DWATCH_HANG_MS
Minimum sleep duration, in milliseconds, displayed by the hang profiles (default 1000; 0 to show everything).
Watch processes as they trigger a particular dtrace probe
dwatch [-1defFmnPqRvVwxy]
[-B
num]
[-E
code]
[-g
group]
[-j
jail]
[-k
name]
[-K
num]
[-N
count]
[-o
file]
[-O
cmd]
[-p
pid]
[-r
regex]
[-t
test]
[-T
time]
[-u
user]
[-X
profile]
[-z
regex]
[--]
[probe
[, ...]]
[args ...]
dwatch -l
[-fmnPqy]
[-r
regex]
[probe ...]
dwatch -Q
[-1qy]
[-r
regex]
The dwatch utility uses dtrace(1) to display process info when a given DTrace probe point is triggered. Only the root user or users with either mdo(1) or sudo(8) (ports/security/sudo) access can run this command.
dwatch automates the process of generating DTrace scripts to coalesce trace output by date/time, process info, and [optionally] probe-specific data.
Output format without options is:
date/time uid.gid execname[pid]: psargs
For example, the command ‘dwatch BEGIN’ produces:
INFO Watching 'dtrace:::BEGIN' ...
2017 May 29 08:23:20 0.0 dtrace[60671]: dtrace -s /dev/stdin
The -F option causes dwatch to instead coalesce trace output by date/time, process info, and probe traversal.
Output format with the ‘-F’ option is:
date/time uid.gid execname[pid]: {->,<-, |} prov:mod:func:name ...
For example, the command ‘dwatch -F BEGIN’ produces:
INFO Watching 'dtrace:::BEGIN' ...
2017 May 29 21:34:41 0.0 dtrace[86593]: | dtrace:::BEGIN ...
The -R option causes dwatch to display a process tree containing the parent, grandparent, and ancestor process info.
Output format with the ‘-R’ option is:
date/time uid0.gid0 execname[pid0]: psargs0
-+= pid3 uid3.gid3 psargs3
\\-+= pid2 uid2.gid2 psargs2
\\-+= pid1 uid1.gid1 psargs1
\\-+= pid0 uid0.guid0 psargs0
For example, the command ‘dwatch -R BEGIN’ produces:
INFO Watching 'dtrace:::BEGIN' ...
2017 May 29 21:38:54 0.0 dtrace[86899]: dtrace -s /dev/stdin
-+= 86855 604.604 -bash
\\-+= 86857 604.604 /bin/sh /usr/sbin/dwatch -R BEGIN
\\-+= 86897 0.0 sudo dtrace -s /dev/stdin
\\-+= 86899 0.0 dtrace -s /dev/stdin
Of particular interest is the ability to filter using regular expressions. The ‘-g group’, ‘-p pid’, ‘-r regex’, ‘-u user’, and ‘-z regex’ options can be combined with ‘-R’ to match on parent process criteria as well as current process info.
In contrast, the ‘-j jail’, and ‘-k name’ options apply only to the current process even if ‘-R’ is given.
The ‘-E code’ option gives the ability to customize probe-specific data. For example, the command:
dwatch -E 'printf("%s", copyinstr(arg0))' chdir
displays the path argument sent to chdir(2) calls.
Profiles can be written for more complex routines and/or convenience. To list available profiles use the ‘-Q’ option. Use the ‘-X profile’ option to use a particular profile.
For example, the command ‘dwatch -X kill’ displays arguments sent to kill(2).
If a probe argument does not contain colon (":") and none of ‘-P’, ‘-m’, ‘-f’, or ‘-n’ are given, the probe argument is intelligently mapped to its most-likely value. Use ‘dwatch -l name’ to see what probes will match a given name.
Multiple probes must be given as a single (quoted) argument, separated by comma and/or whitespace. Any/all arguments following said probes will be passed to dtrace(1) unmodified.
-1
Print one line per process/profile (Default; disables ‘-R’).
-B num
Maximum number of arguments to display (Default 64).
-d
Debug. Send dtrace(1) script to stdout instead of executing.
-e
Exit after compiling request but prior to enabling probes.
-E code
DTrace code for event details. If `-', read from stdin. This allows customization of what is printed after date/time and process info. By default, the name and arguments of the program triggering the probe are shown. Can be specified multiple times.
-f
Enable probes matching the specified function names.
-F
Coalesce trace output by probe.
-g group
Group filter. Only show processes matching group name/gid. This can be an awk(1) regular expression to match a numerical gid.
-j jail
Jail filter. Only show processes matching jail name/jid.
-k name
Only show processes matching name. Can also be of the format ‘name*’ to indicate “begins with”, ‘*name’ to indicate “ends with”, or ‘*name*’ to indicate “contains”. Can be specified multiple times.
-K num
Maximum directory depth to display (Default 64).
-l
List available probes on standard output and exit.
-m
Enable probes matching the specified module names.
-X profile
Load profile from DWATCH_PROFILES_PATH.
-n
Enable probes matching the specified probe names.
-N count
Exit after count matching entries (Default 0 for disabled).
-o file
Set output file. If ‘-’, the path ‘/dev/stdout’ is used.
-O cmd
Execute cmd for each event. This can be any valid sh(1) command. The environment variables ‘$TAG’ and ‘$DETAILS’ are set for the given cmd.
-p pid
Process id filter. Only show processes with matching pid. This can be an awk(1) regular expression.
-P
Enable probe matching the specified provider name.
-q
Quiet. Hide informational messages and all dtrace(1) errors.
-Q
List available profiles in DWATCH_PROFILES_PATH and exit.
-r regex
Filter. Only show blocks matching awk(1) regular expression.
-R
Show parent, grandparent, and ancestor of process.
-t test
Test clause (predicate) to limit events (Default none). Can be specified multiple times.
-T time
Timeout. The format is ‘#[smhd]’ or just ‘#’ for seconds.
-u user
User filter. Only show processes matching user name/uid. This can be an awk(1) regular expression to match a numerical UID.
-v
Verbose. Show all errors from dtrace(1).
-V
Report dwatch version on standard output and exit.
-w
Permit destructive actions (copyout*, stop, panic, etc.).
-x
Trace. Print ‘<probe-id>’ when a probe is triggered.
-y
Always treat stdout as console (enable colors/columns/etc.).
-z regex
Only show processes matching awk(1) regular expression.
Profiles customize the data printed during events. Profiles are loaded from a colon-separated list of directories in DWATCH_PROFILES_PATH. This is an incomplete list of profiles with basic descriptions:
Print mode and path from chmod(2), lchmod(2), fchmodat(2)
Print coredump-worthy signal deliveries with a verdict of whether a core will actually be written
Maintain a running catalog of coredump-worthy signals by process and signal
Maintain a running catalog of net bytes and allocation balance by malloc type (leak suspects rise)
Print non-zero errno results from system calls
Print system calls returning the given errno, by symbolic name or number (e.g., errno-ENOENT, errno-ENOTCAPABLE, errno-13)
Print threads that were asleep longer than a threshold, as they wake, naming the sleeper and the waker
Maintain a running catalog of long sleeps by process
Print disk I/O details provided by dtrace_io(4)
Print disk I/O requests exceeding a latency threshold
Print IPv4 and IPv6 details provided by dtrace_ip(4)
Print jail management activity from jail(2), jail_set(2), jail_get(2), jail_attach(2), jail_remove(2)
Print signal and pid from kill(2)
Print kernel lock contention details provided by dtrace_lockstat(4)
Print kernel lock spinning details provided by dtrace_lockstat(4)
Print network stack statistics counters as they increment (requires a kernel built with options KDTRACE_MIB_SDT)
Print pathname resolution details provided by the vfs:namei:lookup() probes
Print failed pathname resolutions (also namei-enoent for ENOENT only)
Print requested time from nanosleep(2)
Print privilege requests denied by priv_check(9) (also priv-ok for those granted)
Print process execution details provided by dtrace_proc(4)
Print process signal details provided by dtrace_proc(4)
Print only signals whose default disposition is to terminate the receiving process
Print CPU scheduling details provided by dtrace_sched(4)
Print threads that waited longer than a threshold between becoming runnable and running
Print filesystem-related syscalls exceeding a latency threshold
Print any syscall exceeding a latency threshold
Print TCP address/port details provided by dtrace_tcp(4)
Print TCP I/O details provided by dtrace_tcp(4)
Print TCP retransmissions, retransmit timeouts, and connections dropped by retransmit exhaustion (requires a kernel built with options KDTRACE_MIB_SDT)
Print UDP I/O details provided by dtrace_udp(4)
Print filesystem paths being created by VOP_CREATE(9)
Print filesystem paths being looked-up by VOP_LOOKUP(9)
Print directory paths being created by VOP_MKDIR(9)
Print device node paths being created by VOP_MKNOD(9)
Print directory paths being read by VOP_READDIR(9)
Print filesystem paths being removed by VOP_REMOVE(9)
Print filesystem paths being renamed by VOP_RENAME(9)
Print directory paths being removed by VOP_RMDIR(9)
Print symlink paths being created by VOP_SYMLINK(9)
These environment variables affect the execution of dwatch:
DWATCH_HANG_MS
Minimum sleep duration, in milliseconds, displayed by the hang profiles (default 1000; 0 to show everything).
DWATCH_IO_MS
Minimum disk I/O request latency, in milliseconds, displayed by the io-slow profile (default 100; 0 to show everything).
DWATCH_LOCK_MS
Minimum lock holdoff duration, in milliseconds, displayed by the lock profiles (default 1; 0 to show everything).
DWATCH_MALLOC_MIN
Minimum allocation size, in bytes, displayed by the dtmalloc profiles (default 65536; 0 to show everything).
DWATCH_PROFILES_PATH
If DWATCH_PROFILES_PATH is set, dwatch searches for profiles in the colon-separated list of directories in that variable instead of the default ‘/usr/libexec/dwatch:/usr/local/libexec/dwatch’. If set to NULL, profiles are not loaded.
DWATCH_SCHED_MS
Minimum run-queue wait, in milliseconds, displayed by the sched-latency profile (default 10; 0 to show everything).
DWATCH_SLOW_MS
Minimum syscall latency, in milliseconds, displayed by the slow profiles (default 100).
The dwatch utility exits 0 on success, and >0 if an error occurs.
Watch processes entering system CPU scheduler.
dwatch on-cpu
List available profiles, one line per profile.
dwatch -1 -Q
Do not execute dtrace(1) but display script on stdout and exit.
dwatch -d fsync
Compile and test but do not execute code generated with given probe.
dwatch -e test_probe
Print argument one being passed to each call of zfs_sync().
dwatch -E 'printf("%i", arg1)' zfs_sync
Watch all functions named ‘read’.
dwatch -f read
Watch all probe traversal.
dwatch -F :
Watch syscall probe traversal.
dwatch -F syscall
Display only processes belonging to wheel super-group.
dwatch -g wheel execve
Display only processes belonging to groups ‘daemon’ or ‘nobody’.
dwatch -g '1|65534' execve
Ignore jails, displaying only base system processes.
dwatch -j 0 execve
Display only processes running inside the jail named ‘myjail’.
dwatch -j myjail execve
Watch syscall traversal by ruby processes.
dwatch -k 'ruby*' -F syscall
Watch syscall traversal by processes containing ‘daemon’ in their name.
dwatch -k '*daemon*' -F syscall
Watch signals being passed to kill(2).
dwatch -X kill
Watch signals being passed between bash(1) (ports/shells/bash) and vi(1).
dwatch -k bash -k vi -X kill
Display a list of unique functions available.
dwatch -l -f
List available probes for functions ending in ‘read’.
dwatch -l -f '*read'
List available probes ending in “read”.
dwatch -l -r 'read$'
Display a list of unique providers.
dwatch -l -P
Watch paths being removed by VOP_REMOVE(9).
dwatch -X vop_remove
Watch the name ‘read’ instead of the function ‘read’. The dwatch selection algorithm will commonly favor the function named ‘read’ when not given a type (using ‘-P’, ‘-m’, ‘-f’, or ‘-n’) because there are more probes matching the function named ‘read’ than probes matching ‘read’ for any other type.
dwatch -n read
Display the first process to call kill(2) and then exit.
dwatch -N 1 kill
Watch processes forked by pid 1234.
dwatch -p 1234 execve
Watch processes forked by either pid 1234 or pid 5678.
dwatch -p '1234|5678' execve
Watch the provider ‘random’ instead of the function ‘random’. The dwatch selection algorithm will commonly favor the function named ‘random’ when not given a type (using ‘-P’, ‘-m’, ‘-f’, or ‘-n’) because there are more probes matching the function named ‘random’ than probes matching the provider named ‘random’.
dwatch -P random
Display available profiles matching ‘vop’.
dwatch -Q -r vop
Watch VOP_LOOKUP(9) paths containing ‘/lib/’.
dwatch -r /lib/ -X vop_lookup
Show process tree for each command as it is executed.
dwatch -R execve
Watch processes forked by pid 1234 or children thereof.
dwatch -R -p 1234 execve
Display processes calling write(2) with “nbytes” less than 10.
dwatch -t 'arg2<10' -E 'printf("%d",arg2)' write
Display write(2) buffer when “execname” is not ‘dtrace’ and “nbytes” is less than 10.
dwatch -X write -t 'execname != "dtrace" && this->nbytes < 10'
Watch ‘statfs’ for 5 minutes and exit.
dwatch -T 5m statfs
Display only processes belonging to the root super-user.
dwatch -u root execve
Display only processes belonging to users ‘daemon’ or ‘nobody’.
dwatch -u '1|65534' execve
Print version and exit.
dwatch -V
View the first 100 scheduler preemptions.
dwatch -y -N 100 preempt | less -R
Display processes matching either “mkdir” or “rmdir”.
dwatch -z '(mk|rm)dir' execve
Run a command and watch network activity only while that command runs.
dwatch -X tcp -- -c "nc -zvw10 google.com 22"
Watch open(2) and openat(2) calls only while pid 1234 is active.
dwatch -X open -- -p 1234
Watch probe traversal for a given command. Note that “-c true” is passed to dtrace(1) since it appears after the dwatch probe argument.
dwatch -F 'pid$target:::entry' -c true
23 releases since FreeBSD 11.2-RELEASE; the text changed 12 times.