1 /*
2 * Copyright (C) 2017 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #include "src/tracing/service/tracing_service_impl.h"
18
19 #include <limits.h>
20 #include <string.h>
21
22 #include <algorithm>
23 #include <cinttypes>
24 #include <cstdint>
25 #include <limits>
26 #include <optional>
27 #include <string>
28 #include <unordered_set>
29
30 #if !PERFETTO_BUILDFLAG(PERFETTO_OS_WIN) && \
31 !PERFETTO_BUILDFLAG(PERFETTO_OS_NACL)
32 #include <sys/uio.h>
33 #include <sys/utsname.h>
34 #include <unistd.h>
35 #endif
36
37 #if PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID) && \
38 PERFETTO_BUILDFLAG(PERFETTO_ANDROID_BUILD)
39 #include "src/android_internal/lazy_library_loader.h" // nogncheck
40 #include "src/android_internal/tracing_service_proxy.h" // nogncheck
41 #endif
42
43 #if PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID) || \
44 PERFETTO_BUILDFLAG(PERFETTO_OS_LINUX) || \
45 PERFETTO_BUILDFLAG(PERFETTO_OS_APPLE)
46 #define PERFETTO_HAS_CHMOD
47 #include <sys/stat.h>
48 #endif
49
50 #include "perfetto/base/build_config.h"
51 #include "perfetto/base/status.h"
52 #include "perfetto/base/task_runner.h"
53 #include "perfetto/ext/base/android_utils.h"
54 #include "perfetto/ext/base/clock_snapshots.h"
55 #include "perfetto/ext/base/file_utils.h"
56 #include "perfetto/ext/base/metatrace.h"
57 #include "perfetto/ext/base/string_utils.h"
58 #include "perfetto/ext/base/string_view.h"
59 #include "perfetto/ext/base/sys_types.h"
60 #include "perfetto/ext/base/utils.h"
61 #include "perfetto/ext/base/uuid.h"
62 #include "perfetto/ext/base/version.h"
63 #include "perfetto/ext/base/watchdog.h"
64 #include "perfetto/ext/tracing/core/basic_types.h"
65 #include "perfetto/ext/tracing/core/client_identity.h"
66 #include "perfetto/ext/tracing/core/consumer.h"
67 #include "perfetto/ext/tracing/core/observable_events.h"
68 #include "perfetto/ext/tracing/core/producer.h"
69 #include "perfetto/ext/tracing/core/shared_memory.h"
70 #include "perfetto/ext/tracing/core/shared_memory_abi.h"
71 #include "perfetto/ext/tracing/core/trace_packet.h"
72 #include "perfetto/ext/tracing/core/trace_writer.h"
73 #include "perfetto/protozero/scattered_heap_buffer.h"
74 #include "perfetto/protozero/static_buffer.h"
75 #include "perfetto/tracing/core/data_source_descriptor.h"
76 #include "perfetto/tracing/core/tracing_service_capabilities.h"
77 #include "perfetto/tracing/core/tracing_service_state.h"
78 #include "src/android_stats/statsd_logging_helper.h"
79 #include "src/protozero/filtering/message_filter.h"
80 #include "src/protozero/filtering/string_filter.h"
81 #include "src/tracing/core/shared_memory_arbiter_impl.h"
82 #include "src/tracing/service/packet_stream_validator.h"
83 #include "src/tracing/service/trace_buffer.h"
84
85 #include "protos/perfetto/common/builtin_clock.gen.h"
86 #include "protos/perfetto/common/builtin_clock.pbzero.h"
87 #include "protos/perfetto/common/trace_stats.pbzero.h"
88 #include "protos/perfetto/config/trace_config.pbzero.h"
89 #include "protos/perfetto/trace/clock_snapshot.pbzero.h"
90 #include "protos/perfetto/trace/perfetto/tracing_service_event.pbzero.h"
91 #include "protos/perfetto/trace/remote_clock_sync.pbzero.h"
92 #include "protos/perfetto/trace/system_info.pbzero.h"
93 #include "protos/perfetto/trace/trace_packet.pbzero.h"
94 #include "protos/perfetto/trace/trace_uuid.pbzero.h"
95 #include "protos/perfetto/trace/trigger.pbzero.h"
96
97 // General note: this class must assume that Producers are malicious and will
98 // try to crash / exploit this class. We can trust pointers because they come
99 // from the IPC layer, but we should never assume that that the producer calls
100 // come in the right order or their arguments are sane / within bounds.
101
102 // This is a macro because we want the call-site line number for the ELOG.
103 #define PERFETTO_SVC_ERR(...) \
104 (PERFETTO_ELOG(__VA_ARGS__), ::perfetto::base::ErrStatus(__VA_ARGS__))
105
106 namespace perfetto {
107
108 namespace {
109 constexpr int kMaxBuffersPerConsumer = 128;
110 constexpr uint32_t kDefaultSnapshotsIntervalMs = 10 * 1000;
111 constexpr int kDefaultWriteIntoFilePeriodMs = 5000;
112 constexpr int kMinWriteIntoFilePeriodMs = 100;
113 constexpr uint32_t kAllDataSourceStartedTimeout = 20000;
114 constexpr int kMaxConcurrentTracingSessions = 15;
115 constexpr int kMaxConcurrentTracingSessionsPerUid = 5;
116 constexpr int kMaxConcurrentTracingSessionsForStatsdUid = 10;
117 constexpr int64_t kMinSecondsBetweenTracesGuardrail = 5 * 60;
118
119 constexpr uint32_t kMillisPerHour = 3600000;
120 constexpr uint32_t kMillisPerDay = kMillisPerHour * 24;
121 constexpr uint32_t kMaxTracingDurationMillis = 7 * 24 * kMillisPerHour;
122
123 // These apply only if enable_extra_guardrails is true.
124 constexpr uint32_t kGuardrailsMaxTracingBufferSizeKb = 128 * 1024;
125 constexpr uint32_t kGuardrailsMaxTracingDurationMillis = 24 * kMillisPerHour;
126
127 constexpr size_t kMaxLifecycleEventsListedDataSources = 32;
128
129 #if PERFETTO_BUILDFLAG(PERFETTO_OS_WIN) || PERFETTO_BUILDFLAG(PERFETTO_OS_NACL)
130 struct iovec {
131 void* iov_base; // Address
132 size_t iov_len; // Block size
133 };
134
135 // Simple implementation of writev. Note that this does not give the atomicity
136 // guarantees of a real writev, but we don't depend on these (we aren't writing
137 // to the same file from another thread).
writev(int fd,const struct iovec * iov,int iovcnt)138 ssize_t writev(int fd, const struct iovec* iov, int iovcnt) {
139 ssize_t total_size = 0;
140 for (int i = 0; i < iovcnt; ++i) {
141 ssize_t current_size = base::WriteAll(fd, iov[i].iov_base, iov[i].iov_len);
142 if (current_size != static_cast<ssize_t>(iov[i].iov_len))
143 return -1;
144 total_size += current_size;
145 }
146 return total_size;
147 }
148
149 #define IOV_MAX 1024 // Linux compatible limit.
150
151 #elif PERFETTO_BUILDFLAG(PERFETTO_OS_QNX)
152 #define IOV_MAX 1024 // Linux compatible limit.
153 #endif
154
155 // Partially encodes a CommitDataRequest in an int32 for the purposes of
156 // metatracing. Note that it encodes only the bottom 10 bits of the producer id
157 // (which is technically 16 bits wide).
158 //
159 // Format (by bit range):
160 // [ 31 ][ 30 ][ 29:20 ][ 19:10 ][ 9:0]
161 // [unused][has flush id][num chunks to patch][num chunks to move][producer id]
EncodeCommitDataRequest(ProducerID producer_id,const CommitDataRequest & req_untrusted)162 int32_t EncodeCommitDataRequest(ProducerID producer_id,
163 const CommitDataRequest& req_untrusted) {
164 uint32_t cmov = static_cast<uint32_t>(req_untrusted.chunks_to_move_size());
165 uint32_t cpatch = static_cast<uint32_t>(req_untrusted.chunks_to_patch_size());
166 uint32_t has_flush_id = req_untrusted.flush_request_id() != 0;
167
168 uint32_t mask = (1 << 10) - 1;
169 uint32_t acc = 0;
170 acc |= has_flush_id << 30;
171 acc |= (cpatch & mask) << 20;
172 acc |= (cmov & mask) << 10;
173 acc |= (producer_id & mask);
174 return static_cast<int32_t>(acc);
175 }
176
SerializeAndAppendPacket(std::vector<TracePacket> * packets,std::vector<uint8_t> packet)177 void SerializeAndAppendPacket(std::vector<TracePacket>* packets,
178 std::vector<uint8_t> packet) {
179 Slice slice = Slice::Allocate(packet.size());
180 memcpy(slice.own_data(), packet.data(), packet.size());
181 packets->emplace_back();
182 packets->back().AddSlice(std::move(slice));
183 }
184
EnsureValidShmSizes(size_t shm_size,size_t page_size)185 std::tuple<size_t /*shm_size*/, size_t /*page_size*/> EnsureValidShmSizes(
186 size_t shm_size,
187 size_t page_size) {
188 // Theoretically the max page size supported by the ABI is 64KB.
189 // However, the current implementation of TraceBuffer (the non-shared
190 // userspace buffer where the service copies data) supports at most
191 // 32K. Setting 64K "works" from the producer<>consumer viewpoint
192 // but then causes the data to be discarded when copying it into
193 // TraceBuffer.
194 constexpr size_t kMaxPageSize = 32 * 1024;
195 static_assert(kMaxPageSize <= SharedMemoryABI::kMaxPageSize, "");
196
197 if (page_size == 0)
198 page_size = TracingServiceImpl::kDefaultShmPageSize;
199 if (shm_size == 0)
200 shm_size = TracingServiceImpl::kDefaultShmSize;
201
202 page_size = std::min<size_t>(page_size, kMaxPageSize);
203 shm_size = std::min<size_t>(shm_size, TracingServiceImpl::kMaxShmSize);
204
205 // The tracing page size has to be multiple of 4K. On some systems (e.g. Mac
206 // on Arm64) the system page size can be larger (e.g., 16K). That doesn't
207 // matter here, because the tracing page size is just a logical partitioning
208 // and does not have any dependencies on kernel mm syscalls (read: it's fine
209 // to have trace page sizes of 4K on a system where the kernel page size is
210 // 16K).
211 bool page_size_is_valid = page_size >= SharedMemoryABI::kMinPageSize;
212 page_size_is_valid &= page_size % SharedMemoryABI::kMinPageSize == 0;
213
214 // Only allow power of two numbers of pages, i.e. 1, 2, 4, 8 pages.
215 size_t num_pages = page_size / SharedMemoryABI::kMinPageSize;
216 page_size_is_valid &= (num_pages & (num_pages - 1)) == 0;
217
218 if (!page_size_is_valid || shm_size < page_size ||
219 shm_size % page_size != 0) {
220 return std::make_tuple(TracingServiceImpl::kDefaultShmSize,
221 TracingServiceImpl::kDefaultShmPageSize);
222 }
223 return std::make_tuple(shm_size, page_size);
224 }
225
NameMatchesFilter(const std::string & name,const std::vector<std::string> & name_filter,const std::vector<std::string> & name_regex_filter)226 bool NameMatchesFilter(const std::string& name,
227 const std::vector<std::string>& name_filter,
228 const std::vector<std::string>& name_regex_filter) {
229 bool filter_is_set = !name_filter.empty() || !name_regex_filter.empty();
230 if (!filter_is_set)
231 return true;
232 bool filter_matches = std::find(name_filter.begin(), name_filter.end(),
233 name) != name_filter.end();
234 bool filter_regex_matches =
235 std::find_if(name_regex_filter.begin(), name_regex_filter.end(),
236 [&](const std::string& regex) {
237 return std::regex_match(
238 name, std::regex(regex, std::regex::extended));
239 }) != name_regex_filter.end();
240 return filter_matches || filter_regex_matches;
241 }
242
243 // Used when TraceConfig.write_into_file == true and output_path is not empty.
CreateTraceFile(const std::string & path,bool overwrite)244 base::ScopedFile CreateTraceFile(const std::string& path, bool overwrite) {
245 #if PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID) && \
246 PERFETTO_BUILDFLAG(PERFETTO_ANDROID_BUILD)
247 // This is NOT trying to preserve any security property, SELinux does that.
248 // It just improves the actionability of the error when people try to save the
249 // trace in a location that is not SELinux-allowed (a generic "permission
250 // denied" vs "don't put it here, put it there").
251 // These are the only SELinux approved dir for trace files that are created
252 // directly by traced.
253 static const char* kTraceDirBasePath = "/data/misc/perfetto-traces/";
254 if (!base::StartsWith(path, kTraceDirBasePath)) {
255 PERFETTO_ELOG("Invalid output_path %s. On Android it must be within %s.",
256 path.c_str(), kTraceDirBasePath);
257 return base::ScopedFile();
258 }
259 #endif
260 // O_CREAT | O_EXCL will fail if the file exists already.
261 const int flags = O_RDWR | O_CREAT | (overwrite ? O_TRUNC : O_EXCL);
262 auto fd = base::OpenFile(path, flags, 0600);
263 if (fd) {
264 #if defined(PERFETTO_HAS_CHMOD)
265 // Passing 0644 directly above won't work because of umask.
266 PERFETTO_CHECK(fchmod(*fd, 0644) == 0);
267 #endif
268 } else {
269 PERFETTO_PLOG("Failed to create %s", path.c_str());
270 }
271 return fd;
272 }
273
ShouldLogEvent(const TraceConfig & cfg)274 bool ShouldLogEvent(const TraceConfig& cfg) {
275 switch (cfg.statsd_logging()) {
276 case TraceConfig::STATSD_LOGGING_ENABLED:
277 return true;
278 case TraceConfig::STATSD_LOGGING_DISABLED:
279 return false;
280 case TraceConfig::STATSD_LOGGING_UNSPECIFIED:
281 break;
282 }
283 // For backward compatibility with older versions of perfetto_cmd.
284 return cfg.enable_extra_guardrails();
285 }
286
287 // Appends `data` (which has `size` bytes), to `*packet`. Splits the data in
288 // slices no larger than `max_slice_size`.
AppendOwnedSlicesToPacket(std::unique_ptr<uint8_t[]> data,size_t size,size_t max_slice_size,perfetto::TracePacket * packet)289 void AppendOwnedSlicesToPacket(std::unique_ptr<uint8_t[]> data,
290 size_t size,
291 size_t max_slice_size,
292 perfetto::TracePacket* packet) {
293 if (size <= max_slice_size) {
294 packet->AddSlice(Slice::TakeOwnership(std::move(data), size));
295 return;
296 }
297 uint8_t* src_ptr = data.get();
298 for (size_t size_left = size; size_left > 0;) {
299 const size_t slice_size = std::min(size_left, max_slice_size);
300
301 Slice slice = Slice::Allocate(slice_size);
302 memcpy(slice.own_data(), src_ptr, slice_size);
303 packet->AddSlice(std::move(slice));
304
305 src_ptr += slice_size;
306 size_left -= slice_size;
307 }
308 }
309
310 using TraceFilter = protos::gen::TraceConfig::TraceFilter;
ConvertPolicy(TraceFilter::StringFilterPolicy policy)311 std::optional<protozero::StringFilter::Policy> ConvertPolicy(
312 TraceFilter::StringFilterPolicy policy) {
313 switch (policy) {
314 case TraceFilter::SFP_UNSPECIFIED:
315 return std::nullopt;
316 case TraceFilter::SFP_MATCH_REDACT_GROUPS:
317 return protozero::StringFilter::Policy::kMatchRedactGroups;
318 case TraceFilter::SFP_ATRACE_MATCH_REDACT_GROUPS:
319 return protozero::StringFilter::Policy::kAtraceMatchRedactGroups;
320 case TraceFilter::SFP_MATCH_BREAK:
321 return protozero::StringFilter::Policy::kMatchBreak;
322 case TraceFilter::SFP_ATRACE_MATCH_BREAK:
323 return protozero::StringFilter::Policy::kAtraceMatchBreak;
324 case TraceFilter::SFP_ATRACE_REPEATED_SEARCH_REDACT_GROUPS:
325 return protozero::StringFilter::Policy::kAtraceRepeatedSearchRedactGroups;
326 }
327 return std::nullopt;
328 }
329
330 } // namespace
331
332 // static
CreateInstance(std::unique_ptr<SharedMemory::Factory> shm_factory,base::TaskRunner * task_runner,InitOpts init_opts)333 std::unique_ptr<TracingService> TracingService::CreateInstance(
334 std::unique_ptr<SharedMemory::Factory> shm_factory,
335 base::TaskRunner* task_runner,
336 InitOpts init_opts) {
337 tracing_service::Dependencies deps;
338 deps.clock = std::make_unique<tracing_service::ClockImpl>();
339 uint32_t seed = static_cast<uint32_t>(deps.clock->GetWallTimeMs().count());
340 deps.random = std::make_unique<tracing_service::RandomImpl>(seed);
341 return std::unique_ptr<TracingService>(new TracingServiceImpl(
342 std::move(shm_factory), task_runner, std::move(deps), init_opts));
343 }
344
TracingServiceImpl(std::unique_ptr<SharedMemory::Factory> shm_factory,base::TaskRunner * task_runner,tracing_service::Dependencies deps,InitOpts init_opts)345 TracingServiceImpl::TracingServiceImpl(
346 std::unique_ptr<SharedMemory::Factory> shm_factory,
347 base::TaskRunner* task_runner,
348 tracing_service::Dependencies deps,
349 InitOpts init_opts)
350 : clock_(std::move(deps.clock)),
351 random_(std::move(deps.random)),
352 init_opts_(init_opts),
353 shm_factory_(std::move(shm_factory)),
354 uid_(base::GetCurrentUserId()),
355 buffer_ids_(kMaxTraceBufferID),
356 weak_runner_(task_runner) {
357 PERFETTO_DCHECK(task_runner);
358 }
359
~TracingServiceImpl()360 TracingServiceImpl::~TracingServiceImpl() {
361 // TODO(fmayer): handle teardown of all Producer.
362 }
363
364 std::unique_ptr<TracingService::ProducerEndpoint>
ConnectProducer(Producer * producer,const ClientIdentity & client_identity,const std::string & producer_name,size_t shared_memory_size_hint_bytes,bool in_process,ProducerSMBScrapingMode smb_scraping_mode,size_t shared_memory_page_size_hint_bytes,std::unique_ptr<SharedMemory> shm,const std::string & sdk_version)365 TracingServiceImpl::ConnectProducer(Producer* producer,
366 const ClientIdentity& client_identity,
367 const std::string& producer_name,
368 size_t shared_memory_size_hint_bytes,
369 bool in_process,
370 ProducerSMBScrapingMode smb_scraping_mode,
371 size_t shared_memory_page_size_hint_bytes,
372 std::unique_ptr<SharedMemory> shm,
373 const std::string& sdk_version) {
374 PERFETTO_DCHECK_THREAD(thread_checker_);
375
376 auto uid = client_identity.uid();
377 if (lockdown_mode_ && uid != base::GetCurrentUserId()) {
378 PERFETTO_DLOG("Lockdown mode. Rejecting producer with UID %ld",
379 static_cast<unsigned long>(uid));
380 return nullptr;
381 }
382
383 if (producers_.size() >= kMaxProducerID) {
384 PERFETTO_DFATAL("Too many producers.");
385 return nullptr;
386 }
387 const ProducerID id = GetNextProducerID();
388 PERFETTO_DLOG("Producer %" PRIu16 " connected, uid=%d", id,
389 static_cast<int>(uid));
390 bool smb_scraping_enabled = smb_scraping_enabled_;
391 switch (smb_scraping_mode) {
392 case ProducerSMBScrapingMode::kDefault:
393 break;
394 case ProducerSMBScrapingMode::kEnabled:
395 smb_scraping_enabled = true;
396 break;
397 case ProducerSMBScrapingMode::kDisabled:
398 smb_scraping_enabled = false;
399 break;
400 }
401
402 std::unique_ptr<ProducerEndpointImpl> endpoint(new ProducerEndpointImpl(
403 id, client_identity, this, weak_runner_.task_runner(), producer,
404 producer_name, sdk_version, in_process, smb_scraping_enabled));
405 auto it_and_inserted = producers_.emplace(id, endpoint.get());
406 PERFETTO_DCHECK(it_and_inserted.second);
407 endpoint->shmem_size_hint_bytes_ = shared_memory_size_hint_bytes;
408 endpoint->shmem_page_size_hint_bytes_ = shared_memory_page_size_hint_bytes;
409
410 // Producer::OnConnect() should run before Producer::OnTracingSetup(). The
411 // latter may be posted by SetupSharedMemory() below, so post OnConnect() now.
412 endpoint->weak_runner_.PostTask(
413 [endpoint = endpoint.get()] { endpoint->producer_->OnConnect(); });
414
415 if (shm) {
416 // The producer supplied an SMB. This is used only by Chrome; in the most
417 // common cases the SMB is created by the service and passed via
418 // OnTracingSetup(). Verify that it is correctly sized before we attempt to
419 // use it. The transport layer has to verify the integrity of the SMB (e.g.
420 // ensure that the producer can't resize if after the fact).
421 size_t shm_size, page_size;
422 std::tie(shm_size, page_size) =
423 EnsureValidShmSizes(shm->size(), endpoint->shmem_page_size_hint_bytes_);
424 if (shm_size == shm->size() &&
425 page_size == endpoint->shmem_page_size_hint_bytes_) {
426 PERFETTO_DLOG(
427 "Adopting producer-provided SMB of %zu kB for producer \"%s\"",
428 shm_size / 1024, endpoint->name_.c_str());
429 endpoint->SetupSharedMemory(std::move(shm), page_size,
430 /*provided_by_producer=*/true);
431 } else {
432 PERFETTO_LOG(
433 "Discarding incorrectly sized producer-provided SMB for producer "
434 "\"%s\", falling back to service-provided SMB. Requested sizes: %zu "
435 "B total, %zu B page size; suggested corrected sizes: %zu B total, "
436 "%zu B page size",
437 endpoint->name_.c_str(), shm->size(),
438 endpoint->shmem_page_size_hint_bytes_, shm_size, page_size);
439 shm.reset();
440 }
441 }
442
443 return std::unique_ptr<ProducerEndpoint>(std::move(endpoint));
444 }
445
DisconnectProducer(ProducerID id)446 void TracingServiceImpl::DisconnectProducer(ProducerID id) {
447 PERFETTO_DCHECK_THREAD(thread_checker_);
448 PERFETTO_DLOG("Producer %" PRIu16 " disconnected", id);
449 PERFETTO_DCHECK(producers_.count(id));
450
451 // Scrape remaining chunks for this producer to ensure we don't lose data.
452 if (auto* producer = GetProducer(id)) {
453 for (auto& session_id_and_session : tracing_sessions_)
454 ScrapeSharedMemoryBuffers(&session_id_and_session.second, producer);
455 }
456
457 for (auto it = data_sources_.begin(); it != data_sources_.end();) {
458 auto next = it;
459 next++;
460 if (it->second.producer_id == id)
461 UnregisterDataSource(id, it->second.descriptor.name());
462 it = next;
463 }
464
465 producers_.erase(id);
466 UpdateMemoryGuardrail();
467 }
468
GetProducer(ProducerID id) const469 TracingServiceImpl::ProducerEndpointImpl* TracingServiceImpl::GetProducer(
470 ProducerID id) const {
471 PERFETTO_DCHECK_THREAD(thread_checker_);
472 auto it = producers_.find(id);
473 if (it == producers_.end())
474 return nullptr;
475 return it->second;
476 }
477
478 std::unique_ptr<TracingService::ConsumerEndpoint>
ConnectConsumer(Consumer * consumer,uid_t uid)479 TracingServiceImpl::ConnectConsumer(Consumer* consumer, uid_t uid) {
480 PERFETTO_DCHECK_THREAD(thread_checker_);
481 PERFETTO_DLOG("Consumer %p connected from UID %" PRIu64,
482 reinterpret_cast<void*>(consumer), static_cast<uint64_t>(uid));
483 std::unique_ptr<ConsumerEndpointImpl> endpoint(new ConsumerEndpointImpl(
484 this, weak_runner_.task_runner(), consumer, uid));
485 // Consumer might go away before we're able to send the connect notification,
486 // if that is the case just bail out.
487 auto weak_ptr = endpoint->weak_ptr_factory_.GetWeakPtr();
488 weak_runner_.task_runner()->PostTask([weak_ptr = std::move(weak_ptr)] {
489 if (weak_ptr)
490 weak_ptr->consumer_->OnConnect();
491 });
492 return std::unique_ptr<ConsumerEndpoint>(std::move(endpoint));
493 }
494
DisconnectConsumer(ConsumerEndpointImpl * consumer)495 void TracingServiceImpl::DisconnectConsumer(ConsumerEndpointImpl* consumer) {
496 PERFETTO_DCHECK_THREAD(thread_checker_);
497 PERFETTO_DLOG("Consumer %p disconnected", reinterpret_cast<void*>(consumer));
498
499 if (consumer->tracing_session_id_)
500 FreeBuffers(consumer->tracing_session_id_); // Will also DisableTracing().
501
502 // At this point no more pointers to |consumer| should be around.
503 PERFETTO_DCHECK(!std::any_of(
504 tracing_sessions_.begin(), tracing_sessions_.end(),
505 [consumer](const std::pair<const TracingSessionID, TracingSession>& kv) {
506 return kv.second.consumer_maybe_null == consumer;
507 }));
508 }
509
DetachConsumer(ConsumerEndpointImpl * consumer,const std::string & key)510 bool TracingServiceImpl::DetachConsumer(ConsumerEndpointImpl* consumer,
511 const std::string& key) {
512 PERFETTO_DCHECK_THREAD(thread_checker_);
513 PERFETTO_DLOG("Consumer %p detached", reinterpret_cast<void*>(consumer));
514
515 TracingSessionID tsid = consumer->tracing_session_id_;
516 TracingSession* tracing_session;
517 if (!tsid || !(tracing_session = GetTracingSession(tsid)))
518 return false;
519
520 if (GetDetachedSession(consumer->uid_, key)) {
521 PERFETTO_ELOG("Another session has been detached with the same key \"%s\"",
522 key.c_str());
523 return false;
524 }
525
526 PERFETTO_DCHECK(tracing_session->consumer_maybe_null == consumer);
527 tracing_session->consumer_maybe_null = nullptr;
528 tracing_session->detach_key = key;
529 consumer->tracing_session_id_ = 0;
530 return true;
531 }
532
533 std::unique_ptr<TracingService::RelayEndpoint>
ConnectRelayClient(RelayClientID relay_client_id)534 TracingServiceImpl::ConnectRelayClient(RelayClientID relay_client_id) {
535 PERFETTO_DCHECK_THREAD(thread_checker_);
536
537 auto endpoint = std::make_unique<RelayEndpointImpl>(relay_client_id, this);
538 relay_clients_[relay_client_id] = endpoint.get();
539
540 return std::move(endpoint);
541 }
542
DisconnectRelayClient(RelayClientID relay_client_id)543 void TracingServiceImpl::DisconnectRelayClient(RelayClientID relay_client_id) {
544 PERFETTO_DCHECK_THREAD(thread_checker_);
545
546 if (relay_clients_.find(relay_client_id) == relay_clients_.end())
547 return;
548 relay_clients_.erase(relay_client_id);
549 }
550
AttachConsumer(ConsumerEndpointImpl * consumer,const std::string & key)551 bool TracingServiceImpl::AttachConsumer(ConsumerEndpointImpl* consumer,
552 const std::string& key) {
553 PERFETTO_DCHECK_THREAD(thread_checker_);
554 PERFETTO_DLOG("Consumer %p attaching to session %s",
555 reinterpret_cast<void*>(consumer), key.c_str());
556
557 if (consumer->tracing_session_id_) {
558 PERFETTO_ELOG(
559 "Cannot reattach consumer to session %s"
560 " while it already attached tracing session ID %" PRIu64,
561 key.c_str(), consumer->tracing_session_id_);
562 return false;
563 }
564
565 auto* tracing_session = GetDetachedSession(consumer->uid_, key);
566 if (!tracing_session) {
567 PERFETTO_ELOG(
568 "Failed to attach consumer, session '%s' not found for uid %d",
569 key.c_str(), static_cast<int>(consumer->uid_));
570 return false;
571 }
572
573 consumer->tracing_session_id_ = tracing_session->id;
574 tracing_session->consumer_maybe_null = consumer;
575 tracing_session->detach_key.clear();
576 return true;
577 }
578
EnableTracing(ConsumerEndpointImpl * consumer,const TraceConfig & cfg,base::ScopedFile fd)579 base::Status TracingServiceImpl::EnableTracing(ConsumerEndpointImpl* consumer,
580 const TraceConfig& cfg,
581 base::ScopedFile fd) {
582 PERFETTO_DCHECK_THREAD(thread_checker_);
583
584 // If the producer is specifying a UUID, respect that (at least for the first
585 // snapshot). Otherwise generate a new UUID.
586 base::Uuid uuid(cfg.trace_uuid_lsb(), cfg.trace_uuid_msb());
587 if (!uuid)
588 uuid = base::Uuidv4();
589
590 PERFETTO_DLOG("Enabling tracing for consumer %p, UUID: %s",
591 reinterpret_cast<void*>(consumer),
592 uuid.ToPrettyString().c_str());
593 MaybeLogUploadEvent(cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracing);
594 if (cfg.lockdown_mode() == TraceConfig::LOCKDOWN_SET)
595 lockdown_mode_ = true;
596 if (cfg.lockdown_mode() == TraceConfig::LOCKDOWN_CLEAR)
597 lockdown_mode_ = false;
598
599 // Scope |tracing_session| to this block to prevent accidental use of a null
600 // pointer later in this function.
601 {
602 TracingSession* tracing_session =
603 GetTracingSession(consumer->tracing_session_id_);
604 if (tracing_session) {
605 MaybeLogUploadEvent(
606 cfg, uuid,
607 PerfettoStatsdAtom::kTracedEnableTracingExistingTraceSession);
608 return PERFETTO_SVC_ERR(
609 "A Consumer is trying to EnableTracing() but another tracing "
610 "session is already active (forgot a call to FreeBuffers() ?)");
611 }
612 }
613
614 const uint32_t max_duration_ms = cfg.enable_extra_guardrails()
615 ? kGuardrailsMaxTracingDurationMillis
616 : kMaxTracingDurationMillis;
617 if (cfg.duration_ms() > max_duration_ms) {
618 MaybeLogUploadEvent(cfg, uuid,
619 PerfettoStatsdAtom::kTracedEnableTracingTooLongTrace);
620 return PERFETTO_SVC_ERR("Requested too long trace (%" PRIu32
621 "ms > %" PRIu32 " ms)",
622 cfg.duration_ms(), max_duration_ms);
623 }
624
625 const bool has_trigger_config =
626 GetTriggerMode(cfg) != TraceConfig::TriggerConfig::UNSPECIFIED;
627 if (has_trigger_config &&
628 (cfg.trigger_config().trigger_timeout_ms() == 0 ||
629 cfg.trigger_config().trigger_timeout_ms() > max_duration_ms)) {
630 MaybeLogUploadEvent(
631 cfg, uuid,
632 PerfettoStatsdAtom::kTracedEnableTracingInvalidTriggerTimeout);
633 return PERFETTO_SVC_ERR(
634 "Traces with START_TRACING triggers must provide a positive "
635 "trigger_timeout_ms < 7 days (received %" PRIu32 "ms)",
636 cfg.trigger_config().trigger_timeout_ms());
637 }
638
639 // This check has been introduced in May 2023 after finding b/274931668.
640 if (static_cast<int>(cfg.trigger_config().trigger_mode()) >
641 TraceConfig::TriggerConfig::TriggerMode_MAX) {
642 MaybeLogUploadEvent(
643 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingInvalidTriggerMode);
644 return PERFETTO_SVC_ERR(
645 "The trace config specified an invalid trigger_mode");
646 }
647
648 if (cfg.trigger_config().use_clone_snapshot_if_available() &&
649 cfg.trigger_config().trigger_mode() !=
650 TraceConfig::TriggerConfig::STOP_TRACING) {
651 MaybeLogUploadEvent(
652 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingInvalidTriggerMode);
653 return PERFETTO_SVC_ERR(
654 "trigger_mode must be STOP_TRACING when "
655 "use_clone_snapshot_if_available=true");
656 }
657
658 if (has_trigger_config && cfg.duration_ms() != 0) {
659 MaybeLogUploadEvent(
660 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingDurationWithTrigger);
661 return PERFETTO_SVC_ERR(
662 "duration_ms was set, this must not be set for traces with triggers.");
663 }
664
665 for (char c : cfg.bugreport_filename()) {
666 if (!((c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') ||
667 (c >= '0' && c <= '9') || c == '-' || c == '_' || c == '.')) {
668 MaybeLogUploadEvent(
669 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingInvalidBrFilename);
670 return PERFETTO_SVC_ERR(
671 "bugreport_filename contains invalid chars. Use [a-zA-Z0-9-_.]+");
672 }
673 }
674
675 if ((GetTriggerMode(cfg) == TraceConfig::TriggerConfig::STOP_TRACING ||
676 GetTriggerMode(cfg) == TraceConfig::TriggerConfig::CLONE_SNAPSHOT) &&
677 cfg.write_into_file()) {
678 // We don't support this usecase because there are subtle assumptions which
679 // break around TracingServiceEvents and windowed sorting (i.e. if we don't
680 // drain the events in ReadBuffersIntoFile because we are waiting for
681 // STOP_TRACING, we can end up queueing up a lot of TracingServiceEvents and
682 // emitting them wildy out of order breaking windowed sorting in trace
683 // processor).
684 MaybeLogUploadEvent(
685 cfg, uuid,
686 PerfettoStatsdAtom::kTracedEnableTracingStopTracingWriteIntoFile);
687 return PERFETTO_SVC_ERR(
688 "Specifying trigger mode STOP_TRACING/CLONE_SNAPSHOT and "
689 "write_into_file together is unsupported");
690 }
691
692 std::unordered_set<std::string> triggers;
693 for (const auto& trigger : cfg.trigger_config().triggers()) {
694 if (!triggers.insert(trigger.name()).second) {
695 MaybeLogUploadEvent(
696 cfg, uuid,
697 PerfettoStatsdAtom::kTracedEnableTracingDuplicateTriggerName);
698 return PERFETTO_SVC_ERR("Duplicate trigger name: %s",
699 trigger.name().c_str());
700 }
701 }
702
703 if (cfg.enable_extra_guardrails()) {
704 if (cfg.deferred_start()) {
705 MaybeLogUploadEvent(
706 cfg, uuid,
707 PerfettoStatsdAtom::kTracedEnableTracingInvalidDeferredStart);
708 return PERFETTO_SVC_ERR(
709 "deferred_start=true is not supported in unsupervised traces");
710 }
711 uint64_t buf_size_sum = 0;
712 for (const auto& buf : cfg.buffers()) {
713 if (buf.size_kb() % 4 != 0) {
714 MaybeLogUploadEvent(
715 cfg, uuid,
716 PerfettoStatsdAtom::kTracedEnableTracingInvalidBufferSize);
717 return PERFETTO_SVC_ERR(
718 "buffers.size_kb must be a multiple of 4, got %" PRIu32,
719 buf.size_kb());
720 }
721 buf_size_sum += buf.size_kb();
722 }
723
724 uint32_t max_tracing_buffer_size_kb =
725 std::max(kGuardrailsMaxTracingBufferSizeKb,
726 cfg.guardrail_overrides().max_tracing_buffer_size_kb());
727 if (buf_size_sum > max_tracing_buffer_size_kb) {
728 MaybeLogUploadEvent(
729 cfg, uuid,
730 PerfettoStatsdAtom::kTracedEnableTracingBufferSizeTooLarge);
731 return PERFETTO_SVC_ERR("Requested too large trace buffer (%" PRIu64
732 "kB > %" PRIu32 " kB)",
733 buf_size_sum, max_tracing_buffer_size_kb);
734 }
735 }
736
737 if (cfg.buffers_size() > kMaxBuffersPerConsumer) {
738 MaybeLogUploadEvent(cfg, uuid,
739 PerfettoStatsdAtom::kTracedEnableTracingTooManyBuffers);
740 return PERFETTO_SVC_ERR("Too many buffers configured (%d)",
741 cfg.buffers_size());
742 }
743 // Check that the config specifies all buffers for its data sources. This
744 // is also checked in SetupDataSource, but it is simpler to return a proper
745 // error to the consumer from here (and there will be less state to undo).
746 for (const TraceConfig::DataSource& cfg_data_source : cfg.data_sources()) {
747 size_t num_buffers = static_cast<size_t>(cfg.buffers_size());
748 size_t target_buffer = cfg_data_source.config().target_buffer();
749 if (target_buffer >= num_buffers) {
750 MaybeLogUploadEvent(
751 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingOobTargetBuffer);
752 return PERFETTO_SVC_ERR(
753 "Data source \"%s\" specified an out of bounds target_buffer (%zu >= "
754 "%zu)",
755 cfg_data_source.config().name().c_str(), target_buffer, num_buffers);
756 }
757 }
758
759 if (!cfg.unique_session_name().empty()) {
760 const std::string& name = cfg.unique_session_name();
761 for (auto& kv : tracing_sessions_) {
762 if (kv.second.state == TracingSession::CLONED_READ_ONLY)
763 continue; // Don't consider cloned sessions in uniqueness checks.
764 if (kv.second.config.unique_session_name() == name) {
765 MaybeLogUploadEvent(
766 cfg, uuid,
767 PerfettoStatsdAtom::kTracedEnableTracingDuplicateSessionName);
768 static const char fmt[] =
769 "A trace with this unique session name (%s) already exists";
770 // This happens frequently, don't make it an "E"LOG.
771 PERFETTO_LOG(fmt, name.c_str());
772 return base::ErrStatus(fmt, name.c_str());
773 }
774 }
775 }
776
777 if (!cfg.session_semaphores().empty()) {
778 struct SemaphoreSessionsState {
779 uint64_t smallest_max_other_session_count =
780 std::numeric_limits<uint64_t>::max();
781 uint64_t session_count = 0;
782 };
783 // For each semaphore, compute the number of active sessions and the
784 // MIN(limit).
785 std::unordered_map<std::string, SemaphoreSessionsState>
786 sem_to_sessions_state;
787 for (const auto& id_and_session : tracing_sessions_) {
788 const auto& session = id_and_session.second;
789 if (session.state == TracingSession::CLONED_READ_ONLY ||
790 session.state == TracingSession::DISABLED) {
791 // Don't consider cloned or disabled sessions in checks.
792 continue;
793 }
794 for (const auto& sem : session.config.session_semaphores()) {
795 auto& sessions_state = sem_to_sessions_state[sem.name()];
796 sessions_state.smallest_max_other_session_count =
797 std::min(sessions_state.smallest_max_other_session_count,
798 sem.max_other_session_count());
799 sessions_state.session_count++;
800 }
801 }
802
803 // Check if any of the semaphores declared by the config clashes with any of
804 // the currently active semaphores.
805 for (const auto& semaphore : cfg.session_semaphores()) {
806 auto it = sem_to_sessions_state.find(semaphore.name());
807 if (it == sem_to_sessions_state.end()) {
808 continue;
809 }
810 uint64_t max_other_session_count =
811 std::min(semaphore.max_other_session_count(),
812 it->second.smallest_max_other_session_count);
813 if (it->second.session_count > max_other_session_count) {
814 MaybeLogUploadEvent(
815 cfg, uuid,
816 PerfettoStatsdAtom::
817 kTracedEnableTracingFailedSessionSemaphoreCheck);
818 return PERFETTO_SVC_ERR(
819 "Semaphore \"%s\" exceeds maximum allowed other session count "
820 "(%" PRIu64 " > min(%" PRIu64 ", %" PRIu64 "))",
821 semaphore.name().c_str(), it->second.session_count,
822 semaphore.max_other_session_count(),
823 it->second.smallest_max_other_session_count);
824 }
825 }
826 }
827
828 if (cfg.enable_extra_guardrails()) {
829 // unique_session_name can be empty
830 const std::string& name = cfg.unique_session_name();
831 int64_t now_s = clock_->GetBootTimeS().count();
832
833 // Remove any entries where the time limit has passed so this map doesn't
834 // grow indefinitely:
835 std::map<std::string, int64_t>& sessions = session_to_last_trace_s_;
836 for (auto it = sessions.cbegin(); it != sessions.cend();) {
837 if (now_s - it->second > kMinSecondsBetweenTracesGuardrail) {
838 it = sessions.erase(it);
839 } else {
840 ++it;
841 }
842 }
843
844 int64_t& previous_s = session_to_last_trace_s_[name];
845 if (previous_s == 0) {
846 previous_s = now_s;
847 } else {
848 MaybeLogUploadEvent(
849 cfg, uuid,
850 PerfettoStatsdAtom::kTracedEnableTracingSessionNameTooRecent);
851 return PERFETTO_SVC_ERR(
852 "A trace with unique session name \"%s\" began less than %" PRId64
853 "s ago (%" PRId64 "s)",
854 name.c_str(), kMinSecondsBetweenTracesGuardrail, now_s - previous_s);
855 }
856 }
857
858 const int sessions_for_uid = static_cast<int>(std::count_if(
859 tracing_sessions_.begin(), tracing_sessions_.end(),
860 [consumer](const decltype(tracing_sessions_)::value_type& s) {
861 return s.second.consumer_uid == consumer->uid_;
862 }));
863
864 int per_uid_limit = kMaxConcurrentTracingSessionsPerUid;
865 if (consumer->uid_ == 1066 /* AID_STATSD*/) {
866 per_uid_limit = kMaxConcurrentTracingSessionsForStatsdUid;
867 }
868 if (sessions_for_uid >= per_uid_limit) {
869 MaybeLogUploadEvent(
870 cfg, uuid,
871 PerfettoStatsdAtom::kTracedEnableTracingTooManySessionsForUid);
872 return PERFETTO_SVC_ERR(
873 "Too many concurrent tracing sesions (%d) for uid %d limit is %d",
874 sessions_for_uid, static_cast<int>(consumer->uid_), per_uid_limit);
875 }
876
877 // TODO(primiano): This is a workaround to prevent that a producer gets stuck
878 // in a state where it stalls by design by having more TraceWriterImpl
879 // instances than free pages in the buffer. This is really a bug in
880 // trace_probes and the way it handles stalls in the shmem buffer.
881 if (tracing_sessions_.size() >= kMaxConcurrentTracingSessions) {
882 MaybeLogUploadEvent(
883 cfg, uuid,
884 PerfettoStatsdAtom::kTracedEnableTracingTooManyConcurrentSessions);
885 return PERFETTO_SVC_ERR("Too many concurrent tracing sesions (%zu)",
886 tracing_sessions_.size());
887 }
888
889 // If the trace config provides a filter bytecode, setup the filter now.
890 // If the filter loading fails, abort the tracing session rather than running
891 // unfiltered.
892 std::unique_ptr<protozero::MessageFilter> trace_filter;
893 if (cfg.has_trace_filter()) {
894 const auto& filt = cfg.trace_filter();
895 trace_filter.reset(new protozero::MessageFilter());
896
897 protozero::StringFilter& string_filter = trace_filter->string_filter();
898 for (const auto& rule : filt.string_filter_chain().rules()) {
899 auto opt_policy = ConvertPolicy(rule.policy());
900 if (!opt_policy.has_value()) {
901 MaybeLogUploadEvent(
902 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingInvalidFilter);
903 return PERFETTO_SVC_ERR(
904 "Trace filter has invalid string filtering rules, aborting");
905 }
906 string_filter.AddRule(*opt_policy, rule.regex_pattern(),
907 rule.atrace_payload_starts_with());
908 }
909
910 const std::string& bytecode_v1 = filt.bytecode();
911 const std::string& bytecode_v2 = filt.bytecode_v2();
912 const std::string& bytecode =
913 bytecode_v2.empty() ? bytecode_v1 : bytecode_v2;
914 if (!trace_filter->LoadFilterBytecode(bytecode.data(), bytecode.size())) {
915 MaybeLogUploadEvent(
916 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingInvalidFilter);
917 return PERFETTO_SVC_ERR("Trace filter bytecode invalid, aborting");
918 }
919
920 // The filter is created using perfetto.protos.Trace as root message
921 // (because that makes it possible to play around with the `proto_filter`
922 // tool on actual traces). Here in the service, however, we deal with
923 // perfetto.protos.TracePacket(s), which are one level down (Trace.packet).
924 // The IPC client (or the write_into_filte logic in here) are responsible
925 // for pre-pending the packet preamble (See GetProtoPreamble() calls), but
926 // the preamble is not there at ReadBuffer time. Hence we change the root of
927 // the filtering to start at the Trace.packet level.
928 if (!trace_filter->SetFilterRoot({TracePacket::kPacketFieldNumber})) {
929 MaybeLogUploadEvent(
930 cfg, uuid, PerfettoStatsdAtom::kTracedEnableTracingInvalidFilter);
931 return PERFETTO_SVC_ERR("Failed to set filter root.");
932 }
933 }
934
935 const TracingSessionID tsid = ++last_tracing_session_id_;
936 TracingSession* tracing_session =
937 &tracing_sessions_
938 .emplace(std::piecewise_construct, std::forward_as_tuple(tsid),
939 std::forward_as_tuple(tsid, consumer, cfg,
940 weak_runner_.task_runner()))
941 .first->second;
942
943 tracing_session->trace_uuid = uuid;
944
945 if (trace_filter)
946 tracing_session->trace_filter = std::move(trace_filter);
947
948 if (cfg.write_into_file()) {
949 if (!fd ^ !cfg.output_path().empty()) {
950 MaybeLogUploadEvent(
951 tracing_session->config, uuid,
952 PerfettoStatsdAtom::kTracedEnableTracingInvalidFdOutputFile);
953 tracing_sessions_.erase(tsid);
954 return PERFETTO_SVC_ERR(
955 "When write_into_file==true either a FD needs to be passed or "
956 "output_path must be populated (but not both)");
957 }
958 if (!cfg.output_path().empty()) {
959 fd = CreateTraceFile(cfg.output_path(), /*overwrite=*/false);
960 if (!fd) {
961 MaybeLogUploadEvent(
962 tracing_session->config, uuid,
963 PerfettoStatsdAtom::kTracedEnableTracingFailedToCreateFile);
964 tracing_sessions_.erase(tsid);
965 return PERFETTO_SVC_ERR("Failed to create the trace file %s",
966 cfg.output_path().c_str());
967 }
968 }
969 tracing_session->write_into_file = std::move(fd);
970 uint32_t write_period_ms = cfg.file_write_period_ms();
971 if (write_period_ms == 0)
972 write_period_ms = kDefaultWriteIntoFilePeriodMs;
973 if (write_period_ms < kMinWriteIntoFilePeriodMs)
974 write_period_ms = kMinWriteIntoFilePeriodMs;
975 tracing_session->write_period_ms = write_period_ms;
976 tracing_session->max_file_size_bytes = cfg.max_file_size_bytes();
977 tracing_session->bytes_written_into_file = 0;
978 }
979
980 if (cfg.compression_type() == TraceConfig::COMPRESSION_TYPE_DEFLATE) {
981 if (init_opts_.compressor_fn) {
982 tracing_session->compress_deflate = true;
983 } else {
984 PERFETTO_LOG(
985 "COMPRESSION_TYPE_DEFLATE is not supported in the current build "
986 "configuration. Skipping compression");
987 }
988 }
989
990 // Initialize the log buffers.
991 bool did_allocate_all_buffers = true;
992 bool invalid_buffer_config = false;
993
994 // Allocate the trace buffers. Also create a map to translate a consumer
995 // relative index (TraceConfig.DataSourceConfig.target_buffer) into the
996 // corresponding BufferID, which is a global ID namespace for the service and
997 // all producers.
998 size_t total_buf_size_kb = 0;
999 const size_t num_buffers = static_cast<size_t>(cfg.buffers_size());
1000 tracing_session->buffers_index.reserve(num_buffers);
1001 for (size_t i = 0; i < num_buffers; i++) {
1002 const TraceConfig::BufferConfig& buffer_cfg = cfg.buffers()[i];
1003 BufferID global_id = buffer_ids_.Allocate();
1004 if (!global_id) {
1005 did_allocate_all_buffers = false; // We ran out of IDs.
1006 break;
1007 }
1008 tracing_session->buffers_index.push_back(global_id);
1009 // TraceBuffer size is limited to 32-bit.
1010 const uint32_t buf_size_kb = buffer_cfg.size_kb();
1011 const uint64_t buf_size_bytes = buf_size_kb * static_cast<uint64_t>(1024);
1012 const size_t buf_size = static_cast<size_t>(buf_size_bytes);
1013 if (buf_size_bytes == 0 ||
1014 buf_size_bytes > std::numeric_limits<uint32_t>::max() ||
1015 buf_size != buf_size_bytes) {
1016 invalid_buffer_config = true;
1017 did_allocate_all_buffers = false;
1018 break;
1019 }
1020 total_buf_size_kb += buf_size_kb;
1021 TraceBuffer::OverwritePolicy policy =
1022 buffer_cfg.fill_policy() == TraceConfig::BufferConfig::DISCARD
1023 ? TraceBuffer::kDiscard
1024 : TraceBuffer::kOverwrite;
1025 auto it_and_inserted =
1026 buffers_.emplace(global_id, TraceBuffer::Create(buf_size, policy));
1027 PERFETTO_DCHECK(it_and_inserted.second); // buffers_.count(global_id) == 0.
1028 std::unique_ptr<TraceBuffer>& trace_buffer = it_and_inserted.first->second;
1029 if (!trace_buffer) {
1030 did_allocate_all_buffers = false;
1031 break;
1032 }
1033 }
1034
1035 // This can happen if either:
1036 // - All the kMaxTraceBufferID slots are taken.
1037 // - OOM, or, more realistically, we exhausted virtual memory.
1038 // - The buffer size in the config is invalid.
1039 // In any case, free all the previously allocated buffers and abort.
1040 if (!did_allocate_all_buffers) {
1041 for (BufferID global_id : tracing_session->buffers_index) {
1042 buffer_ids_.Free(global_id);
1043 buffers_.erase(global_id);
1044 }
1045 MaybeLogUploadEvent(tracing_session->config, uuid,
1046 PerfettoStatsdAtom::kTracedEnableTracingOom);
1047 tracing_sessions_.erase(tsid);
1048 if (invalid_buffer_config) {
1049 return PERFETTO_SVC_ERR(
1050 "Failed to allocate tracing buffers: Invalid buffer sizes");
1051 }
1052 return PERFETTO_SVC_ERR(
1053 "Failed to allocate tracing buffers: OOM or too many buffers");
1054 }
1055
1056 UpdateMemoryGuardrail();
1057
1058 consumer->tracing_session_id_ = tsid;
1059
1060 // Setup the data sources on the producers without starting them.
1061 for (const TraceConfig::DataSource& cfg_data_source : cfg.data_sources()) {
1062 // Scan all the registered data sources with a matching name.
1063 auto range = data_sources_.equal_range(cfg_data_source.config().name());
1064 for (auto it = range.first; it != range.second; it++) {
1065 TraceConfig::ProducerConfig producer_config;
1066 for (const auto& config : cfg.producers()) {
1067 if (GetProducer(it->second.producer_id)->name_ ==
1068 config.producer_name()) {
1069 producer_config = config;
1070 break;
1071 }
1072 }
1073 SetupDataSource(cfg_data_source, producer_config, it->second,
1074 tracing_session);
1075 }
1076 }
1077
1078 bool has_start_trigger = false;
1079 switch (GetTriggerMode(cfg)) {
1080 case TraceConfig::TriggerConfig::UNSPECIFIED:
1081 // no triggers are specified so this isn't a trace that is using triggers.
1082 PERFETTO_DCHECK(!has_trigger_config);
1083 break;
1084 case TraceConfig::TriggerConfig::START_TRACING:
1085 // For traces which use START_TRACE triggers we need to ensure that the
1086 // tracing session will be cleaned up when it times out.
1087 has_start_trigger = true;
1088 weak_runner_.PostDelayedTask(
1089 [tsid, this]() { OnStartTriggersTimeout(tsid); },
1090 cfg.trigger_config().trigger_timeout_ms());
1091 break;
1092 case TraceConfig::TriggerConfig::STOP_TRACING:
1093 case TraceConfig::TriggerConfig::CLONE_SNAPSHOT:
1094 // Update the tracing_session's duration_ms to ensure that if no trigger
1095 // is received the session will end and be cleaned up equal to the
1096 // timeout.
1097 //
1098 // TODO(nuskos): Refactor this so that rather then modifying the config we
1099 // have a field we look at on the tracing_session.
1100 tracing_session->config.set_duration_ms(
1101 cfg.trigger_config().trigger_timeout_ms());
1102 break;
1103
1104 // The case of unknown modes (coming from future versions of the service)
1105 // is handled few lines above (search for TriggerMode_MAX).
1106 }
1107
1108 tracing_session->state = TracingSession::CONFIGURED;
1109 PERFETTO_LOG(
1110 "Configured tracing session %" PRIu64
1111 ", #sources:%zu, duration:%u ms%s, #buffers:%d, total "
1112 "buffer size:%zu KB, total sessions:%zu, uid:%u session name: \"%s\"",
1113 tsid, cfg.data_sources().size(), tracing_session->config.duration_ms(),
1114 tracing_session->config.prefer_suspend_clock_for_duration()
1115 ? " (suspend_clock)"
1116 : "",
1117 cfg.buffers_size(), total_buf_size_kb, tracing_sessions_.size(),
1118 static_cast<unsigned int>(consumer->uid_),
1119 cfg.unique_session_name().c_str());
1120
1121 // Start the data sources, unless this is a case of early setup + fast
1122 // triggering, either through TraceConfig.deferred_start or
1123 // TraceConfig.trigger_config(). If both are specified which ever one occurs
1124 // first will initiate the trace.
1125 if (!cfg.deferred_start() && !has_start_trigger)
1126 StartTracing(tsid);
1127
1128 return base::OkStatus();
1129 }
1130
ChangeTraceConfig(ConsumerEndpointImpl * consumer,const TraceConfig & updated_cfg)1131 void TracingServiceImpl::ChangeTraceConfig(ConsumerEndpointImpl* consumer,
1132 const TraceConfig& updated_cfg) {
1133 PERFETTO_DCHECK_THREAD(thread_checker_);
1134 TracingSession* tracing_session =
1135 GetTracingSession(consumer->tracing_session_id_);
1136 PERFETTO_DCHECK(tracing_session);
1137
1138 if ((tracing_session->state != TracingSession::STARTED) &&
1139 (tracing_session->state != TracingSession::CONFIGURED)) {
1140 PERFETTO_ELOG(
1141 "ChangeTraceConfig() was called for a tracing session which isn't "
1142 "running.");
1143 return;
1144 }
1145
1146 // We only support updating producer_name_{,regex}_filter (and pass-through
1147 // configs) for now; null out any changeable fields and make sure the rest are
1148 // identical.
1149 TraceConfig new_config_copy(updated_cfg);
1150 for (auto& ds_cfg : *new_config_copy.mutable_data_sources()) {
1151 ds_cfg.clear_producer_name_filter();
1152 ds_cfg.clear_producer_name_regex_filter();
1153 }
1154
1155 TraceConfig current_config_copy(tracing_session->config);
1156 for (auto& ds_cfg : *current_config_copy.mutable_data_sources()) {
1157 ds_cfg.clear_producer_name_filter();
1158 ds_cfg.clear_producer_name_regex_filter();
1159 }
1160
1161 if (new_config_copy != current_config_copy) {
1162 PERFETTO_LOG(
1163 "ChangeTraceConfig() was called with a config containing unsupported "
1164 "changes; only adding to the producer_name_{,regex}_filter is "
1165 "currently supported and will have an effect.");
1166 }
1167
1168 for (TraceConfig::DataSource& cfg_data_source :
1169 *tracing_session->config.mutable_data_sources()) {
1170 // Find the updated producer_filter in the new config.
1171 std::vector<std::string> new_producer_name_filter;
1172 std::vector<std::string> new_producer_name_regex_filter;
1173 bool found_data_source = false;
1174 for (const auto& it : updated_cfg.data_sources()) {
1175 if (cfg_data_source.config().name() == it.config().name()) {
1176 new_producer_name_filter = it.producer_name_filter();
1177 new_producer_name_regex_filter = it.producer_name_regex_filter();
1178 found_data_source = true;
1179 break;
1180 }
1181 }
1182
1183 // Bail out if data source not present in the new config.
1184 if (!found_data_source) {
1185 PERFETTO_ELOG(
1186 "ChangeTraceConfig() called without a current data source also "
1187 "present in the new config: %s",
1188 cfg_data_source.config().name().c_str());
1189 continue;
1190 }
1191
1192 // TODO(oysteine): Just replacing the filter means that if
1193 // there are any filter entries which were present in the original config,
1194 // but removed from the config passed to ChangeTraceConfig, any matching
1195 // producers will keep producing but newly added producers after this
1196 // point will never start.
1197 *cfg_data_source.mutable_producer_name_filter() = new_producer_name_filter;
1198 *cfg_data_source.mutable_producer_name_regex_filter() =
1199 new_producer_name_regex_filter;
1200
1201 // Get the list of producers that are already set up.
1202 std::unordered_set<uint16_t> set_up_producers;
1203 auto& ds_instances = tracing_session->data_source_instances;
1204 for (auto instance_it = ds_instances.begin();
1205 instance_it != ds_instances.end(); ++instance_it) {
1206 set_up_producers.insert(instance_it->first);
1207 }
1208
1209 // Scan all the registered data sources with a matching name.
1210 auto range = data_sources_.equal_range(cfg_data_source.config().name());
1211 for (auto it = range.first; it != range.second; it++) {
1212 ProducerEndpointImpl* producer = GetProducer(it->second.producer_id);
1213 PERFETTO_DCHECK(producer);
1214
1215 // Check if the producer name of this data source is present
1216 // in the name filters. We currently only support new filters, not
1217 // removing old ones.
1218 if (!NameMatchesFilter(producer->name_, new_producer_name_filter,
1219 new_producer_name_regex_filter)) {
1220 continue;
1221 }
1222
1223 // If this producer is already set up, we assume that all datasources
1224 // in it started already.
1225 if (set_up_producers.count(it->second.producer_id))
1226 continue;
1227
1228 // If it wasn't previously setup, set it up now.
1229 // (The per-producer config is optional).
1230 TraceConfig::ProducerConfig producer_config;
1231 for (const auto& config : tracing_session->config.producers()) {
1232 if (producer->name_ == config.producer_name()) {
1233 producer_config = config;
1234 break;
1235 }
1236 }
1237
1238 DataSourceInstance* ds_inst = SetupDataSource(
1239 cfg_data_source, producer_config, it->second, tracing_session);
1240
1241 if (ds_inst && tracing_session->state == TracingSession::STARTED)
1242 StartDataSourceInstance(producer, tracing_session, ds_inst);
1243 }
1244 }
1245 }
1246
DelayToNextWritePeriodMs(const TracingSession & session)1247 uint32_t TracingServiceImpl::DelayToNextWritePeriodMs(
1248 const TracingSession& session) {
1249 PERFETTO_DCHECK(session.write_period_ms > 0);
1250 return session.write_period_ms -
1251 static_cast<uint32_t>(clock_->GetWallTimeMs().count() %
1252 session.write_period_ms);
1253 }
1254
StartTracing(TracingSessionID tsid)1255 void TracingServiceImpl::StartTracing(TracingSessionID tsid) {
1256 PERFETTO_DCHECK_THREAD(thread_checker_);
1257
1258 TracingSession* tracing_session = GetTracingSession(tsid);
1259 if (!tracing_session) {
1260 PERFETTO_ELOG("StartTracing() failed, invalid session ID %" PRIu64, tsid);
1261 return;
1262 }
1263
1264 MaybeLogUploadEvent(tracing_session->config, tracing_session->trace_uuid,
1265 PerfettoStatsdAtom::kTracedStartTracing);
1266
1267 if (tracing_session->state != TracingSession::CONFIGURED) {
1268 MaybeLogUploadEvent(
1269 tracing_session->config, tracing_session->trace_uuid,
1270 PerfettoStatsdAtom::kTracedStartTracingInvalidSessionState);
1271 PERFETTO_ELOG("StartTracing() failed, invalid session state: %d",
1272 tracing_session->state);
1273 return;
1274 }
1275
1276 tracing_session->state = TracingSession::STARTED;
1277
1278 // We store the start of trace snapshot separately as it's important to make
1279 // sure we can interpret all the data in the trace and storing it in the ring
1280 // buffer means it could be overwritten by a later snapshot.
1281 if (!tracing_session->config.builtin_data_sources()
1282 .disable_clock_snapshotting()) {
1283 SnapshotClocks(&tracing_session->initial_clock_snapshot);
1284 }
1285
1286 // We don't snapshot the clocks here because we just did this above.
1287 SnapshotLifecycleEvent(
1288 tracing_session,
1289 protos::pbzero::TracingServiceEvent::kTracingStartedFieldNumber,
1290 false /* snapshot_clocks */);
1291
1292 // Periodically snapshot clocks, stats, sync markers while the trace is
1293 // active. The snapshots are emitted on the future ReadBuffers() calls, which
1294 // means that:
1295 // (a) If we're streaming to a file (or to a consumer) while tracing, we
1296 // write snapshots periodically into the trace.
1297 // (b) If ReadBuffers() is only called after tracing ends, we emit the latest
1298 // snapshot into the trace. For clock snapshots, we keep track of the
1299 // snapshot recorded at the beginning of the session
1300 // (initial_clock_snapshot above), as well as the most recent sampled
1301 // snapshots that showed significant new drift between different clocks.
1302 // The latter clock snapshots are sampled periodically and at lifecycle
1303 // events.
1304 base::PeriodicTask::Args snapshot_task_args;
1305 snapshot_task_args.start_first_task_immediately = true;
1306 snapshot_task_args.use_suspend_aware_timer =
1307 tracing_session->config.builtin_data_sources()
1308 .prefer_suspend_clock_for_snapshot();
1309 snapshot_task_args.task = [this, tsid] { PeriodicSnapshotTask(tsid); };
1310 snapshot_task_args.period_ms =
1311 tracing_session->config.builtin_data_sources().snapshot_interval_ms();
1312 if (!snapshot_task_args.period_ms)
1313 snapshot_task_args.period_ms = kDefaultSnapshotsIntervalMs;
1314 tracing_session->snapshot_periodic_task.Start(snapshot_task_args);
1315
1316 // Trigger delayed task if the trace is time limited.
1317 const uint32_t trace_duration_ms = tracing_session->config.duration_ms();
1318 if (trace_duration_ms > 0) {
1319 auto stop_task =
1320 std::bind(&TracingServiceImpl::StopOnDurationMsExpiry, this, tsid);
1321 if (tracing_session->config.prefer_suspend_clock_for_duration()) {
1322 base::PeriodicTask::Args stop_args;
1323 stop_args.use_suspend_aware_timer = true;
1324 stop_args.period_ms = trace_duration_ms;
1325 stop_args.one_shot = true;
1326 stop_args.task = std::move(stop_task);
1327 tracing_session->timed_stop_task.Start(stop_args);
1328 } else {
1329 weak_runner_.PostDelayedTask(std::move(stop_task), trace_duration_ms);
1330 }
1331 } // if (trace_duration_ms > 0).
1332
1333 // Start the periodic drain tasks if we should to save the trace into a file.
1334 if (tracing_session->config.write_into_file()) {
1335 weak_runner_.PostDelayedTask([this, tsid] { ReadBuffersIntoFile(tsid); },
1336 DelayToNextWritePeriodMs(*tracing_session));
1337 }
1338
1339 // Start the periodic flush tasks if the config specified a flush period.
1340 if (tracing_session->config.flush_period_ms())
1341 PeriodicFlushTask(tsid, /*post_next_only=*/true);
1342
1343 // Start the periodic incremental state clear tasks if the config specified a
1344 // period.
1345 if (tracing_session->config.incremental_state_config().clear_period_ms()) {
1346 PeriodicClearIncrementalStateTask(tsid, /*post_next_only=*/true);
1347 }
1348
1349 for (auto& [prod_id, data_source] : tracing_session->data_source_instances) {
1350 ProducerEndpointImpl* producer = GetProducer(prod_id);
1351 if (!producer) {
1352 PERFETTO_DFATAL("Producer does not exist.");
1353 continue;
1354 }
1355 StartDataSourceInstance(producer, tracing_session, &data_source);
1356 }
1357
1358 MaybeNotifyAllDataSourcesStarted(tracing_session);
1359
1360 // `did_notify_all_data_source_started` is only set if a consumer is
1361 // connected.
1362 if (tracing_session->consumer_maybe_null) {
1363 weak_runner_.PostDelayedTask(
1364 [this, tsid] { OnAllDataSourceStartedTimeout(tsid); },
1365 kAllDataSourceStartedTimeout);
1366 }
1367 }
1368
StopOnDurationMsExpiry(TracingSessionID tsid)1369 void TracingServiceImpl::StopOnDurationMsExpiry(TracingSessionID tsid) {
1370 auto* tracing_session_ptr = GetTracingSession(tsid);
1371 if (!tracing_session_ptr)
1372 return;
1373 // If this trace was using STOP_TRACING triggers and we've seen
1374 // one, then the trigger overrides the normal timeout. In this
1375 // case we just return and let the other task clean up this trace.
1376 if (GetTriggerMode(tracing_session_ptr->config) ==
1377 TraceConfig::TriggerConfig::STOP_TRACING &&
1378 !tracing_session_ptr->received_triggers.empty())
1379 return;
1380 // In all other cases (START_TRACING or no triggers) we flush
1381 // after |trace_duration_ms| unconditionally.
1382 FlushAndDisableTracing(tsid);
1383 }
1384
StartDataSourceInstance(ProducerEndpointImpl * producer,TracingSession * tracing_session,TracingServiceImpl::DataSourceInstance * instance)1385 void TracingServiceImpl::StartDataSourceInstance(
1386 ProducerEndpointImpl* producer,
1387 TracingSession* tracing_session,
1388 TracingServiceImpl::DataSourceInstance* instance) {
1389 PERFETTO_DCHECK(instance->state == DataSourceInstance::CONFIGURED);
1390
1391 bool start_immediately = !instance->will_notify_on_start;
1392
1393 if (producer->IsAndroidProcessFrozen()) {
1394 PERFETTO_DLOG(
1395 "skipping waiting of data source \"%s\" on producer \"%s\" (pid=%u) "
1396 "because it is frozen",
1397 instance->data_source_name.c_str(), producer->name_.c_str(),
1398 producer->pid());
1399 start_immediately = true;
1400 }
1401
1402 if (!start_immediately) {
1403 instance->state = DataSourceInstance::STARTING;
1404 } else {
1405 instance->state = DataSourceInstance::STARTED;
1406 }
1407 if (tracing_session->consumer_maybe_null) {
1408 tracing_session->consumer_maybe_null->OnDataSourceInstanceStateChange(
1409 *producer, *instance);
1410 }
1411 producer->StartDataSource(instance->instance_id, instance->config);
1412
1413 // If all data sources are started, notify the consumer.
1414 if (instance->state == DataSourceInstance::STARTED)
1415 MaybeNotifyAllDataSourcesStarted(tracing_session);
1416 }
1417
1418 // DisableTracing just stops the data sources but doesn't free up any buffer.
1419 // This is to allow the consumer to freeze the buffers (by stopping the trace)
1420 // and then drain the buffers. The actual teardown of the TracingSession happens
1421 // in FreeBuffers().
DisableTracing(TracingSessionID tsid,bool disable_immediately)1422 void TracingServiceImpl::DisableTracing(TracingSessionID tsid,
1423 bool disable_immediately) {
1424 PERFETTO_DCHECK_THREAD(thread_checker_);
1425 TracingSession* tracing_session = GetTracingSession(tsid);
1426 if (!tracing_session) {
1427 // Can happen if the consumer calls this before EnableTracing() or after
1428 // FreeBuffers().
1429 PERFETTO_DLOG("DisableTracing() failed, invalid session ID %" PRIu64, tsid);
1430 return;
1431 }
1432
1433 MaybeLogUploadEvent(tracing_session->config, tracing_session->trace_uuid,
1434 PerfettoStatsdAtom::kTracedDisableTracing);
1435
1436 switch (tracing_session->state) {
1437 // Spurious call to DisableTracing() while already disabled, nothing to do.
1438 case TracingSession::DISABLED:
1439 PERFETTO_DCHECK(tracing_session->AllDataSourceInstancesStopped());
1440 return;
1441
1442 case TracingSession::CLONED_READ_ONLY:
1443 return;
1444
1445 // This is either:
1446 // A) The case of a graceful DisableTracing() call followed by a call to
1447 // FreeBuffers(), iff |disable_immediately| == true. In this case we want
1448 // to forcefully transition in the disabled state without waiting for the
1449 // outstanding acks because the buffers are going to be destroyed soon.
1450 // B) A spurious call, iff |disable_immediately| == false, in which case
1451 // there is nothing to do.
1452 case TracingSession::DISABLING_WAITING_STOP_ACKS:
1453 PERFETTO_DCHECK(!tracing_session->AllDataSourceInstancesStopped());
1454 if (disable_immediately)
1455 DisableTracingNotifyConsumerAndFlushFile(tracing_session);
1456 return;
1457
1458 // Continues below.
1459 case TracingSession::CONFIGURED:
1460 // If the session didn't even start there is no need to orchestrate a
1461 // graceful stop of data sources.
1462 disable_immediately = true;
1463 break;
1464
1465 // This is the nominal case, continues below.
1466 case TracingSession::STARTED:
1467 break;
1468 }
1469
1470 for (auto& data_source_inst : tracing_session->data_source_instances) {
1471 const ProducerID producer_id = data_source_inst.first;
1472 DataSourceInstance& instance = data_source_inst.second;
1473 ProducerEndpointImpl* producer = GetProducer(producer_id);
1474 PERFETTO_DCHECK(producer);
1475 PERFETTO_DCHECK(instance.state == DataSourceInstance::CONFIGURED ||
1476 instance.state == DataSourceInstance::STARTING ||
1477 instance.state == DataSourceInstance::STARTED);
1478 StopDataSourceInstance(producer, tracing_session, &instance,
1479 disable_immediately);
1480 }
1481
1482 // If the periodic task is running, we can stop the periodic snapshot timer
1483 // here instead of waiting until FreeBuffers to prevent useless snapshots
1484 // which won't be read.
1485 tracing_session->snapshot_periodic_task.Reset();
1486
1487 // Either this request is flagged with |disable_immediately| or there are no
1488 // data sources that are requesting a final handshake. In both cases just mark
1489 // the session as disabled immediately, notify the consumer and flush the
1490 // trace file (if used).
1491 if (tracing_session->AllDataSourceInstancesStopped())
1492 return DisableTracingNotifyConsumerAndFlushFile(tracing_session);
1493
1494 tracing_session->state = TracingSession::DISABLING_WAITING_STOP_ACKS;
1495 weak_runner_.PostDelayedTask([this, tsid] { OnDisableTracingTimeout(tsid); },
1496 tracing_session->data_source_stop_timeout_ms());
1497
1498 // Deliberately NOT removing the session from |tracing_session_|, it's still
1499 // needed to call ReadBuffers(). FreeBuffers() will erase() the session.
1500 }
1501
NotifyDataSourceStarted(ProducerID producer_id,DataSourceInstanceID instance_id)1502 void TracingServiceImpl::NotifyDataSourceStarted(
1503 ProducerID producer_id,
1504 DataSourceInstanceID instance_id) {
1505 PERFETTO_DCHECK_THREAD(thread_checker_);
1506 for (auto& kv : tracing_sessions_) {
1507 TracingSession& tracing_session = kv.second;
1508 DataSourceInstance* instance =
1509 tracing_session.GetDataSourceInstance(producer_id, instance_id);
1510
1511 if (!instance)
1512 continue;
1513
1514 // If the tracing session was already stopped, ignore this notification.
1515 if (tracing_session.state != TracingSession::STARTED)
1516 continue;
1517
1518 if (instance->state != DataSourceInstance::STARTING) {
1519 PERFETTO_ELOG("Started data source instance in incorrect state: %d",
1520 instance->state);
1521 continue;
1522 }
1523
1524 instance->state = DataSourceInstance::STARTED;
1525
1526 ProducerEndpointImpl* producer = GetProducer(producer_id);
1527 PERFETTO_DCHECK(producer);
1528 if (tracing_session.consumer_maybe_null) {
1529 tracing_session.consumer_maybe_null->OnDataSourceInstanceStateChange(
1530 *producer, *instance);
1531 }
1532
1533 // If all data sources are started, notify the consumer.
1534 MaybeNotifyAllDataSourcesStarted(&tracing_session);
1535 } // for (tracing_session)
1536 }
1537
OnAllDataSourceStartedTimeout(TracingSessionID tsid)1538 void TracingServiceImpl::OnAllDataSourceStartedTimeout(TracingSessionID tsid) {
1539 PERFETTO_DCHECK_THREAD(thread_checker_);
1540 TracingSession* tracing_session = GetTracingSession(tsid);
1541 // It would be possible to check for `AllDataSourceInstancesStarted()` here,
1542 // but it doesn't make much sense, because a data source can be registered
1543 // after the session has started. Therefore this is tied to
1544 // `did_notify_all_data_source_started`: if that notification happened, do not
1545 // record slow data sources.
1546 if (!tracing_session || !tracing_session->consumer_maybe_null ||
1547 tracing_session->did_notify_all_data_source_started) {
1548 return;
1549 }
1550
1551 int64_t timestamp = clock_->GetBootTimeNs().count();
1552
1553 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
1554 packet->set_timestamp(static_cast<uint64_t>(timestamp));
1555 packet->set_trusted_uid(static_cast<int32_t>(uid_));
1556 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
1557
1558 size_t i = 0;
1559 protos::pbzero::TracingServiceEvent::DataSources* slow_data_sources =
1560 packet->set_service_event()->set_slow_starting_data_sources();
1561 for (const auto& [producer_id, ds_instance] :
1562 tracing_session->data_source_instances) {
1563 if (ds_instance.state == DataSourceInstance::STARTED) {
1564 continue;
1565 }
1566 ProducerEndpointImpl* producer = GetProducer(producer_id);
1567 if (!producer) {
1568 continue;
1569 }
1570 if (++i > kMaxLifecycleEventsListedDataSources) {
1571 break;
1572 }
1573 auto* ds = slow_data_sources->add_data_source();
1574 ds->set_producer_name(producer->name_);
1575 ds->set_data_source_name(ds_instance.data_source_name);
1576 PERFETTO_LOG(
1577 "Data source failed to start within 20s data_source=\"%s\", "
1578 "producer=\"%s\", tsid=%" PRIu64,
1579 ds_instance.data_source_name.c_str(), producer->name_.c_str(), tsid);
1580 }
1581
1582 tracing_session->slow_start_event = TracingSession::ArbitraryLifecycleEvent{
1583 timestamp, packet.SerializeAsArray()};
1584 }
1585
MaybeNotifyAllDataSourcesStarted(TracingSession * tracing_session)1586 void TracingServiceImpl::MaybeNotifyAllDataSourcesStarted(
1587 TracingSession* tracing_session) {
1588 if (!tracing_session->consumer_maybe_null)
1589 return;
1590
1591 if (!tracing_session->AllDataSourceInstancesStarted())
1592 return;
1593
1594 // In some rare cases, we can get in this state more than once. Consider the
1595 // following scenario: 3 data sources are registered -> trace starts ->
1596 // all 3 data sources ack -> OnAllDataSourcesStarted() is called.
1597 // Imagine now that a 4th data source registers while the trace is ongoing.
1598 // This would hit the AllDataSourceInstancesStarted() condition again.
1599 // In this case, however, we don't want to re-notify the consumer again.
1600 // That would be unexpected (even if, perhaps, technically correct) and
1601 // trigger bugs in the consumer.
1602 if (tracing_session->did_notify_all_data_source_started)
1603 return;
1604
1605 PERFETTO_DLOG("All data sources started");
1606
1607 SnapshotLifecycleEvent(
1608 tracing_session,
1609 protos::pbzero::TracingServiceEvent::kAllDataSourcesStartedFieldNumber,
1610 true /* snapshot_clocks */);
1611
1612 tracing_session->did_notify_all_data_source_started = true;
1613 tracing_session->consumer_maybe_null->OnAllDataSourcesStarted();
1614 }
1615
NotifyDataSourceStopped(ProducerID producer_id,DataSourceInstanceID instance_id)1616 void TracingServiceImpl::NotifyDataSourceStopped(
1617 ProducerID producer_id,
1618 DataSourceInstanceID instance_id) {
1619 PERFETTO_DCHECK_THREAD(thread_checker_);
1620 for (auto& kv : tracing_sessions_) {
1621 TracingSession& tracing_session = kv.second;
1622 DataSourceInstance* instance =
1623 tracing_session.GetDataSourceInstance(producer_id, instance_id);
1624
1625 if (!instance)
1626 continue;
1627
1628 if (instance->state != DataSourceInstance::STOPPING) {
1629 PERFETTO_ELOG("Stopped data source instance in incorrect state: %d",
1630 instance->state);
1631 continue;
1632 }
1633
1634 instance->state = DataSourceInstance::STOPPED;
1635
1636 ProducerEndpointImpl* producer = GetProducer(producer_id);
1637 PERFETTO_DCHECK(producer);
1638 if (tracing_session.consumer_maybe_null) {
1639 tracing_session.consumer_maybe_null->OnDataSourceInstanceStateChange(
1640 *producer, *instance);
1641 }
1642
1643 if (!tracing_session.AllDataSourceInstancesStopped())
1644 continue;
1645
1646 if (tracing_session.state != TracingSession::DISABLING_WAITING_STOP_ACKS)
1647 continue;
1648
1649 // All data sources acked the termination.
1650 DisableTracingNotifyConsumerAndFlushFile(&tracing_session);
1651 } // for (tracing_session)
1652 }
1653
ActivateTriggers(ProducerID producer_id,const std::vector<std::string> & triggers)1654 void TracingServiceImpl::ActivateTriggers(
1655 ProducerID producer_id,
1656 const std::vector<std::string>& triggers) {
1657 PERFETTO_DCHECK_THREAD(thread_checker_);
1658 auto* producer = GetProducer(producer_id);
1659 PERFETTO_DCHECK(producer);
1660
1661 int64_t now_ns = clock_->GetBootTimeNs().count();
1662 for (const auto& trigger_name : triggers) {
1663 PERFETTO_DLOG("Received ActivateTriggers request for \"%s\"",
1664 trigger_name.c_str());
1665 android_stats::MaybeLogTriggerEvent(PerfettoTriggerAtom::kTracedTrigger,
1666 trigger_name);
1667
1668 base::Hasher hash;
1669 hash.Update(trigger_name.c_str(), trigger_name.size());
1670 std::string triggered_session_name;
1671 base::Uuid triggered_session_uuid;
1672 TracingSessionID triggered_session_id = 0;
1673 auto trigger_mode = TraceConfig::TriggerConfig::UNSPECIFIED;
1674
1675 uint64_t trigger_name_hash = hash.digest();
1676 size_t count_in_window =
1677 PurgeExpiredAndCountTriggerInWindow(now_ns, trigger_name_hash);
1678
1679 bool trigger_matched = false;
1680 bool trigger_activated = false;
1681 for (auto& id_and_tracing_session : tracing_sessions_) {
1682 auto& tracing_session = id_and_tracing_session.second;
1683 TracingSessionID tsid = id_and_tracing_session.first;
1684 auto iter = std::find_if(
1685 tracing_session.config.trigger_config().triggers().begin(),
1686 tracing_session.config.trigger_config().triggers().end(),
1687 [&trigger_name](const TraceConfig::TriggerConfig::Trigger& trigger) {
1688 return trigger.name() == trigger_name;
1689 });
1690 if (iter == tracing_session.config.trigger_config().triggers().end())
1691 continue;
1692 if (tracing_session.state == TracingSession::CLONED_READ_ONLY)
1693 continue;
1694
1695 // If this trigger requires a certain producer to have sent it
1696 // (non-empty producer_name()) ensure the producer who sent this trigger
1697 // matches.
1698 if (!iter->producer_name_regex().empty() &&
1699 !std::regex_match(
1700 producer->name_,
1701 std::regex(iter->producer_name_regex(), std::regex::extended))) {
1702 continue;
1703 }
1704
1705 // Use a random number between 0 and 1 to check if we should allow this
1706 // trigger through or not.
1707 double trigger_rnd = random_->GetValue();
1708 PERFETTO_DCHECK(trigger_rnd >= 0 && trigger_rnd < 1);
1709 if (trigger_rnd < iter->skip_probability()) {
1710 MaybeLogTriggerEvent(tracing_session.config,
1711 PerfettoTriggerAtom::kTracedLimitProbability,
1712 trigger_name);
1713 continue;
1714 }
1715
1716 // If we already triggered more times than the limit, silently ignore
1717 // this trigger.
1718 if (iter->max_per_24_h() > 0 && count_in_window >= iter->max_per_24_h()) {
1719 MaybeLogTriggerEvent(tracing_session.config,
1720 PerfettoTriggerAtom::kTracedLimitMaxPer24h,
1721 trigger_name);
1722 continue;
1723 }
1724 trigger_matched = true;
1725 triggered_session_id = tracing_session.id;
1726 triggered_session_name = tracing_session.config.unique_session_name();
1727 triggered_session_uuid.set_lsb_msb(tracing_session.trace_uuid.lsb(),
1728 tracing_session.trace_uuid.msb());
1729 trigger_mode = GetTriggerMode(tracing_session.config);
1730
1731 const bool triggers_already_received =
1732 !tracing_session.received_triggers.empty();
1733 const TriggerInfo trigger = {static_cast<uint64_t>(now_ns), iter->name(),
1734 producer->name_, producer->uid()};
1735 tracing_session.received_triggers.push_back(trigger);
1736 switch (trigger_mode) {
1737 case TraceConfig::TriggerConfig::START_TRACING:
1738 // If the session has already been triggered and moved past
1739 // CONFIGURED then we don't need to repeat StartTracing. This would
1740 // work fine (StartTracing would return false) but would add error
1741 // logs.
1742 if (tracing_session.state != TracingSession::CONFIGURED)
1743 break;
1744
1745 trigger_activated = true;
1746 MaybeLogUploadEvent(
1747 tracing_session.config, tracing_session.trace_uuid,
1748 PerfettoStatsdAtom::kTracedTriggerStartTracing, iter->name());
1749
1750 // We override the trace duration to be the trigger's requested
1751 // value, this ensures that the trace will end after this amount
1752 // of time has passed.
1753 tracing_session.config.set_duration_ms(iter->stop_delay_ms());
1754 StartTracing(tsid);
1755 break;
1756 case TraceConfig::TriggerConfig::STOP_TRACING:
1757 // Only stop the trace once to avoid confusing log messages. I.E.
1758 // when we've already hit the first trigger we've already Posted the
1759 // task to FlushAndDisable. So all future triggers will just break
1760 // out.
1761 if (triggers_already_received)
1762 break;
1763
1764 trigger_activated = true;
1765 MaybeLogUploadEvent(
1766 tracing_session.config, tracing_session.trace_uuid,
1767 PerfettoStatsdAtom::kTracedTriggerStopTracing, iter->name());
1768
1769 // Now that we've seen a trigger we need to stop, flush, and disable
1770 // this session after the configured |stop_delay_ms|.
1771 weak_runner_.PostDelayedTask(
1772 [this, tsid] {
1773 // Skip entirely the flush if the trace session doesn't exist
1774 // anymore. This is to prevent misleading error messages to be
1775 // logged.
1776 if (GetTracingSession(tsid))
1777 FlushAndDisableTracing(tsid);
1778 },
1779 // If this trigger is zero this will immediately executable and
1780 // will happen shortly.
1781 iter->stop_delay_ms());
1782 break;
1783
1784 case TraceConfig::TriggerConfig::CLONE_SNAPSHOT:
1785 trigger_activated = true;
1786 MaybeLogUploadEvent(
1787 tracing_session.config, tracing_session.trace_uuid,
1788 PerfettoStatsdAtom::kTracedTriggerCloneSnapshot, iter->name());
1789 weak_runner_.PostDelayedTask(
1790 [this, tsid, trigger] {
1791 auto* tsess = GetTracingSession(tsid);
1792 if (!tsess || !tsess->consumer_maybe_null)
1793 return;
1794 tsess->consumer_maybe_null->NotifyCloneSnapshotTrigger(trigger);
1795 },
1796 iter->stop_delay_ms());
1797 break;
1798
1799 case TraceConfig::TriggerConfig::UNSPECIFIED:
1800 PERFETTO_ELOG("Trigger activated but trigger mode unspecified.");
1801 break;
1802 }
1803 } // for (.. : tracing_sessions_)
1804
1805 if (trigger_matched) {
1806 trigger_history_.emplace_back(TriggerHistory{now_ns, trigger_name_hash});
1807 }
1808
1809 if (trigger_activated) {
1810 // Log only the trigger that actually caused a trace stop/start, don't log
1811 // the follow-up ones, even if they matched.
1812 PERFETTO_LOG(
1813 "Trace trigger activated: trigger_name=\"%s\" trigger_mode=%d "
1814 "trace_name=\"%s\" trace_uuid=\"%s\" tsid=%" PRIu64,
1815 trigger_name.c_str(), trigger_mode, triggered_session_name.c_str(),
1816 triggered_session_uuid.ToPrettyString().c_str(),
1817 triggered_session_id);
1818 }
1819 } // for (trigger_name : triggers)
1820 }
1821
1822 // Always invoked TraceConfig.data_source_stop_timeout_ms (by default
1823 // kDataSourceStopTimeoutMs) after DisableTracing(). In nominal conditions all
1824 // data sources should have acked the stop and this will early out.
OnDisableTracingTimeout(TracingSessionID tsid)1825 void TracingServiceImpl::OnDisableTracingTimeout(TracingSessionID tsid) {
1826 PERFETTO_DCHECK_THREAD(thread_checker_);
1827 TracingSession* tracing_session = GetTracingSession(tsid);
1828 if (!tracing_session ||
1829 tracing_session->state != TracingSession::DISABLING_WAITING_STOP_ACKS) {
1830 return; // Tracing session was successfully disabled.
1831 }
1832
1833 PERFETTO_ILOG("Timeout while waiting for ACKs for tracing session %" PRIu64,
1834 tsid);
1835 PERFETTO_DCHECK(!tracing_session->AllDataSourceInstancesStopped());
1836 DisableTracingNotifyConsumerAndFlushFile(tracing_session);
1837 }
1838
DisableTracingNotifyConsumerAndFlushFile(TracingSession * tracing_session)1839 void TracingServiceImpl::DisableTracingNotifyConsumerAndFlushFile(
1840 TracingSession* tracing_session) {
1841 PERFETTO_DCHECK(tracing_session->state != TracingSession::DISABLED);
1842 for (auto& inst_kv : tracing_session->data_source_instances) {
1843 if (inst_kv.second.state == DataSourceInstance::STOPPED)
1844 continue;
1845 inst_kv.second.state = DataSourceInstance::STOPPED;
1846 ProducerEndpointImpl* producer = GetProducer(inst_kv.first);
1847 PERFETTO_DCHECK(producer);
1848 if (tracing_session->consumer_maybe_null) {
1849 tracing_session->consumer_maybe_null->OnDataSourceInstanceStateChange(
1850 *producer, inst_kv.second);
1851 }
1852 }
1853 tracing_session->state = TracingSession::DISABLED;
1854
1855 // Scrape any remaining chunks that weren't flushed by the producers.
1856 for (auto& producer_id_and_producer : producers_)
1857 ScrapeSharedMemoryBuffers(tracing_session, producer_id_and_producer.second);
1858
1859 SnapshotLifecycleEvent(
1860 tracing_session,
1861 protos::pbzero::TracingServiceEvent::kTracingDisabledFieldNumber,
1862 true /* snapshot_clocks */);
1863
1864 if (tracing_session->write_into_file) {
1865 tracing_session->write_period_ms = 0;
1866 ReadBuffersIntoFile(tracing_session->id);
1867 }
1868
1869 MaybeLogUploadEvent(tracing_session->config, tracing_session->trace_uuid,
1870 PerfettoStatsdAtom::kTracedNotifyTracingDisabled);
1871
1872 if (tracing_session->consumer_maybe_null)
1873 tracing_session->consumer_maybe_null->NotifyOnTracingDisabled("");
1874 }
1875
Flush(TracingSessionID tsid,uint32_t timeout_ms,ConsumerEndpoint::FlushCallback callback,FlushFlags flush_flags)1876 void TracingServiceImpl::Flush(TracingSessionID tsid,
1877 uint32_t timeout_ms,
1878 ConsumerEndpoint::FlushCallback callback,
1879 FlushFlags flush_flags) {
1880 PERFETTO_DCHECK_THREAD(thread_checker_);
1881 TracingSession* tracing_session = GetTracingSession(tsid);
1882 if (!tracing_session) {
1883 PERFETTO_DLOG("Flush() failed, invalid session ID %" PRIu64, tsid);
1884 return;
1885 }
1886
1887 SnapshotLifecycleEvent(
1888 tracing_session,
1889 protos::pbzero::TracingServiceEvent::kFlushStartedFieldNumber,
1890 false /* snapshot_clocks */);
1891
1892 std::map<ProducerID, std::vector<DataSourceInstanceID>> data_source_instances;
1893 for (const auto& [producer_id, ds_inst] :
1894 tracing_session->data_source_instances) {
1895 if (ds_inst.no_flush)
1896 continue;
1897 data_source_instances[producer_id].push_back(ds_inst.instance_id);
1898 }
1899 FlushDataSourceInstances(tracing_session, timeout_ms, data_source_instances,
1900 std::move(callback), flush_flags);
1901 }
1902
FlushDataSourceInstances(TracingSession * tracing_session,uint32_t timeout_ms,const std::map<ProducerID,std::vector<DataSourceInstanceID>> & data_source_instances,ConsumerEndpoint::FlushCallback callback,FlushFlags flush_flags)1903 void TracingServiceImpl::FlushDataSourceInstances(
1904 TracingSession* tracing_session,
1905 uint32_t timeout_ms,
1906 const std::map<ProducerID, std::vector<DataSourceInstanceID>>&
1907 data_source_instances,
1908 ConsumerEndpoint::FlushCallback callback,
1909 FlushFlags flush_flags) {
1910 PERFETTO_DCHECK_THREAD(thread_checker_);
1911 if (!timeout_ms)
1912 timeout_ms = tracing_session->flush_timeout_ms();
1913
1914 if (tracing_session->pending_flushes.size() > 1000) {
1915 PERFETTO_ELOG("Too many flushes (%zu) pending for the tracing session",
1916 tracing_session->pending_flushes.size());
1917 callback(false);
1918 return;
1919 }
1920
1921 if (tracing_session->state != TracingSession::STARTED) {
1922 PERFETTO_LOG("Flush() called, but tracing has not been started");
1923 callback(false);
1924 return;
1925 }
1926
1927 tracing_session->last_flush_events.clear();
1928
1929 ++tracing_session->flushes_requested;
1930 FlushRequestID flush_request_id = ++last_flush_request_id_;
1931 PendingFlush& pending_flush =
1932 tracing_session->pending_flushes
1933 .emplace_hint(tracing_session->pending_flushes.end(),
1934 flush_request_id, PendingFlush(std::move(callback)))
1935 ->second;
1936
1937 // Send a flush request to each producer involved in the tracing session. In
1938 // order to issue a flush request we have to build a map of all data source
1939 // instance ids enabled for each producer.
1940
1941 for (const auto& [producer_id, data_sources] : data_source_instances) {
1942 ProducerEndpointImpl* producer = GetProducer(producer_id);
1943 producer->Flush(flush_request_id, data_sources, flush_flags);
1944 if (!producer->IsAndroidProcessFrozen()) {
1945 pending_flush.producers.insert(producer_id);
1946 } else {
1947 PERFETTO_DLOG(
1948 "skipping waiting flush for on producer \"%s\" (pid=%" PRIu32
1949 ") because it is frozen",
1950 producer->name_.c_str(), static_cast<uint32_t>(producer->pid()));
1951 }
1952 }
1953
1954 // If there are no producers to flush (realistically this happens only in
1955 // some tests) fire OnFlushTimeout() straight away, without waiting.
1956 if (data_source_instances.empty())
1957 timeout_ms = 0;
1958
1959 weak_runner_.PostDelayedTask(
1960 [this, tsid = tracing_session->id, flush_request_id, flush_flags] {
1961 OnFlushTimeout(tsid, flush_request_id, flush_flags);
1962 },
1963 timeout_ms);
1964 }
1965
NotifyFlushDoneForProducer(ProducerID producer_id,FlushRequestID flush_request_id)1966 void TracingServiceImpl::NotifyFlushDoneForProducer(
1967 ProducerID producer_id,
1968 FlushRequestID flush_request_id) {
1969 for (auto& kv : tracing_sessions_) {
1970 // Remove all pending flushes <= |flush_request_id| for |producer_id|.
1971 auto& pending_flushes = kv.second.pending_flushes;
1972 auto end_it = pending_flushes.upper_bound(flush_request_id);
1973 for (auto it = pending_flushes.begin(); it != end_it;) {
1974 PendingFlush& pending_flush = it->second;
1975 pending_flush.producers.erase(producer_id);
1976 if (pending_flush.producers.empty()) {
1977 TracingSessionID tsid = kv.first;
1978 auto callback = std::move(pending_flush.callback);
1979 weak_runner_.PostTask([this, tsid, callback = std::move(callback)]() {
1980 CompleteFlush(tsid, std::move(callback),
1981 /*success=*/true);
1982 });
1983 it = pending_flushes.erase(it);
1984 } else {
1985 it++;
1986 }
1987 } // for (pending_flushes)
1988 } // for (tracing_session)
1989 }
1990
OnFlushTimeout(TracingSessionID tsid,FlushRequestID flush_request_id,FlushFlags flush_flags)1991 void TracingServiceImpl::OnFlushTimeout(TracingSessionID tsid,
1992 FlushRequestID flush_request_id,
1993 FlushFlags flush_flags) {
1994 TracingSession* tracing_session = GetTracingSession(tsid);
1995 if (!tracing_session)
1996 return;
1997 auto it = tracing_session->pending_flushes.find(flush_request_id);
1998 if (it == tracing_session->pending_flushes.end())
1999 return; // Nominal case: flush was completed and acked on time.
2000
2001 PendingFlush& pending_flush = it->second;
2002
2003 // If there were no producers to flush, consider it a success.
2004 bool success = pending_flush.producers.empty();
2005 auto callback = std::move(pending_flush.callback);
2006 // If flush failed and this is a "final" flush, log which data sources were
2007 // slow.
2008 if ((flush_flags.reason() == FlushFlags::Reason::kTraceClone ||
2009 flush_flags.reason() == FlushFlags::Reason::kTraceStop) &&
2010 !success) {
2011 int64_t timestamp = clock_->GetBootTimeNs().count();
2012
2013 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
2014 packet->set_timestamp(static_cast<uint64_t>(timestamp));
2015 packet->set_trusted_uid(static_cast<int32_t>(uid_));
2016 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
2017
2018 size_t i = 0;
2019 protos::pbzero::TracingServiceEvent::DataSources* event =
2020 packet->set_service_event()->set_last_flush_slow_data_sources();
2021 for (const auto& producer_id : pending_flush.producers) {
2022 ProducerEndpointImpl* producer = GetProducer(producer_id);
2023 if (!producer) {
2024 continue;
2025 }
2026 if (++i > kMaxLifecycleEventsListedDataSources) {
2027 break;
2028 }
2029
2030 auto ds_id_range =
2031 tracing_session->data_source_instances.equal_range(producer_id);
2032 for (auto ds_it = ds_id_range.first; ds_it != ds_id_range.second;
2033 ds_it++) {
2034 auto* ds = event->add_data_source();
2035 ds->set_producer_name(producer->name_);
2036 ds->set_data_source_name(ds_it->second.data_source_name);
2037 if (++i > kMaxLifecycleEventsListedDataSources) {
2038 break;
2039 }
2040 }
2041 }
2042
2043 tracing_session->last_flush_events.push_back(
2044 {timestamp, packet.SerializeAsArray()});
2045 }
2046 tracing_session->pending_flushes.erase(it);
2047 CompleteFlush(tsid, std::move(callback), success);
2048 }
2049
CompleteFlush(TracingSessionID tsid,ConsumerEndpoint::FlushCallback callback,bool success)2050 void TracingServiceImpl::CompleteFlush(TracingSessionID tsid,
2051 ConsumerEndpoint::FlushCallback callback,
2052 bool success) {
2053 TracingSession* tracing_session = GetTracingSession(tsid);
2054 if (!tracing_session) {
2055 callback(false);
2056 return;
2057 }
2058 // Producers may not have been able to flush all their data, even if they
2059 // indicated flush completion. If possible, also collect uncommitted chunks
2060 // to make sure we have everything they wrote so far.
2061 for (auto& producer_id_and_producer : producers_) {
2062 ScrapeSharedMemoryBuffers(tracing_session, producer_id_and_producer.second);
2063 }
2064 SnapshotLifecycleEvent(
2065 tracing_session,
2066 protos::pbzero::TracingServiceEvent::kAllDataSourcesFlushedFieldNumber,
2067 true /* snapshot_clocks */);
2068
2069 tracing_session->flushes_succeeded += success ? 1 : 0;
2070 tracing_session->flushes_failed += success ? 0 : 1;
2071 callback(success);
2072 }
2073
ScrapeSharedMemoryBuffers(TracingSession * tracing_session,ProducerEndpointImpl * producer)2074 void TracingServiceImpl::ScrapeSharedMemoryBuffers(
2075 TracingSession* tracing_session,
2076 ProducerEndpointImpl* producer) {
2077 if (!producer->smb_scraping_enabled_)
2078 return;
2079
2080 // Can't copy chunks if we don't know about any trace writers.
2081 if (producer->writers_.empty())
2082 return;
2083
2084 // Performance optimization: On flush or session disconnect, this method is
2085 // called for each producer. If the producer doesn't participate in the
2086 // session, there's no need to scape its chunks right now. We can tell if a
2087 // producer participates in the session by checking if the producer is allowed
2088 // to write into the session's log buffers.
2089 const auto& session_buffers = tracing_session->buffers_index;
2090 bool producer_in_session =
2091 std::any_of(session_buffers.begin(), session_buffers.end(),
2092 [producer](BufferID buffer_id) {
2093 return producer->allowed_target_buffers_.count(buffer_id);
2094 });
2095 if (!producer_in_session)
2096 return;
2097
2098 PERFETTO_DLOG("Scraping SMB for producer %" PRIu16, producer->id_);
2099
2100 // Find and copy any uncommitted chunks from the SMB.
2101 //
2102 // In nominal conditions, the page header bitmap of the used SMB pages should
2103 // never change because the service is the only one who is supposed to modify
2104 // used pages (to make them free again).
2105 //
2106 // However, the code here needs to deal with the case of a malicious producer
2107 // altering the SMB in unpredictable ways. Thankfully the SMB size is
2108 // immutable, so a chunk will always point to some valid memory, even if the
2109 // producer alters the intended layout and chunk header concurrently.
2110 // Ultimately a malicious producer altering the SMB's chunk header bitamp
2111 // while we are iterating in this function is not any different from the case
2112 // of a malicious producer asking to commit a chunk made of random data,
2113 // which is something this class has to deal with regardless.
2114 //
2115 // The only legitimate mutations that can happen from sane producers,
2116 // concurrently to this function, are:
2117 // A. free pages being partitioned,
2118 // B. free chunks being migrated to kChunkBeingWritten,
2119 // C. kChunkBeingWritten chunks being migrated to kChunkCompleted.
2120
2121 SharedMemoryABI* abi = &producer->shmem_abi_;
2122 // num_pages() is immutable after the SMB is initialized and cannot be changed
2123 // even by a producer even if malicious.
2124 for (size_t page_idx = 0; page_idx < abi->num_pages(); page_idx++) {
2125 uint32_t header_bitmap = abi->GetPageHeaderBitmap(page_idx);
2126
2127 uint32_t used_chunks =
2128 abi->GetUsedChunks(header_bitmap); // Returns a bitmap.
2129 // Skip empty pages.
2130 if (used_chunks == 0)
2131 continue;
2132
2133 // Scrape the chunks that are currently used. These should be either in
2134 // state kChunkBeingWritten or kChunkComplete.
2135 for (uint32_t chunk_idx = 0; used_chunks; chunk_idx++, used_chunks >>= 1) {
2136 if (!(used_chunks & 1))
2137 continue;
2138
2139 SharedMemoryABI::ChunkState state =
2140 SharedMemoryABI::GetChunkStateFromHeaderBitmap(header_bitmap,
2141 chunk_idx);
2142 PERFETTO_DCHECK(state == SharedMemoryABI::kChunkBeingWritten ||
2143 state == SharedMemoryABI::kChunkComplete);
2144 bool chunk_complete = state == SharedMemoryABI::kChunkComplete;
2145
2146 SharedMemoryABI::Chunk chunk =
2147 abi->GetChunkUnchecked(page_idx, header_bitmap, chunk_idx);
2148
2149 uint16_t packet_count;
2150 uint8_t flags;
2151 // GetPacketCountAndFlags has acquire_load semantics.
2152 std::tie(packet_count, flags) = chunk.GetPacketCountAndFlags();
2153
2154 // It only makes sense to copy an incomplete chunk if there's at least
2155 // one full packet available. (The producer may not have completed the
2156 // last packet in it yet, so we need at least 2.)
2157 if (!chunk_complete && packet_count < 2)
2158 continue;
2159
2160 // At this point, it is safe to access the remaining header fields of
2161 // the chunk. Even if the chunk was only just transferred from
2162 // kChunkFree into kChunkBeingWritten state, the header should be
2163 // written completely once the packet count increased above 1 (it was
2164 // reset to 0 by the service when the chunk was freed).
2165
2166 WriterID writer_id = chunk.writer_id();
2167 std::optional<BufferID> target_buffer_id =
2168 producer->buffer_id_for_writer(writer_id);
2169
2170 // We can only scrape this chunk if we know which log buffer to copy it
2171 // into.
2172 if (!target_buffer_id)
2173 continue;
2174
2175 // Skip chunks that don't belong to the requested tracing session.
2176 bool target_buffer_belongs_to_session =
2177 std::find(session_buffers.begin(), session_buffers.end(),
2178 *target_buffer_id) != session_buffers.end();
2179 if (!target_buffer_belongs_to_session)
2180 continue;
2181
2182 uint32_t chunk_id =
2183 chunk.header()->chunk_id.load(std::memory_order_relaxed);
2184
2185 CopyProducerPageIntoLogBuffer(
2186 producer->id_, producer->client_identity_, writer_id, chunk_id,
2187 *target_buffer_id, packet_count, flags, chunk_complete,
2188 chunk.payload_begin(), chunk.payload_size());
2189 }
2190 }
2191 }
2192
FlushAndDisableTracing(TracingSessionID tsid)2193 void TracingServiceImpl::FlushAndDisableTracing(TracingSessionID tsid) {
2194 PERFETTO_DCHECK_THREAD(thread_checker_);
2195 PERFETTO_DLOG("Triggering final flush for %" PRIu64, tsid);
2196 Flush(
2197 tsid, 0,
2198 [this, tsid](bool success) {
2199 // This was a DLOG up to Jun 2021 (v16, Android S).
2200 PERFETTO_LOG("FlushAndDisableTracing(%" PRIu64 ") done, success=%d",
2201 tsid, success);
2202 TracingSession* session = GetTracingSession(tsid);
2203 if (!session) {
2204 return;
2205 }
2206 session->final_flush_outcome = success
2207 ? TraceStats::FINAL_FLUSH_SUCCEEDED
2208 : TraceStats::FINAL_FLUSH_FAILED;
2209 if (session->consumer_maybe_null) {
2210 // If the consumer is still attached, just disable the session but
2211 // give it a chance to read the contents.
2212 DisableTracing(tsid);
2213 } else {
2214 // If the consumer detached, destroy the session. If the consumer did
2215 // start the session in long-tracing mode, the service will have saved
2216 // the contents to the passed file. If not, the contents will be
2217 // destroyed.
2218 FreeBuffers(tsid);
2219 }
2220 },
2221 FlushFlags(FlushFlags::Initiator::kTraced,
2222 FlushFlags::Reason::kTraceStop));
2223 }
2224
PeriodicFlushTask(TracingSessionID tsid,bool post_next_only)2225 void TracingServiceImpl::PeriodicFlushTask(TracingSessionID tsid,
2226 bool post_next_only) {
2227 PERFETTO_DCHECK_THREAD(thread_checker_);
2228 TracingSession* tracing_session = GetTracingSession(tsid);
2229 if (!tracing_session || tracing_session->state != TracingSession::STARTED)
2230 return;
2231
2232 uint32_t flush_period_ms = tracing_session->config.flush_period_ms();
2233 weak_runner_.PostDelayedTask(
2234 [this, tsid] { PeriodicFlushTask(tsid, /*post_next_only=*/false); },
2235 flush_period_ms - static_cast<uint32_t>(clock_->GetWallTimeMs().count() %
2236 flush_period_ms));
2237
2238 if (post_next_only)
2239 return;
2240
2241 PERFETTO_DLOG("Triggering periodic flush for trace session %" PRIu64, tsid);
2242 Flush(
2243 tsid, 0,
2244 [](bool success) {
2245 if (!success)
2246 PERFETTO_ELOG("Periodic flush timed out");
2247 },
2248 FlushFlags(FlushFlags::Initiator::kTraced,
2249 FlushFlags::Reason::kPeriodic));
2250 }
2251
PeriodicClearIncrementalStateTask(TracingSessionID tsid,bool post_next_only)2252 void TracingServiceImpl::PeriodicClearIncrementalStateTask(
2253 TracingSessionID tsid,
2254 bool post_next_only) {
2255 PERFETTO_DCHECK_THREAD(thread_checker_);
2256 TracingSession* tracing_session = GetTracingSession(tsid);
2257 if (!tracing_session || tracing_session->state != TracingSession::STARTED)
2258 return;
2259
2260 uint32_t clear_period_ms =
2261 tracing_session->config.incremental_state_config().clear_period_ms();
2262 weak_runner_.PostDelayedTask(
2263 [this, tsid] {
2264 PeriodicClearIncrementalStateTask(tsid, /*post_next_only=*/false);
2265 },
2266 clear_period_ms - static_cast<uint32_t>(clock_->GetWallTimeMs().count() %
2267 clear_period_ms));
2268
2269 if (post_next_only)
2270 return;
2271
2272 PERFETTO_DLOG(
2273 "Performing periodic incremental state clear for trace session %" PRIu64,
2274 tsid);
2275
2276 // Queue the IPCs to producers with active data sources that opted in.
2277 std::map<ProducerID, std::vector<DataSourceInstanceID>> clear_map;
2278 for (const auto& kv : tracing_session->data_source_instances) {
2279 ProducerID producer_id = kv.first;
2280 const DataSourceInstance& data_source = kv.second;
2281 if (data_source.handles_incremental_state_clear) {
2282 clear_map[producer_id].push_back(data_source.instance_id);
2283 }
2284 }
2285
2286 for (const auto& kv : clear_map) {
2287 ProducerID producer_id = kv.first;
2288 const std::vector<DataSourceInstanceID>& data_sources = kv.second;
2289 ProducerEndpointImpl* producer = GetProducer(producer_id);
2290 if (!producer) {
2291 PERFETTO_DFATAL("Producer does not exist.");
2292 continue;
2293 }
2294 producer->ClearIncrementalState(data_sources);
2295 }
2296 }
2297
ReadBuffersIntoConsumer(TracingSessionID tsid,ConsumerEndpointImpl * consumer)2298 bool TracingServiceImpl::ReadBuffersIntoConsumer(
2299 TracingSessionID tsid,
2300 ConsumerEndpointImpl* consumer) {
2301 PERFETTO_DCHECK(consumer);
2302 PERFETTO_DCHECK_THREAD(thread_checker_);
2303 TracingSession* tracing_session = GetTracingSession(tsid);
2304 if (!tracing_session) {
2305 PERFETTO_DLOG(
2306 "Cannot ReadBuffersIntoConsumer(): no tracing session is active");
2307 return false;
2308 }
2309
2310 if (tracing_session->write_into_file) {
2311 // If the consumer enabled tracing and asked to save the contents into the
2312 // passed file makes little sense to also try to read the buffers over IPC,
2313 // as that would just steal data from the periodic draining task.
2314 PERFETTO_ELOG("Consumer trying to read from write_into_file session.");
2315 return false;
2316 }
2317
2318 if (IsWaitingForTrigger(tracing_session))
2319 return false;
2320
2321 // This is a rough threshold to determine how much to read from the buffer in
2322 // each task. This is to avoid executing a single huge sending task for too
2323 // long and risk to hit the watchdog. This is *not* an upper bound: we just
2324 // stop accumulating new packets and PostTask *after* we cross this threshold.
2325 // This constant essentially balances the PostTask and IPC overhead vs the
2326 // responsiveness of the service. An extremely small value will cause one IPC
2327 // and one PostTask for each slice but will keep the service extremely
2328 // responsive. An extremely large value will batch the send for the full
2329 // buffer in one large task, will hit the blocking send() once the socket
2330 // buffers are full and hang the service for a bit (until the consumer
2331 // catches up).
2332 static constexpr size_t kApproxBytesPerTask = 32768;
2333 bool has_more;
2334 std::vector<TracePacket> packets =
2335 ReadBuffers(tracing_session, kApproxBytesPerTask, &has_more);
2336
2337 if (has_more) {
2338 auto weak_consumer = consumer->weak_ptr_factory_.GetWeakPtr();
2339 weak_runner_.PostTask(
2340 [this, weak_consumer = std::move(weak_consumer), tsid] {
2341 if (!weak_consumer)
2342 return;
2343 ReadBuffersIntoConsumer(tsid, weak_consumer.get());
2344 });
2345 }
2346
2347 // Keep this as tail call, just in case the consumer re-enters.
2348 consumer->consumer_->OnTraceData(std::move(packets), has_more);
2349 return true;
2350 }
2351
ReadBuffersIntoFile(TracingSessionID tsid)2352 bool TracingServiceImpl::ReadBuffersIntoFile(TracingSessionID tsid) {
2353 PERFETTO_DCHECK_THREAD(thread_checker_);
2354 TracingSession* tracing_session = GetTracingSession(tsid);
2355 if (!tracing_session) {
2356 // This will be hit systematically from the PostDelayedTask. Avoid logging,
2357 // it would be just spam.
2358 return false;
2359 }
2360
2361 // This can happen if the file is closed by a previous task because it reaches
2362 // |max_file_size_bytes|.
2363 if (!tracing_session->write_into_file)
2364 return false;
2365
2366 if (IsWaitingForTrigger(tracing_session))
2367 return false;
2368
2369 // ReadBuffers() can allocate memory internally, for filtering. By limiting
2370 // the data that ReadBuffers() reads to kWriteIntoChunksSize per iteration,
2371 // we limit the amount of memory used on each iteration.
2372 //
2373 // It would be tempting to split this into multiple tasks like in
2374 // ReadBuffersIntoConsumer, but that's not currently possible.
2375 // ReadBuffersIntoFile has to read the whole available data before returning,
2376 // to support the disable_immediately=true code paths.
2377 bool has_more = true;
2378 bool stop_writing_into_file = false;
2379 do {
2380 std::vector<TracePacket> packets =
2381 ReadBuffers(tracing_session, kWriteIntoFileChunkSize, &has_more);
2382
2383 stop_writing_into_file = WriteIntoFile(tracing_session, std::move(packets));
2384 } while (has_more && !stop_writing_into_file);
2385
2386 if (stop_writing_into_file || tracing_session->write_period_ms == 0) {
2387 // Ensure all data was written to the file before we close it.
2388 base::FlushFile(tracing_session->write_into_file.get());
2389 tracing_session->write_into_file.reset();
2390 tracing_session->write_period_ms = 0;
2391 if (tracing_session->state == TracingSession::STARTED)
2392 DisableTracing(tsid);
2393 return true;
2394 }
2395
2396 weak_runner_.PostDelayedTask([this, tsid] { ReadBuffersIntoFile(tsid); },
2397 DelayToNextWritePeriodMs(*tracing_session));
2398 return true;
2399 }
2400
IsWaitingForTrigger(TracingSession * tracing_session)2401 bool TracingServiceImpl::IsWaitingForTrigger(TracingSession* tracing_session) {
2402 // Ignore the logic below for cloned tracing sessions. In this case we
2403 // actually want to read the (cloned) trace buffers even if no trigger was
2404 // hit.
2405 if (tracing_session->state == TracingSession::CLONED_READ_ONLY) {
2406 return false;
2407 }
2408
2409 // When a tracing session is waiting for a trigger, it is considered empty. If
2410 // a tracing session finishes and moves into DISABLED without ever receiving a
2411 // trigger, the trace should never return any data. This includes the
2412 // synthetic packets like TraceConfig and Clock snapshots. So we bail out
2413 // early and let the consumer know there is no data.
2414 if (!tracing_session->config.trigger_config().triggers().empty() &&
2415 tracing_session->received_triggers.empty()) {
2416 PERFETTO_DLOG(
2417 "ReadBuffers(): tracing session has not received a trigger yet.");
2418 return true;
2419 }
2420
2421 // Traces with CLONE_SNAPSHOT triggers are a special case of the above. They
2422 // can be read only via a CloneSession() request. This is to keep the
2423 // behavior consistent with the STOP_TRACING+triggers case and avoid periodic
2424 // finalizations and uploads of the main CLONE_SNAPSHOT triggers.
2425 if (GetTriggerMode(tracing_session->config) ==
2426 TraceConfig::TriggerConfig::CLONE_SNAPSHOT) {
2427 PERFETTO_DLOG(
2428 "ReadBuffers(): skipping because the tracing session has "
2429 "CLONE_SNAPSHOT triggers defined");
2430 return true;
2431 }
2432
2433 return false;
2434 }
2435
ReadBuffers(TracingSession * tracing_session,size_t threshold,bool * has_more)2436 std::vector<TracePacket> TracingServiceImpl::ReadBuffers(
2437 TracingSession* tracing_session,
2438 size_t threshold,
2439 bool* has_more) {
2440 PERFETTO_DCHECK_THREAD(thread_checker_);
2441 PERFETTO_DCHECK(tracing_session);
2442 *has_more = false;
2443
2444 std::vector<TracePacket> packets;
2445 packets.reserve(1024); // Just an educated guess to avoid trivial expansions.
2446
2447 if (!tracing_session->initial_clock_snapshot.empty()) {
2448 EmitClockSnapshot(tracing_session,
2449 std::move(tracing_session->initial_clock_snapshot),
2450 &packets);
2451 }
2452
2453 for (auto& snapshot : tracing_session->clock_snapshot_ring_buffer) {
2454 PERFETTO_DCHECK(!snapshot.empty());
2455 EmitClockSnapshot(tracing_session, std::move(snapshot), &packets);
2456 }
2457 tracing_session->clock_snapshot_ring_buffer.clear();
2458
2459 if (tracing_session->should_emit_sync_marker) {
2460 EmitSyncMarker(&packets);
2461 tracing_session->should_emit_sync_marker = false;
2462 }
2463
2464 if (!tracing_session->config.builtin_data_sources().disable_trace_config()) {
2465 MaybeEmitTraceConfig(tracing_session, &packets);
2466 MaybeEmitCloneTrigger(tracing_session, &packets);
2467 MaybeEmitReceivedTriggers(tracing_session, &packets);
2468 }
2469 if (!tracing_session->did_emit_initial_packets) {
2470 EmitUuid(tracing_session, &packets);
2471 if (!tracing_session->config.builtin_data_sources().disable_system_info())
2472 EmitSystemInfo(&packets);
2473 }
2474 tracing_session->did_emit_initial_packets = true;
2475
2476 // Note that in the proto comment, we guarantee that the tracing_started
2477 // lifecycle event will be emitted before any data packets so make sure to
2478 // keep this before reading the tracing buffers.
2479 if (!tracing_session->config.builtin_data_sources().disable_service_events())
2480 EmitLifecycleEvents(tracing_session, &packets);
2481
2482 // In a multi-machine tracing session, emit clock synchronization messages for
2483 // remote machines.
2484 if (!relay_clients_.empty())
2485 MaybeEmitRemoteClockSync(tracing_session, &packets);
2486
2487 size_t packets_bytes = 0; // SUM(slice.size() for each slice in |packets|).
2488
2489 // Add up size for packets added by the Maybe* calls above.
2490 for (const TracePacket& packet : packets) {
2491 packets_bytes += packet.size();
2492 }
2493
2494 bool did_hit_threshold = false;
2495
2496 for (size_t buf_idx = 0;
2497 buf_idx < tracing_session->num_buffers() && !did_hit_threshold;
2498 buf_idx++) {
2499 auto tbuf_iter = buffers_.find(tracing_session->buffers_index[buf_idx]);
2500 if (tbuf_iter == buffers_.end()) {
2501 PERFETTO_DFATAL("Buffer not found.");
2502 continue;
2503 }
2504 TraceBuffer& tbuf = *tbuf_iter->second;
2505 tbuf.BeginRead();
2506 while (!did_hit_threshold) {
2507 TracePacket packet;
2508 TraceBuffer::PacketSequenceProperties sequence_properties{};
2509 bool previous_packet_dropped;
2510 if (!tbuf.ReadNextTracePacket(&packet, &sequence_properties,
2511 &previous_packet_dropped)) {
2512 break;
2513 }
2514 packet.set_buffer_index_for_stats(static_cast<uint32_t>(buf_idx));
2515 PERFETTO_DCHECK(sequence_properties.producer_id_trusted != 0);
2516 PERFETTO_DCHECK(sequence_properties.writer_id != 0);
2517 PERFETTO_DCHECK(sequence_properties.client_identity_trusted.has_uid());
2518 // Not checking sequence_properties.client_identity_trusted.has_pid():
2519 // it is false if the platform doesn't support it.
2520
2521 PERFETTO_DCHECK(packet.size() > 0);
2522 if (!PacketStreamValidator::Validate(packet.slices())) {
2523 tracing_session->invalid_packets++;
2524 PERFETTO_DLOG("Dropping invalid packet");
2525 continue;
2526 }
2527
2528 // Append a slice with the trusted field data. This can't be spoofed
2529 // because above we validated that the existing slices don't contain any
2530 // trusted fields. For added safety we append instead of prepending
2531 // because according to protobuf semantics, if the same field is
2532 // encountered multiple times the last instance takes priority. Note that
2533 // truncated packets are also rejected, so the producer can't give us a
2534 // partial packet (e.g., a truncated string) which only becomes valid when
2535 // the trusted data is appended here.
2536 Slice slice = Slice::Allocate(32);
2537 protozero::StaticBuffered<protos::pbzero::TracePacket> trusted_packet(
2538 slice.own_data(), slice.size);
2539 const auto& client_identity_trusted =
2540 sequence_properties.client_identity_trusted;
2541 trusted_packet->set_trusted_uid(
2542 static_cast<int32_t>(client_identity_trusted.uid()));
2543 trusted_packet->set_trusted_packet_sequence_id(
2544 tracing_session->GetPacketSequenceID(
2545 client_identity_trusted.machine_id(),
2546 sequence_properties.producer_id_trusted,
2547 sequence_properties.writer_id));
2548 if (client_identity_trusted.has_pid()) {
2549 // Not supported on all platforms.
2550 trusted_packet->set_trusted_pid(
2551 static_cast<int32_t>(client_identity_trusted.pid()));
2552 }
2553 if (client_identity_trusted.has_non_default_machine_id()) {
2554 trusted_packet->set_machine_id(client_identity_trusted.machine_id());
2555 }
2556 if (previous_packet_dropped)
2557 trusted_packet->set_previous_packet_dropped(previous_packet_dropped);
2558 slice.size = trusted_packet.Finalize();
2559 packet.AddSlice(std::move(slice));
2560
2561 // Append the packet (inclusive of the trusted uid) to |packets|.
2562 packets_bytes += packet.size();
2563 did_hit_threshold = packets_bytes >= threshold;
2564 packets.emplace_back(std::move(packet));
2565 } // for(packets...)
2566 } // for(buffers...)
2567
2568 *has_more = did_hit_threshold;
2569
2570 // Only emit the "read complete" lifetime event when there is no more trace
2571 // data available to read. These events are used as safe points to limit
2572 // sorting in trace processor: the code shouldn't emit the event unless the
2573 // buffers are empty.
2574 if (!*has_more && !tracing_session->config.builtin_data_sources()
2575 .disable_service_events()) {
2576 // We don't bother snapshotting clocks here because we wouldn't be able to
2577 // emit it and we shouldn't have significant drift from the last snapshot in
2578 // any case.
2579 SnapshotLifecycleEvent(tracing_session,
2580 protos::pbzero::TracingServiceEvent::
2581 kReadTracingBuffersCompletedFieldNumber,
2582 false /* snapshot_clocks */);
2583 EmitLifecycleEvents(tracing_session, &packets);
2584 }
2585
2586 // Only emit the stats when there is no more trace data is available to read.
2587 // That way, any problems that occur while reading from the buffers are
2588 // reflected in the emitted stats. This is particularly important for use
2589 // cases where ReadBuffers is only ever called after the tracing session is
2590 // stopped.
2591 if (!*has_more && tracing_session->should_emit_stats) {
2592 EmitStats(tracing_session, &packets);
2593 tracing_session->should_emit_stats = false;
2594 }
2595
2596 MaybeFilterPackets(tracing_session, &packets);
2597
2598 MaybeCompressPackets(tracing_session, &packets);
2599
2600 if (!*has_more) {
2601 // We've observed some extremely high memory usage by scudo after
2602 // MaybeFilterPackets in the past. The original bug (b/195145848) is fixed
2603 // now, but this code asks scudo to release memory just in case.
2604 base::MaybeReleaseAllocatorMemToOS();
2605 }
2606
2607 return packets;
2608 }
2609
MaybeFilterPackets(TracingSession * tracing_session,std::vector<TracePacket> * packets)2610 void TracingServiceImpl::MaybeFilterPackets(TracingSession* tracing_session,
2611 std::vector<TracePacket>* packets) {
2612 // If the tracing session specified a filter, run all packets through the
2613 // filter and replace them with the filter results.
2614 // The process below mantains the cardinality of input packets. Even if an
2615 // entire packet is filtered out, we emit a zero-sized TracePacket proto. That
2616 // makes debugging and reasoning about the trace stats easier.
2617 // This place swaps the contents of each |packets| entry in place.
2618 if (!tracing_session->trace_filter) {
2619 return;
2620 }
2621 protozero::MessageFilter& trace_filter = *tracing_session->trace_filter;
2622 // The filter root should be reset from protos.Trace to protos.TracePacket
2623 // by the earlier call to SetFilterRoot() in EnableTracing().
2624 PERFETTO_DCHECK(trace_filter.config().root_msg_index() != 0);
2625 std::vector<protozero::MessageFilter::InputSlice> filter_input;
2626 auto start = clock_->GetWallTimeNs();
2627 for (TracePacket& packet : *packets) {
2628 const auto& packet_slices = packet.slices();
2629 const size_t input_packet_size = packet.size();
2630 filter_input.clear();
2631 filter_input.resize(packet_slices.size());
2632 ++tracing_session->filter_input_packets;
2633 tracing_session->filter_input_bytes += input_packet_size;
2634 for (size_t i = 0; i < packet_slices.size(); ++i)
2635 filter_input[i] = {packet_slices[i].start, packet_slices[i].size};
2636 auto filtered_packet = trace_filter.FilterMessageFragments(
2637 &filter_input[0], filter_input.size());
2638
2639 // Replace the packet in-place with the filtered one (unless failed).
2640 std::optional<uint32_t> maybe_buffer_idx = packet.buffer_index_for_stats();
2641 packet = TracePacket();
2642 if (filtered_packet.error) {
2643 ++tracing_session->filter_errors;
2644 PERFETTO_DLOG("Trace packet filtering failed @ packet %" PRIu64,
2645 tracing_session->filter_input_packets);
2646 continue;
2647 }
2648 tracing_session->filter_output_bytes += filtered_packet.size;
2649 if (maybe_buffer_idx.has_value()) {
2650 // Keep the per-buffer stats updated. Also propagate the
2651 // buffer_index_for_stats in the output packet to allow accounting by
2652 // other parts of the ReadBuffer pipeline.
2653 uint32_t buffer_idx = maybe_buffer_idx.value();
2654 packet.set_buffer_index_for_stats(buffer_idx);
2655 auto& vec = tracing_session->filter_bytes_discarded_per_buffer;
2656 if (static_cast<size_t>(buffer_idx) >= vec.size())
2657 vec.resize(buffer_idx + 1);
2658 PERFETTO_DCHECK(input_packet_size >= filtered_packet.size);
2659 size_t bytes_filtered_out = input_packet_size - filtered_packet.size;
2660 vec[buffer_idx] += bytes_filtered_out;
2661 }
2662 AppendOwnedSlicesToPacket(std::move(filtered_packet.data),
2663 filtered_packet.size, kMaxTracePacketSliceSize,
2664 &packet);
2665 }
2666 auto end = clock_->GetWallTimeNs();
2667 tracing_session->filter_time_taken_ns +=
2668 static_cast<uint64_t>((end - start).count());
2669 }
2670
MaybeCompressPackets(TracingSession * tracing_session,std::vector<TracePacket> * packets)2671 void TracingServiceImpl::MaybeCompressPackets(
2672 TracingSession* tracing_session,
2673 std::vector<TracePacket>* packets) {
2674 if (!tracing_session->compress_deflate) {
2675 return;
2676 }
2677
2678 init_opts_.compressor_fn(packets);
2679 }
2680
WriteIntoFile(TracingSession * tracing_session,std::vector<TracePacket> packets)2681 bool TracingServiceImpl::WriteIntoFile(TracingSession* tracing_session,
2682 std::vector<TracePacket> packets) {
2683 if (!tracing_session->write_into_file) {
2684 return false;
2685 }
2686 const uint64_t max_size = tracing_session->max_file_size_bytes
2687 ? tracing_session->max_file_size_bytes
2688 : std::numeric_limits<size_t>::max();
2689
2690 size_t total_slices = 0;
2691 for (const TracePacket& packet : packets) {
2692 total_slices += packet.slices().size();
2693 }
2694 // When writing into a file, the file should look like a root trace.proto
2695 // message. Each packet should be prepended with a proto preamble stating
2696 // its field id (within trace.proto) and size. Hence the addition below.
2697 const size_t max_iovecs = total_slices + packets.size();
2698
2699 size_t num_iovecs = 0;
2700 bool stop_writing_into_file = false;
2701 std::unique_ptr<struct iovec[]> iovecs(new struct iovec[max_iovecs]);
2702 size_t num_iovecs_at_last_packet = 0;
2703 uint64_t bytes_about_to_be_written = 0;
2704 for (TracePacket& packet : packets) {
2705 std::tie(iovecs[num_iovecs].iov_base, iovecs[num_iovecs].iov_len) =
2706 packet.GetProtoPreamble();
2707 bytes_about_to_be_written += iovecs[num_iovecs].iov_len;
2708 num_iovecs++;
2709 for (const Slice& slice : packet.slices()) {
2710 // writev() doesn't change the passed pointer. However, struct iovec
2711 // take a non-const ptr because it's the same struct used by readv().
2712 // Hence the const_cast here.
2713 char* start = static_cast<char*>(const_cast<void*>(slice.start));
2714 bytes_about_to_be_written += slice.size;
2715 iovecs[num_iovecs++] = {start, slice.size};
2716 }
2717
2718 if (tracing_session->bytes_written_into_file + bytes_about_to_be_written >=
2719 max_size) {
2720 stop_writing_into_file = true;
2721 num_iovecs = num_iovecs_at_last_packet;
2722 break;
2723 }
2724
2725 num_iovecs_at_last_packet = num_iovecs;
2726 }
2727 PERFETTO_DCHECK(num_iovecs <= max_iovecs);
2728 int fd = *tracing_session->write_into_file;
2729
2730 uint64_t total_wr_size = 0;
2731
2732 // writev() can take at most IOV_MAX entries per call. Batch them.
2733 constexpr size_t kIOVMax = IOV_MAX;
2734 for (size_t i = 0; i < num_iovecs; i += kIOVMax) {
2735 int iov_batch_size = static_cast<int>(std::min(num_iovecs - i, kIOVMax));
2736 ssize_t wr_size = PERFETTO_EINTR(writev(fd, &iovecs[i], iov_batch_size));
2737 if (wr_size <= 0) {
2738 PERFETTO_PLOG("writev() failed");
2739 stop_writing_into_file = true;
2740 break;
2741 }
2742 total_wr_size += static_cast<size_t>(wr_size);
2743 }
2744
2745 tracing_session->bytes_written_into_file += total_wr_size;
2746
2747 PERFETTO_DLOG("Draining into file, written: %" PRIu64 " KB, stop: %d",
2748 (total_wr_size + 1023) / 1024, stop_writing_into_file);
2749 return stop_writing_into_file;
2750 }
2751
FreeBuffers(TracingSessionID tsid)2752 void TracingServiceImpl::FreeBuffers(TracingSessionID tsid) {
2753 PERFETTO_DCHECK_THREAD(thread_checker_);
2754 PERFETTO_DLOG("Freeing buffers for session %" PRIu64, tsid);
2755 TracingSession* tracing_session = GetTracingSession(tsid);
2756 if (!tracing_session) {
2757 PERFETTO_DLOG("FreeBuffers() failed, invalid session ID %" PRIu64, tsid);
2758 return; // TODO(primiano): signal failure?
2759 }
2760 DisableTracing(tsid, /*disable_immediately=*/true);
2761
2762 PERFETTO_DCHECK(tracing_session->AllDataSourceInstancesStopped());
2763 tracing_session->data_source_instances.clear();
2764
2765 for (auto& producer_entry : producers_) {
2766 ProducerEndpointImpl* producer = producer_entry.second;
2767 producer->OnFreeBuffers(tracing_session->buffers_index);
2768 }
2769
2770 for (BufferID buffer_id : tracing_session->buffers_index) {
2771 buffer_ids_.Free(buffer_id);
2772 PERFETTO_DCHECK(buffers_.count(buffer_id) == 1);
2773 buffers_.erase(buffer_id);
2774 }
2775 bool notify_traceur =
2776 tracing_session->config.notify_traceur() &&
2777 tracing_session->state != TracingSession::CLONED_READ_ONLY;
2778 bool is_long_trace =
2779 (tracing_session->config.write_into_file() &&
2780 tracing_session->config.file_write_period_ms() < kMillisPerDay);
2781 auto pending_clones = std::move(tracing_session->pending_clones);
2782 tracing_sessions_.erase(tsid);
2783 tracing_session = nullptr;
2784 UpdateMemoryGuardrail();
2785
2786 for (const auto& id_to_clone_op : pending_clones) {
2787 const PendingClone& clone_op = id_to_clone_op.second;
2788 if (clone_op.weak_consumer) {
2789 weak_runner_.task_runner()->PostTask(
2790 [weak_consumer = clone_op.weak_consumer] {
2791 if (weak_consumer) {
2792 weak_consumer->consumer_->OnSessionCloned(
2793 {false, "Original session ended", {}});
2794 }
2795 });
2796 }
2797 }
2798
2799 PERFETTO_LOG("Tracing session %" PRIu64 " ended, total sessions:%zu", tsid,
2800 tracing_sessions_.size());
2801 #if PERFETTO_BUILDFLAG(PERFETTO_ANDROID_BUILD) && \
2802 PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID)
2803 if (notify_traceur && is_long_trace) {
2804 PERFETTO_LAZY_LOAD(android_internal::NotifyTraceSessionEnded, notify_fn);
2805 if (!notify_fn || !notify_fn(/*session_stolen=*/false))
2806 PERFETTO_ELOG("Failed to notify Traceur long tracing has ended");
2807 }
2808 #else
2809 base::ignore_result(notify_traceur);
2810 base::ignore_result(is_long_trace);
2811 #endif
2812 }
2813
RegisterDataSource(ProducerID producer_id,const DataSourceDescriptor & desc)2814 void TracingServiceImpl::RegisterDataSource(ProducerID producer_id,
2815 const DataSourceDescriptor& desc) {
2816 PERFETTO_DCHECK_THREAD(thread_checker_);
2817 if (desc.name().empty()) {
2818 PERFETTO_DLOG("Received RegisterDataSource() with empty name");
2819 return;
2820 }
2821
2822 ProducerEndpointImpl* producer = GetProducer(producer_id);
2823 if (!producer) {
2824 PERFETTO_DFATAL("Producer not found.");
2825 return;
2826 }
2827
2828 // Check that the producer doesn't register two data sources with the same ID.
2829 // Note that we tolerate |id| == 0 because until Android T / v22 the |id|
2830 // field didn't exist.
2831 for (const auto& kv : data_sources_) {
2832 if (desc.id() && kv.second.producer_id == producer_id &&
2833 kv.second.descriptor.id() == desc.id()) {
2834 PERFETTO_ELOG(
2835 "Failed to register data source \"%s\". A data source with the same "
2836 "id %" PRIu64 " (name=\"%s\") is already registered for producer %d",
2837 desc.name().c_str(), desc.id(), kv.second.descriptor.name().c_str(),
2838 producer_id);
2839 return;
2840 }
2841 }
2842
2843 PERFETTO_DLOG("Producer %" PRIu16 " registered data source \"%s\"",
2844 producer_id, desc.name().c_str());
2845
2846 auto reg_ds = data_sources_.emplace(desc.name(),
2847 RegisteredDataSource{producer_id, desc});
2848
2849 // If there are existing tracing sessions, we need to check if the new
2850 // data source is enabled by any of them.
2851 for (auto& iter : tracing_sessions_) {
2852 TracingSession& tracing_session = iter.second;
2853 if (tracing_session.state != TracingSession::STARTED &&
2854 tracing_session.state != TracingSession::CONFIGURED) {
2855 continue;
2856 }
2857
2858 TraceConfig::ProducerConfig producer_config;
2859 for (const auto& config : tracing_session.config.producers()) {
2860 if (producer->name_ == config.producer_name()) {
2861 producer_config = config;
2862 break;
2863 }
2864 }
2865 for (const TraceConfig::DataSource& cfg_data_source :
2866 tracing_session.config.data_sources()) {
2867 if (cfg_data_source.config().name() != desc.name())
2868 continue;
2869 DataSourceInstance* ds_inst = SetupDataSource(
2870 cfg_data_source, producer_config, reg_ds->second, &tracing_session);
2871 if (ds_inst && tracing_session.state == TracingSession::STARTED)
2872 StartDataSourceInstance(producer, &tracing_session, ds_inst);
2873 }
2874 } // for(iter : tracing_sessions_)
2875 }
2876
UpdateDataSource(ProducerID producer_id,const DataSourceDescriptor & new_desc)2877 void TracingServiceImpl::UpdateDataSource(
2878 ProducerID producer_id,
2879 const DataSourceDescriptor& new_desc) {
2880 if (new_desc.id() == 0) {
2881 PERFETTO_ELOG("UpdateDataSource() must have a non-zero id");
2882 return;
2883 }
2884
2885 // If this producer has already registered a matching descriptor name and id,
2886 // just update the descriptor.
2887 RegisteredDataSource* data_source = nullptr;
2888 auto range = data_sources_.equal_range(new_desc.name());
2889 for (auto it = range.first; it != range.second; ++it) {
2890 if (it->second.producer_id == producer_id &&
2891 it->second.descriptor.id() == new_desc.id()) {
2892 data_source = &it->second;
2893 break;
2894 }
2895 }
2896
2897 if (!data_source) {
2898 PERFETTO_ELOG(
2899 "UpdateDataSource() failed, could not find an existing data source "
2900 "with name=\"%s\" id=%" PRIu64,
2901 new_desc.name().c_str(), new_desc.id());
2902 return;
2903 }
2904
2905 data_source->descriptor = new_desc;
2906 }
2907
StopDataSourceInstance(ProducerEndpointImpl * producer,TracingSession * tracing_session,DataSourceInstance * instance,bool disable_immediately)2908 void TracingServiceImpl::StopDataSourceInstance(ProducerEndpointImpl* producer,
2909 TracingSession* tracing_session,
2910 DataSourceInstance* instance,
2911 bool disable_immediately) {
2912 const DataSourceInstanceID ds_inst_id = instance->instance_id;
2913 if (producer->IsAndroidProcessFrozen()) {
2914 PERFETTO_DLOG(
2915 "skipping waiting of data source \"%s\" on producer \"%s\" (pid=%u) "
2916 "because it is frozen",
2917 instance->data_source_name.c_str(), producer->name_.c_str(),
2918 producer->pid());
2919 disable_immediately = true;
2920 }
2921 if (instance->will_notify_on_stop && !disable_immediately) {
2922 instance->state = DataSourceInstance::STOPPING;
2923 } else {
2924 instance->state = DataSourceInstance::STOPPED;
2925 }
2926 if (tracing_session->consumer_maybe_null) {
2927 tracing_session->consumer_maybe_null->OnDataSourceInstanceStateChange(
2928 *producer, *instance);
2929 }
2930 producer->StopDataSource(ds_inst_id);
2931 }
2932
UnregisterDataSource(ProducerID producer_id,const std::string & name)2933 void TracingServiceImpl::UnregisterDataSource(ProducerID producer_id,
2934 const std::string& name) {
2935 PERFETTO_DCHECK_THREAD(thread_checker_);
2936 PERFETTO_DLOG("Producer %" PRIu16 " unregistered data source \"%s\"",
2937 producer_id, name.c_str());
2938 PERFETTO_CHECK(producer_id);
2939 ProducerEndpointImpl* producer = GetProducer(producer_id);
2940 PERFETTO_DCHECK(producer);
2941 for (auto& kv : tracing_sessions_) {
2942 auto& ds_instances = kv.second.data_source_instances;
2943 bool removed = false;
2944 for (auto it = ds_instances.begin(); it != ds_instances.end();) {
2945 if (it->first == producer_id && it->second.data_source_name == name) {
2946 DataSourceInstanceID ds_inst_id = it->second.instance_id;
2947 if (it->second.state != DataSourceInstance::STOPPED) {
2948 if (it->second.state != DataSourceInstance::STOPPING) {
2949 StopDataSourceInstance(producer, &kv.second, &it->second,
2950 /* disable_immediately = */ false);
2951 }
2952
2953 // Mark the instance as stopped immediately, since we are
2954 // unregistering it below.
2955 //
2956 // The StopDataSourceInstance above might have set the state to
2957 // STOPPING so this condition isn't an else.
2958 if (it->second.state == DataSourceInstance::STOPPING)
2959 NotifyDataSourceStopped(producer_id, ds_inst_id);
2960 }
2961 it = ds_instances.erase(it);
2962 removed = true;
2963 } else {
2964 ++it;
2965 }
2966 } // for (data_source_instances)
2967 if (removed)
2968 MaybeNotifyAllDataSourcesStarted(&kv.second);
2969 } // for (tracing_session)
2970
2971 for (auto it = data_sources_.begin(); it != data_sources_.end(); ++it) {
2972 if (it->second.producer_id == producer_id &&
2973 it->second.descriptor.name() == name) {
2974 data_sources_.erase(it);
2975 return;
2976 }
2977 }
2978
2979 PERFETTO_DFATAL(
2980 "Tried to unregister a non-existent data source \"%s\" for "
2981 "producer %" PRIu16,
2982 name.c_str(), producer_id);
2983 }
2984
IsInitiatorPrivileged(const TracingSession & tracing_session)2985 bool TracingServiceImpl::IsInitiatorPrivileged(
2986 const TracingSession& tracing_session) {
2987 #if PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID)
2988 if (tracing_session.consumer_uid == 1066 /* AID_STATSD */ &&
2989 tracing_session.config.statsd_metadata().triggering_config_uid() !=
2990 2000 /* AID_SHELL */
2991 && tracing_session.config.statsd_metadata().triggering_config_uid() !=
2992 0 /* AID_ROOT */) {
2993 // StatsD can be triggered either by shell, root or an app that has DUMP and
2994 // USAGE_STATS permission. When triggered by shell or root, we do not want
2995 // to consider the trace a trusted system trace, as it was initiated by the
2996 // user. Otherwise, it has to come from an app with DUMP and
2997 // PACKAGE_USAGE_STATS, which has to be preinstalled and trusted by the
2998 // system.
2999 // Check for shell / root: https://bit.ly/3b7oZNi
3000 // Check for DUMP or PACKAGE_USAGE_STATS: https://bit.ly/3ep0NrR
3001 return true;
3002 }
3003 if (tracing_session.consumer_uid == 1000 /* AID_SYSTEM */) {
3004 // AID_SYSTEM is considered a privileged initiator so that system_server can
3005 // profile apps that are not profileable by shell. Other AID_SYSTEM
3006 // processes are not allowed by SELinux to connect to the consumer socket or
3007 // to exec perfetto.
3008 return true;
3009 }
3010 #else
3011 base::ignore_result(tracing_session);
3012 #endif
3013 return false;
3014 }
3015
SetupDataSource(const TraceConfig::DataSource & cfg_data_source,const TraceConfig::ProducerConfig & producer_config,const RegisteredDataSource & data_source,TracingSession * tracing_session)3016 TracingServiceImpl::DataSourceInstance* TracingServiceImpl::SetupDataSource(
3017 const TraceConfig::DataSource& cfg_data_source,
3018 const TraceConfig::ProducerConfig& producer_config,
3019 const RegisteredDataSource& data_source,
3020 TracingSession* tracing_session) {
3021 PERFETTO_DCHECK_THREAD(thread_checker_);
3022 ProducerEndpointImpl* producer = GetProducer(data_source.producer_id);
3023 PERFETTO_DCHECK(producer);
3024 // An existing producer that is not ftrace could have registered itself as
3025 // ftrace, we must not enable it in that case.
3026 if (lockdown_mode_ && producer->uid() != uid_) {
3027 PERFETTO_DLOG("Lockdown mode: not enabling producer %hu", producer->id_);
3028 return nullptr;
3029 }
3030 // TODO(primiano): Add tests for registration ordering (data sources vs
3031 // consumers).
3032 if (!NameMatchesFilter(producer->name_,
3033 cfg_data_source.producer_name_filter(),
3034 cfg_data_source.producer_name_regex_filter())) {
3035 PERFETTO_DLOG("Data source: %s is filtered out for producer: %s",
3036 cfg_data_source.config().name().c_str(),
3037 producer->name_.c_str());
3038 return nullptr;
3039 }
3040
3041 auto relative_buffer_id = cfg_data_source.config().target_buffer();
3042 if (relative_buffer_id >= tracing_session->num_buffers()) {
3043 PERFETTO_LOG(
3044 "The TraceConfig for DataSource %s specified a target_buffer out of "
3045 "bound (%u). Skipping it.",
3046 cfg_data_source.config().name().c_str(), relative_buffer_id);
3047 return nullptr;
3048 }
3049
3050 // Create a copy of the DataSourceConfig specified in the trace config. This
3051 // will be passed to the producer after translating the |target_buffer| id.
3052 // The |target_buffer| parameter passed by the consumer in the trace config is
3053 // relative to the buffers declared in the same trace config. This has to be
3054 // translated to the global BufferID before passing it to the producers, which
3055 // don't know anything about tracing sessions and consumers.
3056
3057 DataSourceInstanceID inst_id = ++last_data_source_instance_id_;
3058 auto insert_iter = tracing_session->data_source_instances.emplace(
3059 std::piecewise_construct, //
3060 std::forward_as_tuple(producer->id_),
3061 std::forward_as_tuple(
3062 inst_id,
3063 cfg_data_source.config(), // Deliberate copy.
3064 data_source.descriptor.name(),
3065 data_source.descriptor.will_notify_on_start(),
3066 data_source.descriptor.will_notify_on_stop(),
3067 data_source.descriptor.handles_incremental_state_clear(),
3068 data_source.descriptor.no_flush()));
3069 DataSourceInstance* ds_instance = &insert_iter->second;
3070
3071 // New data source instance starts out in CONFIGURED state.
3072 if (tracing_session->consumer_maybe_null) {
3073 tracing_session->consumer_maybe_null->OnDataSourceInstanceStateChange(
3074 *producer, *ds_instance);
3075 }
3076
3077 DataSourceConfig& ds_config = ds_instance->config;
3078 ds_config.set_trace_duration_ms(tracing_session->config.duration_ms());
3079
3080 // Rationale for `if (prefer) set_prefer(true)`, rather than `set(prefer)`:
3081 // ComputeStartupConfigHash() in tracing_muxer_impl.cc compares hashes of the
3082 // DataSourceConfig and expects to know (and clear) the fields generated by
3083 // the tracing service. Unconditionally adding a new field breaks backward
3084 // compatibility of startup tracing with older SDKs, because the serialization
3085 // also propagates unkonwn fields, breaking the hash matching check.
3086 if (tracing_session->config.prefer_suspend_clock_for_duration())
3087 ds_config.set_prefer_suspend_clock_for_duration(true);
3088
3089 ds_config.set_stop_timeout_ms(tracing_session->data_source_stop_timeout_ms());
3090 ds_config.set_enable_extra_guardrails(
3091 tracing_session->config.enable_extra_guardrails());
3092 if (IsInitiatorPrivileged(*tracing_session)) {
3093 ds_config.set_session_initiator(
3094 DataSourceConfig::SESSION_INITIATOR_TRUSTED_SYSTEM);
3095 } else {
3096 // Unset in case the consumer set it.
3097 // We need to be able to trust this field.
3098 ds_config.set_session_initiator(
3099 DataSourceConfig::SESSION_INITIATOR_UNSPECIFIED);
3100 }
3101 ds_config.set_tracing_session_id(tracing_session->id);
3102 BufferID global_id = tracing_session->buffers_index[relative_buffer_id];
3103 PERFETTO_DCHECK(global_id);
3104 ds_config.set_target_buffer(global_id);
3105
3106 PERFETTO_DLOG("Setting up data source %s with target buffer %" PRIu16,
3107 ds_config.name().c_str(), global_id);
3108 if (!producer->shared_memory()) {
3109 // Determine the SMB page size. Must be an integer multiple of 4k.
3110 // As for the SMB size below, the decision tree is as follows:
3111 // 1. Give priority to what is defined in the trace config.
3112 // 2. If unset give priority to the hint passed by the producer.
3113 // 3. Keep within bounds and ensure it's a multiple of 4k.
3114 size_t page_size = producer_config.page_size_kb() * 1024;
3115 if (page_size == 0)
3116 page_size = producer->shmem_page_size_hint_bytes_;
3117
3118 // Determine the SMB size. Must be an integer multiple of the SMB page size.
3119 // The decision tree is as follows:
3120 // 1. Give priority to what defined in the trace config.
3121 // 2. If unset give priority to the hint passed by the producer.
3122 // 3. Keep within bounds and ensure it's a multiple of the page size.
3123 size_t shm_size = producer_config.shm_size_kb() * 1024;
3124 if (shm_size == 0)
3125 shm_size = producer->shmem_size_hint_bytes_;
3126
3127 auto valid_sizes = EnsureValidShmSizes(shm_size, page_size);
3128 if (valid_sizes != std::tie(shm_size, page_size)) {
3129 PERFETTO_DLOG(
3130 "Invalid configured SMB sizes: shm_size %zu page_size %zu. Falling "
3131 "back to shm_size %zu page_size %zu.",
3132 shm_size, page_size, std::get<0>(valid_sizes),
3133 std::get<1>(valid_sizes));
3134 }
3135 std::tie(shm_size, page_size) = valid_sizes;
3136
3137 // TODO(primiano): right now Create() will suicide in case of OOM if the
3138 // mmap fails. We should instead gracefully fail the request and tell the
3139 // client to go away.
3140 PERFETTO_DLOG("Creating SMB of %zu KB for producer \"%s\"", shm_size / 1024,
3141 producer->name_.c_str());
3142 auto shared_memory = shm_factory_->CreateSharedMemory(shm_size);
3143 producer->SetupSharedMemory(std::move(shared_memory), page_size,
3144 /*provided_by_producer=*/false);
3145 }
3146 producer->SetupDataSource(inst_id, ds_config);
3147 return ds_instance;
3148 }
3149
3150 // Note: all the fields % *_trusted ones are untrusted, as in, the Producer
3151 // might be lying / returning garbage contents. |src| and |size| can be trusted
3152 // in terms of being a valid pointer, but not the contents.
CopyProducerPageIntoLogBuffer(ProducerID producer_id_trusted,const ClientIdentity & client_identity_trusted,WriterID writer_id,ChunkID chunk_id,BufferID buffer_id,uint16_t num_fragments,uint8_t chunk_flags,bool chunk_complete,const uint8_t * src,size_t size)3153 void TracingServiceImpl::CopyProducerPageIntoLogBuffer(
3154 ProducerID producer_id_trusted,
3155 const ClientIdentity& client_identity_trusted,
3156 WriterID writer_id,
3157 ChunkID chunk_id,
3158 BufferID buffer_id,
3159 uint16_t num_fragments,
3160 uint8_t chunk_flags,
3161 bool chunk_complete,
3162 const uint8_t* src,
3163 size_t size) {
3164 PERFETTO_DCHECK_THREAD(thread_checker_);
3165
3166 ProducerEndpointImpl* producer = GetProducer(producer_id_trusted);
3167 if (!producer) {
3168 PERFETTO_DFATAL("Producer not found.");
3169 chunks_discarded_++;
3170 return;
3171 }
3172
3173 TraceBuffer* buf = GetBufferByID(buffer_id);
3174 if (!buf) {
3175 PERFETTO_DLOG("Could not find target buffer %" PRIu16
3176 " for producer %" PRIu16,
3177 buffer_id, producer_id_trusted);
3178 chunks_discarded_++;
3179 return;
3180 }
3181
3182 // Verify that the producer is actually allowed to write into the target
3183 // buffer specified in the request. This prevents a malicious producer from
3184 // injecting data into a log buffer that belongs to a tracing session the
3185 // producer is not part of.
3186 if (!producer->is_allowed_target_buffer(buffer_id)) {
3187 PERFETTO_ELOG("Producer %" PRIu16
3188 " tried to write into forbidden target buffer %" PRIu16,
3189 producer_id_trusted, buffer_id);
3190 PERFETTO_DFATAL("Forbidden target buffer");
3191 chunks_discarded_++;
3192 return;
3193 }
3194
3195 // If the writer was registered by the producer, it should only write into the
3196 // buffer it was registered with.
3197 std::optional<BufferID> associated_buffer =
3198 producer->buffer_id_for_writer(writer_id);
3199 if (associated_buffer && *associated_buffer != buffer_id) {
3200 PERFETTO_ELOG("Writer %" PRIu16 " of producer %" PRIu16
3201 " was registered to write into target buffer %" PRIu16
3202 ", but tried to write into buffer %" PRIu16,
3203 writer_id, producer_id_trusted, *associated_buffer,
3204 buffer_id);
3205 PERFETTO_DFATAL("Wrong target buffer");
3206 chunks_discarded_++;
3207 return;
3208 }
3209
3210 buf->CopyChunkUntrusted(producer_id_trusted, client_identity_trusted,
3211 writer_id, chunk_id, num_fragments, chunk_flags,
3212 chunk_complete, src, size);
3213 }
3214
ApplyChunkPatches(ProducerID producer_id_trusted,const std::vector<CommitDataRequest::ChunkToPatch> & chunks_to_patch)3215 void TracingServiceImpl::ApplyChunkPatches(
3216 ProducerID producer_id_trusted,
3217 const std::vector<CommitDataRequest::ChunkToPatch>& chunks_to_patch) {
3218 PERFETTO_DCHECK_THREAD(thread_checker_);
3219
3220 for (const auto& chunk : chunks_to_patch) {
3221 const ChunkID chunk_id = static_cast<ChunkID>(chunk.chunk_id());
3222 const WriterID writer_id = static_cast<WriterID>(chunk.writer_id());
3223 TraceBuffer* buf =
3224 GetBufferByID(static_cast<BufferID>(chunk.target_buffer()));
3225 static_assert(std::numeric_limits<ChunkID>::max() == kMaxChunkID,
3226 "Add a '|| chunk_id > kMaxChunkID' below if this fails");
3227 if (!writer_id || writer_id > kMaxWriterID || !buf) {
3228 // This can genuinely happen when the trace is stopped. The producers
3229 // might see the stop signal with some delay and try to keep sending
3230 // patches left soon after.
3231 PERFETTO_DLOG(
3232 "Received invalid chunks_to_patch request from Producer: %" PRIu16
3233 ", BufferID: %" PRIu32 " ChunkdID: %" PRIu32 " WriterID: %" PRIu16,
3234 producer_id_trusted, chunk.target_buffer(), chunk_id, writer_id);
3235 patches_discarded_ += static_cast<uint64_t>(chunk.patches_size());
3236 continue;
3237 }
3238
3239 // Note, there's no need to validate that the producer is allowed to write
3240 // to the specified buffer ID (or that it's the correct buffer ID for a
3241 // registered TraceWriter). That's because TraceBuffer uses the producer ID
3242 // and writer ID to look up the chunk to patch. If the producer specifies an
3243 // incorrect buffer, this lookup will fail and TraceBuffer will ignore the
3244 // patches. Because the producer ID is trusted, there's also no way for a
3245 // malicious producer to patch another producer's data.
3246
3247 // Speculate on the fact that there are going to be a limited amount of
3248 // patches per request, so we can allocate the |patches| array on the stack.
3249 std::array<TraceBuffer::Patch, 1024> patches; // Uninitialized.
3250 if (chunk.patches().size() > patches.size()) {
3251 PERFETTO_ELOG("Too many patches (%zu) batched in the same request",
3252 patches.size());
3253 PERFETTO_DFATAL("Too many patches");
3254 patches_discarded_ += static_cast<uint64_t>(chunk.patches_size());
3255 continue;
3256 }
3257
3258 size_t i = 0;
3259 for (const auto& patch : chunk.patches()) {
3260 const std::string& patch_data = patch.data();
3261 if (patch_data.size() != patches[i].data.size()) {
3262 PERFETTO_ELOG("Received patch from producer: %" PRIu16
3263 " of unexpected size %zu",
3264 producer_id_trusted, patch_data.size());
3265 patches_discarded_++;
3266 continue;
3267 }
3268 patches[i].offset_untrusted = patch.offset();
3269 memcpy(&patches[i].data[0], patch_data.data(), patches[i].data.size());
3270 i++;
3271 }
3272 buf->TryPatchChunkContents(producer_id_trusted, writer_id, chunk_id,
3273 &patches[0], i, chunk.has_more_patches());
3274 }
3275 }
3276
GetDetachedSession(uid_t uid,const std::string & key)3277 TracingServiceImpl::TracingSession* TracingServiceImpl::GetDetachedSession(
3278 uid_t uid,
3279 const std::string& key) {
3280 PERFETTO_DCHECK_THREAD(thread_checker_);
3281 for (auto& kv : tracing_sessions_) {
3282 TracingSession* session = &kv.second;
3283 if (session->consumer_uid == uid && session->detach_key == key) {
3284 PERFETTO_DCHECK(session->consumer_maybe_null == nullptr);
3285 return session;
3286 }
3287 }
3288 return nullptr;
3289 }
3290
GetTracingSession(TracingSessionID tsid)3291 TracingServiceImpl::TracingSession* TracingServiceImpl::GetTracingSession(
3292 TracingSessionID tsid) {
3293 PERFETTO_DCHECK_THREAD(thread_checker_);
3294 auto it = tsid ? tracing_sessions_.find(tsid) : tracing_sessions_.end();
3295 if (it == tracing_sessions_.end())
3296 return nullptr;
3297 return &it->second;
3298 }
3299
3300 TracingServiceImpl::TracingSession*
GetTracingSessionByUniqueName(const std::string & unique_session_name)3301 TracingServiceImpl::GetTracingSessionByUniqueName(
3302 const std::string& unique_session_name) {
3303 PERFETTO_DCHECK_THREAD(thread_checker_);
3304 if (unique_session_name.empty()) {
3305 return nullptr;
3306 }
3307 for (auto& session_id_and_session : tracing_sessions_) {
3308 TracingSession& session = session_id_and_session.second;
3309 if (session.state == TracingSession::CLONED_READ_ONLY) {
3310 continue;
3311 }
3312 if (session.config.unique_session_name() == unique_session_name) {
3313 return &session;
3314 }
3315 }
3316 return nullptr;
3317 }
3318
3319 TracingServiceImpl::TracingSession*
FindTracingSessionWithMaxBugreportScore()3320 TracingServiceImpl::FindTracingSessionWithMaxBugreportScore() {
3321 TracingSession* max_session = nullptr;
3322 for (auto& session_id_and_session : tracing_sessions_) {
3323 auto& session = session_id_and_session.second;
3324 const int32_t score = session.config.bugreport_score();
3325 // Exclude sessions with 0 (or below) score. By default tracing sessions
3326 // should NOT be eligible to be attached to bugreports.
3327 if (score <= 0 || session.state != TracingSession::STARTED)
3328 continue;
3329
3330 if (!max_session || score > max_session->config.bugreport_score())
3331 max_session = &session;
3332 }
3333 return max_session;
3334 }
3335
GetNextProducerID()3336 ProducerID TracingServiceImpl::GetNextProducerID() {
3337 PERFETTO_DCHECK_THREAD(thread_checker_);
3338 PERFETTO_CHECK(producers_.size() < kMaxProducerID);
3339 do {
3340 ++last_producer_id_;
3341 } while (producers_.count(last_producer_id_) || last_producer_id_ == 0);
3342 PERFETTO_DCHECK(last_producer_id_ > 0 && last_producer_id_ <= kMaxProducerID);
3343 return last_producer_id_;
3344 }
3345
GetBufferByID(BufferID buffer_id)3346 TraceBuffer* TracingServiceImpl::GetBufferByID(BufferID buffer_id) {
3347 auto buf_iter = buffers_.find(buffer_id);
3348 if (buf_iter == buffers_.end())
3349 return nullptr;
3350 return &*buf_iter->second;
3351 }
3352
OnStartTriggersTimeout(TracingSessionID tsid)3353 void TracingServiceImpl::OnStartTriggersTimeout(TracingSessionID tsid) {
3354 // Skip entirely the flush if the trace session doesn't exist anymore.
3355 // This is to prevent misleading error messages to be logged.
3356 //
3357 // if the trace has started from the trigger we rely on
3358 // the |stop_delay_ms| from the trigger so don't flush and
3359 // disable if we've moved beyond a CONFIGURED state
3360 auto* tracing_session_ptr = GetTracingSession(tsid);
3361 if (tracing_session_ptr &&
3362 tracing_session_ptr->state == TracingSession::CONFIGURED) {
3363 PERFETTO_DLOG("Disabling TracingSession %" PRIu64
3364 " since no triggers activated.",
3365 tsid);
3366 // No data should be returned from ReadBuffers() regardless of if we
3367 // call FreeBuffers() or DisableTracing(). This is because in
3368 // STOP_TRACING we need this promise in either case, and using
3369 // DisableTracing() allows a graceful shutdown. Consumers can follow
3370 // their normal path and check the buffers through ReadBuffers() and
3371 // the code won't hang because the tracing session will still be
3372 // alive just disabled.
3373 DisableTracing(tsid);
3374 }
3375 }
3376
UpdateMemoryGuardrail()3377 void TracingServiceImpl::UpdateMemoryGuardrail() {
3378 #if PERFETTO_BUILDFLAG(PERFETTO_WATCHDOG)
3379 uint64_t total_buffer_bytes = 0;
3380
3381 // Sum up all the shared memory buffers.
3382 for (const auto& id_to_producer : producers_) {
3383 if (id_to_producer.second->shared_memory())
3384 total_buffer_bytes += id_to_producer.second->shared_memory()->size();
3385 }
3386
3387 // Sum up all the trace buffers.
3388 for (const auto& id_to_buffer : buffers_) {
3389 total_buffer_bytes += id_to_buffer.second->size();
3390 }
3391
3392 // Sum up all the cloned traced buffers.
3393 for (const auto& id_to_ts : tracing_sessions_) {
3394 const TracingSession& ts = id_to_ts.second;
3395 for (const auto& id_to_clone_op : ts.pending_clones) {
3396 const PendingClone& clone_op = id_to_clone_op.second;
3397 for (const std::unique_ptr<TraceBuffer>& buf : clone_op.buffers) {
3398 if (buf) {
3399 total_buffer_bytes += buf->size();
3400 }
3401 }
3402 }
3403 }
3404
3405 // Set the guard rail to 32MB + the sum of all the buffers over a 30 second
3406 // interval.
3407 uint64_t guardrail = base::kWatchdogDefaultMemorySlack + total_buffer_bytes;
3408 base::Watchdog::GetInstance()->SetMemoryLimit(guardrail, 30 * 1000);
3409 #endif
3410 }
3411
PeriodicSnapshotTask(TracingSessionID tsid)3412 void TracingServiceImpl::PeriodicSnapshotTask(TracingSessionID tsid) {
3413 auto* tracing_session = GetTracingSession(tsid);
3414 if (!tracing_session)
3415 return;
3416 if (tracing_session->state != TracingSession::STARTED)
3417 return;
3418 tracing_session->should_emit_sync_marker = true;
3419 tracing_session->should_emit_stats = true;
3420 MaybeSnapshotClocksIntoRingBuffer(tracing_session);
3421 }
3422
SnapshotLifecycleEvent(TracingSession * tracing_session,uint32_t field_id,bool snapshot_clocks)3423 void TracingServiceImpl::SnapshotLifecycleEvent(TracingSession* tracing_session,
3424 uint32_t field_id,
3425 bool snapshot_clocks) {
3426 // field_id should be an id of a field in TracingServiceEvent.
3427 auto& lifecycle_events = tracing_session->lifecycle_events;
3428 auto event_it =
3429 std::find_if(lifecycle_events.begin(), lifecycle_events.end(),
3430 [field_id](const TracingSession::LifecycleEvent& event) {
3431 return event.field_id == field_id;
3432 });
3433
3434 TracingSession::LifecycleEvent* event;
3435 if (event_it == lifecycle_events.end()) {
3436 lifecycle_events.emplace_back(field_id);
3437 event = &lifecycle_events.back();
3438 } else {
3439 event = &*event_it;
3440 }
3441
3442 // Snapshot the clocks before capturing the timestamp for the event so we can
3443 // use this snapshot to resolve the event timestamp if necessary.
3444 if (snapshot_clocks)
3445 MaybeSnapshotClocksIntoRingBuffer(tracing_session);
3446
3447 // Erase before emplacing to prevent a unncessary doubling of memory if
3448 // not needed.
3449 if (event->timestamps.size() >= event->max_size) {
3450 event->timestamps.erase_front(1 + event->timestamps.size() -
3451 event->max_size);
3452 }
3453 event->timestamps.emplace_back(clock_->GetBootTimeNs().count());
3454 }
3455
SetSingleLifecycleEvent(TracingSession * tracing_session,uint32_t field_id,int64_t boot_timestamp_ns)3456 void TracingServiceImpl::SetSingleLifecycleEvent(
3457 TracingSession* tracing_session,
3458 uint32_t field_id,
3459 int64_t boot_timestamp_ns) {
3460 // field_id should be an id of a field in TracingServiceEvent.
3461 auto& lifecycle_events = tracing_session->lifecycle_events;
3462 auto event_it =
3463 std::find_if(lifecycle_events.begin(), lifecycle_events.end(),
3464 [field_id](const TracingSession::LifecycleEvent& event) {
3465 return event.field_id == field_id;
3466 });
3467
3468 TracingSession::LifecycleEvent* event;
3469 if (event_it == lifecycle_events.end()) {
3470 lifecycle_events.emplace_back(field_id);
3471 event = &lifecycle_events.back();
3472 } else {
3473 event = &*event_it;
3474 }
3475
3476 event->timestamps.clear();
3477 event->timestamps.emplace_back(boot_timestamp_ns);
3478 }
3479
MaybeSnapshotClocksIntoRingBuffer(TracingSession * tracing_session)3480 void TracingServiceImpl::MaybeSnapshotClocksIntoRingBuffer(
3481 TracingSession* tracing_session) {
3482 if (tracing_session->config.builtin_data_sources()
3483 .disable_clock_snapshotting()) {
3484 return;
3485 }
3486
3487 // We are making an explicit copy of the latest snapshot (if it exists)
3488 // because SnapshotClocks reads this data and computes the drift based on its
3489 // content. If the clock drift is high enough, it will update the contents of
3490 // |snapshot| and return true. Otherwise, it will return false.
3491 TracingSession::ClockSnapshotData snapshot =
3492 tracing_session->clock_snapshot_ring_buffer.empty()
3493 ? TracingSession::ClockSnapshotData()
3494 : tracing_session->clock_snapshot_ring_buffer.back();
3495 bool did_update = SnapshotClocks(&snapshot);
3496 if (did_update) {
3497 // This means clocks drifted enough since last snapshot. See the comment
3498 // in SnapshotClocks.
3499 auto* snapshot_buffer = &tracing_session->clock_snapshot_ring_buffer;
3500
3501 // Erase before emplacing to prevent a unncessary doubling of memory if
3502 // not needed.
3503 static constexpr uint32_t kClockSnapshotRingBufferSize = 16;
3504 if (snapshot_buffer->size() >= kClockSnapshotRingBufferSize) {
3505 snapshot_buffer->erase_front(1 + snapshot_buffer->size() -
3506 kClockSnapshotRingBufferSize);
3507 }
3508 snapshot_buffer->emplace_back(std::move(snapshot));
3509 }
3510 }
3511
3512 // Returns true when the data in |snapshot_data| is updated with the new state
3513 // of the clocks and false otherwise.
SnapshotClocks(TracingSession::ClockSnapshotData * snapshot_data)3514 bool TracingServiceImpl::SnapshotClocks(
3515 TracingSession::ClockSnapshotData* snapshot_data) {
3516 // Minimum drift that justifies replacing a prior clock snapshot that hasn't
3517 // been emitted into the trace yet (see comment below).
3518 static constexpr int64_t kSignificantDriftNs = 10 * 1000 * 1000; // 10 ms
3519
3520 TracingSession::ClockSnapshotData new_snapshot_data =
3521 base::CaptureClockSnapshots();
3522 // If we're about to update a session's latest clock snapshot that hasn't been
3523 // emitted into the trace yet, check whether the clocks have drifted enough to
3524 // warrant overriding the current snapshot values. The older snapshot would be
3525 // valid for a larger part of the currently buffered trace data because the
3526 // clock sync protocol in trace processor uses the latest clock <= timestamp
3527 // to translate times (see https://perfetto.dev/docs/concepts/clock-sync), so
3528 // we try to keep it if we can.
3529 if (!snapshot_data->empty()) {
3530 PERFETTO_DCHECK(snapshot_data->size() == new_snapshot_data.size());
3531 PERFETTO_DCHECK((*snapshot_data)[0].clock_id ==
3532 protos::gen::BUILTIN_CLOCK_BOOTTIME);
3533
3534 bool update_snapshot = false;
3535 uint64_t old_boot_ns = (*snapshot_data)[0].timestamp;
3536 uint64_t new_boot_ns = new_snapshot_data[0].timestamp;
3537 int64_t boot_diff =
3538 static_cast<int64_t>(new_boot_ns) - static_cast<int64_t>(old_boot_ns);
3539
3540 for (size_t i = 1; i < snapshot_data->size(); i++) {
3541 uint64_t old_ns = (*snapshot_data)[i].timestamp;
3542 uint64_t new_ns = new_snapshot_data[i].timestamp;
3543
3544 int64_t diff =
3545 static_cast<int64_t>(new_ns) - static_cast<int64_t>(old_ns);
3546
3547 // Compare the boottime delta against the delta of this clock.
3548 if (std::abs(boot_diff - diff) >= kSignificantDriftNs) {
3549 update_snapshot = true;
3550 break;
3551 }
3552 }
3553 if (!update_snapshot)
3554 return false;
3555 snapshot_data->clear();
3556 }
3557
3558 *snapshot_data = std::move(new_snapshot_data);
3559 return true;
3560 }
3561
EmitClockSnapshot(TracingSession * tracing_session,TracingSession::ClockSnapshotData snapshot_data,std::vector<TracePacket> * packets)3562 void TracingServiceImpl::EmitClockSnapshot(
3563 TracingSession* tracing_session,
3564 TracingSession::ClockSnapshotData snapshot_data,
3565 std::vector<TracePacket>* packets) {
3566 PERFETTO_DCHECK(!tracing_session->config.builtin_data_sources()
3567 .disable_clock_snapshotting());
3568
3569 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3570 auto* snapshot = packet->set_clock_snapshot();
3571
3572 protos::gen::BuiltinClock trace_clock =
3573 tracing_session->config.builtin_data_sources().primary_trace_clock();
3574 if (!trace_clock)
3575 trace_clock = protos::gen::BUILTIN_CLOCK_BOOTTIME;
3576 snapshot->set_primary_trace_clock(
3577 static_cast<protos::pbzero::BuiltinClock>(trace_clock));
3578
3579 for (auto& clock_id_and_ts : snapshot_data) {
3580 auto* c = snapshot->add_clocks();
3581 c->set_clock_id(clock_id_and_ts.clock_id);
3582 c->set_timestamp(clock_id_and_ts.timestamp);
3583 }
3584
3585 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3586 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3587 SerializeAndAppendPacket(packets, packet.SerializeAsArray());
3588 }
3589
EmitSyncMarker(std::vector<TracePacket> * packets)3590 void TracingServiceImpl::EmitSyncMarker(std::vector<TracePacket>* packets) {
3591 // The sync marks are used to tokenize large traces efficiently.
3592 // See description in trace_packet.proto.
3593 if (sync_marker_packet_size_ == 0) {
3594 // The marker ABI expects that the marker is written after the uid.
3595 // Protozero guarantees that fields are written in the same order of the
3596 // calls. The ResynchronizeTraceStreamUsingSyncMarker test verifies the ABI.
3597 protozero::StaticBuffered<protos::pbzero::TracePacket> packet(
3598 &sync_marker_packet_[0], sizeof(sync_marker_packet_));
3599 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3600 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3601
3602 // Keep this last.
3603 packet->set_synchronization_marker(kSyncMarker, sizeof(kSyncMarker));
3604 sync_marker_packet_size_ = packet.Finalize();
3605 }
3606 packets->emplace_back();
3607 packets->back().AddSlice(&sync_marker_packet_[0], sync_marker_packet_size_);
3608 }
3609
EmitStats(TracingSession * tracing_session,std::vector<TracePacket> * packets)3610 void TracingServiceImpl::EmitStats(TracingSession* tracing_session,
3611 std::vector<TracePacket>* packets) {
3612 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3613 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3614 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3615 GetTraceStats(tracing_session).Serialize(packet->set_trace_stats());
3616 SerializeAndAppendPacket(packets, packet.SerializeAsArray());
3617 }
3618
GetTraceStats(TracingSession * tracing_session)3619 TraceStats TracingServiceImpl::GetTraceStats(TracingSession* tracing_session) {
3620 TraceStats trace_stats;
3621 trace_stats.set_producers_connected(static_cast<uint32_t>(producers_.size()));
3622 trace_stats.set_producers_seen(last_producer_id_);
3623 trace_stats.set_data_sources_registered(
3624 static_cast<uint32_t>(data_sources_.size()));
3625 trace_stats.set_data_sources_seen(last_data_source_instance_id_);
3626 trace_stats.set_tracing_sessions(
3627 static_cast<uint32_t>(tracing_sessions_.size()));
3628 trace_stats.set_total_buffers(static_cast<uint32_t>(buffers_.size()));
3629 trace_stats.set_chunks_discarded(chunks_discarded_);
3630 trace_stats.set_patches_discarded(patches_discarded_);
3631 trace_stats.set_invalid_packets(tracing_session->invalid_packets);
3632 trace_stats.set_flushes_requested(tracing_session->flushes_requested);
3633 trace_stats.set_flushes_succeeded(tracing_session->flushes_succeeded);
3634 trace_stats.set_flushes_failed(tracing_session->flushes_failed);
3635 trace_stats.set_final_flush_outcome(tracing_session->final_flush_outcome);
3636
3637 if (tracing_session->trace_filter) {
3638 auto* filt_stats = trace_stats.mutable_filter_stats();
3639 filt_stats->set_input_packets(tracing_session->filter_input_packets);
3640 filt_stats->set_input_bytes(tracing_session->filter_input_bytes);
3641 filt_stats->set_output_bytes(tracing_session->filter_output_bytes);
3642 filt_stats->set_errors(tracing_session->filter_errors);
3643 filt_stats->set_time_taken_ns(tracing_session->filter_time_taken_ns);
3644 for (uint64_t value : tracing_session->filter_bytes_discarded_per_buffer)
3645 filt_stats->add_bytes_discarded_per_buffer(value);
3646 }
3647
3648 for (BufferID buf_id : tracing_session->buffers_index) {
3649 TraceBuffer* buf = GetBufferByID(buf_id);
3650 if (!buf) {
3651 PERFETTO_DFATAL("Buffer not found.");
3652 continue;
3653 }
3654 *trace_stats.add_buffer_stats() = buf->stats();
3655 } // for (buf in session).
3656
3657 if (!tracing_session->config.builtin_data_sources()
3658 .disable_chunk_usage_histograms()) {
3659 // Emit chunk usage stats broken down by sequence ID (i.e. by trace-writer).
3660 // Writer stats are updated by each TraceBuffer object at ReadBuffers time,
3661 // and there can be >1 buffer per session. A trace writer never writes to
3662 // more than one buffer (it's technically allowed but doesn't happen in the
3663 // current impl of the tracing SDK).
3664
3665 bool has_written_bucket_definition = false;
3666 uint32_t buf_idx = static_cast<uint32_t>(-1);
3667 for (const BufferID buf_id : tracing_session->buffers_index) {
3668 ++buf_idx;
3669 const TraceBuffer* buf = GetBufferByID(buf_id);
3670 if (!buf)
3671 continue;
3672 for (auto it = buf->writer_stats().GetIterator(); it; ++it) {
3673 const auto& hist = it.value().used_chunk_hist;
3674 ProducerID p;
3675 WriterID w;
3676 GetProducerAndWriterID(it.key(), &p, &w);
3677 if (!has_written_bucket_definition) {
3678 // Serialize one-off the histogram bucket definition, which is the
3679 // same for all entries in the map.
3680 has_written_bucket_definition = true;
3681 // The -1 in the loop below is to skip the implicit overflow bucket.
3682 for (size_t i = 0; i < hist.num_buckets() - 1; ++i) {
3683 trace_stats.add_chunk_payload_histogram_def(hist.GetBucketThres(i));
3684 }
3685 } // if(!has_written_bucket_definition)
3686 auto* wri_stats = trace_stats.add_writer_stats();
3687 wri_stats->set_sequence_id(
3688 tracing_session->GetPacketSequenceID(kDefaultMachineID, p, w));
3689 wri_stats->set_buffer(buf_idx);
3690 for (size_t i = 0; i < hist.num_buckets(); ++i) {
3691 wri_stats->add_chunk_payload_histogram_counts(hist.GetBucketCount(i));
3692 wri_stats->add_chunk_payload_histogram_sum(hist.GetBucketSum(i));
3693 }
3694 } // for each sequence (writer).
3695 } // for each buffer.
3696 } // if (!disable_chunk_usage_histograms)
3697
3698 return trace_stats;
3699 }
3700
EmitUuid(TracingSession * tracing_session,std::vector<TracePacket> * packets)3701 void TracingServiceImpl::EmitUuid(TracingSession* tracing_session,
3702 std::vector<TracePacket>* packets) {
3703 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3704 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3705 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3706 auto* uuid = packet->set_trace_uuid();
3707 uuid->set_lsb(tracing_session->trace_uuid.lsb());
3708 uuid->set_msb(tracing_session->trace_uuid.msb());
3709 SerializeAndAppendPacket(packets, packet.SerializeAsArray());
3710 }
3711
MaybeEmitTraceConfig(TracingSession * tracing_session,std::vector<TracePacket> * packets)3712 void TracingServiceImpl::MaybeEmitTraceConfig(
3713 TracingSession* tracing_session,
3714 std::vector<TracePacket>* packets) {
3715 if (tracing_session->did_emit_initial_packets)
3716 return;
3717 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3718 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3719 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3720 tracing_session->config.Serialize(packet->set_trace_config());
3721 SerializeAndAppendPacket(packets, packet.SerializeAsArray());
3722 }
3723
EmitSystemInfo(std::vector<TracePacket> * packets)3724 void TracingServiceImpl::EmitSystemInfo(std::vector<TracePacket>* packets) {
3725 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3726 auto* info = packet->set_system_info();
3727 info->set_tracing_service_version(base::GetVersionString());
3728
3729 std::optional<int32_t> tzoff = base::GetTimezoneOffsetMins();
3730 if (tzoff.has_value())
3731 info->set_timezone_off_mins(*tzoff);
3732
3733 #if !PERFETTO_BUILDFLAG(PERFETTO_OS_WIN) && \
3734 !PERFETTO_BUILDFLAG(PERFETTO_OS_NACL)
3735 struct utsname uname_info;
3736 if (uname(&uname_info) == 0) {
3737 auto* utsname_info = info->set_utsname();
3738 utsname_info->set_sysname(uname_info.sysname);
3739 utsname_info->set_version(uname_info.version);
3740 utsname_info->set_machine(uname_info.machine);
3741 utsname_info->set_release(uname_info.release);
3742 }
3743 info->set_page_size(static_cast<uint32_t>(sysconf(_SC_PAGESIZE)));
3744 info->set_num_cpus(static_cast<uint32_t>(sysconf(_SC_NPROCESSORS_CONF)));
3745 #endif // !PERFETTO_BUILDFLAG(PERFETTO_OS_WIN)
3746 #if PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID)
3747 std::string fingerprint_value = base::GetAndroidProp("ro.build.fingerprint");
3748 if (!fingerprint_value.empty()) {
3749 info->set_android_build_fingerprint(fingerprint_value);
3750 } else {
3751 PERFETTO_ELOG("Unable to read ro.build.fingerprint");
3752 }
3753
3754 std::string device_manufacturer_value =
3755 base::GetAndroidProp("ro.product.manufacturer");
3756 if (!device_manufacturer_value.empty()) {
3757 info->set_android_device_manufacturer(device_manufacturer_value);
3758 } else {
3759 PERFETTO_ELOG("Unable to read ro.product.manufacturer");
3760 }
3761
3762 std::string sdk_str_value = base::GetAndroidProp("ro.build.version.sdk");
3763 std::optional<uint64_t> sdk_value = base::StringToUInt64(sdk_str_value);
3764 if (sdk_value.has_value()) {
3765 info->set_android_sdk_version(*sdk_value);
3766 } else {
3767 PERFETTO_ELOG("Unable to read ro.build.version.sdk");
3768 }
3769
3770 std::string soc_model_value = base::GetAndroidProp("ro.soc.model");
3771 if (!soc_model_value.empty()) {
3772 info->set_android_soc_model(soc_model_value);
3773 } else {
3774 PERFETTO_ELOG("Unable to read ro.soc.model");
3775 }
3776
3777 // guest_soc model is not always present
3778 std::string guest_soc_model_value =
3779 base::GetAndroidProp("ro.boot.guest_soc.model");
3780 if (!guest_soc_model_value.empty()) {
3781 info->set_android_guest_soc_model(guest_soc_model_value);
3782 }
3783
3784 std::string hw_rev_value = base::GetAndroidProp("ro.boot.hardware.revision");
3785 if (!hw_rev_value.empty()) {
3786 info->set_android_hardware_revision(hw_rev_value);
3787 } else {
3788 PERFETTO_ELOG("Unable to read ro.boot.hardware.revision");
3789 }
3790
3791 std::string hw_ufs_value = base::GetAndroidProp("ro.boot.hardware.ufs");
3792 if (!hw_ufs_value.empty()) {
3793 info->set_android_storage_model(hw_ufs_value);
3794 } else {
3795 PERFETTO_ELOG("Unable to read ro.boot.hardware.ufs");
3796 }
3797
3798 std::string hw_ddr_value = base::GetAndroidProp("ro.boot.hardware.ddr");
3799 if (!hw_ddr_value.empty()) {
3800 info->set_android_ram_model(hw_ddr_value);
3801 } else {
3802 PERFETTO_ELOG("Unable to read ro.boot.hardware.ddr");
3803 }
3804
3805 #endif // PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID)
3806 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3807 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3808 SerializeAndAppendPacket(packets, packet.SerializeAsArray());
3809 }
3810
EmitLifecycleEvents(TracingSession * tracing_session,std::vector<TracePacket> * packets)3811 void TracingServiceImpl::EmitLifecycleEvents(
3812 TracingSession* tracing_session,
3813 std::vector<TracePacket>* packets) {
3814 using TimestampedPacket =
3815 std::pair<int64_t /* ts */, std::vector<uint8_t> /* serialized packet */>;
3816
3817 std::vector<TimestampedPacket> timestamped_packets;
3818 for (auto& event : tracing_session->lifecycle_events) {
3819 for (int64_t ts : event.timestamps) {
3820 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3821 packet->set_timestamp(static_cast<uint64_t>(ts));
3822 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3823 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3824
3825 auto* service_event = packet->set_service_event();
3826 service_event->AppendVarInt(event.field_id, 1);
3827 timestamped_packets.emplace_back(ts, packet.SerializeAsArray());
3828 }
3829 event.timestamps.clear();
3830 }
3831
3832 if (tracing_session->slow_start_event.has_value()) {
3833 const TracingSession::ArbitraryLifecycleEvent& event =
3834 *tracing_session->slow_start_event;
3835 timestamped_packets.emplace_back(event.timestamp, std::move(event.data));
3836 }
3837 tracing_session->slow_start_event.reset();
3838
3839 for (auto& event : tracing_session->last_flush_events) {
3840 timestamped_packets.emplace_back(event.timestamp, std::move(event.data));
3841 }
3842 tracing_session->last_flush_events.clear();
3843
3844 for (size_t i = 0; i < tracing_session->buffer_cloned_timestamps.size();
3845 i++) {
3846 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3847 int64_t ts = tracing_session->buffer_cloned_timestamps[i];
3848 packet->set_timestamp(static_cast<uint64_t>(ts));
3849 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3850 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3851
3852 auto* service_event = packet->set_service_event();
3853 service_event->set_buffer_cloned(static_cast<uint32_t>(i));
3854
3855 timestamped_packets.emplace_back(ts, packet.SerializeAsArray());
3856 }
3857 tracing_session->buffer_cloned_timestamps.clear();
3858
3859 // We sort by timestamp here to ensure that the "sequence" of lifecycle
3860 // packets has monotonic timestamps like other sequences in the trace.
3861 // Note that these events could still be out of order with respect to other
3862 // events on the service packet sequence (e.g. trigger received packets).
3863 std::sort(timestamped_packets.begin(), timestamped_packets.end(),
3864 [](const TimestampedPacket& a, const TimestampedPacket& b) {
3865 return a.first < b.first;
3866 });
3867
3868 for (auto& pair : timestamped_packets)
3869 SerializeAndAppendPacket(packets, std::move(pair.second));
3870 }
3871
MaybeEmitRemoteClockSync(TracingSession * tracing_session,std::vector<TracePacket> * packets)3872 void TracingServiceImpl::MaybeEmitRemoteClockSync(
3873 TracingSession* tracing_session,
3874 std::vector<TracePacket>* packets) {
3875 if (tracing_session->did_emit_remote_clock_sync_)
3876 return;
3877
3878 std::unordered_set<MachineID> did_emit_machines;
3879 for (const auto& id_and_relay_client : relay_clients_) {
3880 const auto& relay_client = id_and_relay_client.second;
3881 auto machine_id = relay_client->machine_id();
3882 if (did_emit_machines.find(machine_id) != did_emit_machines.end())
3883 continue; // Already emitted for the machine (e.g. multiple clients).
3884
3885 auto& sync_clock_snapshots = relay_client->synced_clocks();
3886 if (sync_clock_snapshots.empty()) {
3887 PERFETTO_DLOG("Clock not synchronized for machine ID = %" PRIu32,
3888 machine_id);
3889 continue;
3890 }
3891
3892 // Don't emit twice for the same machine.
3893 did_emit_machines.insert(machine_id);
3894
3895 protozero::HeapBuffered<protos::pbzero::TracePacket> sync_packet;
3896 sync_packet->set_machine_id(machine_id);
3897 sync_packet->set_trusted_uid(static_cast<int32_t>(uid_));
3898 auto* remote_clock_sync = sync_packet->set_remote_clock_sync();
3899 for (const auto& sync_exchange : relay_client->synced_clocks()) {
3900 auto* sync_exchange_msg = remote_clock_sync->add_synced_clocks();
3901
3902 auto* client_snapshots = sync_exchange_msg->set_client_clocks();
3903 for (const auto& client_clock : sync_exchange.client_clocks) {
3904 auto* clock = client_snapshots->add_clocks();
3905 clock->set_clock_id(client_clock.clock_id);
3906 clock->set_timestamp(client_clock.timestamp);
3907 }
3908
3909 auto* host_snapshots = sync_exchange_msg->set_host_clocks();
3910 for (const auto& host_clock : sync_exchange.host_clocks) {
3911 auto* clock = host_snapshots->add_clocks();
3912 clock->set_clock_id(host_clock.clock_id);
3913 clock->set_timestamp(host_clock.timestamp);
3914 }
3915 }
3916
3917 SerializeAndAppendPacket(packets, sync_packet.SerializeAsArray());
3918 }
3919
3920 tracing_session->did_emit_remote_clock_sync_ = true;
3921 }
3922
MaybeEmitCloneTrigger(TracingSession * tracing_session,std::vector<TracePacket> * packets)3923 void TracingServiceImpl::MaybeEmitCloneTrigger(
3924 TracingSession* tracing_session,
3925 std::vector<TracePacket>* packets) {
3926 if (tracing_session->clone_trigger.has_value()) {
3927 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3928 auto* trigger = packet->set_clone_snapshot_trigger();
3929 const auto& info = tracing_session->clone_trigger.value();
3930 trigger->set_trigger_name(info.trigger_name);
3931 trigger->set_producer_name(info.producer_name);
3932 trigger->set_trusted_producer_uid(static_cast<int32_t>(info.producer_uid));
3933
3934 packet->set_timestamp(info.boot_time_ns);
3935 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3936 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3937 SerializeAndAppendPacket(packets, packet.SerializeAsArray());
3938 }
3939 }
3940
MaybeEmitReceivedTriggers(TracingSession * tracing_session,std::vector<TracePacket> * packets)3941 void TracingServiceImpl::MaybeEmitReceivedTriggers(
3942 TracingSession* tracing_session,
3943 std::vector<TracePacket>* packets) {
3944 PERFETTO_DCHECK(tracing_session->num_triggers_emitted_into_trace <=
3945 tracing_session->received_triggers.size());
3946 for (size_t i = tracing_session->num_triggers_emitted_into_trace;
3947 i < tracing_session->received_triggers.size(); ++i) {
3948 const auto& info = tracing_session->received_triggers[i];
3949 protozero::HeapBuffered<protos::pbzero::TracePacket> packet;
3950 auto* trigger = packet->set_trigger();
3951 trigger->set_trigger_name(info.trigger_name);
3952 trigger->set_producer_name(info.producer_name);
3953 trigger->set_trusted_producer_uid(static_cast<int32_t>(info.producer_uid));
3954
3955 packet->set_timestamp(info.boot_time_ns);
3956 packet->set_trusted_uid(static_cast<int32_t>(uid_));
3957 packet->set_trusted_packet_sequence_id(kServicePacketSequenceID);
3958 SerializeAndAppendPacket(packets, packet.SerializeAsArray());
3959 ++tracing_session->num_triggers_emitted_into_trace;
3960 }
3961 }
3962
MaybeLogUploadEvent(const TraceConfig & cfg,const base::Uuid & uuid,PerfettoStatsdAtom atom,const std::string & trigger_name)3963 void TracingServiceImpl::MaybeLogUploadEvent(const TraceConfig& cfg,
3964 const base::Uuid& uuid,
3965 PerfettoStatsdAtom atom,
3966 const std::string& trigger_name) {
3967 if (!ShouldLogEvent(cfg))
3968 return;
3969
3970 PERFETTO_DCHECK(uuid); // The UUID must be set at this point.
3971 android_stats::MaybeLogUploadEvent(atom, uuid.lsb(), uuid.msb(),
3972 trigger_name);
3973 }
3974
MaybeLogTriggerEvent(const TraceConfig & cfg,PerfettoTriggerAtom atom,const std::string & trigger_name)3975 void TracingServiceImpl::MaybeLogTriggerEvent(const TraceConfig& cfg,
3976 PerfettoTriggerAtom atom,
3977 const std::string& trigger_name) {
3978 if (!ShouldLogEvent(cfg))
3979 return;
3980 android_stats::MaybeLogTriggerEvent(atom, trigger_name);
3981 }
3982
PurgeExpiredAndCountTriggerInWindow(int64_t now_ns,uint64_t trigger_name_hash)3983 size_t TracingServiceImpl::PurgeExpiredAndCountTriggerInWindow(
3984 int64_t now_ns,
3985 uint64_t trigger_name_hash) {
3986 constexpr int64_t kOneDayInNs = 24ll * 60 * 60 * 1000 * 1000 * 1000;
3987 PERFETTO_DCHECK(
3988 std::is_sorted(trigger_history_.begin(), trigger_history_.end()));
3989 size_t remove_count = 0;
3990 size_t trigger_count = 0;
3991 for (const TriggerHistory& h : trigger_history_) {
3992 if (h.timestamp_ns < now_ns - kOneDayInNs) {
3993 remove_count++;
3994 } else if (h.name_hash == trigger_name_hash) {
3995 trigger_count++;
3996 }
3997 }
3998 trigger_history_.erase_front(remove_count);
3999 return trigger_count;
4000 }
4001
FlushAndCloneSession(ConsumerEndpointImpl * consumer,ConsumerEndpoint::CloneSessionArgs args)4002 base::Status TracingServiceImpl::FlushAndCloneSession(
4003 ConsumerEndpointImpl* consumer,
4004 ConsumerEndpoint::CloneSessionArgs args) {
4005 PERFETTO_DCHECK_THREAD(thread_checker_);
4006 auto clone_target = FlushFlags::CloneTarget::kUnknown;
4007
4008 TracingSession* session = nullptr;
4009 if (args.for_bugreport) {
4010 clone_target = FlushFlags::CloneTarget::kBugreport;
4011 }
4012 if (args.tsid != 0) {
4013 if (args.tsid == kBugreportSessionId) {
4014 // This branch is only here to support the legacy protocol where we could
4015 // clone only a single session using the magic ID kBugreportSessionId.
4016 // The newer perfetto --clone-all-for-bugreport first queries the existing
4017 // sessions and then issues individual clone requests specifying real
4018 // session IDs, setting args.{for_bugreport,skip_trace_filter}=true.
4019 PERFETTO_LOG("Looking for sessions for bugreport");
4020 session = FindTracingSessionWithMaxBugreportScore();
4021 if (!session) {
4022 return base::ErrStatus(
4023 "No tracing sessions eligible for bugreport found");
4024 }
4025 args.tsid = session->id;
4026 clone_target = FlushFlags::CloneTarget::kBugreport;
4027 args.skip_trace_filter = true;
4028 } else {
4029 session = GetTracingSession(args.tsid);
4030 }
4031 } else if (!args.unique_session_name.empty()) {
4032 session = GetTracingSessionByUniqueName(args.unique_session_name);
4033 }
4034
4035 if (!session) {
4036 return base::ErrStatus("Tracing session not found");
4037 }
4038
4039 // Skip the UID check for sessions marked with a bugreport_score > 0.
4040 // Those sessions, by design, can be stolen by any other consumer for the
4041 // sake of creating snapshots for bugreports.
4042 if (!session->IsCloneAllowed(consumer->uid_)) {
4043 return PERFETTO_SVC_ERR("Not allowed to clone a session from another UID");
4044 }
4045
4046 // If any of the buffers are marked as clear_before_clone, reset them before
4047 // issuing the Flush(kCloneReason).
4048 size_t buf_idx = 0;
4049 for (BufferID src_buf_id : session->buffers_index) {
4050 if (!session->config.buffers()[buf_idx++].clear_before_clone())
4051 continue;
4052 auto buf_iter = buffers_.find(src_buf_id);
4053 PERFETTO_CHECK(buf_iter != buffers_.end());
4054 std::unique_ptr<TraceBuffer>& buf = buf_iter->second;
4055
4056 // No need to reset the buffer if nothing has been written into it yet.
4057 // This is the canonical case if producers behive nicely and don't timeout
4058 // the handling of writes during the flush.
4059 // This check avoids a useless re-mmap upon every Clone() if the buffer is
4060 // already empty (when used in combination with `transfer_on_clone`).
4061 if (!buf->has_data())
4062 continue;
4063
4064 // Some leftover data was left in the buffer. Recreate it to empty it.
4065 const auto buf_policy = buf->overwrite_policy();
4066 const auto buf_size = buf->size();
4067 std::unique_ptr<TraceBuffer> old_buf = std::move(buf);
4068 buf = TraceBuffer::Create(buf_size, buf_policy);
4069 if (!buf) {
4070 // This is extremely rare but could happen on 32-bit. If the new buffer
4071 // allocation failed, put back the buffer where it was and fail the clone.
4072 // We cannot leave the original tracing session buffer-less as it would
4073 // cause crashes when data sources commit new data.
4074 buf = std::move(old_buf);
4075 return base::ErrStatus(
4076 "Buffer allocation failed while attempting to clone");
4077 }
4078 }
4079
4080 auto weak_consumer = consumer->GetWeakPtr();
4081
4082 const PendingCloneID clone_id = session->last_pending_clone_id_++;
4083
4084 auto& clone_op = session->pending_clones[clone_id];
4085 clone_op.pending_flush_cnt = 0;
4086 // Pre-initialize these vectors just as an optimization to avoid reallocations
4087 // in DoCloneBuffers().
4088 clone_op.buffers.reserve(session->buffers_index.size());
4089 clone_op.buffer_cloned_timestamps.reserve(session->buffers_index.size());
4090 clone_op.weak_consumer = weak_consumer;
4091 clone_op.skip_trace_filter = args.skip_trace_filter;
4092 if (!args.clone_trigger_name.empty()) {
4093 clone_op.clone_trigger = {args.clone_trigger_boot_time_ns,
4094 args.clone_trigger_name,
4095 args.clone_trigger_producer_name,
4096 args.clone_trigger_trusted_producer_uid};
4097 }
4098
4099 // Issue separate flush requests for separate buffer groups. The buffer marked
4100 // as transfer_on_clone will be flushed and cloned separately: even if they're
4101 // slower (like in the case of Winscope tracing), they will not delay the
4102 // snapshot of the other buffers.
4103 //
4104 // In the future we might want to split the buffer into more groups and maybe
4105 // allow this to be configurable.
4106 std::array<std::set<BufferID>, 2> bufs_groups;
4107 for (size_t i = 0; i < session->buffers_index.size(); i++) {
4108 if (session->config.buffers()[i].transfer_on_clone()) {
4109 bufs_groups[0].insert(session->buffers_index[i]);
4110 } else {
4111 bufs_groups[1].insert(session->buffers_index[i]);
4112 }
4113 }
4114
4115 SnapshotLifecycleEvent(
4116 session, protos::pbzero::TracingServiceEvent::kFlushStartedFieldNumber,
4117 false /* snapshot_clocks */);
4118 clone_op.pending_flush_cnt = bufs_groups.size();
4119 clone_op.clone_started_timestamp_ns = clock_->GetBootTimeNs().count();
4120 for (const std::set<BufferID>& buf_group : bufs_groups) {
4121 FlushDataSourceInstances(
4122 session, 0,
4123 GetFlushableDataSourceInstancesForBuffers(session, buf_group),
4124 [tsid = session->id, clone_id, buf_group, this](bool final_flush) {
4125 OnFlushDoneForClone(tsid, clone_id, buf_group, final_flush);
4126 },
4127 FlushFlags(FlushFlags::Initiator::kTraced,
4128 FlushFlags::Reason::kTraceClone, clone_target));
4129 }
4130
4131 return base::OkStatus();
4132 }
4133
4134 std::map<ProducerID, std::vector<DataSourceInstanceID>>
GetFlushableDataSourceInstancesForBuffers(TracingSession * session,const std::set<BufferID> & bufs)4135 TracingServiceImpl::GetFlushableDataSourceInstancesForBuffers(
4136 TracingSession* session,
4137 const std::set<BufferID>& bufs) {
4138 std::map<ProducerID, std::vector<DataSourceInstanceID>> data_source_instances;
4139
4140 for (const auto& [producer_id, ds_inst] : session->data_source_instances) {
4141 // TODO(ddiproietto): Consider if we should skip instances if ds_inst.state
4142 // != DataSourceInstance::STARTED
4143 if (ds_inst.no_flush) {
4144 continue;
4145 }
4146 if (!bufs.count(static_cast<BufferID>(ds_inst.config.target_buffer()))) {
4147 continue;
4148 }
4149 data_source_instances[producer_id].push_back(ds_inst.instance_id);
4150 }
4151
4152 return data_source_instances;
4153 }
4154
OnFlushDoneForClone(TracingSessionID tsid,PendingCloneID clone_id,const std::set<BufferID> & buf_ids,bool final_flush_outcome)4155 void TracingServiceImpl::OnFlushDoneForClone(TracingSessionID tsid,
4156 PendingCloneID clone_id,
4157 const std::set<BufferID>& buf_ids,
4158 bool final_flush_outcome) {
4159 TracingSession* src = GetTracingSession(tsid);
4160 // The session might be gone by the time we try to clone it.
4161 if (!src) {
4162 return;
4163 }
4164
4165 auto it = src->pending_clones.find(clone_id);
4166 if (it == src->pending_clones.end()) {
4167 return;
4168 }
4169 auto& clone_op = it->second;
4170
4171 if (final_flush_outcome == false) {
4172 clone_op.flush_failed = true;
4173 }
4174
4175 base::Status result;
4176 base::Uuid uuid;
4177
4178 // First clone the flushed TraceBuffer(s). This can fail because of ENOMEM. If
4179 // it happens bail out early before creating any session.
4180 if (!DoCloneBuffers(*src, buf_ids, &clone_op)) {
4181 result = PERFETTO_SVC_ERR("Buffer allocation failed");
4182 }
4183
4184 if (result.ok()) {
4185 UpdateMemoryGuardrail();
4186
4187 if (--clone_op.pending_flush_cnt != 0) {
4188 // Wait for more pending flushes.
4189 return;
4190 }
4191
4192 PERFETTO_LOG("FlushAndCloneSession(%" PRIu64 ") started, success=%d", tsid,
4193 final_flush_outcome);
4194
4195 if (clone_op.weak_consumer) {
4196 result = FinishCloneSession(
4197 &*clone_op.weak_consumer, tsid, std::move(clone_op.buffers),
4198 std::move(clone_op.buffer_cloned_timestamps),
4199 clone_op.skip_trace_filter, !clone_op.flush_failed,
4200 clone_op.clone_trigger, &uuid, clone_op.clone_started_timestamp_ns);
4201 }
4202 } // if (result.ok())
4203
4204 if (clone_op.weak_consumer) {
4205 clone_op.weak_consumer->consumer_->OnSessionCloned(
4206 {result.ok(), result.message(), uuid});
4207 }
4208
4209 src->pending_clones.erase(it);
4210 UpdateMemoryGuardrail();
4211 }
4212
DoCloneBuffers(const TracingSession & src,const std::set<BufferID> & buf_ids,PendingClone * clone_op)4213 bool TracingServiceImpl::DoCloneBuffers(const TracingSession& src,
4214 const std::set<BufferID>& buf_ids,
4215 PendingClone* clone_op) {
4216 PERFETTO_DCHECK(src.num_buffers() == src.config.buffers().size());
4217 clone_op->buffers.resize(src.buffers_index.size());
4218 clone_op->buffer_cloned_timestamps.resize(src.buffers_index.size());
4219
4220 int64_t now = clock_->GetBootTimeNs().count();
4221
4222 for (size_t buf_idx = 0; buf_idx < src.buffers_index.size(); buf_idx++) {
4223 BufferID src_buf_id = src.buffers_index[buf_idx];
4224 if (buf_ids.count(src_buf_id) == 0)
4225 continue;
4226 auto buf_iter = buffers_.find(src_buf_id);
4227 PERFETTO_CHECK(buf_iter != buffers_.end());
4228 std::unique_ptr<TraceBuffer>& src_buf = buf_iter->second;
4229 std::unique_ptr<TraceBuffer> new_buf;
4230 if (src.config.buffers()[buf_idx].transfer_on_clone()) {
4231 const auto buf_policy = src_buf->overwrite_policy();
4232 const auto buf_size = src_buf->size();
4233 new_buf = std::move(src_buf);
4234 src_buf = TraceBuffer::Create(buf_size, buf_policy);
4235 if (!src_buf) {
4236 // If the allocation fails put the buffer back and let the code below
4237 // handle the failure gracefully.
4238 src_buf = std::move(new_buf);
4239 }
4240 } else {
4241 new_buf = src_buf->CloneReadOnly();
4242 }
4243 if (!new_buf.get()) {
4244 return false;
4245 }
4246 clone_op->buffers[buf_idx] = std::move(new_buf);
4247 clone_op->buffer_cloned_timestamps[buf_idx] = now;
4248 }
4249 return true;
4250 }
4251
FinishCloneSession(ConsumerEndpointImpl * consumer,TracingSessionID src_tsid,std::vector<std::unique_ptr<TraceBuffer>> buf_snaps,std::vector<int64_t> buf_cloned_timestamps,bool skip_trace_filter,bool final_flush_outcome,std::optional<TriggerInfo> clone_trigger,base::Uuid * new_uuid,int64_t clone_started_timestamp_ns)4252 base::Status TracingServiceImpl::FinishCloneSession(
4253 ConsumerEndpointImpl* consumer,
4254 TracingSessionID src_tsid,
4255 std::vector<std::unique_ptr<TraceBuffer>> buf_snaps,
4256 std::vector<int64_t> buf_cloned_timestamps,
4257 bool skip_trace_filter,
4258 bool final_flush_outcome,
4259 std::optional<TriggerInfo> clone_trigger,
4260 base::Uuid* new_uuid,
4261 int64_t clone_started_timestamp_ns) {
4262 PERFETTO_DLOG("CloneSession(%" PRIu64
4263 ", skip_trace_filter=%d) started, consumer uid: %d",
4264 src_tsid, skip_trace_filter, static_cast<int>(consumer->uid_));
4265
4266 TracingSession* src = GetTracingSession(src_tsid);
4267
4268 // The session might be gone by the time we try to clone it.
4269 if (!src)
4270 return PERFETTO_SVC_ERR("session not found");
4271
4272 if (consumer->tracing_session_id_) {
4273 return PERFETTO_SVC_ERR(
4274 "The consumer is already attached to another tracing session");
4275 }
4276
4277 std::vector<BufferID> buf_ids =
4278 buffer_ids_.AllocateMultiple(buf_snaps.size());
4279 if (buf_ids.size() != buf_snaps.size()) {
4280 return PERFETTO_SVC_ERR("Buffer id allocation failed");
4281 }
4282
4283 PERFETTO_CHECK(std::none_of(
4284 buf_snaps.begin(), buf_snaps.end(),
4285 [](const std::unique_ptr<TraceBuffer>& buf) { return buf == nullptr; }));
4286
4287 const TracingSessionID tsid = ++last_tracing_session_id_;
4288 TracingSession* cloned_session =
4289 &tracing_sessions_
4290 .emplace(std::piecewise_construct, std::forward_as_tuple(tsid),
4291 std::forward_as_tuple(tsid, consumer, src->config,
4292 weak_runner_.task_runner()))
4293 .first->second;
4294
4295 // Generate a new UUID for the cloned session, but preserve the LSB. In some
4296 // contexts the LSB is used to tie the trace back to the statsd subscription
4297 // that triggered it. See the corresponding code in perfetto_cmd.cc which
4298 // reads at triggering_subscription_id().
4299 const int64_t orig_uuid_lsb = src->trace_uuid.lsb();
4300 cloned_session->state = TracingSession::CLONED_READ_ONLY;
4301 cloned_session->trace_uuid = base::Uuidv4();
4302 cloned_session->trace_uuid.set_lsb(orig_uuid_lsb);
4303 *new_uuid = cloned_session->trace_uuid;
4304
4305 for (size_t i = 0; i < buf_snaps.size(); i++) {
4306 BufferID buf_global_id = buf_ids[i];
4307 std::unique_ptr<TraceBuffer>& buf = buf_snaps[i];
4308 // This is only needed for transfer_on_clone. Other buffers are already
4309 // marked as read-only by CloneReadOnly(). We cannot do this early because
4310 // in case of an allocation failure we will put std::move() the original
4311 // buffer back in its place and in that case should not be made read-only.
4312 buf->set_read_only();
4313 buffers_.emplace(buf_global_id, std::move(buf));
4314 cloned_session->buffers_index.emplace_back(buf_global_id);
4315 }
4316 UpdateMemoryGuardrail();
4317
4318 // Copy over relevant state that we want to persist in the cloned session.
4319 // Mostly stats and metadata that is emitted in the trace file by the service.
4320 // Also clear the received trigger list in the main tracing session. A
4321 // CLONE_SNAPSHOT session can go in ring buffer mode for several hours and get
4322 // snapshotted several times. This causes two issues with `received_triggers`:
4323 // 1. Adding noise in the cloned trace emitting triggers that happened too
4324 // far back (see b/290799105).
4325 // 2. Bloating memory (see b/290798988).
4326 cloned_session->should_emit_stats = true;
4327 cloned_session->clone_trigger = clone_trigger;
4328 cloned_session->received_triggers = std::move(src->received_triggers);
4329 src->received_triggers.clear();
4330 src->num_triggers_emitted_into_trace = 0;
4331 cloned_session->lifecycle_events =
4332 std::vector<TracingSession::LifecycleEvent>(src->lifecycle_events);
4333 cloned_session->slow_start_event = src->slow_start_event;
4334 cloned_session->last_flush_events = src->last_flush_events;
4335 cloned_session->initial_clock_snapshot = src->initial_clock_snapshot;
4336 cloned_session->clock_snapshot_ring_buffer = src->clock_snapshot_ring_buffer;
4337 cloned_session->invalid_packets = src->invalid_packets;
4338 cloned_session->flushes_requested = src->flushes_requested;
4339 cloned_session->flushes_succeeded = src->flushes_succeeded;
4340 cloned_session->flushes_failed = src->flushes_failed;
4341 cloned_session->compress_deflate = src->compress_deflate;
4342 if (src->trace_filter && !skip_trace_filter) {
4343 // Copy the trace filter, unless it's a clone-for-bugreport (b/317065412).
4344 cloned_session->trace_filter.reset(
4345 new protozero::MessageFilter(src->trace_filter->config()));
4346 }
4347
4348 cloned_session->buffer_cloned_timestamps = std::move(buf_cloned_timestamps);
4349
4350 SetSingleLifecycleEvent(
4351 cloned_session,
4352 protos::pbzero::TracingServiceEvent::kCloneStartedFieldNumber,
4353 clone_started_timestamp_ns);
4354
4355 SnapshotLifecycleEvent(
4356 cloned_session,
4357 protos::pbzero::TracingServiceEvent::kTracingDisabledFieldNumber,
4358 true /* snapshot_clocks */);
4359
4360 PERFETTO_DLOG("Consumer (uid:%d) cloned tracing session %" PRIu64
4361 " -> %" PRIu64,
4362 static_cast<int>(consumer->uid_), src_tsid, tsid);
4363
4364 consumer->tracing_session_id_ = tsid;
4365 cloned_session->final_flush_outcome = final_flush_outcome
4366 ? TraceStats::FINAL_FLUSH_SUCCEEDED
4367 : TraceStats::FINAL_FLUSH_FAILED;
4368 return base::OkStatus();
4369 }
4370
IsCloneAllowed(uid_t clone_uid) const4371 bool TracingServiceImpl::TracingSession::IsCloneAllowed(uid_t clone_uid) const {
4372 if (clone_uid == 0)
4373 return true; // Root is always allowed to clone everything.
4374 if (clone_uid == this->consumer_uid)
4375 return true; // Allow cloning if the uids match.
4376 #if PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID)
4377 // On Android allow shell to clone sessions marked as exported for bugreport.
4378 // Dumpstate (invoked by adb bugreport) invokes commands as shell.
4379 if (clone_uid == AID_SHELL && this->config.bugreport_score() > 0)
4380 return true;
4381 #endif
4382 return false;
4383 }
4384
4385 ////////////////////////////////////////////////////////////////////////////////
4386 // TracingServiceImpl::ConsumerEndpointImpl implementation
4387 ////////////////////////////////////////////////////////////////////////////////
4388
ConsumerEndpointImpl(TracingServiceImpl * service,base::TaskRunner * task_runner,Consumer * consumer,uid_t uid)4389 TracingServiceImpl::ConsumerEndpointImpl::ConsumerEndpointImpl(
4390 TracingServiceImpl* service,
4391 base::TaskRunner* task_runner,
4392 Consumer* consumer,
4393 uid_t uid)
4394 : task_runner_(task_runner),
4395 service_(service),
4396 consumer_(consumer),
4397 uid_(uid),
4398 weak_ptr_factory_(this) {}
4399
~ConsumerEndpointImpl()4400 TracingServiceImpl::ConsumerEndpointImpl::~ConsumerEndpointImpl() {
4401 service_->DisconnectConsumer(this);
4402 consumer_->OnDisconnect();
4403 }
4404
NotifyOnTracingDisabled(const std::string & error)4405 void TracingServiceImpl::ConsumerEndpointImpl::NotifyOnTracingDisabled(
4406 const std::string& error) {
4407 PERFETTO_DCHECK_THREAD(thread_checker_);
4408 task_runner_->PostTask([weak_this = weak_ptr_factory_.GetWeakPtr(),
4409 error /* deliberate copy */] {
4410 if (weak_this)
4411 weak_this->consumer_->OnTracingDisabled(error);
4412 });
4413 }
4414
EnableTracing(const TraceConfig & cfg,base::ScopedFile fd)4415 void TracingServiceImpl::ConsumerEndpointImpl::EnableTracing(
4416 const TraceConfig& cfg,
4417 base::ScopedFile fd) {
4418 PERFETTO_DCHECK_THREAD(thread_checker_);
4419 auto status = service_->EnableTracing(this, cfg, std::move(fd));
4420 if (!status.ok())
4421 NotifyOnTracingDisabled(status.message());
4422 }
4423
ChangeTraceConfig(const TraceConfig & cfg)4424 void TracingServiceImpl::ConsumerEndpointImpl::ChangeTraceConfig(
4425 const TraceConfig& cfg) {
4426 if (!tracing_session_id_) {
4427 PERFETTO_LOG(
4428 "Consumer called ChangeTraceConfig() but tracing was "
4429 "not active");
4430 return;
4431 }
4432 service_->ChangeTraceConfig(this, cfg);
4433 }
4434
StartTracing()4435 void TracingServiceImpl::ConsumerEndpointImpl::StartTracing() {
4436 PERFETTO_DCHECK_THREAD(thread_checker_);
4437 if (!tracing_session_id_) {
4438 PERFETTO_LOG("Consumer called StartTracing() but tracing was not active");
4439 return;
4440 }
4441 service_->StartTracing(tracing_session_id_);
4442 }
4443
DisableTracing()4444 void TracingServiceImpl::ConsumerEndpointImpl::DisableTracing() {
4445 PERFETTO_DCHECK_THREAD(thread_checker_);
4446 if (!tracing_session_id_) {
4447 PERFETTO_LOG("Consumer called DisableTracing() but tracing was not active");
4448 return;
4449 }
4450 service_->DisableTracing(tracing_session_id_);
4451 }
4452
ReadBuffers()4453 void TracingServiceImpl::ConsumerEndpointImpl::ReadBuffers() {
4454 PERFETTO_DCHECK_THREAD(thread_checker_);
4455 if (!tracing_session_id_) {
4456 PERFETTO_LOG("Consumer called ReadBuffers() but tracing was not active");
4457 consumer_->OnTraceData({}, /* has_more = */ false);
4458 return;
4459 }
4460 if (!service_->ReadBuffersIntoConsumer(tracing_session_id_, this)) {
4461 consumer_->OnTraceData({}, /* has_more = */ false);
4462 }
4463 }
4464
FreeBuffers()4465 void TracingServiceImpl::ConsumerEndpointImpl::FreeBuffers() {
4466 PERFETTO_DCHECK_THREAD(thread_checker_);
4467 if (!tracing_session_id_) {
4468 PERFETTO_LOG("Consumer called FreeBuffers() but tracing was not active");
4469 return;
4470 }
4471 service_->FreeBuffers(tracing_session_id_);
4472 tracing_session_id_ = 0;
4473 }
4474
Flush(uint32_t timeout_ms,FlushCallback callback,FlushFlags flush_flags)4475 void TracingServiceImpl::ConsumerEndpointImpl::Flush(uint32_t timeout_ms,
4476 FlushCallback callback,
4477 FlushFlags flush_flags) {
4478 PERFETTO_DCHECK_THREAD(thread_checker_);
4479 if (!tracing_session_id_) {
4480 PERFETTO_LOG("Consumer called Flush() but tracing was not active");
4481 return;
4482 }
4483 service_->Flush(tracing_session_id_, timeout_ms, callback, flush_flags);
4484 }
4485
Detach(const std::string & key)4486 void TracingServiceImpl::ConsumerEndpointImpl::Detach(const std::string& key) {
4487 PERFETTO_DCHECK_THREAD(thread_checker_);
4488 bool success = service_->DetachConsumer(this, key);
4489 auto weak_this = weak_ptr_factory_.GetWeakPtr();
4490 task_runner_->PostTask([weak_this = std::move(weak_this), success] {
4491 if (weak_this)
4492 weak_this->consumer_->OnDetach(success);
4493 });
4494 }
4495
Attach(const std::string & key)4496 void TracingServiceImpl::ConsumerEndpointImpl::Attach(const std::string& key) {
4497 PERFETTO_DCHECK_THREAD(thread_checker_);
4498 bool success = service_->AttachConsumer(this, key);
4499 task_runner_->PostTask([weak_this = weak_ptr_factory_.GetWeakPtr(), success] {
4500 if (!weak_this)
4501 return;
4502 Consumer* consumer = weak_this->consumer_;
4503 TracingSession* session =
4504 weak_this->service_->GetTracingSession(weak_this->tracing_session_id_);
4505 if (!session) {
4506 consumer->OnAttach(false, TraceConfig());
4507 return;
4508 }
4509 consumer->OnAttach(success, session->config);
4510 });
4511 }
4512
GetTraceStats()4513 void TracingServiceImpl::ConsumerEndpointImpl::GetTraceStats() {
4514 PERFETTO_DCHECK_THREAD(thread_checker_);
4515 bool success = false;
4516 TraceStats stats;
4517 TracingSession* session = service_->GetTracingSession(tracing_session_id_);
4518 if (session) {
4519 success = true;
4520 stats = service_->GetTraceStats(session);
4521 }
4522 auto weak_this = weak_ptr_factory_.GetWeakPtr();
4523 task_runner_->PostTask(
4524 [weak_this = std::move(weak_this), success, stats = std::move(stats)] {
4525 if (weak_this)
4526 weak_this->consumer_->OnTraceStats(success, stats);
4527 });
4528 }
4529
ObserveEvents(uint32_t events_mask)4530 void TracingServiceImpl::ConsumerEndpointImpl::ObserveEvents(
4531 uint32_t events_mask) {
4532 PERFETTO_DCHECK_THREAD(thread_checker_);
4533 observable_events_mask_ = events_mask;
4534 TracingSession* session = service_->GetTracingSession(tracing_session_id_);
4535 if (!session)
4536 return;
4537
4538 if (observable_events_mask_ & ObservableEvents::TYPE_DATA_SOURCES_INSTANCES) {
4539 // Issue initial states.
4540 for (const auto& kv : session->data_source_instances) {
4541 ProducerEndpointImpl* producer = service_->GetProducer(kv.first);
4542 PERFETTO_DCHECK(producer);
4543 OnDataSourceInstanceStateChange(*producer, kv.second);
4544 }
4545 }
4546
4547 // If the ObserveEvents() call happens after data sources have acked already
4548 // notify immediately.
4549 if (observable_events_mask_ &
4550 ObservableEvents::TYPE_ALL_DATA_SOURCES_STARTED) {
4551 service_->MaybeNotifyAllDataSourcesStarted(session);
4552 }
4553 }
4554
OnDataSourceInstanceStateChange(const ProducerEndpointImpl & producer,const DataSourceInstance & instance)4555 void TracingServiceImpl::ConsumerEndpointImpl::OnDataSourceInstanceStateChange(
4556 const ProducerEndpointImpl& producer,
4557 const DataSourceInstance& instance) {
4558 if (!(observable_events_mask_ &
4559 ObservableEvents::TYPE_DATA_SOURCES_INSTANCES)) {
4560 return;
4561 }
4562
4563 if (instance.state != DataSourceInstance::CONFIGURED &&
4564 instance.state != DataSourceInstance::STARTED &&
4565 instance.state != DataSourceInstance::STOPPED) {
4566 return;
4567 }
4568
4569 auto* observable_events = AddObservableEvents();
4570 auto* change = observable_events->add_instance_state_changes();
4571 change->set_producer_name(producer.name_);
4572 change->set_data_source_name(instance.data_source_name);
4573 if (instance.state == DataSourceInstance::STARTED) {
4574 change->set_state(ObservableEvents::DATA_SOURCE_INSTANCE_STATE_STARTED);
4575 } else {
4576 change->set_state(ObservableEvents::DATA_SOURCE_INSTANCE_STATE_STOPPED);
4577 }
4578 }
4579
OnAllDataSourcesStarted()4580 void TracingServiceImpl::ConsumerEndpointImpl::OnAllDataSourcesStarted() {
4581 if (!(observable_events_mask_ &
4582 ObservableEvents::TYPE_ALL_DATA_SOURCES_STARTED)) {
4583 return;
4584 }
4585 auto* observable_events = AddObservableEvents();
4586 observable_events->set_all_data_sources_started(true);
4587 }
4588
NotifyCloneSnapshotTrigger(const TriggerInfo & trigger)4589 void TracingServiceImpl::ConsumerEndpointImpl::NotifyCloneSnapshotTrigger(
4590 const TriggerInfo& trigger) {
4591 if (!(observable_events_mask_ & ObservableEvents::TYPE_CLONE_TRIGGER_HIT)) {
4592 return;
4593 }
4594 auto* observable_events = AddObservableEvents();
4595 auto* clone_trig = observable_events->mutable_clone_trigger_hit();
4596 clone_trig->set_tracing_session_id(static_cast<int64_t>(tracing_session_id_));
4597 clone_trig->set_trigger_name(trigger.trigger_name);
4598 clone_trig->set_producer_name(trigger.producer_name);
4599 clone_trig->set_producer_uid(static_cast<uint32_t>(trigger.producer_uid));
4600 clone_trig->set_boot_time_ns(trigger.boot_time_ns);
4601 }
4602
4603 ObservableEvents*
AddObservableEvents()4604 TracingServiceImpl::ConsumerEndpointImpl::AddObservableEvents() {
4605 PERFETTO_DCHECK_THREAD(thread_checker_);
4606 if (!observable_events_) {
4607 observable_events_.reset(new ObservableEvents());
4608 task_runner_->PostTask([weak_this = weak_ptr_factory_.GetWeakPtr()] {
4609 if (!weak_this)
4610 return;
4611
4612 // Move into a temporary to allow reentrancy in OnObservableEvents.
4613 auto observable_events = std::move(weak_this->observable_events_);
4614 weak_this->consumer_->OnObservableEvents(*observable_events);
4615 });
4616 }
4617 return observable_events_.get();
4618 }
4619
QueryServiceState(QueryServiceStateArgs args,QueryServiceStateCallback callback)4620 void TracingServiceImpl::ConsumerEndpointImpl::QueryServiceState(
4621 QueryServiceStateArgs args,
4622 QueryServiceStateCallback callback) {
4623 PERFETTO_DCHECK_THREAD(thread_checker_);
4624 TracingServiceState svc_state;
4625
4626 const auto& sessions = service_->tracing_sessions_;
4627 svc_state.set_tracing_service_version(base::GetVersionString());
4628 svc_state.set_num_sessions(static_cast<int>(sessions.size()));
4629
4630 int num_started = 0;
4631 for (const auto& kv : sessions)
4632 num_started += kv.second.state == TracingSession::State::STARTED ? 1 : 0;
4633 svc_state.set_num_sessions_started(num_started);
4634
4635 for (const auto& kv : service_->producers_) {
4636 if (args.sessions_only)
4637 break;
4638 auto* producer = svc_state.add_producers();
4639 producer->set_id(static_cast<int>(kv.first));
4640 producer->set_name(kv.second->name_);
4641 producer->set_sdk_version(kv.second->sdk_version_);
4642 producer->set_uid(static_cast<int32_t>(kv.second->uid()));
4643 producer->set_pid(static_cast<int32_t>(kv.second->pid()));
4644 producer->set_frozen(kv.second->IsAndroidProcessFrozen());
4645 }
4646
4647 for (const auto& kv : service_->data_sources_) {
4648 if (args.sessions_only)
4649 break;
4650 const auto& registered_data_source = kv.second;
4651 auto* data_source = svc_state.add_data_sources();
4652 *data_source->mutable_ds_descriptor() = registered_data_source.descriptor;
4653 data_source->set_producer_id(
4654 static_cast<int>(registered_data_source.producer_id));
4655 }
4656
4657 svc_state.set_supports_tracing_sessions(true);
4658 for (const auto& kv : service_->tracing_sessions_) {
4659 const TracingSession& s = kv.second;
4660 if (!s.IsCloneAllowed(uid_))
4661 continue;
4662 auto* session = svc_state.add_tracing_sessions();
4663 session->set_id(s.id);
4664 session->set_consumer_uid(static_cast<int>(s.consumer_uid));
4665 session->set_duration_ms(s.config.duration_ms());
4666 session->set_num_data_sources(
4667 static_cast<uint32_t>(s.data_source_instances.size()));
4668 session->set_unique_session_name(s.config.unique_session_name());
4669 if (s.config.has_bugreport_score())
4670 session->set_bugreport_score(s.config.bugreport_score());
4671 if (s.config.has_bugreport_filename())
4672 session->set_bugreport_filename(s.config.bugreport_filename());
4673 for (const auto& snap_kv : s.initial_clock_snapshot) {
4674 if (snap_kv.clock_id == protos::pbzero::BUILTIN_CLOCK_REALTIME)
4675 session->set_start_realtime_ns(static_cast<int64_t>(snap_kv.timestamp));
4676 }
4677 for (const auto& buf : s.config.buffers())
4678 session->add_buffer_size_kb(buf.size_kb());
4679
4680 switch (s.state) {
4681 case TracingSession::State::DISABLED:
4682 session->set_state("DISABLED");
4683 break;
4684 case TracingSession::State::CONFIGURED:
4685 session->set_state("CONFIGURED");
4686 break;
4687 case TracingSession::State::STARTED:
4688 session->set_is_started(true);
4689 session->set_state("STARTED");
4690 break;
4691 case TracingSession::State::DISABLING_WAITING_STOP_ACKS:
4692 session->set_state("STOP_WAIT");
4693 break;
4694 case TracingSession::State::CLONED_READ_ONLY:
4695 session->set_state("CLONED_READ_ONLY");
4696 break;
4697 }
4698 }
4699 callback(/*success=*/true, svc_state);
4700 }
4701
QueryCapabilities(QueryCapabilitiesCallback callback)4702 void TracingServiceImpl::ConsumerEndpointImpl::QueryCapabilities(
4703 QueryCapabilitiesCallback callback) {
4704 PERFETTO_DCHECK_THREAD(thread_checker_);
4705 TracingServiceCapabilities caps;
4706 caps.set_has_query_capabilities(true);
4707 caps.set_has_trace_config_output_path(true);
4708 caps.set_has_clone_session(true);
4709 caps.add_observable_events(ObservableEvents::TYPE_DATA_SOURCES_INSTANCES);
4710 caps.add_observable_events(ObservableEvents::TYPE_ALL_DATA_SOURCES_STARTED);
4711 caps.add_observable_events(ObservableEvents::TYPE_CLONE_TRIGGER_HIT);
4712 static_assert(
4713 ObservableEvents::Type_MAX == ObservableEvents::TYPE_CLONE_TRIGGER_HIT,
4714 "");
4715 callback(caps);
4716 }
4717
SaveTraceForBugreport(SaveTraceForBugreportCallback consumer_callback)4718 void TracingServiceImpl::ConsumerEndpointImpl::SaveTraceForBugreport(
4719 SaveTraceForBugreportCallback consumer_callback) {
4720 consumer_callback(false,
4721 "SaveTraceForBugreport is deprecated. Use "
4722 "CloneSession(kBugreportSessionId) instead.");
4723 }
4724
CloneSession(CloneSessionArgs args)4725 void TracingServiceImpl::ConsumerEndpointImpl::CloneSession(
4726 CloneSessionArgs args) {
4727 PERFETTO_DCHECK_THREAD(thread_checker_);
4728 // FlushAndCloneSession will call OnSessionCloned after the async flush.
4729 base::Status result = service_->FlushAndCloneSession(this, std::move(args));
4730
4731 if (!result.ok()) {
4732 consumer_->OnSessionCloned({false, result.message(), {}});
4733 }
4734 }
4735
4736 ////////////////////////////////////////////////////////////////////////////////
4737 // TracingServiceImpl::ProducerEndpointImpl implementation
4738 ////////////////////////////////////////////////////////////////////////////////
4739
ProducerEndpointImpl(ProducerID id,const ClientIdentity & client_identity,TracingServiceImpl * service,base::TaskRunner * task_runner,Producer * producer,const std::string & producer_name,const std::string & sdk_version,bool in_process,bool smb_scraping_enabled)4740 TracingServiceImpl::ProducerEndpointImpl::ProducerEndpointImpl(
4741 ProducerID id,
4742 const ClientIdentity& client_identity,
4743 TracingServiceImpl* service,
4744 base::TaskRunner* task_runner,
4745 Producer* producer,
4746 const std::string& producer_name,
4747 const std::string& sdk_version,
4748 bool in_process,
4749 bool smb_scraping_enabled)
4750 : id_(id),
4751 client_identity_(client_identity),
4752 service_(service),
4753 producer_(producer),
4754 name_(producer_name),
4755 sdk_version_(sdk_version),
4756 in_process_(in_process),
4757 smb_scraping_enabled_(smb_scraping_enabled),
4758 weak_runner_(task_runner) {}
4759
~ProducerEndpointImpl()4760 TracingServiceImpl::ProducerEndpointImpl::~ProducerEndpointImpl() {
4761 service_->DisconnectProducer(id_);
4762 producer_->OnDisconnect();
4763 }
4764
Disconnect()4765 void TracingServiceImpl::ProducerEndpointImpl::Disconnect() {
4766 PERFETTO_DCHECK_THREAD(thread_checker_);
4767 // Disconnection is only supported via destroying the ProducerEndpoint.
4768 PERFETTO_FATAL("Not supported");
4769 }
4770
RegisterDataSource(const DataSourceDescriptor & desc)4771 void TracingServiceImpl::ProducerEndpointImpl::RegisterDataSource(
4772 const DataSourceDescriptor& desc) {
4773 PERFETTO_DCHECK_THREAD(thread_checker_);
4774 service_->RegisterDataSource(id_, desc);
4775 }
4776
UpdateDataSource(const DataSourceDescriptor & desc)4777 void TracingServiceImpl::ProducerEndpointImpl::UpdateDataSource(
4778 const DataSourceDescriptor& desc) {
4779 PERFETTO_DCHECK_THREAD(thread_checker_);
4780 service_->UpdateDataSource(id_, desc);
4781 }
4782
UnregisterDataSource(const std::string & name)4783 void TracingServiceImpl::ProducerEndpointImpl::UnregisterDataSource(
4784 const std::string& name) {
4785 PERFETTO_DCHECK_THREAD(thread_checker_);
4786 service_->UnregisterDataSource(id_, name);
4787 }
4788
RegisterTraceWriter(uint32_t writer_id,uint32_t target_buffer)4789 void TracingServiceImpl::ProducerEndpointImpl::RegisterTraceWriter(
4790 uint32_t writer_id,
4791 uint32_t target_buffer) {
4792 PERFETTO_DCHECK_THREAD(thread_checker_);
4793 writers_[static_cast<WriterID>(writer_id)] =
4794 static_cast<BufferID>(target_buffer);
4795 }
4796
UnregisterTraceWriter(uint32_t writer_id)4797 void TracingServiceImpl::ProducerEndpointImpl::UnregisterTraceWriter(
4798 uint32_t writer_id) {
4799 PERFETTO_DCHECK_THREAD(thread_checker_);
4800 writers_.erase(static_cast<WriterID>(writer_id));
4801 }
4802
CommitData(const CommitDataRequest & req_untrusted,CommitDataCallback callback)4803 void TracingServiceImpl::ProducerEndpointImpl::CommitData(
4804 const CommitDataRequest& req_untrusted,
4805 CommitDataCallback callback) {
4806 PERFETTO_DCHECK_THREAD(thread_checker_);
4807
4808 if (metatrace::IsEnabled(metatrace::TAG_TRACE_SERVICE)) {
4809 PERFETTO_METATRACE_COUNTER(TAG_TRACE_SERVICE, TRACE_SERVICE_COMMIT_DATA,
4810 EncodeCommitDataRequest(id_, req_untrusted));
4811 }
4812
4813 if (!shared_memory_) {
4814 PERFETTO_DLOG(
4815 "Attempted to commit data before the shared memory was allocated.");
4816 return;
4817 }
4818 PERFETTO_DCHECK(shmem_abi_.is_valid());
4819 for (const auto& entry : req_untrusted.chunks_to_move()) {
4820 const uint32_t page_idx = entry.page();
4821 if (page_idx >= shmem_abi_.num_pages())
4822 continue; // A buggy or malicious producer.
4823
4824 SharedMemoryABI::Chunk chunk;
4825 bool commit_data_over_ipc = entry.has_data();
4826 if (PERFETTO_UNLIKELY(commit_data_over_ipc)) {
4827 // Chunk data is passed over the wire. Create a chunk using the serialized
4828 // protobuf message.
4829 const std::string& data = entry.data();
4830 if (data.size() > SharedMemoryABI::Chunk::kMaxSize) {
4831 PERFETTO_DFATAL("IPC data commit too large: %zu", data.size());
4832 continue; // A malicious or buggy producer
4833 }
4834 // |data| is not altered, but we need to const_cast becasue Chunk data
4835 // members are non-const.
4836 chunk = SharedMemoryABI::MakeChunkFromSerializedData(
4837 reinterpret_cast<uint8_t*>(const_cast<char*>(data.data())),
4838 static_cast<uint16_t>(entry.data().size()),
4839 static_cast<uint8_t>(entry.chunk()));
4840 } else
4841 chunk = shmem_abi_.TryAcquireChunkForReading(page_idx, entry.chunk());
4842 if (!chunk.is_valid()) {
4843 PERFETTO_DLOG("Asked to move chunk %d:%d, but it's not complete",
4844 entry.page(), entry.chunk());
4845 continue;
4846 }
4847
4848 // TryAcquireChunkForReading() has load-acquire semantics. Once acquired,
4849 // the ABI contract expects the producer to not touch the chunk anymore
4850 // (until the service marks that as free). This is why all the reads below
4851 // are just memory_order_relaxed. Also, the code here assumes that all this
4852 // data can be malicious and just gives up if anything is malformed.
4853 BufferID buffer_id = static_cast<BufferID>(entry.target_buffer());
4854 const SharedMemoryABI::ChunkHeader& chunk_header = *chunk.header();
4855 WriterID writer_id = chunk_header.writer_id.load(std::memory_order_relaxed);
4856 ChunkID chunk_id = chunk_header.chunk_id.load(std::memory_order_relaxed);
4857 auto packets = chunk_header.packets.load(std::memory_order_relaxed);
4858 uint16_t num_fragments = packets.count;
4859 uint8_t chunk_flags = packets.flags;
4860
4861 service_->CopyProducerPageIntoLogBuffer(
4862 id_, client_identity_, writer_id, chunk_id, buffer_id, num_fragments,
4863 chunk_flags,
4864 /*chunk_complete=*/true, chunk.payload_begin(), chunk.payload_size());
4865
4866 if (!commit_data_over_ipc) {
4867 // This one has release-store semantics.
4868 shmem_abi_.ReleaseChunkAsFree(std::move(chunk));
4869 }
4870 } // for(chunks_to_move)
4871
4872 service_->ApplyChunkPatches(id_, req_untrusted.chunks_to_patch());
4873
4874 if (req_untrusted.flush_request_id()) {
4875 service_->NotifyFlushDoneForProducer(id_, req_untrusted.flush_request_id());
4876 }
4877
4878 // Keep this invocation last. ProducerIPCService::CommitData() relies on this
4879 // callback being invoked within the same callstack and not posted. If this
4880 // changes, the code there needs to be changed accordingly.
4881 if (callback)
4882 callback();
4883 }
4884
SetupSharedMemory(std::unique_ptr<SharedMemory> shared_memory,size_t page_size_bytes,bool provided_by_producer)4885 void TracingServiceImpl::ProducerEndpointImpl::SetupSharedMemory(
4886 std::unique_ptr<SharedMemory> shared_memory,
4887 size_t page_size_bytes,
4888 bool provided_by_producer) {
4889 PERFETTO_DCHECK(!shared_memory_ && !shmem_abi_.is_valid());
4890 PERFETTO_DCHECK(page_size_bytes % 1024 == 0);
4891
4892 shared_memory_ = std::move(shared_memory);
4893 shared_buffer_page_size_kb_ = page_size_bytes / 1024;
4894 is_shmem_provided_by_producer_ = provided_by_producer;
4895
4896 shmem_abi_.Initialize(reinterpret_cast<uint8_t*>(shared_memory_->start()),
4897 shared_memory_->size(),
4898 shared_buffer_page_size_kb() * 1024,
4899 SharedMemoryABI::ShmemMode::kDefault);
4900 if (in_process_) {
4901 inproc_shmem_arbiter_.reset(new SharedMemoryArbiterImpl(
4902 shared_memory_->start(), shared_memory_->size(),
4903 SharedMemoryABI::ShmemMode::kDefault,
4904 shared_buffer_page_size_kb_ * 1024, this, weak_runner_.task_runner()));
4905 inproc_shmem_arbiter_->SetDirectSMBPatchingSupportedByService();
4906 }
4907
4908 OnTracingSetup();
4909 service_->UpdateMemoryGuardrail();
4910 }
4911
shared_memory() const4912 SharedMemory* TracingServiceImpl::ProducerEndpointImpl::shared_memory() const {
4913 PERFETTO_DCHECK_THREAD(thread_checker_);
4914 return shared_memory_.get();
4915 }
4916
shared_buffer_page_size_kb() const4917 size_t TracingServiceImpl::ProducerEndpointImpl::shared_buffer_page_size_kb()
4918 const {
4919 return shared_buffer_page_size_kb_;
4920 }
4921
ActivateTriggers(const std::vector<std::string> & triggers)4922 void TracingServiceImpl::ProducerEndpointImpl::ActivateTriggers(
4923 const std::vector<std::string>& triggers) {
4924 service_->ActivateTriggers(id_, triggers);
4925 }
4926
StopDataSource(DataSourceInstanceID ds_inst_id)4927 void TracingServiceImpl::ProducerEndpointImpl::StopDataSource(
4928 DataSourceInstanceID ds_inst_id) {
4929 // TODO(primiano): When we'll support tearing down the SMB, at this point we
4930 // should send the Producer a TearDownTracing if all its data sources have
4931 // been disabled (see b/77532839 and aosp/655179 PS1).
4932 PERFETTO_DCHECK_THREAD(thread_checker_);
4933 weak_runner_.PostTask(
4934 [this, ds_inst_id] { producer_->StopDataSource(ds_inst_id); });
4935 }
4936
4937 SharedMemoryArbiter*
MaybeSharedMemoryArbiter()4938 TracingServiceImpl::ProducerEndpointImpl::MaybeSharedMemoryArbiter() {
4939 if (!inproc_shmem_arbiter_) {
4940 PERFETTO_FATAL(
4941 "The in-process SharedMemoryArbiter can only be used when "
4942 "CreateProducer has been called with in_process=true and after tracing "
4943 "has started.");
4944 }
4945
4946 PERFETTO_DCHECK(in_process_);
4947 return inproc_shmem_arbiter_.get();
4948 }
4949
IsShmemProvidedByProducer() const4950 bool TracingServiceImpl::ProducerEndpointImpl::IsShmemProvidedByProducer()
4951 const {
4952 return is_shmem_provided_by_producer_;
4953 }
4954
4955 // Can be called on any thread.
4956 std::unique_ptr<TraceWriter>
CreateTraceWriter(BufferID buf_id,BufferExhaustedPolicy buffer_exhausted_policy)4957 TracingServiceImpl::ProducerEndpointImpl::CreateTraceWriter(
4958 BufferID buf_id,
4959 BufferExhaustedPolicy buffer_exhausted_policy) {
4960 PERFETTO_DCHECK(MaybeSharedMemoryArbiter());
4961 return MaybeSharedMemoryArbiter()->CreateTraceWriter(buf_id,
4962 buffer_exhausted_policy);
4963 }
4964
NotifyFlushComplete(FlushRequestID id)4965 void TracingServiceImpl::ProducerEndpointImpl::NotifyFlushComplete(
4966 FlushRequestID id) {
4967 PERFETTO_DCHECK_THREAD(thread_checker_);
4968 PERFETTO_DCHECK(MaybeSharedMemoryArbiter());
4969 return MaybeSharedMemoryArbiter()->NotifyFlushComplete(id);
4970 }
4971
OnTracingSetup()4972 void TracingServiceImpl::ProducerEndpointImpl::OnTracingSetup() {
4973 weak_runner_.PostTask([this] { producer_->OnTracingSetup(); });
4974 }
4975
Flush(FlushRequestID flush_request_id,const std::vector<DataSourceInstanceID> & data_sources,FlushFlags flush_flags)4976 void TracingServiceImpl::ProducerEndpointImpl::Flush(
4977 FlushRequestID flush_request_id,
4978 const std::vector<DataSourceInstanceID>& data_sources,
4979 FlushFlags flush_flags) {
4980 PERFETTO_DCHECK_THREAD(thread_checker_);
4981 weak_runner_.PostTask([this, flush_request_id, data_sources, flush_flags] {
4982 producer_->Flush(flush_request_id, data_sources.data(), data_sources.size(),
4983 flush_flags);
4984 });
4985 }
4986
SetupDataSource(DataSourceInstanceID ds_id,const DataSourceConfig & config)4987 void TracingServiceImpl::ProducerEndpointImpl::SetupDataSource(
4988 DataSourceInstanceID ds_id,
4989 const DataSourceConfig& config) {
4990 PERFETTO_DCHECK_THREAD(thread_checker_);
4991 allowed_target_buffers_.insert(static_cast<BufferID>(config.target_buffer()));
4992 weak_runner_.PostTask([this, ds_id, config] {
4993 producer_->SetupDataSource(ds_id, std::move(config));
4994 });
4995 }
4996
StartDataSource(DataSourceInstanceID ds_id,const DataSourceConfig & config)4997 void TracingServiceImpl::ProducerEndpointImpl::StartDataSource(
4998 DataSourceInstanceID ds_id,
4999 const DataSourceConfig& config) {
5000 PERFETTO_DCHECK_THREAD(thread_checker_);
5001 weak_runner_.PostTask([this, ds_id, config] {
5002 producer_->StartDataSource(ds_id, std::move(config));
5003 });
5004 }
5005
NotifyDataSourceStarted(DataSourceInstanceID data_source_id)5006 void TracingServiceImpl::ProducerEndpointImpl::NotifyDataSourceStarted(
5007 DataSourceInstanceID data_source_id) {
5008 PERFETTO_DCHECK_THREAD(thread_checker_);
5009 service_->NotifyDataSourceStarted(id_, data_source_id);
5010 }
5011
NotifyDataSourceStopped(DataSourceInstanceID data_source_id)5012 void TracingServiceImpl::ProducerEndpointImpl::NotifyDataSourceStopped(
5013 DataSourceInstanceID data_source_id) {
5014 PERFETTO_DCHECK_THREAD(thread_checker_);
5015 service_->NotifyDataSourceStopped(id_, data_source_id);
5016 }
5017
OnFreeBuffers(const std::vector<BufferID> & target_buffers)5018 void TracingServiceImpl::ProducerEndpointImpl::OnFreeBuffers(
5019 const std::vector<BufferID>& target_buffers) {
5020 if (allowed_target_buffers_.empty())
5021 return;
5022 for (BufferID buffer : target_buffers)
5023 allowed_target_buffers_.erase(buffer);
5024 }
5025
ClearIncrementalState(const std::vector<DataSourceInstanceID> & data_sources)5026 void TracingServiceImpl::ProducerEndpointImpl::ClearIncrementalState(
5027 const std::vector<DataSourceInstanceID>& data_sources) {
5028 PERFETTO_DCHECK_THREAD(thread_checker_);
5029 weak_runner_.PostTask([this, data_sources] {
5030 base::StringView producer_name(name_);
5031 producer_->ClearIncrementalState(data_sources.data(), data_sources.size());
5032 });
5033 }
5034
Sync(std::function<void ()> callback)5035 void TracingServiceImpl::ProducerEndpointImpl::Sync(
5036 std::function<void()> callback) {
5037 weak_runner_.task_runner()->PostTask(callback);
5038 }
5039
IsAndroidProcessFrozen()5040 bool TracingServiceImpl::ProducerEndpointImpl::IsAndroidProcessFrozen() {
5041 #if PERFETTO_BUILDFLAG(PERFETTO_OS_ANDROID)
5042 if (in_process_ || uid() == base::kInvalidUid || pid() == base::kInvalidPid)
5043 return false;
5044
5045 // As per aosp/3406861, there are three possible mount points for the cgroup.
5046 // Look at all of them.
5047 // - Historically everything was in /uid_xxx/pid_yyy (and still is if
5048 // PRODUCT_CGROUP_V2_SYS_APP_ISOLATION_ENABLED = false)
5049 // - cgroup isolation introduces /apps /system subdirectories.
5050 base::StackString<255> path_v1(
5051 "/sys/fs/cgroup/uid_%" PRIu32 "/pid_%" PRIu32 "/cgroup.freeze",
5052 static_cast<uint32_t>(uid()), static_cast<uint32_t>(pid()));
5053 base::StackString<255> path_v2_app(
5054 "/sys/fs/cgroup/apps/uid_%" PRIu32 "/pid_%" PRIu32 "/cgroup.freeze",
5055 static_cast<uint32_t>(uid()), static_cast<uint32_t>(pid()));
5056 base::StackString<255> path_v2_system(
5057 "/sys/fs/cgroup/system/uid_%" PRIu32 "/pid_%" PRIu32 "/cgroup.freeze",
5058 static_cast<uint32_t>(uid()), static_cast<uint32_t>(pid()));
5059 const char* paths[] = {path_v1.c_str(), path_v2_app.c_str(),
5060 path_v2_system.c_str()};
5061
5062 for (const char* path : paths) {
5063 char frozen = '0';
5064 auto fd = base::OpenFile(path, O_RDONLY);
5065 ssize_t rsize = 0;
5066 if (fd) {
5067 rsize = base::Read(*fd, &frozen, sizeof(frozen));
5068 if (rsize > 0) {
5069 return frozen == '1';
5070 }
5071 }
5072 }
5073 PERFETTO_DLOG("Failed to read cgroup.freeze from [%s, %s, %s]",
5074 path_v1.c_str(), path_v2_app.c_str(), path_v2_system.c_str());
5075
5076 #endif
5077 return false;
5078 }
5079
5080 ////////////////////////////////////////////////////////////////////////////////
5081 // TracingServiceImpl::TracingSession implementation
5082 ////////////////////////////////////////////////////////////////////////////////
5083
TracingSession(TracingSessionID session_id,ConsumerEndpointImpl * consumer,const TraceConfig & new_config,base::TaskRunner * task_runner)5084 TracingServiceImpl::TracingSession::TracingSession(
5085 TracingSessionID session_id,
5086 ConsumerEndpointImpl* consumer,
5087 const TraceConfig& new_config,
5088 base::TaskRunner* task_runner)
5089 : id(session_id),
5090 consumer_maybe_null(consumer),
5091 consumer_uid(consumer->uid_),
5092 config(new_config),
5093 snapshot_periodic_task(task_runner),
5094 timed_stop_task(task_runner) {
5095 // all_data_sources_flushed (and flush_started) is special because we store up
5096 // to 64 events of this type. Other events will go through the default case in
5097 // SnapshotLifecycleEvent() where they will be given a max history of 1.
5098 lifecycle_events.emplace_back(
5099 protos::pbzero::TracingServiceEvent::kAllDataSourcesFlushedFieldNumber,
5100 64 /* max_size */);
5101 lifecycle_events.emplace_back(
5102 protos::pbzero::TracingServiceEvent::kFlushStartedFieldNumber,
5103 64 /* max_size */);
5104 }
5105
5106 ////////////////////////////////////////////////////////////////////////////////
5107 // TracingServiceImpl::RelayEndpointImpl implementation
5108 ////////////////////////////////////////////////////////////////////////////////
RelayEndpointImpl(RelayClientID relay_client_id,TracingServiceImpl * service)5109 TracingServiceImpl::RelayEndpointImpl::RelayEndpointImpl(
5110 RelayClientID relay_client_id,
5111 TracingServiceImpl* service)
5112 : relay_client_id_(relay_client_id), service_(service) {}
5113 TracingServiceImpl::RelayEndpointImpl::~RelayEndpointImpl() = default;
5114
SyncClocks(SyncMode sync_mode,base::ClockSnapshotVector client_clocks,base::ClockSnapshotVector host_clocks)5115 void TracingServiceImpl::RelayEndpointImpl::SyncClocks(
5116 SyncMode sync_mode,
5117 base::ClockSnapshotVector client_clocks,
5118 base::ClockSnapshotVector host_clocks) {
5119 // We keep only the most recent 5 clock sync snapshots.
5120 static constexpr size_t kNumSyncClocks = 5;
5121 if (synced_clocks_.size() >= kNumSyncClocks)
5122 synced_clocks_.pop_front();
5123
5124 synced_clocks_.emplace_back(sync_mode, std::move(client_clocks),
5125 std::move(host_clocks));
5126 }
5127
Disconnect()5128 void TracingServiceImpl::RelayEndpointImpl::Disconnect() {
5129 service_->DisconnectRelayClient(relay_client_id_);
5130 }
5131
5132 } // namespace perfetto
5133