X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=snapshot.cc;h=fe5110a03fe5a7b0e45e3b62471788eb396a9aed;hb=2ada1a461101f16a0be8ba75200c8ae4f9ff9d84;hp=114473d50d27e21b476f9ab48e57232f888e3eda;hpb=f0f7b26d861767e59a07f3542cf810deb7dd52b0;p=model-checker.git diff --git a/snapshot.cc b/snapshot.cc index 114473d..fe5110a 100644 --- a/snapshot.cc +++ b/snapshot.cc @@ -15,14 +15,17 @@ #define FAILURE(mesg) { model_print("failed in the API: %s with errno relative message: %s\n", mesg, strerror(errno)); exit(EXIT_FAILURE); } -#ifdef CONFIG_SSDEBUG -#define SSDEBUG model_print -#else -#define SSDEBUG(...) do { } while (0) -#endif +/** PageAlignedAdressUpdate return a page aligned address for the + * address being added as a side effect the numBytes are also changed. + */ +static void * PageAlignAddressUpward(void *addr) +{ + return (void *)((((uintptr_t)addr) + PAGESIZE - 1) & ~(PAGESIZE - 1)); +} #if USE_MPROTECT_SNAPSHOT -/* Each snapshotrecord lists the firstbackingpage that must be written to + +/* Each SnapShotRecord lists the firstbackingpage that must be written to * revert to that snapshot */ struct SnapShotRecord { unsigned int firstBackingPage; @@ -44,7 +47,7 @@ struct MemoryRegion { }; /* Primary struct for snapshotting system */ -struct SnapShot { +struct mprot_snapshotter { struct MemoryRegion *regionsToSnapShot; //This pointer references an array of memory regions to snapshot snapshot_page_t *backingStore; //This pointer references an array of snapshotpage's that form the backing store void *backingStoreBasePtr; //This pointer references an array of snapshotpage's that form the backing store @@ -60,50 +63,7 @@ struct SnapShot { unsigned int maxSnapShots; //Stores the total number of snapshots we allow }; -#else - -#include - -#define SHARED_MEMORY_DEFAULT (100 * ((size_t)1 << 20)) // 100mb for the shared memory -#define STACK_SIZE_DEFAULT (((size_t)1 << 20) * 20) // 20 mb out of the above 100 mb for my stack - -struct SnapShot { - void *mSharedMemoryBase; - void *mStackBase; - size_t mStackSize; - volatile snapshot_id mIDToRollback; - ucontext_t mContextToRollback; - snapshot_id currSnapShotID; -}; -#endif - -static struct SnapShot *snapshotrecord = NULL; - -/** PageAlignedAdressUpdate return a page aligned address for the - * address being added as a side effect the numBytes are also changed. - */ -static void * PageAlignAddressUpward(void *addr) -{ - return (void *)((((uintptr_t)addr) + PAGESIZE - 1) & ~(PAGESIZE - 1)); -} - -#if !USE_MPROTECT_SNAPSHOT -/** @statics -* These variables are necessary because the stack is shared region and -* there exists a race between all processes executing the same function. -* To avoid the problem above, we require variables allocated in 'safe' regions. -* The bug was actually observed with the forkID, these variables below are -* used to indicate the various contexts to which to switch to. -* -* @savedSnapshotContext: contains the point to which takesnapshot() call should switch to. -* @savedUserSnapshotContext: contains the point to which the process whose snapshotid is equal to the rollbackid should switch to -* @snapshotid: it is a running counter for the various forked processes snapshotid. it is incremented and set in a persistently shared record -*/ -static ucontext_t savedSnapshotContext; -static ucontext_t savedUserSnapshotContext; -static snapshot_id snapshotid = 0; - -#else /* USE_MPROTECT_SNAPSHOT */ +static struct mprot_snapshotter *mprot_snap = NULL; /** ReturnPageAlignedAddress returns a page aligned address for the * address being added as a side effect the numBytes are also changed. @@ -118,19 +78,19 @@ static void * ReturnPageAlignedAddress(void *addr) */ static void initSnapShotRecord(unsigned int numbackingpages, unsigned int numsnapshots, unsigned int nummemoryregions) { - snapshotrecord = (struct SnapShot *)model_malloc(sizeof(struct SnapShot)); - snapshotrecord->regionsToSnapShot = (struct MemoryRegion *)model_malloc(sizeof(struct MemoryRegion) * nummemoryregions); - snapshotrecord->backingStoreBasePtr = (void *)model_malloc(sizeof(snapshot_page_t) * (numbackingpages + 1)); + mprot_snap = (struct mprot_snapshotter *)model_malloc(sizeof(struct mprot_snapshotter)); + mprot_snap->regionsToSnapShot = (struct MemoryRegion *)model_malloc(sizeof(struct MemoryRegion) * nummemoryregions); + mprot_snap->backingStoreBasePtr = (void *)model_malloc(sizeof(snapshot_page_t) * (numbackingpages + 1)); //Page align the backingstorepages - snapshotrecord->backingStore = (snapshot_page_t *)PageAlignAddressUpward(snapshotrecord->backingStoreBasePtr); - snapshotrecord->backingRecords = (struct BackingPageRecord *)model_malloc(sizeof(struct BackingPageRecord) * numbackingpages); - snapshotrecord->snapShots = (struct SnapShotRecord *)model_malloc(sizeof(struct SnapShotRecord) * numsnapshots); - snapshotrecord->lastSnapShot = 0; - snapshotrecord->lastBackingPage = 0; - snapshotrecord->lastRegion = 0; - snapshotrecord->maxRegions = nummemoryregions; - snapshotrecord->maxBackingPages = numbackingpages; - snapshotrecord->maxSnapShots = numsnapshots; + mprot_snap->backingStore = (snapshot_page_t *)PageAlignAddressUpward(mprot_snap->backingStoreBasePtr); + mprot_snap->backingRecords = (struct BackingPageRecord *)model_malloc(sizeof(struct BackingPageRecord) * numbackingpages); + mprot_snap->snapShots = (struct SnapShotRecord *)model_malloc(sizeof(struct SnapShotRecord) * numsnapshots); + mprot_snap->lastSnapShot = 0; + mprot_snap->lastBackingPage = 0; + mprot_snap->lastRegion = 0; + mprot_snap->maxRegions = nummemoryregions; + mprot_snap->maxBackingPages = numbackingpages; + mprot_snap->maxSnapShots = numsnapshots; } /** HandlePF is the page fault handler for mprotect based snapshotting @@ -147,62 +107,24 @@ static void HandlePF(int sig, siginfo_t *si, void *unused) } void* addr = ReturnPageAlignedAddress(si->si_addr); - unsigned int backingpage = snapshotrecord->lastBackingPage++; //Could run out of pages... - if (backingpage == snapshotrecord->maxBackingPages) { + unsigned int backingpage = mprot_snap->lastBackingPage++; //Could run out of pages... + if (backingpage == mprot_snap->maxBackingPages) { model_print("Out of backing pages at %p\n", si->si_addr); exit(EXIT_FAILURE); } //copy page - memcpy(&(snapshotrecord->backingStore[backingpage]), addr, sizeof(snapshot_page_t)); + memcpy(&(mprot_snap->backingStore[backingpage]), addr, sizeof(snapshot_page_t)); //remember where to copy page back to - snapshotrecord->backingRecords[backingpage].basePtrOfPage = addr; + mprot_snap->backingRecords[backingpage].basePtrOfPage = addr; //set protection to read/write if (mprotect(addr, sizeof(snapshot_page_t), PROT_READ | PROT_WRITE)) { perror("mprotect"); // Handle error by quitting? } } -#endif /* USE_MPROTECT_SNAPSHOT */ - -#if !USE_MPROTECT_SNAPSHOT -static void createSharedMemory() -{ - //step 1. create shared memory. - void *memMapBase = mmap(0, SHARED_MEMORY_DEFAULT + STACK_SIZE_DEFAULT, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); - if (MAP_FAILED == memMapBase) - FAILURE("mmap"); - - //Setup snapshot record at top of free region - snapshotrecord = (struct SnapShot *)memMapBase; - snapshotrecord->mSharedMemoryBase = (void *)((uintptr_t)memMapBase + sizeof(struct SnapShot)); - snapshotrecord->mStackBase = (void *)((uintptr_t)memMapBase + SHARED_MEMORY_DEFAULT); - snapshotrecord->mStackSize = STACK_SIZE_DEFAULT; - snapshotrecord->mIDToRollback = -1; - snapshotrecord->currSnapShotID = 0; -} - -/** - * Create a new mspace pointer for the non-snapshotting (i.e., inter-process - * shared) memory region. Only for fork-based snapshotting. - * - * @return The shared memory mspace - */ -mspace create_shared_mspace() -{ - if (!snapshotrecord) - createSharedMemory(); - return create_mspace_with_base((void *)(snapshotrecord->mSharedMemoryBase), SHARED_MEMORY_DEFAULT - sizeof(struct SnapShot), 1); -} -#endif - - -/** The initSnapshotLibrary function initializes the snapshot library. - * @param entryPoint the function that should run the program. - */ -#if USE_MPROTECT_SNAPSHOT -void initSnapshotLibrary(unsigned int numbackingpages, +static void mprot_snapshot_init(unsigned int numbackingpages, unsigned int numsnapshots, unsigned int nummemoryregions, unsigned int numheappages, VoidFuncPtr entryPoint) { @@ -238,7 +160,7 @@ void initSnapshotLibrary(unsigned int numbackingpages, memset(&si, 0, sizeof(si)); si.si_addr = ss.ss_sp; HandlePF(SIGSEGV, &si, NULL); - snapshotrecord->lastBackingPage--; //remove the fake page we copied + mprot_snap->lastBackingPage--; //remove the fake page we copied void *basemySpace = model_malloc((numheappages + 1) * PAGESIZE); void *pagealignedbase = PageAlignAddressUpward(basemySpace); @@ -252,12 +174,135 @@ void initSnapshotLibrary(unsigned int numbackingpages, entryPoint(); } -#else -void initSnapshotLibrary(unsigned int numbackingpages, + +static void mprot_add_to_snapshot(void *addr, unsigned int numPages) +{ + unsigned int memoryregion = mprot_snap->lastRegion++; + if (memoryregion == mprot_snap->maxRegions) { + model_print("Exceeded supported number of memory regions!\n"); + exit(EXIT_FAILURE); + } + + mprot_snap->regionsToSnapShot[memoryregion].basePtr = addr; + mprot_snap->regionsToSnapShot[memoryregion].sizeInPages = numPages; +} + +static snapshot_id mprot_take_snapshot() +{ + for (unsigned int region = 0; region < mprot_snap->lastRegion; region++) { + if (mprotect(mprot_snap->regionsToSnapShot[region].basePtr, mprot_snap->regionsToSnapShot[region].sizeInPages * sizeof(snapshot_page_t), PROT_READ) == -1) { + perror("mprotect"); + model_print("Failed to mprotect inside of takeSnapShot\n"); + exit(EXIT_FAILURE); + } + } + unsigned int snapshot = mprot_snap->lastSnapShot++; + if (snapshot == mprot_snap->maxSnapShots) { + model_print("Out of snapshots\n"); + exit(EXIT_FAILURE); + } + mprot_snap->snapShots[snapshot].firstBackingPage = mprot_snap->lastBackingPage; + + return snapshot; +} + +static void mprot_roll_back(snapshot_id theID) +{ +#if USE_MPROTECT_SNAPSHOT == 2 + if (mprot_snap->lastSnapShot == (theID + 1)) { + for (unsigned int page = mprot_snap->snapShots[theID].firstBackingPage; page < mprot_snap->lastBackingPage; page++) { + memcpy(mprot_snap->backingRecords[page].basePtrOfPage, &mprot_snap->backingStore[page], sizeof(snapshot_page_t)); + } + return; + } +#endif + + HashTable< void *, bool, uintptr_t, 4, model_malloc, model_calloc, model_free> duplicateMap; + for (unsigned int region = 0; region < mprot_snap->lastRegion; region++) { + if (mprotect(mprot_snap->regionsToSnapShot[region].basePtr, mprot_snap->regionsToSnapShot[region].sizeInPages * sizeof(snapshot_page_t), PROT_READ | PROT_WRITE) == -1) { + perror("mprotect"); + model_print("Failed to mprotect inside of takeSnapShot\n"); + exit(EXIT_FAILURE); + } + } + for (unsigned int page = mprot_snap->snapShots[theID].firstBackingPage; page < mprot_snap->lastBackingPage; page++) { + if (!duplicateMap.contains(mprot_snap->backingRecords[page].basePtrOfPage)) { + duplicateMap.put(mprot_snap->backingRecords[page].basePtrOfPage, true); + memcpy(mprot_snap->backingRecords[page].basePtrOfPage, &mprot_snap->backingStore[page], sizeof(snapshot_page_t)); + } + } + mprot_snap->lastSnapShot = theID; + mprot_snap->lastBackingPage = mprot_snap->snapShots[theID].firstBackingPage; + mprot_take_snapshot(); //Make sure current snapshot is still good...All later ones are cleared +} + +#else /* !USE_MPROTECT_SNAPSHOT */ + +#include + +#define SHARED_MEMORY_DEFAULT (100 * ((size_t)1 << 20)) // 100mb for the shared memory +#define STACK_SIZE_DEFAULT (((size_t)1 << 20) * 20) // 20 mb out of the above 100 mb for my stack + +struct fork_snapshotter { + void *mSharedMemoryBase; + void *mStackBase; + size_t mStackSize; + volatile snapshot_id mIDToRollback; + ucontext_t mContextToRollback; + snapshot_id currSnapShotID; +}; + +static struct fork_snapshotter *fork_snap = NULL; + +/** @statics +* These variables are necessary because the stack is shared region and +* there exists a race between all processes executing the same function. +* To avoid the problem above, we require variables allocated in 'safe' regions. +* The bug was actually observed with the forkID, these variables below are +* used to indicate the various contexts to which to switch to. +* +* @savedSnapshotContext: contains the point to which takesnapshot() call should switch to. +* @savedUserSnapshotContext: contains the point to which the process whose snapshotid is equal to the rollbackid should switch to +* @snapshotid: it is a running counter for the various forked processes snapshotid. it is incremented and set in a persistently shared record +*/ +static ucontext_t savedSnapshotContext; +static ucontext_t savedUserSnapshotContext; +static snapshot_id snapshotid = 0; + +static void createSharedMemory() +{ + //step 1. create shared memory. + void *memMapBase = mmap(0, SHARED_MEMORY_DEFAULT + STACK_SIZE_DEFAULT, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); + if (MAP_FAILED == memMapBase) + FAILURE("mmap"); + + //Setup snapshot record at top of free region + fork_snap = (struct fork_snapshotter *)memMapBase; + fork_snap->mSharedMemoryBase = (void *)((uintptr_t)memMapBase + sizeof(struct fork_snapshotter)); + fork_snap->mStackBase = (void *)((uintptr_t)memMapBase + SHARED_MEMORY_DEFAULT); + fork_snap->mStackSize = STACK_SIZE_DEFAULT; + fork_snap->mIDToRollback = -1; + fork_snap->currSnapShotID = 0; +} + +/** + * Create a new mspace pointer for the non-snapshotting (i.e., inter-process + * shared) memory region. Only for fork-based snapshotting. + * + * @return The shared memory mspace + */ +mspace create_shared_mspace() +{ + if (!fork_snap) + createSharedMemory(); + return create_mspace_with_base((void *)(fork_snap->mSharedMemoryBase), SHARED_MEMORY_DEFAULT - sizeof(struct fork_snapshotter), 1); +} + +static void fork_snapshot_init(unsigned int numbackingpages, unsigned int numsnapshots, unsigned int nummemoryregions, unsigned int numheappages, VoidFuncPtr entryPoint) { - if (!snapshotrecord) + if (!fork_snap) createSharedMemory(); void *base_model_snapshot_space = malloc((numheappages + 1) * PAGESIZE); @@ -267,7 +312,7 @@ void initSnapshotLibrary(unsigned int numbackingpages, //step 2 setup the stack context. ucontext_t newContext; getcontext(&newContext); - newContext.uc_stack.ss_sp = snapshotrecord->mStackBase; + newContext.uc_stack.ss_sp = fork_snap->mStackBase; newContext.uc_stack.ss_size = STACK_SIZE_DEFAULT; makecontext(&newContext, entryPoint, 0); /* switch to a new entryPoint context, on a new stack */ @@ -275,20 +320,20 @@ void initSnapshotLibrary(unsigned int numbackingpages, /* switch back here when takesnapshot is called */ pid_t forkedID = 0; - snapshotid = snapshotrecord->currSnapShotID; + snapshotid = fork_snap->currSnapShotID; /* This bool indicates that the current process's snapshotid is same as the id to which the rollback needs to occur */ bool rollback = false; while (true) { - snapshotrecord->currSnapShotID = snapshotid + 1; + fork_snap->currSnapShotID = snapshotid + 1; forkedID = fork(); if (0 == forkedID) { /* If the rollback bool is set, switch to the context we need to return to during a rollback. */ if (rollback) { - setcontext(&(snapshotrecord->mContextToRollback)); + setcontext(&(fork_snap->mContextToRollback)); } else { /*Child process which is forked as a result of takesnapshot call should switch back to the takesnapshot context*/ @@ -298,36 +343,74 @@ void initSnapshotLibrary(unsigned int numbackingpages, int status; int retVal; - SSDEBUG("The process id of child is %d and the process id of this process is %d and snapshot id is %d\n", + DEBUG("The process id of child is %d and the process id of this process is %d and snapshot id is %d\n", forkedID, getpid(), snapshotid); do { retVal = waitpid(forkedID, &status, 0); } while (-1 == retVal && errno == EINTR); - if (snapshotrecord->mIDToRollback != snapshotid) { + if (fork_snap->mIDToRollback != snapshotid) { exit(EXIT_SUCCESS); } rollback = true; } } } -#endif -/** The addMemoryRegionToSnapShot function assumes that addr is page aligned. +static snapshot_id fork_take_snapshot() +{ + swapcontext(&savedUserSnapshotContext, &savedSnapshotContext); + DEBUG("TAKESNAPSHOT RETURN\n"); + return snapshotid; +} + +static void fork_roll_back(snapshot_id theID) +{ + fork_snap->mIDToRollback = theID; + volatile int sTemp = 0; + getcontext(&fork_snap->mContextToRollback); + /* + * This is used to quit the process on rollback, so that the process + * which needs to rollback can quit allowing the process whose + * snapshotid matches the rollbackid to switch to this context and + * continue.... + */ + if (!sTemp) { + sTemp = 1; + DEBUG("Invoked rollback\n"); + exit(EXIT_SUCCESS); + } + /* + * This fix obviates the need for a finalize call. hence less dependences for model-checker.... + */ + fork_snap->mIDToRollback = -1; +} + +#endif /* !USE_MPROTECT_SNAPSHOT */ + +/** The initSnapshotLibrary function initializes the snapshot library. + * @param entryPoint the function that should run the program. */ -void addMemoryRegionToSnapShot(void *addr, unsigned int numPages) +void initSnapshotLibrary(unsigned int numbackingpages, + unsigned int numsnapshots, unsigned int nummemoryregions, + unsigned int numheappages, VoidFuncPtr entryPoint) { #if USE_MPROTECT_SNAPSHOT - unsigned int memoryregion = snapshotrecord->lastRegion++; - if (memoryregion == snapshotrecord->maxRegions) { - model_print("Exceeded supported number of memory regions!\n"); - exit(EXIT_FAILURE); - } + mprot_snapshot_init(numbackingpages, numsnapshots, nummemoryregions, numheappages, entryPoint); +#else + fork_snapshot_init(numbackingpages, numsnapshots, nummemoryregions, numheappages, entryPoint); +#endif +} - snapshotrecord->regionsToSnapShot[memoryregion].basePtr = addr; - snapshotrecord->regionsToSnapShot[memoryregion].sizeInPages = numPages; -#endif //NOT REQUIRED IN THE CASE OF FORK BASED SNAPSHOTS. +/** The addMemoryRegionToSnapShot function assumes that addr is page aligned. */ +void addMemoryRegionToSnapShot(void *addr, unsigned int numPages) +{ +#if USE_MPROTECT_SNAPSHOT + mprot_add_to_snapshot(addr, numPages); +#else + /* not needed for fork-based snapshotting */ +#endif } /** The takeSnapshot function takes a snapshot. @@ -336,25 +419,9 @@ void addMemoryRegionToSnapShot(void *addr, unsigned int numPages) snapshot_id takeSnapshot() { #if USE_MPROTECT_SNAPSHOT - for (unsigned int region = 0; region < snapshotrecord->lastRegion; region++) { - if (mprotect(snapshotrecord->regionsToSnapShot[region].basePtr, snapshotrecord->regionsToSnapShot[region].sizeInPages * sizeof(snapshot_page_t), PROT_READ) == -1) { - perror("mprotect"); - model_print("Failed to mprotect inside of takeSnapShot\n"); - exit(EXIT_FAILURE); - } - } - unsigned int snapshot = snapshotrecord->lastSnapShot++; - if (snapshot == snapshotrecord->maxSnapShots) { - model_print("Out of snapshots\n"); - exit(EXIT_FAILURE); - } - snapshotrecord->snapShots[snapshot].firstBackingPage = snapshotrecord->lastBackingPage; - - return snapshot; + return mprot_take_snapshot(); #else - swapcontext(&savedUserSnapshotContext, &savedSnapshotContext); - SSDEBUG("TAKESNAPSHOT RETURN\n"); - return snapshotid; + return fork_take_snapshot(); #endif } @@ -363,51 +430,9 @@ snapshot_id takeSnapshot() */ void rollBack(snapshot_id theID) { -#if USE_MPROTECT_SNAPSHOT == 2 - if (snapshotrecord->lastSnapShot == (theID + 1)) { - for (unsigned int page = snapshotrecord->snapShots[theID].firstBackingPage; page < snapshotrecord->lastBackingPage; page++) { - memcpy(snapshotrecord->backingRecords[page].basePtrOfPage, &snapshotrecord->backingStore[page], sizeof(snapshot_page_t)); - } - return; - } -#endif - #if USE_MPROTECT_SNAPSHOT - HashTable< void *, bool, uintptr_t, 4, model_malloc, model_calloc, model_free> duplicateMap; - for (unsigned int region = 0; region < snapshotrecord->lastRegion; region++) { - if (mprotect(snapshotrecord->regionsToSnapShot[region].basePtr, snapshotrecord->regionsToSnapShot[region].sizeInPages * sizeof(snapshot_page_t), PROT_READ | PROT_WRITE) == -1) { - perror("mprotect"); - model_print("Failed to mprotect inside of takeSnapShot\n"); - exit(EXIT_FAILURE); - } - } - for (unsigned int page = snapshotrecord->snapShots[theID].firstBackingPage; page < snapshotrecord->lastBackingPage; page++) { - if (!duplicateMap.contains(snapshotrecord->backingRecords[page].basePtrOfPage)) { - duplicateMap.put(snapshotrecord->backingRecords[page].basePtrOfPage, true); - memcpy(snapshotrecord->backingRecords[page].basePtrOfPage, &snapshotrecord->backingStore[page], sizeof(snapshot_page_t)); - } - } - snapshotrecord->lastSnapShot = theID; - snapshotrecord->lastBackingPage = snapshotrecord->snapShots[theID].firstBackingPage; - takeSnapshot(); //Make sure current snapshot is still good...All later ones are cleared + mprot_roll_back(theID); #else - snapshotrecord->mIDToRollback = theID; - volatile int sTemp = 0; - getcontext(&snapshotrecord->mContextToRollback); - /* - * This is used to quit the process on rollback, so that the process - * which needs to rollback can quit allowing the process whose - * snapshotid matches the rollbackid to switch to this context and - * continue.... - */ - if (!sTemp) { - sTemp = 1; - SSDEBUG("Invoked rollback\n"); - exit(EXIT_SUCCESS); - } - /* - * This fix obviates the need for a finalize call. hence less dependences for model-checker.... - */ - snapshotrecord->mIDToRollback = -1; + fork_roll_back(theID); #endif }