Commit Graph

9 Commits

Author SHA1 Message Date
Wei Fu
fd3f3d5a13 pkg/sys: add GetUsernsForNamespace interface
Signed-off-by: Wei Fu <fuweid89@gmail.com>
2024-09-11 07:21:43 +08:00
Wei Fu
490e45a08a pkg/sys: Add UnshareAfterEnterUserns function
It allows to disassociate parts of its execution context within a user
namespace.

Signed-off-by: Wei Fu <fuweid89@gmail.com>
2024-09-11 07:21:39 +08:00
Wei Fu
3cd8f9734d core/mount: use ptrace instead of go:linkname
The Go runtime has started to [lock down future uses of linkname][1] since
go1.23. In the go source code, containerd project has been marked in the
comment, [hall of shame][2]. Well, the go:linkname is used to fork no-op
subprocess efficiently. However, since that comment, I would like to use
ptrace and remove go:linkname in the whole repository.

With go1.22 `go:linkname`:

```bash
$ go test -bench=.  -benchmem ./ -exec sudo
goos: linux
goarch: amd64
pkg: github.com/containerd/containerd/v2/core/mount
cpu: AMD Ryzen 7 5800H with Radeon Graphics
BenchmarkBatchRunGetUsernsFD_Concurrent1-16                 2440            533320 ns/op            1145 B/op         43 allocs/op
BenchmarkBatchRunGetUsernsFD_Concurrent10-16                 342           3661616 ns/op           11562 B/op        421 allocs/op
PASS
ok      github.com/containerd/containerd/v2/core/mount  2.983s
```

With go1.22 `ptrace`:

```bash
$ go test -bench=.  -benchmem ./ -exec sudo
goos: linux
goarch: amd64
pkg: github.com/containerd/containerd/v2/core/mount
cpu: AMD Ryzen 7 5800H with Radeon Graphics
BenchmarkBatchRunGetUsernsFD_Concurrent1-16                 1785            739557 ns/op            3948 B/op         68 allocs/op
BenchmarkBatchRunGetUsernsFD_Concurrent10-16                 328           4024300 ns/op           39601 B/op        671 allocs/op
PASS
ok      github.com/containerd/containerd/v2/core/mount  3.104s
```

With go1.23 `ptrace`:

```bash
$ go test -bench=.  -benchmem ./ -exec sudo
goos: linux
goarch: amd64
pkg: github.com/containerd/containerd/v2/core/mount
cpu: AMD Ryzen 7 5800H with Radeon Graphics
BenchmarkBatchRunGetUsernsFD_Concurrent1-16                 1815            723252 ns/op            4220 B/op         69 allocs/op
BenchmarkBatchRunGetUsernsFD_Concurrent10-16                 319           3957157 ns/op           42351 B/op        682 allocs/op
PASS
ok      github.com/containerd/containerd/v2/core/mount  3.051s
```

Diff:

The `ptrace` is slower than `go:linkname` mode. However, it's accepctable.

```
goos: linux
goarch: amd64
pkg: github.com/containerd/containerd/v2/core/mount
cpu: AMD Ryzen 7 5800H with Radeon Graphics
                                    │ go122-golinkname │             go122-ptrace              │             go123-ptrace              │
                                    │      sec/op      │    sec/op     vs base                 │    sec/op     vs base                 │
BatchRunGetUsernsFD_Concurrent1-16        533.3µ ± ∞ ¹   739.6µ ± ∞ ¹        ~ (p=1.000 n=1) ²   723.3µ ± ∞ ¹        ~ (p=1.000 n=1) ²
BatchRunGetUsernsFD_Concurrent10-16       3.662m ± ∞ ¹   4.024m ± ∞ ¹        ~ (p=1.000 n=1) ²   3.957m ± ∞ ¹        ~ (p=1.000 n=1) ²
geomean                                   1.397m         1.725m        +23.45%                   1.692m        +21.06%
¹ need >= 6 samples for confidence interval at level 0.95
² need >= 4 samples to detect a difference at alpha level 0.05

                                    │ go122-golinkname │              go122-ptrace               │              go123-ptrace               │
                                    │       B/op       │     B/op       vs base                  │     B/op       vs base                  │
BatchRunGetUsernsFD_Concurrent1-16       1.118Ki ± ∞ ¹   3.855Ki ± ∞ ¹         ~ (p=1.000 n=1) ²   4.121Ki ± ∞ ¹         ~ (p=1.000 n=1) ²
BatchRunGetUsernsFD_Concurrent10-16      11.29Ki ± ∞ ¹   38.67Ki ± ∞ ¹         ~ (p=1.000 n=1) ²   41.36Ki ± ∞ ¹         ~ (p=1.000 n=1) ²
geomean                                  3.553Ki         12.21Ki        +243.65%                   13.06Ki        +267.43%
¹ need >= 6 samples for confidence interval at level 0.95
² need >= 4 samples to detect a difference at alpha level 0.05

                                    │ go122-golinkname │             go122-ptrace             │             go123-ptrace             │
                                    │    allocs/op     │  allocs/op   vs base                 │  allocs/op   vs base                 │
BatchRunGetUsernsFD_Concurrent1-16         43.00 ± ∞ ¹   68.00 ± ∞ ¹        ~ (p=1.000 n=1) ²   69.00 ± ∞ ¹        ~ (p=1.000 n=1) ²
BatchRunGetUsernsFD_Concurrent10-16        421.0 ± ∞ ¹   671.0 ± ∞ ¹        ~ (p=1.000 n=1) ²   682.0 ± ∞ ¹        ~ (p=1.000 n=1) ²
geomean                                    134.5         213.6        +58.76%                   216.9        +61.23%
¹ need >= 6 samples for confidence interval at level 0.95
² need >= 4 samples to detect a difference at alpha level 0.05
```

[1]: <https://github.com/golang/go/issues/67401>
[2]: <https://github.com/golang/go/blob/release-branch.go1.23/src/runtime/proc.go#L4820>

Signed-off-by: Wei Fu <fuweid89@gmail.com>
2024-08-26 21:19:50 +08:00
Sebastiaan van Stijn
9776047243
migrate to github.com/moby/sys/userns
Commit 8437c567d8 migrated the use of the
userns package to the github.com/moby/sys/user module.

After further discussion with maintainers, it was decided to move the
userns package to a separate module, as it has no direct relation with
"user" operations (other than having "user" in its name).

This patch migrates our code to use the new module.

Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2024-08-08 12:48:54 +02:00
Sebastiaan van Stijn
8437c567d8
pkg/userns: deprecate and migrate to github.com/moby/sys/user/userns
The userns package in libcontainer was integrated into the moby/sys/user
module at commit [3778ae603c706494fd1e2c2faf83b406e38d687d][1].

This patch deprecates the containerd fork of that package, and adds it as
an alias for the moby/sys/user/userns package.

[1]: 3778ae603c

Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2024-07-26 09:47:50 +02:00
Danny Canter
3ea69db8e9 Add helper to ignore eintr
We have quite a few pieces of code laying around containerd
that all loop and ignore eintr as they make syscalls directly
(or use a unix/syscall wrapper) because there's no stdlib
equivalent. This adds a small utility to pkg/sys that we can
use for all of these spots.

Signed-off-by: Danny Canter <danny@dcantah.dev>
2024-04-10 11:16:23 -07:00
Phil Estes
723306d0ed
Disable OOM set score unpriv test temporarily
Temporary skip while we find root cause of GHA environment changes
causing failure.

Signed-off-by: Phil Estes <estesp@amazon.com>
2024-03-06 11:38:53 -05:00
Maksym Pavlenko
1ce7b99511 Remove deprecated filesys funcs
Signed-off-by: Maksym Pavlenko <pavlenko.maksym@gmail.com>
2024-02-10 17:55:57 -08:00
Derek McGowan
6be90158cd
Move sys to pkg/sys
Signed-off-by: Derek McGowan <derek@mcg.dev>
2024-01-17 09:56:16 -08:00