Dynamically allocate UIDs

Rather than rely on a nixbld group, we now allocate UIDs/GIDs
dynamically starting at a configurable ID (872415232 by default).

Also, we allocate 2^18 UIDs and GIDs per build, and run the build as
root in its UID namespace. (This should not be the default since it
breaks some builds. We probably should enable this conditional on a
requiredSystemFeature.) The goal is to be able to run (NixOS)
containers in a build. However, this will also require some cgroup
initialisation.

The 2^18 UIDs/GIDs is intended to provide enough ID space to run
multiple containers per build, e.g. for distributed NixOS tests.
This commit is contained in:
Eelco Dolstra 2017-10-31 12:22:29 +01:00
parent 14227aeb32
commit 836573a9a2
2 changed files with 61 additions and 8 deletions

View file

@ -513,7 +513,6 @@ private:
AutoCloseFD fdUserLock; AutoCloseFD fdUserLock;
bool isEnabled = false; bool isEnabled = false;
string user;
uid_t uid = 0; uid_t uid = 0;
gid_t gid = 0; gid_t gid = 0;
std::vector<gid_t> supplementaryGIDs; std::vector<gid_t> supplementaryGIDs;
@ -523,9 +522,9 @@ public:
void kill(); void kill();
string getUser() { return user; }
uid_t getUID() { assert(uid); return uid; } uid_t getUID() { assert(uid); return uid; }
uid_t getGID() { assert(gid); return gid; } gid_t getGID() { assert(gid); return gid; }
uint32_t getIDCount() { return 1; }
std::vector<gid_t> getSupplementaryGIDs() { return supplementaryGIDs; } std::vector<gid_t> getSupplementaryGIDs() { return supplementaryGIDs; }
bool findFreeUser(); bool findFreeUser();
@ -537,13 +536,16 @@ public:
UserLock::UserLock() UserLock::UserLock()
{ {
#if 0
assert(settings.buildUsersGroup != ""); assert(settings.buildUsersGroup != "");
createDirs(settings.nixStateDir + "/userpool"); createDirs(settings.nixStateDir + "/userpool");
#endif
} }
bool UserLock::findFreeUser() { bool UserLock::findFreeUser() {
if (enabled()) return true; if (enabled()) return true;
#if 0
/* Get the members of the build-users-group. */ /* Get the members of the build-users-group. */
struct group * gr = getgrnam(settings.buildUsersGroup.get().c_str()); struct group * gr = getgrnam(settings.buildUsersGroup.get().c_str());
if (!gr) if (!gr)
@ -607,12 +609,46 @@ bool UserLock::findFreeUser() {
} }
} }
return false;
#endif
assert(settings.startId > 0);
assert(settings.startId % settings.idsPerBuild == 0);
assert(settings.uidCount % settings.idsPerBuild == 0);
assert((uint64_t) settings.startId + (uint64_t) settings.uidCount <= std::numeric_limits<uid_t>::max());
// FIXME: check whether the id range overlaps any known users
size_t nrSlots = settings.uidCount / settings.idsPerBuild;
for (size_t i = 0; i < nrSlots; i++) {
debug("trying user slot '%d'", i);
createDirs(settings.nixStateDir + "/userpool");
fnUserLock = fmt("%s/userpool/slot-%d", settings.nixStateDir, i);
AutoCloseFD fd = open(fnUserLock.c_str(), O_RDWR | O_CREAT | O_CLOEXEC, 0600);
if (!fd)
throw SysError("opening user lock '%1%'", fnUserLock);
if (lockFile(fd.get(), ltWrite, false)) {
fdUserLock = std::move(fd);
uid = settings.startId + i * settings.idsPerBuild;
gid = settings.startId + i * settings.idsPerBuild;
return true;
}
}
return false; return false;
} }
void UserLock::kill() void UserLock::kill()
{ {
// FIXME: use a cgroup to kill all processes in the build?
#if 0
killUser(uid); killUser(uid);
#endif
} }
@ -1523,7 +1559,7 @@ void DerivationGoal::tryLocalBuild() {
/* If `build-users-group' is not empty, then we have to build as /* If `build-users-group' is not empty, then we have to build as
one of the members of that group. */ one of the members of that group. */
if (settings.buildUsersGroup != "" && getuid() == 0) { if ((settings.buildUsersGroup != "" || settings.startId.get() != 0) && getuid() == 0) {
#if defined(__linux__) || defined(__APPLE__) #if defined(__linux__) || defined(__APPLE__)
if (!buildUser) buildUser = std::make_unique<UserLock>(); if (!buildUser) buildUser = std::make_unique<UserLock>();
@ -2129,7 +2165,7 @@ void DerivationGoal::startBuilder()
printMsg(lvlChatty, format("setting up chroot environment in '%1%'") % chrootRootDir); printMsg(lvlChatty, format("setting up chroot environment in '%1%'") % chrootRootDir);
if (mkdir(chrootRootDir.c_str(), 0750) == -1) if (mkdir(chrootRootDir.c_str(), 0755) == -1)
throw SysError("cannot create '%1%'", chrootRootDir); throw SysError("cannot create '%1%'", chrootRootDir);
if (buildUser && chown(chrootRootDir.c_str(), 0, buildUser->getGID()) == -1) if (buildUser && chown(chrootRootDir.c_str(), 0, buildUser->getGID()) == -1)
@ -2444,14 +2480,15 @@ void DerivationGoal::startBuilder()
the calling user (if build users are disabled). */ the calling user (if build users are disabled). */
uid_t hostUid = buildUser ? buildUser->getUID() : getuid(); uid_t hostUid = buildUser ? buildUser->getUID() : getuid();
uid_t hostGid = buildUser ? buildUser->getGID() : getgid(); uid_t hostGid = buildUser ? buildUser->getGID() : getgid();
uint32_t nrIds = settings.idsPerBuild; // FIXME
writeFile("/proc/" + std::to_string(pid) + "/uid_map", writeFile("/proc/" + std::to_string(pid) + "/uid_map",
(format("%d %d 1") % sandboxUid % hostUid).str()); fmt("%d %d %d", /* sandboxUid */ 0, hostUid, nrIds));
writeFile("/proc/" + std::to_string(pid) + "/setgroups", "deny"); //writeFile("/proc/" + std::to_string(pid) + "/setgroups", "deny");
writeFile("/proc/" + std::to_string(pid) + "/gid_map", writeFile("/proc/" + std::to_string(pid) + "/gid_map",
(format("%d %d 1") % sandboxGid % hostGid).str()); fmt("%d %d %d", /* sandboxGid */ 0, hostGid, nrIds));
/* Save the mount namespace of the child. We have to do this /* Save the mount namespace of the child. We have to do this
*before* the child does a chroot. */ *before* the child does a chroot. */
@ -3306,10 +3343,16 @@ void DerivationGoal::runChild()
/* Switch to the sandbox uid/gid in the user namespace, /* Switch to the sandbox uid/gid in the user namespace,
which corresponds to the build user or calling user in which corresponds to the build user or calling user in
the parent namespace. */ the parent namespace. */
#if 0
if (setgid(sandboxGid) == -1) if (setgid(sandboxGid) == -1)
throw SysError("setgid failed"); throw SysError("setgid failed");
if (setuid(sandboxUid) == -1) if (setuid(sandboxUid) == -1)
throw SysError("setuid failed"); throw SysError("setuid failed");
#endif
if (setgid(0) == -1)
throw SysError("setgid failed");
if (setuid(0) == -1)
throw SysError("setuid failed");
setUser = false; setUser = false;
} }

View file

@ -148,6 +148,16 @@ public:
Setting<std::string> buildUsersGroup{this, "", "build-users-group", Setting<std::string> buildUsersGroup{this, "", "build-users-group",
"The Unix group that contains the build users."}; "The Unix group that contains the build users."};
#if __linux__
const uint32_t idsPerBuild = 1 << 18;
Setting<uint32_t> startId{this, 872415232, "start-id",
"The first UID and GID to use for dynamic ID allocation. (0 means disable.)"};
Setting<uint32_t> uidCount{this, idsPerBuild * 128, "id-count",
"The number of UIDs/GIDs to use for dynamic ID allocation."};
#endif
Setting<bool> impersonateLinux26{this, false, "impersonate-linux-26", Setting<bool> impersonateLinux26{this, false, "impersonate-linux-26",
"Whether to impersonate a Linux 2.6 machine on newer kernels.", "Whether to impersonate a Linux 2.6 machine on newer kernels.",
{"build-impersonate-linux-26"}}; {"build-impersonate-linux-26"}};