1 /* Recursively descend the filesystem and visit each file. */
30 static TarSegmentStore *tss = NULL;
32 /* Buffer for holding a single block of data read from a file. */
33 static const int LBS_BLOCK_SIZE = 1024 * 1024;
34 static char *block_buf;
36 /* Contents of the root object. This will contain a set of indirect links to
37 * the metadata objects. */
38 std::ostringstream metadata_root;
40 /* Buffer for building up metadata. */
41 std::ostringstream metadata;
43 /* Keep track of all segments which are needed to reconstruct the snapshot. */
44 std::set<string> segment_list;
46 void scandir(const string& path);
48 /* Ensure contents of metadata are flushed to an object. */
51 string m = metadata.str();
55 /* Write current metadata information to a new object. */
56 LbsObject *meta = new LbsObject;
57 meta->set_group("root");
58 meta->set_data(m.data(), m.size());
62 /* Write a reference to this block in the root. */
63 ObjectReference ref = meta->get_ref();
64 metadata_root << "@" << ref.to_string() << "\n";
65 segment_list.insert(ref.get_segment());
72 /* Read data from a file descriptor and return the amount of data read. A
73 * short read (less than the requested size) will only occur if end-of-file is
75 size_t file_read(int fd, char *buf, size_t maxlen)
77 size_t bytes_read = 0;
80 ssize_t res = read(fd, buf, maxlen);
84 throw IOException("file_read: error reading");
85 } else if (res == 0) {
97 /* Read the contents of a file (specified by an open file descriptor) and copy
98 * the data to the store. */
99 void dumpfile(int fd, dictionary &file_info)
101 struct stat stat_buf;
102 fstat(fd, &stat_buf);
104 list<string> object_list;
106 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
107 printf("file is no longer a regular file!\n");
111 /* The index data consists of a sequence of pointers to the data blocks
112 * that actually comprise the file data. This level of indirection is used
113 * so that the same data block can be used in multiple files, or multiple
114 * versions of the same file. */
117 size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
121 hash.process(block_buf, bytes);
123 // tarstore processing
124 LbsObject *o = new LbsObject;
125 o->set_group("data");
126 o->set_data(block_buf, bytes);
128 object_list.push_back(o->get_name());
129 segment_list.insert(o->get_ref().get_segment());
135 file_info["checksum"] = hash.checksum_str();
137 /* For files that only need to be broken apart into a few objects, store
138 * the list of objects directly. For larger files, store the data
139 * out-of-line and provide a pointer to the indrect object. */
140 if (object_list.size() < 8) {
141 string blocklist = "";
142 for (list<string>::iterator i = object_list.begin();
143 i != object_list.end(); ++i) {
144 if (i != object_list.begin())
148 file_info["data"] = blocklist;
150 string blocklist = "";
151 for (list<string>::iterator i = object_list.begin();
152 i != object_list.end(); ++i) {
153 blocklist += *i + "\n";
156 LbsObject *i = new LbsObject;
157 i->set_group("indirect");
158 i->set_data(blocklist.data(), blocklist.size());
160 file_info["data"] = "@" + i->get_name();
161 segment_list.insert(i->get_ref().get_segment());
166 void scanfile(const string& path)
170 struct stat stat_buf;
175 // Set to true if the item is a directory and we should recursively scan
176 bool recurse = false;
178 dictionary file_info;
180 lstat(path.c_str(), &stat_buf);
182 printf("%s\n", path.c_str());
184 metadata << "name: " << uri_encode(path) << "\n";
186 file_info["mode"] = encode_int(stat_buf.st_mode & 07777);
187 file_info["atime"] = encode_int(stat_buf.st_atime);
188 file_info["ctime"] = encode_int(stat_buf.st_ctime);
189 file_info["mtime"] = encode_int(stat_buf.st_mtime);
190 file_info["user"] = encode_int(stat_buf.st_uid);
191 file_info["group"] = encode_int(stat_buf.st_gid);
195 switch (stat_buf.st_mode & S_IFMT) {
211 /* Use the reported file size to allocate a buffer large enough to read
212 * the symlink. Allocate slightly more space, so that we ask for more
213 * bytes than we expect and so check for truncation. */
214 buf = new char[stat_buf.st_size + 2];
215 len = readlink(path.c_str(), buf, stat_buf.st_size + 1);
217 printf("error reading symlink: %m\n");
218 } else if (len <= stat_buf.st_size) {
220 printf(" contents=%s\n", buf);
221 } else if (len > stat_buf.st_size) {
222 printf("error reading symlink: name truncated\n");
225 file_info["contents"] = uri_encode(buf);
232 /* Be paranoid when opening the file. We have no guarantee that the
233 * file was not replaced between the stat() call above and the open()
234 * call below, so we might not even be opening a regular file. That
235 * the file descriptor refers to a regular file is checked in
236 * dumpfile(). But we also supply flags to open to to guard against
237 * various conditions before we can perform that verification:
238 * - O_NOFOLLOW: in the event the file was replaced by a symlink
239 * - O_NONBLOCK: prevents open() from blocking if the file was
241 * We also add in O_NOATIME, since this may reduce disk writes (for
243 fd = open(path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
245 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
247 flags = fcntl(fd, F_GETFL);
248 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
250 file_info["size"] = encode_int(stat_buf.st_size);
251 dumpfile(fd, file_info);
261 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
265 file_info["type"] = string(1, inode_type);
266 metadata << "type: " << inode_type << "\n";
268 dict_output(metadata, file_info);
271 // Break apart metadata listing if it becomes too large.
272 if (metadata.str().size() > 4096)
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 printf("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];
315 tss = new TarSegmentStore(argv[1]);
317 tss = new TarSegmentStore(".");
322 } catch (IOException e) {
323 fprintf(stderr, "IOException: %s\n", e.getError().c_str());
327 const string md = metadata_root.str();
329 LbsObject *root = new LbsObject;
330 root->set_group("root");
331 root->set_data(md.data(), md.size());
335 segment_list.insert(root->get_ref().get_segment());
336 string r = root->get_ref().to_string();
337 printf("\nroot: %s\n\n", r.c_str());
340 printf("segments:\n");
341 for (std::set<string>::iterator i = segment_list.begin();
342 i != segment_list.end(); ++i) {
343 printf(" %s\n", i->c_str());