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/metaspace.hpp"
31 #include "memory/metaspaceClosure.hpp"
32 #include "oops/array.hpp"
33 #include "oops/klass.hpp"
34 #include "runtime/os.hpp"
35 #include "utilities/bitMap.hpp"
36 #include "utilities/growableArray.hpp"
37 #include "utilities/resizeableResourceHash.hpp"
38 #include "utilities/resourceHash.hpp"
39
40 class ArchiveHeapInfo;
41 class CHeapBitMap;
42 class FileMapInfo;
43 class Klass;
44 class MemRegion;
45 class Symbol;
46
47 // The minimum alignment for non-Klass objects inside the CDS archive. Klass objects need
48 // to follow CompressedKlassPointers::klass_alignment_in_bytes().
49 constexpr size_t SharedSpaceObjectAlignment = Metaspace::min_allocation_alignment_bytes;
50
51 // Overview of CDS archive creation (for both static and dynamic dump):
52 //
53 // [1] Load all classes (static dump: from the classlist, dynamic dump: as part of app execution)
54 // [2] Allocate "output buffer"
55 // [3] Copy contents of the 2 "core" regions (rw/ro) into the output buffer.
56 // - allocate the cpp vtables in rw (static dump only)
57 // - memcpy the MetaspaceObjs into rw/ro:
58 // dump_rw_region();
59 // dump_ro_region();
60 // - fix all the pointers in the MetaspaceObjs to point to the copies
61 // relocate_metaspaceobj_embedded_pointers()
62 // [4] Copy symbol table, dictionary, etc, into the ro region
63 // [5] Relocate all the pointers in rw/ro, so that the archive can be mapped to
64 // the "requested" location without runtime relocation. See relocate_to_requested()
65 //
66 // "source" vs "buffered" vs "requested"
67 //
68 // The ArchiveBuilder deals with three types of addresses.
69 //
70 // "source": These are the addresses of objects created in step [1] above. They are the actual
71 // InstanceKlass*, Method*, etc, of the Java classes that are loaded for executing
72 // Java bytecodes in the JVM process that's dumping the CDS archive.
73 //
74 // It may be necessary to contiue Java execution after ArchiveBuilder is finished.
75 // Therefore, we don't modify any of the "source" objects.
76 //
77 // "buffered": The "source" objects that are deemed archivable are copied into a temporary buffer.
78 // Objects in the buffer are modified in steps [2, 3, 4] (e.g., unshareable info is
79 // removed, pointers are relocated, etc) to prepare them to be loaded at runtime.
80 //
81 // "requested": These are the addreses where the "buffered" objects should be loaded at runtime.
82 // When the "buffered" objects are written into the archive file, their addresses
83 // are adjusted in step [5] such that the lowest of these objects would be mapped
84 // at SharedBaseAddress.
85 //
86 // Translation between "source" and "buffered" addresses is done with two hashtables:
87 // _src_obj_table : "source" -> "buffered"
88 // _buffered_to_src_table : "buffered" -> "source"
89 //
90 // Translation between "buffered" and "requested" addresses is done with a simple shift:
91 // buffered_address + _buffer_to_requested_delta == requested_address
92 //
93 class ArchiveBuilder : public StackObj {
94 protected:
95 DumpRegion* _current_dump_region;
96 address _buffer_bottom; // for writing the contents of rw/ro regions
97 address _last_verified_top;
98 int _num_dump_regions_used;
99 size_t _other_region_used_bytes;
100
101 // These are the addresses where we will request the static and dynamic archives to be
102 // mapped at run time. If the request fails (due to ASLR), we will map the archives at
103 // os-selected addresses.
104 address _requested_static_archive_bottom; // This is determined solely by the value of
105 // SharedBaseAddress during -Xshare:dump.
106 address _requested_static_archive_top;
107 address _requested_dynamic_archive_bottom; // Used only during dynamic dump. It's placed
108 // immediately above _requested_static_archive_top.
109 address _requested_dynamic_archive_top;
110
111 // (Used only during dynamic dump) where the static archive is actually mapped. This
112 // may be different than _requested_static_archive_{bottom,top} due to ASLR
113 address _mapped_static_archive_bottom;
114 address _mapped_static_archive_top;
115
116 intx _buffer_to_requested_delta;
117
118 DumpRegion* current_dump_region() const { return _current_dump_region; }
119
120 public:
121 enum FollowMode {
122 make_a_copy, point_to_it, set_to_null
123 };
124
125 private:
126 class SourceObjInfo {
127 uintx _ptrmap_start; // The bit-offset of the start of this object (inclusive)
128 uintx _ptrmap_end; // The bit-offset of the end of this object (exclusive)
129 bool _read_only;
130 bool _has_embedded_pointer;
131 FollowMode _follow_mode;
132 int _size_in_bytes;
133 int _id; // Each object has a unique serial ID, starting from zero. The ID is assigned
134 // when the object is added into _source_objs.
135 MetaspaceObj::Type _msotype;
136 address _source_addr; // The source object to be copied.
137 address _buffered_addr; // The copy of this object insider the buffer.
138 public:
139 SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
140 _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _has_embedded_pointer(false), _follow_mode(follow_mode),
141 _size_in_bytes(ref->size() * BytesPerWord), _id(0), _msotype(ref->msotype()),
142 _source_addr(ref->obj()) {
143 if (follow_mode == point_to_it) {
144 _buffered_addr = ref->obj();
145 } else {
146 _buffered_addr = nullptr;
147 }
148 }
149
150 // This constructor is only used for regenerated objects (created by LambdaFormInvokers, etc).
151 // src = address of a Method or InstanceKlass that has been regenerated.
152 // renegerated_obj_info = info for the regenerated version of src.
153 SourceObjInfo(address src, SourceObjInfo* renegerated_obj_info) :
154 _ptrmap_start(0), _ptrmap_end(0), _read_only(false),
155 _follow_mode(renegerated_obj_info->_follow_mode),
156 _size_in_bytes(0), _msotype(renegerated_obj_info->_msotype),
157 _source_addr(src), _buffered_addr(renegerated_obj_info->_buffered_addr) {}
158
159 bool should_copy() const { return _follow_mode == make_a_copy; }
160 void set_buffered_addr(address addr) {
161 assert(should_copy(), "must be");
162 assert(_buffered_addr == nullptr, "cannot be copied twice");
163 assert(addr != nullptr, "must be a valid copy");
164 _buffered_addr = addr;
165 }
166 void set_ptrmap_start(uintx v) { _ptrmap_start = v; }
167 void set_ptrmap_end(uintx v) { _ptrmap_end = v; }
168 uintx ptrmap_start() const { return _ptrmap_start; } // inclusive
169 uintx ptrmap_end() const { return _ptrmap_end; } // exclusive
170 bool read_only() const { return _read_only; }
171 bool has_embedded_pointer() const { return _has_embedded_pointer; }
172 void set_has_embedded_pointer() { _has_embedded_pointer = true; }
173 int size_in_bytes() const { return _size_in_bytes; }
174 int id() const { return _id; }
175 void set_id(int i) { _id = i; }
176 address source_addr() const { return _source_addr; }
177 address buffered_addr() const {
178 if (_follow_mode != set_to_null) {
179 assert(_buffered_addr != nullptr, "must be initialized");
180 }
181 return _buffered_addr;
182 }
183 MetaspaceObj::Type msotype() const { return _msotype; }
184 };
185
186 class SourceObjList {
187 uintx _total_bytes;
188 GrowableArray<SourceObjInfo*>* _objs; // Source objects to be archived
189 CHeapBitMap _ptrmap; // Marks the addresses of the pointer fields
190 // in the source objects
191 public:
192 SourceObjList();
193 ~SourceObjList();
194
195 GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
196
197 void append(SourceObjInfo* src_info);
198 void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
199 void relocate(int i, ArchiveBuilder* builder);
200
201 // convenience accessor
202 SourceObjInfo* at(int i) const { return objs()->at(i); }
203 };
204
205 class CDSMapLogger;
206
207 static const int INITIAL_TABLE_SIZE = 15889;
208 static const int MAX_TABLE_SIZE = 1000000;
209
210 ReservedSpace _shared_rs;
211 VirtualSpace _shared_vs;
212
213 DumpRegion _rw_region;
214 DumpRegion _ro_region;
215
216 // Combined bitmap to track pointers in both RW and RO regions. This is updated
217 // as objects are copied into RW and RO.
218 CHeapBitMap _ptrmap;
219
220 // _ptrmap is split into these two bitmaps which are written into the archive.
221 CHeapBitMap _rw_ptrmap; // marks pointers in the RW region
222 CHeapBitMap _ro_ptrmap; // marks pointers in the RO region
223
224 SourceObjList _rw_src_objs; // objs to put in rw region
225 SourceObjList _ro_src_objs; // objs to put in ro region
226 ResizeableResourceHashtable<address, SourceObjInfo, AnyObj::C_HEAP, mtClassShared> _src_obj_table;
227 ResizeableResourceHashtable<address, address, AnyObj::C_HEAP, mtClassShared> _buffered_to_src_table;
228 GrowableArray<Klass*>* _klasses;
229 GrowableArray<Symbol*>* _symbols;
230 unsigned int _entropy_seed;
231
232 // statistics
233 DumpAllocStats _alloc_stats;
234 size_t _total_heap_region_size;
235
236 void print_region_stats(FileMapInfo *map_info, ArchiveHeapInfo* heap_info);
237 void print_bitmap_region_stats(size_t size, size_t total_size);
238 void print_heap_region_stats(ArchiveHeapInfo* heap_info, size_t total_size);
239
240 // For global access.
241 static ArchiveBuilder* _current;
242
243 public:
244 // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
245 // These are usually for misc tables that are allocated in the RO space.
246 class OtherROAllocMark {
247 char* _oldtop;
248 public:
249 OtherROAllocMark() {
250 _oldtop = _current->_ro_region.top();
251 }
252 ~OtherROAllocMark();
253 };
254
255 private:
256 FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
257
258 void iterate_sorted_roots(MetaspaceClosure* it);
259 void sort_klasses();
260 static int compare_symbols_by_address(Symbol** a, Symbol** b);
261 static int compare_klass_by_name(Klass** a, Klass** b);
262
263 void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
264 void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
265
266 void relocate_embedded_pointers(SourceObjList* src_objs);
267
268 bool is_excluded(Klass* k);
269 void clean_up_src_obj_table();
270
271 protected:
272 virtual void iterate_roots(MetaspaceClosure* it) = 0;
273
274 // Conservative estimate for number of bytes needed for:
275 size_t _estimated_metaspaceobj_bytes; // all archived MetaspaceObj's.
276 size_t _estimated_hashtable_bytes; // symbol table and dictionaries
277
278 static const int _total_dump_regions = 2;
279
280 size_t estimate_archive_size();
281
282 void start_dump_region(DumpRegion* next);
283 void verify_estimate_size(size_t estimate, const char* which);
284
285 public:
286 address reserve_buffer();
287
288 address buffer_bottom() const { return _buffer_bottom; }
289 address buffer_top() const { return (address)current_dump_region()->top(); }
290 address requested_static_archive_bottom() const { return _requested_static_archive_bottom; }
291 address mapped_static_archive_bottom() const { return _mapped_static_archive_bottom; }
292 intx buffer_to_requested_delta() const { return _buffer_to_requested_delta; }
293
294 bool is_in_buffer_space(address p) const {
295 return (buffer_bottom() <= p && p < buffer_top());
296 }
297
298 template <typename T> bool is_in_requested_static_archive(T p) const {
299 return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
300 }
301
302 template <typename T> bool is_in_mapped_static_archive(T p) const {
303 return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
304 }
305
306 template <typename T> bool is_in_buffer_space(T obj) const {
307 return is_in_buffer_space(address(obj));
308 }
309
310 template <typename T> T to_requested(T obj) const {
311 assert(is_in_buffer_space(obj), "must be");
312 return (T)(address(obj) + _buffer_to_requested_delta);
313 }
314
315 static intx get_buffer_to_requested_delta() {
316 return current()->buffer_to_requested_delta();
317 }
318
319 inline static u4 to_offset_u4(uintx offset) {
320 guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
321 return (u4)offset;
322 }
323
324 public:
325 static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
326
327 // The address p points to an object inside the output buffer. When the archive is mapped
328 // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
329 uintx buffer_to_offset(address p) const;
330
331 // Same as buffer_to_offset, except that the address p points to either (a) an object
332 // inside the output buffer, or (b), an object in the currently mapped static archive.
333 uintx any_to_offset(address p) const;
334
335 template <typename T>
336 u4 buffer_to_offset_u4(T p) const {
337 uintx offset = buffer_to_offset((address)p);
338 return to_offset_u4(offset);
339 }
340
341 template <typename T>
342 u4 any_to_offset_u4(T p) const {
343 uintx offset = any_to_offset((address)p);
344 return to_offset_u4(offset);
345 }
346
347 static void assert_is_vm_thread() PRODUCT_RETURN;
348
349 public:
350 ArchiveBuilder();
351 ~ArchiveBuilder();
352
353 int entropy();
354 void gather_klasses_and_symbols();
355 void gather_source_objs();
356 bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
357 bool gather_one_source_obj(MetaspaceClosure::Ref* ref, bool read_only);
358 void remember_embedded_pointer_in_enclosing_obj(MetaspaceClosure::Ref* ref);
359 static void serialize_dynamic_archivable_items(SerializeClosure* soc);
360
361 DumpRegion* rw_region() { return &_rw_region; }
362 DumpRegion* ro_region() { return &_ro_region; }
363
364 static char* rw_region_alloc(size_t num_bytes) {
365 return current()->rw_region()->allocate(num_bytes);
366 }
367 static char* ro_region_alloc(size_t num_bytes) {
368 return current()->ro_region()->allocate(num_bytes);
369 }
370
371 template <typename T>
372 static Array<T>* new_ro_array(int length) {
373 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
374 Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
375 array->initialize(length);
376 return array;
377 }
378
379 template <typename T>
380 static Array<T>* new_rw_array(int length) {
381 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
382 Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
383 array->initialize(length);
384 return array;
385 }
386
387 template <typename T>
388 static size_t ro_array_bytesize(int length) {
389 size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
390 return align_up(byte_size, SharedSpaceObjectAlignment);
391 }
392
393 char* ro_strdup(const char* s);
394
395 static int compare_src_objs(SourceObjInfo** a, SourceObjInfo** b);
396 void sort_metadata_objs();
397 void dump_rw_metadata();
398 void dump_ro_metadata();
399 void relocate_metaspaceobj_embedded_pointers();
400 void record_regenerated_object(address orig_src_obj, address regen_src_obj);
401 void make_klasses_shareable();
402 void relocate_to_requested();
403 void write_archive(FileMapInfo* mapinfo, ArchiveHeapInfo* heap_info);
404 void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
405 bool read_only, bool allow_exec);
406
407 void write_pointer_in_buffer(address* ptr_location, address src_addr);
408 template <typename T> void write_pointer_in_buffer(T* ptr_location, T src_addr) {
409 write_pointer_in_buffer((address*)ptr_location, (address)src_addr);
410 }
411
412 void mark_and_relocate_to_buffered_addr(address* ptr_location);
413 template <typename T> void mark_and_relocate_to_buffered_addr(T ptr_location) {
414 mark_and_relocate_to_buffered_addr((address*)ptr_location);
415 }
416
417 address get_buffered_addr(address src_addr) const;
418 template <typename T> T get_buffered_addr(T src_addr) const {
419 return (T)get_buffered_addr((address)src_addr);
420 }
421
422 address get_source_addr(address buffered_addr) const;
423 template <typename T> T get_source_addr(T buffered_addr) const {
424 return (T)get_source_addr((address)buffered_addr);
425 }
426
427 // All klasses and symbols that will be copied into the archive
428 GrowableArray<Klass*>* klasses() const { return _klasses; }
429 GrowableArray<Symbol*>* symbols() const { return _symbols; }
430
431 static bool is_active() {
432 return (_current != nullptr);
433 }
434
435 static ArchiveBuilder* current() {
436 assert_is_vm_thread();
437 assert(_current != nullptr, "ArchiveBuilder must be active");
438 return _current;
439 }
440
441 static DumpAllocStats* alloc_stats() {
442 return &(current()->_alloc_stats);
443 }
444
445 static CompactHashtableStats* symbol_stats() {
446 return alloc_stats()->symbol_stats();
447 }
448
449 static CompactHashtableStats* string_stats() {
450 return alloc_stats()->string_stats();
451 }
452
453 narrowKlass get_requested_narrow_klass(Klass* k);
454
455 static Klass* get_buffered_klass(Klass* src_klass) {
456 Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
457 assert(klass != nullptr && klass->is_klass(), "must be");
458 return klass;
459 }
460
461 static Symbol* get_buffered_symbol(Symbol* src_symbol) {
462 return (Symbol*)current()->get_buffered_addr((address)src_symbol);
463 }
464
465 void print_stats();
466 void report_out_of_space(const char* name, size_t needed_bytes);
467
468 #ifdef _LP64
469 // Archived heap object headers (and soon, with Lilliput, markword prototypes) carry pre-computed
470 // narrow Klass ids calculated with the following scheme:
471 // 1) the encoding base must be the mapping start address.
472 // 2) shift must be large enough to result in an encoding range that covers the runtime Klass range.
473 // That Klass range is defined by CDS archive size and runtime class space size. Luckily, the maximum
474 // size can be predicted: archive size is assumed to be <1G, class space size capped at 3G, and at
475 // runtime we put both regions adjacent to each other. Therefore, runtime Klass range size < 4G.
476 // The value of this precomputed shift depends on the class pointer mode at dump time.
477 // Legacy Mode:
478 // Since nKlass itself is 32 bit, our encoding range len is 4G, and since we set the base directly
479 // at mapping start, these 4G are enough. Therefore, we don't need to shift at all (shift=0).
480 // TinyClassPointer Mode:
481 // To cover the 4G, we need the highest possible shift value. That may change in the future, if
482 // we decide to correct the pre-calculated narrow Klass IDs at load time.
483 static int precomputed_narrow_klass_shift();
484 #endif // _LP64
485
486 };
487
488 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP