1 /* Copyright (c) 2008-2015. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
13 #include "src/internal_config.h"
14 #include "src/mc/mc_private.h"
15 #include "xbt/module.h"
16 #include <xbt/mmalloc.h>
17 #include "src/smpi/private.h"
20 #include "src/xbt/mmalloc/mmprivate.h"
22 #include "src/simix/smx_private.h"
24 #include <libunwind.h>
27 #include "src/mc/mc_private.h"
30 #include "src/mc/mc_snapshot.h"
31 #include "src/mc/mc_object_info.h"
32 #include "src/mc/mc_mmu.h"
33 #include "src/mc/mc_unw.h"
34 #include "src/mc/mc_protocol.h"
35 #include "src/mc/mc_smx.h"
36 #include "mc_hash.hpp"
38 #include "src/mc/RegionSnapshot.hpp"
39 #include "src/mc/ObjectInformation.hpp"
40 #include "src/mc/Frame.hpp"
41 #include "src/mc/Variable.hpp"
43 using simgrid::mc::remote;
47 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc,
48 "Logging specific to mc_checkpoint");
55 /************************************ Free functions **************************************/
56 /*****************************************************************************************/
58 /** @brief Restore a region from a snapshot
60 * @param reg Target region
62 static void restore(mc_mem_region_t region)
64 switch(region->storage_type()) {
65 case simgrid::mc::StorageType::NoData:
67 xbt_die("Storage type not supported");
70 case simgrid::mc::StorageType::Flat:
71 mc_model_checker->process().write_bytes(region->flat_data(),
72 region->size(), region->permanent_address());
75 case simgrid::mc::StorageType::Chunked:
76 mc_region_restore_sparse(&mc_model_checker->process(), region);
79 case simgrid::mc::StorageType::Privatized:
80 for (auto& p : region->privatized_data())
87 RegionSnapshot privatized_region(
88 RegionType region_type, void *start_addr, void* permanent_addr,
89 std::size_t size, const RegionSnapshot* ref_region
92 size_t process_count = MC_smpi_process_count();
94 // Read smpi_privatisation_regions from MCed:
95 smpi_privatisation_region_t remote_smpi_privatisation_regions;
96 mc_model_checker->process().read_variable(
97 "smpi_privatisation_regions",
98 &remote_smpi_privatisation_regions, sizeof(remote_smpi_privatisation_regions));
99 s_smpi_privatisation_region_t privatisation_regions[process_count];
100 mc_model_checker->process().read_bytes(
101 &privatisation_regions, sizeof(privatisation_regions),
102 remote(remote_smpi_privatisation_regions));
104 std::vector<simgrid::mc::RegionSnapshot> data;
105 data.reserve(process_count);
106 for (size_t i = 0; i < process_count; i++) {
107 const simgrid::mc::RegionSnapshot* ref_privatized_region = nullptr;
108 if (ref_region && ref_region->storage_type() == StorageType::Privatized)
109 ref_privatized_region = &ref_region->privatized_data()[i];
110 data.push_back(simgrid::mc::region(region_type, start_addr,
111 privatisation_regions[i].address, size, ref_privatized_region));
114 simgrid::mc::RegionSnapshot region = simgrid::mc::RegionSnapshot(
115 region_type, start_addr, permanent_addr, size);
116 region.privatized_data(std::move(data));
117 return std::move(region);
122 void add_region(int index, mc_snapshot_t snapshot,
123 simgrid::mc::RegionType type,
124 simgrid::mc::ObjectInformation* object_info,
125 void *start_addr, void* permanent_addr,
128 if (type == simgrid::mc::RegionType::Data)
129 xbt_assert(object_info, "Missing object info for object.");
130 else if (type == simgrid::mc::RegionType::Heap)
131 xbt_assert(!object_info, "Unexpected object info for heap region.");
133 simgrid::mc::RegionSnapshot const* ref_region = nullptr;
134 if (mc_model_checker->parent_snapshot_)
135 ref_region = mc_model_checker->parent_snapshot_->snapshot_regions[index].get();
137 simgrid::mc::RegionSnapshot region;
139 const bool privatization_aware = object_info
140 && mc_model_checker->process().privatized(*object_info);
141 if (privatization_aware && MC_smpi_process_count())
142 region = simgrid::mc::privatized_region(
143 type, start_addr, permanent_addr, size, ref_region);
146 region = simgrid::mc::region(type, start_addr, permanent_addr, size, ref_region);
148 region.object_info(object_info);
149 snapshot->snapshot_regions[index]
150 = std::unique_ptr<simgrid::mc::RegionSnapshot>(
151 new simgrid::mc::RegionSnapshot(std::move(region)));
155 static void get_memory_regions(simgrid::mc::Process* process, mc_snapshot_t snapshot)
157 const size_t n = process->object_infos.size();
158 snapshot->snapshot_regions.resize(n + 1);
160 for (auto const& object_info : process->object_infos) {
161 add_region(i, snapshot, simgrid::mc::RegionType::Data,
163 object_info->start_rw, object_info->start_rw,
164 object_info->end_rw - object_info->start_rw);
168 xbt_mheap_t heap = process->get_heap();
169 void *start_heap = heap->base;
170 void *end_heap = heap->breakval;
172 add_region(n, snapshot, simgrid::mc::RegionType::Heap, NULL,
173 start_heap, start_heap,
174 (char *) end_heap - (char *) start_heap);
175 snapshot->heap_bytes_used = mmalloc_get_bytes_used_remote(
177 process->get_malloc_info());
180 if (mc_model_checker->process().privatized() && MC_smpi_process_count()) {
181 // snapshot->privatization_index = smpi_loaded_page
182 mc_model_checker->process().read_variable(
183 "smpi_loaded_page", &snapshot->privatization_index,
184 sizeof(snapshot->privatization_index));
188 snapshot->privatization_index = simgrid::mc::ProcessIndexMissing;
192 /** \brief Fills the position of the segments (executable, read-only, read/write).
194 * `dl_iterate_phdr` would be more robust but would not work in cross-process.
196 void find_object_address(
197 std::vector<simgrid::xbt::VmMap> const& maps,
198 simgrid::mc::ObjectInformation* result)
200 char* file_name = xbt_strdup(result->file_name.c_str());
201 const char *name = xbt_basename(file_name);
202 for (size_t i = 0; i < maps.size(); ++i) {
203 simgrid::xbt::VmMap const& reg = maps[i];
204 if (maps[i].pathname.empty()
205 || strcmp(xbt_basename(maps[i].pathname.c_str()), name)) {
207 } else if ((reg.prot & PROT_WRITE)) {
208 xbt_assert(!result->start_rw,
209 "Multiple read-write segments for %s, not supported",
210 maps[i].pathname.c_str());
211 result->start_rw = (char*) reg.start_addr;
212 result->end_rw = (char*) reg.end_addr;
213 // .bss is usually after the .data:
214 simgrid::xbt::VmMap const& next = maps[i + 1];
215 if (next.pathname.empty() && (next.prot & PROT_WRITE)
216 && next.start_addr == reg.end_addr) {
217 result->end_rw = (char*) maps[i + 1].end_addr;
219 } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)) {
220 xbt_assert(!result->start_exec,
221 "Multiple executable segments for %s, not supported",
222 maps[i].pathname.c_str());
223 result->start_exec = (char*) reg.start_addr;
224 result->end_exec = (char*) reg.end_addr;
225 } else if ((reg.prot & PROT_READ) && !(reg.prot & PROT_EXEC)) {
226 xbt_assert(!result->start_ro,
227 "Multiple read only segments for %s, not supported",
228 maps[i].pathname.c_str());
229 result->start_ro = (char*) reg.start_addr;
230 result->end_ro = (char*) reg.end_addr;
234 result->start = result->start_rw;
235 if ((const void*) result->start_ro > result->start)
236 result->start = result->start_ro;
237 if ((const void*) result->start_exec > result->start)
238 result->start = result->start_exec;
240 result->end = result->end_rw;
241 if (result->end_ro && (const void*) result->end_ro < result->end)
242 result->end = result->end_ro;
243 if (result->end_exec && (const void*) result->end_exec > result->end)
244 result->end = result->end_exec;
246 xbt_assert(result->start_rw);
247 xbt_assert(result->start_exec);
251 /************************************* Take Snapshot ************************************/
252 /****************************************************************************************/
254 /** \brief Checks whether the variable is in scope for a given IP.
256 * A variable may be defined only from a given value of IP.
258 * \param var Variable description
259 * \param frame Scope description
260 * \param ip Instruction pointer
261 * \return true if the variable is valid
263 static bool valid_variable(simgrid::mc::Variable* var,
264 simgrid::mc::Frame* scope,
267 // The variable is not yet valid:
268 if (scope->range.begin() + var->start_scope > (std::uint64_t) ip)
274 static void fill_local_variables_values(mc_stack_frame_t stack_frame,
275 simgrid::mc::Frame* scope,
277 std::vector<s_local_variable>& result)
279 simgrid::mc::Process* process = &mc_model_checker->process();
281 if (!scope->range.contain(stack_frame->ip))
284 for(simgrid::mc::Variable& current_variable :
287 if (!valid_variable(¤t_variable, scope, (void *) stack_frame->ip))
291 // FIXME, get rid of `region_type`
292 if ((long) stack_frame->ip > (long) process->libsimgrid_info->start_exec)
297 s_local_variable_t new_var;
298 new_var.subprogram = stack_frame->frame;
299 new_var.ip = stack_frame->ip;
300 new_var.name = current_variable.name;
301 new_var.type = current_variable.type;
302 new_var.region = region_type;
303 new_var.address = nullptr;
305 if (current_variable.address != NULL) {
306 new_var.address = current_variable.address;
307 } else if (!current_variable.location_list.empty()) {
308 simgrid::dwarf::Location location =
309 simgrid::dwarf::resolve(
310 current_variable.location_list,
311 current_variable.object_info,
312 &(stack_frame->unw_cursor),
313 (void *) stack_frame->frame_base,
314 &mc_model_checker->process(), process_index);
316 if (!location.in_memory())
317 xbt_die("Cannot handle non-address variable");
318 new_var.address = location.address();
321 xbt_die("No address");
324 result.push_back(std::move(new_var));
327 // Recursive processing of nested scopes:
328 for(simgrid::mc::Frame& nested_scope : scope->scopes)
329 fill_local_variables_values(
330 stack_frame, &nested_scope, process_index, result);
333 static std::vector<s_local_variable> get_local_variables_values(
334 std::vector<s_mc_stack_frame_t>& stack_frames, int process_index)
336 std::vector<s_local_variable> variables;
337 for (s_mc_stack_frame_t& stack_frame : stack_frames)
338 fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables);
339 return std::move(variables);
342 static std::vector<s_mc_stack_frame_t> unwind_stack_frames(mc_unw_context_t stack_context)
344 simgrid::mc::Process* process = &mc_model_checker->process();
345 std::vector<s_mc_stack_frame_t> result;
349 // TODO, check condition check (unw_init_local==0 means end of frame)
350 if (mc_unw_init_cursor(&c, stack_context) != 0) {
352 xbt_die("Could not initialize stack unwinding");
357 s_mc_stack_frame_t stack_frame;
359 stack_frame.unw_cursor = c;
363 unw_get_reg(&c, UNW_REG_IP, &ip);
364 unw_get_reg(&c, UNW_REG_SP, &sp);
369 // TODO, use real addresses in frame_t instead of fixing it here
371 simgrid::mc::Frame* frame = process->find_function(remote(ip));
372 stack_frame.frame = frame;
375 stack_frame.frame_name = frame->name;
376 stack_frame.frame_base =
377 (unw_word_t) frame->frame_base(c);
379 stack_frame.frame_base = 0;
380 stack_frame.frame_name = std::string();
383 result.push_back(std::move(stack_frame));
385 /* Stop before context switch with maestro */
386 if (frame != nullptr &&
387 frame->name == "smx_ctx_sysv_wrapper")
390 int ret = unw_step(&c);
392 xbt_die("Unexpected end of stack.");
393 } else if (ret < 0) {
394 xbt_die("Error while unwinding stack");
398 if (result.empty()) {
399 XBT_INFO("unw_init_local failed");
403 return std::move(result);
406 static std::vector<s_mc_snapshot_stack_t> take_snapshot_stacks(mc_snapshot_t * snapshot)
408 std::vector<s_mc_snapshot_stack_t> res;
410 for (auto const& stack : mc_model_checker->process().stack_areas()) {
411 s_mc_snapshot_stack_t st;
413 // Read the context from remote process:
414 unw_context_t context;
415 mc_model_checker->process().read_bytes(
416 &context, sizeof(context), remote(stack.context));
418 if (mc_unw_init_context(&st.context, &mc_model_checker->process(),
420 xbt_die("Could not initialise the libunwind context.");
422 st.stack_frames = unwind_stack_frames(&st.context);
423 st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index);
424 st.process_index = stack.process_index;
426 unw_word_t sp = st.stack_frames[0].sp;
428 res.push_back(std::move(st));
431 (char*) stack.address + stack.size - (char*) sp;
432 (*snapshot)->stack_sizes.push_back(stack_size);
435 return std::move(res);
439 static void snapshot_handle_ignore(mc_snapshot_t snapshot)
441 xbt_assert(snapshot->process());
444 for (auto const& region : mc_model_checker->process().ignored_regions()) {
445 s_mc_snapshot_ignored_data_t ignored_data;
446 ignored_data.start = (void*)region.addr;
447 ignored_data.data.resize(region.size);
448 // TODO, we should do this once per privatization segment:
449 snapshot->process()->read_bytes(
450 ignored_data.data.data(), region.size, remote(region.addr),
451 simgrid::mc::ProcessIndexDisabled);
452 snapshot->ignored_data.push_back(std::move(ignored_data));
456 for(auto const& region : mc_model_checker->process().ignored_regions()) {
457 snapshot->process()->clear_bytes(remote(region.addr), region.size);
462 static void snapshot_ignore_restore(mc_snapshot_t snapshot)
464 for (auto const& ignored_data : snapshot->ignored_data)
465 snapshot->process()->write_bytes(
466 ignored_data.data.data(), ignored_data.data.size(),
467 remote(ignored_data.start));
470 static std::vector<s_fd_infos_t> get_current_fds(pid_t pid)
472 const size_t fd_dir_path_size = 20;
473 char fd_dir_path[fd_dir_path_size];
474 int res = snprintf(fd_dir_path, fd_dir_path_size,
475 "/proc/%lli/fd", (long long int) pid);
476 xbt_assert(res >= 0);
477 if ((size_t) res > fd_dir_path_size)
478 xbt_die("Unexpected buffer is too small for fd_dir_path");
480 DIR* fd_dir = opendir(fd_dir_path);
482 xbt_die("Cannot open directory '/proc/self/fd'\n");
484 std::vector<s_fd_infos_t> fds;
486 struct dirent* fd_number;
487 while ((fd_number = readdir(fd_dir))) {
489 int fd_value = xbt_str_parse_int(fd_number->d_name, "Found a non-numerical FD: %s. Freaking out!");
494 const size_t source_size = 25;
496 int res = snprintf(source, source_size, "/proc/%lli/fd/%s",
497 (long long int) pid, fd_number->d_name);
498 xbt_assert(res >= 0);
499 if ((size_t) res > source_size)
500 xbt_die("Unexpected buffer is too small for fd %s", fd_number->d_name);
502 const size_t link_size = 200;
504 res = readlink(source, link, link_size);
506 xbt_die("Could not read link for %s", source);
509 xbt_die("Buffer to small for link of %s", source);
514 if(smpi_is_privatisation_file(link))
518 // This is (probably) the DIR* we are reading:
519 // TODO, read all the file entries at once and close the DIR.*
520 if(strcmp(fd_dir_path, link) == 0)
523 // We don't handle them.
524 // It does not mean we should silently ignore them however.
525 if (strncmp(link, "pipe:", 5) == 0 || strncmp(link, "socket:", 7) == 0)
528 // If dot_output enabled, do not handle the corresponding file
529 if (dot_output != NULL && strcmp(xbt_basename(link), _sg_mc_dot_output_file) == 0)
532 // This is probably a shared memory used by lttng-ust:
533 if(strncmp("/dev/shm/ust-shm-tmp-", link, 21)==0)
536 // Add an entry for this FD in the snapshot:
538 fd.filename = std::string(link);
539 fd.number = fd_value;
540 fd.flags = fcntl(fd_value, F_GETFL) | fcntl(fd_value, F_GETFD) ;
541 fd.current_position = lseek(fd_value, 0, SEEK_CUR);
542 fds.push_back(std::move(fd));
546 return std::move(fds);
549 mc_snapshot_t take_snapshot(int num_state)
551 XBT_DEBUG("Taking snapshot %i", num_state);
553 simgrid::mc::Process* mc_process = &mc_model_checker->process();
555 mc_snapshot_t snapshot = new simgrid::mc::Snapshot(mc_process);
557 snapshot->num_state = num_state;
559 smx_process_t process;
560 MC_EACH_SIMIX_PROCESS(process,
561 snapshot->enabled_processes.insert(process->pid));
563 snapshot_handle_ignore(snapshot);
565 if (_sg_mc_snapshot_fds)
566 snapshot->current_fds = get_current_fds(process->pid);
568 const bool use_soft_dirty = _sg_mc_sparse_checkpoint && _sg_mc_soft_dirty;
570 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
571 get_memory_regions(mc_process, snapshot);
573 mc_process->reset_soft_dirty();
575 snapshot->to_ignore = mc_model_checker->process().ignored_heap();
577 if (_sg_mc_visited > 0 || strcmp(_sg_mc_property_file, "")) {
579 take_snapshot_stacks(&snapshot);
581 snapshot->hash = simgrid::mc::hash(*snapshot);
589 snapshot_ignore_restore(snapshot);
591 mc_model_checker->parent_snapshot_ = snapshot;
596 void restore_snapshot_regions(mc_snapshot_t snapshot)
598 for(std::unique_ptr<s_mc_mem_region_t> const& region : snapshot->snapshot_regions) {
599 // For privatized, variables we decided it was not necessary to take the snapshot:
601 restore(region.get());
605 // TODO, send a message to implement this in the MCed process
606 if(snapshot->privatization_index >= 0) {
607 // Fix the privatization mmap:
608 s_mc_restore_message message;
609 message.type = MC_MESSAGE_RESTORE;
610 message.index = snapshot->privatization_index;
611 mc_model_checker->process().send_message(message);
617 void restore_snapshot_fds(mc_snapshot_t snapshot)
619 if (mc_mode == MC_MODE_SERVER)
620 xbt_die("FD snapshot not implemented in client/server mode.");
622 for (auto const& fd : snapshot->current_fds) {
624 int new_fd = open(fd.filename.c_str(), fd.flags);
626 xbt_die("Could not reopen the file %s fo restoring the file descriptor",
627 fd.filename.c_str());
629 if (new_fd != fd.number) {
630 dup2(new_fd, fd.number);
633 lseek(fd.number, fd.current_position, SEEK_SET);
637 void restore_snapshot(mc_snapshot_t snapshot)
639 XBT_DEBUG("Restore snapshot %i", snapshot->num_state);
640 const bool use_soft_dirty = _sg_mc_sparse_checkpoint && _sg_mc_soft_dirty;
641 restore_snapshot_regions(snapshot);
642 if (_sg_mc_snapshot_fds)
643 restore_snapshot_fds(snapshot);
645 mc_model_checker->process().reset_soft_dirty();
646 snapshot_ignore_restore(snapshot);
647 mc_model_checker->process().cache_flags = 0;
649 mc_model_checker->parent_snapshot_ = snapshot;
656 mc_snapshot_t simcall_HANDLER_mc_snapshot(smx_simcall_t simcall)
658 return simgrid::mc::take_snapshot(1);