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 void scandir(const string& path, std::ostream& metadata);
38 /* Converts time to microseconds since the epoch. */
39 int64_t encode_time(time_t time)
41 return (int64_t)time * 1000000;
44 /* Read data from a file descriptor and return the amount of data read. A
45 * short read (less than the requested size) will only occur if end-of-file is
47 size_t file_read(int fd, char *buf, size_t maxlen)
49 size_t bytes_read = 0;
52 ssize_t res = read(fd, buf, maxlen);
56 throw IOException("file_read: error reading");
57 } else if (res == 0) {
69 /* Read the contents of a file (specified by an open file descriptor) and copy
70 * the data to the store. */
71 void dumpfile(int fd, dictionary &file_info, ostream &metadata)
76 list<string> segment_list;
78 if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
79 printf("file is no longer a regular file!\n");
83 /* The index data consists of a sequence of pointers to the data blocks
84 * that actually comprise the file data. This level of indirection is used
85 * so that the same data block can be used in multiple files, or multiple
86 * versions of the same file. */
89 size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
93 hash.process(block_buf, bytes);
95 // tarstore processing
96 string blockid = tss->write_object(block_buf, bytes, "data");
97 segment_list.push_back(blockid);
102 file_info["checksum"] = hash.checksum_str();
104 /* For files that only need to be broken apart into a few objects, store
105 * the list of objects directly. For larger files, store the data
106 * out-of-line and provide a pointer to the indrect object. */
107 if (segment_list.size() < 8) {
108 string blocklist = "";
109 for (list<string>::iterator i = segment_list.begin();
110 i != segment_list.end(); ++i) {
111 if (i != segment_list.begin())
115 file_info["data"] = blocklist;
117 string blocklist = "";
118 for (list<string>::iterator i = segment_list.begin();
119 i != segment_list.end(); ++i) {
120 blocklist += *i + "\n";
122 string indirect = tss->write_object(blocklist.data(), blocklist.size(),
124 file_info["data"] = "@" + indirect;
128 void scanfile(const string& path, ostream &metadata)
132 struct stat stat_buf;
136 // Set to true if the item is a directory and we should recursively scan
137 bool recurse = false;
139 dictionary file_info;
141 lstat(path.c_str(), &stat_buf);
143 printf("%s\n", path.c_str());
145 metadata << "name: " << uri_encode(path) << "\n";
147 file_info["mode"] = encode_int(stat_buf.st_mode & 07777);
148 file_info["atime"] = encode_int(stat_buf.st_atime);
149 file_info["ctime"] = encode_int(stat_buf.st_ctime);
150 file_info["mtime"] = encode_int(stat_buf.st_mtime);
151 file_info["user"] = encode_int(stat_buf.st_uid);
152 file_info["group"] = encode_int(stat_buf.st_gid);
156 switch (stat_buf.st_mode & S_IFMT) {
172 /* Use the reported file size to allocate a buffer large enough to read
173 * the symlink. Allocate slightly more space, so that we ask for more
174 * bytes than we expect and so check for truncation. */
175 buf = new char[stat_buf.st_size + 2];
176 len = readlink(path.c_str(), buf, stat_buf.st_size + 1);
178 printf("error reading symlink: %m\n");
179 } else if (len <= stat_buf.st_size) {
181 printf(" contents=%s\n", buf);
182 } else if (len > stat_buf.st_size) {
183 printf("error reading symlink: name truncated\n");
186 file_info["contents"] = uri_encode(buf);
193 /* Be paranoid when opening the file. We have no guarantee that the
194 * file was not replaced between the stat() call above and the open()
195 * call below, so we might not even be opening a regular file. That
196 * the file descriptor refers to a regular file is checked in
197 * dumpfile(). But we also supply flags to open to to guard against
198 * various conditions before we can perform that verification:
199 * - O_NOFOLLOW: in the event the file was replaced by a symlink
200 * - O_NONBLOCK: prevents open() from blocking if the file was
202 * We also add in O_NOATIME, since this may reduce disk writes (for
204 fd = open(path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
206 /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
208 flags = fcntl(fd, F_GETFL);
209 fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
211 file_info["size"] = encode_int(stat_buf.st_size);
212 dumpfile(fd, file_info, metadata);
222 fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode);
226 file_info["type"] = string(1, inode_type);
227 metadata << "type: " << inode_type << "\n";
229 dict_output(metadata, file_info);
232 // If we hit a directory, now that we've written the directory itself,
233 // recursively scan the directory.
235 scandir(path, metadata);
238 void scandir(const string& path, ostream &metadata)
240 DIR *dir = opendir(path.c_str());
243 printf("Error: %m\n");
248 vector<string> contents;
249 while ((ent = readdir(dir)) != NULL) {
250 string filename(ent->d_name);
251 if (filename == "." || filename == "..")
253 contents.push_back(filename);
256 sort(contents.begin(), contents.end());
258 for (vector<string>::iterator i = contents.begin();
259 i != contents.end(); ++i) {
260 const string& filename = *i;
261 scanfile(path + "/" + filename, metadata);
267 int main(int argc, char *argv[])
269 block_buf = new char[LBS_BLOCK_SIZE];
271 tss = new TarSegmentStore(".");
273 std::ostringstream metadata;
276 scanfile(".", metadata);
277 } catch (IOException e) {
278 fprintf(stderr, "IOException: %s\n", e.getError().c_str());
281 const string md = metadata.str();
282 string root = tss->write_object(md.data(), md.size(), "root");