1 /*
2 * Copyright (c) 2020, 2024, 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 #ifndef SHARE_CDS_ARCHIVEBUILDER_HPP
26 #define SHARE_CDS_ARCHIVEBUILDER_HPP
27
28 #include "cds/archiveUtils.hpp"
29 #include "cds/dumpAllocStats.hpp"
30 #include "memory/metaspaceClosure.hpp"
31 #include "oops/array.hpp"
32 #include "oops/klass.hpp"
33 #include "runtime/os.hpp"
34 #include "utilities/bitMap.hpp"
35 #include "utilities/growableArray.hpp"
36 #include "utilities/resizeableResourceHash.hpp"
37 #include "utilities/resourceHash.hpp"
38
39 class ArchiveHeapInfo;
40 class CHeapBitMap;
41 class FileMapInfo;
42 class Klass;
43 class MemRegion;
44 class Symbol;
45
46 // Metaspace::allocate() requires that all blocks must be aligned with KlassAlignmentInBytes.
47 // We enforce the same alignment rule in blocks allocated from the shared space.
48 const int SharedSpaceObjectAlignment = KlassAlignmentInBytes;
49
50 // Overview of CDS archive creation (for both static and dynamic dump):
51 //
52 // [1] Load all classes (static dump: from the classlist, dynamic dump: as part of app execution)
53 // [2] Allocate "output buffer"
54 // [3] Copy contents of the 2 "core" regions (rw/ro) into the output buffer.
55 // - allocate the cpp vtables in rw (static dump only)
56 // - memcpy the MetaspaceObjs into rw/ro:
57 // dump_rw_region();
58 // dump_ro_region();
59 // - fix all the pointers in the MetaspaceObjs to point to the copies
60 // relocate_metaspaceobj_embedded_pointers()
61 // [4] Copy symbol table, dictionary, etc, into the ro region
62 // [5] Relocate all the pointers in rw/ro, so that the archive can be mapped to
63 // the "requested" location without runtime relocation. See relocate_to_requested()
64 //
65 // "source" vs "buffered" vs "requested"
66 //
67 // The ArchiveBuilder deals with three types of addresses.
68 //
69 // "source": These are the addresses of objects created in step [1] above. They are the actual
70 // InstanceKlass*, Method*, etc, of the Java classes that are loaded for executing
71 // Java bytecodes in the JVM process that's dumping the CDS archive.
72 //
73 // It may be necessary to contiue Java execution after ArchiveBuilder is finished.
74 // Therefore, we don't modify any of the "source" objects.
75 //
76 // "buffered": The "source" objects that are deemed archivable are copied into a temporary buffer.
77 // Objects in the buffer are modified in steps [2, 3, 4] (e.g., unshareable info is
78 // removed, pointers are relocated, etc) to prepare them to be loaded at runtime.
79 //
80 // "requested": These are the addreses where the "buffered" objects should be loaded at runtime.
81 // When the "buffered" objects are written into the archive file, their addresses
82 // are adjusted in step [5] such that the lowest of these objects would be mapped
83 // at SharedBaseAddress.
84 //
85 // Translation between "source" and "buffered" addresses is done with two hashtables:
86 // _src_obj_table : "source" -> "buffered"
87 // _buffered_to_src_table : "buffered" -> "source"
88 //
89 // Translation between "buffered" and "requested" addresses is done with a simple shift:
90 // buffered_address + _buffer_to_requested_delta == requested_address
91 //
92 class ArchiveBuilder : public StackObj {
93 protected:
94 DumpRegion* _current_dump_region;
95 address _buffer_bottom; // for writing the contents of rw/ro regions
96 address _last_verified_top;
97 int _num_dump_regions_used;
98 size_t _other_region_used_bytes;
99
100 // These are the addresses where we will request the static and dynamic archives to be
101 // mapped at run time. If the request fails (due to ASLR), we will map the archives at
102 // os-selected addresses.
103 address _requested_static_archive_bottom; // This is determined solely by the value of
104 // SharedBaseAddress during -Xshare:dump.
105 address _requested_static_archive_top;
106 address _requested_dynamic_archive_bottom; // Used only during dynamic dump. It's placed
107 // immediately above _requested_static_archive_top.
108 address _requested_dynamic_archive_top;
109
110 // (Used only during dynamic dump) where the static archive is actually mapped. This
111 // may be different than _requested_static_archive_{bottom,top} due to ASLR
112 address _mapped_static_archive_bottom;
113 address _mapped_static_archive_top;
114
115 intx _buffer_to_requested_delta;
116
117 DumpRegion* current_dump_region() const { return _current_dump_region; }
118
119 public:
120 enum FollowMode {
121 make_a_copy, point_to_it, set_to_null
122 };
123
124 private:
125 class SourceObjInfo {
126 uintx _ptrmap_start; // The bit-offset of the start of this object (inclusive)
127 uintx _ptrmap_end; // The bit-offset of the end of this object (exclusive)
128 bool _read_only;
129 bool _has_embedded_pointer;
130 FollowMode _follow_mode;
131 int _size_in_bytes;
132 int _id; // Each object has a unique serial ID, starting from zero. The ID is assigned
133 // when the object is added into _source_objs.
134 MetaspaceObj::Type _msotype;
135 address _source_addr; // The source object to be copied.
136 address _buffered_addr; // The copy of this object insider the buffer.
137 public:
138 SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
139 _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _has_embedded_pointer(false), _follow_mode(follow_mode),
140 _size_in_bytes(ref->size() * BytesPerWord), _id(0), _msotype(ref->msotype()),
141 _source_addr(ref->obj()) {
142 if (follow_mode == point_to_it) {
143 _buffered_addr = ref->obj();
144 } else {
145 _buffered_addr = nullptr;
146 }
147 }
148 SourceObjInfo(address src, address buf) {
149 _source_addr = src;
150 _buffered_addr = buf;
151 }
152
153 // This constructor is only used for regenerated objects (created by LambdaFormInvokers, etc).
154 // src = address of a Method or InstanceKlass that has been regenerated.
155 // renegerated_obj_info = info for the regenerated version of src.
156 SourceObjInfo(address src, SourceObjInfo* renegerated_obj_info) :
157 _ptrmap_start(0), _ptrmap_end(0), _read_only(false),
158 _follow_mode(renegerated_obj_info->_follow_mode),
159 _size_in_bytes(0), _msotype(renegerated_obj_info->_msotype),
160 _source_addr(src), _buffered_addr(renegerated_obj_info->_buffered_addr) {}
161
162 bool should_copy() const { return _follow_mode == make_a_copy; }
163 void set_buffered_addr(address addr) {
164 assert(should_copy(), "must be");
165 assert(_buffered_addr == nullptr, "cannot be copied twice");
166 assert(addr != nullptr, "must be a valid copy");
167 _buffered_addr = addr;
168 }
169 void set_ptrmap_start(uintx v) { _ptrmap_start = v; }
170 void set_ptrmap_end(uintx v) { _ptrmap_end = v; }
171 uintx ptrmap_start() const { return _ptrmap_start; } // inclusive
172 uintx ptrmap_end() const { return _ptrmap_end; } // exclusive
173 bool read_only() const { return _read_only; }
174 bool has_embedded_pointer() const { return _has_embedded_pointer; }
175 void set_has_embedded_pointer() { _has_embedded_pointer = true; }
176 int size_in_bytes() const { return _size_in_bytes; }
177 int id() const { return _id; }
178 void set_id(int i) { _id = i; }
179 address source_addr() const { return _source_addr; }
180 address buffered_addr() const {
181 if (_follow_mode != set_to_null) {
182 assert(_buffered_addr != nullptr, "must be initialized");
183 }
184 return _buffered_addr;
185 }
186 MetaspaceObj::Type msotype() const { return _msotype; }
187 };
188
189 class SourceObjList {
190 uintx _total_bytes;
191 GrowableArray<SourceObjInfo*>* _objs; // Source objects to be archived
192 CHeapBitMap _ptrmap; // Marks the addresses of the pointer fields
193 // in the source objects
194 public:
195 SourceObjList();
196 ~SourceObjList();
197
198 GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
199
200 void append(SourceObjInfo* src_info);
201 void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
202 void relocate(int i, ArchiveBuilder* builder);
203
204 // convenience accessor
205 SourceObjInfo* at(int i) const { return objs()->at(i); }
206 };
207
208 class CDSMapLogger;
209
210 static const int INITIAL_TABLE_SIZE = 15889;
211 static const int MAX_TABLE_SIZE = 1000000;
212
213 ReservedSpace _shared_rs;
214 VirtualSpace _shared_vs;
215
216 DumpRegion _rw_region;
217 DumpRegion _ro_region;
218 DumpRegion _cc_region;
219
220 // Combined bitmap to track pointers in both RW and RO regions. This is updated
221 // as objects are copied into RW and RO.
222 CHeapBitMap _ptrmap;
223
224 // _ptrmap is split into these two bitmaps which are written into the archive.
225 CHeapBitMap _rw_ptrmap; // marks pointers in the RW region
226 CHeapBitMap _ro_ptrmap; // marks pointers in the RO region
227 CHeapBitMap _cc_ptrmap; // marks pointers in the CC region
228
229 SourceObjList _rw_src_objs; // objs to put in rw region
230 SourceObjList _ro_src_objs; // objs to put in ro region
231 ResizeableResourceHashtable<address, SourceObjInfo, AnyObj::C_HEAP, mtClassShared> _src_obj_table;
232 ResizeableResourceHashtable<address, address, AnyObj::C_HEAP, mtClassShared> _buffered_to_src_table;
233 GrowableArray<Klass*>* _klasses;
234 GrowableArray<Symbol*>* _symbols;
235 unsigned int _entropy_seed;
236
237 // statistics
238 DumpAllocStats _alloc_stats;
239 size_t _total_heap_region_size;
240
241 void print_region_stats(FileMapInfo *map_info, ArchiveHeapInfo* heap_info);
242 void print_bitmap_region_stats(size_t size, size_t total_size);
243 void print_heap_region_stats(ArchiveHeapInfo* heap_info, size_t total_size);
244
245 // For global access.
246 static ArchiveBuilder* _current;
247
248 public:
249 // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
250 // These are usually for misc tables that are allocated in the RO space.
251 class OtherROAllocMark {
252 char* _oldtop;
253 public:
254 OtherROAllocMark() {
255 _oldtop = _current->_ro_region.top();
256 }
257 ~OtherROAllocMark();
258 };
259
260 private:
261 FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
262
263 void iterate_sorted_roots(MetaspaceClosure* it);
264 void sort_klasses();
265 static int compare_symbols_by_address(Symbol** a, Symbol** b);
266 static int compare_klass_by_name(Klass** a, Klass** b);
267 void update_hidden_class_loader_type(InstanceKlass* ik) NOT_CDS_JAVA_HEAP_RETURN;
268
269 void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
270 void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
271
272 void relocate_embedded_pointers(SourceObjList* src_objs);
273
274 bool is_excluded(Klass* k);
275 void clean_up_src_obj_table();
276
277 protected:
278 virtual void iterate_roots(MetaspaceClosure* it) = 0;
279
280 // Conservative estimate for number of bytes needed for:
281 size_t _estimated_metaspaceobj_bytes; // all archived MetaspaceObj's.
282 size_t _estimated_hashtable_bytes; // symbol table and dictionaries
283
284 static const int _total_dump_regions = 2;
285
286 size_t estimate_archive_size();
287
288 void start_dump_region(DumpRegion* next);
289 void verify_estimate_size(size_t estimate, const char* which);
290
291 public:
292 address reserve_buffer();
293
294 address buffer_bottom() const { return _buffer_bottom; }
295 address buffer_top() const { return (address)current_dump_region()->top(); }
296 address requested_static_archive_bottom() const { return _requested_static_archive_bottom; }
297 address mapped_static_archive_bottom() const { return _mapped_static_archive_bottom; }
298 intx buffer_to_requested_delta() const { return _buffer_to_requested_delta; }
299
300 bool is_in_buffer_space(address p) const {
301 if (current_dump_region() == nullptr) {
302 return false;
303 }
304 return (buffer_bottom() <= p && p < buffer_top());
305 }
306
307 template <typename T> bool is_in_requested_static_archive(T p) const {
308 return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
309 }
310
311 template <typename T> bool is_in_mapped_static_archive(T p) const {
312 return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
313 }
314
315 template <typename T> bool is_in_buffer_space(T obj) const {
316 return is_in_buffer_space(address(obj));
317 }
318
319 template <typename T> T to_requested(T obj) const {
320 assert(is_in_buffer_space(obj), "must be");
321 return (T)(address(obj) + _buffer_to_requested_delta);
322 }
323
324 static intx get_buffer_to_requested_delta() {
325 return current()->buffer_to_requested_delta();
326 }
327
328 inline static u4 to_offset_u4(uintx offset) {
329 guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
330 return (u4)offset;
331 }
332
333 public:
334 static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
335
336 // The address p points to an object inside the output buffer. When the archive is mapped
337 // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
338 uintx buffer_to_offset(address p) const;
339
340 // Same as buffer_to_offset, except that the address p points to either (a) an object
341 // inside the output buffer, or (b), an object in the currently mapped static archive.
342 uintx any_to_offset(address p) const;
343
344 template <typename T>
345 u4 buffer_to_offset_u4(T p) const {
346 uintx offset = buffer_to_offset((address)p);
347 return to_offset_u4(offset);
348 }
349
350 template <typename T>
351 u4 any_to_offset_u4(T p) const {
352 uintx offset = any_to_offset((address)p);
353 return to_offset_u4(offset);
354 }
355
356 public:
357 ArchiveBuilder();
358 ~ArchiveBuilder();
359
360 int entropy();
361 void gather_klasses_and_symbols();
362 void gather_source_objs();
363 bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
364 bool gather_one_source_obj(MetaspaceClosure::Ref* ref, bool read_only);
365 void remember_embedded_pointer_in_enclosing_obj(MetaspaceClosure::Ref* ref);
366 static void serialize_dynamic_archivable_items(SerializeClosure* soc);
367
368 DumpRegion* rw_region() { return &_rw_region; }
369 DumpRegion* ro_region() { return &_ro_region; }
370 DumpRegion* cc_region() { return &_cc_region; }
371
372 void start_cc_region();
373 void end_cc_region();
374
375 static char* rw_region_alloc(size_t num_bytes) {
376 return current()->rw_region()->allocate(num_bytes);
377 }
378 static char* ro_region_alloc(size_t num_bytes) {
379 return current()->ro_region()->allocate(num_bytes);
380 }
381 static char* cc_region_alloc(size_t num_bytes) {
382 return current()->cc_region()->allocate(num_bytes);
383 }
384
385 template <typename T>
386 static Array<T>* new_ro_array(int length) {
387 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
388 Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
389 array->initialize(length);
390 return array;
391 }
392
393 template <typename T>
394 static Array<T>* new_rw_array(int length) {
395 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
396 Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
397 array->initialize(length);
398 return array;
399 }
400
401 template <typename T>
402 static size_t ro_array_bytesize(int length) {
403 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
404 return align_up(byte_size, SharedSpaceObjectAlignment);
405 }
406
407 char* ro_strdup(const char* s);
408
409 static int compare_src_objs(SourceObjInfo** a, SourceObjInfo** b);
410 void sort_metadata_objs();
411 void dump_rw_metadata();
412 void dump_ro_metadata();
413 void relocate_metaspaceobj_embedded_pointers();
414 void record_regenerated_object(address orig_src_obj, address regen_src_obj);
415 void make_klasses_shareable();
416 void make_training_data_shareable();
417 void relocate_to_requested();
418 void write_archive(FileMapInfo* mapinfo, ArchiveHeapInfo* heap_info);
419 void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
420 bool read_only, bool allow_exec);
421
422 void write_pointer_in_buffer(address* ptr_location, address src_addr);
423 template <typename T> void write_pointer_in_buffer(T* ptr_location, T src_addr) {
424 write_pointer_in_buffer((address*)ptr_location, (address)src_addr);
425 }
426
427 void mark_and_relocate_to_buffered_addr(address* ptr_location);
428 template <typename T> void mark_and_relocate_to_buffered_addr(T ptr_location) {
429 mark_and_relocate_to_buffered_addr((address*)ptr_location);
430 }
431
432 bool has_been_archived(address src_addr) const;
433 address get_buffered_addr(address src_addr) const;
434 template <typename T> T get_buffered_addr(T src_addr) const {
435 return (T)get_buffered_addr((address)src_addr);
436 }
437
438 address get_source_addr(address buffered_addr) const;
439 template <typename T> T get_source_addr(T buffered_addr) const {
440 return (T)get_source_addr((address)buffered_addr);
441 }
442
443 // All klasses and symbols that will be copied into the archive
444 GrowableArray<Klass*>* klasses() const { return _klasses; }
445 GrowableArray<Symbol*>* symbols() const { return _symbols; }
446
447 static bool is_active() {
448 CDS_ONLY(return (_current != nullptr));
449 NOT_CDS(return false;)
450 }
451
452 static ArchiveBuilder* current() {
453 assert(_current != nullptr, "ArchiveBuilder must be active");
454 return _current;
455 }
456
457 static DumpAllocStats* alloc_stats() {
458 return &(current()->_alloc_stats);
459 }
460
461 static CompactHashtableStats* symbol_stats() {
462 return alloc_stats()->symbol_stats();
463 }
464
465 static CompactHashtableStats* string_stats() {
466 return alloc_stats()->string_stats();
467 }
468
469 narrowKlass get_requested_narrow_klass(Klass* k);
470
471 static Klass* get_buffered_klass(Klass* src_klass) {
472 Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
473 assert(klass != nullptr && klass->is_klass(), "must be");
474 return klass;
475 }
476
477 static Symbol* get_buffered_symbol(Symbol* src_symbol) {
478 return (Symbol*)current()->get_buffered_addr((address)src_symbol);
479 }
480
481 void print_stats();
482 void report_out_of_space(const char* name, size_t needed_bytes);
483 };
484
485 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP