kubelet: use new securejoin API

Using securejoin.SecureJoin() ensures that paths are bound within a
given root, but it doesn't protect from changes happening between the
construction of the path and its use. securejoin 0.3 introduces a new
Linux-specific API which avoids this by making rooted open operations
explicit; this migrates kubelet's log retrieval to use that.

Signed-off-by: Stephen Kitt <skitt@redhat.com>

LICENSES/vendor/github.com/cyphar/filepath-securejoin/LICENSE

@@ -1,7 +1,7 @@
 = vendor/github.com/cyphar/filepath-securejoin licensed under: =
 
 Copyright (C) 2014-2015 Docker Inc & Go Authors. All rights reserved.
-Copyright (C) 2017 SUSE LLC. All rights reserved.
+Copyright (C) 2017-2024 SUSE LLC. All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are
@@ -29,4 +29,4 @@ THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-= vendor/github.com/cyphar/filepath-securejoin/LICENSE 8d322afab99e1998dbfcc712f94e824d
+= vendor/github.com/cyphar/filepath-securejoin/LICENSE 7e05df0b39896d74600ef94ab46dce89

go.mod

@@ -24,7 +24,7 @@ require (
 	github.com/coreos/go-oidc v2.2.1+incompatible
 	github.com/coreos/go-systemd/v22 v22.5.0
 	github.com/cpuguy83/go-md2man/v2 v2.0.4
-	github.com/cyphar/filepath-securejoin v0.2.4
+	github.com/cyphar/filepath-securejoin v0.3.4
 	github.com/distribution/reference v0.6.0
 	github.com/docker/go-units v0.5.0
 	github.com/emicklei/go-restful/v3 v3.11.0

go.sum

@@ -210,8 +210,8 @@ github.com/cpuguy83/go-md2man/v2 v2.0.4/go.mod h1:tgQtvFlXSQOSOSIRvRPT7W67SCa46t
 github.com/creack/pty v1.1.9/go.mod h1:oKZEueFk5CKHvIhNR5MUki03XCEU+Q6VDXinZuGJ33E=
 github.com/creack/pty v1.1.18 h1:n56/Zwd5o6whRC5PMGretI4IdRLlmBXYNjScPaBgsbY=
 github.com/creack/pty v1.1.18/go.mod h1:MOBLtS5ELjhRRrroQr9kyvTxUAFNvYEK993ew/Vr4O4=
-github.com/cyphar/filepath-securejoin v0.2.4 h1:Ugdm7cg7i6ZK6x3xDF1oEu1nfkyfH53EtKeQYTC3kyg=
+github.com/cyphar/filepath-securejoin v0.3.4 h1:VBWugsJh2ZxJmLFSM06/0qzQyiQX2Qs0ViKrUAcqdZ8=
-github.com/cyphar/filepath-securejoin v0.2.4/go.mod h1:aPGpWjXOXUn2NCNjFvBE6aRxGGx79pTxQpKOJNYHHl4=
+github.com/cyphar/filepath-securejoin v0.3.4/go.mod h1:8s/MCNJREmFK0H02MF6Ihv1nakJe4L/w3WZLHNkvlYM=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc h1:U9qPSI2PIWSS1VwoXQT9A3Wy9MM3WgvqSxFWenqJduM=

pkg/kubelet/kubelet_server_journal.go

@@ -361,12 +361,7 @@ func heuristicsCopyFileLogs(ctx context.Context, w io.Writer, logDir, service st
 
 	var err error
 	for _, logFileName := range logFileNames {
-		var logFile string
+		err = heuristicsCopyFileLog(ctx, w, logDir, logFileName)
-		logFile, err = securejoin.SecureJoin(logDir, logFileName)
-		if err != nil {
-			break
-		}
-		err = heuristicsCopyFileLog(ctx, w, logFile)
 		if err == nil {
 			break
 		} else if errors.Is(err, os.ErrNotExist) {
@@ -407,30 +402,6 @@ func newReaderCtx(ctx context.Context, r io.Reader) io.Reader {
 	}
 }
 
-// heuristicsCopyFileLog returns the contents of the given logFile
-func heuristicsCopyFileLog(ctx context.Context, w io.Writer, logFile string) error {
-	fInfo, err := os.Stat(logFile)
-	if err != nil {
-		return err
-	}
-	// This is to account for the heuristics where logs for service foo
-	// could be in /var/log/foo/
-	if fInfo.IsDir() {
-		return os.ErrNotExist
-	}
-
-	f, err := os.Open(logFile)
-	if err != nil {
-		return err
-	}
-	defer f.Close()
-
-	if _, err := io.Copy(w, newReaderCtx(ctx, f)); err != nil {
-		return err
-	}
-	return nil
-}
-
 func safeServiceName(s string) error {
 	// Max length of a service name is 256 across supported OSes
 	if len(s) > maxServiceLength {

pkg/kubelet/kubelet_server_journal_linux.go

@@ -21,8 +21,13 @@ package kubelet
 import (
 	"context"
 	"fmt"
+	"io"
+	"os"
 	"os/exec"
 	"strings"
+
+	securejoin "github.com/cyphar/filepath-securejoin"
+	unix "golang.org/x/sys/unix"
 )
 
 // getLoggingCmd returns the journalctl cmd and arguments for the given nodeLogQuery and boot. Note that
@@ -72,3 +77,32 @@ func checkForNativeLogger(ctx context.Context, service string) bool {
 	// hence we search for it in the list of services known to journalctl
 	return strings.Contains(string(output), service+".service")
 }
+
+// heuristicsCopyFileLog returns the contents of the given logFile
+func heuristicsCopyFileLog(ctx context.Context, w io.Writer, logDir, logFileName string) error {
+	f, err := securejoin.OpenInRoot(logDir, logFileName)
+	if err != nil {
+		return err
+	}
+	// Ignoring errors when closing a file opened read-only doesn't cause data loss
+	defer func() { _ = f.Close() }()
+	fInfo, err := f.Stat()
+	if err != nil {
+		return err
+	}
+	// This is to account for the heuristics where logs for service foo
+	// could be in /var/log/foo/
+	if fInfo.IsDir() {
+		return os.ErrNotExist
+	}
+	rf, err := securejoin.Reopen(f, unix.O_RDONLY)
+	if err != nil {
+		return err
+	}
+	defer func() { _ = rf.Close() }()
+
+	if _, err := io.Copy(w, newReaderCtx(ctx, rf)); err != nil {
+		return err
+	}
+	return nil
+}

pkg/kubelet/kubelet_server_journal_nonlinux.go

@@ -0,0 +1,56 @@
+//go:build !linux
+
+/*
+Copyright 2024 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package kubelet
+
+import (
+	"context"
+	"io"
+	"os"
+
+	securejoin "github.com/cyphar/filepath-securejoin"
+)
+
+// heuristicsCopyFileLog returns the contents of the given logFile
+func heuristicsCopyFileLog(ctx context.Context, w io.Writer, logDir, logFileName string) error {
+	logFile, err := securejoin.SecureJoin(logDir, logFileName)
+	if err != nil {
+		return err
+	}
+	fInfo, err := os.Stat(logFile)
+	if err != nil {
+		return err
+	}
+	// This is to account for the heuristics where logs for service foo
+	// could be in /var/log/foo/
+	if fInfo.IsDir() {
+		return os.ErrNotExist
+	}
+
+	f, err := os.Open(logFile)
+	if err != nil {
+		return err
+	}
+	// Ignoring errors when closing a file opened read-only doesn't cause data loss
+	defer func() { _ = f.Close() }()
+
+	if _, err := io.Copy(w, newReaderCtx(ctx, f)); err != nil {
+		return err
+	}
+	return nil
+}

pkg/kubelet/kubelet_server_journal_others.go

@@ -24,11 +24,11 @@ import (
 )
 
 // getLoggingCmd on unsupported operating systems returns the echo command and a warning message (as strings)
-func getLoggingCmd(n *nodeLogQuery, services []string) (string, []string, error) {
+func getLoggingCmd(_ *nodeLogQuery, _ []string) (string, []string, error) {
 	return "", []string{}, errors.New("Operating System Not Supported")
 }
 
 // checkForNativeLogger on unsupported operating systems returns false
-func checkForNativeLogger(ctx context.Context, service string) bool {
+func checkForNativeLogger(_ context.Context, _ string) bool {
 	return false
 }

vendor/github.com/cyphar/filepath-securejoin/CHANGELOG.md

@@ -0,0 +1,178 @@
+# Changelog #
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](http://keepachangelog.com/)
+and this project adheres to [Semantic Versioning](http://semver.org/).
+
+## [Unreleased] ##
+
+## [0.3.4] - 2024-10-09 ##
+
+### Fixed ###
+- Previously, some testing mocks we had resulted in us doing `import "testing"`
+  in non-`_test.go` code, which made some downstreams like Kubernetes unhappy.
+  This has been fixed. (#32)
+
+## [0.3.3] - 2024-09-30 ##
+
+### Fixed ###
+- The mode and owner verification logic in `MkdirAll` has been removed. This
+  was originally intended to protect against some theoretical attacks but upon
+  further consideration these protections don't actually buy us anything and
+  they were causing spurious errors with more complicated filesystem setups.
+- The "is the created directory empty" logic in `MkdirAll` has also been
+  removed. This was not causing us issues yet, but some pseudofilesystems (such
+  as `cgroup`) create non-empty directories and so this logic would've been
+  wrong for such cases.
+
+## [0.3.2] - 2024-09-13 ##
+
+### Changed ###
+- Passing the `S_ISUID` or `S_ISGID` modes to `MkdirAllInRoot` will now return
+  an explicit error saying that those bits are ignored by `mkdirat(2)`. In the
+  past a different error was returned, but since the silent ignoring behaviour
+  is codified in the man pages a more explicit error seems apt. While silently
+  ignoring these bits would be the most compatible option, it could lead to
+  users thinking their code sets these bits when it doesn't. Programs that need
+  to deal with compatibility can mask the bits themselves. (#23, #25)
+
+### Fixed ###
+- If a directory has `S_ISGID` set, then all child directories will have
+  `S_ISGID` set when created and a different gid will be used for any inode
+  created under the directory. Previously, the "expected owner and mode"
+  validation in `securejoin.MkdirAll` did not correctly handle this. We now
+  correctly handle this case. (#24, #25)
+
+## [0.3.1] - 2024-07-23 ##
+
+### Changed ###
+- By allowing `Open(at)InRoot` to opt-out of the extra work done by `MkdirAll`
+  to do the necessary "partial lookups", `Open(at)InRoot` now does less work
+  for both implementations (resulting in a many-fold decrease in the number of
+  operations for `openat2`, and a modest improvement for non-`openat2`) and is
+  far more guaranteed to match the correct `openat2(RESOLVE_IN_ROOT)`
+  behaviour.
+- We now use `readlinkat(fd, "")` where possible. For `Open(at)InRoot` this
+  effectively just means that we no longer risk getting spurious errors during
+  rename races. However, for our hardened procfs handler, this in theory should
+  prevent mount attacks from tricking us when doing magic-link readlinks (even
+  when using the unsafe host `/proc` handle). Unfortunately `Reopen` is still
+  potentially vulnerable to those kinds of somewhat-esoteric attacks.
+
+  Technically this [will only work on post-2.6.39 kernels][linux-readlinkat-emptypath]
+  but it seems incredibly unlikely anyone is using `filepath-securejoin` on a
+  pre-2011 kernel.
+
+### Fixed ###
+- Several improvements were made to the errors returned by `Open(at)InRoot` and
+  `MkdirAll` when dealing with invalid paths under the emulated (ie.
+  non-`openat2`) implementation. Previously, some paths would return the wrong
+  error (`ENOENT` when the last component was a non-directory), and other paths
+  would be returned as though they were acceptable (trailing-slash components
+  after a non-directory would be ignored by `Open(at)InRoot`).
+
+  These changes were done to match `openat2`'s behaviour and purely is a
+  consistency fix (most users are going to be using `openat2` anyway).
+
+[linux-readlinkat-emptypath]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=65cfc6722361570bfe255698d9cd4dccaf47570d
+
+## [0.3.0] - 2024-07-11 ##
+
+### Added ###
+- A new set of `*os.File`-based APIs have been added. These are adapted from
+  [libpathrs][] and we strongly suggest using them if possible (as they provide
+  far more protection against attacks than `SecureJoin`):
+
+   - `Open(at)InRoot` resolves a path inside a rootfs and returns an `*os.File`
+     handle to the path. Note that the handle returned is an `O_PATH` handle,
+     which cannot be used for reading or writing (as well as some other
+     operations -- [see open(2) for more details][open.2])
+
+   - `Reopen` takes an `O_PATH` file handle and safely re-opens it to upgrade
+     it to a regular handle. This can also be used with non-`O_PATH` handles,
+     but `O_PATH` is the most obvious application.
+
+   - `MkdirAll` is an implementation of `os.MkdirAll` that is safe to use to
+     create a directory tree within a rootfs.
+
+  As these are new APIs, they may change in the future. However, they should be
+  safe to start migrating to as we have extensive tests ensuring they behave
+  correctly and are safe against various races and other attacks.
+
+[libpathrs]: https://github.com/openSUSE/libpathrs
+[open.2]: https://www.man7.org/linux/man-pages/man2/open.2.html
+
+## [0.2.5] - 2024-05-03 ##
+
+### Changed ###
+- Some minor changes were made to how lexical components (like `..` and `.`)
+  are handled during path generation in `SecureJoin`. There is no behaviour
+  change as a result of this fix (the resulting paths are the same).
+
+### Fixed ###
+- The error returned when we hit a symlink loop now references the correct
+  path. (#10)
+
+## [0.2.4] - 2023-09-06 ##
+
+### Security ###
+- This release fixes a potential security issue in filepath-securejoin when
+  used on Windows ([GHSA-6xv5-86q9-7xr8][], which could be used to generate
+  paths outside of the provided rootfs in certain cases), as well as improving
+  the overall behaviour of filepath-securejoin when dealing with Windows paths
+  that contain volume names. Thanks to Paulo Gomes for discovering and fixing
+  these issues.
+
+### Fixed ###
+- Switch to GitHub Actions for CI so we can test on Windows as well as Linux
+  and MacOS.
+
+[GHSA-6xv5-86q9-7xr8]: https://github.com/advisories/GHSA-6xv5-86q9-7xr8
+
+## [0.2.3] - 2021-06-04 ##
+
+### Changed ###
+- Switch to Go 1.13-style `%w` error wrapping, letting us drop the dependency
+  on `github.com/pkg/errors`.
+
+## [0.2.2] - 2018-09-05 ##
+
+### Changed ###
+- Use `syscall.ELOOP` as the base error for symlink loops, rather than our own
+  (internal) error. This allows callers to more easily use `errors.Is` to check
+  for this case.
+
+## [0.2.1] - 2018-09-05 ##
+
+### Fixed ###
+- Use our own `IsNotExist` implementation, which lets us handle `ENOTDIR`
+  properly within `SecureJoin`.
+
+## [0.2.0] - 2017-07-19 ##
+
+We now have 100% test coverage!
+
+### Added ###
+- Add a `SecureJoinVFS` API that can be used for mocking (as we do in our new
+  tests) or for implementing custom handling of lookup operations (such as for
+  rootless containers, where work is necessary to access directories with weird
+  modes because we don't have `CAP_DAC_READ_SEARCH` or `CAP_DAC_OVERRIDE`).
+
+## 0.1.0 - 2017-07-19
+
+This is our first release of `github.com/cyphar/filepath-securejoin`,
+containing a full implementation with a coverage of 93.5% (the only missing
+cases are the error cases, which are hard to mocktest at the moment).
+
+[Unreleased]: https://github.com/cyphar/filepath-securejoin/compare/v0.3.4...HEAD
+[0.3.3]: https://github.com/cyphar/filepath-securejoin/compare/v0.3.3...v0.3.4
+[0.3.3]: https://github.com/cyphar/filepath-securejoin/compare/v0.3.2...v0.3.3
+[0.3.2]: https://github.com/cyphar/filepath-securejoin/compare/v0.3.1...v0.3.2
+[0.3.1]: https://github.com/cyphar/filepath-securejoin/compare/v0.3.0...v0.3.1
+[0.3.0]: https://github.com/cyphar/filepath-securejoin/compare/v0.2.5...v0.3.0
+[0.2.5]: https://github.com/cyphar/filepath-securejoin/compare/v0.2.4...v0.2.5
+[0.2.4]: https://github.com/cyphar/filepath-securejoin/compare/v0.2.3...v0.2.4
+[0.2.3]: https://github.com/cyphar/filepath-securejoin/compare/v0.2.2...v0.2.3
+[0.2.2]: https://github.com/cyphar/filepath-securejoin/compare/v0.2.1...v0.2.2
+[0.2.1]: https://github.com/cyphar/filepath-securejoin/compare/v0.2.0...v0.2.1
+[0.2.0]: https://github.com/cyphar/filepath-securejoin/compare/v0.1.0...v0.2.0

vendor/github.com/cyphar/filepath-securejoin/LICENSE

@@ -1,5 +1,5 @@
 Copyright (C) 2014-2015 Docker Inc & Go Authors. All rights reserved.
-Copyright (C) 2017 SUSE LLC. All rights reserved.
+Copyright (C) 2017-2024 SUSE LLC. All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are

vendor/github.com/cyphar/filepath-securejoin/README.md

@@ -1,32 +1,26 @@
 ## `filepath-securejoin` ##
 
+[![Go Documentation](https://pkg.go.dev/badge/github.com/cyphar/filepath-securejoin.svg)](https://pkg.go.dev/github.com/cyphar/filepath-securejoin)
 [![Build Status](https://github.com/cyphar/filepath-securejoin/actions/workflows/ci.yml/badge.svg)](https://github.com/cyphar/filepath-securejoin/actions/workflows/ci.yml)
 
-An implementation of `SecureJoin`, a [candidate for inclusion in the Go
+### Old API ###
-standard library][go#20126]. The purpose of this function is to be a "secure"
-alternative to `filepath.Join`, and in particular it provides certain
-guarantees that are not provided by `filepath.Join`.
 
-> **NOTE**: This code is *only* safe if you are not at risk of other processes
+This library was originally just an implementation of `SecureJoin` which was
-> modifying path components after you've used `SecureJoin`. If it is possible
+[intended to be included in the Go standard library][go#20126] as a safer
-> for a malicious process to modify path components of the resolved path, then
+`filepath.Join` that would restrict the path lookup to be inside a root
-> you will be vulnerable to some fairly trivial TOCTOU race conditions. [There
+directory.
-> are some Linux kernel patches I'm working on which might allow for a better
-> solution.][lwn-obeneath]
->
-> In addition, with a slightly modified API it might be possible to use
-> `O_PATH` and verify that the opened path is actually the resolved one -- but
-> I have not done that yet. I might add it in the future as a helper function
-> to help users verify the path (we can't just return `/proc/self/fd/<foo>`
-> because that doesn't always work transparently for all users).
 
-This is the function prototype:
+The implementation was based on code that existed in several container
+runtimes. Unfortunately, this API is **fundamentally unsafe** against attackers
+that can modify path components after `SecureJoin` returns and before the
+caller uses the path, allowing for some fairly trivial TOCTOU attacks.
 
-```go
+`SecureJoin` (and `SecureJoinVFS`) are still provided by this library to
-func SecureJoin(root, unsafePath string) (string, error)
+support legacy users, but new users are strongly suggested to avoid using
-```
+`SecureJoin` and instead use the [new api](#new-api) or switch to
+[libpathrs][libpathrs].
 
-This library **guarantees** the following:
+With the above limitations in mind, this library guarantees the following:
 
 * If no error is set, the resulting string **must** be a child path of
   `root` and will not contain any symlink path components (they will all be
@@ -47,7 +41,7 @@ This library **guarantees** the following:
 A (trivial) implementation of this function on GNU/Linux systems could be done
 with the following (note that this requires root privileges and is far more
 opaque than the implementation in this library, and also requires that
-`readlink` is inside the `root` path):
+`readlink` is inside the `root` path and is trustworthy):
 
 ```go
 package securejoin
@@ -70,9 +64,105 @@ func SecureJoin(root, unsafePath string) (string, error) {
 }
 ```
 
-[lwn-obeneath]: https://lwn.net/Articles/767547/
+[libpathrs]: https://github.com/openSUSE/libpathrs
 [go#20126]: https://github.com/golang/go/issues/20126
 
+### New API ###
+
+While we recommend users switch to [libpathrs][libpathrs] as soon as it has a
+stable release, some methods implemented by libpathrs have been ported to this
+library to ease the transition. These APIs are only supported on Linux.
+
+These APIs are implemented such that `filepath-securejoin` will
+opportunistically use certain newer kernel APIs that make these operations far
+more secure. In particular:
+
+* All of the lookup operations will use [`openat2`][openat2.2] on new enough
+  kernels (Linux 5.6 or later) to restrict lookups through magic-links and
+  bind-mounts (for certain operations) and to make use of `RESOLVE_IN_ROOT` to
+  efficiently resolve symlinks within a rootfs.
+
+* The APIs provide hardening against a malicious `/proc` mount to either detect
+  or avoid being tricked by a `/proc` that is not legitimate. This is done
+  using [`openat2`][openat2.2] for all users, and privileged users will also be
+  further protected by using [`fsopen`][fsopen.2] and [`open_tree`][open_tree.2]
+  (Linux 5.2 or later).
+
+[openat2.2]: https://www.man7.org/linux/man-pages/man2/openat2.2.html
+[fsopen.2]: https://github.com/brauner/man-pages-md/blob/main/fsopen.md
+[open_tree.2]: https://github.com/brauner/man-pages-md/blob/main/open_tree.md
+
+#### `OpenInRoot` ####
+
+```go
+func OpenInRoot(root, unsafePath string) (*os.File, error)
+func OpenatInRoot(root *os.File, unsafePath string) (*os.File, error)
+func Reopen(handle *os.File, flags int) (*os.File, error)
+```
+
+`OpenInRoot` is a much safer version of
+
+```go
+path, err := securejoin.SecureJoin(root, unsafePath)
+file, err := os.OpenFile(path, unix.O_PATH|unix.O_CLOEXEC)
+```
+
+that protects against various race attacks that could lead to serious security
+issues, depending on the application. Note that the returned `*os.File` is an
+`O_PATH` file descriptor, which is quite restricted. Callers will probably need
+to use `Reopen` to get a more usable handle (this split is done to provide
+useful features like PTY spawning and to avoid users accidentally opening bad
+inodes that could cause a DoS).
+
+Callers need to be careful in how they use the returned `*os.File`. Usually it
+is only safe to operate on the handle directly, and it is very easy to create a
+security issue. [libpathrs][libpathrs] provides far more helpers to make using
+these handles safer -- there is currently no plan to port them to
+`filepath-securejoin`.
+
+`OpenatInRoot` is like `OpenInRoot` except that the root is provided using an
+`*os.File`. This allows you to ensure that multiple `OpenatInRoot` (or
+`MkdirAllHandle`) calls are operating on the same rootfs.
+
+> **NOTE**: Unlike `SecureJoin`, `OpenInRoot` will error out as soon as it hits
+> a dangling symlink or non-existent path. This is in contrast to `SecureJoin`
+> which treated non-existent components as though they were real directories,
+> and would allow for partial resolution of dangling symlinks. These behaviours
+> are at odds with how Linux treats non-existent paths and dangling symlinks,
+> and so these are no longer allowed.
+
+#### `MkdirAll` ####
+
+```go
+func MkdirAll(root, unsafePath string, mode int) error
+func MkdirAllHandle(root *os.File, unsafePath string, mode int) (*os.File, error)
+```
+
+`MkdirAll` is a much safer version of
+
+```go
+path, err := securejoin.SecureJoin(root, unsafePath)
+err = os.MkdirAll(path, mode)
+```
+
+that protects against the same kinds of races that `OpenInRoot` protects
+against.
+
+`MkdirAllHandle` is like `MkdirAll` except that the root is provided using an
+`*os.File` (the reason for this is the same as with `OpenatInRoot`) and an
+`*os.File` of the final created directory is returned (this directory is
+guaranteed to be effectively identical to the directory created by
+`MkdirAllHandle`, which is not possible to ensure by just using `OpenatInRoot`
+after `MkdirAll`).
+
+> **NOTE**: Unlike `SecureJoin`, `MkdirAll` will error out as soon as it hits
+> a dangling symlink or non-existent path. This is in contrast to `SecureJoin`
+> which treated non-existent components as though they were real directories,
+> and would allow for partial resolution of dangling symlinks. These behaviours
+> are at odds with how Linux treats non-existent paths and dangling symlinks,
+> and so these are no longer allowed. This means that `MkdirAll` will not
+> create non-existent directories referenced by a dangling symlink.
+
 ### License ###
 
 The license of this project is the same as Go, which is a BSD 3-clause license

vendor/github.com/cyphar/filepath-securejoin/VERSION

@@ -1 +1 @@
-0.2.4
+0.3.4

vendor/github.com/cyphar/filepath-securejoin/doc.go

@@ -0,0 +1,39 @@
+// Copyright (C) 2014-2015 Docker Inc & Go Authors. All rights reserved.
+// Copyright (C) 2017-2024 SUSE LLC. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package securejoin implements a set of helpers to make it easier to write Go
+// code that is safe against symlink-related escape attacks. The primary idea
+// is to let you resolve a path within a rootfs directory as if the rootfs was
+// a chroot.
+//
+// securejoin has two APIs, a "legacy" API and a "modern" API.
+//
+// The legacy API is [SecureJoin] and [SecureJoinVFS]. These methods are
+// **not** safe against race conditions where an attacker changes the
+// filesystem after (or during) the [SecureJoin] operation.
+//
+// The new API is made up of [OpenInRoot] and [MkdirAll] (and derived
+// functions). These are safe against racing attackers and have several other
+// protections that are not provided by the legacy API. There are many more
+// operations that most programs expect to be able to do safely, but we do not
+// provide explicit support for them because we want to encourage users to
+// switch to [libpathrs](https://github.com/openSUSE/libpathrs) which is a
+// cross-language next-generation library that is entirely designed around
+// operating on paths safely.
+//
+// securejoin has been used by several container runtimes (Docker, runc,
+// Kubernetes, etc) for quite a few years as a de-facto standard for operating
+// on container filesystem paths "safely". However, most users still use the
+// legacy API which is unsafe against various attacks (there is a fairly long
+// history of CVEs in dependent as a result). Users should switch to the modern
+// API as soon as possible (or even better, switch to libpathrs).
+//
+// This project was initially intended to be included in the Go standard
+// library, but [it was rejected](https://go.dev/issue/20126). There is now a
+// [new Go proposal](https://go.dev/issue/67002) for a safe path resolution API
+// that shares some of the goals of filepath-securejoin. However, that design
+// is intended to work like `openat2(RESOLVE_BENEATH)` which does not fit the
+// usecase of container runtimes and most system tools.
+package securejoin

vendor/github.com/cyphar/filepath-securejoin/join.go

@@ -1,17 +1,11 @@
 // Copyright (C) 2014-2015 Docker Inc & Go Authors. All rights reserved.
-// Copyright (C) 2017 SUSE LLC. All rights reserved.
+// Copyright (C) 2017-2024 SUSE LLC. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// Package securejoin is an implementation of the hopefully-soon-to-be-included
-// SecureJoin helper that is meant to be part of the "path/filepath" package.
-// The purpose of this project is to provide a PoC implementation to make the
-// SecureJoin proposal (https://github.com/golang/go/issues/20126) more
-// tangible.
 package securejoin
 
 import (
-	"bytes"
 	"errors"
 	"os"
 	"path/filepath"
@@ -19,26 +13,34 @@ import (
 	"syscall"
 )
 
+const maxSymlinkLimit = 255
+
 // IsNotExist tells you if err is an error that implies that either the path
 // accessed does not exist (or path components don't exist). This is
-// effectively a more broad version of os.IsNotExist.
+// effectively a more broad version of [os.IsNotExist].
 func IsNotExist(err error) bool {
 	// Check that it's not actually an ENOTDIR, which in some cases is a more
 	// convoluted case of ENOENT (usually involving weird paths).
 	return errors.Is(err, os.ErrNotExist) || errors.Is(err, syscall.ENOTDIR) || errors.Is(err, syscall.ENOENT)
 }
 
-// SecureJoinVFS joins the two given path components (similar to Join) except
+// SecureJoinVFS joins the two given path components (similar to [filepath.Join]) except
 // that the returned path is guaranteed to be scoped inside the provided root
 // path (when evaluated). Any symbolic links in the path are evaluated with the
 // given root treated as the root of the filesystem, similar to a chroot. The
-// filesystem state is evaluated through the given VFS interface (if nil, the
+// filesystem state is evaluated through the given [VFS] interface (if nil, the
-// standard os.* family of functions are used).
+// standard [os].* family of functions are used).
 //
 // Note that the guarantees provided by this function only apply if the path
 // components in the returned string are not modified (in other words are not
 // replaced with symlinks on the filesystem) after this function has returned.
-// Such a symlink race is necessarily out-of-scope of SecureJoin.
+// Such a symlink race is necessarily out-of-scope of SecureJoinVFS.
+//
+// NOTE: Due to the above limitation, Linux users are strongly encouraged to
+// use [OpenInRoot] instead, which does safely protect against these kinds of
+// attacks. There is no way to solve this problem with SecureJoinVFS because
+// the API is fundamentally wrong (you cannot return a "safe" path string and
+// guarantee it won't be modified afterwards).
 //
 // Volume names in unsafePath are always discarded, regardless if they are
 // provided via direct input or when evaluating symlinks. Therefore:
@@ -51,75 +53,73 @@ func SecureJoinVFS(root, unsafePath string, vfs VFS) (string, error) {
 	}
 
 	unsafePath = filepath.FromSlash(unsafePath)
-	var path bytes.Buffer
+	var (
-	n := 0
+		currentPath   string
-	for unsafePath != "" {
+		remainingPath = unsafePath
-		if n > 255 {
+		linksWalked   int
-			return "", &os.PathError{Op: "SecureJoin", Path: root + string(filepath.Separator) + unsafePath, Err: syscall.ELOOP}
+	)
+	for remainingPath != "" {
+		if v := filepath.VolumeName(remainingPath); v != "" {
+			remainingPath = remainingPath[len(v):]
 		}
 
-		if v := filepath.VolumeName(unsafePath); v != "" {
+		// Get the next path component.
-			unsafePath = unsafePath[len(v):]
+		var part string
-		}
+		if i := strings.IndexRune(remainingPath, filepath.Separator); i == -1 {
-
+			part, remainingPath = remainingPath, ""
-		// Next path component, p.
-		i := strings.IndexRune(unsafePath, filepath.Separator)
-		var p string
-		if i == -1 {
-			p, unsafePath = unsafePath, ""
 		} else {
-			p, unsafePath = unsafePath[:i], unsafePath[i+1:]
+			part, remainingPath = remainingPath[:i], remainingPath[i+1:]
 		}
 
-		// Create a cleaned path, using the lexical semantics of /../a, to
+		// Apply the component lexically to the path we are building.
-		// create a "scoped" path component which can safely be joined to fullP
+		// currentPath does not contain any symlinks, and we are lexically
-		// for evaluation. At this point, path.String() doesn't contain any
+		// dealing with a single component, so it's okay to do a filepath.Clean
-		// symlink components.
+		// here.
-		cleanP := filepath.Clean(string(filepath.Separator) + path.String() + p)
+		nextPath := filepath.Join(string(filepath.Separator), currentPath, part)
-		if cleanP == string(filepath.Separator) {
+		if nextPath == string(filepath.Separator) {
-			path.Reset()
+			currentPath = ""
 			continue
 		}
-		fullP := filepath.Clean(root + cleanP)
+		fullPath := root + string(filepath.Separator) + nextPath
 
 		// Figure out whether the path is a symlink.
-		fi, err := vfs.Lstat(fullP)
+		fi, err := vfs.Lstat(fullPath)
 		if err != nil && !IsNotExist(err) {
 			return "", err
 		}
 		// Treat non-existent path components the same as non-symlinks (we
 		// can't do any better here).
 		if IsNotExist(err) || fi.Mode()&os.ModeSymlink == 0 {
-			path.WriteString(p)
+			currentPath = nextPath
-			path.WriteRune(filepath.Separator)
 			continue
 		}
 
-		// Only increment when we actually dereference a link.
+		// It's a symlink, so get its contents and expand it by prepending it
-		n++
+		// to the yet-unparsed path.
+		linksWalked++
+		if linksWalked > maxSymlinkLimit {
+			return "", &os.PathError{Op: "SecureJoin", Path: root + string(filepath.Separator) + unsafePath, Err: syscall.ELOOP}
+		}
 
-		// It's a symlink, expand it by prepending it to the yet-unparsed path.
+		dest, err := vfs.Readlink(fullPath)
-		dest, err := vfs.Readlink(fullP)
 		if err != nil {
 			return "", err
 		}
+		remainingPath = dest + string(filepath.Separator) + remainingPath
 		// Absolute symlinks reset any work we've already done.
 		if filepath.IsAbs(dest) {
-			path.Reset()
+			currentPath = ""
 		}
-		unsafePath = dest + string(filepath.Separator) + unsafePath
 	}
 
-	// We have to clean path.String() here because it may contain '..'
+	// There should be no lexical components like ".." left in the path here,
-	// components that are entirely lexical, but would be misleading otherwise.
+	// but for safety clean up the path before joining it to the root.
-	// And finally do a final clean to ensure that root is also lexically
+	finalPath := filepath.Join(string(filepath.Separator), currentPath)
-	// clean.
+	return filepath.Join(root, finalPath), nil
-	fullP := filepath.Clean(string(filepath.Separator) + path.String())
-	return filepath.Clean(root + fullP), nil
 }
 
-// SecureJoin is a wrapper around SecureJoinVFS that just uses the os.* library
+// SecureJoin is a wrapper around [SecureJoinVFS] that just uses the [os].* library
-// of functions as the VFS. If in doubt, use this function over SecureJoinVFS.
+// of functions as the [VFS]. If in doubt, use this function over [SecureJoinVFS].
 func SecureJoin(root, unsafePath string) (string, error) {
 	return SecureJoinVFS(root, unsafePath, nil)
 }

vendor/github.com/cyphar/filepath-securejoin/lookup_linux.go

@@ -0,0 +1,389 @@
+//go:build linux
+
+// Copyright (C) 2024 SUSE LLC. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package securejoin
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path"
+	"path/filepath"
+	"slices"
+	"strings"
+
+	"golang.org/x/sys/unix"
+)
+
+type symlinkStackEntry struct {
+	// (dir, remainingPath) is what we would've returned if the link didn't
+	// exist. This matches what openat2(RESOLVE_IN_ROOT) would return in
+	// this case.
+	dir           *os.File
+	remainingPath string
+	// linkUnwalked is the remaining path components from the original
+	// Readlink which we have yet to walk. When this slice is empty, we
+	// drop the link from the stack.
+	linkUnwalked []string
+}
+
+func (se symlinkStackEntry) String() string {
+	return fmt.Sprintf("<%s>/%s [->%s]", se.dir.Name(), se.remainingPath, strings.Join(se.linkUnwalked, "/"))
+}
+
+func (se symlinkStackEntry) Close() {
+	_ = se.dir.Close()
+}
+
+type symlinkStack []*symlinkStackEntry
+
+func (s *symlinkStack) IsEmpty() bool {
+	return s == nil || len(*s) == 0
+}
+
+func (s *symlinkStack) Close() {
+	if s != nil {
+		for _, link := range *s {
+			link.Close()
+		}
+		// TODO: Switch to clear once we switch to Go 1.21.
+		*s = nil
+	}
+}
+
+var (
+	errEmptyStack         = errors.New("[internal] stack is empty")
+	errBrokenSymlinkStack = errors.New("[internal error] broken symlink stack")
+)
+
+func (s *symlinkStack) popPart(part string) error {
+	if s == nil || s.IsEmpty() {
+		// If there is nothing in the symlink stack, then the part was from the
+		// real path provided by the user, and this is a no-op.
+		return errEmptyStack
+	}
+	if part == "." {
+		// "." components are no-ops -- we drop them when doing SwapLink.
+		return nil
+	}
+
+	tailEntry := (*s)[len(*s)-1]
+
+	// Double-check that we are popping the component we expect.
+	if len(tailEntry.linkUnwalked) == 0 {
+		return fmt.Errorf("%w: trying to pop component %q of empty stack entry %s", errBrokenSymlinkStack, part, tailEntry)
+	}
+	headPart := tailEntry.linkUnwalked[0]
+	if headPart != part {
+		return fmt.Errorf("%w: trying to pop component %q but the last stack entry is %s (%q)", errBrokenSymlinkStack, part, tailEntry, headPart)
+	}
+
+	// Drop the component, but keep the entry around in case we are dealing
+	// with a "tail-chained" symlink.
+	tailEntry.linkUnwalked = tailEntry.linkUnwalked[1:]
+	return nil
+}
+
+func (s *symlinkStack) PopPart(part string) error {
+	if err := s.popPart(part); err != nil {
+		if errors.Is(err, errEmptyStack) {
+			// Skip empty stacks.
+			err = nil
+		}
+		return err
+	}
+
+	// Clean up any of the trailing stack entries that are empty.
+	for lastGood := len(*s) - 1; lastGood >= 0; lastGood-- {
+		entry := (*s)[lastGood]
+		if len(entry.linkUnwalked) > 0 {
+			break
+		}
+		entry.Close()
+		(*s) = (*s)[:lastGood]
+	}
+	return nil
+}
+
+func (s *symlinkStack) push(dir *os.File, remainingPath, linkTarget string) error {
+	if s == nil {
+		return nil
+	}
+	// Split the link target and clean up any "" parts.
+	linkTargetParts := slices.DeleteFunc(
+		strings.Split(linkTarget, "/"),
+		func(part string) bool { return part == "" || part == "." })
+
+	// Copy the directory so the caller doesn't close our copy.
+	dirCopy, err := dupFile(dir)
+	if err != nil {
+		return err
+	}
+
+	// Add to the stack.
+	*s = append(*s, &symlinkStackEntry{
+		dir:           dirCopy,
+		remainingPath: remainingPath,
+		linkUnwalked:  linkTargetParts,
+	})
+	return nil
+}
+
+func (s *symlinkStack) SwapLink(linkPart string, dir *os.File, remainingPath, linkTarget string) error {
+	// If we are currently inside a symlink resolution, remove the symlink
+	// component from the last symlink entry, but don't remove the entry even
+	// if it's empty. If we are a "tail-chained" symlink (a trailing symlink we
+	// hit during a symlink resolution) we need to keep the old symlink until
+	// we finish the resolution.
+	if err := s.popPart(linkPart); err != nil {
+		if !errors.Is(err, errEmptyStack) {
+			return err
+		}
+		// Push the component regardless of whether the stack was empty.
+	}
+	return s.push(dir, remainingPath, linkTarget)
+}
+
+func (s *symlinkStack) PopTopSymlink() (*os.File, string, bool) {
+	if s == nil || s.IsEmpty() {
+		return nil, "", false
+	}
+	tailEntry := (*s)[0]
+	*s = (*s)[1:]
+	return tailEntry.dir, tailEntry.remainingPath, true
+}
+
+// partialLookupInRoot tries to lookup as much of the request path as possible
+// within the provided root (a-la RESOLVE_IN_ROOT) and opens the final existing
+// component of the requested path, returning a file handle to the final
+// existing component and a string containing the remaining path components.
+func partialLookupInRoot(root *os.File, unsafePath string) (*os.File, string, error) {
+	return lookupInRoot(root, unsafePath, true)
+}
+
+func completeLookupInRoot(root *os.File, unsafePath string) (*os.File, error) {
+	handle, remainingPath, err := lookupInRoot(root, unsafePath, false)
+	if remainingPath != "" && err == nil {
+		// should never happen
+		err = fmt.Errorf("[bug] non-empty remaining path when doing a non-partial lookup: %q", remainingPath)
+	}
+	// lookupInRoot(partial=false) will always close the handle if an error is
+	// returned, so no need to double-check here.
+	return handle, err
+}
+
+func lookupInRoot(root *os.File, unsafePath string, partial bool) (Handle *os.File, _ string, _ error) {
+	unsafePath = filepath.ToSlash(unsafePath) // noop
+
+	// This is very similar to SecureJoin, except that we operate on the
+	// components using file descriptors. We then return the last component we
+	// managed open, along with the remaining path components not opened.
+
+	// Try to use openat2 if possible.
+	if hasOpenat2() {
+		return lookupOpenat2(root, unsafePath, partial)
+	}
+
+	// Get the "actual" root path from /proc/self/fd. This is necessary if the
+	// root is some magic-link like /proc/$pid/root, in which case we want to
+	// make sure when we do checkProcSelfFdPath that we are using the correct
+	// root path.
+	logicalRootPath, err := procSelfFdReadlink(root)
+	if err != nil {
+		return nil, "", fmt.Errorf("get real root path: %w", err)
+	}
+
+	currentDir, err := dupFile(root)
+	if err != nil {
+		return nil, "", fmt.Errorf("clone root fd: %w", err)
+	}
+	defer func() {
+		// If a handle is not returned, close the internal handle.
+		if Handle == nil {
+			_ = currentDir.Close()
+		}
+	}()
+
+	// symlinkStack is used to emulate how openat2(RESOLVE_IN_ROOT) treats
+	// dangling symlinks. If we hit a non-existent path while resolving a
+	// symlink, we need to return the (dir, remainingPath) that we had when we
+	// hit the symlink (treating the symlink as though it were a regular file).
+	// The set of (dir, remainingPath) sets is stored within the symlinkStack
+	// and we add and remove parts when we hit symlink and non-symlink
+	// components respectively. We need a stack because of recursive symlinks
+	// (symlinks that contain symlink components in their target).
+	//
+	// Note that the stack is ONLY used for book-keeping. All of the actual
+	// path walking logic is still based on currentPath/remainingPath and
+	// currentDir (as in SecureJoin).
+	var symStack *symlinkStack
+	if partial {
+		symStack = new(symlinkStack)
+		defer symStack.Close()
+	}
+
+	var (
+		linksWalked   int
+		currentPath   string
+		remainingPath = unsafePath
+	)
+	for remainingPath != "" {
+		// Save the current remaining path so if the part is not real we can
+		// return the path including the component.
+		oldRemainingPath := remainingPath
+
+		// Get the next path component.
+		var part string
+		if i := strings.IndexByte(remainingPath, '/'); i == -1 {
+			part, remainingPath = remainingPath, ""
+		} else {
+			part, remainingPath = remainingPath[:i], remainingPath[i+1:]
+		}
+		// If we hit an empty component, we need to treat it as though it is
+		// "." so that trailing "/" and "//" components on a non-directory
+		// correctly return the right error code.
+		if part == "" {
+			part = "."
+		}
+
+		// Apply the component lexically to the path we are building.
+		// currentPath does not contain any symlinks, and we are lexically
+		// dealing with a single component, so it's okay to do a filepath.Clean
+		// here.
+		nextPath := path.Join("/", currentPath, part)
+		// If we logically hit the root, just clone the root rather than
+		// opening the part and doing all of the other checks.
+		if nextPath == "/" {
+			if err := symStack.PopPart(part); err != nil {
+				return nil, "", fmt.Errorf("walking into root with part %q failed: %w", part, err)
+			}
+			// Jump to root.
+			rootClone, err := dupFile(root)
+			if err != nil {
+				return nil, "", fmt.Errorf("clone root fd: %w", err)
+			}
+			_ = currentDir.Close()
+			currentDir = rootClone
+			currentPath = nextPath
+			continue
+		}
+
+		// Try to open the next component.
+		nextDir, err := openatFile(currentDir, part, unix.O_PATH|unix.O_NOFOLLOW|unix.O_CLOEXEC, 0)
+		switch {
+		case err == nil:
+			st, err := nextDir.Stat()
+			if err != nil {
+				_ = nextDir.Close()
+				return nil, "", fmt.Errorf("stat component %q: %w", part, err)
+			}
+
+			switch st.Mode() & os.ModeType {
+			case os.ModeSymlink:
+				// readlinkat implies AT_EMPTY_PATH since Linux 2.6.39. See
+				// Linux commit 65cfc6722361 ("readlinkat(), fchownat() and
+				// fstatat() with empty relative pathnames").
+				linkDest, err := readlinkatFile(nextDir, "")
+				// We don't need the handle anymore.
+				_ = nextDir.Close()
+				if err != nil {
+					return nil, "", err
+				}
+
+				linksWalked++
+				if linksWalked > maxSymlinkLimit {
+					return nil, "", &os.PathError{Op: "securejoin.lookupInRoot", Path: logicalRootPath + "/" + unsafePath, Err: unix.ELOOP}
+				}
+
+				// Swap out the symlink's component for the link entry itself.
+				if err := symStack.SwapLink(part, currentDir, oldRemainingPath, linkDest); err != nil {
+					return nil, "", fmt.Errorf("walking into symlink %q failed: push symlink: %w", part, err)
+				}
+
+				// Update our logical remaining path.
+				remainingPath = linkDest + "/" + remainingPath
+				// Absolute symlinks reset any work we've already done.
+				if path.IsAbs(linkDest) {
+					// Jump to root.
+					rootClone, err := dupFile(root)
+					if err != nil {
+						return nil, "", fmt.Errorf("clone root fd: %w", err)
+					}
+					_ = currentDir.Close()
+					currentDir = rootClone
+					currentPath = "/"
+				}
+
+			default:
+				// If we are dealing with a directory, simply walk into it.
+				_ = currentDir.Close()
+				currentDir = nextDir
+				currentPath = nextPath
+
+				// The part was real, so drop it from the symlink stack.
+				if err := symStack.PopPart(part); err != nil {
+					return nil, "", fmt.Errorf("walking into directory %q failed: %w", part, err)
+				}
+
+				// If we are operating on a .., make sure we haven't escaped.
+				// We only have to check for ".." here because walking down
+				// into a regular component component cannot cause you to
+				// escape. This mirrors the logic in RESOLVE_IN_ROOT, except we
+				// have to check every ".." rather than only checking after a
+				// rename or mount on the system.
+				if part == ".." {
+					// Make sure the root hasn't moved.
+					if err := checkProcSelfFdPath(logicalRootPath, root); err != nil {
+						return nil, "", fmt.Errorf("root path moved during lookup: %w", err)
+					}
+					// Make sure the path is what we expect.
+					fullPath := logicalRootPath + nextPath
+					if err := checkProcSelfFdPath(fullPath, currentDir); err != nil {
+						return nil, "", fmt.Errorf("walking into %q had unexpected result: %w", part, err)
+					}
+				}
+			}
+
+		default:
+			if !partial {
+				return nil, "", err
+			}
+			// If there are any remaining components in the symlink stack, we
+			// are still within a symlink resolution and thus we hit a dangling
+			// symlink. So pretend that the first symlink in the stack we hit
+			// was an ENOENT (to match openat2).
+			if oldDir, remainingPath, ok := symStack.PopTopSymlink(); ok {
+				_ = currentDir.Close()
+				return oldDir, remainingPath, err
+			}
+			// We have hit a final component that doesn't exist, so we have our
+			// partial open result. Note that we have to use the OLD remaining
+			// path, since the lookup failed.
+			return currentDir, oldRemainingPath, err
+		}
+	}
+
+	// If the unsafePath had a trailing slash, we need to make sure we try to
+	// do a relative "." open so that we will correctly return an error when
+	// the final component is a non-directory (to match openat2). In the
+	// context of openat2, a trailing slash and a trailing "/." are completely
+	// equivalent.
+	if strings.HasSuffix(unsafePath, "/") {
+		nextDir, err := openatFile(currentDir, ".", unix.O_PATH|unix.O_NOFOLLOW|unix.O_CLOEXEC, 0)
+		if err != nil {
+			if !partial {
+				_ = currentDir.Close()
+				currentDir = nil
+			}
+			return currentDir, "", err
+		}
+		_ = currentDir.Close()
+		currentDir = nextDir
+	}
+
+	// All of the components existed!
+	return currentDir, "", nil
+}

vendor/github.com/cyphar/filepath-securejoin/mkdir_linux.go

@@ -0,0 +1,207 @@
+//go:build linux
+
+// Copyright (C) 2024 SUSE LLC. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package securejoin
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+
+	"golang.org/x/sys/unix"
+)
+
+var (
+	errInvalidMode    = errors.New("invalid permission mode")
+	errPossibleAttack = errors.New("possible attack detected")
+)
+
+// MkdirAllHandle is equivalent to [MkdirAll], except that it is safer to use
+// in two respects:
+//
+//   - The caller provides the root directory as an *[os.File] (preferably O_PATH)
+//     handle. This means that the caller can be sure which root directory is
+//     being used. Note that this can be emulated by using /proc/self/fd/... as
+//     the root path with [os.MkdirAll].
+//
+//   - Once all of the directories have been created, an *[os.File] O_PATH handle
+//     to the directory at unsafePath is returned to the caller. This is done in
+//     an effectively-race-free way (an attacker would only be able to swap the
+//     final directory component), which is not possible to emulate with
+//     [MkdirAll].
+//
+// In addition, the returned handle is obtained far more efficiently than doing
+// a brand new lookup of unsafePath (such as with [SecureJoin] or openat2) after
+// doing [MkdirAll]. If you intend to open the directory after creating it, you
+// should use MkdirAllHandle.
+func MkdirAllHandle(root *os.File, unsafePath string, mode int) (_ *os.File, Err error) {
+	// Make sure there are no os.FileMode bits set.
+	if mode&^0o7777 != 0 {
+		return nil, fmt.Errorf("%w for mkdir 0o%.3o", errInvalidMode, mode)
+	}
+	// On Linux, mkdirat(2) (and os.Mkdir) silently ignore the suid and sgid
+	// bits. We could also silently ignore them but since we have very few
+	// users it seems more prudent to return an error so users notice that
+	// these bits will not be set.
+	if mode&^0o1777 != 0 {
+		return nil, fmt.Errorf("%w for mkdir 0o%.3o: suid and sgid are ignored by mkdir", errInvalidMode, mode)
+	}
+
+	// Try to open as much of the path as possible.
+	currentDir, remainingPath, err := partialLookupInRoot(root, unsafePath)
+	defer func() {
+		if Err != nil {
+			_ = currentDir.Close()
+		}
+	}()
+	if err != nil && !errors.Is(err, unix.ENOENT) {
+		return nil, fmt.Errorf("find existing subpath of %q: %w", unsafePath, err)
+	}
+
+	// If there is an attacker deleting directories as we walk into them,
+	// detect this proactively. Note this is guaranteed to detect if the
+	// attacker deleted any part of the tree up to currentDir.
+	//
+	// Once we walk into a dead directory, partialLookupInRoot would not be
+	// able to walk further down the tree (directories must be empty before
+	// they are deleted), and if the attacker has removed the entire tree we
+	// can be sure that anything that was originally inside a dead directory
+	// must also be deleted and thus is a dead directory in its own right.
+	//
+	// This is mostly a quality-of-life check, because mkdir will simply fail
+	// later if the attacker deletes the tree after this check.
+	if err := isDeadInode(currentDir); err != nil {
+		return nil, fmt.Errorf("finding existing subpath of %q: %w", unsafePath, err)
+	}
+
+	// Re-open the path to match the O_DIRECTORY reopen loop later (so that we
+	// always return a non-O_PATH handle). We also check that we actually got a
+	// directory.
+	if reopenDir, err := Reopen(currentDir, unix.O_DIRECTORY|unix.O_CLOEXEC); errors.Is(err, unix.ENOTDIR) {
+		return nil, fmt.Errorf("cannot create subdirectories in %q: %w", currentDir.Name(), unix.ENOTDIR)
+	} else if err != nil {
+		return nil, fmt.Errorf("re-opening handle to %q: %w", currentDir.Name(), err)
+	} else {
+		_ = currentDir.Close()
+		currentDir = reopenDir
+	}
+
+	remainingParts := strings.Split(remainingPath, string(filepath.Separator))
+	if slices.Contains(remainingParts, "..") {
+		// The path contained ".." components after the end of the "real"
+		// components. We could try to safely resolve ".." here but that would
+		// add a bunch of extra logic for something that it's not clear even
+		// needs to be supported. So just return an error.
+		//
+		// If we do filepath.Clean(remainingPath) then we end up with the
+		// problem that ".." can erase a trailing dangling symlink and produce
+		// a path that doesn't quite match what the user asked for.
+		return nil, fmt.Errorf("%w: yet-to-be-created path %q contains '..' components", unix.ENOENT, remainingPath)
+	}
+
+	// Make sure the mode doesn't have any type bits.
+	mode &^= unix.S_IFMT
+
+	// Create the remaining components.
+	for _, part := range remainingParts {
+		switch part {
+		case "", ".":
+			// Skip over no-op paths.
+			continue
+		}
+
+		// NOTE: mkdir(2) will not follow trailing symlinks, so we can safely
+		// create the final component without worrying about symlink-exchange
+		// attacks.
+		if err := unix.Mkdirat(int(currentDir.Fd()), part, uint32(mode)); err != nil {
+			err = &os.PathError{Op: "mkdirat", Path: currentDir.Name() + "/" + part, Err: err}
+			// Make the error a bit nicer if the directory is dead.
+			if err2 := isDeadInode(currentDir); err2 != nil {
+				err = fmt.Errorf("%w (%w)", err, err2)
+			}
+			return nil, err
+		}
+
+		// Get a handle to the next component. O_DIRECTORY means we don't need
+		// to use O_PATH.
+		var nextDir *os.File
+		if hasOpenat2() {
+			nextDir, err = openat2File(currentDir, part, &unix.OpenHow{
+				Flags:   unix.O_NOFOLLOW | unix.O_DIRECTORY | unix.O_CLOEXEC,
+				Resolve: unix.RESOLVE_BENEATH | unix.RESOLVE_NO_SYMLINKS | unix.RESOLVE_NO_XDEV,
+			})
+		} else {
+			nextDir, err = openatFile(currentDir, part, unix.O_NOFOLLOW|unix.O_DIRECTORY|unix.O_CLOEXEC, 0)
+		}
+		if err != nil {
+			return nil, err
+		}
+		_ = currentDir.Close()
+		currentDir = nextDir
+
+		// It's possible that the directory we just opened was swapped by an
+		// attacker. Unfortunately there isn't much we can do to protect
+		// against this, and MkdirAll's behaviour is that we will reuse
+		// existing directories anyway so the need to protect against this is
+		// incredibly limited (and arguably doesn't even deserve mention here).
+		//
+		// Ideally we might want to check that the owner and mode match what we
+		// would've created -- unfortunately, it is non-trivial to verify that
+		// the owner and mode of the created directory match. While plain Unix
+		// DAC rules seem simple enough to emulate, there are a bunch of other
+		// factors that can change the mode or owner of created directories
+		// (default POSIX ACLs, mount options like uid=1,gid=2,umask=0 on
+		// filesystems like vfat, etc etc). We used to try to verify this but
+		// it just lead to a series of spurious errors.
+		//
+		// We could also check that the directory is non-empty, but
+		// unfortunately some pseduofilesystems (like cgroupfs) create
+		// non-empty directories, which would result in different spurious
+		// errors.
+	}
+	return currentDir, nil
+}
+
+// MkdirAll is a race-safe alternative to the [os.MkdirAll] function,
+// where the new directory is guaranteed to be within the root directory (if an
+// attacker can move directories from inside the root to outside the root, the
+// created directory tree might be outside of the root but the key constraint
+// is that at no point will we walk outside of the directory tree we are
+// creating).
+//
+// Effectively, MkdirAll(root, unsafePath, mode) is equivalent to
+//
+//	path, _ := securejoin.SecureJoin(root, unsafePath)
+//	err := os.MkdirAll(path, mode)
+//
+// But is much safer. The above implementation is unsafe because if an attacker
+// can modify the filesystem tree between [SecureJoin] and [os.MkdirAll], it is
+// possible for MkdirAll to resolve unsafe symlink components and create
+// directories outside of the root.
+//
+// If you plan to open the directory after you have created it or want to use
+// an open directory handle as the root, you should use [MkdirAllHandle] instead.
+// This function is a wrapper around [MkdirAllHandle].
+//
+// NOTE: The mode argument must be set the unix mode bits (unix.S_I...), not
+// the Go generic mode bits ([os.FileMode]...).
+func MkdirAll(root, unsafePath string, mode int) error {
+	rootDir, err := os.OpenFile(root, unix.O_PATH|unix.O_DIRECTORY|unix.O_CLOEXEC, 0)
+	if err != nil {
+		return err
+	}
+	defer rootDir.Close()
+
+	f, err := MkdirAllHandle(rootDir, unsafePath, mode)
+	if err != nil {
+		return err
+	}
+	_ = f.Close()
+	return nil
+}

vendor/github.com/cyphar/filepath-securejoin/open_linux.go

@@ -0,0 +1,103 @@
+//go:build linux
+
+// Copyright (C) 2024 SUSE LLC. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package securejoin
+
+import (
+	"fmt"
+	"os"
+	"strconv"
+
+	"golang.org/x/sys/unix"
+)
+
+// OpenatInRoot is equivalent to [OpenInRoot], except that the root is provided
+// using an *[os.File] handle, to ensure that the correct root directory is used.
+func OpenatInRoot(root *os.File, unsafePath string) (*os.File, error) {
+	handle, err := completeLookupInRoot(root, unsafePath)
+	if err != nil {
+		return nil, &os.PathError{Op: "securejoin.OpenInRoot", Path: unsafePath, Err: err}
+	}
+	return handle, nil
+}
+
+// OpenInRoot safely opens the provided unsafePath within the root.
+// Effectively, OpenInRoot(root, unsafePath) is equivalent to
+//
+//	path, _ := securejoin.SecureJoin(root, unsafePath)
+//	handle, err := os.OpenFile(path, unix.O_PATH|unix.O_CLOEXEC)
+//
+// But is much safer. The above implementation is unsafe because if an attacker
+// can modify the filesystem tree between [SecureJoin] and [os.OpenFile], it is
+// possible for the returned file to be outside of the root.
+//
+// Note that the returned handle is an O_PATH handle, meaning that only a very
+// limited set of operations will work on the handle. This is done to avoid
+// accidentally opening an untrusted file that could cause issues (such as a
+// disconnected TTY that could cause a DoS, or some other issue). In order to
+// use the returned handle, you can "upgrade" it to a proper handle using
+// [Reopen].
+func OpenInRoot(root, unsafePath string) (*os.File, error) {
+	rootDir, err := os.OpenFile(root, unix.O_PATH|unix.O_DIRECTORY|unix.O_CLOEXEC, 0)
+	if err != nil {
+		return nil, err
+	}
+	defer rootDir.Close()
+	return OpenatInRoot(rootDir, unsafePath)
+}
+
+// Reopen takes an *[os.File] handle and re-opens it through /proc/self/fd.
+// Reopen(file, flags) is effectively equivalent to
+//
+//	fdPath := fmt.Sprintf("/proc/self/fd/%d", file.Fd())
+//	os.OpenFile(fdPath, flags|unix.O_CLOEXEC)
+//
+// But with some extra hardenings to ensure that we are not tricked by a
+// maliciously-configured /proc mount. While this attack scenario is not
+// common, in container runtimes it is possible for higher-level runtimes to be
+// tricked into configuring an unsafe /proc that can be used to attack file
+// operations. See [CVE-2019-19921] for more details.
+//
+// [CVE-2019-19921]: https://github.com/advisories/GHSA-fh74-hm69-rqjw
+func Reopen(handle *os.File, flags int) (*os.File, error) {
+	procRoot, err := getProcRoot()
+	if err != nil {
+		return nil, err
+	}
+
+	// We can't operate on /proc/thread-self/fd/$n directly when doing a
+	// re-open, so we need to open /proc/thread-self/fd and then open a single
+	// final component.
+	procFdDir, closer, err := procThreadSelf(procRoot, "fd/")
+	if err != nil {
+		return nil, fmt.Errorf("get safe /proc/thread-self/fd handle: %w", err)
+	}
+	defer procFdDir.Close()
+	defer closer()
+
+	// Try to detect if there is a mount on top of the magic-link we are about
+	// to open. If we are using unsafeHostProcRoot(), this could change after
+	// we check it (and there's nothing we can do about that) but for
+	// privateProcRoot() this should be guaranteed to be safe (at least since
+	// Linux 5.12[1], when anonymous mount namespaces were completely isolated
+	// from external mounts including mount propagation events).
+	//
+	// [1]: Linux commit ee2e3f50629f ("mount: fix mounting of detached mounts
+	// onto targets that reside on shared mounts").
+	fdStr := strconv.Itoa(int(handle.Fd()))
+	if err := checkSymlinkOvermount(procRoot, procFdDir, fdStr); err != nil {
+		return nil, fmt.Errorf("check safety of /proc/thread-self/fd/%s magiclink: %w", fdStr, err)
+	}
+
+	flags |= unix.O_CLOEXEC
+	// Rather than just wrapping openatFile, open-code it so we can copy
+	// handle.Name().
+	reopenFd, err := unix.Openat(int(procFdDir.Fd()), fdStr, flags, 0)
+	if err != nil {
+		return nil, fmt.Errorf("reopen fd %d: %w", handle.Fd(), err)
+	}
+	return os.NewFile(uintptr(reopenFd), handle.Name()), nil
+}

vendor/github.com/cyphar/filepath-securejoin/openat2_linux.go

@@ -0,0 +1,128 @@
+//go:build linux
+
+// Copyright (C) 2024 SUSE LLC. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package securejoin
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+
+	"golang.org/x/sys/unix"
+)
+
+var hasOpenat2 = sync.OnceValue(func() bool {
+	fd, err := unix.Openat2(unix.AT_FDCWD, ".", &unix.OpenHow{
+		Flags:   unix.O_PATH | unix.O_CLOEXEC,
+		Resolve: unix.RESOLVE_NO_SYMLINKS | unix.RESOLVE_IN_ROOT,
+	})
+	if err != nil {
+		return false
+	}
+	_ = unix.Close(fd)
+	return true
+})
+
+func scopedLookupShouldRetry(how *unix.OpenHow, err error) bool {
+	// RESOLVE_IN_ROOT (and RESOLVE_BENEATH) can return -EAGAIN if we resolve
+	// ".." while a mount or rename occurs anywhere on the system. This could
+	// happen spuriously, or as the result of an attacker trying to mess with
+	// us during lookup.
+	//
+	// In addition, scoped lookups have a "safety check" at the end of
+	// complete_walk which will return -EXDEV if the final path is not in the
+	// root.
+	return how.Resolve&(unix.RESOLVE_IN_ROOT|unix.RESOLVE_BENEATH) != 0 &&
+		(errors.Is(err, unix.EAGAIN) || errors.Is(err, unix.EXDEV))
+}
+
+const scopedLookupMaxRetries = 10
+
+func openat2File(dir *os.File, path string, how *unix.OpenHow) (*os.File, error) {
+	fullPath := dir.Name() + "/" + path
+	// Make sure we always set O_CLOEXEC.
+	how.Flags |= unix.O_CLOEXEC
+	var tries int
+	for tries < scopedLookupMaxRetries {
+		fd, err := unix.Openat2(int(dir.Fd()), path, how)
+		if err != nil {
+			if scopedLookupShouldRetry(how, err) {
+				// We retry a couple of times to avoid the spurious errors, and
+				// if we are being attacked then returning -EAGAIN is the best
+				// we can do.
+				tries++
+				continue
+			}
+			return nil, &os.PathError{Op: "openat2", Path: fullPath, Err: err}
+		}
+		// If we are using RESOLVE_IN_ROOT, the name we generated may be wrong.
+		// NOTE: The procRoot code MUST NOT use RESOLVE_IN_ROOT, otherwise
+		//       you'll get infinite recursion here.
+		if how.Resolve&unix.RESOLVE_IN_ROOT == unix.RESOLVE_IN_ROOT {
+			if actualPath, err := rawProcSelfFdReadlink(fd); err == nil {
+				fullPath = actualPath
+			}
+		}
+		return os.NewFile(uintptr(fd), fullPath), nil
+	}
+	return nil, &os.PathError{Op: "openat2", Path: fullPath, Err: errPossibleAttack}
+}
+
+func lookupOpenat2(root *os.File, unsafePath string, partial bool) (*os.File, string, error) {
+	if !partial {
+		file, err := openat2File(root, unsafePath, &unix.OpenHow{
+			Flags:   unix.O_PATH | unix.O_CLOEXEC,
+			Resolve: unix.RESOLVE_IN_ROOT | unix.RESOLVE_NO_MAGICLINKS,
+		})
+		return file, "", err
+	}
+	return partialLookupOpenat2(root, unsafePath)
+}
+
+// partialLookupOpenat2 is an alternative implementation of
+// partialLookupInRoot, using openat2(RESOLVE_IN_ROOT) to more safely get a
+// handle to the deepest existing child of the requested path within the root.
+func partialLookupOpenat2(root *os.File, unsafePath string) (*os.File, string, error) {
+	// TODO: Implement this as a git-bisect-like binary search.
+
+	unsafePath = filepath.ToSlash(unsafePath) // noop
+	endIdx := len(unsafePath)
+	var lastError error
+	for endIdx > 0 {
+		subpath := unsafePath[:endIdx]
+
+		handle, err := openat2File(root, subpath, &unix.OpenHow{
+			Flags:   unix.O_PATH | unix.O_CLOEXEC,
+			Resolve: unix.RESOLVE_IN_ROOT | unix.RESOLVE_NO_MAGICLINKS,
+		})
+		if err == nil {
+			// Jump over the slash if we have a non-"" remainingPath.
+			if endIdx < len(unsafePath) {
+				endIdx += 1
+			}
+			// We found a subpath!
+			return handle, unsafePath[endIdx:], lastError
+		}
+		if errors.Is(err, unix.ENOENT) || errors.Is(err, unix.ENOTDIR) {
+			// That path doesn't exist, let's try the next directory up.
+			endIdx = strings.LastIndexByte(subpath, '/')
+			lastError = err
+			continue
+		}
+		return nil, "", fmt.Errorf("open subpath: %w", err)
+	}
+	// If we couldn't open anything, the whole subpath is missing. Return a
+	// copy of the root fd so that the caller doesn't close this one by
+	// accident.
+	rootClone, err := dupFile(root)
+	if err != nil {
+		return nil, "", err
+	}
+	return rootClone, unsafePath, lastError
+}

vendor/github.com/cyphar/filepath-securejoin/openat_linux.go

@@ -0,0 +1,59 @@
+//go:build linux
+
+// Copyright (C) 2024 SUSE LLC. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package securejoin
+
+import (
+	"os"
+	"path/filepath"
+
+	"golang.org/x/sys/unix"
+)
+
+func dupFile(f *os.File) (*os.File, error) {
+	fd, err := unix.FcntlInt(f.Fd(), unix.F_DUPFD_CLOEXEC, 0)
+	if err != nil {
+		return nil, os.NewSyscallError("fcntl(F_DUPFD_CLOEXEC)", err)
+	}
+	return os.NewFile(uintptr(fd), f.Name()), nil
+}
+
+func openatFile(dir *os.File, path string, flags int, mode int) (*os.File, error) {
+	// Make sure we always set O_CLOEXEC.
+	flags |= unix.O_CLOEXEC
+	fd, err := unix.Openat(int(dir.Fd()), path, flags, uint32(mode))
+	if err != nil {
+		return nil, &os.PathError{Op: "openat", Path: dir.Name() + "/" + path, Err: err}
+	}
+	// All of the paths we use with openatFile(2) are guaranteed to be
+	// lexically safe, so we can use path.Join here.
+	fullPath := filepath.Join(dir.Name(), path)
+	return os.NewFile(uintptr(fd), fullPath), nil
+}
+
+func fstatatFile(dir *os.File, path string, flags int) (unix.Stat_t, error) {
+	var stat unix.Stat_t
+	if err := unix.Fstatat(int(dir.Fd()), path, &stat, flags); err != nil {
+		return stat, &os.PathError{Op: "fstatat", Path: dir.Name() + "/" + path, Err: err}
+	}
+	return stat, nil
+}
+
+func readlinkatFile(dir *os.File, path string) (string, error) {
+	size := 4096
+	for {
+		linkBuf := make([]byte, size)
+		n, err := unix.Readlinkat(int(dir.Fd()), path, linkBuf)
+		if err != nil {
+			return "", &os.PathError{Op: "readlinkat", Path: dir.Name() + "/" + path, Err: err}
+		}
+		if n != size {
+			return string(linkBuf[:n]), nil
+		}
+		// Possible truncation, resize the buffer.
+		size *= 2
+	}
+}

vendor/github.com/cyphar/filepath-securejoin/procfs_linux.go

@@ -0,0 +1,440 @@
+//go:build linux
+
+// Copyright (C) 2024 SUSE LLC. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package securejoin
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"runtime"
+	"strconv"
+	"sync"
+
+	"golang.org/x/sys/unix"
+)
+
+func fstat(f *os.File) (unix.Stat_t, error) {
+	var stat unix.Stat_t
+	if err := unix.Fstat(int(f.Fd()), &stat); err != nil {
+		return stat, &os.PathError{Op: "fstat", Path: f.Name(), Err: err}
+	}
+	return stat, nil
+}
+
+func fstatfs(f *os.File) (unix.Statfs_t, error) {
+	var statfs unix.Statfs_t
+	if err := unix.Fstatfs(int(f.Fd()), &statfs); err != nil {
+		return statfs, &os.PathError{Op: "fstatfs", Path: f.Name(), Err: err}
+	}
+	return statfs, nil
+}
+
+// The kernel guarantees that the root inode of a procfs mount has an
+// f_type of PROC_SUPER_MAGIC and st_ino of PROC_ROOT_INO.
+const (
+	procSuperMagic = 0x9fa0 // PROC_SUPER_MAGIC
+	procRootIno    = 1      // PROC_ROOT_INO
+)
+
+func verifyProcRoot(procRoot *os.File) error {
+	if statfs, err := fstatfs(procRoot); err != nil {
+		return err
+	} else if statfs.Type != procSuperMagic {
+		return fmt.Errorf("%w: incorrect procfs root filesystem type 0x%x", errUnsafeProcfs, statfs.Type)
+	}
+	if stat, err := fstat(procRoot); err != nil {
+		return err
+	} else if stat.Ino != procRootIno {
+		return fmt.Errorf("%w: incorrect procfs root inode number %d", errUnsafeProcfs, stat.Ino)
+	}
+	return nil
+}
+
+var hasNewMountApi = sync.OnceValue(func() bool {
+	// All of the pieces of the new mount API we use (fsopen, fsconfig,
+	// fsmount, open_tree) were added together in Linux 5.1[1,2], so we can
+	// just check for one of the syscalls and the others should also be
+	// available.
+	//
+	// Just try to use open_tree(2) to open a file without OPEN_TREE_CLONE.
+	// This is equivalent to openat(2), but tells us if open_tree is
+	// available (and thus all of the other basic new mount API syscalls).
+	// open_tree(2) is most light-weight syscall to test here.
+	//
+	// [1]: merge commit 400913252d09
+	// [2]: <https://lore.kernel.org/lkml/153754740781.17872.7869536526927736855.stgit@warthog.procyon.org.uk/>
+	fd, err := unix.OpenTree(-int(unix.EBADF), "/", unix.OPEN_TREE_CLOEXEC)
+	if err != nil {
+		return false
+	}
+	_ = unix.Close(fd)
+	return true
+})
+
+func fsopen(fsName string, flags int) (*os.File, error) {
+	// Make sure we always set O_CLOEXEC.
+	flags |= unix.FSOPEN_CLOEXEC
+	fd, err := unix.Fsopen(fsName, flags)
+	if err != nil {
+		return nil, os.NewSyscallError("fsopen "+fsName, err)
+	}
+	return os.NewFile(uintptr(fd), "fscontext:"+fsName), nil
+}
+
+func fsmount(ctx *os.File, flags, mountAttrs int) (*os.File, error) {
+	// Make sure we always set O_CLOEXEC.
+	flags |= unix.FSMOUNT_CLOEXEC
+	fd, err := unix.Fsmount(int(ctx.Fd()), flags, mountAttrs)
+	if err != nil {
+		return nil, os.NewSyscallError("fsmount "+ctx.Name(), err)
+	}
+	return os.NewFile(uintptr(fd), "fsmount:"+ctx.Name()), nil
+}
+
+func newPrivateProcMount() (*os.File, error) {
+	procfsCtx, err := fsopen("proc", unix.FSOPEN_CLOEXEC)
+	if err != nil {
+		return nil, err
+	}
+	defer procfsCtx.Close()
+
+	// Try to configure hidepid=ptraceable,subset=pid if possible, but ignore errors.
+	_ = unix.FsconfigSetString(int(procfsCtx.Fd()), "hidepid", "ptraceable")
+	_ = unix.FsconfigSetString(int(procfsCtx.Fd()), "subset", "pid")
+
+	// Get an actual handle.
+	if err := unix.FsconfigCreate(int(procfsCtx.Fd())); err != nil {
+		return nil, os.NewSyscallError("fsconfig create procfs", err)
+	}
+	return fsmount(procfsCtx, unix.FSMOUNT_CLOEXEC, unix.MS_RDONLY|unix.MS_NODEV|unix.MS_NOEXEC|unix.MS_NOSUID)
+}
+
+func openTree(dir *os.File, path string, flags uint) (*os.File, error) {
+	dirFd := -int(unix.EBADF)
+	dirName := "."
+	if dir != nil {
+		dirFd = int(dir.Fd())
+		dirName = dir.Name()
+	}
+	// Make sure we always set O_CLOEXEC.
+	flags |= unix.OPEN_TREE_CLOEXEC
+	fd, err := unix.OpenTree(dirFd, path, flags)
+	if err != nil {
+		return nil, &os.PathError{Op: "open_tree", Path: path, Err: err}
+	}
+	return os.NewFile(uintptr(fd), dirName+"/"+path), nil
+}
+
+func clonePrivateProcMount() (_ *os.File, Err error) {
+	// Try to make a clone without using AT_RECURSIVE if we can. If this works,
+	// we can be sure there are no over-mounts and so if the root is valid then
+	// we're golden. Otherwise, we have to deal with over-mounts.
+	procfsHandle, err := openTree(nil, "/proc", unix.OPEN_TREE_CLONE)
+	if err != nil || hookForcePrivateProcRootOpenTreeAtRecursive(procfsHandle) {
+		procfsHandle, err = openTree(nil, "/proc", unix.OPEN_TREE_CLONE|unix.AT_RECURSIVE)
+	}
+	if err != nil {
+		return nil, fmt.Errorf("creating a detached procfs clone: %w", err)
+	}
+	defer func() {
+		if Err != nil {
+			_ = procfsHandle.Close()
+		}
+	}()
+	if err := verifyProcRoot(procfsHandle); err != nil {
+		return nil, err
+	}
+	return procfsHandle, nil
+}
+
+func privateProcRoot() (*os.File, error) {
+	if !hasNewMountApi() || hookForceGetProcRootUnsafe() {
+		return nil, fmt.Errorf("new mount api: %w", unix.ENOTSUP)
+	}
+	// Try to create a new procfs mount from scratch if we can. This ensures we
+	// can get a procfs mount even if /proc is fake (for whatever reason).
+	procRoot, err := newPrivateProcMount()
+	if err != nil || hookForcePrivateProcRootOpenTree(procRoot) {
+		// Try to clone /proc then...
+		procRoot, err = clonePrivateProcMount()
+	}
+	return procRoot, err
+}
+
+func unsafeHostProcRoot() (_ *os.File, Err error) {
+	procRoot, err := os.OpenFile("/proc", unix.O_PATH|unix.O_NOFOLLOW|unix.O_DIRECTORY|unix.O_CLOEXEC, 0)
+	if err != nil {
+		return nil, err
+	}
+	defer func() {
+		if Err != nil {
+			_ = procRoot.Close()
+		}
+	}()
+	if err := verifyProcRoot(procRoot); err != nil {
+		return nil, err
+	}
+	return procRoot, nil
+}
+
+func doGetProcRoot() (*os.File, error) {
+	procRoot, err := privateProcRoot()
+	if err != nil {
+		// Fall back to using a /proc handle if making a private mount failed.
+		// If we have openat2, at least we can avoid some kinds of over-mount
+		// attacks, but without openat2 there's not much we can do.
+		procRoot, err = unsafeHostProcRoot()
+	}
+	return procRoot, err
+}
+
+var getProcRoot = sync.OnceValues(func() (*os.File, error) {
+	return doGetProcRoot()
+})
+
+var hasProcThreadSelf = sync.OnceValue(func() bool {
+	return unix.Access("/proc/thread-self/", unix.F_OK) == nil
+})
+
+var errUnsafeProcfs = errors.New("unsafe procfs detected")
+
+type procThreadSelfCloser func()
+
+// procThreadSelf returns a handle to /proc/thread-self/<subpath> (or an
+// equivalent handle on older kernels where /proc/thread-self doesn't exist).
+// Once finished with the handle, you must call the returned closer function
+// (runtime.UnlockOSThread). You must not pass the returned *os.File to other
+// Go threads or use the handle after calling the closer.
+//
+// This is similar to ProcThreadSelf from runc, but with extra hardening
+// applied and using *os.File.
+func procThreadSelf(procRoot *os.File, subpath string) (_ *os.File, _ procThreadSelfCloser, Err error) {
+	// We need to lock our thread until the caller is done with the handle
+	// because between getting the handle and using it we could get interrupted
+	// by the Go runtime and hit the case where the underlying thread is
+	// swapped out and the original thread is killed, resulting in
+	// pull-your-hair-out-hard-to-debug issues in the caller.
+	runtime.LockOSThread()
+	defer func() {
+		if Err != nil {
+			runtime.UnlockOSThread()
+		}
+	}()
+
+	// Figure out what prefix we want to use.
+	threadSelf := "thread-self/"
+	if !hasProcThreadSelf() || hookForceProcSelfTask() {
+		/// Pre-3.17 kernels don't have /proc/thread-self, so do it manually.
+		threadSelf = "self/task/" + strconv.Itoa(unix.Gettid()) + "/"
+		if _, err := fstatatFile(procRoot, threadSelf, unix.AT_SYMLINK_NOFOLLOW); err != nil || hookForceProcSelf() {
+			// In this case, we running in a pid namespace that doesn't match
+			// the /proc mount we have. This can happen inside runc.
+			//
+			// Unfortunately, there is no nice way to get the correct TID to
+			// use here because of the age of the kernel, so we have to just
+			// use /proc/self and hope that it works.
+			threadSelf = "self/"
+		}
+	}
+
+	// Grab the handle.
+	var (
+		handle *os.File
+		err    error
+	)
+	if hasOpenat2() {
+		// We prefer being able to use RESOLVE_NO_XDEV if we can, to be
+		// absolutely sure we are operating on a clean /proc handle that
+		// doesn't have any cheeky overmounts that could trick us (including
+		// symlink mounts on top of /proc/thread-self). RESOLVE_BENEATH isn't
+		// strictly needed, but just use it since we have it.
+		//
+		// NOTE: /proc/self is technically a magic-link (the contents of the
+		//       symlink are generated dynamically), but it doesn't use
+		//       nd_jump_link() so RESOLVE_NO_MAGICLINKS allows it.
+		//
+		// NOTE: We MUST NOT use RESOLVE_IN_ROOT here, as openat2File uses
+		//       procSelfFdReadlink to clean up the returned f.Name() if we use
+		//       RESOLVE_IN_ROOT (which would lead to an infinite recursion).
+		handle, err = openat2File(procRoot, threadSelf+subpath, &unix.OpenHow{
+			Flags:   unix.O_PATH | unix.O_NOFOLLOW | unix.O_CLOEXEC,
+			Resolve: unix.RESOLVE_BENEATH | unix.RESOLVE_NO_XDEV | unix.RESOLVE_NO_MAGICLINKS,
+		})
+		if err != nil {
+			return nil, nil, fmt.Errorf("%w: %w", errUnsafeProcfs, err)
+		}
+	} else {
+		handle, err = openatFile(procRoot, threadSelf+subpath, unix.O_PATH|unix.O_NOFOLLOW|unix.O_CLOEXEC, 0)
+		if err != nil {
+			return nil, nil, fmt.Errorf("%w: %w", errUnsafeProcfs, err)
+		}
+		defer func() {
+			if Err != nil {
+				_ = handle.Close()
+			}
+		}()
+		// We can't detect bind-mounts of different parts of procfs on top of
+		// /proc (a-la RESOLVE_NO_XDEV), but we can at least be sure that we
+		// aren't on the wrong filesystem here.
+		if statfs, err := fstatfs(handle); err != nil {
+			return nil, nil, err
+		} else if statfs.Type != procSuperMagic {
+			return nil, nil, fmt.Errorf("%w: incorrect /proc/self/fd filesystem type 0x%x", errUnsafeProcfs, statfs.Type)
+		}
+	}
+	return handle, runtime.UnlockOSThread, nil
+}
+
+var hasStatxMountId = sync.OnceValue(func() bool {
+	var (
+		stx unix.Statx_t
+		// We don't care which mount ID we get. The kernel will give us the
+		// unique one if it is supported.
+		wantStxMask uint32 = unix.STATX_MNT_ID_UNIQUE | unix.STATX_MNT_ID
+	)
+	err := unix.Statx(-int(unix.EBADF), "/", 0, int(wantStxMask), &stx)
+	return err == nil && stx.Mask&wantStxMask != 0
+})
+
+func getMountId(dir *os.File, path string) (uint64, error) {
+	// If we don't have statx(STATX_MNT_ID*) support, we can't do anything.
+	if !hasStatxMountId() {
+		return 0, nil
+	}
+
+	var (
+		stx unix.Statx_t
+		// We don't care which mount ID we get. The kernel will give us the
+		// unique one if it is supported.
+		wantStxMask uint32 = unix.STATX_MNT_ID_UNIQUE | unix.STATX_MNT_ID
+	)
+
+	err := unix.Statx(int(dir.Fd()), path, unix.AT_EMPTY_PATH|unix.AT_SYMLINK_NOFOLLOW, int(wantStxMask), &stx)
+	if stx.Mask&wantStxMask == 0 {
+		// It's not a kernel limitation, for some reason we couldn't get a
+		// mount ID. Assume it's some kind of attack.
+		err = fmt.Errorf("%w: could not get mount id", errUnsafeProcfs)
+	}
+	if err != nil {
+		return 0, &os.PathError{Op: "statx(STATX_MNT_ID_...)", Path: dir.Name() + "/" + path, Err: err}
+	}
+	return stx.Mnt_id, nil
+}
+
+func checkSymlinkOvermount(procRoot *os.File, dir *os.File, path string) error {
+	// Get the mntId of our procfs handle.
+	expectedMountId, err := getMountId(procRoot, "")
+	if err != nil {
+		return err
+	}
+	// Get the mntId of the target magic-link.
+	gotMountId, err := getMountId(dir, path)
+	if err != nil {
+		return err
+	}
+	// As long as the directory mount is alive, even with wrapping mount IDs,
+	// we would expect to see a different mount ID here. (Of course, if we're
+	// using unsafeHostProcRoot() then an attaker could change this after we
+	// did this check.)
+	if expectedMountId != gotMountId {
+		return fmt.Errorf("%w: symlink %s/%s has an overmount obscuring the real link (mount ids do not match %d != %d)", errUnsafeProcfs, dir.Name(), path, expectedMountId, gotMountId)
+	}
+	return nil
+}
+
+func doRawProcSelfFdReadlink(procRoot *os.File, fd int) (string, error) {
+	fdPath := fmt.Sprintf("fd/%d", fd)
+	procFdLink, closer, err := procThreadSelf(procRoot, fdPath)
+	if err != nil {
+		return "", fmt.Errorf("get safe /proc/thread-self/%s handle: %w", fdPath, err)
+	}
+	defer procFdLink.Close()
+	defer closer()
+
+	// Try to detect if there is a mount on top of the magic-link. Since we use the handle directly
+	// provide to the closure. If the closure uses the handle directly, this
+	// should be safe in general (a mount on top of the path afterwards would
+	// not affect the handle itself) and will definitely be safe if we are
+	// using privateProcRoot() (at least since Linux 5.12[1], when anonymous
+	// mount namespaces were completely isolated from external mounts including
+	// mount propagation events).
+	//
+	// [1]: Linux commit ee2e3f50629f ("mount: fix mounting of detached mounts
+	// onto targets that reside on shared mounts").
+	if err := checkSymlinkOvermount(procRoot, procFdLink, ""); err != nil {
+		return "", fmt.Errorf("check safety of /proc/thread-self/fd/%d magiclink: %w", fd, err)
+	}
+
+	// readlinkat implies AT_EMPTY_PATH since Linux 2.6.39. See Linux commit
+	// 65cfc6722361 ("readlinkat(), fchownat() and fstatat() with empty
+	// relative pathnames").
+	return readlinkatFile(procFdLink, "")
+}
+
+func rawProcSelfFdReadlink(fd int) (string, error) {
+	procRoot, err := getProcRoot()
+	if err != nil {
+		return "", err
+	}
+	return doRawProcSelfFdReadlink(procRoot, fd)
+}
+
+func procSelfFdReadlink(f *os.File) (string, error) {
+	return rawProcSelfFdReadlink(int(f.Fd()))
+}
+
+var (
+	errPossibleBreakout = errors.New("possible breakout detected")
+	errInvalidDirectory = errors.New("wandered into deleted directory")
+	errDeletedInode     = errors.New("cannot verify path of deleted inode")
+)
+
+func isDeadInode(file *os.File) error {
+	// If the nlink of a file drops to 0, there is an attacker deleting
+	// directories during our walk, which could result in weird /proc values.
+	// It's better to error out in this case.
+	stat, err := fstat(file)
+	if err != nil {
+		return fmt.Errorf("check for dead inode: %w", err)
+	}
+	if stat.Nlink == 0 {
+		err := errDeletedInode
+		if stat.Mode&unix.S_IFMT == unix.S_IFDIR {
+			err = errInvalidDirectory
+		}
+		return fmt.Errorf("%w %q", err, file.Name())
+	}
+	return nil
+}
+
+func checkProcSelfFdPath(path string, file *os.File) error {
+	if err := isDeadInode(file); err != nil {
+		return err
+	}
+	actualPath, err := procSelfFdReadlink(file)
+	if err != nil {
+		return fmt.Errorf("get path of handle: %w", err)
+	}
+	if actualPath != path {
+		return fmt.Errorf("%w: handle path %q doesn't match expected path %q", errPossibleBreakout, actualPath, path)
+	}
+	return nil
+}
+
+// Test hooks used in the procfs tests to verify that the fallback logic works.
+// See testing_mocks_linux_test.go and procfs_linux_test.go for more details.
+var (
+	hookForcePrivateProcRootOpenTree            = hookDummyFile
+	hookForcePrivateProcRootOpenTreeAtRecursive = hookDummyFile
+	hookForceGetProcRootUnsafe                  = hookDummy
+
+	hookForceProcSelfTask = hookDummy
+	hookForceProcSelf     = hookDummy
+)
+
+func hookDummy() bool               { return false }
+func hookDummyFile(_ *os.File) bool { return false }

vendor/github.com/cyphar/filepath-securejoin/vfs.go

@@ -1,4 +1,4 @@
-// Copyright (C) 2017 SUSE LLC. All rights reserved.
+// Copyright (C) 2017-2024 SUSE LLC. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
@@ -10,19 +10,19 @@ import "os"
 // are several projects (umoci and go-mtree) that are using this sort of
 // interface.
 
-// VFS is the minimal interface necessary to use SecureJoinVFS. A nil VFS is
+// VFS is the minimal interface necessary to use [SecureJoinVFS]. A nil VFS is
-// equivalent to using the standard os.* family of functions. This is mainly
+// equivalent to using the standard [os].* family of functions. This is mainly
 // used for the purposes of mock testing, but also can be used to otherwise use
-// SecureJoin with VFS-like system.
+// [SecureJoinVFS] with VFS-like system.
 type VFS interface {
-	// Lstat returns a FileInfo describing the named file. If the file is a
+	// Lstat returns an [os.FileInfo] describing the named file. If the
-	// symbolic link, the returned FileInfo describes the symbolic link. Lstat
+	// file is a symbolic link, the returned [os.FileInfo] describes the
-	// makes no attempt to follow the link. These semantics are identical to
+	// symbolic link. Lstat makes no attempt to follow the link.
-	// os.Lstat.
+	// The semantics are identical to [os.Lstat].
 	Lstat(name string) (os.FileInfo, error)
 
-	// Readlink returns the destination of the named symbolic link. These
+	// Readlink returns the destination of the named symbolic link.
-	// semantics are identical to os.Readlink.
+	// The semantics are identical to [os.Readlink].
 	Readlink(name string) (string, error)
 }
 
@@ -30,12 +30,6 @@ type VFS interface {
 // module.
 type osVFS struct{}
 
-// Lstat returns a FileInfo describing the named file. If the file is a
-// symbolic link, the returned FileInfo describes the symbolic link. Lstat
-// makes no attempt to follow the link. These semantics are identical to
-// os.Lstat.
 func (o osVFS) Lstat(name string) (os.FileInfo, error) { return os.Lstat(name) }
 
-// Readlink returns the destination of the named symbolic link. These
-// semantics are identical to os.Readlink.
 func (o osVFS) Readlink(name string) (string, error) { return os.Readlink(name) }

vendor/modules.txt

@@ -127,8 +127,8 @@ github.com/coreos/go-systemd/v22/util
 # github.com/cpuguy83/go-md2man/v2 v2.0.4
 ## explicit; go 1.11
 github.com/cpuguy83/go-md2man/v2/md2man
-# github.com/cyphar/filepath-securejoin v0.2.4
+# github.com/cyphar/filepath-securejoin v0.3.4
-## explicit; go 1.13
+## explicit; go 1.21
 github.com/cyphar/filepath-securejoin
 # github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc
 ## explicit