Print a timed per-thread microsecond-timed callgraph, complete with function parameters and return values. The first parameter names the function probe points to trace. The optional second parameter names the probe points for trigger functions, which acts to enable tracing for only those functions that occur while the current thread is nested within the trigger.
This script places a set of probes (specified by $1), each of which monitors the state of some context $variable expression (specified by $2). Whenever the value changes, with respect to the active thread, the event is traced.
The whythefail.stp script prints a statement-execution trace for a given function, but only for those runs of the function that ended up with a (configurable) post-return condition.
The script watches all IPv4 network traffic on the system. On exit the script prints a list showing the number of packets sent along source IP address / destination IP address pair encountered, and the total number of bytes sent among the pair. The list is ordered from greatest to least number of packets seen among the source/destination pairs.
The script watches all IPv4 network traffic on the system. The data is output in JSON format and includes the number of packets sent along source IP address / destination IP address pair encountered, and the total number of bytes sent among the pair.
The cycle_thief.stp script instruments the scheduler and IRQ handler to determine which processes and interrupts are competing with the specified task for the cpu cycles. This script uses the '-c' or '-x' options to focus on a specific task. The script output the number of times the task migrates between processors, histograms showing the length of time on and off processor, lists of processes running while the task is off the processor, and the interrupts that occurred while the task was running.
Every 5 seconds, print out a list of 25 processes that took the most system time with information about the processes. Includes information on processes that may have exited while the script was running. The script contains configuration options listed in the script source.
The script loosely emulates strace, when applied to individual processes or hierarchies (via -c/-x), or the entire system (without -c/-x). A few output configuration parameters may be set with -G.
Prints a periodic tabular report about the counts of syscall activity of all threads on the system, along with a textual recent-syscall-history for each
The fntimes.stp script monitors the execution time history of a given function family (assumed non-recursive). Each time (beyond a warmup interval) is then compared to the historical maximum. If it exceeds a certain threshold (250%), a message is printed.
The latencytap.stp script collects data on the intervals processes are deactivated (sleeping). The script categorizes the reasons for the sleeps by analyzing the backtraces and displays a sorted list of the top 20 causes from largest total sum time sleeping to smallest. The output is updated every 30 seconds. The script needs to be compiled with the '--all-modules' option to produce reasons for sleeps caused by modules. Optionally, this script can be used with the '-c' or '-x' options to focus on a specific PID.
The linetimes.stp script takes two arguments: where to find the function and the function name. linetimes.stp will instrument each line in the function. It will print out the number of times that the function is called, a table with the average and maximum time each line takes, and control flow information when the script exits.
The pf4.stp script sets up time-based sampling. Every five seconds it prints out a sorted list with the top twenty kernel and/or user stack backtraces (on a per-cpu basis). Use any of --ldd, --all-modules, -d MODULE, -d /PATH/TO/EXEC to add more symbolic info.
The thread-times.stp script sets up time-based sampling. Every five seconds it prints out a sorted list with the top twenty threads occupying the CPUs, broken down as a percentage of user and kernel time.
The kvm_service_time.stp script tracks the statistics about the amount of time that the processor left the guest virtual machine for each exit reason (for example fixing up a page table or handling an IO operation). When the script exits it prints out the number of times each exit reason was encountered, the total duration of time it left the guest VM, the minimum time, the average time, and the maximum time in microseconds for that exit reason. On Linux 2.6.38 and newer kernel the script can automatically determine whether it is running on Intel or AMD processors. For older kernels with a kernel.trace("kvm_exit") tracepoint that does not have the $isa parameter you can explicitly state the kvm type with a "-G kvm=intel" or "-G kvm=amd" on the command line.
The libguestfs_log.stp script prints a log of when various libgueststartup steps are encountered. The first column is microseconds since the script started. The second column is the time elapsed in microseconds between the previous and current events and the third column is the event name.
SystemTap has multiple passes to convert the text of a SystemTap script into instrumentation that actually collect data on the system. The stap_time.stp script uses the markers in SystemTap to note the amount of time that each of the passes requires. When SystemTap completes pass 4 (compiling the instrumentation into a kernel module) the script print out the script name followed by the amount of time in milliseconds required for Pass 0 (command line option parsing), Pass 1 (script parsing), Pass 2 (elaboration), Pass 3 (code generation), and Pass 4 (module compilation). The resulting data can be analyzed to determine if there are issues with the amount of time that systemtap takes to generate instrumentation.
The watchdog.stp script provides a watchdog timer mechanism for arbitrary events. The script takes three arguments: the events to start watchdog timer, the event to stop the watchdog timer, and the time in millseconds for the watchdog. If the watchdog timer is exceed, the script will trigger a stack backtrace of the user-process that timed out using pstack. This script can be used to diagnose what the userspace application is doing when a slower than expected operation occurs.
The Linux kernel function smp_call_function causes expensive inter-processor interrupts (IPIs). The scf.stp script tallies the processes and backtraces causing the interprocessor interrupts to identify the cause of the expensive IPI. On exit the script prints the tallies in descending frequency.
When a reschedule occurs during an AIO io_submit call, accumulate the traceback in a histogram. When the script exits prints out a sorted list from most common to least common backtrace.
This script reports on the last few free(3) libc calls done by processes (possibly restricted by stap -x/-c), along with a userspace backtrace at those moments.
Attaches to the kernel audit-log paths (also used by libaudit), and log every record being sent, along with a user-space backtrace of the process that caused it.
The pstrace_exec.stp script watches each exec operation. If the exec contains a substring that matches the script's command-line argument, it prints out that process and all of its ancestors.
The script monitors the time that threads spend in waiting for IO operations (in "D" state) in the wait_for_completion function. If a thread spends over 10ms, its name and backtrace is printed, and later so is the total delay.
The pf4.stp script sets up time-based sampling. Every five seconds it prints out a sorted list with the top twenty kernel and/or user stack backtraces (on a per-cpu basis). Use any of --ldd, --all-modules, -d MODULE, -d /PATH/TO/EXEC to add more symbolic info.
The libguestfs_log.stp script prints a log of when various libgueststartup steps are encountered. The first column is microseconds since the script started. The second column is the time elapsed in microseconds between the previous and current events and the third column is the event name.
The script prints a table showing the available attributes (bold, underline, and inverse) with color combinations for the ans_set_color() function in the ansi.stp tapset.
The stopwatch.stp script illustrates how to use multiple stopwatches record how much wallclock time a process spends in kernel- and user-space. On exit the script prints out the time in seconds, milliseconds, microseconds, and nanoseconds. Note that this output of this script is not directly comparable to the time command because time records the time that the process is actually active in kernel- and user-space.
The mmfilepage.stp script uses the virtual memory tracepoints available in some kernels to track the number of faults, copy on writes mapping, and unmapping operations for file backed pages. When the script is terminated the counts are printed for each process that allocated pages while the script was running. The mmfilepage.stp script is useful in debugging leaks in the mapped file regions of a process.
The mmwriteback.stp script uses the virtual memory tracepoints available in some kernels to report all of the file writebacks that occur form kupdate, pdflush and kjournald while the script is running. It's useful in determining where writes are coming from on a supposedly idle system that is experiencing unexpected IO.
The sk_stream-wait_memory.stp prints a time stamp, executable, and pid each time a process blocks due to the send buffer being full. A similar entry is printed each time a process continues because there is room in the buffer.
Print a timed per-thread microsecond-timed callgraph, complete with pretty-printed function parameters and return values. The first parameter names the function probe points to trace. The optional second parameter names the probe points for trigger functions, which acts to enable tracing for only those functions that occur while the current thread is nested within the trigger.
Print a timed per-thread microsecond-timed callgraph, complete with function parameters and return values. The first parameter names the function probe points to trace. The optional second parameter names the probe points for trigger functions, which acts to enable tracing for only those functions that occur while the current thread is nested within the trigger.
The container_check.stp script monitors the use of linux capablities and optionally forbidden syscalls by a process and its children. On exit the script prints out lists showing the capabilies used by each executable, which syscall used specific capabilites for each executable, a list of forbidden syscalls used, and details on any syscalls that failed during monitoring. This script is designed to help diagnose issues caused by restricted capabilies and syscalls when running an application in a container. If the script warns about skipped probes, the number of active kretprobes may need to be increased with "-DKRETACTIVE=100" option on the command line
The pstree.stp script generates a process diagram in DOT form. For instance, it may be useful on a 'make' command to see all the processes that are started.
The deviceseeks.stp script generates a periodic histogram showing the frequency of different sized seeks (in sectors) on each device, or optional given device-name regexp
This script monitors a filesystem implementations for early internal indications of ENOSPC, and reports these to the system logger and the systemtap console.
The nfsd_unlink.stp script lists the ip address and file name each time time a file is being removed or unlinked by the nfsd. This script is run on the nfs server.
The script watches each open, close, read, and write syscalls on the system. For each file the scripts observes opened it accumulates the amount of wall clock time spent in read and write operations and the number of bytes read and written. When a file is closed the script prints out a pair of lines for the file. Both lines begin with a timestamp in microseconds, the PID number, and the executable name in parentheses. The first line with the "access" keyword lists the file name, the attempted number of bytes for the read and write operations. The second line with the "iotime" keyword list the file name and the number of microseconds accumulated in the read and write syscalls.
Run pfiles.stp to produce a human-readable summary of all open file descriptors of a given process. Specify the process-id as -x PID for fastest performance.
This script monitors a filesystem implementations for early internal indications of ENOSPC, and reports these to the system logger and the systemtap console.
This dynamically traces two common file system functions: do_sync_read() and do_sync_write(), and reports a histogram distribution of latency. Many, but not all, file systems and workloads use these functions. Tracing their time provides one view of suffered file system latency.
This dynamically traces two common file system functions: do_sync_read() and do_sync_write(), and shows details of each call that is slower than a threshold. Many, but not all, file systems and workloads use these functions. Tracing their time provides one view of suffered file system latency.
A basic "Floating Point" program implemented in SystemTap script. It extracts floating point from benchmark-sdt section in main.cxx and takes one pseudo floating point, printing out results of various simple floating point operations.
The script prints a table showing the available attributes (bold, underline, and inverse) with color combinations for the ans_set_color() function in the ansi.stp tapset.
The func_time_stats.stp script tracks the wall clock time for each invocation of the function probe listed as the first command line argument. When the script exits it prints out the minimum, average, and maximum times in microseconds followed by a count of times that the function was called and a histogram showing the distributions of times.
The whythefail.stp script prints a statement-execution trace for a given function, but only for those runs of the function that ended up with a (configurable) post-return condition.
The functioncallcount.stp script takes one argument, a list of functions to probe. The script will run and count the number of times that each of the functions on the list is called. On exit the script will print a sorted list from most frequently to least frequently called function.
The sched_switch.stp script takes two arguments, first argument can be "pid" or "name" to indicate what is being passed as second argument. The script will trace the process based on pid/name and print the scheduler switches happening with the process. If no arguments are passed, it displays all the scheduler switches. This can be used to understand which tasks schedule out the current process being traced, and when it gets scheduled in again.
The script watches the futex syscall on the system. On exit the futex's address, the number of contentions, and the average time for each contention on the futex are printed from lowest pid number to highest.
The script watches just shared futex syscalls on the system. On exit the futex's key, the number of contentions, and the average time for each contention on the futex are printed from lowest pid number to highest.
Suppresses fsync() syscalls from processes identified by stap -c/-x by turning them into presumed-faster fsync() on some dummy or other file descriptor
The ttyspy.stp script uses tty_audit hooks to monitor recent typing activity on the system, printing a scrolling record of recent keystrokes, on a per-tty basis.
The script drops the specified number of packets of the specified protocol. Valid protocols are TCP, UDP, or ALL. If ALL is specified, all incoming packets are dropped. The number of packets to drop can be specified with a positive integer. A value of 0 indicates that packets should be dropped until the user manually exits.
Run the tcp_init_cwnd.stp script in the background to override a kernel's default tcp cwnd value to 10, which has been found to improve latency for web server type workloads. The script prints a count of cwnd value changes when it is stopped.
Blocks ptrace(2) attempts from processes identified by stap -c/-x, as also specifiable from /proc/systemtap/stap_XXX/ control files. Processes may be added or removed from the blocked list.
Overrides default NPTL pthread_create stack size for all new threads created by target processes. Reports one line per process when the related glibc variable __default_stacksize is updated. Moot for glibc versions that support $LIBC_PTHREAD_DEFAULT_STACKSIZE_NP.
The script periodically prints a count of specified events and their related tid's over the course of execution. Numerous configuration options exist to control filtering / reporting, some of which can be modified at runtime. See the script source for more information.
This script prints the size of a type, based on dwarf debuginfo for any kernel or userspace module, or trial-compilation of a given header file name. Types and corresponding locations are provided to the script at runtime via keyboard input. The format of the input is identical to that of sizeof.stp (see script source for more information). Types and locations can be repeatedly given until the process is terminated.
The Linux kernel function smp_call_function causes expensive inter-processor interrupts (IPIs). The scf.stp script tallies the processes and backtraces causing the interprocessor interrupts to identify the cause of the expensive IPI. On exit the script prints the tallies in descending frequency.
The cycle_thief.stp script instruments the scheduler and IRQ handler to determine which processes and interrupts are competing with the specified task for the cpu cycles. This script uses the '-c' or '-x' options to focus on a specific task. The script output the number of times the task migrates between processors, histograms showing the length of time on and off processor, lists of processes running while the task is off the processor, and the interrupts that occurred while the task was running.
alias_suffixes.stp is a demonstration of how alias suffixes in the systemtap language might be used. The script tracks the wall clock time for each invocation of the system calls open, close, read, and write. When the script exists it prints out the minimum, average, and maximum times in microseconds for each system call, followed by a count of times that each syscall was invoked and a histogram showing the distributions of times.
Suppresses fsync() syscalls from processes identified by stap -c/-x by turning them into presumed-faster fsync() on some dummy or other file descriptor
The inodewatch.stp outputs the executable name and process id each time a read or write occurs to the specified inode on the specified major/minor device.
The inodewatch2.stp script outputs the executable name, process id, and attributes each time the attributes are changed on the specified inode on the specified major/minor device.
When a reschedule occurs during an AIO io_submit call, accumulate the traceback in a histogram. When the script exits prints out a sorted list from most common to least common backtrace.
# stap io_submit.stp -T 1
io/ioblktime.stp - Average Time Block IO Requests Spend in Queue
keywords: IO
The ioblktime.stp script tracks the amount of time that each block IO requests spend waiting for completion. The script computes the average waiting time for block IO per device and prints list every 10 seconds. In some cases there can be too many outstanding block IO operations and the script may exceed the default number of MAXMAPENTRIES allowed. In this case the allowed number can be increased with "-DMAXMAPENTRIES=10000" option on the stap command line.
The iodevstats.stp script measures the amount of data successfully read and written by all the executables for each io device on the system. The output is sorted from greatest sum of bytes read and written to a device by an executable to the least. The output contains device major/minor number, the count of operations (reads and writes), the totals and averages for the number of bytes read and written.
The iostat-scsi.stp script provides a breakdown of the number of blks read and written on the machine's various SCSI devices. The script takes one argument which is the number of seconds between reports.
The iostat.stp script measures the amount of data successfully read and written by all the executables on the system. The output is sorted from most greatest sum of bytes read and written by an executable to the least. The output contains the count of operations (opens, reads, and writes), the totals and averages for the number of bytes read and written.
The script watches each open, close, read, and write syscalls on the system. For each file the scripts observes opened it accumulates the amount of wall clock time spent in read and write operations and the number of bytes read and written. When a file is closed the script prints out a pair of lines for the file. Both lines begin with a timestamp in microseconds, the PID number, and the executable name in parentheses. The first line with the "access" keyword lists the file name, the attempted number of bytes for the read and write operations. The second line with the "iotime" keyword list the file name and the number of microseconds accumulated in the read and write syscalls.
# stap iotime.stp -T 1
io/iotop.stp - Periodically Print IO Activity by Process Name
keywords: IO
Every five seconds print out the top ten executables generating I/O traffic during that interval sorted in descending order.
The nfs_func_users.stp script counts the uses of NFS functions in the kernel on a per process bases. The output is sorted from the process with the greatest number of NFS functions called to the least. The output contains the executable name, the process number, and the total number of NFS functions called by the process.
This script prints a line for every kernel vfs_open operation that takes longer than a configurable number of microseconds. Highly contended or remote filesystems are likelier to hit this.
The ttyspy.stp script uses tty_audit hooks to monitor recent typing activity on the system, printing a scrolling record of recent keystrokes, on a per-tty basis.
This measures block I/O latency (storage I/O, ie, disk I/O), and shows the distribution as a histogram. This can be useful to identify the characteristics of I/O latency, beyond the averages shown by iostat(1). For example, to study I/O latency outliers, or multi-modal distributions.
This uses the kernel tracepoint block_rq_insert to read the size of I/O. The output includes the name of the process or thread that was on-CPU when the I/O request was inserted on the issue queue.
# stap bitesize-nd.stp -T 1
lwtools/execsnoop-nd.stp - Trace process exec() with command line argument details (non-debuginfo)
keywords: IO
This can identify if CPU is consumed by short-lived processes, by tracing new process execution. It works by tracing exec() from the fork()->exec() sequence, which means it will not catch new processes that only fork(). It will also show every exec(), including those if a process re-execs.
This dynamically traces two common file system functions: do_sync_read() and do_sync_write(), and reports a histogram distribution of latency. Many, but not all, file systems and workloads use these functions. Tracing their time provides one view of suffered file system latency.
This dynamically traces two common file system functions: do_sync_read() and do_sync_write(), and shows details of each call that is slower than a threshold. Many, but not all, file systems and workloads use these functions. Tracing their time provides one view of suffered file system latency.
# stap fsslower-nd.stp -T 1
lwtools/killsnoop-nd.stp - Trace kill() signals showing process and signal details (non-debuginfo)
keywords: IO
This traces signals system-wide, including those sent by the kill(1) command, and shows various details.
The script monitors the time that threads spend in waiting for IO operations (in "D" state) in the wait_for_completion function. If a thread spends over 10ms, its name and backtrace is printed, and later so is the total delay.
The qemu_io.stp script tallies the number of times each of the IO port on the guest virtual machines is touched by a input or output operation. When the script exits, it prints a count of the number of times each IO port read and written.
The script watches all IPv4 network traffic on the system. The data is output in JSON format and includes the number of packets sent along source IP address / destination IP address pair encountered, and the total number of bytes sent among the pair.
The kvm_service_time.stp script tracks the statistics about the amount of time that the processor left the guest virtual machine for each exit reason (for example fixing up a page table or handling an IO operation). When the script exits it prints out the number of times each exit reason was encountered, the total duration of time it left the guest VM, the minimum time, the average time, and the maximum time in microseconds for that exit reason. On Linux 2.6.38 and newer kernel the script can automatically determine whether it is running on Intel or AMD processors. For older kernels with a kernel.trace("kvm_exit") tracepoint that does not have the $isa parameter you can explicitly state the kvm type with a "-G kvm=intel" or "-G kvm=amd" on the command line.
The qemu_count.stp script tallies the number of times each of the user-space qemu probepoints is encountered. When the script exits, it prints a list of the number of times each user-space qemu probepoint is encountered.
The qemu_io.stp script tallies the number of times each of the IO port on the guest virtual machines is touched by a input or output operation. When the script exits, it prints a count of the number of times each IO port read and written.
Intel processors have hardware mechanisms that will reduce the effective processors speed to avoid exceeding set thermal or power constraints. The cpu_throttle.stp script monitors when CPU throttling occurs and accumulates the amount of time in milliseconds that each processor is throttled. The script makes the information available via procfs in Prometheus readable format.
The overcommit.stp script prints a line each time the kernel refuses a memory allocation request from a process because of /proc/sys/vm/overcommit* limits.
The bkl.stp script can help determine whether the Big Kernel Lock (BKL) is causing serialization on a multiprocessor system due to excessive contention of the BKL. The bkl.stp script takes two arguments. The first one is optional, and used to enable backtraces, and print them once a process has been holding the BKL for a user specified number of nseconds is reached. The second option is compulsory and is the number of processes waiting for the Big Kernel Lock (BKL). When the number of processes waiting for the BKL is reached or exceeded, the script will print a time stamp, the number of processes waiting for the BKL, the holder of the BKL, and the amount of time the BKL was held. If backtraces are enabled, a backtrace will be printed as well.
The bkl_stats.stp script can indicate which processes have excessive waits for the Big Kernel Lock (BKL) and which processes are taking the BKL for long periods of time. The bkl_stats.stp script prints lists of all the processes that require the BKL. Every five seconds two tables are printed out. The first table lists the processes that waited for the BKL followed by the number of times that the process waited, the minimum time of the wait, the average and the maximum time waited. The second table lists has similar information for the time spent in holding the lock for each of the processes.
The script watches the futex syscall on the system. On exit the futex's address, the number of contentions, and the average time for each contention on the futex are printed from lowest pid number to highest.
The script watches just shared futex syscalls on the system. On exit the futex's key, the number of contentions, and the average time for each contention on the futex are printed from lowest pid number to highest.
Tracks pthread-mutex initialization/use and underlying futex operations, to identify (with backtraces/symbol-names) the mutexes suffering most contention. Invoke with "-d SHLIB --ldd", perhaps with -DMAXMAPENTRIES=NNNN for some large NNNN, if the arrays overflow due to heavy activity.
This script prints the size of a type, based on dwarf debuginfo for any kernel or userspace module, or trial-compilation of a given header file name. Types and corresponding locations are provided to the script at runtime via keyboard input. The format of the input is identical to that of sizeof.stp (see script source for more information). Types and locations can be repeatedly given until the process is terminated.
When a huge page is initially used by a process it must be cleared of its original contents to avoid leaking information between processes. On x86_64 machines the huge pages are 2MB in size, 512 times larger than a normal 4KB page. Thus, clearing a huge page can delay program execution by a hundred or more microseconds and maybe noticeable to latency sensitive programs. This script will tally the number of times that each process triggers a huge page clear, the total time in microseconds spent clearing the pages, and the average time in microseconds of the huge page clear.
The kernel may attempt to group many normal sized pages into a single huge page to improve TLB performance. This operation may take a significant amount of time and affect program performance. You can identify if huge page collapse operations are occurring with the systemtap hugepage_collapse.stp script.
The system may attempt to reduce memory use by having two or more difference processes share the same writeable page in memory. However, when one of the processes write to shared page that is marked copy on write (COW) the page must be duplicated and then modified. On x86_64 machines the huge pages are 2MB in size, 512 times larger than a normal 4KB page. Thus, copy on write operation on a huge page can delay program execution by a hundred or more microseconds and maybe noticeable to latency sensitive programs. This script will tally the number of times that each process triggers a huge page copy on write, the total time in microseconds spent copying the pages, and the average time in microseconds of the huge page copy.
Because some portions of the kernel code only work with normal-sized pages the kernel may convert a huge page into a set of normal-sized pages using a split operation. This operation may be relatively expensive and noticeable to latency sensitive programs. You can identify if split operations are occurring with the systemtap hugepage_split.stp script.
The script will watch accesses to a single kernel address and prints a traceback each time the address is accessed. This script needs to be run as root to allow access to the breakpoint hardware.
The script will watch accesses to the starting address of a single kernel symbol and prints a traceback each time the symbol is accessed. This script needs to be run as root to allow access to the breakpoint hardware.
The kmalloc-top perl program runs a small systemtap script to collect stack traces for each call to the kmalloc function and counts the time that each stack trace is observed. When kmalloc-top exits it prints out sorted list. The output can be filtered to print only the first N stack traces (-t), stack traces with a minimum counts (-m), or exclude certain stack traces (-e).
This script reports on the last few free(3) libc calls done by processes (possibly restricted by stap -x/-c), along with a userspace backtrace at those moments.
The mmanonpage.stp script uses the virtual memory tracepoints available in some kernels to track the number of faults, user space frees, page ins, copy on writes and unmaps for anonymous pages. When the script is terminated the counts are printed for each process that allocated pages while the script was running. This script displays the anonymous page statistics for each process that ran while the script is active. It's useful in debugging leaks in the anonymous regions of a process.
The mmfilepage.stp script uses the virtual memory tracepoints available in some kernels to track the number of faults, copy on writes mapping, and unmapping operations for file backed pages. When the script is terminated the counts are printed for each process that allocated pages while the script was running. The mmfilepage.stp script is useful in debugging leaks in the mapped file regions of a process.
The mmreclaim.stp script uses the virtual memory tracepoints available in some kernels to track page reclaim activity that occurred while the script was running. It's useful in debugging performance problems that occur due to page reclamation.
The mmwriteback.stp script uses the virtual memory tracepoints available in some kernels to report all of the file writebacks that occur form kupdate, pdflush and kjournald while the script is running. It's useful in determining where writes are coming from on a supposedly idle system that is experiencing unexpected IO.
The numa_faults.stp script tracks the read and write pages faults for each process. When the script exits it prints out the total read and write pages faults for each process. The script also provide a break down of page faults per node for each process. This script is useful for determining whether the program has good locality (page faults limited to a single node) on a NUMA computer.
The overcommit.stp script prints a line each time the kernel refuses a memory allocation request from a process because of /proc/sys/vm/overcommit* limits.
The pfaults.stp script generates a simple log for each major and minor page fault that occurs on the system. Each line contains a timestamp (in microseconds) when the page fault servicing was completed, the pid of the process, the address of the page fault, the type of access (read or write), the type of fault (major or minor), and the elapsed time for page fault. This log can be examined to determine where the page faults are occurring.
The script will probe all memory slab/slub allocations and collects information about the size of the object (bytes requested) and user-space process in execution. When run over a period of time, it helps to correlate kernel-space memory consumption owing to user-space processes.
The measureinteval.stp script allows quick creation of instrumentation to measure quantity between different probe points, aggregate the measurements, and display the results. The first argument to the script is the quantity to measure such as time in microseconds via "gettimeofday_us()". There could be multiple intervals being measured concurrently and the second argument is the means to match the start probe points listed in the fourth argument with the end probes listed in the fifth argument. The third argument is how to display the aggregated data. The data could be displayed as a sum (@sum), an average (@avg), or a histogram (@hist_log). The fourth and fifth arguments are lists of probes marking the start and end of an interval. An optional sixth argument allows grouping of the intervals. One could use "tid()" for the matching argument to measure the intervals on a per-thread basis, but then use "execname()" for the optional argument to group the measurements by executable name. If there is no sixth argument, all the intervals are aggregated into a single entry.
This script places a set of probes (specified by $1), each of which monitors the state of some context $variable expression (specified by $2). Whenever the value changes, with respect to the active thread, the event is traced.
The whythefail.stp script prints a statement-execution trace for a given function, but only for those runs of the function that ended up with a (configurable) post-return condition.
The ttyspy.stp script uses tty_audit hooks to monitor recent typing activity on the system, printing a scrolling record of recent keystrokes, on a per-tty basis.
Attaches to the kernel audit-log paths (also used by libaudit), and log every record being sent, along with a user-space backtrace of the process that caused it.
The procmod_watcher.stp script monitors calls to fork(), exec(), exit(), init_module(), and delete_module(). Event-specific details are also printed out (e.g. for exec(), the file being exec'ed). This script does not require debuginfo.
The script watches each nanosleep syscall on the system. At the end of each nanosleep syscall the script prints out a line with a timestamp in microseconds, the pid, the executable name in parentheses, the "nanosleep:" key, and the duration of the sleep in microseconds.
This traces socket duration from the accept() syscall to close(), and provides details on the lifespan of these passive connections, showing the distribution as a histogram.
The connect_stat.stp script prints a task's entire ancestry (parent process name/uid/gid) whenever it attempts an outgoing socket connection to a given IP address.
The script drops the specified number of packets of the specified protocol. Valid protocols are TCP, UDP, or ALL. If ALL is specified, all incoming packets are dropped. The number of packets to drop can be specified with a positive integer. A value of 0 indicates that packets should be dropped until the user manually exits.
The script watches all IPv4 network traffic on the system. On exit the script prints a list showing the number of packets sent along source IP address / destination IP address pair encountered, and the total number of bytes sent among the pair. The list is ordered from greatest to least number of packets seen among the source/destination pairs.
The script watches all IPv4 network traffic on the system. The data is output in JSON format and includes the number of packets sent along source IP address / destination IP address pair encountered, and the total number of bytes sent among the pair.
Every five seconds the nettop.stp script prints out a list of processed (PID and command) with the number of packets sent/received and the amount of data sent/received by the process during that interval.
The packet_contents.stp script displays the length of each network packet and its contents in both hexadecimal and ASCII. Systemtap strings are MAXSTRINGLEN in length by default which may not be enough for larger packets. In order to print larger packets, this limit can be increased by passing in the "-DMAXSTRINGLEN=65536" command line option.
The sk_stream-wait_memory.stp prints a time stamp, executable, and pid each time a process blocks due to the send buffer being full. A similar entry is printed each time a process continues because there is room in the buffer.
The script instruments each of the functions in the Linux kernel's net/socket.c file. The script prints out trace data. The first element of a line is time delta in microseconds from the previous entry. This is followed by the command name and the PID. The "->" and "<-" indicates function entry and function exit, respectively. The last element of the line is the function name.
The socktop script periodically prints out a list of the processes with the highest socket activity. Command line options for the script allow filtering to focus on particular types of sockets. The "-h" option lists socktop script's filtering options.
The stp_dump.stp prints out the packet contents. Each block contains the STP protocol ID, version ID, flags, root and bridge MAC addresses, and various times.
The tcp_connections.stp script prints information for each new incoming TCP connection accepted by the computer. The information includes the UID, the command accepting the connection, the PID of the command, the port the connection is on, and the IP address of the originator of the request.
Run the tcp_init_cwnd.stp script in the background to override a kernel's default tcp cwnd value to 10, which has been found to improve latency for web server type workloads. The script prints a count of cwnd value changes when it is stopped.
This scripts traces a given TCP connection based on the filter parameters given by the user. The indexing is done by the 4 tuples local address, remote address, local port, remote port.
The tcpdumplike.stp prints out a line for each TCP & UDP packet received. Each line includes the source and destination IP addresses, the source and destination ports, and flags.
The tcpipstat script collects and displays network statistics related to individual TCP sockets or groups of sockets. The statistics that are collected are simular to that of the command netstat -s, only sorted and grouped by individual sockets.
This script traces outgoing network packets using the netfilter probes (not requiring debuginfo), printing the source thread name/id and destination host:port. It may be filtered with the_dport and the_daddr globals, e.g., to watch only for DNS traffic (port 53), and/or only to the localhost (127.0.0.1).
This script tracks all nfsd server operations by client_ip address, and periodically lists those clients that have made recent requests. It's a way of finding out which nfs clients might be considered still connected.
The nfsd_unlink.stp script lists the ip address and file name each time time a file is being removed or unlinked by the nfsd. This script is run on the nfs server.
The numa_faults.stp script tracks the read and write pages faults for each process. When the script exits it prints out the total read and write pages faults for each process. The script also provide a break down of page faults per node for each process. This script is useful for determining whether the program has good locality (page faults limited to a single node) on a NUMA computer.
The script drops the specified number of packets of the specified protocol. Valid protocols are TCP, UDP, or ALL. If ALL is specified, all incoming packets are dropped. The number of packets to drop can be specified with a positive integer. A value of 0 indicates that packets should be dropped until the user manually exits.
The script periodically prints a count of specified events and their related tid's over the course of execution. Numerous configuration options exist to control filtering / reporting, some of which can be modified at runtime. See the script source for more information.
This script reports on the last few free(3) libc calls done by processes (possibly restricted by stap -x/-c), along with a userspace backtrace at those moments.
The connect_stat.stp script prints a task's entire ancestry (parent process name/uid/gid) whenever it attempts an outgoing socket connection to a given IP address.
The sk_stream-wait_memory.stp prints a time stamp, executable, and pid each time a process blocks due to the send buffer being full. A similar entry is printed each time a process continues because there is room in the buffer.
The cycle_thief.stp script instruments the scheduler and IRQ handler to determine which processes and interrupts are competing with the specified task for the cpu cycles. This script uses the '-c' or '-x' options to focus on a specific task. The script output the number of times the task migrates between processors, histograms showing the length of time on and off processor, lists of processes running while the task is off the processor, and the interrupts that occurred while the task was running.
Prints a periodic tabular report about failing system calls, by process and by syscall failure. The first optional argument specifies the reporting interval (in seconds, default 5); the second optional argument gives a screen height (number of lines in the report, default 20).
The forktracker.stp script prints out a time-stamped entry showing each fork and exec operation on the machine. This can be useful to determine what process is creating a flurry of short-lived processes.
Blocks ptrace(2) attempts from processes identified by stap -c/-x, as also specifiable from /proc/systemtap/stap_XXX/ control files. Processes may be added or removed from the blocked list.
Run pfiles.stp to produce a human-readable summary of all open file descriptors of a given process. Specify the process-id as -x PID for fastest performance.
The procmod_watcher.stp script monitors calls to fork(), exec(), exit(), init_module(), and delete_module(). Event-specific details are also printed out (e.g. for exec(), the file being exec'ed). This script does not require debuginfo.
Every 5 seconds, print out a list of 25 processes that took the most system time with information about the processes. Includes information on processes that may have exited while the script was running. The script contains configuration options listed in the script source.
Run psig.stp to produce a human-readable summary of the signal handling configuration of a given process. Specify the process-id as -x PID for fastest performance.
The pstrace_exec.stp script watches each exec operation. If the exec contains a substring that matches the script's command-line argument, it prints out that process and all of its ancestors.
The pstree.stp script generates a process diagram in DOT form. For instance, it may be useful on a 'make' command to see all the processes that are started.
This script periodically reports a histogram of the latency between a task (thread) being woken up and it actually being dispatched to a CPU: the amount of time it's spent in the runnable queue.
The schedtimes.stp script instruments the scheduler to track the amount of time that each process spends in running, sleeping, queuing, and waiting for io. On exit the script prints out the accumulated time for each state of processes observed. Optionally, this script can be used with the '-c' or '-x' options to focus on a specific PID and its children.
The spawn_seeker.stp script every minute (and on exit) prints out the local time and sorted lists of which processes and executables spawned tasks during the previous minute. This can be useful to determine what process is creating a flurry of short-lived processes. When a process exits its count of tasks created is added to its parent's count to better account for the indirect task creation by children processes. For more detailed examination of task creation consider using forktracker.stp.
The script loosely emulates strace, when applied to individual processes or hierarchies (via -c/-x), or the entire system (without -c/-x). A few output configuration parameters may be set with -G.
Prints a periodic tabular report about the counts of syscall activity of all threads on the system, along with a textual recent-syscall-history for each
The script watches each wait4 syscall on the system. At the end of each wait4 syscall the script prints out a line with a timestamp in microseconds, the pid, the executable name in parentheses, the "wait4:" key, the duration of the wait and the PID that the wait4 was waiting for. If the waited for PID is not specified , it is "-1".
The syscallerrorsbypid.stp script tallies syscall errors for each running process. This information can be useful to whether there are excessive errors for various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscalllatency.stp script accumulates syscall latency for each running process. This information can be useful to whether excessive time is being spent in particular syscalls on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscallsbypid.stp script tallies each syscall for each running process. This information can be useful to determine the activity of various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The iodevstats.stp script measures the amount of data successfully read and written by all the executables for each io device on the system. The output is sorted from greatest sum of bytes read and written to a device by an executable to the least. The output contains device major/minor number, the count of operations (reads and writes), the totals and averages for the number of bytes read and written.
The iostat-scsi.stp script provides a breakdown of the number of blks read and written on the machine's various SCSI devices. The script takes one argument which is the number of seconds between reports.
The iostat.stp script measures the amount of data successfully read and written by all the executables on the system. The output is sorted from most greatest sum of bytes read and written by an executable to the least. The output contains the count of operations (opens, reads, and writes), the totals and averages for the number of bytes read and written.
The script watches each open, close, read, and write syscalls on the system. For each file the scripts observes opened it accumulates the amount of wall clock time spent in read and write operations and the number of bytes read and written. When a file is closed the script prints out a pair of lines for the file. Both lines begin with a timestamp in microseconds, the PID number, and the executable name in parentheses. The first line with the "access" keyword lists the file name, the attempted number of bytes for the read and write operations. The second line with the "iotime" keyword list the file name and the number of microseconds accumulated in the read and write syscalls.
The nfs_func_users.stp script counts the uses of NFS functions in the kernel on a per process bases. The output is sorted from the process with the greatest number of NFS functions called to the least. The output contains the executable name, the process number, and the total number of NFS functions called by the process.
The container_check.stp script monitors the use of linux capablities and optionally forbidden syscalls by a process and its children. On exit the script prints out lists showing the capabilies used by each executable, which syscall used specific capabilites for each executable, a list of forbidden syscalls used, and details on any syscalls that failed during monitoring. This script is designed to help diagnose issues caused by restricted capabilies and syscalls when running an application in a container. If the script warns about skipped probes, the number of active kretprobes may need to be increased with "-DKRETACTIVE=100" option on the command line
The fileline-profile.stp script ends by printing out a sorted list of the top twenty kernel and/or user processes providing file:line information, if available, from the samples addresses gathered over the time period the script is run. Use any of --ldd, --all-modules, -d MODULE, -d /PATH/TO/EXEC to add more symbolic info. To include the symbol name in the output, specify guru mode (-g) and add symbolname="yes" to the stap command.
The fntimes.stp script monitors the execution time history of a given function family (assumed non-recursive). Each time (beyond a warmup interval) is then compared to the historical maximum. If it exceeds a certain threshold (250%), a message is printed.
The functioncallcount.stp script takes one argument, a list of functions to probe. The script will run and count the number of times that each of the functions on the list is called. On exit the script will print a sorted list from most frequently to least frequently called function.
The ioctl systemcall is used to manipulate devices setting or special files. The way that ioctl syscalls are handled depend greatly on the device the special file is associated with. Using strace to monitor the open and ioctl syscalls may not give a good indication of what kernel code is actually handling the ioctl operations. The ioctl_handler.stp script is designed to provide more details. On exit the ioctl_handler.stp script provides a count of the ioctl syscalls for each executable run on the system. If there was some special device driver code used to handle to the ioctl, the output will have a tally of the times the function name and module was called for that executable. The "--all-modules" option should be included on the command line so the script can provide function name information.
The latencytap.stp script collects data on the intervals processes are deactivated (sleeping). The script categorizes the reasons for the sleeps by analyzing the backtraces and displays a sorted list of the top 20 causes from largest total sum time sleeping to smallest. The output is updated every 30 seconds. The script needs to be compiled with the '--all-modules' option to produce reasons for sleeps caused by modules. Optionally, this script can be used with the '-c' or '-x' options to focus on a specific PID.
The linetimes.stp script takes two arguments: where to find the function and the function name. linetimes.stp will instrument each line in the function. It will print out the number of times that the function is called, a table with the average and maximum time each line takes, and control flow information when the script exits.
The periodic.stp script uses the kernel.trace("timer_expire_entry") tracepoint to collect data on period and frequency of the various timers on the system. The script displays a sorted list of the timers observed on the system from most frequent to least frequent. The script needs to be compiled with the '--all-modules' option to produce list the function names. Optionally, this script can be used with a numerical argument to indicate the interval in seconds between printing output.
The pf3.stp script sets up time-based sampling. Every five seconds it prints out a sorted list with the top twenty kernel and/or user functions with samples. Use any of --ldd, --all-modules, -d MODULE, -d /PATH/TO/EXEC to add more symbolic info.
The pf4.stp script sets up time-based sampling. Every five seconds it prints out a sorted list with the top twenty kernel and/or user stack backtraces (on a per-cpu basis). Use any of --ldd, --all-modules, -d MODULE, -d /PATH/TO/EXEC to add more symbolic info.
The sched_switch.stp script takes two arguments, first argument can be "pid" or "name" to indicate what is being passed as second argument. The script will trace the process based on pid/name and print the scheduler switches happening with the process. If no arguments are passed, it displays all the scheduler switches. This can be used to understand which tasks schedule out the current process being traced, and when it gets scheduled in again.
The thread-times.stp script sets up time-based sampling. Every five seconds it prints out a sorted list with the top twenty threads occupying the CPUs, broken down as a percentage of user and kernel time.
The timeout.stp script is based on a blog entry (http://udrepper.livejournal.com/19041.html) mentioning a need for a tool to help developers find applications that are polling. The timeout.stp script monitors systemcall used for polling and records the systemcalls that timed out rather than returned because some action occurred. The script updates the screen once a second with the top twenty processes.
The ucalls.stp script is modeled after the BCC ucalls script (https://github.com/iovisor/bcc/blob/master/tools/lib/ucalls.py) by Sasha Goldshtein. The ucalls.stp script monitors the process indicated by the -x or -c option. When the scripts exits it prints out information about the number of times that each method is invoked for code written in Java, Perl, Php, Python, Ruby, and Tcl. Include the word "syscalls" on the command line to count syscalls invoked by the process. If you want latency information, include "latency" on the command line. Note that the latency option does not work for recursive functions
The also_ran.stp script tallies each time a executable is started or a shared library is loaded for execution. This information can be useful to determine what software is actually being used on the system. The script makes the information available via procfs in Prometheus readable format.
Intel processors have hardware mechanisms that will reduce the effective processors speed to avoid exceeding set thermal or power constraints. The cpu_throttle.stp script monitors when CPU throttling occurs and accumulates the amount of time in milliseconds that each processor is throttled. The script makes the information available via procfs in Prometheus readable format.
The syscallerrorsbypid.stp script tallies syscall errors for each running process. This information can be useful to whether there are excessive errors for various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscalllatency.stp script accumulates syscall latency for each running process. This information can be useful to whether excessive time is being spent in particular syscalls on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscallsbypid.stp script tallies each syscall for each running process. This information can be useful to determine the activity of various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The qemu_count.stp script tallies the number of times each of the user-space qemu probepoints is encountered. When the script exits, it prints a list of the number of times each user-space qemu probepoint is encountered.
The qemu_io.stp script tallies the number of times each of the IO port on the guest virtual machines is touched by a input or output operation. When the script exits, it prints a count of the number of times each IO port read and written.
Uses the regex functionality to detect opened files whose names match a pattern given on the command line. If no command line parameter is given, demonstrate by filtering for files that end with an extension showing them to be an archive.
The chng_cpu.stp script takes an argument which is the executable name of the task it should monitor. Each time a task with that executable name is found running on a different processor, the script prints out the thread id (tid), the executable name, the processor now running the task, the thread state, and a backtrace showing the kernel functions that triggered the running of the task on the processor.
The cycle_thief.stp script instruments the scheduler and IRQ handler to determine which processes and interrupts are competing with the specified task for the cpu cycles. This script uses the '-c' or '-x' options to focus on a specific task. The script output the number of times the task migrates between processors, histograms showing the length of time on and off processor, lists of processes running while the task is off the processor, and the interrupts that occurred while the task was running.
The forktracker.stp script prints out a time-stamped entry showing each fork and exec operation on the machine. This can be useful to determine what process is creating a flurry of short-lived processes.
The migrate.stp script takes an argument which is the executable name of the task it should monitor. Each time a task with that executable name migrates between processors an entry is printed with the process id (pid), the executable name, the processor off loading the task, and the process taking the task. Note that the task may or may not be executing at the time of the migration.
Every 5 seconds, print out a list of 25 processes that took the most system time with information about the processes. Includes information on processes that may have exited while the script was running. The script contains configuration options listed in the script source.
This script periodically reports a histogram of the latency between a task (thread) being woken up and it actually being dispatched to a CPU: the amount of time it's spent in the runnable queue.
The schedtimes.stp script instruments the scheduler to track the amount of time that each process spends in running, sleeping, queuing, and waiting for io. On exit the script prints out the accumulated time for each state of processes observed. Optionally, this script can be used with the '-c' or '-x' options to focus on a specific PID and its children.
The script monitors the time that threads spend in waiting for IO operations (in "D" state) in the wait_for_completion function. If a thread spends over 10ms, its name and backtrace is printed, and later so is the total delay.
The spawn_seeker.stp script every minute (and on exit) prints out the local time and sorted lists of which processes and executables spawned tasks during the previous minute. This can be useful to determine what process is creating a flurry of short-lived processes. When a process exits its count of tasks created is added to its parent's count to better account for the indirect task creation by children processes. For more detailed examination of task creation consider using forktracker.stp.
The iostat-scsi.stp script provides a breakdown of the number of blks read and written on the machine's various SCSI devices. The script takes one argument which is the number of seconds between reports.
Attaches to the kernel audit-log paths (also used by libaudit), and log every record being sent, along with a user-space backtrace of the process that caused it.
Blocks ptrace(2) attempts from processes identified by stap -c/-x, as also specifiable from /proc/systemtap/stap_XXX/ control files. Processes may be added or removed from the blocked list.
Run psig.stp to produce a human-readable summary of the signal handling configuration of a given process. Specify the process-id as -x PID for fastest performance.
The script traces any SIGKILL signals. When that SIGKILL signal is sent to a process, the script prints out the signal name, the destination executable and process ID, the executable name and user ID that sents the signal.
The script watches for a particular signal sent to a specific process. When that signal is sent to the specified process, the script prints out the PID and executable of the process sending the signal, the PID and executable name of the process receiving the signal, and the signal number and name.
A basic "Floating Point" program implemented in SystemTap script. It extracts floating point from benchmark-sdt section in main.cxx and takes one pseudo floating point, printing out results of various simple floating point operations.
Suppresses fsync() syscalls from processes identified by stap -c/-x by turning them into presumed-faster fsync() on some dummy or other file descriptor
This script prints a line for every kernel vfs_open operation that takes longer than a configurable number of microseconds. Highly contended or remote filesystems are likelier to hit this.
This script traces outgoing network packets using the netfilter probes (not requiring debuginfo), printing the source thread name/id and destination host:port. It may be filtered with the_dport and the_daddr globals, e.g., to watch only for DNS traffic (port 53), and/or only to the localhost (127.0.0.1).
Prints the life cycle of all sockets associated with a process. This includes bytes and timing. The timing information that is tracked includes event completion relative to the start of said event and the end of the previous event. Currently tracks read, write, recv, send, connect and close.
This traces socket duration from the accept() syscall to close(), and provides details on the lifespan of these passive connections, showing the distribution as a histogram.
The connect_stat.stp script prints a task's entire ancestry (parent process name/uid/gid) whenever it attempts an outgoing socket connection to a given IP address.
The script instruments each of the functions in the Linux kernel's net/socket.c file. The script prints out trace data. The first element of a line is time delta in microseconds from the previous entry. This is followed by the command name and the PID. The "->" and "<-" indicates function entry and function exit, respectively. The last element of the line is the function name.
The socktop script periodically prints out a list of the processes with the highest socket activity. Command line options for the script allow filtering to focus on particular types of sockets. The "-h" option lists socktop script's filtering options.
The tcp_connections.stp script prints information for each new incoming TCP connection accepted by the computer. The information includes the UID, the command accepting the connection, the PID of the command, the port the connection is on, and the IP address of the originator of the request.
Run the tcp_init_cwnd.stp script in the background to override a kernel's default tcp cwnd value to 10, which has been found to improve latency for web server type workloads. The script prints a count of cwnd value changes when it is stopped.
The whythefail.stp script prints a statement-execution trace for a given function, but only for those runs of the function that ended up with a (configurable) post-return condition.
alias_suffixes.stp is a demonstration of how alias suffixes in the systemtap language might be used. The script tracks the wall clock time for each invocation of the system calls open, close, read, and write. When the script exists it prints out the minimum, average, and maximum times in microseconds for each system call, followed by a count of times that each syscall was invoked and a histogram showing the distributions of times.
The script periodically prints a count of specified events and their related tid's over the course of execution. Numerous configuration options exist to control filtering / reporting, some of which can be modified at runtime. See the script source for more information.
The func_time_stats.stp script tracks the wall clock time for each invocation of the function probe listed as the first command line argument. When the script exits it prints out the minimum, average, and maximum times in microseconds followed by a count of times that the function was called and a histogram showing the distributions of times.
This script prints the size of a type, based on dwarf debuginfo for any kernel or userspace module, or trial-compilation of a given header file name. Types and corresponding locations are provided to the script at runtime via keyboard input. The format of the input is identical to that of sizeof.stp (see script source for more information). Types and locations can be repeatedly given until the process is terminated.
The script will probe all memory slab/slub allocations and collects information about the size of the object (bytes requested) and user-space process in execution. When run over a period of time, it helps to correlate kernel-space memory consumption owing to user-space processes.
This script tracks all nfsd server operations by client_ip address, and periodically lists those clients that have made recent requests. It's a way of finding out which nfs clients might be considered still connected.
The tcpipstat script collects and displays network statistics related to individual TCP sockets or groups of sockets. The statistics that are collected are simular to that of the command netstat -s, only sorted and grouped by individual sockets.
Prints the life cycle of all sockets associated with a process. This includes bytes and timing. The timing information that is tracked includes event completion relative to the start of said event and the end of the previous event. Currently tracks read, write, recv, send, connect and close.
The script watches each open, close, read, and write syscalls on the system. For each file the scripts observes opened it accumulates the amount of wall clock time spent in read and write operations and the number of bytes read and written. When a file is closed the script prints out a pair of lines for the file. Both lines begin with a timestamp in microseconds, the PID number, and the executable name in parentheses. The first line with the "access" keyword lists the file name, the attempted number of bytes for the read and write operations. The second line with the "iotime" keyword list the file name and the number of microseconds accumulated in the read and write syscalls.
Prints a periodic tabular report about failing system calls, by process and by syscall failure. The first optional argument specifies the reporting interval (in seconds, default 5); the second optional argument gives a screen height (number of lines in the report, default 20).
The script watches the futex syscall on the system. On exit the futex's address, the number of contentions, and the average time for each contention on the futex are printed from lowest pid number to highest.
The script watches just shared futex syscalls on the system. On exit the futex's key, the number of contentions, and the average time for each contention on the futex are printed from lowest pid number to highest.
The procmod_watcher.stp script monitors calls to fork(), exec(), exit(), init_module(), and delete_module(). Event-specific details are also printed out (e.g. for exec(), the file being exec'ed). This script does not require debuginfo.
The script watches each nanosleep syscall on the system. At the end of each nanosleep syscall the script prints out a line with a timestamp in microseconds, the pid, the executable name in parentheses, the "nanosleep:" key, and the duration of the sleep in microseconds.
The script loosely emulates strace, when applied to individual processes or hierarchies (via -c/-x), or the entire system (without -c/-x). A few output configuration parameters may be set with -G.
The script watches all syscall on the system. On exit the script prints a list showing the number of systemcalls executed by each PID ordered from greatest to least number of syscalls.
The script watches all syscall on the system. On exit the script prints a list showing the number of systemcalls executed by each executable ordered from greatest to least number of syscalls.
Prints a periodic tabular report about the counts of syscall activity of all threads on the system, along with a textual recent-syscall-history for each
The script watches each wait4 syscall on the system. At the end of each wait4 syscall the script prints out a line with a timestamp in microseconds, the pid, the executable name in parentheses, the "wait4:" key, the duration of the wait and the PID that the wait4 was waiting for. If the waited for PID is not specified , it is "-1".
The container_check.stp script monitors the use of linux capablities and optionally forbidden syscalls by a process and its children. On exit the script prints out lists showing the capabilies used by each executable, which syscall used specific capabilites for each executable, a list of forbidden syscalls used, and details on any syscalls that failed during monitoring. This script is designed to help diagnose issues caused by restricted capabilies and syscalls when running an application in a container. If the script warns about skipped probes, the number of active kretprobes may need to be increased with "-DKRETACTIVE=100" option on the command line
The syscallerrorsbypid.stp script tallies syscall errors for each running process. This information can be useful to whether there are excessive errors for various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscalllatency.stp script accumulates syscall latency for each running process. This information can be useful to whether excessive time is being spent in particular syscalls on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscallsbypid.stp script tallies each syscall for each running process. This information can be useful to determine the activity of various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The ucalls.stp script is modeled after the BCC ucalls script (https://github.com/iovisor/bcc/blob/master/tools/lib/ucalls.py) by Sasha Goldshtein. The ucalls.stp script monitors the process indicated by the -x or -c option. When the scripts exits it prints out information about the number of times that each method is invoked for code written in Java, Perl, Php, Python, Ruby, and Tcl. Include the word "syscalls" on the command line to count syscalls invoked by the process. If you want latency information, include "latency" on the command line. Note that the latency option does not work for recursive functions
The sk_stream-wait_memory.stp prints a time stamp, executable, and pid each time a process blocks due to the send buffer being full. A similar entry is printed each time a process continues because there is room in the buffer.
The tcp_connections.stp script prints information for each new incoming TCP connection accepted by the computer. The information includes the UID, the command accepting the connection, the PID of the command, the port the connection is on, and the IP address of the originator of the request.
Run the tcp_init_cwnd.stp script in the background to override a kernel's default tcp cwnd value to 10, which has been found to improve latency for web server type workloads. The script prints a count of cwnd value changes when it is stopped.
The script periodically prints a count of specified events and their related tid's over the course of execution. Numerous configuration options exist to control filtering / reporting, some of which can be modified at runtime. See the script source for more information.
Overrides default NPTL pthread_create stack size for all new threads created by target processes. Reports one line per process when the related glibc variable __default_stacksize is updated. Moot for glibc versions that support $LIBC_PTHREAD_DEFAULT_STACKSIZE_NP.
The stopwatch.stp script illustrates how to use multiple stopwatches record how much wallclock time a process spends in kernel- and user-space. On exit the script prints out the time in seconds, milliseconds, microseconds, and nanoseconds. Note that this output of this script is not directly comparable to the time command because time records the time that the process is actually active in kernel- and user-space.
The cycle_thief.stp script instruments the scheduler and IRQ handler to determine which processes and interrupts are competing with the specified task for the cpu cycles. This script uses the '-c' or '-x' options to focus on a specific task. The script output the number of times the task migrates between processors, histograms showing the length of time on and off processor, lists of processes running while the task is off the processor, and the interrupts that occurred while the task was running.
This script periodically reports a histogram of the latency between a task (thread) being woken up and it actually being dispatched to a CPU: the amount of time it's spent in the runnable queue.
The schedtimes.stp script instruments the scheduler to track the amount of time that each process spends in running, sleeping, queuing, and waiting for io. On exit the script prints out the accumulated time for each state of processes observed. Optionally, this script can be used with the '-c' or '-x' options to focus on a specific PID and its children.
The gmalloc_watch.stp script from Colin Walters' blog (https://blog.verbum.org/2011/03/19/analyzing-memory-use-with-systemtap/) traces the allocation of glib2 memory using the markers in glib2.
The libguestfs_log.stp script prints a log of when various libgueststartup steps are encountered. The first column is microseconds since the script started. The second column is the time elapsed in microseconds between the previous and current events and the third column is the event name.
SystemTap has multiple passes to convert the text of a SystemTap script into instrumentation that actually collect data on the system. The stap_time.stp script uses the markers in SystemTap to note the amount of time that each of the passes requires. When SystemTap completes pass 4 (compiling the instrumentation into a kernel module) the script print out the script name followed by the amount of time in milliseconds required for Pass 0 (command line option parsing), Pass 1 (script parsing), Pass 2 (elaboration), Pass 3 (code generation), and Pass 4 (module compilation). The resulting data can be analyzed to determine if there are issues with the amount of time that systemtap takes to generate instrumentation.
Print a timed per-thread microsecond-timed callgraph, complete with pretty-printed function parameters and return values. The first parameter names the function probe points to trace. The optional second parameter names the probe points for trigger functions, which acts to enable tracing for only those functions that occur while the current thread is nested within the trigger.
Print a timed per-thread microsecond-timed callgraph, complete with function parameters and return values. The first parameter names the function probe points to trace. The optional second parameter names the probe points for trigger functions, which acts to enable tracing for only those functions that occur while the current thread is nested within the trigger.
The whythefail.stp script prints a statement-execution trace for a given function, but only for those runs of the function that ended up with a (configurable) post-return condition.
This scripts traces a given TCP connection based on the filter parameters given by the user. The indexing is done by the 4 tuples local address, remote address, local port, remote port.
This script traces outgoing network packets using the netfilter probes (not requiring debuginfo), printing the source thread name/id and destination host:port. It may be filtered with the_dport and the_daddr globals, e.g., to watch only for DNS traffic (port 53), and/or only to the localhost (127.0.0.1).
The cycle_thief.stp script instruments the scheduler and IRQ handler to determine which processes and interrupts are competing with the specified task for the cpu cycles. This script uses the '-c' or '-x' options to focus on a specific task. The script output the number of times the task migrates between processors, histograms showing the length of time on and off processor, lists of processes running while the task is off the processor, and the interrupts that occurred while the task was running.
The procmod_watcher.stp script monitors calls to fork(), exec(), exit(), init_module(), and delete_module(). Event-specific details are also printed out (e.g. for exec(), the file being exec'ed). This script does not require debuginfo.
This script periodically reports a histogram of the latency between a task (thread) being woken up and it actually being dispatched to a CPU: the amount of time it's spent in the runnable queue.
The schedtimes.stp script instruments the scheduler to track the amount of time that each process spends in running, sleeping, queuing, and waiting for io. On exit the script prints out the accumulated time for each state of processes observed. Optionally, this script can be used with the '-c' or '-x' options to focus on a specific PID and its children.
The syscallerrorsbypid.stp script tallies syscall errors for each running process. This information can be useful to whether there are excessive errors for various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscalllatency.stp script accumulates syscall latency for each running process. This information can be useful to whether excessive time is being spent in particular syscalls on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall error counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The syscallsbypid.stp script tallies each syscall for each running process. This information can be useful to determine the activity of various processes on the system. The script makes the information available via procfs in Prometheus readable format. When a process exits its data will be eliminated from the prometheus output. To avoid exceeding the storage limitations of SystemTap older entries maybe overwritten by newer entries. This can lead to some active process syscall counts disappearing and/or later reappearing with a lower value. Also note that the script does not properly name syscalls for 32-bit applications running on 64-bit machines.
The script watches all IPv4 network traffic on the system. On exit the script prints a list showing the number of packets sent along source IP address / destination IP address pair encountered, and the total number of bytes sent among the pair. The list is ordered from greatest to least number of packets seen among the source/destination pairs.
The script watches all IPv4 network traffic on the system. The data is output in JSON format and includes the number of packets sent along source IP address / destination IP address pair encountered, and the total number of bytes sent among the pair.
Every five seconds the nettop.stp script prints out a list of processed (PID and command) with the number of packets sent/received and the amount of data sent/received by the process during that interval.
The packet_contents.stp script displays the length of each network packet and its contents in both hexadecimal and ASCII. Systemtap strings are MAXSTRINGLEN in length by default which may not be enough for larger packets. In order to print larger packets, this limit can be increased by passing in the "-DMAXSTRINGLEN=65536" command line option.
The stp_dump.stp prints out the packet contents. Each block contains the STP protocol ID, version ID, flags, root and bridge MAC addresses, and various times.
The tcpdumplike.stp prints out a line for each TCP & UDP packet received. Each line includes the source and destination IP addresses, the source and destination ports, and flags.
This script traces outgoing network packets using the netfilter probes (not requiring debuginfo), printing the source thread name/id and destination host:port. It may be filtered with the_dport and the_daddr globals, e.g., to watch only for DNS traffic (port 53), and/or only to the localhost (127.0.0.1).
The ttyspy.stp script uses tty_audit hooks to monitor recent typing activity on the system, printing a scrolling record of recent keystrokes, on a per-tty basis.
The kvm_service_time.stp script tracks the statistics about the amount of time that the processor left the guest virtual machine for each exit reason (for example fixing up a page table or handling an IO operation). When the script exits it prints out the number of times each exit reason was encountered, the total duration of time it left the guest VM, the minimum time, the average time, and the maximum time in microseconds for that exit reason. On Linux 2.6.38 and newer kernel the script can automatically determine whether it is running on Intel or AMD processors. For older kernels with a kernel.trace("kvm_exit") tracepoint that does not have the $isa parameter you can explicitly state the kvm type with a "-G kvm=intel" or "-G kvm=amd" on the command line.
The qemu_count.stp script tallies the number of times each of the user-space qemu probepoints is encountered. When the script exits, it prints a list of the number of times each user-space qemu probepoint is encountered.
The qemu_io.stp script tallies the number of times each of the IO port on the guest virtual machines is touched by a input or output operation. When the script exits, it prints a count of the number of times each IO port read and written.
The watchdog.stp script provides a watchdog timer mechanism for arbitrary events. The script takes three arguments: the events to start watchdog timer, the event to stop the watchdog timer, and the time in millseconds for the watchdog. If the watchdog timer is exceed, the script will trigger a stack backtrace of the user-process that timed out using pstack. This script can be used to diagnose what the userspace application is doing when a slower than expected operation occurs.
The script will watch accesses to a single kernel address and prints a traceback each time the address is accessed. This script needs to be run as root to allow access to the breakpoint hardware.
The script will watch accesses to the starting address of a single kernel symbol and prints a traceback each time the symbol is accessed. This script needs to be run as root to allow access to the breakpoint hardware.
