1 /* Cumulus: Smart Filesystem Backup to Dumb Servers
3 * Copyright (C) 2008 The Regents of the University of California
4 * Written by Michael Vrable <mvrable@cs.ucsd.edu>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
21 /* Backup data is stored in a collection of objects, which are grouped together
22 * into segments for storage purposes. This implementation of the object store
23 * represents segments as TAR files and objects as files within them. */
29 #include <sys/types.h>
31 #include <sys/resource.h>
55 /* Default filter program is bzip2 */
56 const char *filter_program = "bzip2 -c";
57 const char *filter_extension = ".bz2";
59 Tarfile::Tarfile(RemoteFile *file, const string &segment)
63 assert(sizeof(struct tar_header) == TAR_BLOCK_SIZE);
66 real_fd = file->get_fd();
67 filter_fd = spawn_filter(real_fd, filter_program, &filter_pid);
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_fd) != 0)
80 fatal("Error closing Tarfile");
82 /* ...and wait for filter process to finish. */
84 waitpid(filter_pid, &status, 0);
86 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
87 fatal("Filter process error");
93 /* Launch a child process which can act as a filter (compress, encrypt, etc.)
94 * on the TAR output. The file descriptor to which output should be written
95 * must be specified; the return value is the file descriptor which will be
96 * attached to the standard input of the filter program. */
97 int spawn_filter(int fd_out, const char *program, pid_t *filter_pid)
102 /* Create a pipe for communicating with the filter process. */
104 fatal("Unable to create pipe for filter");
107 /* Create a child process which can exec() the filter program. */
110 fatal("Unable to fork filter process");
116 if (filter_pid != NULL)
119 /* Child process. Rearrange file descriptors. stdin is fds[0], stdout
120 * is fd_out, stderr is unchanged. */
123 if (dup2(fds[0], 0) < 0)
127 if (dup2(fd_out, 1) < 0)
131 /* Exec the filter program. */
132 execlp("/bin/sh", "/bin/sh", "-c", program, NULL);
134 /* Should not reach here except for error cases. */
135 fprintf(stderr, "Could not exec filter: %m\n");
142 void Tarfile::tar_write(const char *data, size_t len)
147 int res = write(filter_fd, data, len);
152 fprintf(stderr, "Write error: %m\n");
153 fatal("Write error");
161 void Tarfile::write_object(int id, const char *data, size_t len)
163 struct tar_header header;
164 memset(&header, 0, sizeof(header));
167 sprintf(buf, "%08x", id);
168 string path = segment_name + "/" + buf;
170 assert(path.size() < 100);
171 memcpy(header.name, path.data(), path.size());
172 sprintf(header.mode, "%07o", 0600);
173 sprintf(header.uid, "%07o", 0);
174 sprintf(header.gid, "%07o", 0);
175 sprintf(header.size, "%011o", (int)len);
176 sprintf(header.mtime, "%011o", (int)time(NULL));
177 header.typeflag = '0';
178 strcpy(header.magic, "ustar ");
179 strcpy(header.uname, "root");
180 strcpy(header.gname, "root");
182 memset(header.chksum, ' ', sizeof(header.chksum));
184 for (int i = 0; i < TAR_BLOCK_SIZE; i++) {
185 checksum += ((uint8_t *)&header)[i];
187 sprintf(header.chksum, "%06o", checksum);
189 tar_write((const char *)&header, TAR_BLOCK_SIZE);
194 tar_write(data, len);
196 char padbuf[TAR_BLOCK_SIZE];
197 size_t blocks = (len + TAR_BLOCK_SIZE - 1) / TAR_BLOCK_SIZE;
198 size_t padding = blocks * TAR_BLOCK_SIZE - len;
199 memset(padbuf, 0, padding);
200 tar_write(padbuf, padding);
203 /* Estimate the size based on the size of the actual output file on disk.
204 * However, it might be the case that the filter program is buffering all its
205 * data, and might potentially not write a single byte until we have closed
206 * our end of the pipe. If we don't do so until we see data written, we have
207 * a problem. So, arbitrarily pick an upper bound on the compression ratio
208 * that the filter will achieve (128:1), and return a size estimate which is
209 * the larger of a) bytes actually seen written to disk, and b) input
211 size_t Tarfile::size_estimate()
215 if (fstat(real_fd, &statbuf) == 0)
216 return max((int64_t)statbuf.st_size, (int64_t)(size / 128));
218 /* Couldn't stat the file on disk, so just return the actual number of
219 * bytes, before compression. */
223 static const size_t SEGMENT_SIZE = 4 * 1024 * 1024;
225 /* Backup size summary: segment type -> (uncompressed size, compressed size) */
226 static map<string, pair<int64_t, int64_t> > group_sizes;
228 ObjectReference TarSegmentStore::write_object(const char *data, size_t len,
229 const std::string &group)
231 struct segment_info *segment;
233 // Find the segment into which the object should be written, looking up by
234 // group. If no segment exists yet, create one.
235 if (segments.find(group) == segments.end()) {
236 segment = new segment_info;
238 segment->name = generate_uuid();
239 segment->group = group;
240 segment->basename = segment->name + ".tar";
241 segment->basename += filter_extension;
244 segment->rf = remote->alloc_file(segment->basename, "segments");
245 segment->file = new Tarfile(segment->rf, segment->name);
247 segments[group] = segment;
249 segment = segments[group];
252 int id = segment->count;
254 sprintf(id_buf, "%08x", id);
256 segment->file->write_object(id, data, len);
258 segment->size += len;
260 group_sizes[group].first += len;
262 ObjectReference ref(segment->name, id_buf);
264 // If this segment meets or exceeds the size target, close it so that
265 // future objects will go into a new segment.
266 if (segment->file->size_estimate() >= SEGMENT_SIZE)
267 close_segment(group);
272 void TarSegmentStore::sync()
274 while (!segments.empty())
275 close_segment(segments.begin()->first);
278 void TarSegmentStore::dump_stats()
280 printf("Data written:\n");
281 for (map<string, pair<int64_t, int64_t> >::iterator i = group_sizes.begin();
282 i != group_sizes.end(); ++i) {
283 printf(" %s: %lld (%lld compressed)\n", i->first.c_str(),
284 (long long)i->second.first, (long long)i->second.second);
288 void TarSegmentStore::close_segment(const string &group)
290 struct segment_info *segment = segments[group];
292 delete segment->file;
295 SHA1Checksum segment_checksum;
296 if (segment_checksum.process_file(segment->rf->get_local_path().c_str())) {
297 string checksum = segment_checksum.checksum_str();
298 db->SetSegmentChecksum(segment->name, segment->basename, checksum,
302 struct stat stat_buf;
303 if (stat(segment->rf->get_local_path().c_str(), &stat_buf) == 0) {
304 group_sizes[segment->group].second += stat_buf.st_size;
310 segments.erase(segments.find(group));
314 string TarSegmentStore::object_reference_to_segment(const string &object)
319 LbsObject::LbsObject()
320 : group(""), data(NULL), data_len(0), written(false)
324 LbsObject::~LbsObject()
328 void LbsObject::write(TarSegmentStore *store)
330 assert(data != NULL);
333 ref = store->write_object(data, data_len, group);
337 void LbsObject::checksum()
342 hash.process(data, data_len);
343 ref.set_checksum(hash.checksum_str());