1
1
Fork 0
mirror of https://github.com/NixOS/nix.git synced 2025-11-28 13:11:00 +01:00

Merge pull request #3600 from NixOS/auto-uid-allocation

Automatic UID allocation
This commit is contained in:
Eelco Dolstra 2022-11-29 14:01:42 +01:00 committed by GitHub
commit fbc53e97ed
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
25 changed files with 859 additions and 203 deletions

View file

@ -886,6 +886,14 @@ void DerivationGoal::buildDone()
cleanupPostChildKill();
if (buildResult.cpuUser && buildResult.cpuSystem) {
debug("builder for '%s' terminated with status %d, user CPU %.3fs, system CPU %.3fs",
worker.store.printStorePath(drvPath),
status,
((double) buildResult.cpuUser->count()) / 1000000,
((double) buildResult.cpuSystem->count()) / 1000000);
}
bool diskFull = false;
try {

View file

@ -14,6 +14,7 @@
#include "topo-sort.hh"
#include "callback.hh"
#include "json-utils.hh"
#include "cgroup.hh"
#include <regex>
#include <queue>
@ -129,26 +130,44 @@ void LocalDerivationGoal::killChild()
if (pid != -1) {
worker.childTerminated(this);
if (buildUser) {
/* If we're using a build user, then there is a tricky
race condition: if we kill the build user before the
child has done its setuid() to the build user uid, then
it won't be killed, and we'll potentially lock up in
pid.wait(). So also send a conventional kill to the
child. */
::kill(-pid, SIGKILL); /* ignore the result */
buildUser->kill();
pid.wait();
} else
pid.kill();
/* If we're using a build user, then there is a tricky race
condition: if we kill the build user before the child has
done its setuid() to the build user uid, then it won't be
killed, and we'll potentially lock up in pid.wait(). So
also send a conventional kill to the child. */
::kill(-pid, SIGKILL); /* ignore the result */
assert(pid == -1);
killSandbox(true);
pid.wait();
}
DerivationGoal::killChild();
}
void LocalDerivationGoal::killSandbox(bool getStats)
{
if (cgroup) {
#if __linux__
auto stats = destroyCgroup(*cgroup);
if (getStats) {
buildResult.cpuUser = stats.cpuUser;
buildResult.cpuSystem = stats.cpuSystem;
}
#else
abort();
#endif
}
else if (buildUser) {
auto uid = buildUser->getUID();
assert(uid != 0);
killUser(uid);
}
}
void LocalDerivationGoal::tryLocalBuild() {
unsigned int curBuilds = worker.getNrLocalBuilds();
if (curBuilds >= settings.maxBuildJobs) {
@ -158,28 +177,46 @@ void LocalDerivationGoal::tryLocalBuild() {
return;
}
/* If `build-users-group' is not empty, then we have to build as
one of the members of that group. */
if (settings.buildUsersGroup != "" && getuid() == 0) {
#if defined(__linux__) || defined(__APPLE__)
if (!buildUser) buildUser = std::make_unique<UserLock>();
/* Are we doing a chroot build? */
{
auto noChroot = parsedDrv->getBoolAttr("__noChroot");
if (settings.sandboxMode == smEnabled) {
if (noChroot)
throw Error("derivation '%s' has '__noChroot' set, "
"but that's not allowed when 'sandbox' is 'true'", worker.store.printStorePath(drvPath));
#if __APPLE__
if (additionalSandboxProfile != "")
throw Error("derivation '%s' specifies a sandbox profile, "
"but this is only allowed when 'sandbox' is 'relaxed'", worker.store.printStorePath(drvPath));
#endif
useChroot = true;
}
else if (settings.sandboxMode == smDisabled)
useChroot = false;
else if (settings.sandboxMode == smRelaxed)
useChroot = derivationType.isSandboxed() && !noChroot;
}
if (buildUser->findFreeUser()) {
/* Make sure that no other processes are executing under this
uid. */
buildUser->kill();
} else {
auto & localStore = getLocalStore();
if (localStore.storeDir != localStore.realStoreDir.get()) {
#if __linux__
useChroot = true;
#else
throw Error("building using a diverted store is not supported on this platform");
#endif
}
if (useBuildUsers()) {
if (!buildUser)
buildUser = acquireUserLock(parsedDrv->useUidRange() ? 65536 : 1, useChroot);
if (!buildUser) {
if (!actLock)
actLock = std::make_unique<Activity>(*logger, lvlWarn, actBuildWaiting,
fmt("waiting for UID to build '%s'", yellowtxt(worker.store.printStorePath(drvPath))));
worker.waitForAWhile(shared_from_this());
return;
}
#else
/* Don't know how to block the creation of setuid/setgid
binaries on this platform. */
throw Error("build users are not supported on this platform for security reasons");
#endif
}
actLock.reset();
@ -270,7 +307,7 @@ void LocalDerivationGoal::cleanupPostChildKill()
malicious user from leaving behind a process that keeps files
open and modifies them after they have been chown'ed to
root. */
if (buildUser) buildUser->kill();
killSandbox(true);
/* Terminate the recursive Nix daemon. */
stopDaemon();
@ -363,6 +400,60 @@ static void linkOrCopy(const Path & from, const Path & to)
void LocalDerivationGoal::startBuilder()
{
if ((buildUser && buildUser->getUIDCount() != 1)
#if __linux__
|| settings.useCgroups
#endif
)
{
#if __linux__
settings.requireExperimentalFeature(Xp::Cgroups);
auto ourCgroups = getCgroups("/proc/self/cgroup");
auto ourCgroup = ourCgroups[""];
if (ourCgroup == "")
throw Error("cannot determine cgroup name from /proc/self/cgroup");
auto ourCgroupPath = canonPath("/sys/fs/cgroup/" + ourCgroup);
if (!pathExists(ourCgroupPath))
throw Error("expected cgroup directory '%s'", ourCgroupPath);
static std::atomic<unsigned int> counter{0};
cgroup = buildUser
? fmt("%s/nix-build-uid-%d", ourCgroupPath, buildUser->getUID())
: fmt("%s/nix-build-pid-%d-%d", ourCgroupPath, getpid(), counter++);
debug("using cgroup '%s'", *cgroup);
/* When using a build user, record the cgroup we used for that
user so that if we got interrupted previously, we can kill
any left-over cgroup first. */
if (buildUser) {
auto cgroupsDir = settings.nixStateDir + "/cgroups";
createDirs(cgroupsDir);
auto cgroupFile = fmt("%s/%d", cgroupsDir, buildUser->getUID());
if (pathExists(cgroupFile)) {
auto prevCgroup = readFile(cgroupFile);
destroyCgroup(prevCgroup);
}
writeFile(cgroupFile, *cgroup);
}
#else
throw Error("cgroups are not supported on this platform");
#endif
}
/* Make sure that no other processes are executing under the
sandbox uids. This must be done before any chownToBuilder()
calls. */
killSandbox(false);
/* Right platform? */
if (!parsedDrv->canBuildLocally(worker.store))
throw Error("a '%s' with features {%s} is required to build '%s', but I am a '%s' with features {%s}",
@ -376,35 +467,6 @@ void LocalDerivationGoal::startBuilder()
additionalSandboxProfile = parsedDrv->getStringAttr("__sandboxProfile").value_or("");
#endif
/* Are we doing a chroot build? */
{
auto noChroot = parsedDrv->getBoolAttr("__noChroot");
if (settings.sandboxMode == smEnabled) {
if (noChroot)
throw Error("derivation '%s' has '__noChroot' set, "
"but that's not allowed when 'sandbox' is 'true'", worker.store.printStorePath(drvPath));
#if __APPLE__
if (additionalSandboxProfile != "")
throw Error("derivation '%s' specifies a sandbox profile, "
"but this is only allowed when 'sandbox' is 'relaxed'", worker.store.printStorePath(drvPath));
#endif
useChroot = true;
}
else if (settings.sandboxMode == smDisabled)
useChroot = false;
else if (settings.sandboxMode == smRelaxed)
useChroot = derivationType.isSandboxed() && !noChroot;
}
auto & localStore = getLocalStore();
if (localStore.storeDir != localStore.realStoreDir.get()) {
#if __linux__
useChroot = true;
#else
throw Error("building using a diverted store is not supported on this platform");
#endif
}
/* Create a temporary directory where the build will take
place. */
tmpDir = createTempDir("", "nix-build-" + std::string(drvPath.name()), false, false, 0700);
@ -580,10 +642,11 @@ void LocalDerivationGoal::startBuilder()
printMsg(lvlChatty, format("setting up chroot environment in '%1%'") % chrootRootDir);
if (mkdir(chrootRootDir.c_str(), 0750) == -1)
// FIXME: make this 0700
if (mkdir(chrootRootDir.c_str(), buildUser && buildUser->getUIDCount() != 1 ? 0755 : 0750) == -1)
throw SysError("cannot create '%1%'", chrootRootDir);
if (buildUser && chown(chrootRootDir.c_str(), 0, buildUser->getGID()) == -1)
if (buildUser && chown(chrootRootDir.c_str(), buildUser->getUIDCount() != 1 ? buildUser->getUID() : 0, buildUser->getGID()) == -1)
throw SysError("cannot change ownership of '%1%'", chrootRootDir);
/* Create a writable /tmp in the chroot. Many builders need
@ -597,6 +660,10 @@ void LocalDerivationGoal::startBuilder()
nobody account. The latter is kind of a hack to support
Samba-in-QEMU. */
createDirs(chrootRootDir + "/etc");
chownToBuilder(chrootRootDir + "/etc");
if (parsedDrv->useUidRange() && (!buildUser || buildUser->getUIDCount() < 65536))
throw Error("feature 'uid-range' requires the setting '%s' to be enabled", settings.autoAllocateUids.name);
/* Declare the build user's group so that programs get a consistent
view of the system (e.g., "id -gn"). */
@ -647,12 +714,28 @@ void LocalDerivationGoal::startBuilder()
dirsInChroot.erase(worker.store.printStorePath(*i.second.second));
}
#elif __APPLE__
/* We don't really have any parent prep work to do (yet?)
All work happens in the child, instead. */
if (cgroup) {
if (mkdir(cgroup->c_str(), 0755) != 0)
throw SysError("creating cgroup '%s'", *cgroup);
chownToBuilder(*cgroup);
chownToBuilder(*cgroup + "/cgroup.procs");
chownToBuilder(*cgroup + "/cgroup.threads");
//chownToBuilder(*cgroup + "/cgroup.subtree_control");
}
#else
throw Error("sandboxing builds is not supported on this platform");
if (parsedDrv->useUidRange())
throw Error("feature 'uid-range' is not supported on this platform");
#if __APPLE__
/* We don't really have any parent prep work to do (yet?)
All work happens in the child, instead. */
#else
throw Error("sandboxing builds is not supported on this platform");
#endif
#endif
} else {
if (parsedDrv->useUidRange())
throw Error("feature 'uid-range' is only supported in sandboxed builds");
}
if (needsHashRewrite() && pathExists(homeDir))
@ -913,14 +996,16 @@ void LocalDerivationGoal::startBuilder()
the calling user (if build users are disabled). */
uid_t hostUid = buildUser ? buildUser->getUID() : getuid();
uid_t hostGid = buildUser ? buildUser->getGID() : getgid();
uid_t nrIds = buildUser ? buildUser->getUIDCount() : 1;
writeFile("/proc/" + std::to_string(pid) + "/uid_map",
fmt("%d %d 1", sandboxUid(), hostUid));
fmt("%d %d %d", sandboxUid(), hostUid, nrIds));
writeFile("/proc/" + std::to_string(pid) + "/setgroups", "deny");
if (!buildUser || buildUser->getUIDCount() == 1)
writeFile("/proc/" + std::to_string(pid) + "/setgroups", "deny");
writeFile("/proc/" + std::to_string(pid) + "/gid_map",
fmt("%d %d 1", sandboxGid(), hostGid));
fmt("%d %d %d", sandboxGid(), hostGid, nrIds));
} else {
debug("note: not using a user namespace");
if (!buildUser)
@ -947,6 +1032,10 @@ void LocalDerivationGoal::startBuilder()
throw SysError("getting sandbox user namespace");
}
/* Move the child into its own cgroup. */
if (cgroup)
writeFile(*cgroup + "/cgroup.procs", fmt("%d", (pid_t) pid));
/* Signal the builder that we've updated its user namespace. */
writeFull(userNamespaceSync.writeSide.get(), "1");
@ -1779,6 +1868,13 @@ void LocalDerivationGoal::runChild()
if (mount("none", (chrootRootDir + "/proc").c_str(), "proc", 0, 0) == -1)
throw SysError("mounting /proc");
/* Mount sysfs on /sys. */
if (buildUser && buildUser->getUIDCount() != 1) {
createDirs(chrootRootDir + "/sys");
if (mount("none", (chrootRootDir + "/sys").c_str(), "sysfs", 0, 0) == -1)
throw SysError("mounting /sys");
}
/* Mount a new tmpfs on /dev/shm to ensure that whatever
the builder puts in /dev/shm is cleaned up automatically. */
if (pathExists("/dev/shm") && mount("none", (chrootRootDir + "/dev/shm").c_str(), "tmpfs", 0,
@ -1821,6 +1917,12 @@ void LocalDerivationGoal::runChild()
if (unshare(CLONE_NEWNS) == -1)
throw SysError("unsharing mount namespace");
/* Unshare the cgroup namespace. This means
/proc/self/cgroup will show the child's cgroup as '/'
rather than whatever it is in the parent. */
if (cgroup && unshare(CLONE_NEWCGROUP) == -1)
throw SysError("unsharing cgroup namespace");
/* Do the chroot(). */
if (chdir(chrootRootDir.c_str()) == -1)
throw SysError("cannot change directory to '%1%'", chrootRootDir);
@ -1906,9 +2008,8 @@ void LocalDerivationGoal::runChild()
if (setUser && buildUser) {
/* Preserve supplementary groups of the build user, to allow
admins to specify groups such as "kvm". */
if (!buildUser->getSupplementaryGIDs().empty() &&
setgroups(buildUser->getSupplementaryGIDs().size(),
buildUser->getSupplementaryGIDs().data()) == -1)
auto gids = buildUser->getSupplementaryGIDs();
if (setgroups(gids.size(), gids.data()) == -1)
throw SysError("cannot set supplementary groups of build user");
if (setgid(buildUser->getGID()) == -1 ||
@ -2237,7 +2338,10 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
/* Canonicalise first. This ensures that the path we're
rewriting doesn't contain a hard link to /etc/shadow or
something like that. */
canonicalisePathMetaData(actualPath, buildUser ? buildUser->getUID() : -1, inodesSeen);
canonicalisePathMetaData(
actualPath,
buildUser ? std::optional(buildUser->getUIDRange()) : std::nullopt,
inodesSeen);
debug("scanning for references for output '%s' in temp location '%s'", outputName, actualPath);
@ -2330,6 +2434,10 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
sink.s = rewriteStrings(sink.s, outputRewrites);
StringSource source(sink.s);
restorePath(actualPath, source);
/* FIXME: set proper permissions in restorePath() so
we don't have to do another traversal. */
canonicalisePathMetaData(actualPath, {}, inodesSeen);
}
};
@ -2492,7 +2600,7 @@ DrvOutputs LocalDerivationGoal::registerOutputs()
/* FIXME: set proper permissions in restorePath() so
we don't have to do another traversal. */
canonicalisePathMetaData(actualPath, -1, inodesSeen);
canonicalisePathMetaData(actualPath, {}, inodesSeen);
/* Calculate where we'll move the output files. In the checking case we
will leave leave them where they are, for now, rather than move to

View file

@ -15,6 +15,9 @@ struct LocalDerivationGoal : public DerivationGoal
/* The process ID of the builder. */
Pid pid;
/* The cgroup of the builder, if any. */
std::optional<Path> cgroup;
/* The temporary directory. */
Path tmpDir;
@ -92,8 +95,8 @@ struct LocalDerivationGoal : public DerivationGoal
result. */
std::map<Path, ValidPathInfo> prevInfos;
uid_t sandboxUid() { return usingUserNamespace ? 1000 : buildUser->getUID(); }
gid_t sandboxGid() { return usingUserNamespace ? 100 : buildUser->getGID(); }
uid_t sandboxUid() { return usingUserNamespace ? (!buildUser || buildUser->getUIDCount() == 1 ? 1000 : 0) : buildUser->getUID(); }
gid_t sandboxGid() { return usingUserNamespace ? (!buildUser || buildUser->getUIDCount() == 1 ? 100 : 0) : buildUser->getGID(); }
const static Path homeDir;
@ -197,6 +200,10 @@ struct LocalDerivationGoal : public DerivationGoal
/* Forcibly kill the child process, if any. */
void killChild() override;
/* Kill any processes running under the build user UID or in the
cgroup of the build. */
void killSandbox(bool getStats);
/* Create alternative path calculated from but distinct from the
input, so we can avoid overwriting outputs (or other store paths)
that already exist. */