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 "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 "mc_private.h"
30 #include "mc_snapshot.h"
31 #include "mc_object_info.h"
34 #include "mc_protocol.h"
36 #include "mc_hash.hpp"
38 #include "src/mc/ObjectInformation.hpp"
39 #include "src/mc/Frame.hpp"
40 #include "src/mc/Variable.hpp"
42 using simgrid::mc::remote;
46 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc,
47 "Logging specific to mc_checkpoint");
49 /************************************ Free functions **************************************/
50 /*****************************************************************************************/
52 /** @brief Restore a region from a snapshot
54 * @param reg Target region
56 static void MC_region_restore(mc_mem_region_t region)
58 switch(region->storage_type()) {
59 case simgrid::mc::StorageType::NoData:
61 xbt_die("Storage type not supported");
64 case simgrid::mc::StorageType::Flat:
65 mc_model_checker->process().write_bytes(region->flat_data(),
66 region->size(), region->permanent_address());
69 case simgrid::mc::StorageType::Chunked:
70 mc_region_restore_sparse(&mc_model_checker->process(), region);
73 case simgrid::mc::StorageType::Privatized:
74 for (auto& p : region->privatized_data())
75 MC_region_restore(&p);
86 simgrid::mc::RegionSnapshot privatized_region(
87 RegionType region_type, void *start_addr, void* permanent_addr,
88 std::size_t size, const simgrid::mc::RegionSnapshot* ref_region
91 size_t process_count = MC_smpi_process_count();
93 // Read smpi_privatisation_regions from MCed:
94 smpi_privatisation_region_t remote_smpi_privatisation_regions;
95 mc_model_checker->process().read_variable(
96 "smpi_privatisation_regions",
97 &remote_smpi_privatisation_regions, sizeof(remote_smpi_privatisation_regions));
98 s_smpi_privatisation_region_t privatisation_regions[process_count];
99 mc_model_checker->process().read_bytes(
100 &privatisation_regions, sizeof(privatisation_regions),
101 remote(remote_smpi_privatisation_regions));
103 std::vector<simgrid::mc::RegionSnapshot> data;
104 data.reserve(process_count);
105 for (size_t i = 0; i < process_count; i++) {
106 const simgrid::mc::RegionSnapshot* ref_privatized_region = nullptr;
107 if (ref_region && ref_region->storage_type() == StorageType::Privatized)
108 ref_privatized_region = &ref_region->privatized_data()[i];
109 data.push_back(simgrid::mc::region(region_type, start_addr,
110 privatisation_regions[i].address, size, ref_privatized_region));
113 simgrid::mc::RegionSnapshot region = simgrid::mc::RegionSnapshot(
114 region_type, start_addr, permanent_addr, size);
115 region.privatized_data(std::move(data));
116 return std::move(region);
125 static void MC_snapshot_add_region(int index, mc_snapshot_t snapshot,
126 simgrid::mc::RegionType type,
127 simgrid::mc::ObjectInformation* object_info,
128 void *start_addr, void* permanent_addr,
131 if (type == simgrid::mc::RegionType::Data)
132 xbt_assert(object_info, "Missing object info for object.");
133 else if (type == simgrid::mc::RegionType::Heap)
134 xbt_assert(!object_info, "Unexpected object info for heap region.");
136 simgrid::mc::RegionSnapshot const* ref_region = nullptr;
137 if (mc_model_checker->parent_snapshot_)
138 ref_region = mc_model_checker->parent_snapshot_->snapshot_regions[index].get();
140 simgrid::mc::RegionSnapshot region;
142 const bool privatization_aware = object_info
143 && mc_model_checker->process().privatized(*object_info);
144 if (privatization_aware && MC_smpi_process_count())
145 region = simgrid::mc::privatized_region(
146 type, start_addr, permanent_addr, size, ref_region);
149 region = simgrid::mc::region(type, start_addr, permanent_addr, size, ref_region);
151 region.object_info(object_info);
152 snapshot->snapshot_regions[index]
153 = std::unique_ptr<simgrid::mc::RegionSnapshot>(
154 new simgrid::mc::RegionSnapshot(std::move(region)));
158 static void MC_get_memory_regions(simgrid::mc::Process* process, mc_snapshot_t snapshot)
160 const size_t n = process->object_infos.size();
161 snapshot->snapshot_regions.resize(n + 1);
163 for (auto const& object_info : process->object_infos) {
164 MC_snapshot_add_region(i, snapshot, simgrid::mc::RegionType::Data,
166 object_info->start_rw, object_info->start_rw,
167 object_info->end_rw - object_info->start_rw);
171 xbt_mheap_t heap = process->get_heap();
172 void *start_heap = heap->base;
173 void *end_heap = heap->breakval;
175 MC_snapshot_add_region(n, snapshot, simgrid::mc::RegionType::Heap, NULL,
176 start_heap, start_heap,
177 (char *) end_heap - (char *) start_heap);
178 snapshot->heap_bytes_used = mmalloc_get_bytes_used_remote(
180 process->get_malloc_info());
183 if (mc_model_checker->process().privatized() && MC_smpi_process_count()) {
184 // snapshot->privatization_index = smpi_loaded_page
185 mc_model_checker->process().read_variable(
186 "smpi_loaded_page", &snapshot->privatization_index,
187 sizeof(snapshot->privatization_index));
191 snapshot->privatization_index = simgrid::mc::ProcessIndexMissing;
195 /** \brief Fills the position of the segments (executable, read-only, read/write).
197 * `dl_iterate_phdr` would be more robust but would not work in cross-process.
199 void MC_find_object_address(
200 std::vector<simgrid::xbt::VmMap> const& maps,
201 simgrid::mc::ObjectInformation* result)
203 char* file_name = xbt_strdup(result->file_name.c_str());
204 const char *name = basename(file_name);
205 for (size_t i = 0; i < maps.size(); ++i) {
206 simgrid::xbt::VmMap const& reg = maps[i];
207 if (maps[i].pathname.empty()
208 || strcmp(basename(maps[i].pathname.c_str()), name)) {
210 } else if ((reg.prot & PROT_WRITE)) {
211 xbt_assert(!result->start_rw,
212 "Multiple read-write segments for %s, not supported",
213 maps[i].pathname.c_str());
214 result->start_rw = (char*) reg.start_addr;
215 result->end_rw = (char*) reg.end_addr;
216 // .bss is usually after the .data:
217 simgrid::xbt::VmMap const& next = maps[i + 1];
218 if (next.pathname.empty() && (next.prot & PROT_WRITE)
219 && next.start_addr == reg.end_addr) {
220 result->end_rw = (char*) maps[i + 1].end_addr;
222 } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)) {
223 xbt_assert(!result->start_exec,
224 "Multiple executable segments for %s, not supported",
225 maps[i].pathname.c_str());
226 result->start_exec = (char*) reg.start_addr;
227 result->end_exec = (char*) reg.end_addr;
228 } else if ((reg.prot & PROT_READ) && !(reg.prot & PROT_EXEC)) {
229 xbt_assert(!result->start_ro,
230 "Multiple read only segments for %s, not supported",
231 maps[i].pathname.c_str());
232 result->start_ro = (char*) reg.start_addr;
233 result->end_ro = (char*) reg.end_addr;
237 result->start = result->start_rw;
238 if ((const void*) result->start_ro > result->start)
239 result->start = result->start_ro;
240 if ((const void*) result->start_exec > result->start)
241 result->start = result->start_exec;
243 result->end = result->end_rw;
244 if (result->end_ro && (const void*) result->end_ro < result->end)
245 result->end = result->end_ro;
246 if (result->end_exec && (const void*) result->end_exec > result->end)
247 result->end = result->end_exec;
249 xbt_assert(result->start_rw);
250 xbt_assert(result->start_exec);
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 frame Scope description
263 * \param ip Instruction pointer
264 * \return true if the variable is valid
266 static bool mc_valid_variable(simgrid::mc::Variable* var,
267 simgrid::mc::Frame* scope,
270 // The variable is not yet valid:
271 if ((const void *) ((const char *) scope->low_pc + var->start_scope) > ip)
277 static void mc_fill_local_variables_values(mc_stack_frame_t stack_frame,
278 simgrid::mc::Frame* scope,
280 std::vector<s_local_variable>& result)
282 simgrid::mc::Process* process = &mc_model_checker->process();
284 void *ip = (void *) stack_frame->ip;
285 if (ip < scope->low_pc || ip >= scope->high_pc)
288 for(simgrid::mc::Variable& current_variable :
291 if (!mc_valid_variable(¤t_variable, scope, (void *) stack_frame->ip))
295 // FIXME, get rid of `region_type`
296 if ((long) stack_frame->ip > (long) process->libsimgrid_info->start_exec)
301 s_local_variable_t new_var;
302 new_var.subprogram = stack_frame->frame;
303 new_var.ip = stack_frame->ip;
304 new_var.name = current_variable.name;
305 new_var.type = current_variable.type;
306 new_var.region = region_type;
307 new_var.address = nullptr;
309 if (current_variable.address != NULL) {
310 new_var.address = current_variable.address;
311 } else if (!current_variable.location_list.empty()) {
312 simgrid::dwarf::Location location =
313 simgrid::dwarf::resolve(
314 current_variable.location_list,
315 current_variable.object_info,
316 &(stack_frame->unw_cursor),
317 (void *) stack_frame->frame_base,
318 &mc_model_checker->process(), process_index);
320 if (!location.in_memory())
321 xbt_die("Cannot handle non-address variable");
322 new_var.address = location.address();
325 xbt_die("No address");
328 result.push_back(std::move(new_var));
331 // Recursive processing of nested scopes:
332 for(simgrid::mc::Frame& nested_scope : scope->scopes)
333 mc_fill_local_variables_values(
334 stack_frame, &nested_scope, process_index, result);
337 static std::vector<s_local_variable> MC_get_local_variables_values(
338 std::vector<s_mc_stack_frame_t>& stack_frames, int process_index)
340 std::vector<s_local_variable> variables;
341 for (s_mc_stack_frame_t& stack_frame : stack_frames)
342 mc_fill_local_variables_values(&stack_frame, stack_frame.frame, process_index, variables);
343 return std::move(variables);
346 static void MC_stack_frame_free_voipd(void *s)
348 mc_stack_frame_t stack_frame = *(mc_stack_frame_t *) s;
352 static std::vector<s_mc_stack_frame_t> MC_unwind_stack_frames(mc_unw_context_t stack_context)
354 simgrid::mc::Process* process = &mc_model_checker->process();
355 std::vector<s_mc_stack_frame_t> result;
359 // TODO, check condition check (unw_init_local==0 means end of frame)
360 if (mc_unw_init_cursor(&c, stack_context) != 0) {
362 xbt_die("Could not initialize stack unwinding");
367 s_mc_stack_frame_t stack_frame;
369 stack_frame.unw_cursor = c;
373 unw_get_reg(&c, UNW_REG_IP, &ip);
374 unw_get_reg(&c, UNW_REG_SP, &sp);
379 // TODO, use real addresses in frame_t instead of fixing it here
381 simgrid::mc::Frame* frame = process->find_function(remote(ip));
382 stack_frame.frame = frame;
385 stack_frame.frame_name = frame->name;
386 stack_frame.frame_base =
387 (unw_word_t) frame->frame_base(c);
389 stack_frame.frame_base = 0;
390 stack_frame.frame_name = std::string();
393 result.push_back(std::move(stack_frame));
395 /* Stop before context switch with maestro */
396 if (frame != nullptr &&
397 frame->name == "smx_ctx_sysv_wrapper")
400 int ret = unw_step(&c);
402 xbt_die("Unexpected end of stack.");
403 } else if (ret < 0) {
404 xbt_die("Error while unwinding stack");
408 if (result.empty()) {
409 XBT_INFO("unw_init_local failed");
413 return std::move(result);
416 static std::vector<s_mc_snapshot_stack_t> MC_take_snapshot_stacks(mc_snapshot_t * snapshot)
418 std::vector<s_mc_snapshot_stack_t> res;
420 for (auto const& stack : mc_model_checker->process().stack_areas()) {
421 s_mc_snapshot_stack_t st;
423 // Read the context from remote process:
424 unw_context_t context;
425 mc_model_checker->process().read_bytes(
426 &context, sizeof(context), remote(stack.context));
428 if (mc_unw_init_context(&st.context, &mc_model_checker->process(),
430 xbt_die("Could not initialise the libunwind context.");
432 st.stack_frames = MC_unwind_stack_frames(&st.context);
433 st.local_variables = MC_get_local_variables_values(st.stack_frames, stack.process_index);
434 st.process_index = stack.process_index;
436 unw_word_t sp = st.stack_frames[0].sp;
438 res.push_back(std::move(st));
441 (char*) stack.address + stack.size - (char*) sp;
442 (*snapshot)->stack_sizes.push_back(stack_size);
445 return std::move(res);
449 static void MC_snapshot_handle_ignore(mc_snapshot_t snapshot)
451 xbt_assert(snapshot->process());
454 for (auto const& region : mc_model_checker->process().ignored_regions()) {
455 s_mc_snapshot_ignored_data_t ignored_data;
456 ignored_data.start = (void*)region.addr;
457 ignored_data.data.resize(region.size);
458 // TODO, we should do this once per privatization segment:
459 snapshot->process()->read_bytes(
460 ignored_data.data.data(), region.size, remote(region.addr),
461 simgrid::mc::ProcessIndexDisabled);
462 snapshot->ignored_data.push_back(std::move(ignored_data));
466 for(auto const& region : mc_model_checker->process().ignored_regions()) {
467 snapshot->process()->clear_bytes(remote(region.addr), region.size);
472 static void MC_snapshot_ignore_restore(mc_snapshot_t snapshot)
474 for (auto const& ignored_data : snapshot->ignored_data)
475 snapshot->process()->write_bytes(
476 ignored_data.data.data(), ignored_data.data.size(),
477 remote(ignored_data.start));
480 static std::vector<s_fd_infos_t> MC_get_current_fds(pid_t pid)
482 const size_t fd_dir_path_size = 20;
483 char fd_dir_path[fd_dir_path_size];
484 int res = snprintf(fd_dir_path, fd_dir_path_size,
485 "/proc/%lli/fd", (long long int) pid);
486 xbt_assert(res >= 0);
487 if ((size_t) res > fd_dir_path_size)
488 xbt_die("Unexpected buffer is too small for fd_dir_path");
490 DIR* fd_dir = opendir(fd_dir_path);
492 xbt_die("Cannot open directory '/proc/self/fd'\n");
494 std::vector<s_fd_infos_t> fds;
496 struct dirent* fd_number;
497 while ((fd_number = readdir(fd_dir))) {
499 int fd_value = atoi(fd_number->d_name);
504 const size_t source_size = 25;
506 int res = snprintf(source, source_size, "/proc/%lli/fd/%s",
507 (long long int) pid, fd_number->d_name);
508 xbt_assert(res >= 0);
509 if ((size_t) res > source_size)
510 xbt_die("Unexpected buffer is too small for fd %s", fd_number->d_name);
512 const size_t link_size = 200;
514 res = readlink(source, link, link_size);
516 xbt_die("Could not read link for %s", source);
519 xbt_die("Buffer to small for link of %s", source);
524 if(smpi_is_privatisation_file(link))
528 // This is (probably) the DIR* we are reading:
529 // TODO, read all the file entries at once and close the DIR.*
530 if(strcmp(fd_dir_path, link) == 0)
533 // We don't handle them.
534 // It does not mean we should silently ignore them however.
535 if (strncmp(link, "pipe:", 5) == 0 || strncmp(link, "socket:", 7) == 0)
538 // If dot_output enabled, do not handle the corresponding file
539 if (dot_output != NULL && strcmp(basename(link), _sg_mc_dot_output_file) == 0)
542 // This is probably a shared memory used by lttng-ust:
543 if(strncmp("/dev/shm/ust-shm-tmp-", link, 21)==0)
546 // Add an entry for this FD in the snapshot:
548 fd.filename = std::string(link);
549 fd.number = fd_value;
550 fd.flags = fcntl(fd_value, F_GETFL) | fcntl(fd_value, F_GETFD) ;
551 fd.current_position = lseek(fd_value, 0, SEEK_CUR);
552 fds.push_back(std::move(fd));
556 return std::move(fds);
559 mc_snapshot_t MC_take_snapshot(int num_state)
561 XBT_DEBUG("Taking snapshot %i", num_state);
563 simgrid::mc::Process* mc_process = &mc_model_checker->process();
565 mc_snapshot_t snapshot = new simgrid::mc::Snapshot(mc_process);
567 snapshot->num_state = num_state;
569 smx_process_t process;
570 MC_EACH_SIMIX_PROCESS(process,
571 snapshot->enabled_processes.insert(process->pid));
573 MC_snapshot_handle_ignore(snapshot);
575 if (_sg_mc_snapshot_fds)
576 snapshot->current_fds = MC_get_current_fds(process->pid);
578 const bool use_soft_dirty = _sg_mc_sparse_checkpoint && _sg_mc_soft_dirty;
580 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
581 MC_get_memory_regions(mc_process, snapshot);
583 mc_process->reset_soft_dirty();
585 snapshot->to_ignore = mc_model_checker->process().ignored_heap();
587 if (_sg_mc_visited > 0 || strcmp(_sg_mc_property_file, "")) {
589 MC_take_snapshot_stacks(&snapshot);
591 snapshot->hash = simgrid::mc::hash(*snapshot);
599 MC_snapshot_ignore_restore(snapshot);
601 mc_model_checker->parent_snapshot_ = snapshot;
606 void MC_restore_snapshot_regions(mc_snapshot_t snapshot)
608 for(std::unique_ptr<s_mc_mem_region_t> const& region : snapshot->snapshot_regions) {
609 // For privatized, variables we decided it was not necessary to take the snapshot:
611 MC_region_restore(region.get());
615 // TODO, send a message to implement this in the MCed process
616 if(snapshot->privatization_index >= 0) {
617 // Fix the privatization mmap:
618 s_mc_restore_message message;
619 message.type = MC_MESSAGE_RESTORE;
620 message.index = snapshot->privatization_index;
621 mc_model_checker->process().send_message(message);
627 void MC_restore_snapshot_fds(mc_snapshot_t snapshot)
629 if (mc_mode == MC_MODE_SERVER)
630 xbt_die("FD snapshot not implemented in client/server mode.");
632 for (auto const& fd : snapshot->current_fds) {
634 int new_fd = open(fd.filename.c_str(), fd.flags);
636 xbt_die("Could not reopen the file %s fo restoring the file descriptor",
637 fd.filename.c_str());
639 if (new_fd != fd.number) {
640 dup2(new_fd, fd.number);
643 lseek(fd.number, fd.current_position, SEEK_SET);
647 void MC_restore_snapshot(mc_snapshot_t snapshot)
649 XBT_DEBUG("Restore snapshot %i", snapshot->num_state);
650 const bool use_soft_dirty = _sg_mc_sparse_checkpoint && _sg_mc_soft_dirty;
651 MC_restore_snapshot_regions(snapshot);
652 if (_sg_mc_snapshot_fds)
653 MC_restore_snapshot_fds(snapshot);
655 mc_model_checker->process().reset_soft_dirty();
656 MC_snapshot_ignore_restore(snapshot);
657 mc_model_checker->process().cache_flags = 0;
659 mc_model_checker->parent_snapshot_ = snapshot;
662 mc_snapshot_t simcall_HANDLER_mc_snapshot(smx_simcall_t simcall)
664 return MC_take_snapshot(1);