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