1 /* Copyright (c) 2014-2021. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 #define _FILE_OFFSET_BITS 64 /* needed for pread_whole to work as expected on 32bits */
8 #include "src/mc/remote/RemoteProcess.hpp"
10 #include "src/mc/sosp/Snapshot.hpp"
11 #include "xbt/file.hpp"
15 #include <libunwind-ptrace.h>
16 #include <sys/mman.h> // PROT_*
25 using simgrid::mc::remote;
27 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_process, mc, "MC process information");
34 // List of library which memory segments are not considered:
35 static const std::vector<std::string> filtered_libraries = {
39 #elif defined __FreeBSD__
42 "libkvm", /* kernel data access library */
43 "libprocstat", /* process and file information retrieval */
44 "libthr", /* thread library */
47 "libargp", /* workarounds for glibc-less systems */
48 "libasan", /* gcc sanitizers */
52 "libboost_context-mt",
53 "libboost_stacktrace_addr2line",
54 "libboost_stacktrace_backtrace",
100 "libkrb5support", /*odd behaviour on fedora rawhide ... remove these when fixed*/
133 "libtsan", /* gcc sanitizers */
134 "libubsan", /* gcc sanitizers */
144 static bool is_filtered_lib(const std::string& libname)
146 return std::find(begin(filtered_libraries), end(filtered_libraries), libname) != end(filtered_libraries);
149 static std::string get_lib_name(const std::string& pathname)
151 std::string map_basename = simgrid::xbt::Path(pathname).get_base_name();
154 size_t pos = map_basename.rfind(".so");
155 if (pos != std::string::npos) {
156 // strip the extension (matching regex "\.so.*$")
157 libname.assign(map_basename, 0, pos);
159 // strip the version suffix (matching regex "-[.0-9-]*$")
161 pos = libname.rfind('-');
162 if (pos == std::string::npos || libname.find_first_not_of(".0123456789", pos + 1) != std::string::npos)
171 static ssize_t pread_whole(int fd, void* buf, size_t count, off_t offset)
173 auto* buffer = static_cast<char*>(buf);
174 ssize_t real_count = count;
176 ssize_t res = pread(fd, buffer, count, offset);
183 else if (errno != EINTR) {
184 XBT_ERROR("pread_whole: %s", strerror(errno));
191 static ssize_t pwrite_whole(int fd, const void* buf, size_t count, off_t offset)
193 const auto* buffer = static_cast<const char*>(buf);
194 ssize_t real_count = count;
196 ssize_t res = pwrite(fd, buffer, count, offset);
203 else if (errno != EINTR) {
204 XBT_ERROR("pwrite_whole: %s", strerror(errno));
211 int open_vm(pid_t pid, int flags)
213 std::string buffer = "/proc/" + std::to_string(pid) + "/mem";
214 return open(buffer.c_str(), flags);
217 // ***** RemoteProcess
219 RemoteProcess::RemoteProcess(pid_t pid) : AddressSpace(this), pid_(pid), running_(true) {}
221 void RemoteProcess::init(xbt_mheap_t mmalloc_default_mdp, unsigned long* maxpid, xbt_dynar_t actors,
222 xbt_dynar_t dead_actors)
224 this->heap_address = remote(mmalloc_default_mdp);
225 this->maxpid_addr_ = remote(maxpid);
226 this->actors_addr_ = remote(actors);
227 this->dead_actors_addr_ = remote(dead_actors);
229 this->memory_map_ = simgrid::xbt::get_memory_map(this->pid_);
230 this->init_memory_map_info();
232 int fd = open_vm(this->pid_, O_RDWR);
233 xbt_assert(fd >= 0, "Could not open file for process virtual address space");
234 this->memory_file = fd;
236 this->smx_actors_infos.clear();
237 this->smx_dead_actors_infos.clear();
238 this->unw_addr_space = simgrid::mc::UnwindContext::createUnwindAddressSpace();
239 this->unw_underlying_addr_space = simgrid::unw::create_addr_space();
240 this->unw_underlying_context = simgrid::unw::create_context(this->unw_underlying_addr_space, this->pid_);
243 RemoteProcess::~RemoteProcess()
245 if (this->memory_file >= 0)
246 close(this->memory_file);
248 if (this->unw_underlying_addr_space != unw_local_addr_space) {
249 if (this->unw_underlying_addr_space)
250 unw_destroy_addr_space(this->unw_underlying_addr_space);
251 if (this->unw_underlying_context)
252 _UPT_destroy(this->unw_underlying_context);
255 unw_destroy_addr_space(this->unw_addr_space);
258 /** Refresh the information about the process
260 * Do not use directly, this is used by the getters when appropriate
261 * in order to have fresh data.
263 void RemoteProcess::refresh_heap()
265 // Read/dereference/refresh the std_heap pointer:
267 this->heap = std::make_unique<s_xbt_mheap_t>();
268 this->read(this->heap.get(), this->heap_address);
269 this->cache_flags_ |= RemoteProcess::cache_heap;
272 /** Refresh the information about the process
274 * Do not use directly, this is used by the getters when appropriate
275 * in order to have fresh data.
277 void RemoteProcess::refresh_malloc_info()
279 // Refresh process->heapinfo:
280 if (this->cache_flags_ & RemoteProcess::cache_malloc)
282 size_t count = this->heap->heaplimit + 1;
283 if (this->heap_info.size() < count)
284 this->heap_info.resize(count);
285 this->read_bytes(this->heap_info.data(), count * sizeof(malloc_info), remote(this->heap->heapinfo));
286 this->cache_flags_ |= RemoteProcess::cache_malloc;
289 /** @brief Finds the range of the different memory segments and binary paths */
290 void RemoteProcess::init_memory_map_info()
292 XBT_DEBUG("Get debug information ...");
293 this->maestro_stack_start_ = nullptr;
294 this->maestro_stack_end_ = nullptr;
295 this->object_infos.clear();
296 this->binary_info = nullptr;
298 std::vector<simgrid::xbt::VmMap> const& maps = this->memory_map_;
300 const char* current_name = nullptr;
302 for (size_t i = 0; i < maps.size(); i++) {
303 simgrid::xbt::VmMap const& reg = maps[i];
304 const char* pathname = maps[i].pathname.c_str();
307 if (maps[i].pathname.empty()) {
308 current_name = nullptr;
312 // [stack], [vvar], [vsyscall], [vdso] ...
313 if (pathname[0] == '[') {
314 if ((reg.prot & PROT_WRITE) && not memcmp(pathname, "[stack]", 7)) {
315 this->maestro_stack_start_ = remote(reg.start_addr);
316 this->maestro_stack_end_ = remote(reg.end_addr);
318 current_name = nullptr;
322 if (current_name && strcmp(current_name, pathname) == 0)
325 current_name = pathname;
326 if (not(reg.prot & PROT_READ) && (reg.prot & PROT_EXEC))
329 const bool is_executable = not i;
331 if (not is_executable) {
332 libname = get_lib_name(pathname);
333 if (is_filtered_lib(libname)) {
338 std::shared_ptr<simgrid::mc::ObjectInformation> info =
339 simgrid::mc::createObjectInformation(this->memory_map_, pathname);
340 this->object_infos.push_back(info);
342 this->binary_info = info;
345 xbt_assert(this->maestro_stack_start_, "Did not find maestro_stack_start");
346 xbt_assert(this->maestro_stack_end_, "Did not find maestro_stack_end");
348 XBT_DEBUG("Get debug information done !");
351 std::shared_ptr<simgrid::mc::ObjectInformation> RemoteProcess::find_object_info(RemotePtr<void> addr) const
353 for (auto const& object_info : this->object_infos)
354 if (addr.address() >= (std::uint64_t)object_info->start && addr.address() <= (std::uint64_t)object_info->end)
359 std::shared_ptr<ObjectInformation> RemoteProcess::find_object_info_exec(RemotePtr<void> addr) const
361 for (std::shared_ptr<ObjectInformation> const& info : this->object_infos)
362 if (addr.address() >= (std::uint64_t)info->start_exec && addr.address() <= (std::uint64_t)info->end_exec)
367 std::shared_ptr<ObjectInformation> RemoteProcess::find_object_info_rw(RemotePtr<void> addr) const
369 for (std::shared_ptr<ObjectInformation> const& info : this->object_infos)
370 if (addr.address() >= (std::uint64_t)info->start_rw && addr.address() <= (std::uint64_t)info->end_rw)
375 simgrid::mc::Frame* RemoteProcess::find_function(RemotePtr<void> ip) const
377 std::shared_ptr<simgrid::mc::ObjectInformation> info = this->find_object_info_exec(ip);
378 return info ? info->find_function((void*)ip.address()) : nullptr;
381 /** Find (one occurrence of) the named variable definition
383 const simgrid::mc::Variable* RemoteProcess::find_variable(const char* name) const
385 // First lookup the variable in the executable shared object.
386 // A global variable used directly by the executable code from a library
387 // is reinstantiated in the executable memory .data/.bss.
388 // We need to look up the variable in the executable first.
389 if (this->binary_info) {
390 std::shared_ptr<simgrid::mc::ObjectInformation> const& info = this->binary_info;
391 const simgrid::mc::Variable* var = info->find_variable(name);
396 for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos) {
397 const simgrid::mc::Variable* var = info->find_variable(name);
405 void RemoteProcess::read_variable(const char* name, void* target, size_t size) const
407 const simgrid::mc::Variable* var = this->find_variable(name);
408 xbt_assert(var, "Variable %s not found", name);
409 xbt_assert(var->address, "No simple location for this variable");
411 if (not var->type->full_type) // Try to resolve this type. The needed ObjectInfo was maybe (lazily) loaded recently
412 for (auto const& object_info : this->object_infos)
413 postProcessObjectInformation(this, object_info.get());
414 xbt_assert(var->type->full_type, "Partial type for %s (even after re-resolving types), cannot retrieve its size.",
416 xbt_assert((size_t)var->type->full_type->byte_size == size, "Unexpected size for %s (expected %zu, received %zu).",
417 name, size, (size_t)var->type->full_type->byte_size);
418 this->read_bytes(target, size, remote(var->address));
421 std::string RemoteProcess::read_string(RemotePtr<char> address) const
426 std::vector<char> res(128);
430 ssize_t c = pread(this->memory_file, res.data() + off, res.size() - off, (off_t)address.address() + off);
431 if (c == -1 && errno == EINTR)
433 xbt_assert(c > 0, "Could not read string from remote process");
435 const void* p = memchr(res.data() + off, '\0', c);
437 return std::string(res.data());
440 if (off == (off_t)res.size())
441 res.resize(res.size() * 2);
445 void* RemoteProcess::read_bytes(void* buffer, std::size_t size, RemotePtr<void> address, ReadOptions /*options*/) const
447 xbt_assert(pread_whole(this->memory_file, buffer, size, (size_t)address.address()) != -1,
448 "Read at %p from process %lli failed", (void*)address.address(), (long long)this->pid_);
452 /** Write data to a process memory
454 * @param buffer local memory address (source)
455 * @param len data size
456 * @param address target process memory address (target)
458 void RemoteProcess::write_bytes(const void* buffer, size_t len, RemotePtr<void> address) const
460 xbt_assert(pwrite_whole(this->memory_file, buffer, len, (size_t)address.address()) != -1,
461 "Write to process %lli failed", (long long)this->pid_);
464 static void zero_buffer_init(const void** zero_buffer, size_t zero_buffer_size)
466 int fd = open("/dev/zero", O_RDONLY);
467 xbt_assert(fd >= 0, "Could not open /dev/zero");
468 *zero_buffer = mmap(nullptr, zero_buffer_size, PROT_READ, MAP_SHARED, fd, 0);
469 xbt_assert(*zero_buffer != MAP_FAILED, "Could not map the zero buffer");
473 void RemoteProcess::clear_bytes(RemotePtr<void> address, size_t len) const
475 static constexpr size_t zero_buffer_size = 10 * 4096;
476 static const void* zero_buffer;
477 static std::once_flag zero_buffer_flag;
479 std::call_once(zero_buffer_flag, zero_buffer_init, &zero_buffer, zero_buffer_size);
481 size_t s = len > zero_buffer_size ? zero_buffer_size : len;
482 this->write_bytes(zero_buffer, s, address);
483 address = remote((char*)address.address() + s);
488 void RemoteProcess::ignore_region(std::uint64_t addr, std::size_t size)
490 IgnoredRegion region;
494 auto pos = std::lower_bound(ignored_regions_.begin(), ignored_regions_.end(), region,
495 [](auto const& reg1, auto const& reg2) {
496 return reg1.addr < reg2.addr || (reg1.addr == reg2.addr && reg1.size < reg2.size);
498 if (pos == ignored_regions_.end() || pos->addr != addr || pos->size != size)
499 ignored_regions_.insert(pos, region);
502 void RemoteProcess::ignore_heap(IgnoredHeapRegion const& region)
504 // Binary search the position of insertion:
505 auto pos = std::lower_bound(ignored_heap_.begin(), ignored_heap_.end(), region.address,
506 [](auto const& reg, auto const* addr) { return reg.address < addr; });
507 if (pos == ignored_heap_.end() || pos->address != region.address) {
509 ignored_heap_.insert(pos, region);
513 void RemoteProcess::unignore_heap(void* address, size_t size)
516 auto pos = std::lower_bound(ignored_heap_.begin(), ignored_heap_.end(), address,
517 [](auto const& reg, auto const* addr) { return reg.address < addr; });
518 if (pos != ignored_heap_.end() && static_cast<char*>(pos->address) <= static_cast<char*>(address) + size)
519 ignored_heap_.erase(pos);
522 void RemoteProcess::ignore_local_variable(const char* var_name, const char* frame_name) const
524 if (frame_name != nullptr && strcmp(frame_name, "*") == 0)
525 frame_name = nullptr;
526 for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos)
527 info->remove_local_variable(var_name, frame_name);
530 std::vector<simgrid::mc::ActorInformation>& RemoteProcess::actors()
532 this->refresh_simix();
533 return smx_actors_infos;
536 std::vector<simgrid::mc::ActorInformation>& RemoteProcess::dead_actors()
538 this->refresh_simix();
539 return smx_dead_actors_infos;
542 void RemoteProcess::dump_stack() const
544 unw_addr_space_t as = unw_create_addr_space(&_UPT_accessors, BYTE_ORDER);
546 XBT_ERROR("Could not initialize ptrace address space");
550 void* context = _UPT_create(this->pid_);
551 if (context == nullptr) {
552 unw_destroy_addr_space(as);
553 XBT_ERROR("Could not initialize ptrace context");
558 if (unw_init_remote(&cursor, as, context) != 0) {
559 _UPT_destroy(context);
560 unw_destroy_addr_space(as);
561 XBT_ERROR("Could not initialiez ptrace cursor");
565 simgrid::mc::dumpStack(stderr, &cursor);
567 _UPT_destroy(context);
568 unw_destroy_addr_space(as);
571 } // namespace simgrid