diff options
Diffstat (limited to 'toolkit/xre/MultiInstanceLock.cpp')
-rw-r--r-- | toolkit/xre/MultiInstanceLock.cpp | 197 |
1 files changed, 197 insertions, 0 deletions
diff --git a/toolkit/xre/MultiInstanceLock.cpp b/toolkit/xre/MultiInstanceLock.cpp new file mode 100644 index 0000000000..eb1cd01ce6 --- /dev/null +++ b/toolkit/xre/MultiInstanceLock.cpp @@ -0,0 +1,197 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim:set ts=2 sw=2 sts=2 et cindent: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "MultiInstanceLock.h" + +#include "commonupdatedir.h" // for GetInstallHash +#include "mozilla/UniquePtr.h" +#include "nsPrintfCString.h" +#include "nsPromiseFlatString.h" +#include "updatedefines.h" // for NS_t* definitions + +#ifndef XP_WIN +# include <fcntl.h> +# include <sys/stat.h> +# include <sys/types.h> +#endif + +namespace mozilla { + +static bool GetLockFileName(const char* nameToken, const char16_t* installPath, + nsCString& filePath) { + mozilla::UniquePtr<NS_tchar[]> pathHash; + if (!GetInstallHash(installPath, MOZ_APP_VENDOR, pathHash)) { + return false; + } + +#ifdef XP_WIN + // On Windows, the lock file is placed at the path + // ProgramData\[vendor]\[nameToken]-[pathHash], so first we need to get the + // ProgramData path and then append our directory and the file name. + PWSTR programDataPath; + HRESULT hr = SHGetKnownFolderPath(FOLDERID_ProgramData, KF_FLAG_CREATE, + nullptr, &programDataPath); + if (FAILED(hr)) { + return false; + } + mozilla::UniquePtr<wchar_t, CoTaskMemFreeDeleter> programDataPathUnique( + programDataPath); + + filePath = nsPrintfCString("%S\\%s\\%s-%S", programDataPath, MOZ_APP_VENDOR, + nameToken, pathHash.get()); + +#else + // On POSIX platforms the base path is /tmp/[vendor][nameToken]-[pathHash]. + filePath = nsPrintfCString("/tmp/%s%s-%s", MOZ_APP_VENDOR, nameToken, + pathHash.get()); + +#endif + + return true; +} + +MultiInstLockHandle OpenMultiInstanceLock(const char* nameToken, + const char16_t* installPath) { + nsCString filePath; + GetLockFileName(nameToken, installPath, filePath); + + // Open a file handle with full privileges and sharing, and then attempt to + // take a shared (nonexclusive, read-only) lock on it. +#ifdef XP_WIN + HANDLE h = + ::CreateFileW(PromiseFlatString(NS_ConvertUTF8toUTF16(filePath)).get(), + GENERIC_READ | GENERIC_WRITE, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, + nullptr, OPEN_ALWAYS, FILE_FLAG_DELETE_ON_CLOSE, nullptr); + if (h != INVALID_HANDLE_VALUE) { + // The LockFileEx functions always require an OVERLAPPED structure even + // though we did not open the lock file for overlapped I/O. + OVERLAPPED o = {0}; + if (!::LockFileEx(h, LOCKFILE_FAIL_IMMEDIATELY, 0, 1, 0, &o)) { + CloseHandle(h); + h = INVALID_HANDLE_VALUE; + } + } + return h; + +#else + int fd = ::open(PromiseFlatCString(filePath).get(), + O_CLOEXEC | O_CREAT | O_NOFOLLOW, + S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH); + if (fd != -1) { + // We would like to ensure that the lock file is deleted when we are done + // with it. The normal way to do that would be to call unlink on it right + // now, but that would immediately delete the name from the file system, and + // we need other instances to be able to open that name and get the same + // inode, so we can't unlink the file before we're done with it. This means + // we accept some unreliability in getting the file deleted, but it's a zero + // byte file in the tmp directory, so having it stay around isn't the worst. + struct flock l = {0}; + l.l_start = 0; + l.l_len = 0; + l.l_type = F_RDLCK; + if (::fcntl(fd, F_SETLK, &l)) { + ::close(fd); + fd = -1; + } + } + return fd; + +#endif +} + +void ReleaseMultiInstanceLock(MultiInstLockHandle lock) { + if (lock != MULTI_INSTANCE_LOCK_HANDLE_ERROR) { +#ifdef XP_WIN + OVERLAPPED o = {0}; + ::UnlockFileEx(lock, 0, 1, 0, &o); + // We've used FILE_FLAG_DELETE_ON_CLOSE, so if we are the last instance + // with a handle on the lock file, closing it here will delete it. + ::CloseHandle(lock); + +#else + // If we're the last instance, then unlink the lock file. There is a race + // condition here that may cause an instance to fail to open the same inode + // as another even though they use the same path, but there's no reasonable + // way to avoid that without skipping deleting the file at all, so we accept + // that risk. + bool otherInstance = true; + if (IsOtherInstanceRunning(lock, &otherInstance) && !otherInstance) { + // Recover the file's path so we can unlink it. + // There's no error checking in here because we're content to let the file + // hang around if any of this fails (which can happen if for example we're + // on a system where /proc/self/fd does not exist); this is a zero-byte + // file in the tmp directory after all. + UniquePtr<NS_tchar[]> linkPath = MakeUnique<NS_tchar[]>(MAXPATHLEN + 1); + NS_tsnprintf(linkPath.get(), MAXPATHLEN + 1, "/proc/self/fd/%d", lock); + UniquePtr<NS_tchar[]> lockFilePath = + MakeUnique<NS_tchar[]>(MAXPATHLEN + 1); + if (::readlink(linkPath.get(), lockFilePath.get(), MAXPATHLEN + 1) != + -1) { + ::unlink(lockFilePath.get()); + } + } + // Now close the lock file, which will release the lock. + ::close(lock); +#endif + } +} + +bool IsOtherInstanceRunning(MultiInstLockHandle lock, bool* aResult) { + // Every running instance has opened a readonly lock, and read locks prevent + // write locks from being opened, so to see if we are the only instance, we + // attempt to take a write lock, and if it succeeds then that must mean there + // are no other read locks open and therefore no other instances. + if (lock == MULTI_INSTANCE_LOCK_HANDLE_ERROR) { + return false; + } + +#ifdef XP_WIN + // We need to release the lock we're holding before we would be allowed to + // take an exclusive lock, and if that succeeds we need to release it too + // in order to get our shared lock back. This procedure is not atomic, so we + // accept the risk of the scheduler deciding to ruin our day between these + // operations; we'd get a false negative in a different instance's check. + OVERLAPPED o = {0}; + // Release our current shared lock. + if (!::UnlockFileEx(lock, 0, 1, 0, &o)) { + return false; + } + // Attempt to take an exclusive lock. + bool rv = false; + if (::LockFileEx(lock, LOCKFILE_EXCLUSIVE_LOCK | LOCKFILE_FAIL_IMMEDIATELY, 0, + 1, 0, &o)) { + // We got the exclusive lock, so now release it. + ::UnlockFileEx(lock, 0, 1, 0, &o); + *aResult = false; + rv = true; + } else if (::GetLastError() == ERROR_LOCK_VIOLATION) { + // We didn't get the exclusive lock because of outstanding shared locks. + *aResult = true; + rv = true; + } + // Attempt to reclaim the shared lock we released at the beginning. + if (!::LockFileEx(lock, LOCKFILE_FAIL_IMMEDIATELY, 0, 1, 0, &o)) { + rv = false; + } + return rv; + +#else + // See if we would be allowed to set a write lock (no need to actually do so). + struct flock l = {0}; + l.l_start = 0; + l.l_len = 0; + l.l_type = F_WRLCK; + if (::fcntl(lock, F_GETLK, &l)) { + return false; + } + *aResult = l.l_type != F_UNLCK; + return true; + +#endif +} + +}; // namespace mozilla |