DTrace Code Samples
DTrace supports the D programing language. This topic provide D code samples.
For general information about DTrace on Windows, see DTrace.
For detailed information about DTrace see the OpenDTrace Speciļ¬cation version 1.0 at the University Of Cambridge.
Note
DTrace is supported in the Insider builds of Windows after version 18980 and Windows Server Build 18975.
Additional sample scripts
Additional D scripts applicable for Windows scenarios are available in the samples directory of the DTrace source code.
https://github.com/microsoft/DTrace-on-Windows/tree/windows/samples/windows
A set of useful opentrace toolkit scripts is available at https://github.com/opendtrace/toolkit.
Disk Usage by Name
This script provides the disk counters for a given executable name. The executable name is case sensitive and is provided on the command line when the script is started.
#pragma D option quiet
#pragma D option destructive
intptr_t curptr;
struct nt`_EPROCESS *eprocess_ptr;
int found;
int firstpass;
BEGIN
{
curptr = (intptr_t) ((struct nt`_LIST_ENTRY *) (void*)&nt`PsActiveProcessHead)->Flink;
found = 0;
firstpass = 1;
bytesread = 0;
byteswrite = 0;
readcount = 0;
writecount = 0;
flushcount = 0;
}
tick-1ms
/found == 0/
{
/* use this for pointer parsing */
if (found == 0)
{
eprocess_ptr = (struct nt`_EPROCESS *)(curptr - offsetof(nt`_EPROCESS, ActiveProcessLinks));
processid = (string) eprocess_ptr->ImageFileName;
if ($1 == processid)
{
found = 1;
}
else
{
curptr = (intptr_t) ((struct nt`_LIST_ENTRY *) (void*)curptr)->Flink;
}
}
}
tick-2s
{
system ("cls");
if (found == 1)
{
if (firstpass)
{
firstpass = 0;
bytesread = (unsigned long long) eprocess_ptr->DiskCounters->BytesRead;
byteswrite = (unsigned long long) eprocess_ptr->DiskCounters->BytesWritten;
readcount = eprocess_ptr->DiskCounters->ReadOperationCount;
writecount = eprocess_ptr->DiskCounters->WriteOperationCount;
flushcount = eprocess_ptr->DiskCounters->FlushOperationCount;
}
else
{
bytesread = (unsigned long long) (eprocess_ptr->DiskCounters->BytesRead - bytesread);
byteswrite = (unsigned long long) (eprocess_ptr->DiskCounters->BytesWritten - byteswrite);
readcount = eprocess_ptr->DiskCounters->ReadOperationCount - readcount;
writecount = eprocess_ptr->DiskCounters->WriteOperationCount - writecount;
flushcount = eprocess_ptr->DiskCounters->FlushOperationCount - flushcount;
printf("*** Reports disk read/write every second *** \n");
printf("Process name: %s\n", eprocess_ptr->ImageFileName);
printf("Process Id: %d\n", (int) eprocess_ptr->UniqueProcessId);
printf("Bytes Read %llu \n", (unsigned long long) bytesread);
printf("Bytes Written %llu \n", (unsigned long long) byteswrite);
printf("Read Operation Count %d \n", readcount);
printf("Write Operation Count %d \n",writecount);
printf("Flush Operation Count %d \n", flushcount);
bytesread = (unsigned long long) eprocess_ptr->DiskCounters->BytesRead;
byteswrite = (unsigned long long) eprocess_ptr->DiskCounters->BytesWritten;
readcount = eprocess_ptr->DiskCounters->ReadOperationCount;
writecount = eprocess_ptr->DiskCounters->WriteOperationCount;
flushcount = eprocess_ptr->DiskCounters->FlushOperationCount;
}
}
else
{
printf("No matching process found for %s \n", $1);
exit(0);
}
}
Save the file as diskuagebyname.d and use the -s option to run the test script. Provide a case sensitive name of the desired exe such as Notepad.exe
C:\>dtrace -s diskusagebyname.d Notepad.exe
*** Reports disk read/write every second ***
Process name: Notepad.exe
Process Id: 6012
Bytes Read 0
Bytes Written 0
Read Operation Count 0
Write Operation Count 0
Flush Operation Count 0
*** Reports disk read/write every second ***
Process name: cmd.exe
Process Id: 4428
Bytes Read 18446744073709480960
Bytes Written 18446744073709522944
Read Operation Count -5
Write Operation Count -7
Flush Operation Count 0
To keep a running log of disk activity redirect the command out to a text file as shown here.
C:\>dtrace -s diskusagebyname.d Notepad.exe>>diskstats.txt
Dumping Memory Usage
In some diagnostics cases, there is a need to dump kernel structures to understand the situation. This code shows an example to dump system memory usage. The below D-script fires a timer probe every 3 second and refreshes system memory usage.
#pragma D option quiet
#pragma D option destructive
tick-3s
{
system ("cls");
this->pp = ((struct nt`_MI_PARTITION *) &nt`MiSystemPartition);
printf("***** Printing System wide page information ******* \n");
printf( "Total Pages Entire Node: %u MB \n", this->pp->Core.NodeInformation->TotalPagesEntireNode*4096/(1024*1024));
printf("Total Available Pages: %u Mb \n", this->pp->Vp.AvailablePages*4096/(1024*1024));
printf("Total ResAvail Pages: %u Mb \n", this->pp->Vp.ResidentAvailablePages*4096/(1024*1024));
printf("Total Shared Commit: %u Mb \n", this->pp->Vp.SharedCommit*4096/(1024*1024));
printf("Total Pages for PagingFile: %u Mb \n", this->pp->Vp.TotalPagesForPagingFile*4096/(1024*1024));
printf("Modified Pages : %u Mb \n", this->pp->Vp.ModifiedPageListHead.Total*4096/(1024*1024));
printf("Modified No Write Page Count : %u Mb \n", this->pp->Vp.ModifiedNoWritePageListHead.Total*4096/(1024*1024));
printf("Bad Page Count : %d Mb \n", this->pp->PageLists.BadPageListHead.Total*4096/(1024*1024));
printf("Zero Page Count : %d Mb \n", this->pp->PageLists.ZeroedPageListHead.Total*4096/(1024*1024));
printf("Free Page Count : %d Mb \n", this->pp->PageLists.FreePageListHead.Total*4096/(1024*1024));
/********** Printing Commit info ******************/
this->commit = ((struct nt`_MI_PARTITION_COMMIT ) ((struct nt`_MI_PARTITION *) &nt`MiSystemPartition)->Commit);
printf("***** Printing Commit Info ******* \n");
printf("Total Committed Pages: %u Mb \n", this->pp->Vp.TotalCommittedPages*4096/(1024*1024));
printf("Total Commit limit: %u Mb \n", this->pp->Vp.TotalCommitLimit*4096/(1024*1024));
printf("Peak Commitment: %u Mb \n", this->commit.PeakCommitment*4096/(1024*1024));
printf("Low Commit Threshold: %u Mb \n", this->commit.LowCommitThreshold*4096/(1024*1024));
printf("High Commit Threshold: %u Mb \n", this->commit.HighCommitThreshold*4096/(1024*1024));
printf("System Commit Reserve: %u Mb \n", this->commit.SystemCommitReserve*4096/(1024*1024));
}
Save the file as dumpmemoryusage.d.
This sample uses symbols, so set the symbol path as discussed in installation.
set _NT_SYMBOL_PATH=srv*C:\symbols*https://msdl.microsoft.com/download/symbols
Use the -s option to run the test script.
C:\>dtrace -s dumpmemoryusage.d
***** Printing System wide page information *******
Total Pages Entire Node: 2823 MB
Total Available Pages: 622 Mb
Total ResAvail Pages: 2369 Mb
Total Shared Commit: 166 Mb
Total Pages for PagingFile: 30 Mb
Modified Pages : 30 Mb
Modified No Write Page Count : 0 Mb
Bad Page Count : 0 Mb
Zero Page Count : 7 Mb
Free Page Count : 0 Mb
***** Printing Commit Info *******
Total Committed Pages: 2401 Mb
Total Commit limit: 4551 Mb
Peak Commitment: 2989 Mb
Low Commit Threshold: 3413 Mb
High Commit Threshold: 4295 Mb
System Commit Reserve: 5 Mb
This program is designed to continue to monitor memory usage. Press CTRL+C to exit.
Identifying heap free time breakdown for an activity
This sample provides a breakdown of the RtlFreeHeap function into sub-functions and display max time taken for these functions to execute.
/* Mark script destructive as we call "cls" in the tick-1sec provider*/
#pragma D option destructive
/* Program to calculate time taken in RtlFreeHeap function. Hit Ctrl-C to break. */
dtrace:::BEGIN
{
counter1 = 0;
}
tick-1sec
{
system ("cls");
printf("count of function hit = %d << Press Ctrl-C to exit >> \n", counter1);
}
/* Instrument entry of RtlFreeHeap*/
pid$1:ntdll:RtlFreeHeap:entry
{
/* self is thread local */
self->ts = timestamp;
self->depth = 0;
counter1++;
}
pid$1:ntdll:RtlFreeHeap:return
/self->ts/
{
this->ts = timestamp - self->ts;
@disttime = quantize(this->ts);
@funcName[probemod,probefunc] = max(this->ts);
self->ts = 0;
}
pid$1:::entry
/ self->ts /
{
self->functime[self->depth] = timestamp;
self->depth++;
}
pid$1:::return
/ self->ts /
{
if (self->depth > 0)
{
self->depth--;
/* this is a local scope */
this->ts = timestamp - self->functime[self->depth];
@funcName[probemod,probefunc] = max(this->ts);
}
}
END
{
system ("cls");
/* Print overall time distribution for RtlFreeHeap */
printa (@disttime);
/* Print top 15 functions along with average time spent executing them */
trunc (@funcName, 15);
printa( @funcName);
}
Save the file as heapfree.d.
This script takes a process ID or PID. To test the script, run tasklist at the command prompt and locate the desired PID.
C:\>tasklist
Image Name PID Session Name Session# Mem Usage
========================= ======== ================ =========== ============
System Idle Process 0 Services 0 8 K
System 4 Services 0 108 K
Registry 72 Services 0 35,576 K
...
Notepad.exe 7944 31C5CE94259D4006 2 20,736 K
Specify the PID as a parameter and use the -s option to run the test script as shown.
C:\>dtrace -s heapfree.d 7944
count of function hit = 28 <<Press Ctrl-C to exit>>
Perform what ever task is of interest in the program associated with the PID, for example typing in Notepad or changing the font.
Press CRTL-C and the heap usage will be displayed along with stack information.
value ------------- Distribution ------------- count
4096 | 0
8192 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ 23
16384 |@@@@@@ 4
32768 | 0
65536 | 0
131072 |@ 1
262144 | 0
ntdll RtlTryEnterCriticalSection 9444
ntdll RtlGetCurrentServiceSessionId 12302
ntdll RtlFreeHeap 151354
Memory Pool Tracking
Note: Run this script with aggsortkeypos variable set. This variable informs D-Trace to sort the output based on first index (size).
Usage:
dtrace -s <script.d> <Time> <Tag> -x aggsortkey -x aggsortkeypos=1
Example for tracking KSec leaks: dtrace -s PoolTrackingSummary.d "120s" 0x6365734b -x aggsortkey -x aggsortkeypos=1
The <Tag> value is an encoded ASCII value of the memory pool tag. For example to encode the tag value for the nt file system - NtFf, first reverse the two sets of letters, it will be fFtN. Converting the ASCII characters to hex will be 46 66 74 4e, or 0x4666744e as a parameter to our program.
Another example would be converting Ksec.
First reverse the order of the letters: cesK
Then convert those characters to hex, 63 65 73 4b or 0x6365734b as a parameter to our program.
Note that upper lower case is important to match the pool tag.
Output: The script runs for 120 seconds and outputs the KSec allocation/free summary. You can set this to whatever time is necessary.
#pragma D option destructive
#pragma D option quiet
#pragma D option dynvarsize=240m
#pragma D option bufsize=120m
#pragma D option aggsize=120m
fbt:nt:ExAllocatePoolWithTag:entry
{
/* This is E100 in reserve. Convert ASCII to Hex => Hex('001E').*/
if (arg2 == $2)
{
self->size = (unsigned int) arg1;
@allocstack[stack(), self->size] = count();
}
}
fbt:nt:ExAllocatePoolWithTag:return
/self->size/
{
@mem[(uintptr_t) arg1] = sum(1);
addr[(uintptr_t) arg1] = 1;
/* printf("%Y: Execname %s allocated size %d bytes return ptr %x", walltimestamp, execname, (uint64_t) self->size, (uintptr_t) arg1 );*/
size[(uintptr_t) arg1] = self->size;
@sizealloc[self->size] = count();
@delta[self->size] = sum(1);
self->size = 0;
}
fbt:nt:ExFreePoolWithTag:entry
/addr[(uintptr_t) arg0]/
{
@mem[(uintptr_t) arg0] = sum (-1);
addr[(uintptr_t) arg0] -= 1;
@sizefree[size[(uintptr_t) arg0]] = count();
@delta[size[(uintptr_t) arg0]] = sum(-1);
}
tick-$1
{
exit(0);
}
END
{
printf("%10s %10s %10s %10s\n", "SIZE", "ALLOC", "FREE", "DELTA");
printa("%10d %@10d %@10d %@10d\n", @sizealloc, @sizefree, @delta);
printf("Printing stacks \n");
printa (@allocstack);
printf("== REPORT ==\n\n");
printa("0x%x => %@u\n",@mem);
}
Save the file as pooltrackingsummary.d and use the -s option to run the test script, providing a tag value and other parameters discussed above.
The 120s option, the program runs for two minutes. During that time, use Windows to exercise the monitored memory pool. For example load and unload a web browser or other programs.
C:\>dtrace -s pooltrackingsummary.d "120s" 0x4666744e -x aggsortkey -x aggsortkeypos=1
SIZE ALLOC FREE DELTA
1552 16 0 16
Printing stacks
Ntfs.sys`NtfsCreateFcb+0x388
Ntfs.sys`NtfsCreateNewFile+0xaa8
Ntfs.sys`NtfsCommonCreate+0x2303
Ntfs.sys`NtfsFsdCreate+0x284
nt`IofCallDriver+0x55
FLTMGR.SYS`FltpLegacyProcessingAfterPreCallbacksCompleted+0x1b9
FLTMGR.SYS`FltpCreate+0x324
nt`IofCallDriver+0x55
nt`IoCallDriverWithTracing+0x34
nt`IopParseDevice+0x6ac
nt`ObpLookupObjectName+0x3fe
nt`ObOpenObjectByNameEx+0x1fa
nt`IopCreateFile+0x40f
nt`NtCreateFile+0x79
nt`KiSystemServiceCopyEnd+0x38
1552 3
Ntfs.sys`NtfsCreateFcb+0x388
Ntfs.sys`NtfsOpenFile+0x2d7
Ntfs.sys`NtfsCommonCreate+0x25a8
Ntfs.sys`NtfsFsdCreate+0x284
nt`IofCallDriver+0x55
FLTMGR.SYS`FltpLegacyProcessingAfterPreCallbacksCompleted+0x1b9
FLTMGR.SYS`FltpCreate+0x324
nt`IofCallDriver+0x55
nt`IoCallDriverWithTracing+0x34
nt`IopParseDevice+0x6ac
nt`ObpLookupObjectName+0x3fe
nt`ObOpenObjectByNameEx+0x1fa
nt`IopCreateFile+0x40f
nt`NtCreateFile+0x79
nt`KiSystemServiceCopyEnd+0x38
1552 4
Ntfs.sys`NtfsCreateFcb+0x388
Ntfs.sys`NtfsOpenFile+0x2d7
Ntfs.sys`NtfsCommonCreate+0x25a8
Ntfs.sys`NtfsFsdCreate+0x284
nt`IofCallDriver+0x55
FLTMGR.SYS`FltpLegacyProcessingAfterPreCallbacksCompleted+0x1b9
FLTMGR.SYS`FltpCreate+0x324
nt`IofCallDriver+0x55
nt`IoCallDriverWithTracing+0x34
nt`IopParseDevice+0x6ac
nt`ObpLookupObjectName+0x3fe
nt`ObOpenObjectByNameEx+0x1fa
nt`IopCreateFile+0x40f
nt`NtOpenFile+0x58
nt`KiSystemServiceCopyEnd+0x38
1552 3
...
== REPORT ==
0xffffd304c98dd380 => 1
0xffffd304cc4655a0 => 1
0xffffd304cccf15a0 => 1
0xffffd304ccdeb990 => 1
0xffffd304ce048760 => 1
0xffffd304cf1ee990 => 1
0xffffd304d0473010 => 1
0xffffd304d12075a0 => 1
0xffffd304d14135a0 => 1
0xffffd304d1674010 => 1
0xffffd304d33b3660 => 1
0xffffd304d3b29010 => 1
0xffffd304d42c6010 => 1
0xffffd304d48b2010 => 1
0xffffd304de1fa5f0 => 1
0xffffd304e1ad56a0 => 1
Comparing GUIDs
Although, DTrace does not support GUID's natively, you can create struct as show in this sample code to work with them. Here is a sample of how you can compare GUIDs.
nt`_GUID guidcmp;
/* Sleep After GUID: 29f6c1db-86da-48c5-9fdb-f2b67b1f44da */
dtrace:::BEGIN
{
printf("Begin\n");
guidcmp.Data1 = 0x29f6c1db;
guidcmp.Data2 = 0x86da;
guidcmp.Data3 = 0x48c5;
guidcmp.Data4[0] = 0x9f;
guidcmp.Data4[1] = 0xdb;
guidcmp.Data4[2] = 0xf2;
guidcmp.Data4[3] = 0xb6;
guidcmp.Data4[4] = 0x7b;
guidcmp.Data4[5] = 0x1f;
guidcmp.Data4[6] = 0x44;
guidcmp.Data4[7] = 0xda;
}
pid$target:PowrProf:PowerReadACValueIndexEx:entry
{
cidstr = (nt`_GUID *) (copyin(arg2, sizeof(nt`_GUID)));
printf("\nPrinting GUID to compare\n");
print(guidcmp);
printf("\nPrinting GUID received \n");
print(*cidstr);
if ( (cidstr->Data1 == guidcmp.Data1) &&
(cidstr->Data2 == guidcmp.Data2) &&
(cidstr->Data3 == guidcmp.Data3) &&
(cidstr->Data4[0] == guidcmp.Data4[0]) &&
(cidstr->Data4[1] == guidcmp.Data4[1]) &&
(cidstr->Data4[2] == guidcmp.Data4[2]) &&
(cidstr->Data4[3] == guidcmp.Data4[3]) &&
(cidstr->Data4[4] == guidcmp.Data4[4]) &&
(cidstr->Data4[5] == guidcmp.Data4[5]) &&
(cidstr->Data4[6] == guidcmp.Data4[6]) &&
(cidstr->Data4[7] == guidcmp.Data4[7]) )
{
printf("GUID matched \n");
}
else
{
printf("No match");
}
}
dtrace:::END
{
printf("End\n");
}
Save the file as comparequid.d and use the -s option to run the test script, providing the paramter shown below.
C:\Windows\system32>dtrace -s compareguid.d -c "powercfg /qh scheme_current sub_sleep standbyidle"
dtrace: script 'compareguid.d' matched 9 probes
CPU ID FUNCTION:NAME
0 1 :BEGIN Begin
Power Scheme GUID: 381b4222-f694-41f0-9685-ff5bb260df2e (Balanced)
GUID Alias: SCHEME_BALANCED
Subgroup GUID: 238c9fa8-0aad-41ed-83f4-97be242c8f20 (Sleep)
GUID Alias: SUB_SLEEP
Power Setting GUID: 29f6c1db-86da-48c5-9fdb-f2b67b1f44da (Sleep after)
GUID Alias: STANDBYIDLE
Minimum Possible Setting: 0x00000000
Maximum Possible Setting: 0xffffffff
Possible Settings increment: 0x00000001
Possible Settings units: Seconds
Current AC Power Setting Index: 0x00000708
Current DC Power Setting Index: 0x00000384
dtrace: pid 7560 has exited
0 42695 PowerReadACValueIndexEx:entry
Printing GUID to compare
struct _GUID {
UInt32 Data1 = 0x29f6c1db
UInt16 Data2 = 0x86da
UInt16 Data3 = 0x48c5
UInt8 [8] Data4 = [ 0x9f, 0xdb, 0xf2, 0xb6, 0x7b, 0x1f, 0x44, 0xda ]
}
Printing GUID received
struct _GUID {
UInt32 Data1 = 0x29f6c1db
UInt16 Data2 = 0x86da
UInt16 Data3 = 0x48c5
UInt8 [8] Data4 = [ 0x9f, 0xdb, 0xf2, 0xb6, 0x7b, 0x1f, 0x44, 0xda ]
}
0 42695 PowerReadACValueIndexEx:entry GUID matched
See Also
Feedback
Coming soon: Throughout 2024 we will be phasing out GitHub Issues as the feedback mechanism for content and replacing it with a new feedback system. For more information see: https://aka.ms/ContentUserFeedback.
Submit and view feedback for