1 /* Recursively descend the filesystem and visit each file. */
11 #include <sys/types.h>
34 static TarSegmentStore *tss = NULL;
36 /* Buffer for holding a single block of data read from a file. */
37 static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
38 static char *block_buf;
40 static const size_t LBS_METADATA_BLOCK_SIZE = 65536;
42 /* Local database, which tracks objects written in this and previous
43 * invocations to help in creating incremental snapshots. */
46 /* Contents of the root object. This will contain a set of indirect links to
47 * the metadata objects. */
48 std::ostringstream metadata_root;
50 /* Buffer for building up metadata. */
51 std::ostringstream metadata;
53 /* Keep track of all segments which are needed to reconstruct the snapshot. */
54 std::set<string> segment_list;
56 void scandir(const string& path);
58 /* Ensure contents of metadata are flushed to an object. */
61 string m = metadata.str();
65 /* Write current metadata information to a new object. */
66 LbsObject *meta = new LbsObject;
67 meta->set_group("root");
68 meta->set_data(m.data(), m.size());
72 /* Write a reference to this block in the root. */
73 ObjectReference ref = meta->get_ref();
74 metadata_root << "@" << ref.to_string() << "\n";
75 segment_list.insert(ref.get_segment());
82 /* Read data from a file descriptor and return the amount of data read. A
83 * short read (less than the requested size) will only occur if end-of-file is
85 size_t file_read(int fd, char *buf, size_t maxlen)
87 size_t bytes_read = 0;
90 ssize_t res = read(fd, buf, maxlen);
94 throw IOException("file_read: error reading");
95 } else if (res == 0) {
107 /* Read the contents of a file (specified by an open file descriptor) and copy
108 * the data to the store. Returns the size of the file (number of bytes
109 * dumped), or -1 on error. */
110 int64_t dumpfile(int fd, dictionary &file_info)
112 struct stat stat_buf;
113 fstat(fd, &stat_buf);
115 list<string> object_list;
117 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
118 fprintf(stderr, "file is no longer a regular file!\n");
122 /* The index data consists of a sequence of pointers to the data blocks
123 * that actually comprise the file data. This level of indirection is used
124 * so that the same data block can be used in multiple files, or multiple
125 * versions of the same file. */
128 size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
132 hash.process(block_buf, bytes);
134 // tarstore processing
135 LbsObject *o = new LbsObject;
136 o->set_group("data");
137 o->set_data(block_buf, bytes);
139 object_list.push_back(o->get_name());
140 segment_list.insert(o->get_ref().get_segment());
142 // Index this block so it can be used by future snapshots
143 SHA1Checksum block_hash;
144 block_hash.process(block_buf, bytes);
145 db->StoreObject(o->get_ref(), block_hash.checksum_str(), bytes);
152 file_info["checksum"] = hash.checksum_str();
154 /* For files that only need to be broken apart into a few objects, store
155 * the list of objects directly. For larger files, store the data
156 * out-of-line and provide a pointer to the indrect object. */
157 if (object_list.size() < 8) {
158 string blocklist = "";
159 for (list<string>::iterator i = object_list.begin();
160 i != object_list.end(); ++i) {
161 if (i != object_list.begin())
165 file_info["data"] = blocklist;
167 string blocklist = "";
168 for (list<string>::iterator i = object_list.begin();
169 i != object_list.end(); ++i) {
170 blocklist += *i + "\n";
173 LbsObject *i = new LbsObject;
174 i->set_group("indirect");
175 i->set_data(blocklist.data(), blocklist.size());
177 file_info["data"] = "@" + i->get_name();
178 segment_list.insert(i->get_ref().get_segment());
185 void scanfile(const string& path)
189 struct stat stat_buf;
195 // Set to true if the item is a directory and we should recursively scan
196 bool recurse = false;
198 dictionary file_info;
200 lstat(path.c_str(), &stat_buf);
202 printf("%s\n", path.c_str());
204 metadata << "name: " << uri_encode(path) << "\n";
206 file_info["mode"] = encode_int(stat_buf.st_mode & 07777);
207 file_info["mtime"] = encode_int(stat_buf.st_mtime);
208 file_info["user"] = encode_int(stat_buf.st_uid);
209 file_info["group"] = encode_int(stat_buf.st_gid);
211 struct passwd *pwd = getpwuid(stat_buf.st_uid);
213 file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")";
216 struct group *grp = getgrgid(stat_buf.st_gid);
218 file_info["group"] += " (" + uri_encode(grp->gr_name) + ")";
223 switch (stat_buf.st_mode & S_IFMT) {
239 /* Use the reported file size to allocate a buffer large enough to read
240 * the symlink. Allocate slightly more space, so that we ask for more
241 * bytes than we expect and so check for truncation. */
242 buf = new char[stat_buf.st_size + 2];
243 len = readlink(path.c_str(), buf, stat_buf.st_size + 1);
245 fprintf(stderr, "error reading symlink: %m\n");
246 } else if (len <= stat_buf.st_size) {
248 file_info["contents"] = uri_encode(buf);
249 } else if (len > stat_buf.st_size) {
250 fprintf(stderr, "error reading symlink: name truncated\n");
258 /* Be paranoid when opening the file. We have no guarantee that the
259 * file was not replaced between the stat() call above and the open()
260 * call below, so we might not even be opening a regular file. That
261 * the file descriptor refers to a regular file is checked in
262 * dumpfile(). But we also supply flags to open to to guard against
263 * various conditions before we can perform that verification:
264 * - O_NOFOLLOW: in the event the file was replaced by a symlink
265 * - O_NONBLOCK: prevents open() from blocking if the file was
267 * We also add in O_NOATIME, since this may reduce disk writes (for
269 fd = open(path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
271 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
273 flags = fcntl(fd, F_GETFL);
274 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
276 file_size = dumpfile(fd, file_info);
277 file_info["size"] = encode_int(file_size);
281 return; // error occurred; do not dump file
283 if (file_size != stat_buf.st_size) {
284 fprintf(stderr, "Warning: Size of %s changed during reading\n",
295 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
299 file_info["type"] = string(1, inode_type);
301 dict_output(metadata, file_info);
304 // Break apart metadata listing if it becomes too large.
305 if (metadata.str().size() > LBS_METADATA_BLOCK_SIZE)
308 // If we hit a directory, now that we've written the directory itself,
309 // recursively scan the directory.
314 void scandir(const string& path)
316 DIR *dir = opendir(path.c_str());
319 fprintf(stderr, "Error: %m\n");
324 vector<string> contents;
325 while ((ent = readdir(dir)) != NULL) {
326 string filename(ent->d_name);
327 if (filename == "." || filename == "..")
329 contents.push_back(filename);
332 sort(contents.begin(), contents.end());
334 for (vector<string>::iterator i = contents.begin();
335 i != contents.end(); ++i) {
336 const string& filename = *i;
337 scanfile(path + "/" + filename);
343 int main(int argc, char *argv[])
345 block_buf = new char[LBS_BLOCK_SIZE];
347 string backup_dest = ".";
350 backup_dest = argv[1];
352 tss = new TarSegmentStore(backup_dest);
354 string database_path = backup_dest + "/localdb.sqlite";
356 db->Open(database_path.c_str());
358 /* Write a backup descriptor file, which says which segments are needed and
359 * where to start to restore this snapshot. The filename is based on the
365 localtime_r(&now, &time_buf);
366 strftime(desc_buf, sizeof(desc_buf), "%Y%m%dT%H%M%S", &time_buf);
367 string desc_filename = backup_dest + "/" + desc_buf + ".lbs";
368 std::ofstream descriptor(desc_filename.c_str());
372 } catch (IOException e) {
373 fprintf(stderr, "IOException: %s\n", e.getError().c_str());
377 const string md = metadata_root.str();
379 LbsObject *root = new LbsObject;
380 root->set_group("root");
381 root->set_data(md.data(), md.size());
385 segment_list.insert(root->get_ref().get_segment());
386 descriptor << "Root: " << root->get_ref().to_string() << "\n";
387 strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
388 descriptor << "Date: " << desc_buf << "\n";
392 descriptor << "Segments:\n";
393 for (std::set<string>::iterator i = segment_list.begin();
394 i != segment_list.end(); ++i) {
395 descriptor << " " << *i << "\n";