Refine incidentd error handling and logging
Added or changed some error messages and logging for easier debugging.
Log a few section errors to incident metadata in addition to logcat.
Also let incident reporting continue instead of failing when non-fatal
error occurs.
Bug: 119417232
Test: Take an incident report, and logcat.
Change-Id: Id18b7b690100923a8074d99b2312ee80c0bcd760
diff --git a/cmds/incidentd/src/FdBuffer.cpp b/cmds/incidentd/src/FdBuffer.cpp
index a8ef831..74cda1d 100644
--- a/cmds/incidentd/src/FdBuffer.cpp
+++ b/cmds/incidentd/src/FdBuffer.cpp
@@ -47,9 +47,13 @@
while (true) {
if (mBuffer.size() >= MAX_BUFFER_COUNT * BUFFER_SIZE) {
mTruncated = true;
+ VLOG("Truncating data");
break;
}
- if (mBuffer.writeBuffer() == NULL) return NO_MEMORY;
+ if (mBuffer.writeBuffer() == NULL) {
+ VLOG("No memory");
+ return NO_MEMORY;
+ }
int64_t remainingTime = (mStartTime + timeout) - uptimeMillis();
if (remainingTime <= 0) {
@@ -58,7 +62,7 @@
break;
}
- int count = poll(&pfds, 1, remainingTime);
+ int count = TEMP_FAILURE_RETRY(poll(&pfds, 1, remainingTime));
if (count == 0) {
VLOG("timed out due to block calling poll");
mTimedOut = true;
@@ -102,7 +106,10 @@
VLOG("Truncating data");
break;
}
- if (mBuffer.writeBuffer() == NULL) return NO_MEMORY;
+ if (mBuffer.writeBuffer() == NULL) {
+ VLOG("No memory");
+ return NO_MEMORY;
+ }
ssize_t amt =
TEMP_FAILURE_RETRY(::read(fd, mBuffer.writeBuffer(), mBuffer.currentToWrite()));
@@ -144,10 +151,14 @@
// This is the buffer used to store processed data
while (true) {
if (mBuffer.size() >= MAX_BUFFER_COUNT * BUFFER_SIZE) {
+ VLOG("Truncating data");
mTruncated = true;
break;
}
- if (mBuffer.writeBuffer() == NULL) return NO_MEMORY;
+ if (mBuffer.writeBuffer() == NULL) {
+ VLOG("No memory");
+ return NO_MEMORY;
+ }
int64_t remainingTime = (mStartTime + timeoutMs) - uptimeMillis();
if (remainingTime <= 0) {
@@ -157,7 +168,7 @@
}
// wait for any pfds to be ready to perform IO
- int count = poll(pfds, 3, remainingTime);
+ int count = TEMP_FAILURE_RETRY(poll(pfds, 3, remainingTime));
if (count == 0) {
VLOG("timed out due to block calling poll");
mTimedOut = true;
diff --git a/cmds/incidentd/src/IncidentService.cpp b/cmds/incidentd/src/IncidentService.cpp
index b23c1ed..3d892ad 100644
--- a/cmds/incidentd/src/IncidentService.cpp
+++ b/cmds/incidentd/src/IncidentService.cpp
@@ -42,13 +42,11 @@
#define DEFAULT_BYTES_SIZE_LIMIT (20 * 1024 * 1024) // 20MB
#define DEFAULT_REFACTORY_PERIOD_MS (24 * 60 * 60 * 1000) // 1 Day
-// Skip logs (1100 - 1108) because they are already in the bug report
-// Skip 1200, 1201, 1202, 3018 because they take too long
-// TODO(120079956): Skip 3008, 3015 because of error
+// Skip these sections for dumpstate only. Dumpstate allows 10s max for each service to dump.
+// Skip logs (1100 - 1108) and traces (1200 - 1202) because they are already in the bug report.
+// Skip 3018 because it takes too long.
#define SKIPPED_SECTIONS { 1100, 1101, 1102, 1103, 1104, 1105, 1106, 1107, 1108, /* Logs */ \
1200, 1201, 1202, /* Native, hal, java traces */ \
- 3008, /* "package --proto" */ \
- 3015, /* "activity --proto processes" */ \
3018 /* "meminfo -a --proto" */ }
namespace android {
diff --git a/cmds/incidentd/src/PrivacyBuffer.cpp b/cmds/incidentd/src/PrivacyBuffer.cpp
index 7a8ebe3..6967e33 100644
--- a/cmds/incidentd/src/PrivacyBuffer.cpp
+++ b/cmds/incidentd/src/PrivacyBuffer.cpp
@@ -96,7 +96,12 @@
uint64_t token = mProto.start(encode_field_id(policy));
while (mData.rp()->pos() - start != msgSize) {
status_t err = stripField(policy, spec, depth + 1);
- if (err != NO_ERROR) return err;
+ if (err != NO_ERROR) {
+ VLOG("Bad value when stripping id %d, wiretype %d, tag %#x, depth %d, size %d, "
+ "relative pos %zu, ", fieldId, read_wire_type(fieldTag), fieldTag, depth,
+ msgSize, mData.rp()->pos() - start);
+ return err;
+ }
}
mProto.end(token);
return NO_ERROR;
@@ -117,9 +122,13 @@
}
while (mData.hasNext()) {
status_t err = stripField(mPolicy, spec, 0);
- if (err != NO_ERROR) return err;
+ if (err != NO_ERROR) return err; // Error logged in stripField.
}
- if (mData.bytesRead() != mData.size()) return BAD_VALUE;
+ if (mData.bytesRead() != mData.size()) {
+ VLOG("Buffer corrupted: expect %zu bytes, read %zu bytes",
+ mData.size(), mData.bytesRead());
+ return BAD_VALUE;
+ }
mSize = mProto.size();
mData.rp()->rewind(); // rewind the read pointer back to beginning after the strip.
return NO_ERROR;
diff --git a/cmds/incidentd/src/Reporter.cpp b/cmds/incidentd/src/Reporter.cpp
index b3bab0c..8f62da2 100644
--- a/cmds/incidentd/src/Reporter.cpp
+++ b/cmds/incidentd/src/Reporter.cpp
@@ -129,6 +129,7 @@
bool needMainFd = false;
int mainFd = -1;
int mainDest = -1;
+ int sectionCount = 0;
HeaderSection headers;
MetadataSection metadataSection;
std::string buildType = android::base::GetProperty("ro.build.type", "");
@@ -180,12 +181,12 @@
for (const Section** section = SECTION_LIST; *section; section++) {
const int id = (*section)->id;
if ((*section)->userdebugAndEngOnly && !isUserdebugOrEng) {
- ALOGD("Skipping incident report section %d '%s' because it's limited to userdebug/eng",
+ VLOG("Skipping incident report section %d '%s' because it's limited to userdebug/eng",
id, (*section)->name.string());
continue;
}
if (this->batch.containsSection(id)) {
- ALOGD("Taking incident report section %d '%s'", id, (*section)->name.string());
+ VLOG("Taking incident report section %d '%s'", id, (*section)->name.string());
for (ReportRequestSet::iterator it = batch.begin(); it != batch.end(); it++) {
if ((*it)->listener != NULL && (*it)->args.containsSection(id)) {
(*it)->listener->onReportSectionStatus(
@@ -198,11 +199,12 @@
int64_t startTime = uptimeMillis();
err = (*section)->Execute(&batch);
int64_t endTime = uptimeMillis();
- stats->set_success(err == NO_ERROR);
stats->set_exec_duration_ms(endTime - startTime);
if (err != NO_ERROR) {
ALOGW("Incident section %s (%d) failed: %s. Stopping report.",
(*section)->name.string(), id, strerror(-err));
+ // Execute() has already recorded this status. Only update if there's new failure.
+ stats->set_success(false);
goto DONE;
}
(*reportByteSize) += stats->report_size_bytes();
@@ -214,11 +216,13 @@
id, IIncidentReportStatusListener::STATUS_FINISHED);
}
}
- ALOGD("Finish incident report section %d '%s'", id, (*section)->name.string());
+ VLOG("Finish incident report section %d '%s'", id, (*section)->name.string());
+ sectionCount++;
}
}
DONE:
+ ALOGD("Incident reporting took %d sections.", sectionCount);
// Reports the metdadata when taking the incident report.
if (!isTest) metadataSection.Execute(&batch);
diff --git a/cmds/incidentd/src/Section.cpp b/cmds/incidentd/src/Section.cpp
index cd48af9..10d2268 100644
--- a/cmds/incidentd/src/Section.cpp
+++ b/cmds/incidentd/src/Section.cpp
@@ -75,6 +75,7 @@
stats->set_dump_duration_ms(buffer.durationMs());
stats->set_timed_out(buffer.timedOut());
stats->set_is_truncated(buffer.truncated());
+ stats->set_success(!buffer.timedOut() && !buffer.truncated());
}
// Reads data from FdBuffer and writes it to the requests file descriptor.
@@ -83,7 +84,8 @@
status_t err = -EBADF;
EncodedBuffer::iterator data = buffer.data();
PrivacyBuffer privacyBuffer(get_privacy_of_section(id), data);
- int writeable = 0;
+ IncidentMetadata::SectionStats* stats = requests->sectionStats(id);
+ int nPassed = 0;
// The streaming ones, group requests by spec in order to save unnecessary strip operations
map<PrivacySpec, vector<sp<ReportRequest>>> requestsBySpec;
@@ -99,7 +101,18 @@
for (auto mit = requestsBySpec.begin(); mit != requestsBySpec.end(); mit++) {
PrivacySpec spec = mit->first;
err = privacyBuffer.strip(spec);
- if (err != NO_ERROR) return err; // it means the privacyBuffer data is corrupted.
+ if (err != NO_ERROR) {
+ // Privacy Buffer is corrupted, probably due to an ill-formatted proto. This is a
+ // non-fatal error. The whole report is still valid. So just log the failure.
+ ALOGW("Failed to strip section %d with spec %d: %s",
+ id, spec.dest, strerror(-err));
+ stats->set_success(false);
+ stats->set_error_msg("Failed to strip section: privacy buffer corrupted, probably "
+ "due to an ill-formatted proto");
+ nPassed++;
+ continue;
+ }
+
if (privacyBuffer.size() == 0) continue;
for (auto it = mit->second.begin(); it != mit->second.end(); it++) {
@@ -114,7 +127,7 @@
request->err = err;
continue;
}
- writeable++;
+ nPassed++;
VLOG("Section %d flushed %zu bytes to fd %d with spec %d", id, privacyBuffer.size(),
request->fd, spec.dest);
}
@@ -125,7 +138,15 @@
if (requests->mainFd() >= 0) {
PrivacySpec spec = PrivacySpec::new_spec(requests->mainDest());
err = privacyBuffer.strip(spec);
- if (err != NO_ERROR) return err; // the buffer data is corrupted.
+ if (err != NO_ERROR) {
+ ALOGW("Failed to strip section %d with spec %d for dropbox: %s",
+ id, spec.dest, strerror(-err));
+ stats->set_success(false);
+ stats->set_error_msg("Failed to strip section: privacy buffer corrupted, probably "
+ "due to an ill-formatted proto");
+ nPassed++;
+ goto DONE;
+ }
if (privacyBuffer.size() == 0) goto DONE;
err = write_section_header(requests->mainFd(), id, privacyBuffer.size());
@@ -138,7 +159,7 @@
requests->setMainFd(-1);
goto DONE;
}
- writeable++;
+ nPassed++;
VLOG("Section %d flushed %zu bytes to dropbox %d with spec %d", id, privacyBuffer.size(),
requests->mainFd(), spec.dest);
// Reports bytes of the section uploaded via dropbox after filtering.
@@ -147,7 +168,7 @@
DONE:
// only returns error if there is no fd to write to.
- return writeable > 0 ? NO_ERROR : err;
+ return nPassed > 0 ? NO_ERROR : err;
}
// ================================================================================
@@ -213,6 +234,8 @@
stats.timed_out());
proto.write(FIELD_TYPE_BOOL | IncidentMetadata::SectionStats::kIsTruncatedFieldNumber,
stats.is_truncated());
+ proto.write(FIELD_TYPE_STRING | IncidentMetadata::SectionStats::kErrorMsgFieldNumber,
+ stats.error_msg());
proto.end(token);
}
@@ -439,8 +462,9 @@
status_t err = NO_ERROR;
pthread_t thread;
pthread_attr_t attr;
- bool timedOut = false;
+ bool workerDone = false;
FdBuffer buffer;
+ IncidentMetadata::SectionStats* stats = requests->sectionStats(this->id);
// Data shared between this thread and the worker thread.
sp<WorkerThreadData> data = new WorkerThreadData(this);
@@ -475,8 +499,7 @@
// Loop reading until either the timeout or the worker side is done (i.e. eof).
err = buffer.read(data->pipe.readFd().get(), this->timeoutMs);
if (err != NO_ERROR) {
- // TODO: Log this error into the incident report.
- ALOGW("[%s] reader failed with error '%s'", this->name.string(), strerror(-err));
+ ALOGE("[%s] reader failed with error '%s'", this->name.string(), strerror(-err));
}
// Done with the read fd. The worker thread closes the write one so
@@ -484,31 +507,32 @@
data->pipe.readFd().reset();
// If the worker side is finished, then return its error (which may overwrite
- // our possible error -- but it's more interesting anyway). If not, then we timed out.
+ // our possible error -- but it's more interesting anyway). If not, then we timed out.
{
unique_lock<mutex> lock(data->lock);
- if (!data->workerDone) {
- // We timed out
- timedOut = true;
- } else {
- if (data->workerError != NO_ERROR) {
- err = data->workerError;
- // TODO: Log this error into the incident report.
- ALOGW("[%s] worker failed with error '%s'", this->name.string(), strerror(-err));
- }
+ if (data->workerError != NO_ERROR) {
+ err = data->workerError;
+ ALOGE("[%s] worker failed with error '%s'", this->name.string(), strerror(-err));
}
+ workerDone = data->workerDone;
}
- write_section_stats(requests->sectionStats(this->id), buffer);
- if (timedOut || buffer.timedOut()) {
- ALOGW("[%s] timed out", this->name.string());
+ write_section_stats(stats, buffer);
+ if (err != NO_ERROR) {
+ char errMsg[128];
+ snprintf(errMsg, 128, "[%s] failed with error '%s'",
+ this->name.string(), strerror(-err));
+ stats->set_success(false);
+ stats->set_error_msg(errMsg);
return NO_ERROR;
}
-
- // TODO: There was an error with the command or buffering. Report that. For now
- // just exit with a log messasge.
- if (err != NO_ERROR) {
- ALOGW("[%s] failed with error '%s'", this->name.string(), strerror(-err));
+ if (buffer.truncated()) {
+ ALOGW("[%s] too large, truncating", this->name.string());
+ // Do not write a truncated section. It won't pass through the PrivacyBuffer.
+ return NO_ERROR;
+ }
+ if (!workerDone || buffer.timedOut()) {
+ ALOGW("[%s] timed out", this->name.string());
return NO_ERROR;
}
@@ -617,14 +641,8 @@
sp<IBinder> service = defaultServiceManager()->checkService(mService);
if (service == NULL) {
- // Returning an error interrupts the entire incident report, so just
- // log the failure.
- // TODO: have a meta record inside the report that would log this
- // failure inside the report, because the fact that we can't find
- // the service is good data in and of itself. This is running in
- // another thread so lock that carefully...
ALOGW("DumpsysSection: Can't lookup service: %s", String8(mService).string());
- return NO_ERROR;
+ return NAME_NOT_FOUND;
}
service->dump(pipeWriteFd, mArgs);