TxFS is a novel transactional file system that builds upon the file system’s atomic-update mechanism such as journaling. Though prior work has explored a number of transactional file systems, TxFS has a unique set of properties: a simple API, portability across different hardware, high performance, low complexity (by building on the journal), and full ACID transactions.

Please cite the following paper if you use TxFS: TxFS: Leveraging File-System Crash Consistency to Provide ACID Transactions. Yige Hu, Zhiting Zhu, Ian Neal, Youngjin Kwon, Tianyu Cheng, Vijay Chidambaram, and Emmett Witchel ATC 18. Bibtex

This repository includes two main components:

1. linux-3.18.22-mod. This contains the modified Linux kernel for TxFS.
   • fs/user_tx.c defines the system calls provided by TxFS;
   • fs/memlog contains TxFS's internal bookkeeping code;
   • fs/*_memlog.c contains TxFS's changes to the VFS layer;
   • fs/ext4/*_memlog.c contains TxFS's major changes to ext4;
   • fs/jbd2/transaction.c contains TxFS's changes to JBD2.
   • mm/filemap.c contains TxFS's changes to memory page accesses.
2. benchmark. Contains benchmarks used to evaluate TxFS (we have only released microbenchmarks).
   • syscall_wrapper contains a system call wrapper providing API calls fs_tx_begin/end/abort, and an example showing how to use the wrapper;
   • syscall contains a set of tests on the system calls supported by TxFS transactions;
   • multithread contains multi-threaded tests;
   • stress_test randomly executes file system operations with multiple threads.

Inside linux-3.18.22-mod, compiling the kernel follows the standard method of configure and make. To see examples of TxFS in action, check out the examples inside benchmark.

We present an example where two files file1 and file2 are modified atomically. For the sake of brevity, we do not show error cases.

ret = fs_tx_begin();
fd1 = open("file1.txt", O_RDWR | O_APPEND, 0644);
fd2 = open("file2.txt", O_RDWR | O_APPEND, 0644);
write(fd1, "foo\n", 4);
write(fd2, "bar\n", 4);
ret = ret = fs_tx_commit();

The user does not have to add any other code apart from fs_tx_begin() and fs_tx_commit(). Once fs_tx_commit returns successfully, reads of both files will include both foo and bar.

syscall_wrapper in the benchmmark directory has the more complete version of this example. You can find it here.

Please contact us at yige@cs.utexas.edu with any questions. Drop us a note if you would like to use TxFS in your research.