#include "gc-store.hh" #include "signals.hh" #include "platform/darwin.hh" #include "regex.hh" #include #include #include #include namespace nix { void DarwinLocalStore::findPlatformRoots(UncheckedRoots & unchecked) { auto storePathRegex = regex::storePathRegex(storeDir); std::vector pids; int pidBufSize = 1; while (pidBufSize > pids.size() * sizeof(int)) { // Reserve some extra size so we don't fail too much pids.resize((pidBufSize + pidBufSize / 8) / sizeof(int)); pidBufSize = proc_listpids(PROC_ALL_PIDS, 0, pids.data(), pids.size() * sizeof(int)); if (pidBufSize <= 0) { throw SysError("Listing PIDs"); } } pids.resize(pidBufSize / sizeof(int)); for (auto pid : pids) { // It doesn't make sense to ask about the kernel if (pid == 0) { continue; } try { // Process cwd/root directory struct proc_vnodepathinfo vnodeInfo; if (proc_pidinfo(pid, PROC_PIDVNODEPATHINFO, 0, &vnodeInfo, sizeof(vnodeInfo)) <= 0) { throw SysError("Getting pid %1% working directory", pid); } unchecked[std::string(vnodeInfo.pvi_cdir.vip_path)].emplace(fmt("{libproc/%d/cwd}", pid) ); unchecked[std::string(vnodeInfo.pvi_rdir.vip_path)].emplace( fmt("{libproc/%d/rootdir}", pid) ); // File descriptors std::vector fds; int fdBufSize = 1; while (fdBufSize > fds.size() * sizeof(struct proc_fdinfo)) { // Reserve some extra size so we don't fail too much fds.resize((fdBufSize + fdBufSize / 8) / sizeof(struct proc_fdinfo)); fdBufSize = proc_pidinfo( pid, PROC_PIDLISTFDS, 0, fds.data(), fds.size() * sizeof(struct proc_fdinfo) ); if (fdBufSize <= 0) { throw SysError("Listing pid %1% file descriptors", pid); } } fds.resize(fdBufSize / sizeof(struct proc_fdinfo)); for (auto fd : fds) { // By definition, only a vnode is on the filesystem if (fd.proc_fdtype != PROX_FDTYPE_VNODE) { continue; } struct vnode_fdinfowithpath fdInfo; if (proc_pidfdinfo( pid, fd.proc_fd, PROC_PIDFDVNODEPATHINFO, &fdInfo, sizeof(fdInfo) ) <= 0) { // They probably just closed this fd, no need to cancel looking at ranges and // arguments if (errno == EBADF) { continue; } throw SysError("Getting pid %1% fd %2% path", pid, fd.proc_fd); } unchecked[std::string(fdInfo.pvip.vip_path)].emplace( fmt("{libproc/%d/fd/%d}", pid, fd.proc_fd) ); } // Regions (e.g. mmapped files, executables, shared libraries) uint64_t nextAddr = 0; while (true) { // Seriously, what are you doing XNU? // There's 3 flavors of PROC_PIDREGIONPATHINFO: // * PROC_PIDREGIONPATHINFO includes all regions // * PROC_PIDREGIONPATHINFO2 includes regions backed by a vnode // * PROC_PIDREGIONPATHINFO3 includes regions backed by a vnode on a specified // filesystem Only PROC_PIDREGIONPATHINFO is documented. Unfortunately, using it // would make finding gcroots take about 100x as long and tests would fail from // timeout. According to the Frida source code, PROC_PIDREGIONPATHINFO2 has been // available since XNU 2782.1.97 in OS X 10.10 // // 22 means PROC_PIDREGIONPATHINFO2 struct proc_regionwithpathinfo regionInfo; if (proc_pidinfo(pid, 22, nextAddr, ®ionInfo, sizeof(regionInfo)) <= 0) { // PROC_PIDREGIONPATHINFO signals we're done with an error, // so we're expected to hit this once per process if (errno == ESRCH || errno == EINVAL) { break; } throw SysError("Getting pid %1% region path", pid); } unchecked[std::string(regionInfo.prp_vip.vip_path)].emplace( fmt("{libproc/%d/region}", pid) ); nextAddr = regionInfo.prp_prinfo.pri_address + regionInfo.prp_prinfo.pri_size; } // Arguments and environment variables // We can't read environment variables of binaries with entitlements unless // nix has the `com.apple.private.read-environment-variables` entitlement or SIP is off // We can read arguments for all applications though. // Yes, it's a sysctl, the proc_info and sysctl APIs are mostly similar, // but both have exclusive capabilities int sysctlName[3] = {CTL_KERN, KERN_PROCARGS2, pid}; size_t argsSize = 0; if (sysctl(sysctlName, 3, nullptr, &argsSize, nullptr, 0) < 0) { throw SysError("Reading pid %1% arguments", pid); } std::vector args(argsSize); if (sysctl(sysctlName, 3, args.data(), &argsSize, nullptr, 0) < 0) { throw SysError("Reading pid %1% arguments", pid); } if (argsSize < args.size()) { args.resize(argsSize); } // We have these perfectly nice arguments, but have to ignore them because // otherwise we'd see arguments to nix-store commands and // `nix-store --delete /nix/store/whatever` would always fail // First 4 bytes are an int of argc. if (args.size() < sizeof(int)) { continue; } auto argc = reinterpret_cast(args.data())[0]; auto argsIter = args.begin(); std::advance(argsIter, sizeof(int)); // Executable then argc args, each separated by some number of null bytes for (int i = 0; argsIter != args.end() && i < argc + 1; i++) { argsIter = std::find(argsIter, args.end(), '\0'); argsIter = std::find_if(argsIter, args.end(), [](char ch) { return ch != '\0'; }); } if (argsIter != args.end()) { auto env_end = std::sregex_iterator{}; for (auto i = std::sregex_iterator{argsIter, args.end(), storePathRegex}; i != env_end; ++i) { unchecked[i->str()].emplace(fmt("{libproc/%d/environ}", pid)); } }; // Per-thread working directories struct proc_taskallinfo taskAllInfo; if (proc_pidinfo(pid, PROC_PIDTASKALLINFO, 0, &taskAllInfo, sizeof(taskAllInfo)) <= 0) { throw SysError("Reading pid %1% tasks", pid); } // If the process doesn't have the per-thread cwd flag then we already have the // process-wide cwd from PROC_PIDVNODEPATHINFO if (taskAllInfo.pbsd.pbi_flags & PROC_FLAG_THCWD) { std::vector tids(taskAllInfo.ptinfo.pti_threadnum); int tidBufSize = proc_pidinfo( pid, PROC_PIDLISTTHREADS, 0, tids.data(), tids.size() * sizeof(uint64_t) ); if (tidBufSize <= 0) { throw SysError("Listing pid %1% threads", pid); } for (auto tid : tids) { struct proc_threadwithpathinfo threadPathInfo; if (proc_pidinfo( pid, PROC_PIDTHREADPATHINFO, tid, &threadPathInfo, sizeof(threadPathInfo) ) <= 0) { throw SysError("Reading pid %1% thread %2% cwd", pid, tid); } unchecked[std::string(threadPathInfo.pvip.vip_path)].emplace( fmt("{libproc/%d/thread/%d/cwd}", pid, tid) ); } } } catch (SysError & e) { // ENOENT/ESRCH: Process no longer exists (proc_info) // EINVAL: Process no longer exists (sysctl) // EACCESS/EPERM: We don't have permission to read this field (proc_info) // EIO: Kernel failed to read from target process memory during KERN_PROCARGS2 (sysctl) if (errno == ENOENT || errno == ESRCH || errno == EINVAL || errno == EACCES || errno == EPERM || errno == EIO) { continue; } throw; } } } }