1 /* Copyright (c) 2014-2023. 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/sosp/RemoteProcessMemory.hpp"
10 #include "src/mc/explo/Exploration.hpp"
11 #include "src/mc/explo/LivenessChecker.hpp"
12 #include "src/mc/sosp/Snapshot.hpp"
13 #include "xbt/file.hpp"
15 #include "xbt/system_error.hpp"
18 #include <libunwind-ptrace.h>
19 #include <sys/mman.h> // PROT_*
27 #include <string_view>
29 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_process, mc, "MC process information");
31 namespace simgrid::mc {
35 static bool is_filtered_lib(std::string_view libname)
37 return libname != "libsimgrid";
40 static std::string get_lib_name(const std::string& pathname)
42 std::string map_basename = simgrid::xbt::Path(pathname).get_base_name();
45 if (size_t pos = map_basename.rfind(".so"); pos != std::string::npos) {
46 // strip the extension (matching regex "\.so.*$")
47 libname.assign(map_basename, 0, pos);
49 // strip the version suffix (matching regex "-[.0-9-]*$")
51 pos = libname.rfind('-');
52 if (pos == std::string::npos || libname.find_first_not_of(".0123456789", pos + 1) != std::string::npos)
61 static ssize_t pread_whole(int fd, void* buf, size_t count, off_t offset)
63 auto* buffer = static_cast<char*>(buf);
64 ssize_t real_count = count;
66 ssize_t res = pread(fd, buffer, count, offset);
73 else if (errno != EINTR) {
74 XBT_ERROR("pread_whole: %s", strerror(errno));
81 static ssize_t pwrite_whole(int fd, const void* buf, size_t count, off_t offset)
83 const auto* buffer = static_cast<const char*>(buf);
84 ssize_t real_count = count;
86 ssize_t res = pwrite(fd, buffer, count, offset);
93 else if (errno != EINTR) {
94 XBT_ERROR("pwrite_whole: %s", strerror(errno));
101 int open_vm(pid_t pid, int flags)
103 std::string buffer = "/proc/" + std::to_string(pid) + "/mem";
104 return open(buffer.c_str(), flags);
107 // ***** RemoteProcessMemory
109 RemoteProcessMemory::RemoteProcessMemory(pid_t pid) : AddressSpace(this), pid_(pid), running_(true) {}
111 void RemoteProcessMemory::init(xbt_mheap_t mmalloc_default_mdp)
113 this->heap_address = remote(mmalloc_default_mdp);
115 this->memory_map_ = simgrid::xbt::get_memory_map(this->pid_);
116 this->init_memory_map_info();
118 int fd = open_vm(this->pid_, O_RDWR);
119 xbt_assert(fd >= 0, "Could not open file for process virtual address space");
120 this->memory_file = fd;
122 this->unw_addr_space = simgrid::mc::UnwindContext::createUnwindAddressSpace();
123 this->unw_underlying_addr_space = simgrid::unw::create_addr_space();
124 this->unw_underlying_context = simgrid::unw::create_context(this->unw_underlying_addr_space, this->pid_);
126 auto ignored_local_variables = {
127 std::make_pair("e", "*"),
128 std::make_pair("_log_ev", "*"),
130 /* Ignore local variable about time used for tracing */
131 std::make_pair("start_time", "*"),
133 for (auto const& [var, frame] : ignored_local_variables)
134 ignore_local_variable(var, frame);
136 ignore_global_variable("counter"); // Static variable used for tracing
139 RemoteProcessMemory::~RemoteProcessMemory()
141 if (this->memory_file >= 0)
142 close(this->memory_file);
144 if (this->unw_underlying_addr_space != unw_local_addr_space) {
145 if (this->unw_underlying_addr_space)
146 unw_destroy_addr_space(this->unw_underlying_addr_space);
147 if (this->unw_underlying_context)
148 _UPT_destroy(this->unw_underlying_context);
151 unw_destroy_addr_space(this->unw_addr_space);
154 /** Refresh the information about the process
156 * Do not use directly, this is used by the getters when appropriate
157 * in order to have fresh data.
159 void RemoteProcessMemory::refresh_heap()
161 // Read/dereference/refresh the std_heap pointer:
162 this->read(this->heap.get(), this->heap_address);
163 this->cache_flags_ |= RemoteProcessMemory::cache_heap;
166 /** Refresh the information about the process
168 * Do not use directly, this is used by the getters when appropriate
169 * in order to have fresh data.
171 void RemoteProcessMemory::refresh_malloc_info()
173 // Refresh process->heapinfo:
174 if (this->cache_flags_ & RemoteProcessMemory::cache_malloc)
176 size_t count = this->heap->heaplimit + 1;
177 if (this->heap_info.size() < count)
178 this->heap_info.resize(count);
179 this->read_bytes(this->heap_info.data(), count * sizeof(malloc_info), remote(this->heap->heapinfo));
180 this->cache_flags_ |= RemoteProcessMemory::cache_malloc;
182 std::size_t RemoteProcessMemory::get_remote_heap_bytes()
184 return mmalloc_get_bytes_used_remote(get_heap()->heaplimit, get_malloc_info());
187 /** @brief Finds the range of the different memory segments and binary paths */
188 void RemoteProcessMemory::init_memory_map_info()
190 XBT_DEBUG("Get debug information ...");
191 this->maestro_stack_start_ = nullptr;
192 this->maestro_stack_end_ = nullptr;
193 this->object_infos.clear();
194 this->binary_info = nullptr;
196 std::vector<simgrid::xbt::VmMap> const& maps = this->memory_map_;
198 const char* current_name = nullptr;
200 for (size_t i = 0; i < maps.size(); i++) {
201 simgrid::xbt::VmMap const& reg = maps[i];
202 const char* pathname = maps[i].pathname.c_str();
205 if (maps[i].pathname.empty()) {
206 current_name = nullptr;
210 // [stack], [vvar], [vsyscall], [vdso] ...
211 if (pathname[0] == '[') {
212 if ((reg.prot & PROT_WRITE) && not memcmp(pathname, "[stack]", 7)) {
213 this->maestro_stack_start_ = remote(reg.start_addr);
214 this->maestro_stack_end_ = remote(reg.end_addr);
216 current_name = nullptr;
220 if (current_name && strcmp(current_name, pathname) == 0)
223 current_name = pathname;
224 if (not(reg.prot & PROT_READ) && (reg.prot & PROT_EXEC))
227 const bool is_executable = not i;
229 if (not is_executable) {
230 libname = get_lib_name(pathname);
231 if (is_filtered_lib(libname)) {
236 std::shared_ptr<simgrid::mc::ObjectInformation> info =
237 simgrid::mc::createObjectInformation(this->memory_map_, pathname);
238 this->object_infos.push_back(info);
240 this->binary_info = info;
243 xbt_assert(this->maestro_stack_start_, "Did not find maestro_stack_start");
244 xbt_assert(this->maestro_stack_end_, "Did not find maestro_stack_end");
246 XBT_DEBUG("Get debug information done !");
249 std::shared_ptr<simgrid::mc::ObjectInformation> RemoteProcessMemory::find_object_info(RemotePtr<void> addr) const
251 for (auto const& object_info : this->object_infos)
252 if (addr.address() >= (std::uint64_t)object_info->start && addr.address() <= (std::uint64_t)object_info->end)
257 std::shared_ptr<ObjectInformation> RemoteProcessMemory::find_object_info_exec(RemotePtr<void> addr) const
259 for (std::shared_ptr<ObjectInformation> const& info : this->object_infos)
260 if (addr.address() >= (std::uint64_t)info->start_exec && addr.address() <= (std::uint64_t)info->end_exec)
265 std::shared_ptr<ObjectInformation> RemoteProcessMemory::find_object_info_rw(RemotePtr<void> addr) const
267 for (std::shared_ptr<ObjectInformation> const& info : this->object_infos)
268 if (addr.address() >= (std::uint64_t)info->start_rw && addr.address() <= (std::uint64_t)info->end_rw)
273 simgrid::mc::Frame* RemoteProcessMemory::find_function(RemotePtr<void> ip) const
275 std::shared_ptr<simgrid::mc::ObjectInformation> info = this->find_object_info_exec(ip);
276 return info ? info->find_function((void*)ip.address()) : nullptr;
279 /** Find (one occurrence of) the named variable definition
281 const simgrid::mc::Variable* RemoteProcessMemory::find_variable(const char* name) const
283 // First lookup the variable in the executable shared object.
284 // A global variable used directly by the executable code from a library
285 // is reinstantiated in the executable memory .data/.bss.
286 // We need to look up the variable in the executable first.
287 if (this->binary_info) {
288 std::shared_ptr<simgrid::mc::ObjectInformation> const& info = this->binary_info;
289 const simgrid::mc::Variable* var = info->find_variable(name);
294 for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos) {
295 const simgrid::mc::Variable* var = info->find_variable(name);
303 void RemoteProcessMemory::read_variable(const char* name, void* target, size_t size) const
305 const simgrid::mc::Variable* var = this->find_variable(name);
306 xbt_assert(var, "Variable %s not found", name);
307 xbt_assert(var->address, "No simple location for this variable");
309 if (not var->type->full_type) // Try to resolve this type. The needed ObjectInfo was maybe (lazily) loaded recently
310 for (auto const& object_info : this->object_infos)
311 postProcessObjectInformation(this, object_info.get());
312 xbt_assert(var->type->full_type, "Partial type for %s (even after re-resolving types), cannot retrieve its size.",
314 xbt_assert((size_t)var->type->full_type->byte_size == size, "Unexpected size for %s (expected %zu, received %zu).",
315 name, size, (size_t)var->type->full_type->byte_size);
316 this->read_bytes(target, size, remote(var->address));
319 std::string RemoteProcessMemory::read_string(RemotePtr<char> address) const
324 std::vector<char> res(128);
328 ssize_t c = pread(this->memory_file, res.data() + off, res.size() - off, (off_t)address.address() + off);
329 if (c == -1 && errno == EINTR)
331 xbt_assert(c > 0, "Could not read string from remote process");
333 if (memchr(res.data() + off, '\0', c))
337 if (off == (off_t)res.size())
338 res.resize(res.size() * 2);
342 void* RemoteProcessMemory::read_bytes(void* buffer, std::size_t size, RemotePtr<void> address,
343 ReadOptions /*options*/) const
345 xbt_assert(pread_whole(this->memory_file, buffer, size, (size_t)address.address()) != -1,
346 "Read at %p from process %lli failed", (void*)address.address(), (long long)this->pid_);
350 /** Write data to a process memory
352 * @param buffer local memory address (source)
353 * @param len data size
354 * @param address target process memory address (target)
356 void RemoteProcessMemory::write_bytes(const void* buffer, size_t len, RemotePtr<void> address) const
358 xbt_assert(pwrite_whole(this->memory_file, buffer, len, (size_t)address.address()) != -1,
359 "Write to process %lli failed", (long long)this->pid_);
362 static void zero_buffer_init(const void** zero_buffer, size_t zero_buffer_size)
364 int fd = open("/dev/zero", O_RDONLY);
365 xbt_assert(fd >= 0, "Could not open /dev/zero");
366 *zero_buffer = mmap(nullptr, zero_buffer_size, PROT_READ, MAP_SHARED, fd, 0);
367 xbt_assert(*zero_buffer != MAP_FAILED, "Could not map the zero buffer");
371 void RemoteProcessMemory::clear_bytes(RemotePtr<void> address, size_t len) const
373 static constexpr size_t zero_buffer_size = 10 * 4096;
374 static const void* zero_buffer;
375 static std::once_flag zero_buffer_flag;
377 std::call_once(zero_buffer_flag, zero_buffer_init, &zero_buffer, zero_buffer_size);
379 size_t s = len > zero_buffer_size ? zero_buffer_size : len;
380 this->write_bytes(zero_buffer, s, address);
381 address = remote((char*)address.address() + s);
386 void RemoteProcessMemory::ignore_region(std::uint64_t addr, std::size_t size)
388 IgnoredRegion region;
392 auto pos = std::lower_bound(ignored_regions_.begin(), ignored_regions_.end(), region,
393 [](auto const& reg1, auto const& reg2) {
394 return reg1.addr < reg2.addr || (reg1.addr == reg2.addr && reg1.size < reg2.size);
396 if (pos == ignored_regions_.end() || pos->addr != addr || pos->size != size)
397 ignored_regions_.insert(pos, region);
400 void RemoteProcessMemory::ignore_heap(IgnoredHeapRegion const& region)
402 // Binary search the position of insertion:
403 auto pos = std::lower_bound(ignored_heap_.begin(), ignored_heap_.end(), region.address,
404 [](auto const& reg, auto const* addr) { return reg.address < addr; });
405 if (pos == ignored_heap_.end() || pos->address != region.address) {
407 ignored_heap_.insert(pos, region);
411 void RemoteProcessMemory::unignore_heap(void* address, size_t size)
414 auto pos = std::lower_bound(ignored_heap_.begin(), ignored_heap_.end(), address,
415 [](auto const& reg, auto const* addr) { return reg.address < addr; });
416 if (pos != ignored_heap_.end() && static_cast<char*>(pos->address) <= static_cast<char*>(address) + size)
417 ignored_heap_.erase(pos);
420 void RemoteProcessMemory::ignore_local_variable(const char* var_name, const char* frame_name) const
422 if (frame_name != nullptr && strcmp(frame_name, "*") == 0)
423 frame_name = nullptr;
424 for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos)
425 info->remove_local_variable(var_name, frame_name);
428 void RemoteProcessMemory::dump_stack() const
430 unw_addr_space_t as = unw_create_addr_space(&_UPT_accessors, BYTE_ORDER);
432 XBT_ERROR("Could not initialize ptrace address space");
436 void* context = _UPT_create(this->pid_);
437 if (context == nullptr) {
438 unw_destroy_addr_space(as);
439 XBT_ERROR("Could not initialize ptrace context");
444 if (unw_init_remote(&cursor, as, context) != 0) {
445 _UPT_destroy(context);
446 unw_destroy_addr_space(as);
447 XBT_ERROR("Could not initialize ptrace cursor");
451 simgrid::mc::dumpStack(stderr, &cursor);
453 _UPT_destroy(context);
454 unw_destroy_addr_space(as);
457 } // namespace simgrid::mc