1 /* Copyright (c) 2008-2017. 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_hash.hpp"
36 #include "src/mc/mc_mmu.hpp"
37 #include "src/mc/mc_smx.hpp"
38 #include "src/mc/mc_snapshot.hpp"
39 #include "src/mc/mc_unw.hpp"
40 #include "src/mc/remote/mc_protocol.h"
42 #include "src/mc/RegionSnapshot.hpp"
43 #include "src/mc/ObjectInformation.hpp"
44 #include "src/mc/Frame.hpp"
45 #include "src/mc/Variable.hpp"
47 using simgrid::mc::remote;
49 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc, "Logging specific to mc_checkpoint");
51 #define PROT_RWX (PROT_READ | PROT_WRITE | PROT_EXEC)
52 #define PROT_RW (PROT_READ | PROT_WRITE)
53 #define PROT_RX (PROT_READ | PROT_EXEC)
54 extern std::string _sg_mc_property_file;
55 extern std::string _sg_mc_dot_output_file;
60 /************************************ Free functions **************************************/
61 /*****************************************************************************************/
63 /** @brief Restore a region from a snapshot
65 * @param region Target region
67 static void restore(mc_mem_region_t region)
69 switch(region->storage_type()) {
70 case simgrid::mc::StorageType::Flat:
71 mc_model_checker->process().write_bytes(region->flat_data().get(),
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())
84 default: // includes StorageType::NoData
85 xbt_die("Storage type not supported");
91 RegionSnapshot privatized_region(
92 RegionType region_type, void *start_addr, void* permanent_addr,
96 size_t process_count = MC_smpi_process_count();
98 // Read smpi_privatization_regions from MCed:
99 smpi_privatization_region_t remote_smpi_privatization_regions;
100 mc_model_checker->process().read_variable(
101 "smpi_privatization_regions",
102 &remote_smpi_privatization_regions, sizeof(remote_smpi_privatization_regions));
103 s_smpi_privatization_region_t privatization_regions[process_count];
104 mc_model_checker->process().read_bytes(
105 &privatization_regions, sizeof(privatization_regions),
106 remote(remote_smpi_privatization_regions));
108 std::vector<simgrid::mc::RegionSnapshot> data;
109 data.reserve(process_count);
110 for (size_t i = 0; i < process_count; i++)
111 data.push_back(simgrid::mc::region(region_type, start_addr,
112 privatization_regions[i].address, size));
114 simgrid::mc::RegionSnapshot region = simgrid::mc::RegionSnapshot(
115 region_type, start_addr, permanent_addr, size);
116 region.privatized_data(std::move(data));
122 void add_region(int index, simgrid::mc::Snapshot* 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(not object_info, "Unexpected object info for heap region.");
133 simgrid::mc::RegionSnapshot region;
135 const bool privatization_aware = object_info
136 && mc_model_checker->process().privatized(*object_info);
137 if (privatization_aware && MC_smpi_process_count())
138 region = simgrid::mc::privatized_region(
139 type, start_addr, permanent_addr, size);
142 region = simgrid::mc::region(type, start_addr, permanent_addr, size);
144 region.object_info(object_info);
145 snapshot->snapshot_regions[index]
146 = std::unique_ptr<simgrid::mc::RegionSnapshot>(
147 new simgrid::mc::RegionSnapshot(std::move(region)));
150 static void get_memory_regions(simgrid::mc::RemoteClient* process, simgrid::mc::Snapshot* snapshot)
152 const size_t n = process->object_infos.size();
153 snapshot->snapshot_regions.resize(n + 1);
155 for (auto const& object_info : process->object_infos)
156 add_region(i++, snapshot, simgrid::mc::RegionType::Data,
158 object_info->start_rw, object_info->start_rw,
159 object_info->end_rw - object_info->start_rw);
161 xbt_mheap_t heap = process->get_heap();
162 void *start_heap = heap->base;
163 void *end_heap = heap->breakval;
165 add_region(n, snapshot, simgrid::mc::RegionType::Heap, nullptr,
166 start_heap, start_heap,
167 (char *) end_heap - (char *) start_heap);
168 snapshot->heap_bytes_used = mmalloc_get_bytes_used_remote(
170 process->get_malloc_info());
173 if (mc_model_checker->process().privatized() && MC_smpi_process_count())
174 // snapshot->privatization_index = smpi_loaded_page
175 mc_model_checker->process().read_variable(
176 "smpi_loaded_page", &snapshot->privatization_index,
177 sizeof(snapshot->privatization_index));
180 snapshot->privatization_index = simgrid::mc::ProcessIndexMissing;
183 /** \brief Fills the position of the segments (executable, read-only, read/write).
185 // TODO, use the ELF segment information for more robustness
186 void find_object_address(
187 std::vector<simgrid::xbt::VmMap> const& maps,
188 simgrid::mc::ObjectInformation* result)
190 std::string name = simgrid::xbt::Path(result->file_name).getBasename();
192 for (size_t i = 0; i < maps.size(); ++i) {
193 simgrid::xbt::VmMap const& reg = maps[i];
194 if (maps[i].pathname.empty())
196 std::string map_basename = simgrid::xbt::Path(maps[i].pathname).getBasename();
197 if (map_basename != name)
200 // This is the non-GNU_RELRO-part of the data segment:
201 if (reg.prot == PROT_RW) {
202 xbt_assert(not result->start_rw, "Multiple read-write segments for %s, not supported", maps[i].pathname.c_str());
203 result->start_rw = (char*) reg.start_addr;
204 result->end_rw = (char*) reg.end_addr;
206 // The next VMA might be end of the data segment:
207 if (i + 1 < maps.size()
208 && maps[i + 1].pathname.empty()
209 && maps[i + 1].prot == PROT_RW
210 && maps[i + 1].start_addr == reg.end_addr)
211 result->end_rw = (char*) maps[i + 1].end_addr;
214 // This is the text segment:
215 else if (reg.prot == PROT_RX) {
216 xbt_assert(not result->start_exec, "Multiple executable segments for %s, not supported",
217 maps[i].pathname.c_str());
218 result->start_exec = (char*) reg.start_addr;
219 result->end_exec = (char*) reg.end_addr;
221 // The next VMA might be end of the data segment:
222 if (i + 1 < maps.size()
223 && maps[i + 1].pathname.empty()
224 && maps[i + 1].prot == PROT_RW
225 && maps[i + 1].start_addr == reg.end_addr) {
226 result->start_rw = (char*) maps[i + 1].start_addr;
227 result->end_rw = (char*) maps[i + 1].end_addr;
231 // This is the GNU_RELRO-part of the data segment:
232 else if (reg.prot == PROT_READ) {
233 xbt_assert(not result->start_ro, "Multiple read only segments for %s, not supported", maps[i].pathname.c_str());
234 result->start_ro = (char*) reg.start_addr;
235 result->end_ro = (char*) reg.end_addr;
239 result->start = result->start_rw;
240 if ((const void*) result->start_ro < result->start)
241 result->start = result->start_ro;
242 if ((const void*) result->start_exec < result->start)
243 result->start = result->start_exec;
245 result->end = result->end_rw;
246 if (result->end_ro && (const void*) result->end_ro > result->end)
247 result->end = result->end_ro;
248 if (result->end_exec && (const void*) result->end_exec > result->end)
249 result->end = result->end_exec;
251 xbt_assert(result->start_exec || result->start_rw || result->start_ro);
254 /************************************* Take Snapshot ************************************/
255 /****************************************************************************************/
257 /** \brief Checks whether the variable is in scope for a given IP.
259 * A variable may be defined only from a given value of IP.
261 * \param var Variable description
262 * \param scope Scope description
263 * \param ip Instruction pointer
264 * \return true if the variable is valid
266 static bool valid_variable(simgrid::mc::Variable* var,
267 simgrid::mc::Frame* scope,
270 // The variable is not yet valid:
271 if (scope->range.begin() + var->start_scope > (std::uint64_t) ip)
277 static void fill_local_variables_values(mc_stack_frame_t stack_frame, simgrid::mc::Frame* scope, int process_index,
278 std::vector<s_local_variable_t>& result)
280 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
282 if (not scope || not scope->range.contain(stack_frame->ip))
285 for (simgrid::mc::Variable& current_variable : scope->variables) {
287 if (not 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 != nullptr)
306 new_var.address = current_variable.address;
307 else if (not 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 (not location.in_memory())
317 xbt_die("Cannot handle non-address variable");
318 new_var.address = location.address();
321 xbt_die("No address");
323 result.push_back(std::move(new_var));
326 // Recursive processing of nested scopes:
327 for (simgrid::mc::Frame& nested_scope : scope->scopes)
328 fill_local_variables_values(
329 stack_frame, &nested_scope, process_index, result);
332 static std::vector<s_local_variable_t> get_local_variables_values(std::vector<s_mc_stack_frame_t>& stack_frames,
335 std::vector<s_local_variable_t> variables;
336 for (s_mc_stack_frame_t& stack_frame : stack_frames)
337 fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables);
341 static std::vector<s_mc_stack_frame_t> unwind_stack_frames(simgrid::mc::UnwindContext* stack_context)
343 simgrid::mc::RemoteClient* process = &mc_model_checker->process();
344 std::vector<s_mc_stack_frame_t> result;
346 unw_cursor_t c = stack_context->cursor();
348 // TODO, check condition check (unw_init_local==0 means end of frame)
352 s_mc_stack_frame_t stack_frame;
354 stack_frame.unw_cursor = c;
359 unw_get_reg(&c, UNW_REG_IP, &ip);
360 unw_get_reg(&c, UNW_REG_SP, &sp);
365 // TODO, use real addresses in frame_t instead of fixing it here
367 simgrid::mc::Frame* frame = process->find_function(remote(ip));
368 stack_frame.frame = frame;
371 stack_frame.frame_name = frame->name;
372 stack_frame.frame_base =
373 (unw_word_t) frame->frame_base(c);
375 stack_frame.frame_base = 0;
376 stack_frame.frame_name = std::string();
379 result.push_back(std::move(stack_frame));
381 /* Stop before context switch with maestro */
382 if (frame != nullptr &&
383 frame->name == "smx_ctx_sysv_wrapper")
386 int ret = unw_step(&c);
388 xbt_die("Unexpected end of stack.");
390 xbt_die("Error while unwinding stack");
393 if (result.empty()) {
394 XBT_INFO("unw_init_local failed");
401 static std::vector<s_mc_snapshot_stack_t> take_snapshot_stacks(simgrid::mc::Snapshot* snapshot)
403 std::vector<s_mc_snapshot_stack_t> res;
405 for (auto const& stack : mc_model_checker->process().stack_areas()) {
406 s_mc_snapshot_stack_t st;
408 // Read the context from remote process:
409 unw_context_t context;
410 mc_model_checker->process().read_bytes(
411 &context, sizeof(context), remote(stack.context));
413 st.context.initialize(&mc_model_checker->process(), &context);
415 st.stack_frames = unwind_stack_frames(&st.context);
416 st.local_variables = get_local_variables_values(st.stack_frames, stack.process_index);
417 st.process_index = stack.process_index;
419 unw_word_t sp = st.stack_frames[0].sp;
421 res.push_back(std::move(st));
424 (char*) stack.address + stack.size - (char*) sp;
425 snapshot->stack_sizes.push_back(stack_size);
432 static void snapshot_handle_ignore(simgrid::mc::Snapshot* snapshot)
434 xbt_assert(snapshot->process());
437 for (auto const& region : mc_model_checker->process().ignored_regions()) {
438 s_mc_snapshot_ignored_data_t ignored_data;
439 ignored_data.start = (void*)region.addr;
440 ignored_data.data.resize(region.size);
441 // TODO, we should do this once per privatization segment:
442 snapshot->process()->read_bytes(
443 ignored_data.data.data(), region.size, remote(region.addr),
444 simgrid::mc::ProcessIndexDisabled);
445 snapshot->ignored_data.push_back(std::move(ignored_data));
449 for (auto const& region : mc_model_checker->process().ignored_regions())
450 snapshot->process()->clear_bytes(remote(region.addr), region.size);
454 static void snapshot_ignore_restore(simgrid::mc::Snapshot* snapshot)
456 for (auto const& ignored_data : snapshot->ignored_data)
457 snapshot->process()->write_bytes(
458 ignored_data.data.data(), ignored_data.data.size(),
459 remote(ignored_data.start));
462 static std::vector<s_fd_infos_t> get_current_fds(pid_t pid)
464 const size_t fd_dir_path_size = 20;
465 char fd_dir_path[fd_dir_path_size];
466 int res = snprintf(fd_dir_path, fd_dir_path_size,
467 "/proc/%lli/fd", (long long int) pid);
468 xbt_assert(res >= 0);
469 if ((size_t) res > fd_dir_path_size)
470 xbt_die("Unexpected buffer is too small for fd_dir_path");
472 DIR* fd_dir = opendir(fd_dir_path);
473 if (fd_dir == nullptr)
474 xbt_die("Cannot open directory '/proc/self/fd'\n");
476 std::vector<s_fd_infos_t> fds;
478 struct dirent* fd_number;
479 while ((fd_number = readdir(fd_dir))) {
481 int fd_value = xbt_str_parse_int(fd_number->d_name, "Found a non-numerical FD: %s. Freaking out!");
486 const size_t source_size = 25;
488 int res = snprintf(source, source_size, "/proc/%lli/fd/%s",
489 (long long int) pid, fd_number->d_name);
490 xbt_assert(res >= 0);
491 if ((size_t) res > source_size)
492 xbt_die("Unexpected buffer is too small for fd %s", fd_number->d_name);
494 const size_t link_size = 200;
496 res = readlink(source, link, link_size);
499 xbt_die("Could not read link for %s", source);
501 xbt_die("Buffer to small for link of %s", source);
506 if(smpi_is_privatization_file(link))
510 // This is (probably) the DIR* we are reading:
511 // TODO, read all the file entries at once and close the DIR.*
512 if(strcmp(fd_dir_path, link) == 0)
515 // We don't handle them.
516 // It does not mean we should silently ignore them however.
517 if (strncmp(link, "pipe:", std::strlen("pipe:")) == 0 || strncmp(link, "socket:", std::strlen("socket:")) == 0)
520 // If dot_output enabled, do not handle the corresponding file
521 if (dot_output != nullptr) {
522 std::string link_basename = simgrid::xbt::Path(link).getBasename();
523 if (link_basename == _sg_mc_dot_output_file)
527 // This is probably a shared memory used by lttng-ust:
528 if(strncmp("/dev/shm/ust-shm-tmp-", link, std::strlen("/dev/shm/ust-shm-tmp-"))==0)
531 // Add an entry for this FD in the snapshot:
533 fd.filename = std::string(link);
534 fd.number = fd_value;
535 fd.flags = fcntl(fd_value, F_GETFL) | fcntl(fd_value, F_GETFD) ;
536 fd.current_position = lseek(fd_value, 0, SEEK_CUR);
537 fds.push_back(std::move(fd));
544 std::shared_ptr<simgrid::mc::Snapshot> take_snapshot(int num_state)
546 XBT_DEBUG("Taking snapshot %i", num_state);
548 simgrid::mc::RemoteClient* mc_process = &mc_model_checker->process();
550 std::shared_ptr<simgrid::mc::Snapshot> snapshot = std::make_shared<simgrid::mc::Snapshot>(mc_process, num_state);
552 for (auto const& p : mc_model_checker->process().actors())
553 snapshot->enabled_processes.insert(p.copy.getBuffer()->pid);
555 snapshot_handle_ignore(snapshot.get());
557 if (_sg_mc_snapshot_fds)
558 snapshot->current_fds = get_current_fds(mc_model_checker->process().pid());
560 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
561 get_memory_regions(mc_process, snapshot.get());
563 snapshot->to_ignore = mc_model_checker->process().ignored_heap();
565 if (_sg_mc_max_visited_states > 0 || not _sg_mc_property_file.empty()) {
566 snapshot->stacks = take_snapshot_stacks(snapshot.get());
568 snapshot->hash = simgrid::mc::hash(*snapshot);
574 snapshot_ignore_restore(snapshot.get());
579 void restore_snapshot_regions(simgrid::mc::Snapshot* snapshot)
581 for (std::unique_ptr<s_mc_mem_region_t> const& region : snapshot->snapshot_regions) {
582 // For privatized, variables we decided it was not necessary to take the snapshot:
584 restore(region.get());
588 if(snapshot->privatization_index >= 0) {
589 // Fix the privatization mmap:
590 s_mc_message_restore_t message{MC_MESSAGE_RESTORE, snapshot->privatization_index};
591 mc_model_checker->process().getChannel().send(message);
597 void restore_snapshot_fds(simgrid::mc::Snapshot* snapshot)
599 xbt_die("FD snapshot not implemented in client/server mode.");
601 for (auto const& fd : snapshot->current_fds) {
603 int new_fd = open(fd.filename.c_str(), fd.flags);
605 xbt_die("Could not reopen the file %s fo restoring the file descriptor", fd.filename.c_str());
606 if (new_fd != fd.number) {
607 dup2(new_fd, fd.number);
610 lseek(fd.number, fd.current_position, SEEK_SET);
614 void restore_snapshot(std::shared_ptr<simgrid::mc::Snapshot> snapshot)
616 XBT_DEBUG("Restore snapshot %i", snapshot->num_state);
617 restore_snapshot_regions(snapshot.get());
618 if (_sg_mc_snapshot_fds)
619 restore_snapshot_fds(snapshot.get());
620 snapshot_ignore_restore(snapshot.get());
621 mc_model_checker->process().clear_cache();