Craig Small [Sun, 2 Feb 2014 06:02:36 +0000 (17:02 +1100)]
pstree: fclose fopened files
Commit [9de89a] uses fopen to find the tasks, but doesn't fclose
them afterward. Thanks for the anonymous and somewhat obscure
bug commenter for pointing it out.
Craig Small [Sun, 29 Sep 2013 05:16:08 +0000 (15:16 +1000)]
pstree shows threads correct names.
pstree would only show the process name, not the thread name. This
change now makes pstree read /proc/<PID>/task/<TID>/stat to find the
threads' real names.
$ pstree -p 1187
udisksd(1187)─┬─{udisksd}(1189)
├─{udisksd}(1191)
├─{udisksd}(1192)
├─{udisksd}(1195)
└─{udisksd}(27395)
./pstree -p 1187
udisksd(1187)─┬─{cleanup}(1195)
├─{gdbus}(1191)
├─{gmain}(1189)
├─{pool}(27395)
└─{probing-thread}(1192)
Craig Small [Tue, 16 Jul 2013 11:39:09 +0000 (21:39 +1000)]
pstree uses COLUMNS environement variable
pstree previously only used the window size for determining number of
columns, then a default of 132. With this change, pstree now checks
the COLUMNS environment variable first and uses that if valid.
env checking code nicked from top.c, by Jim Warner.
Han Pingtian [Thu, 20 Dec 2012 03:31:50 +0000 (11:31 +0800)]
doesn't complain about "not a directory"
If /proc/#/fd points to a path which has component removed, fuser
will complain like "Cannot stat file /proc/29322/fd/16: Not a
directory". Looks like it can be disabled just like commit 898dc70a1f
did.
Mike Frysinger [Tue, 19 Feb 2013 19:44:50 +0000 (14:44 -0500)]
use AC_CANONICAL_HOST
--host is used for the system where the compiled code will run.
--target is not the correct system to look that up.
http://www.gnu.org/software/autoconf/manual/html_node/Manual-Configuration.html#Manual-Configuration
Signed-off-by: Mike Frysinger <vapier@gentoo.org> Signed-off-by: Craig Small <csmall@enc.com.au>
Craig Small [Sat, 6 Oct 2012 23:52:46 +0000 (10:52 +1100)]
pstree compiles with SE Linux
pstree failed to compile with SE Linux enabled because one of the
scontext was missed and without it enabled the bug doesn't appear.
pstree is now re-worked so scontext is defined as a dummy meaning
most of the code except the reall SE Linux specific stuff is compilied
meaning this sort of thing shouldn't happen too much again.
Craig Small [Fri, 31 Aug 2012 04:21:19 +0000 (14:21 +1000)]
fuser finds unix sockets
in check_dir() stat was replaced with device() which found the right
device but zeroed out the inode, making that check always fail. This
meant fuser never found unix sockets.
Craig Small [Fri, 31 Aug 2012 00:05:24 +0000 (10:05 +1000)]
pstree finds orphans
Commit 26f9b6c1553d021c0bf9dd85f0647dc6e210948d while fixing the missing
root pid problem introduced another problem where PIDs are out of order.
A process that has a lower PID than its PPID would appear twice with
the tree breaking in strange ways.
We now scan the entire process tree first then perform two checks
if PID==1 is missing add it
if a process has no parent, then make it a child of PID==1, either the
real one or the faked one made in step 1
As this is done after all procsses are found, we don't get the previous
breakage.
Craig Small [Thu, 30 Aug 2012 22:09:45 +0000 (08:09 +1000)]
fuser -m regression
While fuser -m /boot worked, fuser -m /dev/blah would show everything.
This was because the filesystem for /dev was added as a target, not the
filesystem /dev/blah.
This simple fix suggested by SF user hanpt works well.
Ref: https://sourceforge.net/tracker/?func=detail&aid=3559232&group_id=15273&atid=115273 Signed-off-by: Craig Small <csmall@users.sourceforge.net>
Craig Small [Fri, 20 Jul 2012 12:32:27 +0000 (22:32 +1000)]
pstree sorts properly on names
pstree added temporary parents to processes, so if a child process
was lexigraphically before the parents or its "aunties and uncles"
then they were sorted with the parent temporary name "?".
This patch reorders the children in the parents list when they
are renamed.
Craig Small [Mon, 11 Jun 2012 01:36:55 +0000 (11:36 +1000)]
sed uses proper BRE
Apparently even though [0-9]+ works most of the time, its not proper BRE
and doesn't work ALL of the time. Replacing this with [0-9][0-9]* makes
everyone happy.
Noticed by Rich from the busybox project and reported by John Spencer
of the same project.
Corinna Vinschen [Sat, 19 May 2012 08:54:35 +0000 (18:54 +1000)]
3 fixes for Cygwin
The attached patch fixes three issues of fuser on Cygwin.:
- Cygwin does not support /proc/net/unix. The call to fill_unix_cache
always generates an error message that /proc/net/unix couldn't be
found, so the patch disables the call for Cygwin. An alternative
fix would be to avoid the error message entirely.
- `ls /dev' on Cygwin only lists manually added entries to /dev, not the
default files suported by Cygwin internally. Also,
/proc/$PID/mountinfo doesn't exist up to Cygwin 1.7.11. This
breaks fuser when using the _LISTS_H option.
- If you don't include lists.h, the code in check_dir doesn't work well.
st.st_ino is never set to 0, so the timeout(stat, ...) is never
called. Also, thedev never gets a valid value in this case, but it's
used subsequently for comparisons.
Samuel Thibault [Sat, 19 May 2012 08:09:41 +0000 (18:09 +1000)]
Remove fuser out of target for hurd-i386
hurd-i386 does not have PATH_MAX which means fuser won't compile.
However the archtecture cannot use fuser anyway so we now don't
even try to compile it for hurd-i386
Bug-Debian: http://bugs.debian.org/673485 Signed-off-by: Craig Small <csmall@enc.com.au>
Craig Small [Fri, 27 Jan 2012 02:11:25 +0000 (13:11 +1100)]
Added hardened CFLAGS
Debian and Ubuntu (at the least) have some CFLAGS that harden the binary
for some typical attacks. psmisc now will have these flags on by default
but checks the compilier supports them.
Sami Kerola [Wed, 4 Jan 2012 13:49:08 +0000 (14:49 +0100)]
build-sys: fix AC_TRY_LINK warnings
configure.ac:71: warning: The macro `AC_TRY_LINK' is obsolete.
configure.ac:71: You should run autoupdate.
../../lib/autoconf/general.m4:2688: AC_TRY_LINK is expanded from...
../../lib/m4sugar/m4sh.m4:606: AS_IF is expanded from...
../../lib/autoconf/general.m4:2032: AC_CACHE_VAL is expanded from...
../../lib/autoconf/general.m4:2053: AC_CACHE_CHECK is expanded from...
m4/gettext.m4:367: gt_INTL_MACOSX is expanded from...
m4/gettext.m4:57: AM_GNU_GETTEXT is expanded from...
configure.ac:71: the top level
configure.ac:71: warning: The macro `AC_TRY_LINK' is obsolete.
configure.ac:71: You should run autoupdate.
../../lib/autoconf/general.m4:2688: AC_TRY_LINK is expanded from...
../../lib/m4sugar/m4sh.m4:606: AS_IF is expanded from...
../../lib/autoconf/general.m4:2032: AC_CACHE_VAL is expanded from...
../../lib/autoconf/general.m4:2053: AC_CACHE_CHECK is expanded from...
m4/gettext.m4:57: AM_GNU_GETTEXT is expanded from...
configure.ac:71: the top level
configure.ac:71: warning: The macro `AC_TRY_LINK' is obsolete.
configure.ac:71: You should run autoupdate.
../../lib/autoconf/general.m4:2688: AC_TRY_LINK is expanded from...
../../lib/m4sugar/m4sh.m4:606: AS_IF is expanded from...
../../lib/autoconf/general.m4:2032: AC_CACHE_VAL is expanded from...
../../lib/autoconf/general.m4:2053: AC_CACHE_CHECK is expanded from...
Signed-off-by: Sami Kerola <sami.kerola@tomtom.com>
Craig Small [Sun, 19 Jun 2011 01:16:43 +0000 (11:16 +1000)]
2 Fixes for Debian Bug#623425
Both patches are by Jonathan Nieder <jrnieder@gmail.com>
Ever since psmisc 22.4 (2007-04-08), "killall <path>" compares
the target of the /proc/<pid>/exe link to <path> when comparison
of inode numbers fails. But the code to do this has two problems:
- readlink does not NUL-terminate, but we use strcmp to
compare strings. Probably this hasn't been a problem so far
because (1) the on-stack buffer happened to have zeroes in the
right places or (2) some implementations of readlink might
happen to NUL-terminate their result when convenient anyway.
- it relies on PATH_MAX to determine the size of the buffer,
so the code fails to build from source on platforms (like the
Hurd) that have no global PATH_MAX.
Fix both by using a buffer of size strlen("<path>") + 1 and comparing
the link target to <path> with memcmp after checking that it fit in
the buffer.
For consistency with the surrounding code, the pid is considered not
to match if malloc or readlink fails.
---------------
The Hurd has most of the expected /proc/<pid> hierarchy but no
/proc/self. Without this change, running "sleep 10 & killall sleep"
on that platform produces the confusing message:
/proc is empty (not mounted ?)
Craig Small [Tue, 22 Feb 2011 09:35:14 +0000 (20:35 +1100)]
Added SF patches and bug reports
* Applied patch from Werner Fink to avoid stat(2) on NFS mounts
* Zeros process group memory - Patch by jgorig SF#3152925 RH#666213
* fuser -m -s flags work - Patch by jgorig SF#31110178 RH#651794
* fuser silent if /proc/swaps not available SF#3072134
* ppc 64 support for peekfd by jgorig SF#3166444
* jiffies now ULL in killall SF#3138538
* pstree can show parents of a process. Patch supplied by Curtis Hawthorne SF#3135157
Werner Fink [Tue, 21 Sep 2010 07:04:16 +0000 (09:04 +0200)]
Add an option -w to fuser to be able to send signals
to processes which have write access to an open file
or directory. This option is silently ignored if -k is
not present too.
projects / psmisc / log