1 /*
2  * Copyright (c) 2020, 2021, 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/hashtable.hpp"
37 #include "utilities/resourceHash.hpp"
38 
39 struct ArchiveHeapOopmapInfo;
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 class ArchiveBuilder : public StackObj {
65 protected:
66   DumpRegion* _current_dump_space;
67   address _buffer_bottom;                      // for writing the contents of rw/ro regions
68   address _last_verified_top;
69   int _num_dump_regions_used;
70   size_t _other_region_used_bytes;
71 
72   // These are the addresses where we will request the static and dynamic archives to be
73   // mapped at run time. If the request fails (due to ASLR), we will map the archives at
74   // os-selected addresses.
75   address _requested_static_archive_bottom;     // This is determined solely by the value of
76                                                 // SharedBaseAddress during -Xshare:dump.
77   address _requested_static_archive_top;
78   address _requested_dynamic_archive_bottom;    // Used only during dynamic dump. It's placed
79                                                 // immediately above _requested_static_archive_top.
80   address _requested_dynamic_archive_top;
81 
82   // (Used only during dynamic dump) where the static archive is actually mapped. This
83   // may be different than _requested_static_archive_{bottom,top} due to ASLR
84   address _mapped_static_archive_bottom;
85   address _mapped_static_archive_top;
86 
87   intx _buffer_to_requested_delta;
88 
current_dump_space() const89   DumpRegion* current_dump_space() const {  return _current_dump_space;  }
90 
91 public:
92   enum FollowMode {
93     make_a_copy, point_to_it, set_to_null
94   };
95 
96 private:
97   class SpecialRefInfo {
98     // We have a "special pointer" of the given _type at _field_offset of _src_obj.
99     // See MetaspaceClosure::push_special().
100     MetaspaceClosure::SpecialRef _type;
101     address _src_obj;
102     size_t _field_offset;
103 
104   public:
SpecialRefInfo()105     SpecialRefInfo() {}
SpecialRefInfo(MetaspaceClosure::SpecialRef type,address src_obj,size_t field_offset)106     SpecialRefInfo(MetaspaceClosure::SpecialRef type, address src_obj, size_t field_offset)
107       : _type(type), _src_obj(src_obj), _field_offset(field_offset) {}
108 
type() const109     MetaspaceClosure::SpecialRef type() const { return _type;         }
src_obj() const110     address src_obj()                   const { return _src_obj;      }
field_offset() const111     size_t field_offset()               const { return _field_offset; }
112   };
113 
114   class SourceObjInfo {
115     MetaspaceClosure::Ref* _ref;
116     uintx _ptrmap_start;     // The bit-offset of the start of this object (inclusive)
117     uintx _ptrmap_end;       // The bit-offset of the end   of this object (exclusive)
118     bool _read_only;
119     FollowMode _follow_mode;
120     int _size_in_bytes;
121     MetaspaceObj::Type _msotype;
122     address _dumped_addr;    // Address this->obj(), as used by the dumped archive.
123     address _orig_obj;       // The value of the original object (_ref->obj()) when this
124                              // SourceObjInfo was created. Note that _ref->obj() may change
125                              // later if _ref is relocated.
126 
127   public:
SourceObjInfo(MetaspaceClosure::Ref * ref,bool read_only,FollowMode follow_mode)128     SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
129       _ref(ref), _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _follow_mode(follow_mode),
130       _size_in_bytes(ref->size() * BytesPerWord), _msotype(ref->msotype()),
131       _orig_obj(ref->obj()) {
132       if (follow_mode == point_to_it) {
133         _dumped_addr = ref->obj();
134       } else {
135         _dumped_addr = NULL;
136       }
137     }
138 
should_copy() const139     bool should_copy() const { return _follow_mode == make_a_copy; }
ref() const140     MetaspaceClosure::Ref* ref() const { return  _ref; }
set_dumped_addr(address dumped_addr)141     void set_dumped_addr(address dumped_addr)  {
142       assert(should_copy(), "must be");
143       assert(_dumped_addr == NULL, "cannot be copied twice");
144       assert(dumped_addr != NULL, "must be a valid copy");
145       _dumped_addr = dumped_addr;
146     }
set_ptrmap_start(uintx v)147     void set_ptrmap_start(uintx v) { _ptrmap_start = v;    }
set_ptrmap_end(uintx v)148     void set_ptrmap_end(uintx v)   { _ptrmap_end = v;      }
ptrmap_start() const149     uintx ptrmap_start()  const    { return _ptrmap_start; } // inclusive
ptrmap_end() const150     uintx ptrmap_end()    const    { return _ptrmap_end;   } // exclusive
read_only() const151     bool read_only()      const    { return _read_only;    }
size_in_bytes() const152     int size_in_bytes()   const    { return _size_in_bytes; }
orig_obj() const153     address orig_obj()    const    { return _orig_obj; }
dumped_addr() const154     address dumped_addr() const    { return _dumped_addr; }
msotype() const155     MetaspaceObj::Type msotype() const { return _msotype; }
156 
157     // convenience accessor
obj() const158     address obj() const { return ref()->obj(); }
159   };
160 
161   class SourceObjList {
162     uintx _total_bytes;
163     GrowableArray<SourceObjInfo*>* _objs;     // Source objects to be archived
164     CHeapBitMap _ptrmap;                      // Marks the addresses of the pointer fields
165                                               // in the source objects
166   public:
167     SourceObjList();
168     ~SourceObjList();
169 
objs() const170     GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
171 
172     void append(MetaspaceClosure::Ref* enclosing_ref, SourceObjInfo* src_info);
173     void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
174     void relocate(int i, ArchiveBuilder* builder);
175 
176     // convenience accessor
at(int i) const177     SourceObjInfo* at(int i) const { return objs()->at(i); }
178   };
179 
180   class SrcObjTableCleaner {
181   public:
do_entry(address key,const SourceObjInfo * value)182     bool do_entry(address key, const SourceObjInfo* value) {
183       delete value->ref();
184       return true;
185     }
186   };
187 
188   class CDSMapLogger;
189 
190   static const int INITIAL_TABLE_SIZE = 15889;
191   static const int MAX_TABLE_SIZE     = 1000000;
192 
193   ReservedSpace _shared_rs;
194   VirtualSpace _shared_vs;
195 
196   DumpRegion _rw_region;
197   DumpRegion _ro_region;
198   CHeapBitMap _ptrmap;    // bitmap used by ArchivePtrMarker
199 
200   SourceObjList _rw_src_objs;                 // objs to put in rw region
201   SourceObjList _ro_src_objs;                 // objs to put in ro region
202   KVHashtable<address, SourceObjInfo, mtClassShared> _src_obj_table;
203   GrowableArray<Klass*>* _klasses;
204   GrowableArray<Symbol*>* _symbols;
205   GrowableArray<SpecialRefInfo>* _special_refs;
206 
207   // statistics
208   int _num_instance_klasses;
209   int _num_obj_array_klasses;
210   int _num_type_array_klasses;
211   DumpAllocStats _alloc_stats;
212   size_t _total_closed_heap_region_size;
213   size_t _total_open_heap_region_size;
214 
215   void print_region_stats(FileMapInfo *map_info,
216                           GrowableArray<MemRegion>* closed_heap_regions,
217                           GrowableArray<MemRegion>* open_heap_regions);
218   void print_bitmap_region_stats(size_t size, size_t total_size);
219   void print_heap_region_stats(GrowableArray<MemRegion> *heap_mem,
220                                const char *name, size_t total_size);
221 
222   // For global access.
223   static ArchiveBuilder* _current;
224 
225 public:
226   // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
227   // These are usually for misc tables that are allocated in the RO space.
228   class OtherROAllocMark {
229     char* _oldtop;
230   public:
OtherROAllocMark()231     OtherROAllocMark() {
232       _oldtop = _current->_ro_region.top();
233     }
234     ~OtherROAllocMark();
235   };
236 
237 private:
238   bool is_dumping_full_module_graph();
239   FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
240 
241   void iterate_sorted_roots(MetaspaceClosure* it, bool is_relocating_pointers);
242   void sort_symbols_and_fix_hash();
243   void sort_klasses();
244   static int compare_symbols_by_address(Symbol** a, Symbol** b);
245   static int compare_klass_by_name(Klass** a, Klass** b);
246 
247   void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
248   void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
249 
250   void update_special_refs();
251   void relocate_embedded_pointers(SourceObjList* src_objs);
252 
253   bool is_excluded(Klass* k);
254   void clean_up_src_obj_table();
255 
256 protected:
257   virtual void iterate_roots(MetaspaceClosure* it, bool is_relocating_pointers) = 0;
258 
259   // Conservative estimate for number of bytes needed for:
260   size_t _estimated_metaspaceobj_bytes;   // all archived MetaspaceObj's.
261   size_t _estimated_hashtable_bytes;     // symbol table and dictionaries
262 
263   static const int _total_dump_regions = 2;
264 
265   size_t estimate_archive_size();
266 
267   void start_dump_space(DumpRegion* next);
268   void verify_estimate_size(size_t estimate, const char* which);
269 
270 public:
271   address reserve_buffer();
272 
buffer_bottom() const273   address buffer_bottom()                    const { return _buffer_bottom;                       }
buffer_top() const274   address buffer_top()                       const { return (address)current_dump_space()->top(); }
requested_static_archive_bottom() const275   address requested_static_archive_bottom()  const { return  _requested_static_archive_bottom;    }
mapped_static_archive_bottom() const276   address mapped_static_archive_bottom()     const { return  _mapped_static_archive_bottom;       }
buffer_to_requested_delta() const277   intx buffer_to_requested_delta()           const { return _buffer_to_requested_delta;           }
278 
is_in_buffer_space(address p) const279   bool is_in_buffer_space(address p) const {
280     return (buffer_bottom() <= p && p < buffer_top());
281   }
282 
is_in_requested_static_archive(T p) const283   template <typename T> bool is_in_requested_static_archive(T p) const {
284     return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
285   }
286 
is_in_mapped_static_archive(T p) const287   template <typename T> bool is_in_mapped_static_archive(T p) const {
288     return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
289   }
290 
is_in_buffer_space(T obj) const291   template <typename T> bool is_in_buffer_space(T obj) const {
292     return is_in_buffer_space(address(obj));
293   }
294 
to_requested(T obj) const295   template <typename T> T to_requested(T obj) const {
296     assert(is_in_buffer_space(obj), "must be");
297     return (T)(address(obj) + _buffer_to_requested_delta);
298   }
299 
get_buffer_to_requested_delta()300   static intx get_buffer_to_requested_delta() {
301     return current()->buffer_to_requested_delta();
302   }
303 
304 public:
305   static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
306 
307   // The address p points to an object inside the output buffer. When the archive is mapped
308   // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
309   uintx buffer_to_offset(address p) const;
310 
311   // Same as buffer_to_offset, except that the address p points to either (a) an object
312   // inside the output buffer, or (b), an object in the currently mapped static archive.
313   uintx any_to_offset(address p) const;
314 
315   template <typename T>
buffer_to_offset_u4(T p) const316   u4 buffer_to_offset_u4(T p) const {
317     uintx offset = buffer_to_offset((address)p);
318     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset");
319     return (u4)offset;
320   }
321 
322   template <typename T>
any_to_offset_u4(T p) const323   u4 any_to_offset_u4(T p) const {
324     uintx offset = any_to_offset((address)p);
325     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset");
326     return (u4)offset;
327   }
328 
329   static void assert_is_vm_thread() PRODUCT_RETURN;
330 
331 public:
332   ArchiveBuilder();
333   ~ArchiveBuilder();
334 
335   void gather_klasses_and_symbols();
336   void gather_source_objs();
337   bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
338   bool gather_one_source_obj(MetaspaceClosure::Ref* enclosing_ref, MetaspaceClosure::Ref* ref, bool read_only);
339   void add_special_ref(MetaspaceClosure::SpecialRef type, address src_obj, size_t field_offset);
340   void remember_embedded_pointer_in_copied_obj(MetaspaceClosure::Ref* enclosing_ref, MetaspaceClosure::Ref* ref);
341 
rw_region()342   DumpRegion* rw_region() { return &_rw_region; }
ro_region()343   DumpRegion* ro_region() { return &_ro_region; }
344 
rw_region_alloc(size_t num_bytes)345   static char* rw_region_alloc(size_t num_bytes) {
346     return current()->rw_region()->allocate(num_bytes);
347   }
ro_region_alloc(size_t num_bytes)348   static char* ro_region_alloc(size_t num_bytes) {
349     return current()->ro_region()->allocate(num_bytes);
350   }
351 
352   template <typename T>
new_ro_array(int length)353   static Array<T>* new_ro_array(int length) {
354     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
355     Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
356     array->initialize(length);
357     return array;
358   }
359 
360   template <typename T>
new_rw_array(int length)361   static Array<T>* new_rw_array(int length) {
362     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
363     Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
364     array->initialize(length);
365     return array;
366   }
367 
368   template <typename T>
ro_array_bytesize(int length)369   static size_t ro_array_bytesize(int length) {
370     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
371     return align_up(byte_size, SharedSpaceObjectAlignment);
372   }
373 
374   void dump_rw_metadata();
375   void dump_ro_metadata();
376   void relocate_metaspaceobj_embedded_pointers();
377   void relocate_roots();
378   void relocate_vm_classes();
379   void make_klasses_shareable();
380   void relocate_to_requested();
381   void write_archive(FileMapInfo* mapinfo,
382                      GrowableArray<MemRegion>* closed_heap_regions,
383                      GrowableArray<MemRegion>* open_heap_regions,
384                      GrowableArray<ArchiveHeapOopmapInfo>* closed_heap_oopmaps,
385                      GrowableArray<ArchiveHeapOopmapInfo>* open_heap_oopmaps);
386   void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
387                     bool read_only,  bool allow_exec);
388 
389   address get_dumped_addr(address src_obj) const;
390 
391   // All klasses and symbols that will be copied into the archive
klasses() const392   GrowableArray<Klass*>*  klasses() const { return _klasses; }
symbols() const393   GrowableArray<Symbol*>* symbols() const { return _symbols; }
394 
is_active()395   static bool is_active() {
396     return (_current != NULL);
397   }
398 
current()399   static ArchiveBuilder* current() {
400     assert_is_vm_thread();
401     assert(_current != NULL, "ArchiveBuilder must be active");
402     return _current;
403   }
404 
alloc_stats()405   static DumpAllocStats* alloc_stats() {
406     return &(current()->_alloc_stats);
407   }
408 
symbol_stats()409   static CompactHashtableStats* symbol_stats() {
410     return alloc_stats()->symbol_stats();
411   }
412 
string_stats()413   static CompactHashtableStats* string_stats() {
414     return alloc_stats()->string_stats();
415   }
416 
417   void relocate_klass_ptr(oop o);
418 
get_relocated_klass(Klass * orig_klass)419   static Klass* get_relocated_klass(Klass* orig_klass) {
420     Klass* klass = (Klass*)current()->get_dumped_addr((address)orig_klass);
421     assert(klass != NULL && klass->is_klass(), "must be");
422     return klass;
423   }
424 
get_relocated_symbol(Symbol * orig_symbol)425   static Symbol* get_relocated_symbol(Symbol* orig_symbol) {
426     return (Symbol*)current()->get_dumped_addr((address)orig_symbol);
427   }
428 
429   void print_stats();
430   void report_out_of_space(const char* name, size_t needed_bytes);
431 };
432 
433 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP
434