1 /* Cumulus: Efficient Filesystem Backup to the Cloud
2 * Copyright (C) 2008-2009 The Cumulus Developers
3 * See the AUTHORS file for a list of contributors.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
20 /* Backup data is stored in a collection of objects, which are grouped together
21 * into segments for storage purposes. This implementation of the object store
22 * represents segments as TAR files and objects as files within them. */
28 #include <sys/types.h>
30 #include <sys/resource.h>
56 /* Default filter program is bzip2 */
57 const char *filter_program = "bzip2 -c";
58 const char *filter_extension = ".bz2";
60 Tarfile::Tarfile(RemoteFile *file, const string &segment)
64 assert(sizeof(struct tar_header) == TAR_BLOCK_SIZE);
67 this->filter.reset(FileFilter::New(file->get_fd(), filter_program));
72 char buf[TAR_BLOCK_SIZE];
74 /* Append the EOF marker: two blocks filled with nulls. */
75 memset(buf, 0, sizeof(buf));
76 tar_write(buf, TAR_BLOCK_SIZE);
77 tar_write(buf, TAR_BLOCK_SIZE);
79 if (close(filter->get_wrapped_fd()) != 0)
80 fatal("Error closing Tarfile");
82 /* ...and wait for filter process to finish. */
83 if (filter->wait() != 0) {
84 fatal("Filter process error");
88 FileFilter::FileFilter(int raw, int wrapped, pid_t pid)
89 : fd_raw(raw), fd_wrapped(wrapped), pid(pid) { }
91 FileFilter *FileFilter::New(int fd, const char *program)
93 if (program == NULL || strlen(program) == 0) {
94 return new FileFilter(fd, fd, -1);
98 int wrapped_fd = spawn_filter(fd, program, &pid);
99 return new FileFilter(fd, wrapped_fd, pid);
102 int FileFilter::wait()
104 // No filter program was launched implies no need to wait.
108 // The raw file descriptor was held open to track the output file size, but
109 // is not needed any longer.
113 if (waitpid(pid, &status, 0) < 0) {
114 fprintf(stderr, "Error waiting for filter process: %m\n");
118 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
119 fprintf(stderr, "Filter process error: %d\n", WEXITSTATUS(status));
125 /* Launch a child process which can act as a filter (compress, encrypt, etc.)
126 * on the TAR output. The file descriptor to which output should be written
127 * must be specified; the return value is the file descriptor which will be
128 * attached to the standard input of the filter program. */
129 int FileFilter::spawn_filter(int fd_out, const char *program, pid_t *filter_pid)
134 /* Create a pipe for communicating with the filter process. */
136 fatal("Unable to create pipe for filter");
139 /* Create a child process which can exec() the filter program. */
142 fatal("Unable to fork filter process");
148 if (filter_pid != NULL)
151 /* Child process. Rearrange file descriptors. stdin is fds[0], stdout
152 * is fd_out, stderr is unchanged. */
155 if (dup2(fds[0], 0) < 0)
159 if (dup2(fd_out, 1) < 0)
163 /* Exec the filter program. */
164 execlp("/bin/sh", "/bin/sh", "-c", program, NULL);
166 /* Should not reach here except for error cases. */
167 fprintf(stderr, "Could not exec filter: %m\n");
174 void Tarfile::tar_write(const char *data, size_t len)
179 int res = write(filter->get_wrapped_fd(), data, len);
184 fprintf(stderr, "Write error: %m\n");
185 fatal("Write error");
193 void Tarfile::write_object(int id, const char *data, size_t len)
195 struct tar_header header;
196 memset(&header, 0, sizeof(header));
199 sprintf(buf, "%08x", id);
200 string path = segment_name + "/" + buf;
202 assert(path.size() < 100);
203 memcpy(header.name, path.data(), path.size());
204 sprintf(header.mode, "%07o", 0600);
205 sprintf(header.uid, "%07o", 0);
206 sprintf(header.gid, "%07o", 0);
207 sprintf(header.size, "%011o", (int)len);
208 sprintf(header.mtime, "%011o", (int)time(NULL));
209 header.typeflag = '0';
210 strcpy(header.magic, "ustar ");
211 strcpy(header.uname, "root");
212 strcpy(header.gname, "root");
214 memset(header.chksum, ' ', sizeof(header.chksum));
216 for (int i = 0; i < TAR_BLOCK_SIZE; i++) {
217 checksum += ((uint8_t *)&header)[i];
219 sprintf(header.chksum, "%06o", checksum);
221 tar_write((const char *)&header, TAR_BLOCK_SIZE);
226 tar_write(data, len);
228 char padbuf[TAR_BLOCK_SIZE];
229 size_t blocks = (len + TAR_BLOCK_SIZE - 1) / TAR_BLOCK_SIZE;
230 size_t padding = blocks * TAR_BLOCK_SIZE - len;
231 memset(padbuf, 0, padding);
232 tar_write(padbuf, padding);
235 /* Estimate the size based on the size of the actual output file on disk.
236 * However, it might be the case that the filter program is buffering all its
237 * data, and might potentially not write a single byte until we have closed
238 * our end of the pipe. If we don't do so until we see data written, we have
239 * a problem. So, arbitrarily pick an upper bound on the compression ratio
240 * that the filter will achieve (128:1), and return a size estimate which is
241 * the larger of a) bytes actually seen written to disk, and b) input
243 size_t Tarfile::size_estimate()
247 if (fstat(filter->get_raw_fd(), &statbuf) == 0)
248 return max((int64_t)statbuf.st_size, (int64_t)(size / 128));
250 /* Couldn't stat the file on disk, so just return the actual number of
251 * bytes, before compression. */
255 static const size_t SEGMENT_SIZE = 4 * 1024 * 1024;
257 /* Backup size summary: segment type -> (uncompressed size, compressed size) */
258 static map<string, pair<int64_t, int64_t> > group_sizes;
260 ObjectReference TarSegmentStore::write_object(const char *data, size_t len,
261 const std::string &group,
262 const std::string &checksum,
265 struct segment_info *segment;
267 // Find the segment into which the object should be written, looking up by
268 // group. If no segment exists yet, create one.
269 if (segments.find(group) == segments.end()) {
270 segment = new segment_info;
272 segment->name = generate_uuid();
273 segment->group = group;
274 segment->basename = segment->name + ".tar";
275 segment->basename += filter_extension;
277 segment->data_size = 0;
278 segment->rf = remote->alloc_file(segment->basename,
279 group == "metadata" ? "segments0"
281 segment->file = new Tarfile(segment->rf, segment->name);
283 segments[group] = segment;
285 segment = segments[group];
288 int id = segment->count;
290 sprintf(id_buf, "%08x", id);
292 segment->file->write_object(id, data, len);
294 segment->data_size += len;
296 group_sizes[group].first += len;
298 ObjectReference ref(segment->name, id_buf);
299 ref.set_range(0, len, true);
300 if (checksum.size() > 0)
301 ref.set_checksum(checksum);
303 db->StoreObject(ref, age);
305 // If this segment meets or exceeds the size target, close it so that
306 // future objects will go into a new segment.
307 if (segment->file->size_estimate() >= SEGMENT_SIZE)
308 close_segment(group);
313 void TarSegmentStore::sync()
315 while (!segments.empty())
316 close_segment(segments.begin()->first);
319 void TarSegmentStore::dump_stats()
321 printf("Data written:\n");
322 for (map<string, pair<int64_t, int64_t> >::iterator i = group_sizes.begin();
323 i != group_sizes.end(); ++i) {
324 printf(" %s: %lld (%lld compressed)\n", i->first.c_str(),
325 (long long)i->second.first, (long long)i->second.second);
329 void TarSegmentStore::close_segment(const string &group)
331 struct segment_info *segment = segments[group];
333 delete segment->file;
336 struct stat stat_buf;
338 if (stat(segment->rf->get_local_path().c_str(), &stat_buf) == 0) {
339 disk_size = stat_buf.st_size;
340 group_sizes[segment->group].second += disk_size;
344 = Hash::hash_file(segment->rf->get_local_path().c_str());
346 db->SetSegmentMetadata(segment->name, segment->rf->get_remote_path(),
347 checksum, group, segment->data_size, disk_size);
352 segments.erase(segments.find(group));
356 string TarSegmentStore::object_reference_to_segment(const string &object)
361 LbsObject::LbsObject()
362 : group(""), age(0.0), data(NULL), data_len(0), written(false)
366 LbsObject::~LbsObject()
370 void LbsObject::set_data(const char *d, size_t len, const char *checksum)
375 if (checksum != NULL) {
376 this->checksum = checksum;
378 Hash *hash = Hash::New();
379 hash->update(data, data_len);
380 this->checksum = hash->digest_str();
385 void LbsObject::write(TarSegmentStore *store)
387 assert(data != NULL);
390 ref = store->write_object(data, data_len, group, checksum, age);