X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;ds=sidebyside;f=snapshot.cc;h=dab1a480d5910d84ed97ec2782e6fa73d7e50a85;hb=809d0f8a8f05ad2464c5e4df5b14ad2650d59fe8;hp=a562b91b4064f57c65f8d63964a0d3ca8038c035;hpb=c3ca7e70616c64ec1d21215e7b5ac51529affca5;p=c11tester.git diff --git a/snapshot.cc b/snapshot.cc index a562b91b..dab1a480 100644 --- a/snapshot.cc +++ b/snapshot.cc @@ -6,17 +6,25 @@ #include #include #include -#include #include "hashtable.h" #include "snapshot.h" #include "mymemory.h" #include "common.h" +#include "context.h" +#include "model.h" -#define FAILURE(mesg) { model_print("failed in the API: %s with errno relative message: %s\n", mesg, strerror(errno)); exit(EXIT_FAILURE); } +/** 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; @@ -33,172 +41,100 @@ struct BackingPageRecord { /* Struct for each memory region */ struct MemoryRegion { - void *basePtr; // base of memory region - int sizeInPages; // size of memory region in pages + void *basePtr; // base of memory region + int sizeInPages; // size of memory region in pages }; +/** ReturnPageAlignedAddress returns a page aligned address for the + * address being added as a side effect the numBytes are also changed. + */ +static void * ReturnPageAlignedAddress(void *addr) +{ + return (void *)(((uintptr_t)addr) & ~(PAGESIZE - 1)); +} + /* Primary struct for snapshotting system */ -struct SnapShot { - 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 - struct BackingPageRecord *backingRecords; //This pointer references an array of backingpagerecord's (same number of elements as backingstore - struct SnapShotRecord *snapShots; //This pointer references the snapshot array - - unsigned int lastSnapShot; //Stores the next snapshot record we should use - unsigned int lastBackingPage; //Stores the next backingpage we should use - unsigned int lastRegion; //Stores the next memory region to be used - - unsigned int maxRegions; //Stores the max number of memory regions we support - unsigned int maxBackingPages; //Stores the total number of backing pages - unsigned int maxSnapShots; //Stores the total number of snapshots we allow -}; +struct mprot_snapshotter { + mprot_snapshotter(unsigned int numbackingpages, unsigned int numsnapshots, unsigned int nummemoryregions); + ~mprot_snapshotter(); -#else + 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 + struct BackingPageRecord *backingRecords; //This pointer references an array of backingpagerecord's (same number of elements as backingstore + struct SnapShotRecord *snapShots; //This pointer references the snapshot array -#include + unsigned int lastSnapShot; //Stores the next snapshot record we should use + unsigned int lastBackingPage; //Stores the next backingpage we should use + unsigned int lastRegion; //Stores the next memory region to be used -#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 + unsigned int maxRegions; //Stores the max number of memory regions we support + unsigned int maxBackingPages; //Stores the total number of backing pages + unsigned int maxSnapShots; //Stores the total number of snapshots we allow -struct SnapShot { - void *mSharedMemoryBase; - void *mStackBase; - size_t mStackSize; - volatile snapshot_id mIDToRollback; - ucontext_t mContextToRollback; - snapshot_id currSnapShotID; + MEMALLOC }; -#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. - */ -static void * ReturnPageAlignedAddress(void *addr) +mprot_snapshotter::mprot_snapshotter(unsigned int backing_pages, unsigned int snapshots, unsigned int regions) : + lastSnapShot(0), + lastBackingPage(0), + lastRegion(0), + maxRegions(regions), + maxBackingPages(backing_pages), + maxSnapShots(snapshots) { - return (void *)(((uintptr_t)addr) & ~(PAGESIZE - 1)); + regionsToSnapShot = (struct MemoryRegion *)model_malloc(sizeof(struct MemoryRegion) * regions); + backingStoreBasePtr = (void *)model_malloc(sizeof(snapshot_page_t) * (backing_pages + 1)); + //Page align the backingstorepages + backingStore = (snapshot_page_t *)PageAlignAddressUpward(backingStoreBasePtr); + backingRecords = (struct BackingPageRecord *)model_malloc(sizeof(struct BackingPageRecord) * backing_pages); + snapShots = (struct SnapShotRecord *)model_malloc(sizeof(struct SnapShotRecord) * snapshots); } -/** The initSnapShotRecord method initialized the snapshotting data - * structures for the mprotect based snapshot. - */ -static void initSnapShotRecord(unsigned int numbackingpages, unsigned int numsnapshots, unsigned int nummemoryregions) +mprot_snapshotter::~mprot_snapshotter() { - 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)); - //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; + model_free(regionsToSnapShot); + model_free(backingStoreBasePtr); + model_free(backingRecords); + model_free(snapShots); } -/** HandlePF is the page fault handler for mprotect based snapshotting +/** mprot_handle_pf is the page fault handler for mprotect based snapshotting * algorithm. */ -static void HandlePF(int sig, siginfo_t *si, void *unused) +static void mprot_handle_pf(int sig, siginfo_t *si, void *unused) { if (si->si_code == SEGV_MAPERR) { - model_print("Real Fault at %p\n", si->si_addr); - print_trace(); + model_print("Segmentation fault at %p\n", si->si_addr); model_print("For debugging, place breakpoint at: %s:%d\n", - __FILE__, __LINE__); + __FILE__, __LINE__); + // print_trace(); // Trace printing may cause dynamic memory allocation exit(EXIT_FAILURE); } 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, - unsigned int numsnapshots, unsigned int nummemoryregions, - unsigned int numheappages, VoidFuncPtr entryPoint) +static void mprot_snapshot_init(unsigned int numbackingpages, + unsigned int numsnapshots, unsigned int nummemoryregions, + unsigned int numheappages) { /* Setup a stack for our signal handler.... */ stack_t ss; @@ -210,198 +146,354 @@ void initSnapshotLibrary(unsigned int numbackingpages, struct sigaction sa; sa.sa_flags = SA_SIGINFO | SA_NODEFER | SA_RESTART | SA_ONSTACK; sigemptyset(&sa.sa_mask); - sa.sa_sigaction = HandlePF; + sa.sa_sigaction = mprot_handle_pf; #ifdef MAC if (sigaction(SIGBUS, &sa, NULL) == -1) { - model_print("SIGACTION CANNOT BE INSTALLED\n"); + perror("sigaction(SIGBUS)"); exit(EXIT_FAILURE); } #endif if (sigaction(SIGSEGV, &sa, NULL) == -1) { - model_print("SIGACTION CANNOT BE INSTALLED\n"); + perror("sigaction(SIGSEGV)"); exit(EXIT_FAILURE); } - initSnapShotRecord(numbackingpages, numsnapshots, nummemoryregions); + mprot_snap = new mprot_snapshotter(numbackingpages, numsnapshots, nummemoryregions); - // EVIL HACK: We need to make sure that calls into the HandlePF method don't cause dynamic links + // EVIL HACK: We need to make sure that calls into the mprot_handle_pf method don't cause dynamic links // The problem is that we end up protecting state in the dynamic linker... // Solution is to call our signal handler before we start protecting stuff... siginfo_t si; memset(&si, 0, sizeof(si)); si.si_addr = ss.ss_sp; - HandlePF(SIGSEGV, &si, NULL); - snapshotrecord->lastBackingPage--; //remove the fake page we copied + mprot_handle_pf(SIGSEGV, &si, NULL); + mprot_snap->lastBackingPage--; //remove the fake page we copied void *basemySpace = model_malloc((numheappages + 1) * PAGESIZE); void *pagealignedbase = PageAlignAddressUpward(basemySpace); user_snapshot_space = create_mspace_with_base(pagealignedbase, numheappages * PAGESIZE, 1); - addMemoryRegionToSnapShot(pagealignedbase, numheappages); + snapshot_add_memory_region(pagealignedbase, numheappages); void *base_model_snapshot_space = model_malloc((numheappages + 1) * PAGESIZE); pagealignedbase = PageAlignAddressUpward(base_model_snapshot_space); model_snapshot_space = create_mspace_with_base(pagealignedbase, numheappages * PAGESIZE, 1); - addMemoryRegionToSnapShot(pagealignedbase, numheappages); - - entryPoint(); + snapshot_add_memory_region(pagealignedbase, numheappages); } -#else -void initSnapshotLibrary(unsigned int numbackingpages, - unsigned int numsnapshots, unsigned int nummemoryregions, - unsigned int numheappages, VoidFuncPtr entryPoint) -{ - if (!snapshotrecord) - createSharedMemory(); - - void *base_model_snapshot_space = malloc((numheappages + 1) * PAGESIZE); - void *pagealignedbase = PageAlignAddressUpward(base_model_snapshot_space); - model_snapshot_space = create_mspace_with_base(pagealignedbase, numheappages * PAGESIZE, 1); - - //step 2 setup the stack context. - ucontext_t newContext; - getcontext(&newContext); - newContext.uc_stack.ss_sp = snapshotrecord->mStackBase; - newContext.uc_stack.ss_size = STACK_SIZE_DEFAULT; - makecontext(&newContext, entryPoint, 0); - /* switch to a new entryPoint context, on a new stack */ - swapcontext(&savedSnapshotContext, &newContext); - - /* switch back here when takesnapshot is called */ - pid_t forkedID = 0; - snapshotid = snapshotrecord->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; - 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)); - } else { - /*Child process which is forked as a result of takesnapshot - call should switch back to the takesnapshot context*/ - setcontext(&savedUserSnapshotContext); - } - } else { - int status; - int retVal; - - 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) { - exit(EXIT_SUCCESS); - } - rollback = true; - } - } +static void mprot_startExecution(ucontext_t * context, VoidFuncPtr entryPoint) { + /* setup the shared-stack context */ + create_context(context, fork_snap->mStackBase, model_calloc(STACK_SIZE_DEFAULT, 1), STACK_SIZE_DEFAULT, entryPoint); } -#endif -/** The addMemoryRegionToSnapShot function assumes that addr is page aligned. - */ -void addMemoryRegionToSnapShot(void *addr, unsigned int numPages) +static void mprot_add_to_snapshot(void *addr, unsigned int numPages) { -#if USE_MPROTECT_SNAPSHOT - unsigned int memoryregion = snapshotrecord->lastRegion++; - if (memoryregion == snapshotrecord->maxRegions) { + unsigned int memoryregion = mprot_snap->lastRegion++; + if (memoryregion == mprot_snap->maxRegions) { model_print("Exceeded supported number of memory regions!\n"); exit(EXIT_FAILURE); } - snapshotrecord->regionsToSnapShot[memoryregion].basePtr = addr; - snapshotrecord->regionsToSnapShot[memoryregion].sizeInPages = numPages; -#endif //NOT REQUIRED IN THE CASE OF FORK BASED SNAPSHOTS. + DEBUG("snapshot region %p-%p (%u page%s)\n", + addr, (char *)addr + numPages * PAGESIZE, numPages, + numPages > 1 ? "s" : ""); + mprot_snap->regionsToSnapShot[memoryregion].basePtr = addr; + mprot_snap->regionsToSnapShot[memoryregion].sizeInPages = numPages; } -/** The takeSnapshot function takes a snapshot. - * @return The snapshot identifier. - */ -snapshot_id takeSnapshot() +static snapshot_id mprot_take_snapshot() { -#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) { + 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 = snapshotrecord->lastSnapShot++; - if (snapshot == snapshotrecord->maxSnapShots) { + unsigned int snapshot = mprot_snap->lastSnapShot++; + if (snapshot == mprot_snap->maxSnapShots) { model_print("Out of snapshots\n"); exit(EXIT_FAILURE); } - snapshotrecord->snapShots[snapshot].firstBackingPage = snapshotrecord->lastBackingPage; + mprot_snap->snapShots[snapshot].firstBackingPage = mprot_snap->lastBackingPage; return snapshot; -#else - swapcontext(&savedUserSnapshotContext, &savedSnapshotContext); - DEBUG("TAKESNAPSHOT RETURN\n"); - return snapshotid; -#endif } -/** The rollBack function rollback to the given snapshot identifier. - * @param theID is the snapshot identifier to rollback to. - */ -void rollBack(snapshot_id theID) +static void mprot_roll_back(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)); + 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 -#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) { + 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 = 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)); + 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)); } } - snapshotrecord->lastSnapShot = theID; - snapshotrecord->lastBackingPage = snapshotrecord->snapShots[theID].firstBackingPage; - takeSnapshot(); //Make sure current snapshot is still good...All later ones are cleared -#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.... + 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 */ + +#define SHARED_MEMORY_DEFAULT (200 * ((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 { + /** @brief Pointer to the shared (non-snapshot) memory heap base + * (NOTE: this has size SHARED_MEMORY_DEFAULT - sizeof(*fork_snap)) */ + void *mSharedMemoryBase; + + /** @brief Pointer to the shared (non-snapshot) stack region */ + void *mStackBase; + + /** @brief Size of the shared stack */ + size_t mStackSize; + + /** + * @brief Stores the ID that we are attempting to roll back to + * + * Used in inter-process communication so that each process can + * determine whether or not to take over execution (w/ matching ID) or + * exit (we're rolling back even further). Dubiously marked 'volatile' + * to prevent compiler optimizations from messing with the + * inter-process behavior. */ - 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.... + volatile snapshot_id mIDToRollback; + + /** + * @brief The context for the shared (non-snapshot) stack + * + * This context is passed between the various processes which represent + * various snapshot states. It should be used primarily for the + * "client-side" code, not the main snapshot loop. */ - snapshotrecord->mIDToRollback = -1; + ucontext_t shared_ctxt; + + /** @brief Inter-process tracking of the next snapshot ID */ + 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. + * + * @private_ctxt: the context which is internal to the current process. Used + * for running the internal snapshot/rollback loop. + * @exit_ctxt: a special context used just for exiting from a process (so we + * can use swapcontext() instead of setcontext() + hacks) + * @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 private_ctxt; +static ucontext_t exit_ctxt; +static snapshot_id snapshotid = 0; + +/** + * @brief Create a new context, with a given stack and entry function + * @param ctxt The context structure to fill + * @param stack The stack to run the new context in + * @param stacksize The size of the stack + * @param func The entry point function for the context + */ +static void create_context(ucontext_t *ctxt, void *stack, size_t stacksize, + void (*func)(void)) +{ + getcontext(ctxt); + ctxt->uc_stack.ss_sp = stack; + ctxt->uc_stack.ss_size = stacksize; + ctxt->uc_link = NULL; + makecontext(ctxt, func, 0); +} + +/** @brief An empty function, used for an "empty" context which just exits a + * process */ +static void fork_exit() +{ + _Exit(EXIT_SUCCESS); +} + +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 (memMapBase == MAP_FAILED) { + perror("mmap"); + exit(EXIT_FAILURE); + } + + //Setup snapshot record at top of free region + fork_snap = (struct fork_snapshotter *)memMapBase; + fork_snap->mSharedMemoryBase = (void *)((uintptr_t)memMapBase + sizeof(*fork_snap)); + 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(*fork_snap), 1); +} + +static void fork_snapshot_init(unsigned int numbackingpages, + unsigned int numsnapshots, unsigned int nummemoryregions, + unsigned int numheappages) +{ + if (!fork_snap) + createSharedMemory(); + + model_snapshot_space = create_mspace(numheappages * PAGESIZE, 1); +} + +volatile int modellock = 0; + +static void fork_loop() { + /* switch back here when takesnapshot is called */ + snapshotid = fork_snap->currSnapShotID; + if (model->params.nofork) { + setcontext(&fork_snap->shared_ctxt); + _Exit(EXIT_SUCCESS); + } + + while (true) { + pid_t forkedID; + fork_snap->currSnapShotID = snapshotid + 1; + + modellock = 1; + forkedID = fork(); + modellock = 0; + + if (0 == forkedID) { + setcontext(&fork_snap->shared_ctxt); + } else { + DEBUG("parent PID: %d, child PID: %d, snapshot ID: %d\n", + getpid(), forkedID, snapshotid); + + while (waitpid(forkedID, NULL, 0) < 0) { + /* waitpid() may be interrupted */ + if (errno != EINTR) { + perror("waitpid"); + exit(EXIT_FAILURE); + } + } + + if (fork_snap->mIDToRollback != snapshotid) + _Exit(EXIT_SUCCESS); + } + } +} + +static void fork_startExecution(ucontext_t *context, VoidFuncPtr entryPoint) { + /* setup an "exiting" context */ + char stack[128]; + create_context(&exit_ctxt, stack, sizeof(stack), fork_exit); + + /* setup the system context */ + create_context(context, fork_snap->mStackBase, STACK_SIZE_DEFAULT, entryPoint); + /* switch to a new entryPoint context, on a new stack */ + create_context(&private_ctxt, snapshot_calloc(STACK_SIZE_DEFAULT, 1), STACK_SIZE_DEFAULT, fork_loop); +} + +static snapshot_id fork_take_snapshot() { + model_swapcontext(&fork_snap->shared_ctxt, &private_ctxt); + DEBUG("TAKESNAPSHOT RETURN\n"); + return snapshotid; +} + +static void fork_roll_back(snapshot_id theID) +{ + DEBUG("Rollback\n"); + fork_snap->mIDToRollback = theID; + model_swapcontext(&fork_snap->shared_ctxt, &exit_ctxt); + fork_snap->mIDToRollback = -1; +} + +#endif /* !USE_MPROTECT_SNAPSHOT */ + +/** + * @brief Initializes the snapshot system + * @param entryPoint the function that should run the program. + */ +void snapshot_system_init(unsigned int numbackingpages, + unsigned int numsnapshots, unsigned int nummemoryregions, + unsigned int numheappages) +{ +#if USE_MPROTECT_SNAPSHOT + mprot_snapshot_init(numbackingpages, numsnapshots, nummemoryregions, numheappages); +#else + fork_snapshot_init(numbackingpages, numsnapshots, nummemoryregions, numheappages); +#endif +} + +void startExecution(ucontext_t *context, VoidFuncPtr entryPoint) +{ +#if USE_MPROTECT_SNAPSHOT + mprot_startExecution(context, entryPoint); +#else + fork_startExecution(context, entryPoint); +#endif +} + +/** Assumes that addr is page aligned. */ +void snapshot_add_memory_region(void *addr, unsigned int numPages) +{ +#if USE_MPROTECT_SNAPSHOT + mprot_add_to_snapshot(addr, numPages); +#else + /* not needed for fork-based snapshotting */ +#endif +} + +/** Takes a snapshot of memory. + * @return The snapshot identifier. + */ +snapshot_id take_snapshot() +{ +#if USE_MPROTECT_SNAPSHOT + return mprot_take_snapshot(); +#else + return fork_take_snapshot(); +#endif +} + +/** Rolls the memory state back to the given snapshot identifier. + * @param theID is the snapshot identifier to rollback to. + */ +void snapshot_roll_back(snapshot_id theID) +{ +#if USE_MPROTECT_SNAPSHOT + mprot_roll_back(theID); +#else + fork_roll_back(theID); #endif }