#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
+#include <sys/resource.h>
+#include <sys/wait.h>
#include <unistd.h>
#include <fcntl.h>
#include <time.h>
+#include <algorithm>
#include <list>
+#include <map>
#include <set>
#include <string>
#include <iostream>
#include "store.h"
#include "ref.h"
+using std::max;
using std::list;
+using std::map;
using std::set;
using std::string;
-list<string> TarSegmentStore::norefs;
+/* Default filter program is bzip2 */
+const char *filter_program = "bzip2 -c";
+const char *filter_extension = ".bz2";
+
+static void cloexec(int fd)
+{
+ long flags = fcntl(fd, F_GETFD);
+
+ if (flags < 0)
+ return;
+
+ fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
+}
Tarfile::Tarfile(const string &path, const string &segment)
: size(0),
segment_name(segment)
{
- if (tar_open(&t, (char *)path.c_str(), NULL, O_WRONLY | O_CREAT, 0600,
- TAR_VERBOSE | TAR_GNU) == -1)
+ real_fd = open(path.c_str(), O_WRONLY | O_CREAT, 0600);
+ if (real_fd < 0)
+ throw IOException("Error opening output file");
+
+ filter_fd = spawn_filter(real_fd);
+
+ if (tar_fdopen(&t, filter_fd, (char *)path.c_str(), NULL,
+ O_WRONLY | O_CREAT, 0666, TAR_VERBOSE | TAR_GNU) == -1)
throw IOException("Error opening Tarfile");
}
Tarfile::~Tarfile()
{
- string checksum_list = checksums.str();
- internal_write_object(segment_name + "/checksums",
- checksum_list.data(), checksum_list.size());
+ /* Close the tar file... */
tar_append_eof(t);
if (tar_close(t) != 0)
throw IOException("Error closing Tarfile");
+
+ /* ...and wait for filter process to finish. */
+ int status;
+ waitpid(filter_pid, &status, 0);
+
+ if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
+ throw IOException("Filter process error");
+ }
+
+ close(real_fd);
+}
+
+/* Launch a child process which can act as a filter (compress, encrypt, etc.)
+ * on the TAR output. The file descriptor to which output should be written
+ * must be specified; the return value is the file descriptor which will be
+ * attached to the standard input of the filter program. */
+int Tarfile::spawn_filter(int fd_out)
+{
+ int fds[2];
+
+ /* Create a pipe for communicating with the filter process. */
+ if (pipe(fds) < 0) {
+ throw IOException("Unable to create pipe for filter");
+ }
+
+ /* Create a child process which can exec() the filter program. */
+ filter_pid = fork();
+ if (filter_pid < 0)
+ throw IOException("Unable to fork filter process");
+
+ if (filter_pid > 0) {
+ /* Parent process */
+ close(fds[0]);
+ cloexec(fds[1]);
+ } else {
+ /* Child process. Rearrange file descriptors. stdin is fds[0], stdout
+ * is fd_out, stderr is unchanged. */
+ close(fds[1]);
+
+ if (dup2(fds[0], 0) < 0)
+ exit(1);
+ close(fds[0]);
+
+ if (dup2(fd_out, 1) < 0)
+ exit(1);
+ close(fd_out);
+
+ /* Exec the filter program. */
+ execlp("/bin/sh", "/bin/sh", "-c", filter_program, NULL);
+
+ /* Should not reach here except for error cases. */
+ fprintf(stderr, "Could not exec filter: %m\n");
+ exit(1);
+ }
+
+ return fds[1];
}
void Tarfile::write_object(int id, const char *data, size_t len)
string path = segment_name + "/" + buf;
internal_write_object(path, data, len);
-
- // Compute a checksum for the data block, which will be stored at the end
- // of the TAR file.
- SHA1Checksum hash;
- hash.process(data, len);
- sprintf(buf, "%08x", id);
- checksums << buf << " " << hash.checksum_str() << "\n";
}
void Tarfile::internal_write_object(const string &path,
size += blocks * T_BLOCKSIZE;
}
+/* Estimate the size based on the size of the actual output file on disk.
+ * However, it might be the case that the filter program is buffering all its
+ * data, and might potentially not write a single byte until we have closed
+ * our end of the pipe. If we don't do so until we see data written, we have
+ * a problem. So, arbitrarily pick an upper bound on the compression ratio
+ * that the filter will achieve (128:1), and return a size estimate which is
+ * the larger of a) bytes actually seen written to disk, and b) input
+ * bytes/128. */
+size_t Tarfile::size_estimate()
+{
+ struct stat statbuf;
+
+ if (fstat(real_fd, &statbuf) == 0)
+ return max((int64_t)statbuf.st_size, (int64_t)(size / 128));
+
+ /* Couldn't stat the file on disk, so just return the actual number of
+ * bytes, before compression. */
+ return size;
+}
+
static const size_t SEGMENT_SIZE = 4 * 1024 * 1024;
-string TarSegmentStore::write_object(const char *data, size_t len, const
- std::string &group,
- const std::list<std::string> &refs)
+static map<string, int64_t> group_sizes;
+
+ObjectReference TarSegmentStore::write_object(const char *data, size_t len,
+ const std::string &group)
{
struct segment_info *segment;
segment->name = generate_uuid();
string filename = path + "/" + segment->name + ".tar";
+ filename += filter_extension;
segment->file = new Tarfile(filename, segment->name);
segment->count = 0;
segment->file->write_object(id, data, len);
segment->count++;
- string full_name = segment->name + "/" + id_buf;
+ group_sizes[group] += len;
- // Store any dependencies this object has on other segments, so they can be
- // written when the segment is closed.
- for (list<string>::const_iterator i = refs.begin(); i != refs.end(); ++i) {
- segment->refs.insert(*i);
- }
+ ObjectReference ref(segment->name, id_buf);
// If this segment meets or exceeds the size target, close it so that
// future objects will go into a new segment.
if (segment->file->size_estimate() >= SEGMENT_SIZE)
close_segment(group);
- return full_name;
+ return ref;
}
void TarSegmentStore::sync()
close_segment(segments.begin()->first);
}
+void TarSegmentStore::dump_stats()
+{
+ printf("Data written:\n");
+ for (map<string, int64_t>::iterator i = group_sizes.begin();
+ i != group_sizes.end(); ++i) {
+ printf(" %s: %lld\n", i->first.c_str(), i->second);
+ }
+}
+
void TarSegmentStore::close_segment(const string &group)
{
struct segment_info *segment = segments[group];
- fprintf(stderr, "Closing segment group %s (%s)\n",
- group.c_str(), segment->name.c_str());
-
- string reflist;
- for (set<string>::iterator i = segment->refs.begin();
- i != segment->refs.end(); ++i) {
- reflist += *i + "\n";
- }
- segment->file->internal_write_object(segment->name + "/references",
- reflist.data(), reflist.size());
delete segment->file;
segments.erase(segments.find(group));
{
}
-void LbsObject::add_reference(const LbsObject *o)
-{
- refs.insert(o->get_name());
-}
-
void LbsObject::write(TarSegmentStore *store)
{
assert(data != NULL);
assert(!written);
- list<string> reflist;
- for (set<string>::iterator i = refs.begin(); i != refs.end(); ++i) {
- reflist.push_back(*i);
- }
+ ref = store->write_object(data, data_len, group);
+ written = true;
+}
- name = store->write_object(data, data_len, group, reflist);
+void LbsObject::checksum()
+{
+ assert(written);
- written = true;
- data = NULL;
+ SHA1Checksum hash;
+ hash.process(data, data_len);
+ ref.set_checksum(hash.checksum_str());
}