update vendor

Signed-off-by: Jess Frazelle <acidburn@microsoft.com>
This commit is contained in:
Jess Frazelle 2018-09-25 12:27:46 -04:00
parent 19a32db84d
commit 94d1cfbfbf
No known key found for this signature in database
GPG key ID: 18F3685C0022BFF3
10501 changed files with 2307943 additions and 29279 deletions

View file

@ -10,8 +10,8 @@ The `nsenter` package will `import "C"` and it uses [cgo](https://golang.org/cmd
package. In cgo, if the import of "C" is immediately preceded by a comment, that comment,
called the preamble, is used as a header when compiling the C parts of the package.
So every time we import package `nsenter`, the C code function `nsexec()` would be
called. And package `nsenter` is now only imported in `main_unix.go`, so every time
before we call `cmd.Start` on linux, that C code would run.
called. And package `nsenter` is only imported in `init.go`, so every time the runc
`init` command is invoked, that C code is run.
Because `nsexec()` must be run before the Go runtime in order to use the
Linux kernel namespace, you must `import` this library into a package if
@ -37,7 +37,7 @@ the parent `nsexec()` will exit and the child `nsexec()` process will
return to allow the Go runtime take over.
NOTE: We do both `setns(2)` and `clone(2)` even if we don't have any
CLONE_NEW* clone flags because we must fork a new process in order to
`CLONE_NEW*` clone flags because we must fork a new process in order to
enter the PID namespace.

View file

@ -211,7 +211,7 @@ static int try_mapping_tool(const char *app, int pid, char *map, size_t map_len)
/*
* If @app is NULL, execve will segfault. Just check it here and bail (if
* we're in this path, the caller is already getting desparate and there
* we're in this path, the caller is already getting desperate and there
* isn't a backup to this failing). This usually would be a configuration
* or programming issue.
*/
@ -505,7 +505,8 @@ void join_namespaces(char *nslist)
ns->fd = fd;
ns->ns = nsflag(namespace);
strncpy(ns->path, path, PATH_MAX);
strncpy(ns->path, path, PATH_MAX - 1);
ns->path[PATH_MAX - 1] = '\0';
} while ((namespace = strtok_r(NULL, ",", &saveptr)) != NULL);
/*
@ -678,17 +679,15 @@ void nsexec(void)
/*
* Enable setgroups(2) if we've been asked to. But we also
* have to explicitly disable setgroups(2) if we're
* creating a rootless container (this is required since
* Linux 3.19).
* creating a rootless container for single-entry mapping.
* i.e. config.is_setgroup == false.
* (this is required since Linux 3.19).
*
* For rootless multi-entry mapping, config.is_setgroup shall be true and
* newuidmap/newgidmap shall be used.
*/
if (config.is_rootless && config.is_setgroup) {
kill(child, SIGKILL);
bail("cannot allow setgroup in an unprivileged user namespace setup");
}
if (config.is_setgroup)
update_setgroups(child, SETGROUPS_ALLOW);
if (config.is_rootless)
if (config.is_rootless && !config.is_setgroup)
update_setgroups(child, SETGROUPS_DENY);
/* Set up mappings. */
@ -809,25 +808,30 @@ void nsexec(void)
if (config.namespaces)
join_namespaces(config.namespaces);
/*
* Unshare all of the namespaces. Now, it should be noted that this
* ordering might break in the future (especially with rootless
* containers). But for now, it's not possible to split this into
* CLONE_NEWUSER + [the rest] because of some RHEL SELinux issues.
*
* Note that we don't merge this with clone() because there were
* some old kernel versions where clone(CLONE_PARENT | CLONE_NEWPID)
* was broken, so we'll just do it the long way anyway.
*/
if (unshare(config.cloneflags) < 0)
bail("failed to unshare namespaces");
/*
* Deal with user namespaces first. They are quite special, as they
* affect our ability to unshare other namespaces and are used as
* context for privilege checks.
*
* We don't unshare all namespaces in one go. The reason for this
* is that, while the kernel documentation may claim otherwise,
* there are certain cases where unsharing all namespaces at once
* will result in namespace objects being owned incorrectly.
* Ideally we should just fix these kernel bugs, but it's better to
* be safe than sorry, and fix them separately.
*
* A specific case of this is that the SELinux label of the
* internal kern-mount that mqueue uses will be incorrect if the
* UTS namespace is cloned before the USER namespace is mapped.
* I've also heard of similar problems with the network namespace
* in some scenarios. This also mirrors how LXC deals with this
* problem.
*/
if (config.cloneflags & CLONE_NEWUSER) {
if (unshare(CLONE_NEWUSER) < 0)
bail("failed to unshare user namespace");
config.cloneflags &= ~CLONE_NEWUSER;
/*
* We don't have the privileges to do any mapping here (see the
* clone_parent rant). So signal our parent to hook us up.
@ -853,8 +857,21 @@ void nsexec(void)
if (prctl(PR_SET_DUMPABLE, 0, 0, 0, 0) < 0)
bail("failed to set process as dumpable");
}
/* Become root in the namespace proper. */
if (setresuid(0, 0, 0) < 0)
bail("failed to become root in user namespace");
}
/*
* Unshare all of the namespaces. Note that we don't merge this
* with clone() because there were some old kernel versions where
* clone(CLONE_PARENT | CLONE_NEWPID) was broken, so we'll just do
* it the long way.
*/
if (unshare(config.cloneflags) < 0)
bail("failed to unshare namespaces");
/*
* TODO: What about non-namespace clone flags that we're dropping here?
*