Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
0172d7064e620000b710ce38627cf65f9ca6a06b
[simgrid.git] / src / mc / remote / RemoteClient.cpp
1 /* Copyright (c) 2014-2017. The SimGrid Team. All rights reserved.          */
2
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. */
5
6 #define _FILE_OFFSET_BITS 64 /* needed for pread_whole to work as expected on 32bits */
7
8 #include <algorithm>
9 #include <cassert>
10 #include <cerrno>
11 #include <cstddef>
12 #include <cstdint>
13 #include <string>
14 #include <vector>
15
16 #include <sys/ptrace.h>
17
18 #include <cstdio>
19
20 #include <fcntl.h>
21 #include <sys/mman.h> // PROT_*
22 #include <sys/types.h>
23 #include <unistd.h>
24
25 #include <pthread.h>
26
27 #include <libgen.h>
28
29 #include <libunwind-ptrace.h>
30 #include <libunwind.h>
31
32 #include "xbt/base.h"
33 #include "xbt/file.hpp"
34 #include "xbt/log.h"
35 #include <xbt/mmalloc.h>
36
37 #include "src/mc/mc_smx.hpp"
38 #include "src/mc/mc_snapshot.hpp"
39 #include "src/mc/mc_unw.hpp"
40
41 #include "src/mc/AddressSpace.hpp"
42 #include "src/mc/ObjectInformation.hpp"
43 #include "src/mc/Variable.hpp"
44 #include "src/mc/remote/RemoteClient.hpp"
45
46 using simgrid::mc::remote;
47
48 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_process, mc, "MC process information");
49
50 namespace simgrid {
51 namespace mc {
52
53 // ***** Helper stuff
54
55 // List of library which memory segments are not considered:
56 static const std::vector<std::string> filtered_libraries = {
57 #ifdef __linux__
58     "ld",
59 #elif defined __FreeBSD__
60     "ld-elf",
61     "ld-elf32",
62     "libkvm",      /* kernel data access library */
63     "libprocstat", /* process and file information retrieval */
64     "libthr",      /* thread library */
65     "libutil",
66 #endif
67     "libasan", /* gcc sanitizers */
68     "libargp", /* workarounds for glibc-less systems */
69     "libtsan",
70     "libubsan",
71     "libbz2",
72     "libboost_chrono",
73     "libboost_context",
74     "libboost_context-mt",
75     "libboost_system",
76     "libboost_thread",
77     "libc",
78     "libc++",
79     "libcdt",
80     "libcgraph",
81     "libcrypto",
82     "libcxxrt",
83     "libdl",
84     "libdw",
85     "libelf",
86     "libevent",
87     "libgcc_s",
88     "liblua5.1",
89     "liblua5.3",
90     "liblzma",
91     "libm",
92     "libpthread",
93     "librt",
94     "libstdc++",
95     "libunwind",
96     "libunwind-x86_64",
97     "libunwind-x86",
98     "libunwind-ptrace",
99     "libz"};
100
101 static bool is_simgrid_lib(const std::string& libname)
102 {
103   return libname == "libsimgrid";
104 }
105
106 static bool is_filtered_lib(const std::string& libname)
107 {
108   return std::find(begin(filtered_libraries), end(filtered_libraries), libname) != end(filtered_libraries);
109 }
110
111 static std::string get_lib_name(const std::string& pathname)
112 {
113   constexpr char digits[]  = ".0123456789";
114   std::string map_basename = simgrid::xbt::Path(pathname).getBasename();
115   std::string libname;
116
117   size_t pos = map_basename.rfind(".so");
118   if (pos != std::string::npos && map_basename.find_first_not_of(digits, pos + 3) == std::string::npos) {
119     // strip the extension (matching regex "\.so[.0-9]*$")
120     libname.assign(map_basename, 0, pos);
121
122     // strip the version suffix (matching regex "-[.0-9-]*$")
123     while (true) {
124       pos = libname.rfind('-');
125       if (pos == std::string::npos || libname.find_first_not_of(digits, pos + 1) != std::string::npos)
126         break;
127       libname.erase(pos);
128     }
129   }
130
131   return libname;
132 }
133
134 static ssize_t pread_whole(int fd, void* buf, size_t count, off_t offset)
135 {
136   char* buffer       = (char*)buf;
137   ssize_t real_count = count;
138   while (count) {
139     ssize_t res = pread(fd, buffer, count, offset);
140     if (res > 0) {
141       count -= res;
142       buffer += res;
143       offset += res;
144     } else if (res == 0)
145       return -1;
146     else if (errno != EINTR) {
147       perror("pread_whole");
148       return -1;
149     }
150   }
151   return real_count;
152 }
153
154 static ssize_t pwrite_whole(int fd, const void* buf, size_t count, off_t offset)
155 {
156   const char* buffer = (const char*)buf;
157   ssize_t real_count = count;
158   while (count) {
159     ssize_t res = pwrite(fd, buffer, count, offset);
160     if (res > 0) {
161       count -= res;
162       buffer += res;
163       offset += res;
164     } else if (res == 0)
165       return -1;
166     else if (errno != EINTR)
167       return -1;
168   }
169   return real_count;
170 }
171
172 static pthread_once_t zero_buffer_flag = PTHREAD_ONCE_INIT;
173 static const void* zero_buffer;
174 static const size_t zero_buffer_size = 10 * 4096;
175
176 static void zero_buffer_init()
177 {
178   int fd = open("/dev/zero", O_RDONLY);
179   if (fd < 0)
180     xbt_die("Could not open /dev/zero");
181   zero_buffer = mmap(nullptr, zero_buffer_size, PROT_READ, MAP_SHARED, fd, 0);
182   if (zero_buffer == MAP_FAILED)
183     xbt_die("Could not map the zero buffer");
184   close(fd);
185 }
186
187 int open_vm(pid_t pid, int flags)
188 {
189   const size_t buffer_size = 30;
190   char buffer[buffer_size];
191   int res = snprintf(buffer, buffer_size, "/proc/%lli/mem", (long long)pid);
192   if (res < 0 || (size_t)res >= buffer_size) {
193     errno = ENAMETOOLONG;
194     return -1;
195   }
196   return open(buffer, flags);
197 }
198
199 // ***** Process
200
201 RemoteClient::RemoteClient(pid_t pid, int sockfd) : AddressSpace(this), pid_(pid), channel_(sockfd), running_(true)
202 {
203 }
204
205 void RemoteClient::init()
206 {
207   this->memory_map_ = simgrid::xbt::get_memory_map(this->pid_);
208   this->init_memory_map_info();
209
210   int fd = open_vm(this->pid_, O_RDWR);
211   if (fd < 0)
212     xbt_die("Could not open file for process virtual address space");
213   this->memory_file = fd;
214
215   // Read std_heap (is a struct mdesc*):
216   simgrid::mc::Variable* std_heap_var = this->find_variable("__mmalloc_default_mdp");
217   if (not std_heap_var)
218     xbt_die("No heap information in the target process");
219   if (not std_heap_var->address)
220     xbt_die("No constant address for this variable");
221   this->read_bytes(&this->heap_address, sizeof(mdesc*), remote(std_heap_var->address),
222                    simgrid::mc::ProcessIndexDisabled);
223
224   this->smx_actors_infos.clear();
225   this->smx_dead_actors_infos.clear();
226   this->unw_addr_space            = simgrid::mc::UnwindContext::createUnwindAddressSpace();
227   this->unw_underlying_addr_space = simgrid::unw::create_addr_space();
228   this->unw_underlying_context    = simgrid::unw::create_context(this->unw_underlying_addr_space, this->pid_);
229 }
230
231 RemoteClient::~RemoteClient()
232 {
233   if (this->memory_file >= 0)
234     close(this->memory_file);
235
236   if (this->unw_underlying_addr_space != unw_local_addr_space) {
237     if (this->unw_underlying_addr_space)
238       unw_destroy_addr_space(this->unw_underlying_addr_space);
239     if (this->unw_underlying_context)
240       _UPT_destroy(this->unw_underlying_context);
241   }
242
243   unw_destroy_addr_space(this->unw_addr_space);
244 }
245
246 /** Refresh the information about the process
247  *
248  *  Do not use directly, this is used by the getters when appropriate
249  *  in order to have fresh data.
250  */
251 void RemoteClient::refresh_heap()
252 {
253   // Read/dereference/refresh the std_heap pointer:
254   if (not this->heap)
255     this->heap = std::unique_ptr<s_xbt_mheap_t>(new s_xbt_mheap_t());
256   this->read_bytes(this->heap.get(), sizeof(mdesc), remote(this->heap_address), simgrid::mc::ProcessIndexDisabled);
257   this->cache_flags_ |= RemoteClient::cache_heap;
258 }
259
260 /** Refresh the information about the process
261  *
262  *  Do not use direclty, this is used by the getters when appropriate
263  *  in order to have fresh data.
264  * */
265 void RemoteClient::refresh_malloc_info()
266 {
267   // Refresh process->heapinfo:
268   if (this->cache_flags_ & RemoteClient::cache_malloc)
269     return;
270   size_t count = this->heap->heaplimit + 1;
271   if (this->heap_info.size() < count)
272     this->heap_info.resize(count);
273   this->read_bytes(this->heap_info.data(), count * sizeof(malloc_info), remote(this->heap->heapinfo),
274                    simgrid::mc::ProcessIndexDisabled);
275   this->cache_flags_ |= RemoteClient::cache_malloc;
276 }
277
278 /** @brief Finds the range of the different memory segments and binary paths */
279 void RemoteClient::init_memory_map_info()
280 {
281   XBT_DEBUG("Get debug information ...");
282   this->maestro_stack_start_ = nullptr;
283   this->maestro_stack_end_   = nullptr;
284   this->object_infos.resize(0);
285   this->binary_info     = nullptr;
286   this->libsimgrid_info = nullptr;
287
288   std::vector<simgrid::xbt::VmMap> const& maps = this->memory_map_;
289
290   const char* current_name = nullptr;
291
292   this->object_infos.clear();
293
294   for (size_t i = 0; i < maps.size(); i++) {
295     simgrid::xbt::VmMap const& reg = maps[i];
296     const char* pathname           = maps[i].pathname.c_str();
297
298     // Nothing to do
299     if (maps[i].pathname.empty()) {
300       current_name = nullptr;
301       continue;
302     }
303
304     // [stack], [vvar], [vsyscall], [vdso] ...
305     if (pathname[0] == '[') {
306       if ((reg.prot & PROT_WRITE) && not memcmp(pathname, "[stack]", 7)) {
307         this->maestro_stack_start_ = remote(reg.start_addr);
308         this->maestro_stack_end_   = remote(reg.end_addr);
309       }
310       current_name = nullptr;
311       continue;
312     }
313
314     if (current_name && strcmp(current_name, pathname) == 0)
315       continue;
316
317     current_name = pathname;
318     if (not(reg.prot & PROT_READ) && (reg.prot & PROT_EXEC))
319       continue;
320
321     const bool is_executable = not i;
322     std::string libname;
323     if (not is_executable) {
324       libname = get_lib_name(pathname);
325       if (is_filtered_lib(libname)) {
326         continue;
327       }
328     }
329
330     std::shared_ptr<simgrid::mc::ObjectInformation> info =
331         simgrid::mc::createObjectInformation(this->memory_map_, pathname);
332     this->object_infos.push_back(info);
333     if (is_executable)
334       this->binary_info = info;
335     else if (is_simgrid_lib(libname))
336       this->libsimgrid_info = info;
337   }
338
339   // Resolve time (including across different objects):
340   for (auto const& object_info : this->object_infos)
341     postProcessObjectInformation(this, object_info.get());
342
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");
345
346   XBT_DEBUG("Get debug information done !");
347 }
348
349 std::shared_ptr<simgrid::mc::ObjectInformation> RemoteClient::find_object_info(RemotePtr<void> addr) const
350 {
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)
353       return object_info;
354   return nullptr;
355 }
356
357 std::shared_ptr<ObjectInformation> RemoteClient::find_object_info_exec(RemotePtr<void> addr) const
358 {
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)
361       return info;
362   return nullptr;
363 }
364
365 std::shared_ptr<ObjectInformation> RemoteClient::find_object_info_rw(RemotePtr<void> addr) const
366 {
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)
369       return info;
370   return nullptr;
371 }
372
373 simgrid::mc::Frame* RemoteClient::find_function(RemotePtr<void> ip) const
374 {
375   std::shared_ptr<simgrid::mc::ObjectInformation> info = this->find_object_info_exec(ip);
376   return info ? info->find_function((void*)ip.address()) : nullptr;
377 }
378
379 /** Find (one occurrence of) the named variable definition
380  */
381 simgrid::mc::Variable* RemoteClient::find_variable(const char* name) const
382 {
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 reinstanciated 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     simgrid::mc::Variable* var                                  = info->find_variable(name);
390     if (var)
391       return var;
392   }
393
394   for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos) {
395     simgrid::mc::Variable* var = info->find_variable(name);
396     if (var)
397       return var;
398   }
399
400   return nullptr;
401 }
402
403 void RemoteClient::read_variable(const char* name, void* target, size_t size) const
404 {
405   simgrid::mc::Variable* var = this->find_variable(name);
406   xbt_assert(var->address, "No simple location for this variable");
407   xbt_assert(var->type->full_type, "Partial type for %s, cannot check size", name);
408   xbt_assert((size_t)var->type->full_type->byte_size == size, "Unexpected size for %s (expected %zu, was %zu)", name,
409              size, (size_t)var->type->full_type->byte_size);
410   this->read_bytes(target, size, remote(var->address));
411 }
412
413 std::string RemoteClient::read_string(RemotePtr<char> address) const
414 {
415   if (not address)
416     return {};
417
418   std::vector<char> res(128);
419   off_t off = 0;
420
421   while (1) {
422     ssize_t c = pread(this->memory_file, res.data() + off, res.size() - off, (off_t)address.address() + off);
423     if (c == -1) {
424       if (errno == EINTR)
425         continue;
426       else
427         xbt_die("Could not read from from remote process");
428     }
429     if (c == 0)
430       xbt_die("Could not read string from remote process");
431
432     void* p = memchr(res.data() + off, '\0', c);
433     if (p)
434       return std::string(res.data());
435
436     off += c;
437     if (off == (off_t)res.size())
438       res.resize(res.size() * 2);
439   }
440 }
441
442 const void* RemoteClient::read_bytes(void* buffer, std::size_t size, RemotePtr<void> address, int process_index,
443                                      ReadOptions options) const
444 {
445   if (process_index != simgrid::mc::ProcessIndexDisabled) {
446     std::shared_ptr<simgrid::mc::ObjectInformation> const& info = this->find_object_info_rw((void*)address.address());
447 // Segment overlap is not handled.
448 #if HAVE_SMPI
449     if (info.get() && this->privatized(*info)) {
450       if (process_index < 0)
451         xbt_die("Missing process index");
452       if (process_index >= (int)MC_smpi_process_count())
453         xbt_die("Invalid process index");
454
455       // Read smpi_privatization_regions from MCed:
456       smpi_privatization_region_t remote_smpi_privatization_regions =
457           mc_model_checker->process().read_variable<smpi_privatization_region_t>("smpi_privatization_regions");
458
459       s_smpi_privatization_region_t privatization_region =
460           mc_model_checker->process().read<s_smpi_privatization_region_t>(
461               remote(remote_smpi_privatization_regions + process_index));
462
463       // Address translation in the privatization segment:
464       size_t offset = address.address() - (std::uint64_t)info->start_rw;
465       address       = remote((char*)privatization_region.address + offset);
466     }
467 #endif
468   }
469   if (pread_whole(this->memory_file, buffer, size, (size_t)address.address()) < 0)
470     xbt_die("Read at %p from process %lli failed", (void*)address.address(), (long long)this->pid_);
471   return buffer;
472 }
473
474 /** Write data to a process memory
475  *
476  *  @param buffer   local memory address (source)
477  *  @param len      data size
478  *  @param address  target process memory address (target)
479  */
480 void RemoteClient::write_bytes(const void* buffer, size_t len, RemotePtr<void> address)
481 {
482   if (pwrite_whole(this->memory_file, buffer, len, (size_t)address.address()) < 0)
483     xbt_die("Write to process %lli failed", (long long)this->pid_);
484 }
485
486 void RemoteClient::clear_bytes(RemotePtr<void> address, size_t len)
487 {
488   pthread_once(&zero_buffer_flag, zero_buffer_init);
489   while (len) {
490     size_t s = len > zero_buffer_size ? zero_buffer_size : len;
491     this->write_bytes(zero_buffer, s, address);
492     address = remote((char*)address.address() + s);
493     len -= s;
494   }
495 }
496
497 void RemoteClient::ignore_region(std::uint64_t addr, std::size_t size)
498 {
499   IgnoredRegion region;
500   region.addr = addr;
501   region.size = size;
502
503   if (ignored_regions_.empty()) {
504     ignored_regions_.push_back(region);
505     return;
506   }
507
508   unsigned int cursor           = 0;
509   IgnoredRegion* current_region = nullptr;
510
511   int start = 0;
512   int end   = ignored_regions_.size() - 1;
513   while (start <= end) {
514     cursor         = (start + end) / 2;
515     current_region = &ignored_regions_[cursor];
516     if (current_region->addr == addr) {
517       if (current_region->size == size)
518         return;
519       else if (current_region->size < size)
520         start = cursor + 1;
521       else
522         end = cursor - 1;
523     } else if (current_region->addr < addr)
524       start = cursor + 1;
525     else
526       end = cursor - 1;
527   }
528
529   std::size_t position;
530   if (current_region->addr == addr) {
531     if (current_region->size < size)
532       position = cursor + 1;
533     else
534       position = cursor;
535   } else if (current_region->addr < addr)
536     position = cursor + 1;
537   else
538     position = cursor;
539   ignored_regions_.insert(ignored_regions_.begin() + position, region);
540 }
541
542 void RemoteClient::ignore_heap(IgnoredHeapRegion const& region)
543 {
544   if (ignored_heap_.empty()) {
545     ignored_heap_.push_back(std::move(region));
546     return;
547   }
548
549   typedef std::vector<IgnoredHeapRegion>::size_type size_type;
550
551   size_type start = 0;
552   size_type end   = ignored_heap_.size() - 1;
553
554   // Binary search the position of insertion:
555   size_type cursor;
556   while (start <= end) {
557     cursor               = start + (end - start) / 2;
558     auto& current_region = ignored_heap_[cursor];
559     if (current_region.address == region.address)
560       return;
561     else if (current_region.address < region.address)
562       start = cursor + 1;
563     else if (cursor != 0)
564       end = cursor - 1;
565     // Avoid underflow:
566     else
567       break;
568   }
569
570   // Insert it mc_heap_ignore_region_t:
571   if (ignored_heap_[cursor].address < region.address)
572     ++cursor;
573   ignored_heap_.insert(ignored_heap_.begin() + cursor, region);
574 }
575
576 void RemoteClient::unignore_heap(void* address, size_t size)
577 {
578   typedef std::vector<IgnoredHeapRegion>::size_type size_type;
579
580   size_type start = 0;
581   size_type end   = ignored_heap_.size() - 1;
582
583   // Binary search:
584   size_type cursor;
585   while (start <= end) {
586     cursor       = (start + end) / 2;
587     auto& region = ignored_heap_[cursor];
588     if (region.address < address)
589       start = cursor + 1;
590     else if ((char*)region.address <= ((char*)address + size)) {
591       ignored_heap_.erase(ignored_heap_.begin() + cursor);
592       return;
593     } else if (cursor != 0)
594       end = cursor - 1;
595     // Avoid underflow:
596     else
597       break;
598   }
599 }
600
601 void RemoteClient::ignore_local_variable(const char* var_name, const char* frame_name)
602 {
603   if (frame_name != nullptr && strcmp(frame_name, "*") == 0)
604     frame_name = nullptr;
605   for (std::shared_ptr<simgrid::mc::ObjectInformation> const& info : this->object_infos)
606     info->remove_local_variable(var_name, frame_name);
607 }
608
609 std::vector<simgrid::mc::ActorInformation>& RemoteClient::actors()
610 {
611   this->refresh_simix();
612   return smx_actors_infos;
613 }
614
615 std::vector<simgrid::mc::ActorInformation>& RemoteClient::dead_actors()
616 {
617   this->refresh_simix();
618   return smx_dead_actors_infos;
619 }
620
621 void RemoteClient::dumpStack()
622 {
623   unw_addr_space_t as = unw_create_addr_space(&_UPT_accessors, BYTE_ORDER);
624   if (as == nullptr) {
625     XBT_ERROR("Could not initialize ptrace address space");
626     return;
627   }
628
629   void* context = _UPT_create(this->pid_);
630   if (context == nullptr) {
631     unw_destroy_addr_space(as);
632     XBT_ERROR("Could not initialize ptrace context");
633     return;
634   }
635
636   unw_cursor_t cursor;
637   if (unw_init_remote(&cursor, as, context) != 0) {
638     _UPT_destroy(context);
639     unw_destroy_addr_space(as);
640     XBT_ERROR("Could not initialiez ptrace cursor");
641     return;
642   }
643
644   simgrid::mc::dumpStack(stderr, cursor);
645
646   _UPT_destroy(context);
647   unw_destroy_addr_space(as);
648 }
649
650 bool RemoteClient::actor_is_enabled(aid_t pid)
651 {
652   s_mc_message_actor_enabled_t msg{MC_MESSAGE_ACTOR_ENABLED, pid};
653   process()->getChannel().send(msg);
654   char buff[MC_MESSAGE_LENGTH];
655   ssize_t received = process()->getChannel().receive(buff, MC_MESSAGE_LENGTH, true);
656   xbt_assert(received == sizeof(s_mc_message_int_t), "Unexpected size in answer to ACTOR_ENABLED");
657   return ((s_mc_message_int_t*)buff)->value;
658 }
659 }
660 }