |
4b8b5e72 GB |
948 | } |
949 | ||
950 | helper->nsid = sb.st_ino; | |
951 | } | |
952 | ||
4b8b5e72 | 953 | return MNSH_FS_HELPER; |
4b8b5e72 GB |
954 | } |
955 | ||
956 | /* See nat/linux-namespaces.h. */ | |
957 | ||
958 | int | |
959 | linux_mntns_open_cloexec (pid_t pid, const char *filename, | |
960 | int flags, mode_t mode) | |
961 | { | |
962 | enum mnsh_fs_code access = linux_mntns_access_fs (pid); | |
963 | struct linux_mnsh *helper; | |
964 | int fd, error; | |
965 | ssize_t size; | |
966 | ||
967 | if (access == MNSH_FS_ERROR) | |
968 | return -1; | |
969 | ||
970 | if (access == MNSH_FS_DIRECT) | |
971 | return gdb_open_cloexec (filename, flags, mode); | |
972 | ||
973 | gdb_assert (access == MNSH_FS_HELPER); | |
974 | ||
975 | helper = linux_mntns_get_helper (); | |
976 | ||
977 | size = mnsh_send_open (helper, filename, flags, mode); | |
978 | if (size < 0) | |
979 | return -1; | |
980 | ||
981 | if (mnsh_recv_fd (helper, &fd, &error) != 0) | |
982 | return -1; | |
983 | ||
984 | if (fd < 0) | |
985 | errno = error; | |
986 | ||
987 | return fd; | |
988 | } | |
989 | ||
990 | /* See nat/linux-namespaces.h. */ | |
991 | ||
992 | int | |
993 | linux_mntns_unlink (pid_t pid, const char *filename) | |
994 | { | |
995 | enum mnsh_fs_code access = linux_mntns_access_fs (pid); | |
996 | struct linux_mnsh *helper; | |
997 | int ret, error; | |
998 | ssize_t size; | |
999 | ||
1000 | if (access == MNSH_FS_ERROR) | |
1001 | return -1; | |
1002 | ||
1003 | if (access == MNSH_FS_DIRECT) | |
1004 | return unlink (filename); | |
1005 | ||
1006 | gdb_assert (access == MNSH_FS_HELPER); | |
1007 | ||
1008 | helper = linux_mntns_get_helper (); | |
1009 | ||
1010 | size = mnsh_send_unlink (helper, filename); | |
1011 | if (size < 0) | |
1012 | return -1; | |
1013 | ||
1014 | if (mnsh_recv_int (helper, &ret, &error) != 0) | |
1015 | return -1; | |
1016 | ||
1017 | if (ret != 0) | |
1018 | errno = error; | |
1019 | ||
1020 | return ret; | |
1021 | } | |
1022 | ||
1023 | /* See nat/linux-namespaces.h. */ | |
1024 | ||
1025 | ssize_t | |
1026 | linux_mntns_readlink (pid_t pid, const char *filename, | |
1027 | char *buf, size_t bufsiz) | |
1028 | { | |
1029 | enum mnsh_fs_code access = linux_mntns_access_fs (pid); | |
1030 | struct linux_mnsh *helper; | |
1031 | int ret, error; | |
1032 | ssize_t size; | |
1033 | ||
1034 | if (access == MNSH_FS_ERROR) | |
1035 | return -1; | |
1036 | ||
1037 | if (access == MNSH_FS_DIRECT) | |
1038 | return readlink (filename, buf, bufsiz); | |
1039 | ||
1040 | gdb_assert (access == MNSH_FS_HELPER); | |
1041 | ||
1042 | helper = linux_mntns_get_helper (); | |
1043 | ||
1044 | size = mnsh_send_readlink (helper, filename); | |
1045 | if (size < 0) | |
1046 | return -1; | |
1047 | ||
1048 | size = mnsh_recv_intstr (helper, &ret, &error, buf, bufsiz); | |
1049 | ||
1050 | if (size < 0) | |
1051 | { | |
1052 | ret = -1; | |
1053 | errno = error; | |
1054 | } | |
1055 | else | |
1056 | gdb_assert (size == ret); | |
1057 | ||
1058 | return ret; | |
1059 | } |