Red Hat ENTERPRISE LINUX 5.5 - TECHNICAL NOTES Manual page 270

Chapter 1. Package Updates
• Using the task_pid() function in a SystemTap script while the kernel-debuginfo package was
not installed could incorrectly result in a semantic error. This update applies an upstream patch to
the task.stp tapset, which fixes the issue.
• This release also includes sample scripts for probing kernel tracepoints, namely:
• memory/mmanonpage.stp
• memory/mmfilepage.stp
• memory/mmreclaim.stp
• memory/mmwriteback.stp
• network/dropwatch.stp
• process/schedtimes.stp
2073
(BZ#497894 )
• Killing concurrent staprun processes could result in a kernel panic. This was because runtime/
procfs.c only checked if /proc/systemtap was being used before deleting it, resulting in a race
condition that made it possible for /proc/systemtap to be deleted while a module was still loaded
inside. This update fixes the race condition by adding instructions to lock the transport directory and
check for files under /proc/systemtap before deleting it.
• The tcpmib.stp and ipmib.stp tapsets have been updated to provide per-socket network
statistics and dynamic TCP connection tracing.
• SystemTap now supports signal-based log file switching. As such, the on-file flight recorder can
easily backup its latest logs on-the-fly.
• Using SIGKILL on the stap process will not terminate its child process, stapio. As such,
users may not be aware that a SystemTap module is still probing the system, which will result in
performance degradation. This update adds a note in man stap warning users of this behavior.
2077
(BZ#523356 )
• Previous updates to SystemTap changed the order of parameters in the output of stap. This could
cause problems in third-party tools that use SystemTap to probe kernel functions. This update
reverts the order of parameters to its original sequence, which is also consistent with their order in
the kernel source code.
• The sys32_pipe function was removed in updated kernels, but the system call tapsets for
the x86_64 kernel still contained an alias that used this function. As a result, using the probe
syscall.* resulted in a semantic error. With this release, the system call tapsets for the x86_64
kernel make the syscall.pipe probe alias (which uses sys32_pipe) optional, thereby avoiding
the error. (BZ#563114
• The unprivileged user mode in this release is stricter, carefully restricting the types of probes an
allowed for unprivileged users. In addition, unprivileged user mode also features clearer diagnostic
messages whenever users attempt to use restricted probes.
• It was possible to call the module shutdown code while a start-up was in progress; this could
leave some kernel callbacks registered after the module has unloaded. As such, running multiple
260
(BZ#517091
2078
(BZ#560890 )
2079
)
2072
(BZ#519314 )
(BZ#510282
2075
(BZ#512202 )
2076
)
(BZ#564443
2074
)
2080
)