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_space;
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_space() const { return _current_dump_space; }
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 FollowMode _follow_mode;
130 int _size_in_bytes;
131 MetaspaceObj::Type _msotype;
132 address _source_addr; // The source object to be copied.
133 address _buffered_addr; // The copy of this object insider the buffer.
134 public:
135 SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
136 _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _follow_mode(follow_mode),
137 _size_in_bytes(ref->size() * BytesPerWord), _msotype(ref->msotype()),
138 _source_addr(ref->obj()) {
139 if (follow_mode == point_to_it) {
140 _buffered_addr = ref->obj();
141 } else {
142 _buffered_addr = nullptr;
143 }
144 }
145
146 // This constructor is only used for regenerated objects (created by LambdaFormInvokers, etc).
147 // src = address of a Method or InstanceKlass that has been regenerated.
148 // renegerated_obj_info = info for the regenerated version of src.
149 SourceObjInfo(address src, SourceObjInfo* renegerated_obj_info) :
150 _ptrmap_start(0), _ptrmap_end(0), _read_only(false),
151 _follow_mode(renegerated_obj_info->_follow_mode),
152 _size_in_bytes(0), _msotype(renegerated_obj_info->_msotype),
153 _source_addr(src), _buffered_addr(renegerated_obj_info->_buffered_addr) {}
154
155 bool should_copy() const { return _follow_mode == make_a_copy; }
156 void set_buffered_addr(address addr) {
157 assert(should_copy(), "must be");
158 assert(_buffered_addr == nullptr, "cannot be copied twice");
159 assert(addr != nullptr, "must be a valid copy");
160 _buffered_addr = addr;
161 }
162 void set_ptrmap_start(uintx v) { _ptrmap_start = v; }
163 void set_ptrmap_end(uintx v) { _ptrmap_end = v; }
164 uintx ptrmap_start() const { return _ptrmap_start; } // inclusive
165 uintx ptrmap_end() const { return _ptrmap_end; } // exclusive
166 bool read_only() const { return _read_only; }
167 int size_in_bytes() const { return _size_in_bytes; }
168 address source_addr() const { return _source_addr; }
169 address buffered_addr() const {
170 if (_follow_mode != set_to_null) {
171 assert(_buffered_addr != nullptr, "must be initialized");
172 }
173 return _buffered_addr;
174 }
175 MetaspaceObj::Type msotype() const { return _msotype; }
176 };
177
178 class SourceObjList {
179 uintx _total_bytes;
180 GrowableArray<SourceObjInfo*>* _objs; // Source objects to be archived
181 CHeapBitMap _ptrmap; // Marks the addresses of the pointer fields
182 // in the source objects
183 public:
184 SourceObjList();
185 ~SourceObjList();
186
187 GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
188
189 void append(SourceObjInfo* src_info);
190 void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
191 void relocate(int i, ArchiveBuilder* builder);
192
193 // convenience accessor
194 SourceObjInfo* at(int i) const { return objs()->at(i); }
195 };
196
197 class CDSMapLogger;
198
199 static const int INITIAL_TABLE_SIZE = 15889;
200 static const int MAX_TABLE_SIZE = 1000000;
201
202 ReservedSpace _shared_rs;
203 VirtualSpace _shared_vs;
204
205 DumpRegion _rw_region;
206 DumpRegion _ro_region;
207
208 // Combined bitmap to track pointers in both RW and RO regions. This is updated
209 // as objects are copied into RW and RO.
210 CHeapBitMap _ptrmap;
211
212 // _ptrmap is split into these two bitmaps which are written into the archive.
213 CHeapBitMap _rw_ptrmap; // marks pointers in the RW region
214 CHeapBitMap _ro_ptrmap; // marks pointers in the RO region
215
216 SourceObjList _rw_src_objs; // objs to put in rw region
217 SourceObjList _ro_src_objs; // objs to put in ro region
218 ResizeableResourceHashtable<address, SourceObjInfo, AnyObj::C_HEAP, mtClassShared> _src_obj_table;
219 ResizeableResourceHashtable<address, address, AnyObj::C_HEAP, mtClassShared> _buffered_to_src_table;
220 GrowableArray<Klass*>* _klasses;
221 GrowableArray<Symbol*>* _symbols;
222
223 // statistics
224 DumpAllocStats _alloc_stats;
225 size_t _total_heap_region_size;
226
227 void print_region_stats(FileMapInfo *map_info, ArchiveHeapInfo* heap_info);
228 void print_bitmap_region_stats(size_t size, size_t total_size);
229 void print_heap_region_stats(ArchiveHeapInfo* heap_info, size_t total_size);
230
231 // For global access.
232 static ArchiveBuilder* _current;
233
234 public:
235 // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
236 // These are usually for misc tables that are allocated in the RO space.
237 class OtherROAllocMark {
238 char* _oldtop;
239 public:
240 OtherROAllocMark() {
241 _oldtop = _current->_ro_region.top();
242 }
243 ~OtherROAllocMark();
244 };
245
246 private:
247 FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
248
249 void iterate_sorted_roots(MetaspaceClosure* it);
250 void sort_klasses();
251 static int compare_symbols_by_address(Symbol** a, Symbol** b);
252 static int compare_klass_by_name(Klass** a, Klass** b);
253
254 void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
255 void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
256
257 void relocate_embedded_pointers(SourceObjList* src_objs);
258
259 bool is_excluded(Klass* k);
260 void clean_up_src_obj_table();
261
262 protected:
263 virtual void iterate_roots(MetaspaceClosure* it) = 0;
264
265 // Conservative estimate for number of bytes needed for:
266 size_t _estimated_metaspaceobj_bytes; // all archived MetaspaceObj's.
267 size_t _estimated_hashtable_bytes; // symbol table and dictionaries
268
269 static const int _total_dump_regions = 2;
270
271 size_t estimate_archive_size();
272
273 void start_dump_space(DumpRegion* next);
274 void verify_estimate_size(size_t estimate, const char* which);
275
276 public:
277 address reserve_buffer();
278
279 address buffer_bottom() const { return _buffer_bottom; }
280 address buffer_top() const { return (address)current_dump_space()->top(); }
281 address requested_static_archive_bottom() const { return _requested_static_archive_bottom; }
282 address mapped_static_archive_bottom() const { return _mapped_static_archive_bottom; }
283 intx buffer_to_requested_delta() const { return _buffer_to_requested_delta; }
284
285 bool is_in_buffer_space(address p) const {
286 return (buffer_bottom() <= p && p < buffer_top());
287 }
288
289 template <typename T> bool is_in_requested_static_archive(T p) const {
290 return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
291 }
292
293 template <typename T> bool is_in_mapped_static_archive(T p) const {
294 return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
295 }
296
297 template <typename T> bool is_in_buffer_space(T obj) const {
298 return is_in_buffer_space(address(obj));
299 }
300
301 template <typename T> T to_requested(T obj) const {
302 assert(is_in_buffer_space(obj), "must be");
303 return (T)(address(obj) + _buffer_to_requested_delta);
304 }
305
306 static intx get_buffer_to_requested_delta() {
307 return current()->buffer_to_requested_delta();
308 }
309
310 inline static u4 to_offset_u4(uintx offset) {
311 guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
312 return (u4)offset;
313 }
314
315 public:
316 static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
317
318 // The address p points to an object inside the output buffer. When the archive is mapped
319 // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
320 uintx buffer_to_offset(address p) const;
321
322 // Same as buffer_to_offset, except that the address p points to either (a) an object
323 // inside the output buffer, or (b), an object in the currently mapped static archive.
324 uintx any_to_offset(address p) const;
325
326 template <typename T>
327 u4 buffer_to_offset_u4(T p) const {
328 uintx offset = buffer_to_offset((address)p);
329 return to_offset_u4(offset);
330 }
331
332 template <typename T>
333 u4 any_to_offset_u4(T p) const {
334 uintx offset = any_to_offset((address)p);
335 return to_offset_u4(offset);
336 }
337
338 static void assert_is_vm_thread() PRODUCT_RETURN;
339
340 public:
341 ArchiveBuilder();
342 ~ArchiveBuilder();
343
344 void gather_klasses_and_symbols();
345 void gather_source_objs();
346 bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
347 bool gather_one_source_obj(MetaspaceClosure::Ref* ref, bool read_only);
348 void remember_embedded_pointer_in_enclosing_obj(MetaspaceClosure::Ref* ref);
349 static void serialize_dynamic_archivable_items(SerializeClosure* soc);
350
351 DumpRegion* rw_region() { return &_rw_region; }
352 DumpRegion* ro_region() { return &_ro_region; }
353
354 static char* rw_region_alloc(size_t num_bytes) {
355 return current()->rw_region()->allocate(num_bytes);
356 }
357 static char* ro_region_alloc(size_t num_bytes) {
358 return current()->ro_region()->allocate(num_bytes);
359 }
360
361 template <typename T>
362 static Array<T>* new_ro_array(int length) {
363 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
364 Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
365 array->initialize(length);
366 return array;
367 }
368
369 template <typename T>
370 static Array<T>* new_rw_array(int length) {
371 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
372 Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
373 array->initialize(length);
374 return array;
375 }
376
377 template <typename T>
378 static size_t ro_array_bytesize(int length) {
379 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
380 return align_up(byte_size, SharedSpaceObjectAlignment);
381 }
382
383 char* ro_strdup(const char* s);
384
385 void dump_rw_metadata();
386 void dump_ro_metadata();
387 void relocate_metaspaceobj_embedded_pointers();
388 void record_regenerated_object(address orig_src_obj, address regen_src_obj);
389 void make_klasses_shareable();
390 void relocate_to_requested();
391 void write_archive(FileMapInfo* mapinfo, ArchiveHeapInfo* heap_info);
392 void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
393 bool read_only, bool allow_exec);
394
395 void write_pointer_in_buffer(address* ptr_location, address src_addr);
396 template <typename T> void write_pointer_in_buffer(T* ptr_location, T src_addr) {
397 write_pointer_in_buffer((address*)ptr_location, (address)src_addr);
398 }
399
400 address get_buffered_addr(address src_addr) const;
401 template <typename T> T get_buffered_addr(T src_addr) const {
402 return (T)get_buffered_addr((address)src_addr);
403 }
404
405 address get_source_addr(address buffered_addr) const;
406 template <typename T> T get_source_addr(T buffered_addr) const {
407 return (T)get_source_addr((address)buffered_addr);
408 }
409
410 // All klasses and symbols that will be copied into the archive
411 GrowableArray<Klass*>* klasses() const { return _klasses; }
412 GrowableArray<Symbol*>* symbols() const { return _symbols; }
413
414 static bool is_active() {
415 return (_current != nullptr);
416 }
417
418 static ArchiveBuilder* current() {
419 assert_is_vm_thread();
420 assert(_current != nullptr, "ArchiveBuilder must be active");
421 return _current;
422 }
423
424 static DumpAllocStats* alloc_stats() {
425 return &(current()->_alloc_stats);
426 }
427
428 static CompactHashtableStats* symbol_stats() {
429 return alloc_stats()->symbol_stats();
430 }
431
432 static CompactHashtableStats* string_stats() {
433 return alloc_stats()->string_stats();
434 }
435
436 narrowKlass get_requested_narrow_klass(Klass* k);
437
438 static Klass* get_buffered_klass(Klass* src_klass) {
439 Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
440 assert(klass != nullptr && klass->is_klass(), "must be");
441 return klass;
442 }
443
444 static Symbol* get_buffered_symbol(Symbol* src_symbol) {
445 return (Symbol*)current()->get_buffered_addr((address)src_symbol);
446 }
447
448 void print_stats();
449 void report_out_of_space(const char* name, size_t needed_bytes);
450 };
451
452 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP