1 /*
2 * Copyright (c) 2012, 2019, 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 */
24
25 #include "precompiled.hpp"
26 #include "jfr/jfrEvents.hpp"
27 #include "jfr/jni/jfrJavaSupport.hpp"
28 #include "jfr/leakprofiler/leakProfiler.hpp"
29 #include "jfr/recorder/repository/jfrEmergencyDump.hpp"
30 #include "jfr/recorder/service/jfrPostBox.hpp"
31 #include "jfr/recorder/service/jfrRecorderService.hpp"
32 #include "jfr/utilities/jfrTypes.hpp"
33 #include "logging/log.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "runtime/atomic.hpp"
36 #include "runtime/handles.inline.hpp"
37 #include "runtime/globals.hpp"
38 #include "runtime/mutexLocker.hpp"
39 #include "runtime/os.hpp"
40 #include "runtime/thread.inline.hpp"
41 #include "utilities/growableArray.hpp"
42
43 static const char vm_error_filename_fmt[] = "hs_err_pid%p.jfr";
44 static const char vm_oom_filename_fmt[] = "hs_oom_pid%p.jfr";
45 static const char vm_soe_filename_fmt[] = "hs_soe_pid%p.jfr";
46 static const char chunk_file_jfr_ext[] = ".jfr";
47 static const size_t iso8601_len = 19; // "YYYY-MM-DDTHH:MM:SS"
48 static fio_fd emergency_fd = invalid_fd;
49
50 static fio_fd open_exclusivly(const char* path) {
51 return os::open(path, O_CREAT | O_RDWR, S_IREAD | S_IWRITE);
52 }
53
54 static int file_sort(const char** const file1, const char** file2) {
55 assert(NULL != *file1 && NULL != *file2, "invariant");
56 int cmp = strncmp(*file1, *file2, iso8601_len);
57 if (0 == cmp) {
58 const char* const dot1 = strchr(*file1, '.');
59 assert(NULL != dot1, "invariant");
60 const char* const dot2 = strchr(*file2, '.');
61 assert(NULL != dot2, "invariant");
62 ptrdiff_t file1_len = dot1 - *file1;
63 ptrdiff_t file2_len = dot2 - *file2;
64 if (file1_len < file2_len) {
65 return -1;
66 }
67 if (file1_len > file2_len) {
68 return 1;
69 }
70 assert(file1_len == file2_len, "invariant");
71 cmp = strncmp(*file1, *file2, file1_len);
72 }
73 assert(cmp != 0, "invariant");
74 return cmp;
75 }
76
77 static void iso8601_to_date_time(char* iso8601_str) {
78 assert(iso8601_str != NULL, "invariant");
79 assert(strlen(iso8601_str) == iso8601_len, "invariant");
80 // "YYYY-MM-DDTHH:MM:SS"
81 for (size_t i = 0; i < iso8601_len; ++i) {
82 switch (iso8601_str[i]) {
83 case 'T':
84 case '-':
85 case ':':
86 iso8601_str[i] = '_';
87 break;
88 }
89 }
90 // "YYYY_MM_DD_HH_MM_SS"
91 }
92
93 static void date_time(char* buffer, size_t buffer_len) {
94 assert(buffer != NULL, "invariant");
95 assert(buffer_len >= iso8601_len, "buffer too small");
96 os::iso8601_time(buffer, buffer_len);
97 assert(strlen(buffer) >= iso8601_len + 1, "invariant");
98 // "YYYY-MM-DDTHH:MM:SS"
99 buffer[iso8601_len] = '\0';
100 iso8601_to_date_time(buffer);
101 }
102
103 static int64_t file_size(fio_fd fd) {
104 assert(fd != invalid_fd, "invariant");
105 const int64_t current_offset = os::current_file_offset(fd);
106 const int64_t size = os::lseek(fd, 0, SEEK_END);
107 os::seek_to_file_offset(fd, current_offset);
108 return size;
109 }
110
111 class RepositoryIterator : public StackObj {
112 private:
113 const char* const _repo;
114 const size_t _repository_len;
115 GrowableArray<const char*>* _files;
116 const char* const fully_qualified(const char* entry) const;
117 mutable int _iterator;
118
119 public:
120 RepositoryIterator(const char* repository, size_t repository_len);
121 ~RepositoryIterator() {}
122 const char* const filter(const char* entry) const;
123 bool has_next() const;
124 const char* const next() const;
125 };
126
127 const char* const RepositoryIterator::fully_qualified(const char* entry) const {
128 assert(NULL != entry, "invariant");
129 char* file_path_entry = NULL;
130 // only use files that have content, not placeholders
131 const char* const file_separator = os::file_separator();
132 if (NULL != file_separator) {
133 const size_t entry_len = strlen(entry);
134 const size_t file_separator_length = strlen(file_separator);
135 const size_t file_path_entry_length = _repository_len + file_separator_length + entry_len;
136 file_path_entry = NEW_RESOURCE_ARRAY_RETURN_NULL(char, file_path_entry_length + 1);
137 if (NULL == file_path_entry) {
138 return NULL;
139 }
140 int position = 0;
141 position += jio_snprintf(&file_path_entry[position], _repository_len + 1, "%s", _repo);
142 position += jio_snprintf(&file_path_entry[position], file_separator_length + 1, "%s", os::file_separator());
143 position += jio_snprintf(&file_path_entry[position], entry_len + 1, "%s", entry);
144 file_path_entry[position] = '\0';
145 assert((size_t)position == file_path_entry_length, "invariant");
146 assert(strlen(file_path_entry) == (size_t)position, "invariant");
147 }
148 return file_path_entry;
149 }
150
151 const char* const RepositoryIterator::filter(const char* entry) const {
152 if (entry == NULL) {
153 return NULL;
154 }
155 const size_t entry_len = strlen(entry);
156 if (entry_len <= 2) {
157 // for "." and ".."
158 return NULL;
159 }
160 char* entry_name = NEW_RESOURCE_ARRAY_RETURN_NULL(char, entry_len + 1);
161 if (entry_name == NULL) {
162 return NULL;
163 }
164 strncpy(entry_name, entry, entry_len + 1);
165 const char* const fully_qualified_path_entry = fully_qualified(entry_name);
166 if (NULL == fully_qualified_path_entry) {
167 return NULL;
168 }
169 const fio_fd entry_fd = open_exclusivly(fully_qualified_path_entry);
170 if (invalid_fd == entry_fd) {
171 return NULL;
172 }
173 const int64_t entry_size = file_size(entry_fd);
174 os::close(entry_fd);
175 if (0 == entry_size) {
176 return NULL;
177 }
178 return entry_name;
179 }
180
181 RepositoryIterator::RepositoryIterator(const char* repository, size_t repository_len) :
182 _repo(repository),
183 _repository_len(repository_len),
184 _files(NULL),
185 _iterator(0) {
186 if (NULL != _repo) {
187 assert(strlen(_repo) == _repository_len, "invariant");
188 _files = new GrowableArray<const char*>(10);
189 DIR* dirp = os::opendir(_repo);
190 if (dirp == NULL) {
191 log_error(jfr, system)("Unable to open repository %s", _repo);
192 return;
193 }
194 struct dirent* dentry;
195 while ((dentry = os::readdir(dirp)) != NULL) {
196 const char* const entry_path = filter(dentry->d_name);
197 if (NULL != entry_path) {
198 _files->append(entry_path);
199 }
200 }
201 os::closedir(dirp);
202 if (_files->length() > 1) {
203 _files->sort(file_sort);
204 }
205 }
206 }
207
208 bool RepositoryIterator::has_next() const {
209 return (_files != NULL && _iterator < _files->length());
210 }
211
212 const char* const RepositoryIterator::next() const {
213 return _iterator >= _files->length() ? NULL : fully_qualified(_files->at(_iterator++));
214 }
215
216 static void write_emergency_file(const RepositoryIterator& iterator) {
217 assert(emergency_fd != invalid_fd, "invariant");
218 const size_t size_of_file_copy_block = 1 * M; // 1 mb
219 jbyte* const file_copy_block = NEW_RESOURCE_ARRAY_RETURN_NULL(jbyte, size_of_file_copy_block);
220 if (file_copy_block == NULL) {
221 return;
222 }
223 while (iterator.has_next()) {
224 fio_fd current_fd = invalid_fd;
225 const char* const fqn = iterator.next();
226 if (fqn != NULL) {
227 current_fd = open_exclusivly(fqn);
228 if (current_fd != invalid_fd) {
229 const int64_t current_filesize = file_size(current_fd);
230 assert(current_filesize > 0, "invariant");
231 int64_t bytes_read = 0;
232 int64_t bytes_written = 0;
233 while (bytes_read < current_filesize) {
234 const ssize_t read_result = os::read_at(current_fd, file_copy_block, size_of_file_copy_block, bytes_read);
235 if (-1 == read_result) {
236 log_info(jfr)( // For user, should not be "jfr, system"
237 "Unable to recover JFR data");
238 break;
239 }
240 bytes_read += (int64_t)read_result;
241 assert(bytes_read - bytes_written <= (int64_t)size_of_file_copy_block, "invariant");
242 bytes_written += (int64_t)os::write(emergency_fd, file_copy_block, bytes_read - bytes_written);
243 assert(bytes_read == bytes_written, "invariant");
244 }
245 os::close(current_fd);
246 }
247 }
248 }
249 }
250
251 // "emergency_dump_path" have to be allocated JVM_MAXPATHLEN or larger.
252 static void create_emergency_dump_path(char* emergency_dump_path) {
253 char buffer[JVM_MAXPATHLEN];
254 emergency_dump_path[0] = '\0';
255 const char* const cwd = os::get_current_directory(buffer, JVM_MAXPATHLEN);
256 if (NULL == cwd) {
257 return;
258 }
259 size_t pos = strlen(cwd);
260 const int fsep_len = jio_snprintf(&buffer[pos], JVM_MAXPATHLEN - pos, "%s", os::file_separator());
261 const char* filename_fmt = NULL;
262 // fetch specific error cause
263 switch (JfrJavaSupport::cause()) {
264 case JfrJavaSupport::OUT_OF_MEMORY:
265 filename_fmt = vm_oom_filename_fmt;
266 break;
267 case JfrJavaSupport::STACK_OVERFLOW:
268 filename_fmt = vm_soe_filename_fmt;
269 break;
270 default:
271 filename_fmt = vm_error_filename_fmt;
272 }
273 pos += fsep_len;
274 if (Arguments::copy_expand_pid(filename_fmt, strlen(filename_fmt), &buffer[pos], JVM_MAXPATHLEN - pos)) {
275 strncpy(emergency_dump_path, buffer, JVM_MAXPATHLEN);
276 }
277 if (emergency_dump_path[0] != '\0') {
278 log_info(jfr)( // For user, should not be "jfr, system"
279 "Attempting to recover JFR data, emergency jfr file: %s", emergency_dump_path);
280 }
281 }
282
283 // Caller needs ResourceMark
284 static const char* create_emergency_chunk_path(const char* repository_path) {
285 assert(repository_path != NULL, "invariant");
286 const size_t repository_path_len = strlen(repository_path);
287 // date time
288 char date_time_buffer[32] = { 0 };
289 date_time(date_time_buffer, sizeof(date_time_buffer));
290 size_t date_time_len = strlen(date_time_buffer);
291 size_t chunkname_max_len = repository_path_len // repository_base_path
292 + 1 // "/"
293 + date_time_len // date_time
294 + strlen(chunk_file_jfr_ext) // .jfr
295 + 1;
296 char* chunk_path = NEW_RESOURCE_ARRAY_RETURN_NULL(char, chunkname_max_len);
297 if (chunk_path == NULL) {
298 return NULL;
299 }
300 // append the individual substrings
301 jio_snprintf(chunk_path, chunkname_max_len, "%s%s%s%s", repository_path, os::file_separator(), date_time_buffer, chunk_file_jfr_ext);
302 return chunk_path;
303 }
304
305 void JfrEmergencyDump::setup_emergency_dump_file_descriptor() {
306 assert(emergency_fd == invalid_fd, "invariant");
307 char emergency_dump_path[JVM_MAXPATHLEN];
308 create_emergency_dump_path(emergency_dump_path);
309 emergency_fd = (emergency_dump_path[0] != '\0') ? open_exclusivly(emergency_dump_path) : invalid_fd;
310 }
311
312 // Caller needs ResourceMark
313 const char* JfrEmergencyDump::build_dump_path(const char* repository_path) {
314 if (repository_path == NULL) {
315 char* dump_path = NEW_RESOURCE_ARRAY_RETURN_NULL(char, JVM_MAXPATHLEN);
316 if (dump_path == NULL) {
317 return NULL;
318 }
319 create_emergency_dump_path(dump_path);
320 return dump_path;
321 } else {
322 return create_emergency_chunk_path(repository_path);
323 }
324 }
325
326 void JfrEmergencyDump::on_vm_error(const char* repository_path) {
327 assert(repository_path != NULL, "invariant");
328 ResourceMark rm;
329 if (emergency_fd != invalid_fd) {
330 RepositoryIterator iterator(repository_path, strlen(repository_path));
331 write_emergency_file(iterator);
332 os::close(emergency_fd);
333 emergency_fd = invalid_fd;
334 }
335 }
336
337 /*
338 * We are just about to exit the VM, so we will be very aggressive
339 * at this point in order to increase overall success of dumping jfr data.
340 *
341 * If we end up deadlocking in the attempt of dumping out jfr data,
342 * we rely on the WatcherThread task "is_error_reported()",
343 * to exit the VM after a hard-coded timeout (disallow WatcherThread to emergency dump).
344 * This "safety net" somewhat explains the aggressiveness in this attempt.
345 *
346 */
347 static bool prepare_for_emergency_dump(Thread* thread) {
348 assert(thread != NULL, "invariant");
349
350 if (thread->is_Watcher_thread()) {
351 // need WatcherThread as a safeguard against potential deadlocks
352 return false;
353 }
354 if (JfrStream_lock->owned_by_self()) {
355 // crashed during jfr rotation, disallow recursion
356 return false;
357 }
358
359 #ifdef ASSERT
360 Mutex* owned_lock = thread->owned_locks();
361 while (owned_lock != NULL) {
362 Mutex* next = owned_lock->next();
363 owned_lock->unlock();
364 owned_lock = next;
365 }
366 #endif // ASSERT
367
368 if (Threads_lock->owned_by_self()) {
369 Threads_lock->unlock();
370 }
371
372 if (Module_lock->owned_by_self()) {
373 Module_lock->unlock();
374 }
375
376 if (ClassLoaderDataGraph_lock->owned_by_self()) {
377 ClassLoaderDataGraph_lock->unlock();
378 }
379
380 if (Heap_lock->owned_by_self()) {
381 Heap_lock->unlock();
382 }
383
384 if (VMOperationQueue_lock->owned_by_self()) {
385 VMOperationQueue_lock->unlock();
386 }
387
388 if (VMOperationRequest_lock->owned_by_self()) {
389 VMOperationRequest_lock->unlock();
390 }
391
392 if (Service_lock->owned_by_self()) {
393 Service_lock->unlock();
394 }
395
396 if (UseNotificationThread && Notification_lock->owned_by_self()) {
397 Notification_lock->unlock();
398 }
399
400 if (CodeCache_lock->owned_by_self()) {
401 CodeCache_lock->unlock();
402 }
403
404 if (PeriodicTask_lock->owned_by_self()) {
405 PeriodicTask_lock->unlock();
406 }
407
408 if (JfrMsg_lock->owned_by_self()) {
409 JfrMsg_lock->unlock();
410 }
411
412 if (JfrBuffer_lock->owned_by_self()) {
413 JfrBuffer_lock->unlock();
414 }
415
416 if (JfrStacktrace_lock->owned_by_self()) {
417 JfrStacktrace_lock->unlock();
418 }
419 return true;
420 }
421
422 static volatile int jfr_shutdown_lock = 0;
423
424 static bool guard_reentrancy() {
425 return Atomic::cmpxchg(&jfr_shutdown_lock, 0, 1) == 0;
426 }
427
428 class JavaThreadInVM : public StackObj {
429 private:
430 JavaThread* const _jt;
431 JavaThreadState _original_state;
432 public:
433
434 JavaThreadInVM(Thread* t) : _jt(t->is_Java_thread() ? (JavaThread*)t : NULL),
435 _original_state(_thread_max_state) {
436 if ((_jt != NULL) && (_jt->thread_state() != _thread_in_vm)) {
437 _original_state = _jt->thread_state();
438 _jt->set_thread_state(_thread_in_vm);
439 }
440 }
441
442 ~JavaThreadInVM() {
443 if (_original_state != _thread_max_state) {
444 _jt->set_thread_state(_original_state);
445 }
446 }
447
448 };
449
450 void JfrEmergencyDump::on_vm_shutdown(bool exception_handler) {
451 if (!guard_reentrancy()) {
452 return;
453 }
454
455 Thread* thread = Thread::current_or_null_safe();
456 if (thread == NULL) {
457 return;
458 }
459 // Ensure a JavaThread is _thread_in_vm when we make this call
460 JavaThreadInVM jtivm(thread);
461 if (!prepare_for_emergency_dump(thread)) {
462 return;
463 }
464
465 EventDumpReason event;
466 if (event.should_commit()) {
467 event.set_reason(exception_handler ? "Crash" : "Out of Memory");
468 event.set_recordingId(-1);
469 event.commit();
470 }
471 if (!exception_handler) {
472 // OOM
473 LeakProfiler::emit_events(max_jlong, false);
474 }
475 const int messages = MSGBIT(MSG_VM_ERROR);
476 JfrRecorderService service;
477 service.rotate(messages);
478 }