/* LBS: An LFS-inspired filesystem backup system
- * Copyright (C) 2006 Michael Vrable
+ * Copyright (C) 2007 Michael Vrable
*
* Backup data is stored in a collection of objects, which are grouped together
- * into segments for storage purposes. This file provides interfaces for
- * reading and writing objects and segments. */
+ * into segments for storage purposes. This implementation of the object store
+ * represents segments as TAR files and objects as files within them. */
#include <assert.h>
+#include <errno.h>
+#include <stdio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/resource.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <time.h>
+
+#include <algorithm>
+#include <list>
+#include <map>
+#include <set>
+#include <string>
+#include <iostream>
#include "store.h"
+#include "ref.h"
+using std::max;
+using std::list;
+using std::map;
+using std::set;
using std::string;
-void OutputStream::write_u8(uint8_t val)
-{
- write(&val, 1);
-}
+/* Default filter program is bzip2 */
+const char *filter_program = "bzip2 -c";
+const char *filter_extension = ".bz2";
-void OutputStream::write_u16(uint16_t val)
+static void cloexec(int fd)
{
- unsigned char buf[2];
+ long flags = fcntl(fd, F_GETFD);
+
+ if (flags < 0)
+ return;
- buf[0] = val & 0xff;
- buf[1] = (val >> 8) & 0xff;
- write(buf, 2);
+ fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
}
-void OutputStream::write_u32(uint32_t val)
+Tarfile::Tarfile(const string &path, const string &segment)
+ : size(0),
+ segment_name(segment)
{
- unsigned char buf[4];
+ assert(sizeof(struct tar_header) == TAR_BLOCK_SIZE);
- buf[0] = val & 0xff;
- buf[1] = (val >> 8) & 0xff;
- buf[2] = (val >> 16) & 0xff;
- buf[3] = (val >> 24) & 0xff;
- write(buf, 4);
+ real_fd = open(path.c_str(), O_WRONLY | O_CREAT, 0666);
+ if (real_fd < 0)
+ throw IOException("Error opening output file");
+
+ filter_fd = spawn_filter(real_fd, filter_program, &filter_pid);
}
-void OutputStream::write_u64(uint64_t val)
+Tarfile::~Tarfile()
{
- unsigned char buf[8];
+ char buf[TAR_BLOCK_SIZE];
+
+ /* Append the EOF marker: two blocks filled with nulls. */
+ memset(buf, 0, sizeof(buf));
+ tar_write(buf, TAR_BLOCK_SIZE);
+ tar_write(buf, TAR_BLOCK_SIZE);
+
+ if (close(filter_fd) != 0)
+ throw IOException("Error closing Tarfile");
- buf[0] = val & 0xff;
- buf[1] = (val >> 8) & 0xff;
- buf[2] = (val >> 16) & 0xff;
- buf[3] = (val >> 24) & 0xff;
- buf[4] = (val >> 32) & 0xff;
- buf[5] = (val >> 40) & 0xff;
- buf[6] = (val >> 48) & 0xff;
- buf[7] = (val >> 56) & 0xff;
- write(buf, 8);
+ /* ...and wait for filter process to finish. */
+ int status;
+ waitpid(filter_pid, &status, 0);
+
+ if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
+ throw IOException("Filter process error");
+ }
+
+ close(real_fd);
}
-/* Writes an integer to an output stream using a variable-sized representation:
- * seven bits are written at a time (little-endian), and the eigth bit of each
- * byte is set if more data follows. */
-void OutputStream::write_varint(uint64_t val)
+/* Launch a child process which can act as a filter (compress, encrypt, etc.)
+ * on the TAR output. The file descriptor to which output should be written
+ * must be specified; the return value is the file descriptor which will be
+ * attached to the standard input of the filter program. */
+int spawn_filter(int fd_out, const char *program, pid_t *filter_pid)
{
- do {
- uint8_t remainder = (val & 0x7f);
- val >>= 7;
- if (val)
- remainder |= 0x80;
- write_u8(remainder);
- } while (val);
+ int fds[2];
+ pid_t pid;
+
+ /* Create a pipe for communicating with the filter process. */
+ if (pipe(fds) < 0) {
+ throw IOException("Unable to create pipe for filter");
+ }
+
+ /* Create a child process which can exec() the filter program. */
+ pid = fork();
+ if (pid < 0)
+ throw IOException("Unable to fork filter process");
+
+ if (pid > 0) {
+ /* Parent process */
+ close(fds[0]);
+ cloexec(fds[1]);
+ if (filter_pid != NULL)
+ *filter_pid = pid;
+ } else {
+ /* Child process. Rearrange file descriptors. stdin is fds[0], stdout
+ * is fd_out, stderr is unchanged. */
+ close(fds[1]);
+
+ if (dup2(fds[0], 0) < 0)
+ exit(1);
+ close(fds[0]);
+
+ if (dup2(fd_out, 1) < 0)
+ exit(1);
+ close(fd_out);
+
+ /* Exec the filter program. */
+ execlp("/bin/sh", "/bin/sh", "-c", program, NULL);
+
+ /* Should not reach here except for error cases. */
+ fprintf(stderr, "Could not exec filter: %m\n");
+ exit(1);
+ }
+
+ return fds[1];
}
-/* Write an arbitrary string by first writing out the length, followed by the
- * data itself. */
-void OutputStream::write_string(const string &s)
+void Tarfile::tar_write(const char *data, size_t len)
{
- size_t len = s.length();
- write_varint(len);
- write(s.data(), len);
+ size += len;
+
+ while (len > 0) {
+ int res = write(filter_fd, data, len);
+
+ if (res < 0) {
+ if (errno == EINTR)
+ continue;
+ fprintf(stderr, "Write error: %m\n");
+ throw IOException("Write error");
+ }
+
+ len -= res;
+ data += res;
+ }
}
-void OutputStream::write_dictionary(const dictionary &d)
+void Tarfile::write_object(int id, const char *data, size_t len)
{
- size_t size = d.size();
- size_t written = 0;
+ struct tar_header header;
+ memset(&header, 0, sizeof(header));
- write_varint(size);
+ char buf[64];
+ sprintf(buf, "%08x", id);
+ string path = segment_name + "/" + buf;
- for (dictionary::const_iterator i = d.begin(); i != d.end(); ++i) {
- write_string(i->first);
- write_string(i->second);
- written++;
+ assert(path.size() < 100);
+ memcpy(header.name, path.data(), path.size());
+ sprintf(header.mode, "%07o", 0600);
+ sprintf(header.uid, "%07o", 0);
+ sprintf(header.gid, "%07o", 0);
+ sprintf(header.size, "%011o", len);
+ sprintf(header.mtime, "%011o", (int)time(NULL));
+ header.typeflag = '0';
+ strcpy(header.magic, "ustar ");
+ strcpy(header.uname, "root");
+ strcpy(header.gname, "root");
+
+ memset(header.chksum, ' ', sizeof(header.chksum));
+ int checksum = 0;
+ for (int i = 0; i < TAR_BLOCK_SIZE; i++) {
+ checksum += ((uint8_t *)&header)[i];
}
+ sprintf(header.chksum, "%06o", checksum);
+
+ tar_write((const char *)&header, TAR_BLOCK_SIZE);
+
+ if (len == 0)
+ return;
- assert(written == size);
+ tar_write(data, len);
+
+ char padbuf[TAR_BLOCK_SIZE];
+ size_t blocks = (len + TAR_BLOCK_SIZE - 1) / TAR_BLOCK_SIZE;
+ size_t padding = blocks * TAR_BLOCK_SIZE - len;
+ memset(padbuf, 0, padding);
+ tar_write(padbuf, padding);
}
-StringOutputStream::StringOutputStream()
- : buf(std::ios_base::out)
+/* Estimate the size based on the size of the actual output file on disk.
+ * However, it might be the case that the filter program is buffering all its
+ * data, and might potentially not write a single byte until we have closed
+ * our end of the pipe. If we don't do so until we see data written, we have
+ * a problem. So, arbitrarily pick an upper bound on the compression ratio
+ * that the filter will achieve (128:1), and return a size estimate which is
+ * the larger of a) bytes actually seen written to disk, and b) input
+ * bytes/128. */
+size_t Tarfile::size_estimate()
{
+ struct stat statbuf;
+
+ if (fstat(real_fd, &statbuf) == 0)
+ return max((int64_t)statbuf.st_size, (int64_t)(size / 128));
+
+ /* Couldn't stat the file on disk, so just return the actual number of
+ * bytes, before compression. */
+ return size;
}
-void StringOutputStream::write(const void *data, size_t len)
+static const size_t SEGMENT_SIZE = 4 * 1024 * 1024;
+
+static map<string, int64_t> group_sizes;
+
+ObjectReference TarSegmentStore::write_object(const char *data, size_t len,
+ const std::string &group)
{
- buf.write((const char *)data, len);
- if (!buf.good())
- throw IOException("error writing to StringOutputStream");
+ struct segment_info *segment;
+
+ // Find the segment into which the object should be written, looking up by
+ // group. If no segment exists yet, create one.
+ if (segments.find(group) == segments.end()) {
+ segment = new segment_info;
+
+ segment->name = generate_uuid();
+ segment->basename = segment->name + ".tar";
+ segment->basename += filter_extension;
+ segment->fullname = path + "/" + segment->basename;
+ segment->file = new Tarfile(segment->fullname, segment->name);
+ segment->count = 0;
+
+ segments[group] = segment;
+ } else {
+ segment = segments[group];
+ }
+
+ int id = segment->count;
+ char id_buf[64];
+ sprintf(id_buf, "%08x", id);
+
+ segment->file->write_object(id, data, len);
+ segment->count++;
+
+ group_sizes[group] += len;
+
+ ObjectReference ref(segment->name, id_buf);
+
+ // If this segment meets or exceeds the size target, close it so that
+ // future objects will go into a new segment.
+ if (segment->file->size_estimate() >= SEGMENT_SIZE)
+ close_segment(group);
+
+ return ref;
}
-FileOutputStream::FileOutputStream(FILE *file)
+void TarSegmentStore::sync()
{
- f = file;
+ while (!segments.empty())
+ close_segment(segments.begin()->first);
}
-FileOutputStream::~FileOutputStream()
+void TarSegmentStore::dump_stats()
{
- fclose(f);
+ printf("Data written:\n");
+ for (map<string, int64_t>::iterator i = group_sizes.begin();
+ i != group_sizes.end(); ++i) {
+ printf(" %s: %lld\n", i->first.c_str(), i->second);
+ }
}
-void FileOutputStream::write(const void *data, size_t len)
+void TarSegmentStore::close_segment(const string &group)
{
- size_t res;
+ struct segment_info *segment = segments[group];
+
+ delete segment->file;
- res = fwrite(data, 1, len, f);
- if (res != len) {
- throw IOException("write error");
+ if (db != NULL) {
+ SHA1Checksum segment_checksum;
+ if (segment_checksum.process_file(segment->fullname.c_str())) {
+ string checksum = segment_checksum.checksum_str();
+ db->SetSegmentChecksum(segment->name, segment->basename, checksum);
+ }
}
+
+ segments.erase(segments.find(group));
+ delete segment;
}
-/* Utility functions, for encoding data types to strings. */
-string encode_u16(uint16_t val)
+string TarSegmentStore::object_reference_to_segment(const string &object)
{
- StringOutputStream s;
- s.write_u16(val);
- return s.contents();
+ return object;
}
-string encode_u32(uint32_t val)
+LbsObject::LbsObject()
+ : group(""), data(NULL), data_len(0), written(false)
{
- StringOutputStream s;
- s.write_u32(val);
- return s.contents();
}
-string encode_u64(uint64_t val)
+LbsObject::~LbsObject()
{
- StringOutputStream s;
- s.write_u64(val);
- return s.contents();
+}
+
+void LbsObject::write(TarSegmentStore *store)
+{
+ assert(data != NULL);
+ assert(!written);
+
+ ref = store->write_object(data, data_len, group);
+ written = true;
+}
+
+void LbsObject::checksum()
+{
+ assert(written);
+
+ SHA1Checksum hash;
+ hash.process(data, data_len);
+ ref.set_checksum(hash.checksum_str());
}