support running tests from travis

Add a main.sh which can run a self-confined testsuite.

Note that without having lxcfs_mkdir installed, we cannot mkdir as
non-root user.

Update the test scripts to support using a lxcfs mountpoint other
than /var/lib/lxcfs

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

enable travis

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

don't let idletime be > reaperage

This is not a good way to do this.  We should decide on a proper
heuristic.  We could take something like reaperage * (idletime/total_uptime),
but that doesn't scale for how much our own container used the cpu nor
for time.

I will open a github issue to fix this.

However as it currently stands the test_proc testcase was failing, this
at least lets it pass.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Merge pull request #49 from hallyn/offset1

fix corner cases in uptime and diskstat read

fix corner cases in uptime and diskstat read

Closes #33

The code for these (a shortcut version of the other proc_*_read ones) was
doing snprintf(buf, size, ...).  If the user only requests one byte, we
just served them a trailing 0.  Fix that.

We weren't handling reads with offset in these cases either.  Fix that.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Merge pull request #48 from bmiklautz/uptime_format

Change uptime format

update .gitignore

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

move lxcfs_mkdir under libexecdir

Note, this will require us to add

       dh_auto_configure -- --libdir=\$${prefix}/lib/$(DEB_HOST_MULTIARCH) \
               --libexecdir=\$${prefix}/lib/$(DEB_HOST_MULTIARCH)

to debian/rules.

Closes #46

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Change "uptime" format

/proc/uptime has the format %lu.%02lu %lu.%02lu. The format used by
lxcfs doesn't consider the nano second portion of the uptime which might
cause programs that rely on that format to fail.

This commit adapts the uptime format to match the kernels by adding
trailing dummy values (.0) to the uptime and idle time values.
The parsing of /proc/uptime was updated.

Signed-off-by: Bernhard Miklautz <bernhard.miklautz@shacknet.at>

configure.ac: drop libcgmanager dependency

Closes #45

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

configure.ac: v0.11

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Remove a memory leak from getattr

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

remove stale comments, update some

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

fix memory leak

each key also has an allocated char*, so use list_key to free the key.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

lxc mount hook: whitespace fixup

Acked-by: Serge Hallyn <serge.hallyn@ubuntu.com>

lxc mount hook: add missing space before ]

Otherwise [ interprets the ] as part of the path and also
errors about a missing ']'

Acked-by: Serge Hallyn <serge.hallyn@ubuntu.com>

swap out libnih and libnih-dbus for glib

The motivation is to make threading possible, to hopefully greatly
speed up systemd startup inside containers.

This required converting all the nih-ified memory tracking.  (Some
of this can probably be done smarter in a more glib-friendly way, i.e.
avoiding doing a glib string alloc followed by strdup followed by
freeing the glib string)

We open a single dbus connection for all threads to use.  If that
connection is closed (i.e. cgmanager exits / restarts) the first
task to find it so takes a mutex and attempts to reconnect, once
per second, until it is reconnected.

When creating a directory for non-root user, execute a new binary
to get a clean dbus session as that user.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

allow systemd to escape its init.scope cgroup

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Merge pull request #38 from sn-x/master

Fixed issue #37

Fixed issue #37

Merge pull request #34 from ralt/patch-1

Better explain what lxcfs does

Better explain what lxcfs does

At first glance, I thought this project was a custom fs for the hosts, not for the containers themselves.

Hopefully this line will clear up the misunderstanding for future
readers, and show that this project is *really* useful!

Signed-off-by: Florian Margaine <florian@margaine.com>

configure.ac: v0.10

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

fix symlink

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

lxc mount hook: implement stgraber's suggestions

revert to using sh and use symlinks instead of extra mounts.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Update lxc mount hook to ovver a mount per co-mounted controller

If cpu,cpuacct are comounted, then in addition to cpu,cpuacct make
mounts for 'cpu' and 'cpuacct'.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

lxc mount hook: don't mount lxcfs fs for any controller already setup

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

don't thread

The pieces are there, but for now it's still not safe with libdbus

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Return host's meminfo file if no memory cgroup

If memory cgroup is not available (that is, we can't find a
memory cgroup for the reading task), then just return the
contents of the host's /proc/meminfo.  (Same for all proc file
reads)

Closes #30

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

fix two threading issues

Make sure to prep dbus for threading before we start.

And use _exit() any time we are exiting from a forked child.  This is
to avoid calling the at_exit() functions.  Once a thread in the main
program has called nih_error_init(), this registers an at_exit fn which
asserts that the nih_context not be null - but after we fork, if libnih
is built with --enable-threading, then the nih_context is in fact null.
The only way to clear the atexit fns would be to exec().  So call
_exit() instead of exit(), because _exit() avoids calling the atexit
fns.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

use threads when safe

libnih, when not built with --enable-threading, cannot be safely
used by a threaded application.  Detect whether it is built to be
threadsafe using a new libnih helper, and, if so, run threaded by
(a) not passing '-s' to fuse, and (b) making the dbus connection
and detected api version thread-local.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

configure.ac: v0.9

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Merge pull request #26 from mikemccracken/maybe-fix-24

Add error handling for snprintf calls inside *_read() funcs

inside *_read() funcs, return 0 for snprintf errors and truncations

Signed-off-by: Michael McCracken <mike.mccracken@canonical.com>

check return value from snprintf to avoid FUSE errors. Potentially fixes #24.

Signed-off-by: Michael McCracken <mike.mccracken@canonical.com>

configure.ac: release 0.8

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

implement --version

Because this is needed for help2man to determine the program name.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Fix the check for cpu being in cpuset range

Also add a testcase to make sure that the new function is correct.
In order to share the cpuset range checking code with with the
test, move it into cpuset.c.  Not sure whether we want that in a
utils.c instead.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Add some more sanity checks

Limit string size in sscanf, convert some sprintfs to snprintfs
(to make them future-proof), and free d in case of error.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

free d at program end

to silence warnings from memory checkers

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Make sure that that '/cgroup' and the controller are sep'd by /

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

cg_read: add ending newline

Closes #23

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

use direct_io

This will prevent a file read for over getattr.

File sizes show up as 0, but they do so in cgroupfs and procfs as well.

Closes #20.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Provide the fuse options we want ourselves

continue to allow the options which in the past we basically required
callers to provide ("-s -f -o allow_other"), exit on anything else.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Cleanup whitespace

and remove commented out debug statement

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Merge pull request #22 from hustcat/cache.1

cache for proc file

info as parent when malloc memory for file buf
reduce one blank for /proc/stat

Signed-off-by: Ye Yin <eyniy@qq.com>

Merge pull request #19 from hallyn/cache.1

Cache.1

add cache for proc file, for support multiple read

Signed-off-by: Ye Yin <eyniy@qq.com>

LXC hook fixes

 - Detects whether cgroups are enabled in the container's config
 - Don't mount a second tmpfs on /sys/fs/cgroup

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

cache: use file_dir object as parent for alloc'ed objects

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

don't check for cgroup perms when opendir'ing /cgroup

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

add a file_info cache pointer for procfile reads

This doesn't actually use them yet, but now fi->fh->buf can be
used in https://github.com/lxc/lxcfs/pull/18

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

add cache types for each procfile type that we provide

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

check that file and dir entries don't get mixed up

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

cache verified controller/cgroup info at cg file read/write

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

cache opendir work and re-use it at readdir

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

release 0.7

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

be robust to glibc fork failures

kill child and retry when __glibc_fork() child hangs on an assert
(newpid != ppid) failure.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

timeout the reception of a scm_cred

Else if the peer is killed at a strange time we will hang forever waiting.

Also don't kill children - always just let them time out

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

nih_local's must be initialized as NULL

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Merge pull request #16 from hustcat/diskstats

Add /proc/diskstats

Add /proc/diskstats
Signed-off-by: Ye Yin <eyniy@qq.com>

Merge pull request #15 from steelbrain/master

Fix formatting of README

Fix formatting of README
Signed-off-by: Anees Iqbal <aneesiqbalbhatti@gmail.com>

Fix test_proc following the stat fix

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Release LXCFS 0.6

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Show cpu-average in /proc/stat and start cup numbering at 0

Signed-off-by: Christian Brauner <christianvanbrauner@gmail.com>
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

clean up after several functions

which weren't freeing line of closing an open file.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Update README

Close #10

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Release LXCFS 0.5

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

hook: Don't spam /etc/mtab

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Make the LXC hook more robust

Don't fail if /sys/fs/cgroup or some proc files don't exist.

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Update gitignore

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Better detect the cgmanager version

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Release LXCFS 0.4

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Remount / recursive ms-slave in our new mntns

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Revert "remount / ms-slave"

This will affect the hostns, let's go with my implementation instead :)

This reverts commit 1ff1e28c04267064c73c5b968d31f6e5f2e1f3fa.

remount / ms-slave

to avoid propagating proc mounts from get_pid1_time() back to host

Signed-ff-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Release LXCFS 0.3

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Fix LXC config install path

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Release LXCFS 0.2

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Update the lxc hook and add a LXC config file

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

tests: Fix return value

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Release LXCFS 0.1

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Fix autotools

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

test: Introduce a basic cgroup tree test

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

test: Introduce a basic proc tree test

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

update docs

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

add lxcfs container hook and comment

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

write: make sure to null terminate the buffer

We are passed a size, respect it.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Add some more missing newlines

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

fix up pid conversion algorithm

Mainly when the pid conversion task gets a ESRCH from sendmsg,
then it should send send the next message (with v='1' meaning
error) without first waiting for a ping from its peer, since
the peer already sent one and has gotten no response.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Update README

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

Add some more information to error messages

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

convert pids at write to tasks and cgroup.procs files

we're not yet distinguishing between thread vs threadgroup moving.

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Merge branch 'readtasks.pass.creds.2' into m.2

read: translate pids into the reader's pidns

When reading a cgroupfs 'tasks' or 'cgroup.procs' file, we must translate
all pids from our namespace into the reader's.

TODO - convert writes

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>

Add missing line breaks

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>

implement read of /proc/uptime

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>