1. Instead of enabling all available controllers, figure out which
ones are required, and only enable those.
2. Amend all setFoo() functions to call isFooSet(). While this might
seem unnecessary, it might actually help to uncover a bug.
Imagine someone:
- adds a cgroup.Resources.CpuFoo setting;
- modifies setCpu() to apply the new setting;
- but forgets to amend isCpuSet() accordingly <-- BUG
In this case, a test case modifying CpuFoo will help
to uncover the BUG. This is the reason why it's added.
This patch *could be* amended by enabling controllers on a best-effort
basis, i.e. :
- do not return an error early if we can't enable some controllers;
- if we fail to enable all controllers at once (usually because one
of them can't be enabled), try enabling them one by one.
Currently this is not implemented, and it's not clear whether this
would be a good way to go or not.
[v2: add/use is${Controller}Set() functions]
[v3: document neededControllers()]
[v4: drop "best-effort" part]
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
1. Rename the files
- v1.go: cgroupv1 aka legacy;
- v2.go: cgroupv2 aka unified hierarchy;
- unsupported.go: when systemd is not available.
2. Move the code that is common between v1 and v2 to common.go
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
This check was removed in commit 5406833a65. Now, when this
function is called from a few places, it is no longer obvious
that the path always starts with /sys/fs/cgroup/, so reinstate
the check just to be on the safe side.
This check also ensures that elements[3:] can be used safely.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
fs2 cgroup driver was not working because it did not enable controllers
while creating cgroup directory; instead it was merely doing MkdirAll()
and gathered the list of available controllers in NewManager().
Also, cgroup should be created in Apply(), not while creating a new
manager instance.
To fix:
1. Move the createCgroupsv2Path function from systemd driver to fs2 driver,
renaming it to CreateCgroupPath. Use in Apply() from both fs2 and
systemd drivers.
2. Delay available controllers map initialization to until it is needed.
With this patch:
- NewManager() only performs minimal initialization (initializin
m.dirPath, if not provided);
- Apply() properly creates cgroup path, enabling the controllers;
- m.controllers is initialized lazily on demand.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
The fs2 cgroup driver has a sanity check for path.
Since systemd driver is relying on the same path,
it makes sense to move this check to the common code.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Having map of per-subsystem paths in systemd unified cgroups
driver does not make sense and makes the code less readable.
To get rid of it, move the systemd v1-or-v2 init code to
libcontainer/factory_linux.go which already has a function
to deduce unified path out of paths map.
End result is much cleaner code. Besides, we no longer write pid
to the same cgroup file 7 times in Apply() like we did before.
While at it
- add `rootless` flag which is passed on to fs2 manager
- merge getv2Path() into GetUnifiedPath(), don't overwrite
path if it is set during initialization (on Load).
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
In this very case, the code is writing to cgroup2 filesystem,
and the file name is well known and can't possibly be a symlink.
So, using securejoin is redundant.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
When the cgroupv2 fs driver is used without setting cgroupsPath,
it picks up a path from /proc/self/cgroup. On a host with systemd,
such a path can look like (examples from my machines):
- /user.slice/user-1000.slice/session-4.scope
- /user.slice/user-1000.slice/user@1000.service/gnome-launched-xfce4-terminal.desktop-4260.scope
- /user.slice/user-1000.slice/user@1000.service/gnome-terminal-server.service
This cgroup already contains processes in it, which prevents to enable
controllers for a sub-cgroup (writing to cgroup.subtree_control fails
with EBUSY or EOPNOTSUPP).
Obviously, a parent cgroup (which does not contain tasks) should be used.
Fixes opencontainers/runc/issues/2298
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Commit 6905b72154 treats all negative values as "max",
citing cgroup v1 compatibility as a reason. In fact, in
cgroup v1 only -1 is treated as "unlimited", and other
negative values usually calse an error.
Treat -1 as "max", pass other negative values as is
(the error will be returned from the kernel).
Fixes: 6905b72154
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
The resources.MemorySwap field from OCI is memory+swap, while cgroupv2
has a separate swap limit, so subtract memory from the limit (and make
sure values are set and sane).
Make sure to set MemorySwapMax for systemd, too. Since systemd does not
have MemorySwapMax for cgroupv1, it is only needed for v2 driver.
[v2: return -1 on any negative value, add unit test]
[v3: treat any negative value other than -1 as error]
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
otherwise the memoryStats and hugetlbStats maps are not initialized
and GetStats() segfaults when using them.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Make use of errors.Is() and errors.As() where appropriate to check
the underlying error. The biggest motivation is to simplify the code.
The feature requires go 1.13 but since merging #2256 we are already
not supporting go 1.12 (which is an unsupported release anyway).
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Using errors.Unwrap() is not the best thing to do, since it returns
nil in case of an error which was not wrapped. More to say,
errors package provides more elegant ways to check for underlying
errors, such as errors.As() and errors.Is().
This reverts commit f8e138855d, reversing
changes made to 6ca9d8e6da.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Tested that the EINTR is still being detected:
> $ go1.14 test -c # 1.14 is needed for EINTR to happen
> $ sudo ./fscommon.test
> INFO[0000] interrupted while writing 1063068 to /sys/fs/cgroup/memory/test-eint-89293785/memory.limit_in_bytes
> PASS
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
The Err() method should be called after the Scan() loop, not inside it.
Found by
git grep -A3 -F '.Scan()' | grep Err
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Remove joinCgroupsV2() function, as its name and second parameter
are misleading. Use createCgroupsv2Path() directly, do not call
getv2Path() twice.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Function getSubsystemPath(), while works for v2 unified case, is
suboptimal, as it does a few unnecessary calls.
Add a simplified version of getSubsystemPath(), called getv2Path(),
and use it.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
This code is a copy-paste from cgroupv1 systemd code. Its aim
is to check whether a subsystem is available, and skip those
that are not.
In case v2 unified hierarchy is used, getSubsystemPath never
returns "not found" error, so calling it is useless.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Cgroup v1 kernel doc [1] says:
> We can write "-1" to reset the ``*.limit_in_bytes(unlimited)``.
and cgroup v2 kernel documentation [2] says:
> - If a controller implements an absolute resource guarantee and/or
> limit, the interface files should be named "min" and "max"
> respectively. If a controller implements best effort resource
> guarantee and/or limit, the interface files should be named "low"
> and "high" respectively.
>
> In the above four control files, the special token "max" should be
> used to represent upward infinity for both reading and writing.
Allow -1 value to still be used for v2, converting it to "max"
where it makes sense to do so.
This fixes the following issue:
> runc update test_update --memory-swap -1:
> error while setting cgroup v2: [write /sys/fs/cgroup/machine.slice/runc-cgroups-integration-test.scope/memory.swap.max: invalid argument
> failed to write "-1" to "/sys/fs/cgroup/machine.slice/runc-cgroups-integration-test.scope/memory.swap.max"
> github.com/opencontainers/runc/libcontainer/cgroups/fscommon.WriteFile
> /home/kir/go/src/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/fscommon.go:21
> github.com/opencontainers/runc/libcontainer/cgroups/fs2.setMemory
> /home/kir/go/src/github.com/opencontainers/runc/libcontainer/cgroups/fs2/memory.go:20
> github.com/opencontainers/runc/libcontainer/cgroups/fs2.(*manager).Set
> /home/kir/go/src/github.com/opencontainers/runc/libcontainer/cgroups/fs2/fs2.go:175
> github.com/opencontainers/runc/libcontainer/cgroups/systemd.(*UnifiedManager).Set
> /home/kir/go/src/github.com/opencontainers/runc/libcontainer/cgroups/systemd/unified_hierarchy.go:290
> github.com/opencontainers/runc/libcontainer.(*linuxContainer).Set
> /home/kir/go/src/github.com/opencontainers/runc/libcontainer/container_linux.go:211
[1] linux/Documentation/admin-guide/cgroup-v1/memory.rst
[2] linux/Documentation/admin-guide/cgroup-v2.rst
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
To the best of my knowledge, it has been decided to drop the kernel
memory controller from the cgroupv2 hierarchy, so "kernel memory limits"
do not exist if we're using v2 unified.
So, we need to ignore kernel memory setting. This was already done in
non-systemd case (see commit 88e8350de), let's do the same for systemd.
This fixes the following error:
> container_linux.go:349: starting container process caused "process_linux.go:306: applying cgroup configuration for process caused \"open /sys/fs/cgroup/machine.slice/runc-cgroups-integration-test.scope/tasks: no such file or directory\""
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
1. Return earlier if there is an error.
2. Do not use filepath.Split on every entry, use info.Name() instead.
3. Make readProcsFile() accept file name as an argument, to avoid
unnecessary file name and directory splitting and merging.
4. Skip on info.IsDir() -- this avoids an error when cgroup name is
set to "cgroup.procs".
This is still not very good since filepath.Walk() performs an unnecessary
stat(2) on every entry, but better than before.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
fmt.Sprintf is slow and is not needed here, string concatenation would
be sufficient. It is also redundant to convert []byte from string and
back, since `bytes` package now provides the same functions as `strings`.
Use Fields() instead of TrimSpace() and Split(), mainly for readability
(note Fields() is somewhat slower than Split() but here it doesn't
matter much).
Use Join() to prepend the plus signs.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Golang 1.14 introduces asynchronous preemption which results into
applications getting frequent EINTR (syscall interrupted) errors when
invoking slow syscalls, e.g. when writing to cgroup files.
As writing to cgroups is idempotent, it is safe to retry writing to the
file whenever the write syscall is interrupted.
Signed-off-by: Mario Nitchev <marionitchev@gmail.com>
Kir Kolyshkin