X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/507785035fffb029577ed608142f0cb8a43f77e9..624f2dace76a17e378a5ceeee2b838d3c3152ed3:/src/mc/mc_checkpoint.cpp diff --git a/src/mc/mc_checkpoint.cpp b/src/mc/mc_checkpoint.cpp index 3e9a705c9f..d8c934365b 100644 --- a/src/mc/mc_checkpoint.cpp +++ b/src/mc/mc_checkpoint.cpp @@ -6,14 +6,20 @@ #include +#include #include #include #include +#ifndef WIN32 +#include +#endif + #include "src/internal_config.h" #include "src/mc/mc_private.h" #include "xbt/module.h" #include +#include #include "src/smpi/private.h" #include "src/xbt/mmalloc/mmprivate.h" @@ -40,13 +46,9 @@ using simgrid::mc::remote; -extern "C" { - XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc, "Logging specific to mc_checkpoint"); -} - namespace simgrid { namespace mc { @@ -81,10 +83,10 @@ static void restore(mc_mem_region_t region) } } -#ifdef HAVE_SMPI +#if HAVE_SMPI RegionSnapshot privatized_region( RegionType region_type, void *start_addr, void* permanent_addr, - std::size_t size, const RegionSnapshot* ref_region + std::size_t size ) { size_t process_count = MC_smpi_process_count(); @@ -101,23 +103,19 @@ RegionSnapshot privatized_region( std::vector data; data.reserve(process_count); - for (size_t i = 0; i < process_count; i++) { - const simgrid::mc::RegionSnapshot* ref_privatized_region = nullptr; - if (ref_region && ref_region->storage_type() == StorageType::Privatized) - ref_privatized_region = &ref_region->privatized_data()[i]; + for (size_t i = 0; i < process_count; i++) data.push_back(simgrid::mc::region(region_type, start_addr, - privatisation_regions[i].address, size, ref_privatized_region)); - } + privatisation_regions[i].address, size)); simgrid::mc::RegionSnapshot region = simgrid::mc::RegionSnapshot( region_type, start_addr, permanent_addr, size); region.privatized_data(std::move(data)); - return std::move(region); + return region; } #endif static -void add_region(int index, mc_snapshot_t snapshot, +void add_region(int index, simgrid::mc::Snapshot* snapshot, simgrid::mc::RegionType type, simgrid::mc::ObjectInformation* object_info, void *start_addr, void* permanent_addr, @@ -128,20 +126,16 @@ void add_region(int index, mc_snapshot_t snapshot, else if (type == simgrid::mc::RegionType::Heap) xbt_assert(!object_info, "Unexpected object info for heap region."); - simgrid::mc::RegionSnapshot const* ref_region = nullptr; - if (mc_model_checker->parent_snapshot_) - ref_region = mc_model_checker->parent_snapshot_->snapshot_regions[index].get(); - simgrid::mc::RegionSnapshot region; -#ifdef HAVE_SMPI +#if HAVE_SMPI const bool privatization_aware = object_info && mc_model_checker->process().privatized(*object_info); if (privatization_aware && MC_smpi_process_count()) region = simgrid::mc::privatized_region( - type, start_addr, permanent_addr, size, ref_region); + type, start_addr, permanent_addr, size); else #endif - region = simgrid::mc::region(type, start_addr, permanent_addr, size, ref_region); + region = simgrid::mc::region(type, start_addr, permanent_addr, size); region.object_info(object_info); snapshot->snapshot_regions[index] @@ -150,7 +144,7 @@ void add_region(int index, mc_snapshot_t snapshot, return; } -static void get_memory_regions(simgrid::mc::Process* process, mc_snapshot_t snapshot) +static void get_memory_regions(simgrid::mc::Process* process, simgrid::mc::Snapshot* snapshot) { const size_t n = process->object_infos.size(); snapshot->snapshot_regions.resize(n + 1); @@ -172,7 +166,7 @@ static void get_memory_regions(simgrid::mc::Process* process, mc_snapshot_t snap heap->heaplimit, process->get_malloc_info()); -#ifdef HAVE_SMPI +#if HAVE_SMPI if (mc_model_checker->process().privatized() && MC_smpi_process_count()) // snapshot->privatization_index = smpi_loaded_page mc_model_checker->process().read_variable( @@ -183,39 +177,66 @@ static void get_memory_regions(simgrid::mc::Process* process, mc_snapshot_t snap snapshot->privatization_index = simgrid::mc::ProcessIndexMissing; } +#define PROT_RWX (PROT_READ | PROT_WRITE | PROT_EXEC) +#define PROT_RW (PROT_READ | PROT_WRITE) +#define PROT_RX (PROT_READ | PROT_EXEC) + /** \brief Fills the position of the segments (executable, read-only, read/write). - * - * `dl_iterate_phdr` would be more robust but would not work in cross-process. * */ +// TODO, use the ELF segment information for more robustness void find_object_address( std::vector const& maps, simgrid::mc::ObjectInformation* result) { - char* file_name = xbt_strdup(result->file_name.c_str()); - const char *name = xbt_basename(file_name); + char* name = xbt_basename(result->file_name.c_str()); + for (size_t i = 0; i < maps.size(); ++i) { simgrid::xbt::VmMap const& reg = maps[i]; - if (maps[i].pathname.empty() - || strcmp(xbt_basename(maps[i].pathname.c_str()), name)) { - // Nothing to do - } else if ((reg.prot & PROT_WRITE)) { + if (maps[i].pathname.empty()) + continue; + char* map_basename = xbt_basename(maps[i].pathname.c_str()); + if (strcmp(name, map_basename) != 0) { + free(map_basename); + continue; + } + free(map_basename); + + // This is the non-GNU_RELRO-part of the data segment: + if (reg.prot == PROT_RW) { xbt_assert(!result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str()); result->start_rw = (char*) reg.start_addr; result->end_rw = (char*) reg.end_addr; - // .bss is usually after the .data: - simgrid::xbt::VmMap const& next = maps[i + 1]; - if (next.pathname.empty() && (next.prot & PROT_WRITE) - && next.start_addr == reg.end_addr) + + // The next VMA might be end of the data segment: + if (i + 1 < maps.size() + && maps[i + 1].pathname.empty() + && maps[i + 1].prot == PROT_RW + && maps[i + 1].start_addr == reg.end_addr) result->end_rw = (char*) maps[i + 1].end_addr; - } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)) { + } + + // This is the text segment: + else if (reg.prot == PROT_RX) { xbt_assert(!result->start_exec, "Multiple executable segments for %s, not supported", maps[i].pathname.c_str()); result->start_exec = (char*) reg.start_addr; result->end_exec = (char*) reg.end_addr; - } else if ((reg.prot & PROT_READ) && !(reg.prot & PROT_EXEC)) { + + // The next VMA might be end of the data segment: + if (i + 1 < maps.size() + && maps[i + 1].pathname.empty() + && maps[i + 1].prot == PROT_RW + && maps[i + 1].start_addr == reg.end_addr) { + result->start_rw = (char*) maps[i + 1].start_addr; + result->end_rw = (char*) maps[i + 1].end_addr; + } + } + + // This is the GNU_RELRO-part of the data segment: + else if (reg.prot == PROT_READ) { xbt_assert(!result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str()); @@ -225,20 +246,20 @@ void find_object_address( } result->start = result->start_rw; - if ((const void*) result->start_ro > result->start) + if ((const void*) result->start_ro < result->start) result->start = result->start_ro; - if ((const void*) result->start_exec > result->start) + if ((const void*) result->start_exec < result->start) result->start = result->start_exec; result->end = result->end_rw; - if (result->end_ro && (const void*) result->end_ro < result->end) + if (result->end_ro && (const void*) result->end_ro > result->end) result->end = result->end_ro; if (result->end_exec && (const void*) result->end_exec > result->end) result->end = result->end_exec; - xbt_assert(result->start_rw); - xbt_assert(result->start_exec); - free(file_name); + xbt_assert(result->start_exec || result->start_rw || result->start_ro); + + free(name); } /************************************* Take Snapshot ************************************/ @@ -328,19 +349,17 @@ static std::vector get_local_variables_values( std::vector variables; for (s_mc_stack_frame_t& stack_frame : stack_frames) fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables); - return std::move(variables); + return variables; } -static std::vector unwind_stack_frames(mc_unw_context_t stack_context) +static std::vector unwind_stack_frames(simgrid::mc::UnwindContext* stack_context) { simgrid::mc::Process* process = &mc_model_checker->process(); std::vector result; - unw_cursor_t c; + unw_cursor_t c = stack_context->cursor(); // TODO, check condition check (unw_init_local==0 means end of frame) - if (mc_unw_init_cursor(&c, stack_context) != 0) - xbt_die("Could not initialize stack unwinding"); while (1) { @@ -389,10 +408,10 @@ static std::vector unwind_stack_frames(mc_unw_context_t stac xbt_abort(); } - return std::move(result); + return result; }; -static std::vector take_snapshot_stacks(mc_snapshot_t * snapshot) +static std::vector take_snapshot_stacks(simgrid::mc::Snapshot* snapshot) { std::vector res; @@ -404,9 +423,7 @@ static std::vector take_snapshot_stacks(mc_snapshot_t * s mc_model_checker->process().read_bytes( &context, sizeof(context), remote(stack.context)); - if (mc_unw_init_context(&st.context, &mc_model_checker->process(), - &context) < 0) - xbt_die("Could not initialise the libunwind context."); + st.context.initialize(&mc_model_checker->process(), &context); st.stack_frames = unwind_stack_frames(&st.context); st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index); @@ -418,14 +435,14 @@ static std::vector take_snapshot_stacks(mc_snapshot_t * s size_t stack_size = (char*) stack.address + stack.size - (char*) sp; - (*snapshot)->stack_sizes.push_back(stack_size); + snapshot->stack_sizes.push_back(stack_size); } - return std::move(res); + return res; } -static void snapshot_handle_ignore(mc_snapshot_t snapshot) +static void snapshot_handle_ignore(simgrid::mc::Snapshot* snapshot) { xbt_assert(snapshot->process()); @@ -447,7 +464,7 @@ static void snapshot_handle_ignore(mc_snapshot_t snapshot) } -static void snapshot_ignore_restore(mc_snapshot_t snapshot) +static void snapshot_ignore_restore(simgrid::mc::Snapshot* snapshot) { for (auto const& ignored_data : snapshot->ignored_data) snapshot->process()->write_bytes( @@ -498,7 +515,7 @@ static std::vector get_current_fds(pid_t pid) link[res] = '\0'; -#ifdef HAVE_SMPI +#if HAVE_SMPI if(smpi_is_privatisation_file(link)) continue; #endif @@ -514,8 +531,14 @@ static std::vector get_current_fds(pid_t pid) continue; // If dot_output enabled, do not handle the corresponding file - if (dot_output != nullptr && strcmp(xbt_basename(link), _sg_mc_dot_output_file) == 0) - continue; + if (dot_output != nullptr) { + char* link_basename = xbt_basename(link); + if (strcmp(link_basename, _sg_mc_dot_output_file) == 0) { + free(link_basename); + continue; + } + free(link_basename); + } // This is probably a shared memory used by lttng-ust: if(strncmp("/dev/shm/ust-shm-tmp-", link, 21)==0) @@ -531,39 +554,35 @@ static std::vector get_current_fds(pid_t pid) } closedir (fd_dir); - return std::move(fds); + return fds; } -mc_snapshot_t take_snapshot(int num_state) +std::shared_ptr take_snapshot(int num_state) { XBT_DEBUG("Taking snapshot %i", num_state); simgrid::mc::Process* mc_process = &mc_model_checker->process(); - mc_snapshot_t snapshot = new simgrid::mc::Snapshot(mc_process); + std::shared_ptr snapshot = + std::make_shared(mc_process); snapshot->num_state = num_state; for (auto& p : mc_model_checker->process().simix_processes()) snapshot->enabled_processes.insert(p.copy.pid); - snapshot_handle_ignore(snapshot); + snapshot_handle_ignore(snapshot.get()); if (_sg_mc_snapshot_fds) snapshot->current_fds = get_current_fds(mc_model_checker->process().pid()); - const bool use_soft_dirty = _sg_mc_sparse_checkpoint && _sg_mc_soft_dirty; - /* Save the std heap and the writable mapped pages of libsimgrid and binary */ - get_memory_regions(mc_process, snapshot); - if (use_soft_dirty) - mc_process->reset_soft_dirty(); + get_memory_regions(mc_process, snapshot.get()); snapshot->to_ignore = mc_model_checker->process().ignored_heap(); if (_sg_mc_visited > 0 || strcmp(_sg_mc_property_file, "")) { - snapshot->stacks = - take_snapshot_stacks(&snapshot); + snapshot->stacks = take_snapshot_stacks(snapshot.get()); if (_sg_mc_hash) snapshot->hash = simgrid::mc::hash(*snapshot); else @@ -571,14 +590,12 @@ mc_snapshot_t take_snapshot(int num_state) } else snapshot->hash = 0; - snapshot_ignore_restore(snapshot); - if (use_soft_dirty) - mc_model_checker->parent_snapshot_ = snapshot; + snapshot_ignore_restore(snapshot.get()); return snapshot; } static inline -void restore_snapshot_regions(mc_snapshot_t snapshot) +void restore_snapshot_regions(simgrid::mc::Snapshot* snapshot) { for(std::unique_ptr const& region : snapshot->snapshot_regions) { // For privatized, variables we decided it was not necessary to take the snapshot: @@ -586,20 +603,20 @@ void restore_snapshot_regions(mc_snapshot_t snapshot) restore(region.get()); } -#ifdef HAVE_SMPI +#if HAVE_SMPI // TODO, send a message to implement this in the MCed process if(snapshot->privatization_index >= 0) { // Fix the privatization mmap: s_mc_restore_message message; message.type = MC_MESSAGE_RESTORE; message.index = snapshot->privatization_index; - mc_model_checker->process().send_message(message); + mc_model_checker->process().getChannel().send(message); } #endif } static inline -void restore_snapshot_fds(mc_snapshot_t snapshot) +void restore_snapshot_fds(simgrid::mc::Snapshot* snapshot) { if (mc_mode == MC_MODE_SERVER) xbt_die("FD snapshot not implemented in client/server mode."); @@ -618,19 +635,14 @@ void restore_snapshot_fds(mc_snapshot_t snapshot) } } -void restore_snapshot(mc_snapshot_t snapshot) +void restore_snapshot(std::shared_ptr snapshot) { XBT_DEBUG("Restore snapshot %i", snapshot->num_state); - const bool use_soft_dirty = _sg_mc_sparse_checkpoint && _sg_mc_soft_dirty; - restore_snapshot_regions(snapshot); + restore_snapshot_regions(snapshot.get()); if (_sg_mc_snapshot_fds) - restore_snapshot_fds(snapshot); - if (use_soft_dirty) - mc_model_checker->process().reset_soft_dirty(); - snapshot_ignore_restore(snapshot); - mc_model_checker->process().cache_flags = 0; - if (use_soft_dirty) - mc_model_checker->parent_snapshot_ = snapshot; + restore_snapshot_fds(snapshot.get()); + snapshot_ignore_restore(snapshot.get()); + mc_model_checker->process().clear_cache(); } }