Fri, 12 Jul 2013 18:43:27 +0200
7162400: Intermittent java.io.IOException: Bad file number during HotSpotVirtualMachine.executeCommand
Summary: Intermittent java.io.IOException: Bad file number during HotSpotVirtualMachine.executeCommand
Reviewed-by: dcubed, dholmes, sspitsyn, mgerdin, ctornqvi, dsamersoff
1 /*
2 * Copyright (c) 2005, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "runtime/interfaceSupport.hpp"
27 #include "runtime/os.hpp"
28 #include "services/attachListener.hpp"
29 #include "services/dtraceAttacher.hpp"
31 #include <unistd.h>
32 #include <signal.h>
33 #include <sys/types.h>
34 #include <sys/socket.h>
35 #include <sys/un.h>
36 #include <sys/stat.h>
38 #ifndef UNIX_PATH_MAX
39 #define UNIX_PATH_MAX sizeof(((struct sockaddr_un *)0)->sun_path)
40 #endif
42 // The attach mechanism on Linux uses a UNIX domain socket. An attach listener
43 // thread is created at startup or is created on-demand via a signal from
44 // the client tool. The attach listener creates a socket and binds it to a file
45 // in the filesystem. The attach listener then acts as a simple (single-
46 // threaded) server - it waits for a client to connect, reads the request,
47 // executes it, and returns the response to the client via the socket
48 // connection.
49 //
50 // As the socket is a UNIX domain socket it means that only clients on the
51 // local machine can connect. In addition there are two other aspects to
52 // the security:
53 // 1. The well known file that the socket is bound to has permission 400
54 // 2. When a client connect, the SO_PEERCRED socket option is used to
55 // obtain the credentials of client. We check that the effective uid
56 // of the client matches this process.
58 // forward reference
59 class LinuxAttachOperation;
61 class LinuxAttachListener: AllStatic {
62 private:
63 // the path to which we bind the UNIX domain socket
64 static char _path[UNIX_PATH_MAX];
65 static bool _has_path;
67 // the file descriptor for the listening socket
68 static int _listener;
70 static void set_path(char* path) {
71 if (path == NULL) {
72 _has_path = false;
73 } else {
74 strncpy(_path, path, UNIX_PATH_MAX);
75 _path[UNIX_PATH_MAX-1] = '\0';
76 _has_path = true;
77 }
78 }
80 static void set_listener(int s) { _listener = s; }
82 // reads a request from the given connected socket
83 static LinuxAttachOperation* read_request(int s);
85 public:
86 enum {
87 ATTACH_PROTOCOL_VER = 1 // protocol version
88 };
89 enum {
90 ATTACH_ERROR_BADVERSION = 101 // error codes
91 };
93 // initialize the listener, returns 0 if okay
94 static int init();
96 static char* path() { return _path; }
97 static bool has_path() { return _has_path; }
98 static int listener() { return _listener; }
100 // write the given buffer to a socket
101 static int write_fully(int s, char* buf, int len);
103 static LinuxAttachOperation* dequeue();
104 };
106 class LinuxAttachOperation: public AttachOperation {
107 private:
108 // the connection to the client
109 int _socket;
111 public:
112 void complete(jint res, bufferedStream* st);
114 void set_socket(int s) { _socket = s; }
115 int socket() const { return _socket; }
117 LinuxAttachOperation(char* name) : AttachOperation(name) {
118 set_socket(-1);
119 }
120 };
122 // statics
123 char LinuxAttachListener::_path[UNIX_PATH_MAX];
124 bool LinuxAttachListener::_has_path;
125 int LinuxAttachListener::_listener = -1;
127 // Supporting class to help split a buffer into individual components
128 class ArgumentIterator : public StackObj {
129 private:
130 char* _pos;
131 char* _end;
132 public:
133 ArgumentIterator(char* arg_buffer, size_t arg_size) {
134 _pos = arg_buffer;
135 _end = _pos + arg_size - 1;
136 }
137 char* next() {
138 if (*_pos == '\0') {
139 return NULL;
140 }
141 char* res = _pos;
142 char* next_pos = strchr(_pos, '\0');
143 if (next_pos < _end) {
144 next_pos++;
145 }
146 _pos = next_pos;
147 return res;
148 }
149 };
152 // atexit hook to stop listener and unlink the file that it is
153 // bound too.
154 extern "C" {
155 static void listener_cleanup() {
156 static int cleanup_done;
157 if (!cleanup_done) {
158 cleanup_done = 1;
159 int s = LinuxAttachListener::listener();
160 if (s != -1) {
161 ::close(s);
162 }
163 if (LinuxAttachListener::has_path()) {
164 ::unlink(LinuxAttachListener::path());
165 }
166 }
167 }
168 }
170 // Initialization - create a listener socket and bind it to a file
172 int LinuxAttachListener::init() {
173 char path[UNIX_PATH_MAX]; // socket file
174 char initial_path[UNIX_PATH_MAX]; // socket file during setup
175 int listener; // listener socket (file descriptor)
177 // register function to cleanup
178 ::atexit(listener_cleanup);
180 int n = snprintf(path, UNIX_PATH_MAX, "%s/.java_pid%d",
181 os::get_temp_directory(), os::current_process_id());
182 if (n < (int)UNIX_PATH_MAX) {
183 n = snprintf(initial_path, UNIX_PATH_MAX, "%s.tmp", path);
184 }
185 if (n >= (int)UNIX_PATH_MAX) {
186 return -1;
187 }
189 // create the listener socket
190 listener = ::socket(PF_UNIX, SOCK_STREAM, 0);
191 if (listener == -1) {
192 return -1;
193 }
195 // bind socket
196 struct sockaddr_un addr;
197 addr.sun_family = AF_UNIX;
198 strcpy(addr.sun_path, initial_path);
199 ::unlink(initial_path);
200 int res = ::bind(listener, (struct sockaddr*)&addr, sizeof(addr));
201 if (res == -1) {
202 ::close(listener);
203 return -1;
204 }
206 // put in listen mode, set permissions, and rename into place
207 res = ::listen(listener, 5);
208 if (res == 0) {
209 RESTARTABLE(::chmod(initial_path, S_IREAD|S_IWRITE), res);
210 if (res == 0) {
211 res = ::rename(initial_path, path);
212 }
213 }
214 if (res == -1) {
215 ::close(listener);
216 ::unlink(initial_path);
217 return -1;
218 }
219 set_path(path);
220 set_listener(listener);
222 return 0;
223 }
225 // Given a socket that is connected to a peer we read the request and
226 // create an AttachOperation. As the socket is blocking there is potential
227 // for a denial-of-service if the peer does not response. However this happens
228 // after the peer credentials have been checked and in the worst case it just
229 // means that the attach listener thread is blocked.
230 //
231 LinuxAttachOperation* LinuxAttachListener::read_request(int s) {
232 char ver_str[8];
233 sprintf(ver_str, "%d", ATTACH_PROTOCOL_VER);
235 // The request is a sequence of strings so we first figure out the
236 // expected count and the maximum possible length of the request.
237 // The request is:
238 // <ver>0<cmd>0<arg>0<arg>0<arg>0
239 // where <ver> is the protocol version (1), <cmd> is the command
240 // name ("load", "datadump", ...), and <arg> is an argument
241 int expected_str_count = 2 + AttachOperation::arg_count_max;
242 const int max_len = (sizeof(ver_str) + 1) + (AttachOperation::name_length_max + 1) +
243 AttachOperation::arg_count_max*(AttachOperation::arg_length_max + 1);
245 char buf[max_len];
246 int str_count = 0;
248 // Read until all (expected) strings have been read, the buffer is
249 // full, or EOF.
251 int off = 0;
252 int left = max_len;
254 do {
255 int n;
256 RESTARTABLE(read(s, buf+off, left), n);
257 if (n == -1) {
258 return NULL; // reset by peer or other error
259 }
260 if (n == 0) {
261 break;
262 }
263 for (int i=0; i<n; i++) {
264 if (buf[off+i] == 0) {
265 // EOS found
266 str_count++;
268 // The first string is <ver> so check it now to
269 // check for protocol mis-match
270 if (str_count == 1) {
271 if ((strlen(buf) != strlen(ver_str)) ||
272 (atoi(buf) != ATTACH_PROTOCOL_VER)) {
273 char msg[32];
274 sprintf(msg, "%d\n", ATTACH_ERROR_BADVERSION);
275 write_fully(s, msg, strlen(msg));
276 return NULL;
277 }
278 }
279 }
280 }
281 off += n;
282 left -= n;
283 } while (left > 0 && str_count < expected_str_count);
285 if (str_count != expected_str_count) {
286 return NULL; // incomplete request
287 }
289 // parse request
291 ArgumentIterator args(buf, (max_len)-left);
293 // version already checked
294 char* v = args.next();
296 char* name = args.next();
297 if (name == NULL || strlen(name) > AttachOperation::name_length_max) {
298 return NULL;
299 }
301 LinuxAttachOperation* op = new LinuxAttachOperation(name);
303 for (int i=0; i<AttachOperation::arg_count_max; i++) {
304 char* arg = args.next();
305 if (arg == NULL) {
306 op->set_arg(i, NULL);
307 } else {
308 if (strlen(arg) > AttachOperation::arg_length_max) {
309 delete op;
310 return NULL;
311 }
312 op->set_arg(i, arg);
313 }
314 }
316 op->set_socket(s);
317 return op;
318 }
321 // Dequeue an operation
322 //
323 // In the Linux implementation there is only a single operation and clients
324 // cannot queue commands (except at the socket level).
325 //
326 LinuxAttachOperation* LinuxAttachListener::dequeue() {
327 for (;;) {
328 int s;
330 // wait for client to connect
331 struct sockaddr addr;
332 socklen_t len = sizeof(addr);
333 RESTARTABLE(::accept(listener(), &addr, &len), s);
334 if (s == -1) {
335 return NULL; // log a warning?
336 }
338 // get the credentials of the peer and check the effective uid/guid
339 // - check with jeff on this.
340 struct ucred cred_info;
341 socklen_t optlen = sizeof(cred_info);
342 if (::getsockopt(s, SOL_SOCKET, SO_PEERCRED, (void*)&cred_info, &optlen) == -1) {
343 ::close(s);
344 continue;
345 }
346 uid_t euid = geteuid();
347 gid_t egid = getegid();
349 if (cred_info.uid != euid || cred_info.gid != egid) {
350 ::close(s);
351 continue;
352 }
354 // peer credential look okay so we read the request
355 LinuxAttachOperation* op = read_request(s);
356 if (op == NULL) {
357 ::close(s);
358 continue;
359 } else {
360 return op;
361 }
362 }
363 }
365 // write the given buffer to the socket
366 int LinuxAttachListener::write_fully(int s, char* buf, int len) {
367 do {
368 int n = ::write(s, buf, len);
369 if (n == -1) {
370 if (errno != EINTR) return -1;
371 } else {
372 buf += n;
373 len -= n;
374 }
375 }
376 while (len > 0);
377 return 0;
378 }
380 // Complete an operation by sending the operation result and any result
381 // output to the client. At this time the socket is in blocking mode so
382 // potentially we can block if there is a lot of data and the client is
383 // non-responsive. For most operations this is a non-issue because the
384 // default send buffer is sufficient to buffer everything. In the future
385 // if there are operations that involves a very big reply then it the
386 // socket could be made non-blocking and a timeout could be used.
388 void LinuxAttachOperation::complete(jint result, bufferedStream* st) {
389 JavaThread* thread = JavaThread::current();
390 ThreadBlockInVM tbivm(thread);
392 thread->set_suspend_equivalent();
393 // cleared by handle_special_suspend_equivalent_condition() or
394 // java_suspend_self() via check_and_wait_while_suspended()
396 // write operation result
397 char msg[32];
398 sprintf(msg, "%d\n", result);
399 int rc = LinuxAttachListener::write_fully(this->socket(), msg, strlen(msg));
401 // write any result data
402 if (rc == 0) {
403 LinuxAttachListener::write_fully(this->socket(), (char*) st->base(), st->size());
404 ::shutdown(this->socket(), 2);
405 }
407 // done
408 ::close(this->socket());
410 // were we externally suspended while we were waiting?
411 thread->check_and_wait_while_suspended();
413 delete this;
414 }
417 // AttachListener functions
419 AttachOperation* AttachListener::dequeue() {
420 JavaThread* thread = JavaThread::current();
421 ThreadBlockInVM tbivm(thread);
423 thread->set_suspend_equivalent();
424 // cleared by handle_special_suspend_equivalent_condition() or
425 // java_suspend_self() via check_and_wait_while_suspended()
427 AttachOperation* op = LinuxAttachListener::dequeue();
429 // were we externally suspended while we were waiting?
430 thread->check_and_wait_while_suspended();
432 return op;
433 }
436 // Performs initialization at vm startup
437 // For Linux we remove any stale .java_pid file which could cause
438 // an attaching process to think we are ready to receive on the
439 // domain socket before we are properly initialized
441 void AttachListener::vm_start() {
442 char fn[UNIX_PATH_MAX];
443 struct stat64 st;
444 int ret;
446 int n = snprintf(fn, UNIX_PATH_MAX, "%s/.java_pid%d",
447 os::get_temp_directory(), os::current_process_id());
448 assert(n < (int)UNIX_PATH_MAX, "java_pid file name buffer overflow");
450 RESTARTABLE(::stat64(fn, &st), ret);
451 if (ret == 0) {
452 ret = ::unlink(fn);
453 if (ret == -1) {
454 debug_only(warning("failed to remove stale attach pid file at %s", fn));
455 }
456 }
457 }
459 int AttachListener::pd_init() {
460 JavaThread* thread = JavaThread::current();
461 ThreadBlockInVM tbivm(thread);
463 thread->set_suspend_equivalent();
464 // cleared by handle_special_suspend_equivalent_condition() or
465 // java_suspend_self() via check_and_wait_while_suspended()
467 int ret_code = LinuxAttachListener::init();
469 // were we externally suspended while we were waiting?
470 thread->check_and_wait_while_suspended();
472 return ret_code;
473 }
475 // Attach Listener is started lazily except in the case when
476 // +ReduseSignalUsage is used
477 bool AttachListener::init_at_startup() {
478 if (ReduceSignalUsage) {
479 return true;
480 } else {
481 return false;
482 }
483 }
485 // If the file .attach_pid<pid> exists in the working directory
486 // or /tmp then this is the trigger to start the attach mechanism
487 bool AttachListener::is_init_trigger() {
488 if (init_at_startup() || is_initialized()) {
489 return false; // initialized at startup or already initialized
490 }
491 char fn[PATH_MAX+1];
492 sprintf(fn, ".attach_pid%d", os::current_process_id());
493 int ret;
494 struct stat64 st;
495 RESTARTABLE(::stat64(fn, &st), ret);
496 if (ret == -1) {
497 snprintf(fn, sizeof(fn), "%s/.attach_pid%d",
498 os::get_temp_directory(), os::current_process_id());
499 RESTARTABLE(::stat64(fn, &st), ret);
500 }
501 if (ret == 0) {
502 // simple check to avoid starting the attach mechanism when
503 // a bogus user creates the file
504 if (st.st_uid == geteuid()) {
505 init();
506 return true;
507 }
508 }
509 return false;
510 }
512 // if VM aborts then remove listener
513 void AttachListener::abort() {
514 listener_cleanup();
515 }
517 void AttachListener::pd_data_dump() {
518 os::signal_notify(SIGQUIT);
519 }
521 AttachOperationFunctionInfo* AttachListener::pd_find_operation(const char* n) {
522 return NULL;
523 }
525 jint AttachListener::pd_set_flag(AttachOperation* op, outputStream* out) {
526 out->print_cr("flag '%s' cannot be changed", op->arg(0));
527 return JNI_ERR;
528 }
530 void AttachListener::pd_detachall() {
531 // do nothing for now
532 }