1 /* Copyright (c) 2008-2018. 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. */
18 #include "src/internal_config.h"
19 #include "src/mc/mc_private.hpp"
20 #include "src/smpi/include/private.hpp"
21 #include "xbt/file.hpp"
22 #include "xbt/mmalloc.h"
23 #include "xbt/module.h"
25 #include "src/xbt/mmalloc/mmprivate.h"
27 #include "src/simix/smx_private.hpp"
29 #include <libunwind.h>
32 #include "src/mc/mc_private.hpp"
35 #include "src/mc/mc_config.hpp"
36 #include "src/mc/mc_hash.hpp"
37 #include "src/mc/mc_mmu.hpp"
38 #include "src/mc/mc_smx.hpp"
39 #include "src/mc/mc_snapshot.hpp"
40 #include "src/mc/mc_unw.hpp"
41 #include "src/mc/remote/mc_protocol.h"
43 #include "src/mc/RegionSnapshot.hpp"
44 #include "src/mc/ObjectInformation.hpp"
45 #include "src/mc/Frame.hpp"
46 #include "src/mc/Variable.hpp"
48 using simgrid::mc::remote;
50 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc, "Logging specific to mc_checkpoint");
52 #define PROT_RWX (PROT_READ | PROT_WRITE | PROT_EXEC)
53 #define PROT_RW (PROT_READ | PROT_WRITE)
54 #define PROT_RX (PROT_READ | PROT_EXEC)
55 extern std::string _sg_mc_property_file;
56 extern std::string _sg_mc_dot_output_file;
61 /************************************ Free functions **************************************/
62 /*****************************************************************************************/
64 /** @brief Restore a region from a snapshot
66 * @param region Target region
68 static void restore(mc_mem_region_t region)
70 switch(region->storage_type()) {
71 case simgrid::mc::StorageType::Flat:
72 mc_model_checker->process().write_bytes(region->flat_data().get(),
73 region->size(), region->permanent_address());
76 case simgrid::mc::StorageType::Chunked:
77 mc_region_restore_sparse(&mc_model_checker->process(), region);
80 case simgrid::mc::StorageType::Privatized:
81 for (auto& p : region->privatized_data())
85 default: // includes StorageType::NoData
86 xbt_die("Storage type not supported");
92 RegionSnapshot privatized_region(
93 RegionType region_type, void *start_addr, void* permanent_addr,
97 size_t process_count = MC_smpi_process_count();
99 // Read smpi_privatization_regions from MCed:
100 smpi_privatization_region_t remote_smpi_privatization_regions;
101 mc_model_checker->process().read_variable(
102 "smpi_privatization_regions",
103 &remote_smpi_privatization_regions, sizeof(remote_smpi_privatization_regions));
104 s_smpi_privatization_region_t privatization_regions[process_count];
105 mc_model_checker->process().read_bytes(
106 &privatization_regions, sizeof(privatization_regions),
107 remote(remote_smpi_privatization_regions));
109 std::vector<simgrid::mc::RegionSnapshot> data;
110 data.reserve(process_count);
111 for (size_t i = 0; i < process_count; i++)
112 data.push_back(simgrid::mc::region(region_type, start_addr,
113 privatization_regions[i].address, size));
115 simgrid::mc::RegionSnapshot region = simgrid::mc::RegionSnapshot(
116 region_type, start_addr, permanent_addr, size);
117 region.privatized_data(std::move(data));
123 void add_region(int index, simgrid::mc::Snapshot* snapshot,
124 simgrid::mc::RegionType type,
125 simgrid::mc::ObjectInformation* object_info,
126 void *start_addr, void* permanent_addr,
129 if (type == simgrid::mc::RegionType::Data)
130 xbt_assert(object_info, "Missing object info for object.");
131 else if (type == simgrid::mc::RegionType::Heap)
132 xbt_assert(not object_info, "Unexpected object info for heap region.");
134 simgrid::mc::RegionSnapshot region;
136 const bool privatization_aware = object_info
137 && mc_model_checker->process().privatized(*object_info);
138 if (privatization_aware && MC_smpi_process_count())
139 region = simgrid::mc::privatized_region(
140 type, start_addr, permanent_addr, size);
143 region = simgrid::mc::region(type, start_addr, permanent_addr, size);
145 region.object_info(object_info);
146 snapshot->snapshot_regions[index]
147 = std::unique_ptr<simgrid::mc::RegionSnapshot>(
148 new simgrid::mc::RegionSnapshot(std::move(region)));
151 static void get_memory_regions(simgrid::mc::RemoteClient* process, simgrid::mc::Snapshot* snapshot)
153 const size_t n = process->object_infos.size();
154 snapshot->snapshot_regions.resize(n + 1);
156 for (auto const& object_info : process->object_infos)
157 add_region(i++, snapshot, simgrid::mc::RegionType::Data,
159 object_info->start_rw, object_info->start_rw,
160 object_info->end_rw - object_info->start_rw);
162 xbt_mheap_t heap = process->get_heap();
163 void *start_heap = heap->base;
164 void *end_heap = heap->breakval;
166 add_region(n, snapshot, simgrid::mc::RegionType::Heap, nullptr,
167 start_heap, start_heap,
168 (char *) end_heap - (char *) start_heap);
169 snapshot->heap_bytes_used = mmalloc_get_bytes_used_remote(
171 process->get_malloc_info());
174 if (mc_model_checker->process().privatized() && MC_smpi_process_count())
175 // snapshot->privatization_index = smpi_loaded_page
176 mc_model_checker->process().read_variable(
177 "smpi_loaded_page", &snapshot->privatization_index,
178 sizeof(snapshot->privatization_index));
181 snapshot->privatization_index = simgrid::mc::ProcessIndexMissing;
184 /** \brief Fills the position of the segments (executable, read-only, read/write).
186 // TODO, use the ELF segment information for more robustness
187 void find_object_address(
188 std::vector<simgrid::xbt::VmMap> const& maps,
189 simgrid::mc::ObjectInformation* result)
191 std::string name = simgrid::xbt::Path(result->file_name).getBasename();
193 for (size_t i = 0; i < maps.size(); ++i) {
194 simgrid::xbt::VmMap const& reg = maps[i];
195 if (maps[i].pathname.empty())
197 std::string map_basename = simgrid::xbt::Path(maps[i].pathname).getBasename();
198 if (map_basename != name)
201 // This is the non-GNU_RELRO-part of the data segment:
202 if (reg.prot == PROT_RW) {
203 xbt_assert(not result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str());
204 result->start_rw = (char*) reg.start_addr;
205 result->end_rw = (char*) reg.end_addr;
207 // The next VMA might be end of the data segment:
208 if (i + 1 < maps.size()
209 && maps[i + 1].pathname.empty()
210 && maps[i + 1].prot == PROT_RW
211 && maps[i + 1].start_addr == reg.end_addr)
212 result->end_rw = (char*) maps[i + 1].end_addr;
215 // This is the text segment:
216 else if (reg.prot == PROT_RX) {
217 xbt_assert(not result->start_exec, "Multiple executable segments for %s, not supported",
218 maps[i].pathname.c_str());
219 result->start_exec = (char*) reg.start_addr;
220 result->end_exec = (char*) reg.end_addr;
222 // The next VMA might be end of the data segment:
223 if (i + 1 < maps.size()
224 && maps[i + 1].pathname.empty()
225 && maps[i + 1].prot == PROT_RW
226 && maps[i + 1].start_addr == reg.end_addr) {
227 result->start_rw = (char*) maps[i + 1].start_addr;
228 result->end_rw = (char*) maps[i + 1].end_addr;
232 // This is the GNU_RELRO-part of the data segment:
233 else if (reg.prot == PROT_READ) {
234 xbt_assert(not result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str());
235 result->start_ro = (char*) reg.start_addr;
236 result->end_ro = (char*) reg.end_addr;
240 result->start = result->start_rw;
241 if ((const void*) result->start_ro < result->start)
242 result->start = result->start_ro;
243 if ((const void*) result->start_exec < result->start)
244 result->start = result->start_exec;
246 result->end = result->end_rw;
247 if (result->end_ro && (const void*) result->end_ro > result->end)
248 result->end = result->end_ro;
249 if (result->end_exec && (const void*) result->end_exec > result->end)
250 result->end = result->end_exec;
252 xbt_assert(result->start_exec || result->start_rw || result->start_ro);
255 /************************************* Take Snapshot ************************************/
256 /****************************************************************************************/
258 /** \brief Checks whether the variable is in scope for a given IP.
260 * A variable may be defined only from a given value of IP.
262 * \param var Variable description
263 * \param scope Scope description
264 * \param ip Instruction pointer
265 * \return true if the variable is valid
267 static bool valid_variable(simgrid::mc::Variable* var,
268 simgrid::mc::Frame* scope,
271 // The variable is not yet valid:
272 if (scope->range.begin() + var->start_scope > (std::uint64_t) ip)
278 static void fill_local_variables_values(mc_stack_frame_t stack_frame, simgrid::mc::Frame* scope, int process_index,
279 std::vector<s_local_variable_t>& result)
281 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
283 if (not scope || not scope->range.contain(stack_frame->ip))
286 for (simgrid::mc::Variable& current_variable : scope->variables) {
288 if (not valid_variable(¤t_variable, scope, (void*)stack_frame->ip))
292 // FIXME, get rid of `region_type`
293 if ((long) stack_frame->ip > (long) process->libsimgrid_info->start_exec)
298 s_local_variable_t new_var;
299 new_var.subprogram = stack_frame->frame;
300 new_var.ip = stack_frame->ip;
301 new_var.name = current_variable.name;
302 new_var.type = current_variable.type;
303 new_var.region = region_type;
304 new_var.address = nullptr;
306 if (current_variable.address != nullptr)
307 new_var.address = current_variable.address;
308 else if (not current_variable.location_list.empty()) {
309 simgrid::dwarf::Location location =
310 simgrid::dwarf::resolve(
311 current_variable.location_list,
312 current_variable.object_info,
313 &(stack_frame->unw_cursor),
314 (void *) stack_frame->frame_base,
315 &mc_model_checker->process(), process_index);
317 if (not location.in_memory())
318 xbt_die("Cannot handle non-address variable");
319 new_var.address = location.address();
322 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_t> get_local_variables_values(std::vector<s_mc_stack_frame_t>& stack_frames,
336 std::vector<s_local_variable_t> 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);
342 static std::vector<s_mc_stack_frame_t> unwind_stack_frames(simgrid::mc::UnwindContext* stack_context)
344 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
345 std::vector<s_mc_stack_frame_t> result;
347 unw_cursor_t c = stack_context->cursor();
349 // TODO, check condition check (unw_init_local==0 means end of frame)
353 s_mc_stack_frame_t stack_frame;
355 stack_frame.unw_cursor = c;
360 unw_get_reg(&c, UNW_REG_IP, &ip);
361 unw_get_reg(&c, UNW_REG_SP, &sp);
366 // TODO, use real addresses in frame_t instead of fixing it here
368 simgrid::mc::Frame* frame = process->find_function(remote(ip));
369 stack_frame.frame = frame;
372 stack_frame.frame_name = frame->name;
373 stack_frame.frame_base =
374 (unw_word_t) frame->frame_base(c);
376 stack_frame.frame_base = 0;
377 stack_frame.frame_name = std::string();
380 result.push_back(std::move(stack_frame));
382 /* Stop before context switch with maestro */
383 if (frame != nullptr &&
384 frame->name == "smx_ctx_sysv_wrapper")
387 int ret = unw_step(&c);
389 xbt_die("Unexpected end of stack.");
391 xbt_die("Error while unwinding stack");
394 if (result.empty()) {
395 XBT_INFO("unw_init_local failed");
402 static std::vector<s_mc_snapshot_stack_t> take_snapshot_stacks(simgrid::mc::Snapshot* snapshot)
404 std::vector<s_mc_snapshot_stack_t> res;
406 for (auto const& stack : mc_model_checker->process().stack_areas()) {
407 s_mc_snapshot_stack_t st;
409 // Read the context from remote process:
410 unw_context_t context;
411 mc_model_checker->process().read_bytes(
412 &context, sizeof(context), remote(stack.context));
414 st.context.initialize(&mc_model_checker->process(), &context);
416 st.stack_frames = unwind_stack_frames(&st.context);
417 st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index);
418 st.process_index = stack.process_index;
420 unw_word_t sp = st.stack_frames[0].sp;
422 res.push_back(std::move(st));
425 (char*) stack.address + stack.size - (char*) sp;
426 snapshot->stack_sizes.push_back(stack_size);
433 static void snapshot_handle_ignore(simgrid::mc::Snapshot* snapshot)
435 xbt_assert(snapshot->process());
438 for (auto const& region : mc_model_checker->process().ignored_regions()) {
439 s_mc_snapshot_ignored_data_t ignored_data;
440 ignored_data.start = (void*)region.addr;
441 ignored_data.data.resize(region.size);
442 // TODO, we should do this once per privatization segment:
443 snapshot->process()->read_bytes(
444 ignored_data.data.data(), region.size, remote(region.addr),
445 simgrid::mc::ProcessIndexDisabled);
446 snapshot->ignored_data.push_back(std::move(ignored_data));
450 for (auto const& region : mc_model_checker->process().ignored_regions())
451 snapshot->process()->clear_bytes(remote(region.addr), region.size);
455 static void snapshot_ignore_restore(simgrid::mc::Snapshot* snapshot)
457 for (auto const& ignored_data : snapshot->ignored_data)
458 snapshot->process()->write_bytes(
459 ignored_data.data.data(), ignored_data.data.size(),
460 remote(ignored_data.start));
463 static std::vector<s_fd_infos_t> get_current_fds(pid_t pid)
465 const size_t fd_dir_path_size = 20;
466 char fd_dir_path[fd_dir_path_size];
467 int res = snprintf(fd_dir_path, fd_dir_path_size,
468 "/proc/%lli/fd", (long long int) pid);
469 xbt_assert(res >= 0);
470 if ((size_t) res > fd_dir_path_size)
471 xbt_die("Unexpected buffer is too small for fd_dir_path");
473 DIR* fd_dir = opendir(fd_dir_path);
474 if (fd_dir == nullptr)
475 xbt_die("Cannot open directory '/proc/self/fd'\n");
477 std::vector<s_fd_infos_t> fds;
479 struct dirent* fd_number;
480 while ((fd_number = readdir(fd_dir))) {
482 int fd_value = xbt_str_parse_int(fd_number->d_name, "Found a non-numerical FD: %s. Freaking out!");
487 const size_t source_size = 25;
489 int res = snprintf(source, source_size, "/proc/%lli/fd/%s",
490 (long long int) pid, fd_number->d_name);
491 xbt_assert(res >= 0);
492 if ((size_t) res > source_size)
493 xbt_die("Unexpected buffer is too small for fd %s", fd_number->d_name);
495 const size_t link_size = 200;
497 res = readlink(source, link, link_size);
500 xbt_die("Could not read link for %s", source);
502 xbt_die("Buffer to small for link of %s", source);
507 if(smpi_is_privatization_file(link))
511 // This is (probably) the DIR* we are reading:
512 // TODO, read all the file entries at once and close the DIR.*
513 if(strcmp(fd_dir_path, link) == 0)
516 // We don't handle them.
517 // It does not mean we should silently ignore them however.
518 if (strncmp(link, "pipe:", std::strlen("pipe:")) == 0 || strncmp(link, "socket:", std::strlen("socket:")) == 0)
521 // If dot_output enabled, do not handle the corresponding file
522 if (dot_output != nullptr) {
523 std::string link_basename = simgrid::xbt::Path(link).getBasename();
524 if (link_basename == _sg_mc_dot_output_file)
528 // This is probably a shared memory used by lttng-ust:
529 if(strncmp("/dev/shm/ust-shm-tmp-", link, std::strlen("/dev/shm/ust-shm-tmp-"))==0)
532 // Add an entry for this FD in the snapshot:
534 fd.filename = std::string(link);
535 fd.number = fd_value;
536 fd.flags = fcntl(fd_value, F_GETFL) | fcntl(fd_value, F_GETFD) ;
537 fd.current_position = lseek(fd_value, 0, SEEK_CUR);
538 fds.push_back(std::move(fd));
545 std::shared_ptr<simgrid::mc::Snapshot> take_snapshot(int num_state)
547 XBT_DEBUG("Taking snapshot %i", num_state);
549 simgrid::mc::RemoteClient* mc_process = &mc_model_checker->process();
551 std::shared_ptr<simgrid::mc::Snapshot> snapshot = std::make_shared<simgrid::mc::Snapshot>(mc_process, num_state);
553 for (auto const& p : mc_model_checker->process().actors())
554 snapshot->enabled_processes.insert(p.copy.getBuffer()->pid);
556 snapshot_handle_ignore(snapshot.get());
558 if (_sg_mc_snapshot_fds)
559 snapshot->current_fds = get_current_fds(mc_model_checker->process().pid());
561 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
562 get_memory_regions(mc_process, snapshot.get());
564 snapshot->to_ignore = mc_model_checker->process().ignored_heap();
566 if (_sg_mc_max_visited_states > 0 || not _sg_mc_property_file.empty()) {
567 snapshot->stacks = take_snapshot_stacks(snapshot.get());
569 snapshot->hash = simgrid::mc::hash(*snapshot);
575 snapshot_ignore_restore(snapshot.get());
580 void restore_snapshot_regions(simgrid::mc::Snapshot* snapshot)
582 for (std::unique_ptr<s_mc_mem_region_t> const& region : snapshot->snapshot_regions) {
583 // For privatized, variables we decided it was not necessary to take the snapshot:
585 restore(region.get());
589 if(snapshot->privatization_index >= 0) {
590 // Fix the privatization mmap:
591 s_mc_message_restore_t message{MC_MESSAGE_RESTORE, snapshot->privatization_index};
592 mc_model_checker->process().getChannel().send(message);
598 void restore_snapshot_fds(simgrid::mc::Snapshot* snapshot)
600 xbt_die("FD snapshot not implemented in client/server mode.");
602 for (auto const& fd : snapshot->current_fds) {
604 int new_fd = open(fd.filename.c_str(), fd.flags);
606 xbt_die("Could not reopen the file %s fo restoring the file descriptor", fd.filename.c_str());
607 if (new_fd != fd.number) {
608 dup2(new_fd, fd.number);
611 lseek(fd.number, fd.current_position, SEEK_SET);
615 void restore_snapshot(std::shared_ptr<simgrid::mc::Snapshot> snapshot)
617 XBT_DEBUG("Restore snapshot %i", snapshot->num_state);
618 restore_snapshot_regions(snapshot.get());
619 if (_sg_mc_snapshot_fds)
620 restore_snapshot_fds(snapshot.get());
621 snapshot_ignore_restore(snapshot.get());
622 mc_model_checker->process().clear_cache();