1 /* Copyright (c) 2014-2022. 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*/
131 "libtsan", /* gcc sanitizers */
132 "libubsan", /* gcc sanitizers */
142 static bool is_filtered_lib(const std::string& libname)
144 return std::find(begin(filtered_libraries), end(filtered_libraries), libname) != end(filtered_libraries);
147 static std::string get_lib_name(const std::string& pathname)
149 std::string map_basename = simgrid::xbt::Path(pathname).get_base_name();
152 size_t pos = map_basename.rfind(".so");
153 if (pos != std::string::npos) {
154 // strip the extension (matching regex "\.so.*$")
155 libname.assign(map_basename, 0, pos);
157 // strip the version suffix (matching regex "-[.0-9-]*$")
159 pos = libname.rfind('-');
160 if (pos == std::string::npos || libname.find_first_not_of(".0123456789", pos + 1) != std::string::npos)
169 static ssize_t pread_whole(int fd, void* buf, size_t count, off_t offset)
171 auto* buffer = static_cast<char*>(buf);
172 ssize_t real_count = count;
174 ssize_t res = pread(fd, buffer, count, offset);
181 else if (errno != EINTR) {
182 XBT_ERROR("pread_whole: %s", strerror(errno));
189 static ssize_t pwrite_whole(int fd, const void* buf, size_t count, off_t offset)
191 const auto* buffer = static_cast<const char*>(buf);
192 ssize_t real_count = count;
194 ssize_t res = pwrite(fd, buffer, count, offset);
201 else if (errno != EINTR) {
202 XBT_ERROR("pwrite_whole: %s", strerror(errno));
209 int open_vm(pid_t pid, int flags)
211 std::string buffer = "/proc/" + std::to_string(pid) + "/mem";
212 return open(buffer.c_str(), flags);
215 // ***** RemoteProcess
217 RemoteProcess::RemoteProcess(pid_t pid) : AddressSpace(this), pid_(pid), running_(true) {}
219 void RemoteProcess::init(xbt_mheap_t mmalloc_default_mdp, unsigned long* maxpid, xbt_dynar_t actors,
220 xbt_dynar_t dead_actors)
222 this->heap_address = remote(mmalloc_default_mdp);
223 this->maxpid_addr_ = remote(maxpid);
224 this->actors_addr_ = remote(actors);
225 this->dead_actors_addr_ = remote(dead_actors);
227 this->memory_map_ = simgrid::xbt::get_memory_map(this->pid_);
228 this->init_memory_map_info();
230 int fd = open_vm(this->pid_, O_RDWR);
231 xbt_assert(fd >= 0, "Could not open file for process virtual address space");
232 this->memory_file = fd;
234 this->smx_actors_infos.clear();
235 this->smx_dead_actors_infos.clear();
236 this->unw_addr_space = simgrid::mc::UnwindContext::createUnwindAddressSpace();
237 this->unw_underlying_addr_space = simgrid::unw::create_addr_space();
238 this->unw_underlying_context = simgrid::unw::create_context(this->unw_underlying_addr_space, this->pid_);
241 RemoteProcess::~RemoteProcess()
243 if (this->memory_file >= 0)
244 close(this->memory_file);
246 if (this->unw_underlying_addr_space != unw_local_addr_space) {
247 if (this->unw_underlying_addr_space)
248 unw_destroy_addr_space(this->unw_underlying_addr_space);
249 if (this->unw_underlying_context)
250 _UPT_destroy(this->unw_underlying_context);
253 unw_destroy_addr_space(this->unw_addr_space);
256 /** Refresh the information about the process
258 * Do not use directly, this is used by the getters when appropriate
259 * in order to have fresh data.
261 void RemoteProcess::refresh_heap()
263 // Read/dereference/refresh the std_heap pointer:
265 this->heap = std::make_unique<s_xbt_mheap_t>();
266 this->read(this->heap.get(), this->heap_address);
267 this->cache_flags_ |= RemoteProcess::cache_heap;
270 /** Refresh the information about the process
272 * Do not use directly, this is used by the getters when appropriate
273 * in order to have fresh data.
275 void RemoteProcess::refresh_malloc_info()
277 // Refresh process->heapinfo:
278 if (this->cache_flags_ & RemoteProcess::cache_malloc)
280 size_t count = this->heap->heaplimit + 1;
281 if (this->heap_info.size() < count)
282 this->heap_info.resize(count);
283 this->read_bytes(this->heap_info.data(), count * sizeof(malloc_info), remote(this->heap->heapinfo));
284 this->cache_flags_ |= RemoteProcess::cache_malloc;
287 /** @brief Finds the range of the different memory segments and binary paths */
288 void RemoteProcess::init_memory_map_info()
290 XBT_DEBUG("Get debug information ...");
291 this->maestro_stack_start_ = nullptr;
292 this->maestro_stack_end_ = nullptr;
293 this->object_infos.clear();
294 this->binary_info = nullptr;
296 std::vector<simgrid::xbt::VmMap> const& maps = this->memory_map_;
298 const char* current_name = nullptr;
300 for (size_t i = 0; i < maps.size(); i++) {
301 simgrid::xbt::VmMap const& reg = maps[i];
302 const char* pathname = maps[i].pathname.c_str();
305 if (maps[i].pathname.empty()) {
306 current_name = nullptr;
310 // [stack], [vvar], [vsyscall], [vdso] ...
311 if (pathname[0] == '[') {
312 if ((reg.prot & PROT_WRITE) && not memcmp(pathname, "[stack]", 7)) {
313 this->maestro_stack_start_ = remote(reg.start_addr);
314 this->maestro_stack_end_ = remote(reg.end_addr);
316 current_name = nullptr;
320 if (current_name && strcmp(current_name, pathname) == 0)
323 current_name = pathname;
324 if (not(reg.prot & PROT_READ) && (reg.prot & PROT_EXEC))
327 const bool is_executable = not i;
329 if (not is_executable) {
330 libname = get_lib_name(pathname);
331 if (is_filtered_lib(libname)) {
336 std::shared_ptr<simgrid::mc::ObjectInformation> info =
337 simgrid::mc::createObjectInformation(this->memory_map_, pathname);
338 this->object_infos.push_back(info);
340 this->binary_info = info;
343 xbt_assert(this->maestro_stack_start_, "Did not find maestro_stack_start");
344 xbt_assert(this->maestro_stack_end_, "Did not find maestro_stack_end");
346 XBT_DEBUG("Get debug information done !");
349 std::shared_ptr<simgrid::mc::ObjectInformation> RemoteProcess::find_object_info(RemotePtr<void> addr) const
351 for (auto const& object_info : this->object_infos)
352 if (addr.address() >= (std::uint64_t)object_info->start && addr.address() <= (std::uint64_t)object_info->end)
357 std::shared_ptr<ObjectInformation> RemoteProcess::find_object_info_exec(RemotePtr<void> addr) const
359 for (std::shared_ptr<ObjectInformation> const& info : this->object_infos)
360 if (addr.address() >= (std::uint64_t)info->start_exec && addr.address() <= (std::uint64_t)info->end_exec)
365 std::shared_ptr<ObjectInformation> RemoteProcess::find_object_info_rw(RemotePtr<void> addr) const
367 for (std::shared_ptr<ObjectInformation> const& info : this->object_infos)
368 if (addr.address() >= (std::uint64_t)info->start_rw && addr.address() <= (std::uint64_t)info->end_rw)
373 simgrid::mc::Frame* RemoteProcess::find_function(RemotePtr<void> ip) const
375 std::shared_ptr<simgrid::mc::ObjectInformation> info = this->find_object_info_exec(ip);
376 return info ? info->find_function((void*)ip.address()) : nullptr;
379 /** Find (one occurrence of) the named variable definition
381 const simgrid::mc::Variable* RemoteProcess::find_variable(const char* name) const
383 // First lookup the variable in the executable shared object.
384 // A global variable used directly by the executable code from a library
385 // is reinstantiated in the executable memory .data/.bss.
386 // We need to look up the variable in the executable first.
387 if (this->binary_info) {
388 std::shared_ptr<simgrid::mc::ObjectInformation> const& info = this->binary_info;
389 const simgrid::mc::Variable* var = info->find_variable(name);
394 for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos) {
395 const simgrid::mc::Variable* var = info->find_variable(name);
403 void RemoteProcess::read_variable(const char* name, void* target, size_t size) const
405 const simgrid::mc::Variable* var = this->find_variable(name);
406 xbt_assert(var, "Variable %s not found", name);
407 xbt_assert(var->address, "No simple location for this variable");
409 if (not var->type->full_type) // Try to resolve this type. The needed ObjectInfo was maybe (lazily) loaded recently
410 for (auto const& object_info : this->object_infos)
411 postProcessObjectInformation(this, object_info.get());
412 xbt_assert(var->type->full_type, "Partial type for %s (even after re-resolving types), cannot retrieve its size.",
414 xbt_assert((size_t)var->type->full_type->byte_size == size, "Unexpected size for %s (expected %zu, received %zu).",
415 name, size, (size_t)var->type->full_type->byte_size);
416 this->read_bytes(target, size, remote(var->address));
419 std::string RemoteProcess::read_string(RemotePtr<char> address) const
424 std::vector<char> res(128);
428 ssize_t c = pread(this->memory_file, res.data() + off, res.size() - off, (off_t)address.address() + off);
429 if (c == -1 && errno == EINTR)
431 xbt_assert(c > 0, "Could not read string from remote process");
433 const void* p = memchr(res.data() + off, '\0', c);
435 return std::string(res.data());
438 if (off == (off_t)res.size())
439 res.resize(res.size() * 2);
443 void* RemoteProcess::read_bytes(void* buffer, std::size_t size, RemotePtr<void> address, ReadOptions /*options*/) const
445 xbt_assert(pread_whole(this->memory_file, buffer, size, (size_t)address.address()) != -1,
446 "Read at %p from process %lli failed", (void*)address.address(), (long long)this->pid_);
450 /** Write data to a process memory
452 * @param buffer local memory address (source)
453 * @param len data size
454 * @param address target process memory address (target)
456 void RemoteProcess::write_bytes(const void* buffer, size_t len, RemotePtr<void> address) const
458 xbt_assert(pwrite_whole(this->memory_file, buffer, len, (size_t)address.address()) != -1,
459 "Write to process %lli failed", (long long)this->pid_);
462 static void zero_buffer_init(const void** zero_buffer, size_t zero_buffer_size)
464 int fd = open("/dev/zero", O_RDONLY);
465 xbt_assert(fd >= 0, "Could not open /dev/zero");
466 *zero_buffer = mmap(nullptr, zero_buffer_size, PROT_READ, MAP_SHARED, fd, 0);
467 xbt_assert(*zero_buffer != MAP_FAILED, "Could not map the zero buffer");
471 void RemoteProcess::clear_bytes(RemotePtr<void> address, size_t len) const
473 static constexpr size_t zero_buffer_size = 10 * 4096;
474 static const void* zero_buffer;
475 static std::once_flag zero_buffer_flag;
477 std::call_once(zero_buffer_flag, zero_buffer_init, &zero_buffer, zero_buffer_size);
479 size_t s = len > zero_buffer_size ? zero_buffer_size : len;
480 this->write_bytes(zero_buffer, s, address);
481 address = remote((char*)address.address() + s);
486 void RemoteProcess::ignore_region(std::uint64_t addr, std::size_t size)
488 IgnoredRegion region;
492 auto pos = std::lower_bound(ignored_regions_.begin(), ignored_regions_.end(), region,
493 [](auto const& reg1, auto const& reg2) {
494 return reg1.addr < reg2.addr || (reg1.addr == reg2.addr && reg1.size < reg2.size);
496 if (pos == ignored_regions_.end() || pos->addr != addr || pos->size != size)
497 ignored_regions_.insert(pos, region);
500 void RemoteProcess::ignore_heap(IgnoredHeapRegion const& region)
502 // Binary search the position of insertion:
503 auto pos = std::lower_bound(ignored_heap_.begin(), ignored_heap_.end(), region.address,
504 [](auto const& reg, auto const* addr) { return reg.address < addr; });
505 if (pos == ignored_heap_.end() || pos->address != region.address) {
507 ignored_heap_.insert(pos, region);
511 void RemoteProcess::unignore_heap(void* address, size_t size)
514 auto pos = std::lower_bound(ignored_heap_.begin(), ignored_heap_.end(), address,
515 [](auto const& reg, auto const* addr) { return reg.address < addr; });
516 if (pos != ignored_heap_.end() && static_cast<char*>(pos->address) <= static_cast<char*>(address) + size)
517 ignored_heap_.erase(pos);
520 void RemoteProcess::ignore_local_variable(const char* var_name, const char* frame_name) const
522 if (frame_name != nullptr && strcmp(frame_name, "*") == 0)
523 frame_name = nullptr;
524 for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos)
525 info->remove_local_variable(var_name, frame_name);
528 std::vector<simgrid::mc::ActorInformation>& RemoteProcess::actors()
530 this->refresh_simix();
531 return smx_actors_infos;
534 std::vector<simgrid::mc::ActorInformation>& RemoteProcess::dead_actors()
536 this->refresh_simix();
537 return smx_dead_actors_infos;
540 void RemoteProcess::dump_stack() const
542 unw_addr_space_t as = unw_create_addr_space(&_UPT_accessors, BYTE_ORDER);
544 XBT_ERROR("Could not initialize ptrace address space");
548 void* context = _UPT_create(this->pid_);
549 if (context == nullptr) {
550 unw_destroy_addr_space(as);
551 XBT_ERROR("Could not initialize ptrace context");
556 if (unw_init_remote(&cursor, as, context) != 0) {
557 _UPT_destroy(context);
558 unw_destroy_addr_space(as);
559 XBT_ERROR("Could not initialize ptrace cursor");
563 simgrid::mc::dumpStack(stderr, &cursor);
565 _UPT_destroy(context);
566 unw_destroy_addr_space(as);
569 } // namespace simgrid