1 /* Recursively descend the filesystem and visit each file. */
31 static TarSegmentStore *tss = NULL;
33 /* Buffer for holding a single block of data read from a file. */
34 static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
35 static char *block_buf;
37 static const size_t LBS_METADATA_BLOCK_SIZE = 65536;
39 /* Contents of the root object. This will contain a set of indirect links to
40 * the metadata objects. */
41 std::ostringstream metadata_root;
43 /* Buffer for building up metadata. */
44 std::ostringstream metadata;
46 /* Keep track of all segments which are needed to reconstruct the snapshot. */
47 std::set<string> segment_list;
49 void scandir(const string& path);
51 /* Ensure contents of metadata are flushed to an object. */
54 string m = metadata.str();
58 /* Write current metadata information to a new object. */
59 LbsObject *meta = new LbsObject;
60 meta->set_group("root");
61 meta->set_data(m.data(), m.size());
65 /* Write a reference to this block in the root. */
66 ObjectReference ref = meta->get_ref();
67 metadata_root << "@" << ref.to_string() << "\n";
68 segment_list.insert(ref.get_segment());
75 /* Read data from a file descriptor and return the amount of data read. A
76 * short read (less than the requested size) will only occur if end-of-file is
78 size_t file_read(int fd, char *buf, size_t maxlen)
80 size_t bytes_read = 0;
83 ssize_t res = read(fd, buf, maxlen);
87 throw IOException("file_read: error reading");
88 } else if (res == 0) {
100 /* Read the contents of a file (specified by an open file descriptor) and copy
101 * the data to the store. */
102 void dumpfile(int fd, dictionary &file_info)
104 struct stat stat_buf;
105 fstat(fd, &stat_buf);
107 list<string> object_list;
109 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
110 fprintf(stderr, "file is no longer a regular file!\n");
114 /* The index data consists of a sequence of pointers to the data blocks
115 * that actually comprise the file data. This level of indirection is used
116 * so that the same data block can be used in multiple files, or multiple
117 * versions of the same file. */
120 size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
124 hash.process(block_buf, bytes);
126 // tarstore processing
127 LbsObject *o = new LbsObject;
128 o->set_group("data");
129 o->set_data(block_buf, bytes);
131 object_list.push_back(o->get_name());
132 segment_list.insert(o->get_ref().get_segment());
138 file_info["checksum"] = hash.checksum_str();
140 /* For files that only need to be broken apart into a few objects, store
141 * the list of objects directly. For larger files, store the data
142 * out-of-line and provide a pointer to the indrect object. */
143 if (object_list.size() < 8) {
144 string blocklist = "";
145 for (list<string>::iterator i = object_list.begin();
146 i != object_list.end(); ++i) {
147 if (i != object_list.begin())
151 file_info["data"] = blocklist;
153 string blocklist = "";
154 for (list<string>::iterator i = object_list.begin();
155 i != object_list.end(); ++i) {
156 blocklist += *i + "\n";
159 LbsObject *i = new LbsObject;
160 i->set_group("indirect");
161 i->set_data(blocklist.data(), blocklist.size());
163 file_info["data"] = "@" + i->get_name();
164 segment_list.insert(i->get_ref().get_segment());
169 void scanfile(const string& path)
173 struct stat stat_buf;
178 // Set to true if the item is a directory and we should recursively scan
179 bool recurse = false;
181 dictionary file_info;
183 lstat(path.c_str(), &stat_buf);
185 printf("%s\n", path.c_str());
187 metadata << "name: " << uri_encode(path) << "\n";
189 file_info["mode"] = encode_int(stat_buf.st_mode & 07777);
190 file_info["atime"] = encode_int(stat_buf.st_atime);
191 file_info["ctime"] = encode_int(stat_buf.st_ctime);
192 file_info["mtime"] = encode_int(stat_buf.st_mtime);
193 file_info["user"] = encode_int(stat_buf.st_uid);
194 file_info["group"] = encode_int(stat_buf.st_gid);
198 switch (stat_buf.st_mode & S_IFMT) {
214 /* Use the reported file size to allocate a buffer large enough to read
215 * the symlink. Allocate slightly more space, so that we ask for more
216 * bytes than we expect and so check for truncation. */
217 buf = new char[stat_buf.st_size + 2];
218 len = readlink(path.c_str(), buf, stat_buf.st_size + 1);
220 fprintf(stderr, "error reading symlink: %m\n");
221 } else if (len <= stat_buf.st_size) {
223 file_info["contents"] = uri_encode(buf);
224 } else if (len > stat_buf.st_size) {
225 fprintf(stderr, "error reading symlink: name truncated\n");
233 /* Be paranoid when opening the file. We have no guarantee that the
234 * file was not replaced between the stat() call above and the open()
235 * call below, so we might not even be opening a regular file. That
236 * the file descriptor refers to a regular file is checked in
237 * dumpfile(). But we also supply flags to open to to guard against
238 * various conditions before we can perform that verification:
239 * - O_NOFOLLOW: in the event the file was replaced by a symlink
240 * - O_NONBLOCK: prevents open() from blocking if the file was
242 * We also add in O_NOATIME, since this may reduce disk writes (for
244 fd = open(path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
246 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
248 flags = fcntl(fd, F_GETFL);
249 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
251 file_info["size"] = encode_int(stat_buf.st_size);
252 dumpfile(fd, file_info);
262 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
266 file_info["type"] = string(1, inode_type);
268 dict_output(metadata, file_info);
271 // Break apart metadata listing if it becomes too large.
272 if (metadata.str().size() > LBS_METADATA_BLOCK_SIZE)
275 // If we hit a directory, now that we've written the directory itself,
276 // recursively scan the directory.
281 void scandir(const string& path)
283 DIR *dir = opendir(path.c_str());
286 fprintf(stderr, "Error: %m\n");
291 vector<string> contents;
292 while ((ent = readdir(dir)) != NULL) {
293 string filename(ent->d_name);
294 if (filename == "." || filename == "..")
296 contents.push_back(filename);
299 sort(contents.begin(), contents.end());
301 for (vector<string>::iterator i = contents.begin();
302 i != contents.end(); ++i) {
303 const string& filename = *i;
304 scanfile(path + "/" + filename);
310 int main(int argc, char *argv[])
312 block_buf = new char[LBS_BLOCK_SIZE];
314 string backup_dest = ".";
317 backup_dest = argv[1];
319 tss = new TarSegmentStore(backup_dest);
321 string desc_filename = backup_dest + "/snapshot.lbs";
322 std::ofstream descriptor(desc_filename.c_str());
326 } catch (IOException e) {
327 fprintf(stderr, "IOException: %s\n", e.getError().c_str());
331 const string md = metadata_root.str();
333 LbsObject *root = new LbsObject;
334 root->set_group("root");
335 root->set_data(md.data(), md.size());
339 segment_list.insert(root->get_ref().get_segment());
340 descriptor << "root: " << root->get_ref().to_string() << "\n\n";
344 descriptor << "segments:\n";
345 for (std::set<string>::iterator i = segment_list.begin();
346 i != segment_list.end(); ++i) {
347 descriptor << " " << *i << "\n";