From c5f27a7cb2ec816f483a65255034a1b57a8aa221 Mon Sep 17 00:00:00 2001 From: Narayan Kamath Date: Fri, 19 Aug 2016 13:45:24 +0100 Subject: [PATCH] Reopen whitelisted zygote file descriptors after a fork. We don't want these descriptors to be shared post-fork, so we'll have to close and reopen them when the zygote forks. The set of open descriptors is checked against a whitelist and it is a fatal error if a non whitelisted FD is opened. It is also a fatal error if anything other than a regular file / character device or socket is opened at the time of forking. This work is done in two stages : - An initial list of FDs is constructed and cached prior to the first zygote fork. - On each subsequent fork, we check whether the list of open FDs has changed. We are currently tolerant of changes, but in the longer term, it should be a fatal error if the set of open file descriptors in the zygote changes. - Post fork, we traverse the list of open descriptors and reopen them if necessary. bug: 30963384 Change-Id: Icfd45c96c660491b554d146a9d70d97dbcc712bc --- core/jni/com_android_internal_os_Zygote.cpp | 22 ++ core/jni/fd_utils-inl.h | 448 ++++++++++++++++++++++++++++ 2 files changed, 470 insertions(+) create mode 100644 core/jni/fd_utils-inl.h diff --git a/core/jni/com_android_internal_os_Zygote.cpp b/core/jni/com_android_internal_os_Zygote.cpp index a04fc2a70686..3207b3c6f187 100644 --- a/core/jni/com_android_internal_os_Zygote.cpp +++ b/core/jni/com_android_internal_os_Zygote.cpp @@ -54,6 +54,7 @@ #include "ScopedLocalRef.h" #include "ScopedPrimitiveArray.h" #include "ScopedUtfChars.h" +#include "fd_utils-inl.h" #include "nativebridge/native_bridge.h" @@ -436,6 +437,9 @@ static void SetForkLoad(bool boost) { } #endif +// The list of open zygote file descriptors. +static FileDescriptorTable* gOpenFdTable = NULL; + // Utility routine to fork zygote and specialize the child process. static pid_t ForkAndSpecializeCommon(JNIEnv* env, uid_t uid, gid_t gid, jintArray javaGids, jint debug_flags, jobjectArray javaRlimits, @@ -450,6 +454,18 @@ static pid_t ForkAndSpecializeCommon(JNIEnv* env, uid_t uid, gid_t gid, jintArra SetForkLoad(true); #endif + // If this is the first fork for this zygote, create the open FD table. + // If it isn't, we just need to check whether the list of open files has + // changed (and it shouldn't in the normal case). + if (gOpenFdTable == NULL) { + gOpenFdTable = FileDescriptorTable::Create(); + if (gOpenFdTable == NULL) { + RuntimeAbort(env, __LINE__, "Unable to construct file descriptor table."); + } + } else if (!gOpenFdTable->Restat()) { + RuntimeAbort(env, __LINE__, "Unable to restat file descriptor table."); + } + pid_t pid = fork(); if (pid == 0) { @@ -459,6 +475,12 @@ static pid_t ForkAndSpecializeCommon(JNIEnv* env, uid_t uid, gid_t gid, jintArra // Clean up any descriptors which must be closed immediately DetachDescriptors(env, fdsToClose); + // Re-open all remaining open file descriptors so that they aren't shared + // with the zygote across a fork. + if (!gOpenFdTable->Reopen()) { + RuntimeAbort(env, __LINE__, "Unable to reopen whitelisted descriptors."); + } + // Keep capabilities across UID change, unless we're staying root. if (uid != 0) { EnableKeepCapabilities(env); diff --git a/core/jni/fd_utils-inl.h b/core/jni/fd_utils-inl.h new file mode 100644 index 000000000000..bd68625514c7 --- /dev/null +++ b/core/jni/fd_utils-inl.h @@ -0,0 +1,448 @@ +/* + * Copyright (C) 2016 The Android Open Source Project + * + * 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. + */ + +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include "JNIHelp.h" +#include "ScopedPrimitiveArray.h" + +// Whitelist of open paths that the zygote is allowed to keep open +// that will be recreated across forks. In addition to files here, +// all files ending with ".jar" under /system/framework" are +// whitelisted. See FileDescriptorInfo::IsWhitelisted for the +// canonical definition. +static const char* kPathWhitelist[] = { + "/dev/null", + "/dev/pmsg0", + "/system/etc/event-log-tags", + "/sys/kernel/debug/tracing/trace_marker", + "/system/framework/framework-res.apk", + "/dev/urandom", + "/dev/ion" +}; + +static const char* kFdPath = "/proc/self/fd"; + +// Keeps track of all relevant information (flags, offset etc.) of an +// open zygote file descriptor. +class FileDescriptorInfo { + public: + // Create a FileDescriptorInfo for a given file descriptor. Returns + // |NULL| if an error occurred. + static FileDescriptorInfo* createFromFd(int fd) { + struct stat f_stat; + // This should never happen; the zygote should always have the right set + // of permissions required to stat all its open files. + if (TEMP_FAILURE_RETRY(fstat(fd, &f_stat)) == -1) { + ALOGE("Unable to stat fd %d : %s", fd, strerror(errno)); + return NULL; + } + + // Ignore (don't reopen or fail on) socket fds for now. + if (S_ISSOCK(f_stat.st_mode)) { + ALOGW("Unsupported socket FD: %d, ignoring.", fd); + return new FileDescriptorInfo(fd); + } + + // We only handle whitelisted regular files and character devices. Whitelisted + // character devices must provide a guarantee of sensible behaviour when + // reopened. + // + // S_ISDIR : Not supported. (We could if we wanted to, but it's unused). + // S_ISLINK : Not supported. + // S_ISBLK : Not supported. + // S_ISFIFO : Not supported. Note that the zygote uses pipes to communicate + // with the child process across forks but those should have been closed + // before we got to this point. + if (!S_ISCHR(f_stat.st_mode) && !S_ISREG(f_stat.st_mode)) { + ALOGE("Unsupported st_mode %d", f_stat.st_mode); + return NULL; + } + + std::string file_path; + if (!Readlink(fd, &file_path)) { + return NULL; + } + + if (!IsWhitelisted(file_path)) { + ALOGE("Not whitelisted : %s", file_path.c_str()); + return NULL; + } + + // File descriptor flags : currently on FD_CLOEXEC. We can set these + // using F_SETFD - we're single threaded at this point of execution so + // there won't be any races. + const int fd_flags = TEMP_FAILURE_RETRY(fcntl(fd, F_GETFD)); + if (fd_flags == -1) { + ALOGE("Failed fcntl(%d, F_GETFD) : %s", fd, strerror(errno)); + return NULL; + } + + // File status flags : + // - File access mode : (O_RDONLY, O_WRONLY...) we'll pass these through + // to the open() call. + // + // - File creation flags : (O_CREAT, O_EXCL...) - there's not much we can + // do about these, since the file has already been created. We shall ignore + // them here. + // + // - Other flags : We'll have to set these via F_SETFL. On linux, F_SETFL + // can only set O_APPEND, O_ASYNC, O_DIRECT, O_NOATIME, and O_NONBLOCK. + // In particular, it can't set O_SYNC and O_DSYNC. We'll have to test for + // their presence and pass them in to open(). + int fs_flags = TEMP_FAILURE_RETRY(fcntl(fd, F_GETFL)); + if (fs_flags == -1) { + ALOGE("Failed fcntl(%d, F_GETFL) : %s", fd, strerror(errno)); + return NULL; + } + + // File offset : Ignore the offset for non seekable files. + const off_t offset = TEMP_FAILURE_RETRY(lseek64(fd, 0, SEEK_CUR)); + + // We pass the flags that open accepts to open, and use F_SETFL for + // the rest of them. + static const int kOpenFlags = (O_RDONLY | O_WRONLY | O_RDWR | O_DSYNC | O_SYNC); + int open_flags = fs_flags & (kOpenFlags); + fs_flags = fs_flags & (~(kOpenFlags)); + + return new FileDescriptorInfo(f_stat, file_path, fd, open_flags, fd_flags, fs_flags, offset); + } + + // Checks whether the file descriptor associated with this object + // refers to the same description. + bool Restat() const { + struct stat f_stat; + if (TEMP_FAILURE_RETRY(fstat(fd, &f_stat)) == -1) { + return false; + } + + return f_stat.st_ino == stat.st_ino && f_stat.st_dev == stat.st_dev; + } + + bool Reopen() const { + // Always skip over socket FDs for now. + if (is_sock) { + return true; + } + + // NOTE: This might happen if the file was unlinked after being opened. + // It's a common pattern in the case of temporary files and the like but + // we should not allow such usage from the zygote. + const int new_fd = TEMP_FAILURE_RETRY(open(file_path.c_str(), open_flags)); + + if (new_fd == -1) { + ALOGE("Failed open(%s, %d) : %s", file_path.c_str(), open_flags, strerror(errno)); + return false; + } + + if (TEMP_FAILURE_RETRY(fcntl(new_fd, F_SETFD, fd_flags)) == -1) { + close(new_fd); + ALOGE("Failed fcntl(%d, F_SETFD, %x) : %s", new_fd, fd_flags, strerror(errno)); + return false; + } + + if (TEMP_FAILURE_RETRY(fcntl(new_fd, F_SETFL, fs_flags)) == -1) { + close(new_fd); + ALOGE("Failed fcntl(%d, F_SETFL, %x) : %s", new_fd, fs_flags, strerror(errno)); + return false; + } + + if (offset != -1 && TEMP_FAILURE_RETRY(lseek64(new_fd, offset, SEEK_SET)) == -1) { + close(new_fd); + ALOGE("Failed lseek64(%d, SEEK_SET) : %s", new_fd, strerror(errno)); + return false; + } + + if (TEMP_FAILURE_RETRY(dup2(new_fd, fd)) == -1) { + close(new_fd); + ALOGE("Failed dup2(%d, %d) : %s", fd, new_fd, strerror(errno)); + return false; + } + + if (close(new_fd) == -1) { + ALOGW("Failed close(%d) : %s", new_fd, strerror(errno)); + } + + return true; + } + + const int fd; + const struct stat stat; + const std::string file_path; + const int open_flags; + const int fd_flags; + const int fs_flags; + const off_t offset; + const bool is_sock; + + private: + FileDescriptorInfo(int fd) : + fd(fd), + stat(), + open_flags(0), + fd_flags(0), + fs_flags(0), + offset(0), + is_sock(true) { + } + + FileDescriptorInfo(struct stat stat, const std::string& file_path, int fd, int open_flags, + int fd_flags, int fs_flags, off_t offset) : + fd(fd), + stat(stat), + file_path(file_path), + open_flags(open_flags), + fd_flags(fd_flags), + fs_flags(fs_flags), + offset(offset), + is_sock(false) { + } + + // Returns true iff. a given path is whitelisted. A path is whitelisted + // if it belongs to the whitelist (see kPathWhitelist) or if it's a path + // under /system/framework that ends with ".jar". + static bool IsWhitelisted(const std::string& path) { + for (size_t i = 0; i < (sizeof(kPathWhitelist) / sizeof(kPathWhitelist[0])); ++i) { + if (kPathWhitelist[i] == path) { + return true; + } + } + + static const std::string kFrameworksPrefix = "/system/framework/"; + static const std::string kJarSuffix = ".jar"; + if (path.compare(0, kFrameworksPrefix.size(), kFrameworksPrefix) == 0 && + path.compare(path.size() - kJarSuffix.size(), kJarSuffix.size(), kJarSuffix) == 0) { + return true; + } + return false; + } + + // TODO: Call android::base::Readlink instead of copying the code here. + static bool Readlink(const int fd, std::string* result) { + char path[64]; + snprintf(path, sizeof(path), "/proc/self/fd/%d", fd); + + // Code copied from android::base::Readlink starts here : + + // Annoyingly, the readlink system call returns EINVAL for a zero-sized buffer, + // and truncates to whatever size you do supply, so it can't be used to query. + // We could call lstat first, but that would introduce a race condition that + // we couldn't detect. + // ext2 and ext4 both have PAGE_SIZE limitations, so we assume that here. + char buf[4096]; + ssize_t len = readlink(path, buf, sizeof(buf)); + if (len == -1) return false; + + result->assign(buf, len); + return true; + } + + DISALLOW_COPY_AND_ASSIGN(FileDescriptorInfo); +}; + +// A FileDescriptorTable is a collection of FileDescriptorInfo objects +// keyed by their FDs. +class FileDescriptorTable { + public: + // Creates a new FileDescriptorTable. This function scans + // /proc/self/fd for the list of open file descriptors and collects + // information about them. Returns NULL if an error occurs. + static FileDescriptorTable* Create() { + DIR* d = opendir(kFdPath); + if (d == NULL) { + ALOGE("Unable to open directory %s: %s", kFdPath, strerror(errno)); + return NULL; + } + int dir_fd = dirfd(d); + dirent* e; + + std::unordered_map open_fd_map; + while ((e = readdir(d)) != NULL) { + const int fd = ParseFd(e, dir_fd); + if (fd == -1) { + continue; + } + + FileDescriptorInfo* info = FileDescriptorInfo::createFromFd(fd); + if (info == NULL) { + if (closedir(d) == -1) { + ALOGE("Unable to close directory : %s", strerror(errno)); + } + return NULL; + } + open_fd_map[fd] = info; + } + + if (closedir(d) == -1) { + ALOGE("Unable to close directory : %s", strerror(errno)); + return NULL; + } + return new FileDescriptorTable(open_fd_map); + } + + bool Restat() { + std::set open_fds; + + // First get the list of open descriptors. + DIR* d = opendir(kFdPath); + if (d == NULL) { + ALOGE("Unable to open directory %s: %s", kFdPath, strerror(errno)); + return false; + } + + int dir_fd = dirfd(d); + dirent* e; + while ((e = readdir(d)) != NULL) { + const int fd = ParseFd(e, dir_fd); + if (fd == -1) { + continue; + } + + open_fds.insert(fd); + } + + if (closedir(d) == -1) { + ALOGE("Unable to close directory : %s", strerror(errno)); + return false; + } + + return RestatInternal(open_fds); + } + + // Reopens all file descriptors that are contained in the table. Returns true + // if all descriptors were re-opened, and false if an error occurred. + bool Reopen() { + std::unordered_map::const_iterator it; + for (it = open_fd_map_.begin(); it != open_fd_map_.end(); ++it) { + const FileDescriptorInfo* info = it->second; + if (info == NULL || !info->Reopen()) { + return false; + } + } + + return true; + } + + private: + FileDescriptorTable(const std::unordered_map& map) + : open_fd_map_(map) { + } + + bool RestatInternal(std::set& open_fds) { + bool error = false; + + // Iterate through the list of file descriptors we've already recorded + // and check whether : + // + // (a) they continue to be open. + // (b) they refer to the same file. + std::unordered_map::iterator it; + for (it = open_fd_map_.begin(); it != open_fd_map_.end(); ++it) { + std::set::const_iterator element = open_fds.find(it->first); + if (element == open_fds.end()) { + // The entry from the file descriptor table is no longer in the list + // of open files. We warn about this condition and remove it from + // the list of FDs under consideration. + // + // TODO(narayan): This will be an error in a future android release. + // error = true; + ALOGW("Zygote closed file descriptor %d.", it->first); + open_fd_map_.erase(it); + } else { + // The entry from the file descriptor table is still open. Restat + // it and check whether it refers to the same file. + open_fds.erase(element); + const bool same_file = it->second->Restat(); + if (!same_file) { + // The file descriptor refers to a different description. We must + // update our entry in the table. + delete it->second; + it->second = FileDescriptorInfo::createFromFd(*element); + if (it->second == NULL) { + // The descriptor no longer no longer refers to a whitelisted file. + // We flag an error and remove it from the list of files we're + // tracking. + error = true; + open_fd_map_.erase(it); + } + } else { + // It's the same file. Nothing to do here. + } + } + } + + if (open_fds.size() > 0) { + // The zygote has opened new file descriptors since our last inspection. + // We warn about this condition and add them to our table. + // + // TODO(narayan): This will be an error in a future android release. + // error = true; + ALOGW("Zygote opened %d new file descriptor(s).", open_fds.size()); + + // TODO(narayan): This code will be removed in a future android release. + std::set::const_iterator it; + for (it = open_fds.begin(); it != open_fds.end(); ++it) { + const int fd = (*it); + FileDescriptorInfo* info = FileDescriptorInfo::createFromFd(fd); + if (info == NULL) { + // A newly opened file is not on the whitelist. Flag an error and + // continue. + error = true; + } else { + // Track the newly opened file. + open_fd_map_[fd] = info; + } + } + } + + return !error; + } + + static int ParseFd(dirent* e, int dir_fd) { + char* end; + const int fd = strtol(e->d_name, &end, 10); + if ((*end) != '\0') { + return -1; + } + + // Don't bother with the standard input/output/error, they're handled + // specially post-fork anyway. + if (fd <= STDERR_FILENO || fd == dir_fd) { + return -1; + } + + return fd; + } + + // Invariant: All values in this unordered_map are non-NULL. + std::unordered_map open_fd_map_; + + DISALLOW_COPY_AND_ASSIGN(FileDescriptorTable); +}; -- 2.11.0