X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/bc48db087894fd960073b3120cebf90e6b2f8c02..3e5588fc9b34239f448a523211086b05c9fee29c:/src/mc/mc_checkpoint.cpp diff --git a/src/mc/mc_checkpoint.cpp b/src/mc/mc_checkpoint.cpp index 99b11bc3ef..aa9f21d46f 100644 --- a/src/mc/mc_checkpoint.cpp +++ b/src/mc/mc_checkpoint.cpp @@ -1,4 +1,4 @@ -/* Copyright (c) 2008-2015. The SimGrid Team. +/* Copyright (c) 2008-2017. The SimGrid Team. * All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it @@ -6,34 +6,38 @@ #include +#include +#include #include -#include #include -#include + +#ifndef WIN32 +#include +#endif #include "src/internal_config.h" -#include "src/mc/mc_private.h" +#include "src/mc/mc_private.hpp" +#include "src/smpi/include/private.hpp" +#include "xbt/file.hpp" +#include "xbt/mmalloc.h" #include "xbt/module.h" -#include -#include -#include "src/smpi/private.h" #include "src/xbt/mmalloc/mmprivate.h" -#include "src/simix/smx_private.h" +#include "src/simix/smx_private.hpp" #include #include -#include "src/mc/mc_private.h" +#include "src/mc/mc_private.hpp" #include -#include "src/mc/mc_snapshot.h" -#include "src/mc/mc_mmu.h" -#include "src/mc/mc_unw.h" -#include "src/mc/mc_protocol.h" -#include "src/mc/mc_smx.h" -#include "mc_hash.hpp" +#include "src/mc/mc_hash.hpp" +#include "src/mc/mc_mmu.hpp" +#include "src/mc/mc_smx.hpp" +#include "src/mc/mc_snapshot.hpp" +#include "src/mc/mc_unw.hpp" +#include "src/mc/remote/mc_protocol.h" #include "src/mc/RegionSnapshot.hpp" #include "src/mc/ObjectInformation.hpp" @@ -42,12 +46,11 @@ using simgrid::mc::remote; -extern "C" { +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc, "Logging specific to mc_checkpoint"); -XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc, - "Logging specific to mc_checkpoint"); - -} +#define PROT_RWX (PROT_READ | PROT_WRITE | PROT_EXEC) +#define PROT_RW (PROT_READ | PROT_WRITE) +#define PROT_RX (PROT_READ | PROT_EXEC) namespace simgrid { namespace mc { @@ -57,16 +60,11 @@ namespace mc { /** @brief Restore a region from a snapshot * - * @param reg Target region + * @param region Target region */ static void restore(mc_mem_region_t region) { switch(region->storage_type()) { - case simgrid::mc::StorageType::NoData: - default: - xbt_die("Storage type not supported"); - break; - case simgrid::mc::StorageType::Flat: mc_model_checker->process().write_bytes(region->flat_data().get(), region->size(), region->permanent_address()); @@ -80,46 +78,46 @@ static void restore(mc_mem_region_t region) for (auto& p : region->privatized_data()) restore(&p); break; + + default: // includes StorageType::NoData + xbt_die("Storage type not supported"); + break; } } #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(); - // Read smpi_privatisation_regions from MCed: - smpi_privatisation_region_t remote_smpi_privatisation_regions; + // Read smpi_privatization_regions from MCed: + smpi_privatization_region_t remote_smpi_privatization_regions; mc_model_checker->process().read_variable( - "smpi_privatisation_regions", - &remote_smpi_privatisation_regions, sizeof(remote_smpi_privatisation_regions)); - s_smpi_privatisation_region_t privatisation_regions[process_count]; + "smpi_privatization_regions", + &remote_smpi_privatization_regions, sizeof(remote_smpi_privatization_regions)); + s_smpi_privatization_region_t privatization_regions[process_count]; mc_model_checker->process().read_bytes( - &privatisation_regions, sizeof(privatisation_regions), - remote(remote_smpi_privatisation_regions)); + &privatization_regions, sizeof(privatization_regions), + remote(remote_smpi_privatization_regions)); 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)); - } + privatization_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, @@ -128,11 +126,7 @@ void add_region(int index, mc_snapshot_t snapshot, if (type == simgrid::mc::RegionType::Data) xbt_assert(object_info, "Missing object info for object."); 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(); + xbt_assert(not object_info, "Unexpected object info for heap region."); simgrid::mc::RegionSnapshot region; #if HAVE_SMPI @@ -140,10 +134,10 @@ void add_region(int index, mc_snapshot_t snapshot, && 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] @@ -152,7 +146,7 @@ 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::RemoteClient* process, simgrid::mc::Snapshot* snapshot) { const size_t n = process->object_infos.size(); snapshot->snapshot_regions.resize(n + 1); @@ -186,67 +180,74 @@ static void get_memory_regions(simgrid::mc::Process* process, mc_snapshot_t snap } /** \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* name = xbt_basename(result->file_name.c_str()); + std::string name = simgrid::xbt::Path(result->file_name).getBasename(); for (size_t i = 0; i < maps.size(); ++i) { simgrid::xbt::VmMap const& reg = maps[i]; 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); + std::string map_basename = simgrid::xbt::Path(maps[i].pathname).getBasename(); + if (map_basename != name) continue; - } - free(map_basename); - if ((reg.prot & PROT_WRITE)) { - xbt_assert(!result->start_rw, - "Multiple read-write segments for %s, not supported", - maps[i].pathname.c_str()); + + // This is the non-GNU_RELRO-part of the data segment: + if (reg.prot == PROT_RW) { + xbt_assert(not 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)) { - xbt_assert(!result->start_exec, - "Multiple executable segments for %s, not supported", + } + + // This is the text segment: + else if (reg.prot == PROT_RX) { + xbt_assert(not 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)) { - xbt_assert(!result->start_ro, - "Multiple read only segments for %s, not supported", - maps[i].pathname.c_str()); + + // 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(not result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str()); result->start_ro = (char*) reg.start_addr; result->end_ro = (char*) reg.end_addr; } } 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(name); + xbt_assert(result->start_exec || result->start_rw || result->start_ro); } /************************************* Take Snapshot ************************************/ @@ -257,7 +258,7 @@ void find_object_address( * A variable may be defined only from a given value of IP. * * \param var Variable description - * \param frame Scope description + * \param scope Scope description * \param ip Instruction pointer * \return true if the variable is valid * */ @@ -272,20 +273,17 @@ static bool valid_variable(simgrid::mc::Variable* var, return true; } -static void fill_local_variables_values(mc_stack_frame_t stack_frame, - simgrid::mc::Frame* scope, - int process_index, - std::vector& result) +static void fill_local_variables_values(mc_stack_frame_t stack_frame, simgrid::mc::Frame* scope, int process_index, + std::vector& result) { - simgrid::mc::Process* process = &mc_model_checker->process(); + simgrid::mc::RemoteClient* process = &mc_model_checker->process(); - if (!scope->range.contain(stack_frame->ip)) + if (not scope || not scope->range.contain(stack_frame->ip)) return; - for(simgrid::mc::Variable& current_variable : - scope->variables) { + for (simgrid::mc::Variable& current_variable : scope->variables) { - if (!valid_variable(¤t_variable, scope, (void *) stack_frame->ip)) + if (not valid_variable(¤t_variable, scope, (void*)stack_frame->ip)) continue; int region_type; @@ -305,7 +303,7 @@ static void fill_local_variables_values(mc_stack_frame_t stack_frame, if (current_variable.address != nullptr) new_var.address = current_variable.address; - else if (!current_variable.location_list.empty()) { + else if (not current_variable.location_list.empty()) { simgrid::dwarf::Location location = simgrid::dwarf::resolve( current_variable.location_list, @@ -314,7 +312,7 @@ static void fill_local_variables_values(mc_stack_frame_t stack_frame, (void *) stack_frame->frame_base, &mc_model_checker->process(), process_index); - if (!location.in_memory()) + if (not location.in_memory()) xbt_die("Cannot handle non-address variable"); new_var.address = location.address(); @@ -325,30 +323,28 @@ static void fill_local_variables_values(mc_stack_frame_t stack_frame, } // Recursive processing of nested scopes: - for(simgrid::mc::Frame& nested_scope : scope->scopes) + for (simgrid::mc::Frame& nested_scope : scope->scopes) fill_local_variables_values( stack_frame, &nested_scope, process_index, result); } -static std::vector get_local_variables_values( - std::vector& stack_frames, int process_index) +static std::vector get_local_variables_values(std::vector& stack_frames, + int process_index) { - std::vector variables; + 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(); + simgrid::mc::RemoteClient* 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) { @@ -356,7 +352,8 @@ static std::vector unwind_stack_frames(mc_unw_context_t stac stack_frame.unw_cursor = c; - unw_word_t ip, sp; + unw_word_t ip; + unw_word_t sp; unw_get_reg(&c, UNW_REG_IP, &ip); unw_get_reg(&c, UNW_REG_SP, &sp); @@ -397,10 +394,10 @@ 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; @@ -412,9 +409,7 @@ 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); @@ -426,17 +421,17 @@ 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()); - + // Copy the memory: for (auto const& region : mc_model_checker->process().ignored_regions()) { s_mc_snapshot_ignored_data_t ignored_data; @@ -450,12 +445,12 @@ static void snapshot_handle_ignore(mc_snapshot_t snapshot) } // Zero the memory: - for(auto const& region : mc_model_checker->process().ignored_regions()) + for (auto const& region : mc_model_checker->process().ignored_regions()) snapshot->process()->clear_bytes(remote(region.addr), region.size); } -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( @@ -507,7 +502,7 @@ static std::vector get_current_fds(pid_t pid) link[res] = '\0'; #if HAVE_SMPI - if(smpi_is_privatisation_file(link)) + if(smpi_is_privatization_file(link)) continue; #endif @@ -518,21 +513,18 @@ static std::vector get_current_fds(pid_t pid) // We don't handle them. // It does not mean we should silently ignore them however. - if (strncmp(link, "pipe:", 5) == 0 || strncmp(link, "socket:", 7) == 0) + if (strncmp(link, "pipe:", std::strlen("pipe:")) == 0 || strncmp(link, "socket:", std::strlen("socket:")) == 0) continue; // If dot_output enabled, do not handle the corresponding file if (dot_output != nullptr) { - char* link_basename = xbt_basename(link); - if (strcmp(link_basename, _sg_mc_dot_output_file) == 0) { - free(link_basename); + std::string link_basename = simgrid::xbt::Path(link).getBasename(); + if (link_basename == _sg_mc_dot_output_file) continue; - } - free(link_basename); } // This is probably a shared memory used by lttng-ust: - if(strncmp("/dev/shm/ust-shm-tmp-", link, 21)==0) + if(strncmp("/dev/shm/ust-shm-tmp-", link, std::strlen("/dev/shm/ust-shm-tmp-"))==0) continue; // Add an entry for this FD in the snapshot: @@ -545,39 +537,32 @@ 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(); + simgrid::mc::RemoteClient* 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, num_state); - snapshot->num_state = num_state; + for (auto const& p : mc_model_checker->process().actors()) + snapshot->enabled_processes.insert(p.copy.getBuffer()->pid); - 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); + if (_sg_mc_max_visited_states > 0 || strcmp(_sg_mc_property_file, "")) { + snapshot->stacks = take_snapshot_stacks(snapshot.get()); if (_sg_mc_hash) snapshot->hash = simgrid::mc::hash(*snapshot); else @@ -585,66 +570,54 @@ 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 (std::unique_ptr const& region : snapshot->snapshot_regions) { // For privatized, variables we decided it was not necessary to take the snapshot: if (region) restore(region.get()); } #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); + s_mc_message_restore message{MC_MESSAGE_RESTORE, snapshot->privatization_index}; + 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."); + xbt_die("FD snapshot not implemented in client/server mode."); for (auto const& fd : snapshot->current_fds) { - + int new_fd = open(fd.filename.c_str(), fd.flags); if (new_fd < 0) - xbt_die("Could not reopen the file %s fo restoring the file descriptor", - fd.filename.c_str()); + xbt_die("Could not reopen the file %s fo restoring the file descriptor", fd.filename.c_str()); if (new_fd != fd.number) { dup2(new_fd, fd.number); close(new_fd); - }; + } lseek(fd.number, fd.current_position, SEEK_SET); } } -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(); } }