1 /* Copyright (c) 2008-2014. 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 "internal_config.h"
14 #include "mc_memory_map.h"
15 #include "mc_private.h"
16 #include "xbt/module.h"
17 #include <xbt/mmalloc.h>
18 #include "../smpi/private.h"
21 #include "xbt/mmalloc/mmprivate.h"
23 #include "../simix/smx_private.h"
25 #include <libunwind.h>
28 #include "mc_private.h"
31 #include "mc_snapshot.h"
32 #include "mc_object_info.h"
35 #include "mc_protocol.h"
38 using simgrid::mc::remote;
42 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc,
43 "Logging specific to mc_checkpoint");
45 /************************************ Free functions **************************************/
46 /*****************************************************************************************/
48 static void MC_snapshot_stack_free(mc_snapshot_stack_t s)
51 xbt_dynar_free(&(s->local_variables));
52 xbt_dynar_free(&(s->stack_frames));
53 mc_unw_destroy_context(s->context);
59 static void MC_snapshot_stack_free_voidp(void *s)
61 mc_snapshot_stack_t stack = (mc_snapshot_stack_t) * (void **) s;
62 MC_snapshot_stack_free(stack);
65 static void local_variable_free(local_variable_t v)
71 static void local_variable_free_voidp(void *v)
73 local_variable_free((local_variable_t) * (void **) v);
80 /** @brief Restore a region from a snapshot
82 * @param reg Target region
84 static void MC_region_restore(mc_mem_region_t region)
86 switch(region->storage_type()) {
87 case MC_REGION_STORAGE_TYPE_NONE:
89 xbt_die("Storage type not supported");
92 case MC_REGION_STORAGE_TYPE_FLAT:
93 mc_model_checker->process().write_bytes(region->flat_data().data(),
94 region->size(), region->permanent_address());
97 case MC_REGION_STORAGE_TYPE_CHUNKED:
98 mc_region_restore_sparse(&mc_model_checker->process(), region);
101 case MC_REGION_STORAGE_TYPE_PRIVATIZED:
102 for (auto& p : region->privatized_data())
103 MC_region_restore(&p);
113 simgrid::mc::RegionSnapshot privatized_region(
114 mc_region_type_t region_type, void *start_addr, void* permanent_addr, size_t size
117 size_t process_count = MC_smpi_process_count();
119 // Read smpi_privatisation_regions from MCed:
120 smpi_privatisation_region_t remote_smpi_privatisation_regions;
121 mc_model_checker->process().read_variable(
122 "smpi_privatisation_regions",
123 &remote_smpi_privatisation_regions, sizeof(remote_smpi_privatisation_regions));
124 s_smpi_privatisation_region_t privatisation_regions[process_count];
125 mc_model_checker->process().read_bytes(
126 &privatisation_regions, sizeof(privatisation_regions),
127 remote(remote_smpi_privatisation_regions));
129 std::vector<simgrid::mc::RegionSnapshot> data;
130 data.reserve(process_count);
131 for (size_t i = 0; i < process_count; i++)
133 simgrid::mc::region(region_type, start_addr,
134 privatisation_regions[i].address, size)
137 simgrid::mc::RegionSnapshot region = simgrid::mc::RegionSnapshot(
138 region_type, start_addr, permanent_addr, size);
139 region.privatized_data(std::move(data));
140 return std::move(region);
148 static void MC_snapshot_add_region(int index, mc_snapshot_t snapshot, mc_region_type_t type,
149 mc_object_info_t object_info,
150 void *start_addr, void* permanent_addr, size_t size)
152 if (type == MC_REGION_TYPE_DATA)
153 xbt_assert(object_info, "Missing object info for object.");
154 else if (type == MC_REGION_TYPE_HEAP)
155 xbt_assert(!object_info, "Unexpected object info for heap region.");
157 const bool privatization_aware = MC_object_info_is_privatized(object_info);
159 simgrid::mc::RegionSnapshot region;
160 if (privatization_aware && MC_smpi_process_count())
161 region = simgrid::mc::privatized_region(type, start_addr, permanent_addr, size);
163 region = simgrid::mc::region(type, start_addr, permanent_addr, size);
165 region.object_info(object_info);
166 snapshot->snapshot_regions[index]
167 = new simgrid::mc::RegionSnapshot(std::move(region));
171 static void MC_get_memory_regions(mc_process_t process, mc_snapshot_t snapshot)
173 const size_t n = process->object_infos_size;
174 snapshot->snapshot_regions_count = n + 1;
175 snapshot->snapshot_regions = xbt_new0(mc_mem_region_t, n + 1);
177 for (size_t i = 0; i!=n; ++i) {
178 mc_object_info_t object_info = process->object_infos[i];
179 MC_snapshot_add_region(i, snapshot, MC_REGION_TYPE_DATA, object_info,
180 object_info->start_rw, object_info->start_rw,
181 object_info->end_rw - object_info->start_rw);
184 xbt_mheap_t heap = process->get_heap();
185 void *start_heap = heap->base;
186 void *end_heap = heap->breakval;
188 MC_snapshot_add_region(n, snapshot, MC_REGION_TYPE_HEAP, NULL,
189 start_heap, start_heap,
190 (char *) end_heap - (char *) start_heap);
191 snapshot->heap_bytes_used = mmalloc_get_bytes_used_remote(
193 process->get_malloc_info());
196 if (smpi_privatize_global_variables && MC_smpi_process_count()) {
197 // snapshot->privatization_index = smpi_loaded_page
198 mc_model_checker->process().read_variable(
199 "smpi_loaded_page", &snapshot->privatization_index,
200 sizeof(snapshot->privatization_index));
204 snapshot->privatization_index = simgrid::mc::ProcessIndexMissing;
208 /** \brief Fills the position of the segments (executable, read-only, read/write).
210 * `dl_iterate_phdr` would be more robust but would not work in cross-process.
212 void MC_find_object_address(std::vector<simgrid::mc::VmMap> const& maps, mc_object_info_t result)
214 const char *name = basename(result->file_name);
215 for (size_t i = 0; i < maps.size(); ++i) {
216 simgrid::mc::VmMap const& reg = maps[i];
217 if (maps[i].pathname.empty()
218 || strcmp(basename(maps[i].pathname.c_str()), name)) {
220 } else if ((reg.prot & PROT_WRITE)) {
221 xbt_assert(!result->start_rw,
222 "Multiple read-write segments for %s, not supported",
223 maps[i].pathname.c_str());
224 result->start_rw = (char*) reg.start_addr;
225 result->end_rw = (char*) reg.end_addr;
226 // .bss is usually after the .data:
227 simgrid::mc::VmMap const& next = maps[i + 1];
228 if (next.pathname.empty() && (next.prot & PROT_WRITE)
229 && next.start_addr == reg.end_addr) {
230 result->end_rw = (char*) maps[i + 1].end_addr;
232 } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)) {
233 xbt_assert(!result->start_exec,
234 "Multiple executable segments for %s, not supported",
235 maps[i].pathname.c_str());
236 result->start_exec = (char*) reg.start_addr;
237 result->end_exec = (char*) reg.end_addr;
238 } else if ((reg.prot & PROT_READ) && !(reg.prot & PROT_EXEC)) {
239 xbt_assert(!result->start_ro,
240 "Multiple read only segments for %s, not supported",
241 maps[i].pathname.c_str());
242 result->start_ro = (char*) reg.start_addr;
243 result->end_ro = (char*) reg.end_addr;
247 result->start = result->start_rw;
248 if ((const void*) result->start_ro > result->start)
249 result->start = result->start_ro;
250 if ((const void*) result->start_exec > result->start)
251 result->start = result->start_exec;
253 result->end = result->end_rw;
254 if (result->end_ro && (const void*) result->end_ro < result->end)
255 result->end = result->end_ro;
256 if (result->end_exec && (const void*) result->end_exec > result->end)
257 result->end = result->end_exec;
259 xbt_assert(result->file_name);
260 xbt_assert(result->start_rw);
261 xbt_assert(result->start_exec);
264 /************************************* Take Snapshot ************************************/
265 /****************************************************************************************/
267 /** \brief Checks whether the variable is in scope for a given IP.
269 * A variable may be defined only from a given value of IP.
271 * \param var Variable description
272 * \param frame Scope description
273 * \param ip Instruction pointer
274 * \return true if the variable is valid
276 static bool mc_valid_variable(dw_variable_t var, dw_frame_t scope,
279 // The variable is not yet valid:
280 if ((const void *) ((const char *) scope->low_pc + var->start_scope) > ip)
286 static void mc_fill_local_variables_values(mc_stack_frame_t stack_frame,
287 dw_frame_t scope, int process_index, xbt_dynar_t result)
289 mc_process_t process = &mc_model_checker->process();
291 void *ip = (void *) stack_frame->ip;
292 if (ip < scope->low_pc || ip >= scope->high_pc)
296 dw_variable_t current_variable;
297 xbt_dynar_foreach(scope->variables, cursor, current_variable) {
299 if (!mc_valid_variable(current_variable, scope, (void *) stack_frame->ip))
303 // FIXME, get rid of `region_type`
304 if ((long) stack_frame->ip > (long) process->libsimgrid_info->start_exec)
309 local_variable_t new_var = xbt_new0(s_local_variable_t, 1);
310 new_var->subprogram = stack_frame->frame;
311 new_var->ip = stack_frame->ip;
312 new_var->name = xbt_strdup(current_variable->name);
313 new_var->type = current_variable->type;
314 new_var->region = region_type;
316 if (current_variable->address != NULL) {
317 new_var->address = current_variable->address;
318 } else if (current_variable->locations.size != 0) {
319 s_mc_location_t location;
320 mc_dwarf_resolve_locations(
321 &location, ¤t_variable->locations,
322 current_variable->object_info,
323 &(stack_frame->unw_cursor),
324 (void *) stack_frame->frame_base,
325 &mc_model_checker->process(), process_index);
327 switch(mc_get_location_type(&location)) {
328 case MC_LOCATION_TYPE_ADDRESS:
329 new_var->address = location.memory_location;
331 case MC_LOCATION_TYPE_REGISTER:
333 xbt_die("Cannot handle non-address variable");
337 xbt_die("No address");
340 xbt_dynar_push(result, &new_var);
343 // Recursive processing of nested scopes:
344 dw_frame_t nested_scope = NULL;
345 xbt_dynar_foreach(scope->scopes, cursor, nested_scope) {
346 mc_fill_local_variables_values(stack_frame, nested_scope, process_index, result);
350 static xbt_dynar_t MC_get_local_variables_values(xbt_dynar_t stack_frames, int process_index)
353 unsigned cursor1 = 0;
354 mc_stack_frame_t stack_frame;
355 xbt_dynar_t variables =
356 xbt_dynar_new(sizeof(local_variable_t), local_variable_free_voidp);
358 xbt_dynar_foreach(stack_frames, cursor1, stack_frame) {
359 mc_fill_local_variables_values(stack_frame, stack_frame->frame, process_index, variables);
365 static void MC_stack_frame_free_voipd(void *s)
367 mc_stack_frame_t stack_frame = *(mc_stack_frame_t *) s;
369 xbt_free(stack_frame->frame_name);
370 xbt_free(stack_frame);
374 static xbt_dynar_t MC_unwind_stack_frames(mc_unw_context_t stack_context)
376 mc_process_t process = &mc_model_checker->process();
378 xbt_dynar_new(sizeof(mc_stack_frame_t), MC_stack_frame_free_voipd);
382 // TODO, check condition check (unw_init_local==0 means end of frame)
383 if (mc_unw_init_cursor(&c, stack_context) != 0) {
385 xbt_die("Could not initialize stack unwinding");
390 mc_stack_frame_t stack_frame = xbt_new(s_mc_stack_frame_t, 1);
391 xbt_dynar_push(result, &stack_frame);
393 stack_frame->unw_cursor = c;
397 unw_get_reg(&c, UNW_REG_IP, &ip);
398 unw_get_reg(&c, UNW_REG_SP, &sp);
400 stack_frame->ip = ip;
401 stack_frame->sp = sp;
403 // TODO, use real addresses in frame_t instead of fixing it here
405 dw_frame_t frame = process->find_function(remote(ip));
406 stack_frame->frame = frame;
409 stack_frame->frame_name = xbt_strdup(frame->name);
410 stack_frame->frame_base =
411 (unw_word_t) mc_find_frame_base(frame, frame->object_info, &c);
413 stack_frame->frame_base = 0;
414 stack_frame->frame_name = NULL;
417 /* Stop before context switch with maestro */
418 if (frame != NULL && frame->name != NULL
419 && !strcmp(frame->name, "smx_ctx_sysv_wrapper"))
422 int ret = unw_step(&c);
424 xbt_die("Unexpected end of stack.");
425 } else if (ret < 0) {
426 xbt_die("Error while unwinding stack");
430 if (xbt_dynar_length(result) == 0) {
431 XBT_INFO("unw_init_local failed");
438 static xbt_dynar_t MC_take_snapshot_stacks(mc_snapshot_t * snapshot)
442 xbt_dynar_new(sizeof(s_mc_snapshot_stack_t),
443 MC_snapshot_stack_free_voidp);
445 unsigned int cursor = 0;
446 stack_region_t current_stack;
448 // FIXME, cross-process support (stack_areas)
449 xbt_dynar_foreach(stacks_areas, cursor, current_stack) {
450 mc_snapshot_stack_t st = xbt_new(s_mc_snapshot_stack_t, 1);
452 // Read the context from remote process:
453 unw_context_t context;
454 mc_model_checker->process().read_bytes(
455 &context, sizeof(context), remote(current_stack->context));
457 st->context = xbt_new0(s_mc_unw_context_t, 1);
458 if (mc_unw_init_context(st->context, &mc_model_checker->process(),
460 xbt_die("Could not initialise the libunwind context.");
463 st->stack_frames = MC_unwind_stack_frames(st->context);
464 st->local_variables = MC_get_local_variables_values(st->stack_frames, current_stack->process_index);
465 st->process_index = current_stack->process_index;
467 unw_word_t sp = xbt_dynar_get_as(st->stack_frames, 0, mc_stack_frame_t)->sp;
469 xbt_dynar_push(res, &st);
470 (*snapshot)->stack_sizes = (size_t*)
471 xbt_realloc((*snapshot)->stack_sizes, (cursor + 1) * sizeof(size_t));
472 (*snapshot)->stack_sizes[cursor] =
473 (char*) current_stack->address + current_stack->size - (char*) sp;
480 static xbt_dynar_t MC_take_snapshot_ignore()
483 if (mc_heap_comparison_ignore == NULL)
487 xbt_dynar_new(sizeof(mc_heap_ignore_region_t),
488 heap_ignore_region_free_voidp);
490 unsigned int cursor = 0;
491 mc_heap_ignore_region_t current_region;
493 xbt_dynar_foreach(mc_heap_comparison_ignore, cursor, current_region) {
494 mc_heap_ignore_region_t new_region = NULL;
495 new_region = xbt_new0(s_mc_heap_ignore_region_t, 1);
496 new_region->address = current_region->address;
497 new_region->size = current_region->size;
498 new_region->block = current_region->block;
499 new_region->fragment = current_region->fragment;
500 xbt_dynar_push(cpy, &new_region);
507 static void mc_free_snapshot_ignored_data_pvoid(void* data) {
508 mc_snapshot_ignored_data_t ignored_data = (mc_snapshot_ignored_data_t) data;
509 free(ignored_data->data);
512 static void MC_snapshot_handle_ignore(mc_snapshot_t snapshot)
514 xbt_assert(snapshot->process);
515 snapshot->ignored_data = xbt_dynar_new(sizeof(s_mc_snapshot_ignored_data_t), mc_free_snapshot_ignored_data_pvoid);
518 for (auto const& region : mc_model_checker->process().ignored_regions()) {
519 s_mc_snapshot_ignored_data_t ignored_data;
520 ignored_data.start = (void*)region.addr;
521 ignored_data.size = region.size;
522 ignored_data.data = malloc(region.size);
523 // TODO, we should do this once per privatization segment:
524 snapshot->process->read_bytes(
525 ignored_data.data, region.size, remote(region.addr),
526 simgrid::mc::ProcessIndexDisabled);
527 xbt_dynar_push(snapshot->ignored_data, &ignored_data);
531 for(auto const& region : mc_model_checker->process().ignored_regions()) {
532 snapshot->process->clear_bytes(remote(region.addr), region.size);
537 static void MC_snapshot_ignore_restore(mc_snapshot_t snapshot)
539 unsigned int cursor = 0;
540 s_mc_snapshot_ignored_data_t ignored_data;
541 xbt_dynar_foreach (snapshot->ignored_data, cursor, ignored_data) {
542 snapshot->process->write_bytes(ignored_data.data, ignored_data.size,
543 remote(ignored_data.start));
547 static void MC_get_current_fd(mc_snapshot_t snapshot)
549 snapshot->total_fd = 0;
551 const size_t fd_dir_path_size = 20;
552 char fd_dir_path[fd_dir_path_size];
553 int res = snprintf(fd_dir_path, fd_dir_path_size,
554 "/proc/%lli/fd", (long long int) snapshot->process->pid());
555 xbt_assert(res >= 0);
556 if ((size_t) res > fd_dir_path_size)
557 xbt_die("Unexpected buffer is too small for fd_dir_path");
559 DIR* fd_dir = opendir(fd_dir_path);
561 xbt_die("Cannot open directory '/proc/self/fd'\n");
564 struct dirent* fd_number;
565 while ((fd_number = readdir(fd_dir))) {
567 int fd_value = atoi(fd_number->d_name);
572 const size_t source_size = 25;
574 int res = snprintf(source, source_size, "/proc/%lli/fd/%s",
575 (long long int) snapshot->process->pid(), fd_number->d_name);
576 xbt_assert(res >= 0);
577 if ((size_t) res > source_size)
578 xbt_die("Unexpected buffer is too small for fd %s", fd_number->d_name);
580 const size_t link_size = 200;
582 res = readlink(source, link, link_size);
584 xbt_die("Could not read link for %s", source);
587 xbt_die("Buffer to small for link of %s", source);
591 if(smpi_is_privatisation_file(link))
594 // This is (probably) the DIR* we are reading:
595 // TODO, read all the file entries at once and close the DIR.*
596 if(strcmp(fd_dir_path, link) == 0)
599 // We don't handle them.
600 // It does not mean we should silently ignore them however.
601 if (strncmp(link, "pipe:", 5) == 0 || strncmp(link, "socket:", 7) == 0)
604 // If dot_output enabled, do not handle the corresponding file
605 if (dot_output != NULL && strcmp(basename(link), _sg_mc_dot_output_file) == 0)
608 // This is probably a shared memory used by lttng-ust:
609 if(strncmp("/dev/shm/ust-shm-tmp-", link, 21)==0)
612 // Add an entry for this FD in the snapshot:
613 fd_infos_t fd = xbt_new0(s_fd_infos_t, 1);
614 fd->filename = strdup(link);
615 fd->number = fd_value;
616 fd->flags = fcntl(fd_value, F_GETFL) | fcntl(fd_value, F_GETFD) ;
617 fd->current_position = lseek(fd_value, 0, SEEK_CUR);
618 snapshot->current_fd = (fd_infos_t*) xbt_realloc(snapshot->current_fd, (total_fd + 1) * sizeof(fd_infos_t));
619 snapshot->current_fd[total_fd] = fd;
623 snapshot->total_fd = total_fd;
627 mc_snapshot_t MC_take_snapshot(int num_state)
629 XBT_DEBUG("Taking snapshot %i", num_state);
631 mc_process_t mc_process = &mc_model_checker->process();
633 mc_snapshot_t snapshot = new simgrid::mc::Snapshot();
635 snapshot->process = mc_process;
636 snapshot->num_state = num_state;
638 snapshot->enabled_processes = xbt_dynar_new(sizeof(int), NULL);
640 smx_process_t process;
641 MC_EACH_SIMIX_PROCESS(process,
642 xbt_dynar_push_as(snapshot->enabled_processes, int, (int)process->pid));
644 MC_snapshot_handle_ignore(snapshot);
646 if (_sg_mc_snapshot_fds)
647 MC_get_current_fd(snapshot);
649 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
650 MC_get_memory_regions(mc_process, snapshot);
652 snapshot->to_ignore = MC_take_snapshot_ignore();
654 if (_sg_mc_visited > 0 || strcmp(_sg_mc_property_file, "")) {
656 MC_take_snapshot_stacks(&snapshot);
657 if (_sg_mc_hash && snapshot->stacks != NULL) {
658 snapshot->hash = mc_hash_processes_state(num_state, snapshot->stacks);
666 MC_snapshot_ignore_restore(snapshot);
671 void MC_restore_snapshot_regions(mc_snapshot_t snapshot)
673 const size_t n = snapshot->snapshot_regions_count;
674 for (size_t i = 0; i < n; i++) {
675 // For privatized, variables we decided it was not necessary to take the snapshot:
676 if (snapshot->snapshot_regions[i])
677 MC_region_restore(snapshot->snapshot_regions[i]);
681 // TODO, send a message to implement this in the MCed process
682 if(snapshot->privatization_index >= 0) {
683 // We just rewrote the global variables.
684 // The privatisation segment SMPI thinks
685 // is mapped might be inconsistent with the segment which
686 // is really mapped in memory (kernel state).
687 // We ask politely SMPI to map the segment anyway,
688 // even if it thinks it is the current one:
689 smpi_really_switch_data_segment(snapshot->privatization_index);
695 void MC_restore_snapshot_fds(mc_snapshot_t snapshot)
697 if (mc_mode == MC_MODE_SERVER)
698 xbt_die("FD snapshot not implemented in client/server mode.");
701 for (int i=0; i < snapshot->total_fd; i++) {
703 new_fd = open(snapshot->current_fd[i]->filename, snapshot->current_fd[i]->flags);
705 xbt_die("Could not reopen the file %s fo restoring the file descriptor",
706 snapshot->current_fd[i]->filename);
708 if(new_fd != -1 && new_fd != snapshot->current_fd[i]->number){
709 dup2(new_fd, snapshot->current_fd[i]->number);
710 //fprintf(stderr, "%p\n", fdopen(snapshot->current_fd[i]->number, "rw"));
713 lseek(snapshot->current_fd[i]->number, snapshot->current_fd[i]->current_position, SEEK_SET);
717 void MC_restore_snapshot(mc_snapshot_t snapshot)
719 XBT_DEBUG("Restore snapshot %i", snapshot->num_state);
720 MC_restore_snapshot_regions(snapshot);
721 if (_sg_mc_snapshot_fds)
722 MC_restore_snapshot_fds(snapshot);
723 MC_snapshot_ignore_restore(snapshot);
724 mc_model_checker->process().cache_flags = 0;
727 mc_snapshot_t simcall_HANDLER_mc_snapshot(smx_simcall_t simcall)
729 return MC_take_snapshot(1);