1 /*
2 * Copyright (c) 2020, 2025, 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/metaspace.hpp"
31 #include "memory/metaspaceClosure.hpp"
32 #include "memory/reservedSpace.hpp"
33 #include "memory/virtualspace.hpp"
34 #include "oops/array.hpp"
35 #include "oops/klass.hpp"
36 #include "runtime/os.hpp"
37 #include "utilities/bitMap.hpp"
38 #include "utilities/growableArray.hpp"
39 #include "utilities/resizeableResourceHash.hpp"
40 #include "utilities/resourceHash.hpp"
41
42 class ArchiveHeapInfo;
43 class CHeapBitMap;
44 class FileMapInfo;
45 class Klass;
46 class MemRegion;
47 class Symbol;
48
49 // The minimum alignment for non-Klass objects inside the CDS archive. Klass objects need
50 // to follow CompressedKlassPointers::klass_alignment_in_bytes().
51 constexpr size_t SharedSpaceObjectAlignment = Metaspace::min_allocation_alignment_bytes;
52
53 // Overview of CDS archive creation (for both static and dynamic dump):
54 //
55 // [1] Load all classes (static dump: from the classlist, dynamic dump: as part of app execution)
56 // [2] Allocate "output buffer"
57 // [3] Copy contents of the 2 "core" regions (rw/ro) into the output buffer.
58 // - allocate the cpp vtables in rw (static dump only)
59 // - memcpy the MetaspaceObjs into rw/ro:
60 // dump_rw_region();
61 // dump_ro_region();
62 // - fix all the pointers in the MetaspaceObjs to point to the copies
63 // relocate_metaspaceobj_embedded_pointers()
64 // [4] Copy symbol table, dictionary, etc, into the ro region
65 // [5] Relocate all the pointers in rw/ro, so that the archive can be mapped to
66 // the "requested" location without runtime relocation. See relocate_to_requested()
67 //
68 // "source" vs "buffered" vs "requested"
69 //
70 // The ArchiveBuilder deals with three types of addresses.
71 //
72 // "source": These are the addresses of objects created in step [1] above. They are the actual
73 // InstanceKlass*, Method*, etc, of the Java classes that are loaded for executing
74 // Java bytecodes in the JVM process that's dumping the CDS archive.
75 //
76 // It may be necessary to contiue Java execution after ArchiveBuilder is finished.
77 // Therefore, we don't modify any of the "source" objects.
78 //
79 // "buffered": The "source" objects that are deemed archivable are copied into a temporary buffer.
80 // Objects in the buffer are modified in steps [2, 3, 4] (e.g., unshareable info is
81 // removed, pointers are relocated, etc) to prepare them to be loaded at runtime.
82 //
83 // "requested": These are the addreses where the "buffered" objects should be loaded at runtime.
84 // When the "buffered" objects are written into the archive file, their addresses
85 // are adjusted in step [5] such that the lowest of these objects would be mapped
86 // at SharedBaseAddress.
87 //
88 // Translation between "source" and "buffered" addresses is done with two hashtables:
89 // _src_obj_table : "source" -> "buffered"
90 // _buffered_to_src_table : "buffered" -> "source"
91 //
92 // Translation between "buffered" and "requested" addresses is done with a simple shift:
93 // buffered_address + _buffer_to_requested_delta == requested_address
94 //
95 class ArchiveBuilder : public StackObj {
96 protected:
97 DumpRegion* _current_dump_region;
98 address _buffer_bottom; // for writing the contents of rw/ro regions
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
149 // This constructor is only used for regenerated objects (created by LambdaFormInvokers, etc).
150 // src = address of a Method or InstanceKlass that has been regenerated.
151 // renegerated_obj_info = info for the regenerated version of src.
152 SourceObjInfo(address src, SourceObjInfo* renegerated_obj_info) :
153 _ptrmap_start(0), _ptrmap_end(0), _read_only(false),
154 _follow_mode(renegerated_obj_info->_follow_mode),
155 _size_in_bytes(0), _msotype(renegerated_obj_info->_msotype),
156 _source_addr(src), _buffered_addr(renegerated_obj_info->_buffered_addr) {}
157
158 bool should_copy() const { return _follow_mode == make_a_copy; }
159 void set_buffered_addr(address addr) {
160 assert(should_copy(), "must be");
161 assert(_buffered_addr == nullptr, "cannot be copied twice");
162 assert(addr != nullptr, "must be a valid copy");
163 _buffered_addr = addr;
164 }
165 void set_ptrmap_start(uintx v) { _ptrmap_start = v; }
166 void set_ptrmap_end(uintx v) { _ptrmap_end = v; }
167 uintx ptrmap_start() const { return _ptrmap_start; } // inclusive
168 uintx ptrmap_end() const { return _ptrmap_end; } // exclusive
169 bool read_only() const { return _read_only; }
170 bool has_embedded_pointer() const { return _has_embedded_pointer; }
171 void set_has_embedded_pointer() { _has_embedded_pointer = true; }
172 int size_in_bytes() const { return _size_in_bytes; }
173 int id() const { return _id; }
174 void set_id(int i) { _id = i; }
175 address source_addr() const { return _source_addr; }
176 address buffered_addr() const {
177 if (_follow_mode != set_to_null) {
178 assert(_buffered_addr != nullptr, "must be initialized");
179 }
180 return _buffered_addr;
181 }
182 MetaspaceObj::Type msotype() const { return _msotype; }
183 };
184
185 class SourceObjList {
186 uintx _total_bytes;
187 GrowableArray<SourceObjInfo*>* _objs; // Source objects to be archived
188 CHeapBitMap _ptrmap; // Marks the addresses of the pointer fields
189 // in the source objects
190 public:
191 SourceObjList();
192 ~SourceObjList();
193
194 GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
195
196 void append(SourceObjInfo* src_info);
197 void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
198 void relocate(int i, ArchiveBuilder* builder);
199
200 // convenience accessor
201 SourceObjInfo* at(int i) const { return objs()->at(i); }
202 };
203
204 class CDSMapLogger;
205
206 static const int INITIAL_TABLE_SIZE = 15889;
207 static const int MAX_TABLE_SIZE = 1000000;
208
209 ReservedSpace _shared_rs;
210 VirtualSpace _shared_vs;
211
212 // The "pz" region is used only during static dumps to reserve an unused space between SharedBaseAddress and
213 // the bottom of the rw region. During runtime, this space will be filled with a reserved area that disallows
214 // read/write/exec, so we can track for bad CompressedKlassPointers encoding.
215 // Note: this region does NOT exist in the cds archive.
216 DumpRegion _pz_region;
217
218 DumpRegion _rw_region;
219 DumpRegion _ro_region;
220 DumpRegion _ac_region; // AOT code
221
222 // Combined bitmap to track pointers in both RW and RO regions. This is updated
223 // as objects are copied into RW and RO.
224 CHeapBitMap _ptrmap;
225
226 // _ptrmap is split into these two bitmaps which are written into the archive.
227 CHeapBitMap _rw_ptrmap; // marks pointers in the RW region
228 CHeapBitMap _ro_ptrmap; // marks pointers in the RO region
229
230 SourceObjList _rw_src_objs; // objs to put in rw region
231 SourceObjList _ro_src_objs; // objs to put in ro region
232 ResizeableResourceHashtable<address, SourceObjInfo, AnyObj::C_HEAP, mtClassShared> _src_obj_table;
233 ResizeableResourceHashtable<address, address, AnyObj::C_HEAP, mtClassShared> _buffered_to_src_table;
234 GrowableArray<Klass*>* _klasses;
235 GrowableArray<Symbol*>* _symbols;
236 unsigned int _entropy_seed;
237
238 // statistics
239 DumpAllocStats _alloc_stats;
240 size_t _total_heap_region_size;
241 struct {
242 size_t _num_ptrs;
243 size_t _num_tagged_ptrs;
244 size_t _num_nulled_ptrs;
245 } _relocated_ptr_info;
246
247 void print_region_stats(FileMapInfo *map_info, ArchiveHeapInfo* heap_info);
248 void print_bitmap_region_stats(size_t size, size_t total_size);
249 void print_heap_region_stats(ArchiveHeapInfo* heap_info, size_t total_size);
250
251 // For global access.
252 static ArchiveBuilder* _current;
253
254 public:
255 // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
256 // These are usually for misc tables that are allocated in the RO space.
257 class OtherROAllocMark {
258 char* _oldtop;
259 public:
260 OtherROAllocMark() {
261 _oldtop = _current->_ro_region.top();
262 }
263 ~OtherROAllocMark();
264 };
265
266 void count_relocated_pointer(bool tagged, bool nulled);
267
268 private:
269 FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
270
271 void iterate_sorted_roots(MetaspaceClosure* it);
272 void sort_klasses();
273 static int compare_symbols_by_address(Symbol** a, Symbol** b);
274 static int compare_klass_by_name(Klass** a, Klass** b);
275
276 void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
277 void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
278
279 void relocate_embedded_pointers(SourceObjList* src_objs);
280
281 bool is_excluded(Klass* k);
282 void clean_up_src_obj_table();
283
284 protected:
285 virtual void iterate_roots(MetaspaceClosure* it) = 0;
286 void start_dump_region(DumpRegion* next);
287
288 public:
289 address reserve_buffer();
290
291 address buffer_bottom() const { return _buffer_bottom; }
292 address buffer_top() const { return (address)current_dump_region()->top(); }
293 address requested_static_archive_bottom() const { return _requested_static_archive_bottom; }
294 address mapped_static_archive_bottom() const { return _mapped_static_archive_bottom; }
295 intx buffer_to_requested_delta() const { return _buffer_to_requested_delta; }
296
297 bool is_in_buffer_space(address p) const {
298 return (buffer_bottom() != nullptr && buffer_bottom() <= p && p < buffer_top());
299 }
300
301 template <typename T> bool is_in_requested_static_archive(T p) const {
302 return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
303 }
304
305 template <typename T> bool is_in_mapped_static_archive(T p) const {
306 return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
307 }
308
309 template <typename T> bool is_in_buffer_space(T obj) const {
310 return is_in_buffer_space(address(obj));
311 }
312
313 template <typename T> T to_requested(T obj) const {
314 assert(is_in_buffer_space(obj), "must be");
315 return (T)(address(obj) + _buffer_to_requested_delta);
316 }
317
318 static intx get_buffer_to_requested_delta() {
319 return current()->buffer_to_requested_delta();
320 }
321
322 inline static u4 to_offset_u4(uintx offset) {
323 guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
324 return (u4)offset;
325 }
326
327 public:
328 static const uintx MAX_SHARED_DELTA = ArchiveUtils::MAX_SHARED_DELTA;;
329
330 // The address p points to an object inside the output buffer. When the archive is mapped
331 // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
332 uintx buffer_to_offset(address p) const;
333
334 // Same as buffer_to_offset, except that the address p points to either (a) an object
335 // inside the output buffer, or (b), an object in the currently mapped static archive.
336 uintx any_to_offset(address p) const;
337
338 // The reverse of buffer_to_offset()
339 address offset_to_buffered_address(u4 offset) const;
340
341 template <typename T>
342 u4 buffer_to_offset_u4(T p) const {
343 uintx offset = buffer_to_offset((address)p);
344 return to_offset_u4(offset);
345 }
346
347 template <typename T>
348 u4 any_to_offset_u4(T p) const {
349 assert(p != nullptr, "must not be null");
350 uintx offset = any_to_offset((address)p);
351 return to_offset_u4(offset);
352 }
353
354 template <typename T>
355 u4 any_or_null_to_offset_u4(T p) const {
356 if (p == nullptr) {
357 return 0;
358 } else {
359 return any_to_offset_u4<T>(p);
360 }
361 }
362
363 template <typename T>
364 T offset_to_buffered(u4 offset) const {
365 return (T)offset_to_buffered_address(offset);
366 }
367
368 public:
369 ArchiveBuilder();
370 ~ArchiveBuilder();
371
372 int entropy();
373 void gather_klasses_and_symbols();
374 void gather_source_objs();
375 bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
376 bool gather_one_source_obj(MetaspaceClosure::Ref* ref, bool read_only);
377 void remember_embedded_pointer_in_enclosing_obj(MetaspaceClosure::Ref* ref);
378 static void serialize_dynamic_archivable_items(SerializeClosure* soc);
379
380 DumpRegion* pz_region() { return &_pz_region; }
381 DumpRegion* rw_region() { return &_rw_region; }
382 DumpRegion* ro_region() { return &_ro_region; }
383 DumpRegion* ac_region() { return &_ac_region; }
384
385 static char* rw_region_alloc(size_t num_bytes) {
386 return current()->rw_region()->allocate(num_bytes);
387 }
388 static char* ro_region_alloc(size_t num_bytes) {
389 return current()->ro_region()->allocate(num_bytes);
390 }
391 static char* ac_region_alloc(size_t num_bytes) {
392 return current()->ac_region()->allocate(num_bytes);
393 }
394
395 void start_ac_region();
396 void end_ac_region();
397
398 template <typename T>
399 static Array<T>* new_ro_array(int length) {
400 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
401 Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
402 array->initialize(length);
403 return array;
404 }
405
406 template <typename T>
407 static Array<T>* new_rw_array(int length) {
408 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
409 Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
410 array->initialize(length);
411 return array;
412 }
413
414 template <typename T>
415 static size_t ro_array_bytesize(int length) {
416 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
417 return align_up(byte_size, SharedSpaceObjectAlignment);
418 }
419
420 char* ro_strdup(const char* s);
421
422 static int compare_src_objs(SourceObjInfo** a, SourceObjInfo** b);
423 void sort_metadata_objs();
424 void dump_rw_metadata();
425 void dump_ro_metadata();
426 void relocate_metaspaceobj_embedded_pointers();
427 void record_regenerated_object(address orig_src_obj, address regen_src_obj);
428 void make_klasses_shareable();
429 void make_training_data_shareable();
430 void relocate_to_requested();
431 void write_archive(FileMapInfo* mapinfo, ArchiveHeapInfo* heap_info);
432 void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
433 bool read_only, bool allow_exec);
434
435 void write_pointer_in_buffer(address* ptr_location, address src_addr);
436 template <typename T> void write_pointer_in_buffer(T* ptr_location, T src_addr) {
437 write_pointer_in_buffer((address*)ptr_location, (address)src_addr);
438 }
439
440 void mark_and_relocate_to_buffered_addr(address* ptr_location);
441 template <typename T> void mark_and_relocate_to_buffered_addr(T ptr_location) {
442 mark_and_relocate_to_buffered_addr((address*)ptr_location);
443 }
444
445 bool has_been_archived(address src_addr) const;
446
447 bool has_been_buffered(address src_addr) const;
448 template <typename T> bool has_been_buffered(T src_addr) const {
449 return has_been_buffered((address)src_addr);
450 }
451
452 address get_buffered_addr(address src_addr) const;
453 template <typename T> T get_buffered_addr(T src_addr) const {
454 CDS_ONLY(return (T)get_buffered_addr((address)src_addr);)
455 NOT_CDS(return nullptr;)
456 }
457
458 address get_source_addr(address buffered_addr) const;
459 template <typename T> T get_source_addr(T buffered_addr) const {
460 return (T)get_source_addr((address)buffered_addr);
461 }
462
463 // All klasses and symbols that will be copied into the archive
464 GrowableArray<Klass*>* klasses() const { return _klasses; }
465 GrowableArray<Symbol*>* symbols() const { return _symbols; }
466
467 static bool is_active() {
468 CDS_ONLY(return (_current != nullptr));
469 NOT_CDS(return false;)
470 }
471
472 static ArchiveBuilder* current() {
473 assert(_current != nullptr, "ArchiveBuilder must be active");
474 return _current;
475 }
476
477 static DumpAllocStats* alloc_stats() {
478 return &(current()->_alloc_stats);
479 }
480
481 static CompactHashtableStats* symbol_stats() {
482 return alloc_stats()->symbol_stats();
483 }
484
485 static CompactHashtableStats* string_stats() {
486 return alloc_stats()->string_stats();
487 }
488
489 narrowKlass get_requested_narrow_klass(Klass* k);
490
491 static Klass* get_buffered_klass(Klass* src_klass) {
492 Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
493 assert(klass != nullptr && klass->is_klass(), "must be");
494 return klass;
495 }
496
497 static Symbol* get_buffered_symbol(Symbol* src_symbol) {
498 return (Symbol*)current()->get_buffered_addr((address)src_symbol);
499 }
500
501 void print_stats();
502 void report_out_of_space(const char* name, size_t needed_bytes);
503
504 #ifdef _LP64
505 // The CDS archive contains pre-computed narrow Klass IDs. It carries them in the headers of
506 // archived heap objects. With +UseCompactObjectHeaders, it also carries them in prototypes
507 // in Klass.
508 // When generating the archive, these narrow Klass IDs are computed using the following scheme:
509 // 1) The future encoding base is assumed to point to the first address of the generated mapping.
510 // That means that at runtime, the narrow Klass encoding must be set up with base pointing to
511 // the start address of the mapped CDS metadata archive (wherever that may be). This precludes
512 // zero-based encoding.
513 // 2) The shift must be large enough to result in an encoding range that covers the future assumed
514 // runtime Klass range. That future Klass range will contain both the CDS metadata archive and
515 // the future runtime class space. Since we do not know the size of the future class space, we
516 // need to chose an encoding base/shift combination that will result in a "large enough" size.
517 // The details depend on whether we use compact object headers or legacy object headers.
518 // In Legacy Mode, a narrow Klass ID is 32 bit. This gives us an encoding range size of 4G even
519 // with shift = 0, which is all we need. Therefore, we use a shift=0 for pre-calculating the
520 // narrow Klass IDs.
521 // TinyClassPointer Mode:
522 // We use the highest possible shift value to maximize the encoding range size.
523 static int precomputed_narrow_klass_shift();
524 #endif // _LP64
525
526 };
527
528 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP