10 #include "snapshotimp.h"
15 #include <semaphore.h>
22 #define FAILURE(mesg) { printf("failed in the API: %s with errno relative message: %s\n", mesg, strerror( errno ) ); exit(EXIT_FAILURE); }
25 #define SSDEBUG printf
27 #define SSDEBUG(...) do { } while (0)
30 /* extern declaration definition */
31 struct SnapShot * snapshotrecord = NULL;
33 #if !USE_MPROTECT_SNAPSHOT
35 * These variables are necessary because the stack is shared region and
36 * there exists a race between all processes executing the same function.
37 * To avoid the problem above, we require variables allocated in 'safe' regions.
38 * The bug was actually observed with the forkID, these variables below are
39 * used to indicate the various contexts to which to switch to.
41 * @savedSnapshotContext: contains the point to which takesnapshot() call should switch to.
42 * @savedUserSnapshotContext: contains the point to which the process whose snapshotid is equal to the rollbackid should switch to
43 * @snapshotid: it is a running counter for the various forked processes snapshotid. it is incremented and set in a persistently shared record
45 static ucontext_t savedSnapshotContext;
46 static ucontext_t savedUserSnapshotContext;
47 static snapshot_id snapshotid = 0;
50 /** PageAlignedAdressUpdate return a page aligned address for the
51 * address being added as a side effect the numBytes are also changed.
53 static void * PageAlignAddressUpward(void * addr) {
54 return (void *)((((uintptr_t)addr)+PAGESIZE-1)&~(PAGESIZE-1));
57 #if USE_MPROTECT_SNAPSHOT
59 /** ReturnPageAlignedAddress returns a page aligned address for the
60 * address being added as a side effect the numBytes are also changed.
62 static void * ReturnPageAlignedAddress(void * addr) {
63 return (void *)(((uintptr_t)addr)&~(PAGESIZE-1));
66 /** The initSnapShotRecord method initialized the snapshotting data
67 * structures for the mprotect based snapshot.
69 static void initSnapShotRecord(unsigned int numbackingpages, unsigned int numsnapshots, unsigned int nummemoryregions) {
70 snapshotrecord=( struct SnapShot * )model_malloc(sizeof(struct SnapShot));
71 snapshotrecord->regionsToSnapShot=( struct MemoryRegion * )model_malloc(sizeof(struct MemoryRegion)*nummemoryregions);
72 snapshotrecord->backingStoreBasePtr= ( struct SnapShotPage * )model_malloc( sizeof( struct SnapShotPage ) * (numbackingpages + 1) );
73 //Page align the backingstorepages
74 snapshotrecord->backingStore=( struct SnapShotPage * )PageAlignAddressUpward(snapshotrecord->backingStoreBasePtr);
75 snapshotrecord->backingRecords=( struct BackingPageRecord * )model_malloc(sizeof(struct BackingPageRecord)*numbackingpages);
76 snapshotrecord->snapShots= ( struct SnapShotRecord * )model_malloc(sizeof(struct SnapShotRecord)*numsnapshots);
77 snapshotrecord->lastSnapShot=0;
78 snapshotrecord->lastBackingPage=0;
79 snapshotrecord->lastRegion=0;
80 snapshotrecord->maxRegions=nummemoryregions;
81 snapshotrecord->maxBackingPages=numbackingpages;
82 snapshotrecord->maxSnapShots=numsnapshots;
85 /** HandlePF is the page fault handler for mprotect based snapshotting
88 static void HandlePF( int sig, siginfo_t *si, void * unused){
89 if( si->si_code == SEGV_MAPERR ){
90 printf("Real Fault at %p\n", si->si_addr);
94 void* addr = ReturnPageAlignedAddress(si->si_addr);
96 unsigned int backingpage=snapshotrecord->lastBackingPage++; //Could run out of pages...
97 if (backingpage==snapshotrecord->maxBackingPages) {
98 printf("Out of backing pages at %p\n", si->si_addr);
103 memcpy(&(snapshotrecord->backingStore[backingpage]), addr, sizeof(struct SnapShotPage));
104 //remember where to copy page back to
105 snapshotrecord->backingRecords[backingpage].basePtrOfPage=addr;
106 //set protection to read/write
107 if (mprotect( addr, sizeof(struct SnapShotPage), PROT_READ | PROT_WRITE )) {
109 // Handle error by quitting?
112 #endif //nothing to handle for non snapshotting case.
114 #if !USE_MPROTECT_SNAPSHOT
115 void createSharedMemory(){
116 //step 1. create shared memory.
117 void * memMapBase = mmap( 0, SHARED_MEMORY_DEFAULT + STACK_SIZE_DEFAULT, PROT_READ | PROT_WRITE, MAP_SHARED|MAP_ANON, -1, 0 );
118 if( MAP_FAILED == memMapBase )
121 //Setup snapshot record at top of free region
122 snapshotrecord = ( struct SnapShot * )memMapBase;
123 snapshotrecord->mSharedMemoryBase = (void *)((uintptr_t)memMapBase + sizeof(struct SnapShot));
124 snapshotrecord->mStackBase = (void *)((uintptr_t)memMapBase + SHARED_MEMORY_DEFAULT);
125 snapshotrecord->mStackSize = STACK_SIZE_DEFAULT;
126 snapshotrecord->mIDToRollback = -1;
127 snapshotrecord->currSnapShotID = 0;
132 /** The initSnapShotLibrary function initializes the Snapshot library.
133 * @param entryPoint the function that should run the program.
135 #if USE_MPROTECT_SNAPSHOT
137 void initSnapShotLibrary(unsigned int numbackingpages,
138 unsigned int numsnapshots, unsigned int nummemoryregions,
139 unsigned int numheappages, VoidFuncPtr entryPoint) {
140 /* Setup a stack for our signal handler.... */
142 ss.ss_sp = model_malloc(SIGSTACKSIZE);
143 ss.ss_size = SIGSTACKSIZE;
145 sigaltstack(&ss, NULL);
148 sa.sa_flags = SA_SIGINFO | SA_NODEFER | SA_RESTART | SA_ONSTACK;
149 sigemptyset( &sa.sa_mask );
150 sa.sa_sigaction = HandlePF;
152 if( sigaction( SIGBUS, &sa, NULL ) == -1 ){
153 printf("SIGACTION CANNOT BE INSTALLED\n");
157 if( sigaction( SIGSEGV, &sa, NULL ) == -1 ){
158 printf("SIGACTION CANNOT BE INSTALLED\n");
162 initSnapShotRecord(numbackingpages, numsnapshots, nummemoryregions);
164 // EVIL HACK: We need to make sure that calls into the HandlePF method don't cause dynamic links
165 // The problem is that we end up protecting state in the dynamic linker...
166 // Solution is to call our signal handler before we start protecting stuff...
169 memset(&si, 0, sizeof(si));
171 HandlePF(SIGSEGV, &si, NULL);
172 snapshotrecord->lastBackingPage--; //remove the fake page we copied
174 basemySpace=model_malloc((numheappages+1)*PAGESIZE);
175 void * pagealignedbase=PageAlignAddressUpward(basemySpace);
176 mySpace = create_mspace_with_base(pagealignedbase, numheappages*PAGESIZE, 1 );
177 addMemoryRegionToSnapShot(pagealignedbase, numheappages);
181 void initSnapShotLibrary(unsigned int numbackingpages,
182 unsigned int numsnapshots, unsigned int nummemoryregions,
183 unsigned int numheappages, VoidFuncPtr entryPoint) {
184 basemySpace=system_malloc((numheappages+1)*PAGESIZE);
185 void * pagealignedbase=PageAlignAddressUpward(basemySpace);
186 mySpace = create_mspace_with_base(pagealignedbase, numheappages*PAGESIZE, 1 );
188 createSharedMemory();
190 //step 2 setup the stack context.
191 ucontext_t newContext;
192 getcontext( &newContext );
193 newContext.uc_stack.ss_sp = snapshotrecord->mStackBase;
194 newContext.uc_stack.ss_size = STACK_SIZE_DEFAULT;
195 makecontext( &newContext, entryPoint, 0 );
196 /* switch to a new entryPoint context, on a new stack */
197 swapcontext(&savedSnapshotContext, &newContext);
199 /* switch back here when takesnapshot is called */
201 snapshotid = snapshotrecord->currSnapShotID;
202 /* This bool indicates that the current process's snapshotid is same
203 as the id to which the rollback needs to occur */
205 bool rollback = false;
207 snapshotrecord->currSnapShotID=snapshotid+1;
211 /* If the rollback bool is set, switch to the context we need to
212 return to during a rollback. */
214 setcontext( &( snapshotrecord->mContextToRollback ) );
216 /*Child process which is forked as a result of takesnapshot
217 call should switch back to the takesnapshot context*/
218 setcontext( &savedUserSnapshotContext );
224 SSDEBUG("The process id of child is %d and the process id of this process is %d and snapshot id is %d\n",
225 forkedID, getpid(), snapshotid );
228 retVal=waitpid( forkedID, &status, 0 );
229 } while( -1 == retVal && errno == EINTR );
231 if( snapshotrecord->mIDToRollback != snapshotid ) {
240 /** The addMemoryRegionToSnapShot function assumes that addr is page aligned.
242 void addMemoryRegionToSnapShot( void * addr, unsigned int numPages) {
243 #if USE_MPROTECT_SNAPSHOT
244 unsigned int memoryregion=snapshotrecord->lastRegion++;
245 if (memoryregion==snapshotrecord->maxRegions) {
246 printf("Exceeded supported number of memory regions!\n");
250 snapshotrecord->regionsToSnapShot[ memoryregion ].basePtr=addr;
251 snapshotrecord->regionsToSnapShot[ memoryregion ].sizeInPages=numPages;
252 #endif //NOT REQUIRED IN THE CASE OF FORK BASED SNAPSHOTS.
255 /** The takeSnapshot function takes a snapshot.
256 * @return The snapshot identifier.
258 snapshot_id takeSnapshot( ){
259 #if USE_MPROTECT_SNAPSHOT
260 for(unsigned int region=0; region<snapshotrecord->lastRegion;region++) {
261 if( mprotect(snapshotrecord->regionsToSnapShot[region].basePtr, snapshotrecord->regionsToSnapShot[region].sizeInPages*sizeof(struct SnapShotPage), PROT_READ ) == -1 ){
263 printf("Failed to mprotect inside of takeSnapShot\n");
267 unsigned int snapshot=snapshotrecord->lastSnapShot++;
268 if (snapshot==snapshotrecord->maxSnapShots) {
269 printf("Out of snapshots\n");
272 snapshotrecord->snapShots[snapshot].firstBackingPage=snapshotrecord->lastBackingPage;
276 swapcontext( &savedUserSnapshotContext, &savedSnapshotContext );
277 SSDEBUG("TAKESNAPSHOT RETURN\n");
282 /** The rollBack function rollback to the given snapshot identifier.
283 * @param theID is the snapshot identifier to rollback to.
285 void rollBack( snapshot_id theID ){
286 #if USE_MPROTECT_SNAPSHOT
287 HashTable< void *, bool, uintptr_t, 4, model_malloc, model_calloc, model_free> duplicateMap;
288 for(unsigned int region=0; region<snapshotrecord->lastRegion;region++) {
289 if( mprotect(snapshotrecord->regionsToSnapShot[region].basePtr, snapshotrecord->regionsToSnapShot[region].sizeInPages*sizeof(struct SnapShotPage), PROT_READ | PROT_WRITE ) == -1 ){
291 printf("Failed to mprotect inside of takeSnapShot\n");
295 for(unsigned int page=snapshotrecord->snapShots[theID].firstBackingPage; page<snapshotrecord->lastBackingPage; page++) {
296 if( !duplicateMap.contains(snapshotrecord->backingRecords[page].basePtrOfPage )) {
297 duplicateMap.put(snapshotrecord->backingRecords[page].basePtrOfPage, true);
298 memcpy(snapshotrecord->backingRecords[page].basePtrOfPage, &snapshotrecord->backingStore[page], sizeof(struct SnapShotPage));
301 snapshotrecord->lastSnapShot=theID;
302 snapshotrecord->lastBackingPage=snapshotrecord->snapShots[theID].firstBackingPage;
303 takeSnapshot(); //Make sure current snapshot is still good...All later ones are cleared
305 snapshotrecord->mIDToRollback = theID;
306 volatile int sTemp = 0;
307 getcontext( &snapshotrecord->mContextToRollback );
309 * This is used to quit the process on rollback, so that the process
310 * which needs to rollback can quit allowing the process whose
311 * snapshotid matches the rollbackid to switch to this context and
316 SSDEBUG("Invoked rollback\n");
320 * This fix obviates the need for a finalize call. hence less dependences for model-checker....
323 snapshotrecord->mIDToRollback = -1;