/* Recursively descend the filesystem and visit each file. */
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdint.h>
#include <dirent.h>
#include <errno.h>
#include <fcntl.h>
-#include <pwd.h>
+#include <getopt.h>
#include <grp.h>
-#include <sys/types.h>
+#include <pwd.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
#include <sys/stat.h>
+#include <sys/types.h>
#include <unistd.h>
#include <algorithm>
-#include <string>
-#include <list>
-#include <vector>
-#include <iostream>
#include <fstream>
-#include <sstream>
+#include <iostream>
+#include <list>
#include <set>
+#include <sstream>
+#include <string>
+#include <vector>
#include "format.h"
#include "localdb.h"
#include "store.h"
#include "sha1.h"
+#include "statcache.h"
using std::list;
using std::string;
* invocations to help in creating incremental snapshots. */
LocalDb *db;
+/* Stat cache, which stored data locally to speed the backup process by quickly
+ * skipping files which have not changed. */
+StatCache *statcache;
+
/* Contents of the root object. This will contain a set of indirect links to
* the metadata objects. */
std::ostringstream metadata_root;
void scandir(const string& path);
+/* Selection of files to include/exclude in the snapshot. */
+std::list<string> excludes;
+
/* Ensure contents of metadata are flushed to an object. */
void metadata_flush()
{
/* Write current metadata information to a new object. */
LbsObject *meta = new LbsObject;
- meta->set_group("root");
+ meta->set_group("metadata");
meta->set_data(m.data(), m.size());
meta->write(tss);
meta->checksum();
/* Read the contents of a file (specified by an open file descriptor) and copy
* the data to the store. Returns the size of the file (number of bytes
* dumped), or -1 on error. */
-int64_t dumpfile(int fd, dictionary &file_info)
+int64_t dumpfile(int fd, dictionary &file_info, const string &path)
{
struct stat stat_buf;
fstat(fd, &stat_buf);
// Store a copy of the object if one does not yet exist
if (ref.get_segment().size() == 0) {
LbsObject *o = new LbsObject;
- o->set_group("data");
+
+ /* We might still have seen this checksum before, if the object was
+ * stored at some time in the past, but we have decided to clean
+ * the segment the object was originally stored in (FindObject will
+ * not return such objects). When rewriting the object contents,
+ * put it in a separate group, so that old objects get grouped
+ * together. The hope is that these old objects will continue to
+ * be used in the future, and we obtain segments which will
+ * continue to be well-utilized. */
+ if (db->IsOldObject(block_csum, bytes))
+ o->set_group("compacted");
+ else
+ o->set_group("data");
+
o->set_data(block_buf, bytes);
o->write(tss);
ref = o->get_ref();
db->StoreObject(ref, block_csum, bytes);
delete o;
}
+
object_list.push_back(ref.to_string());
segment_list.insert(ref.get_segment());
+ db->UseObject(ref);
size += bytes;
}
file_info["checksum"] = hash.checksum_str();
+ statcache->Save(path, &stat_buf, file_info["checksum"], object_list);
+
/* For files that only need to be broken apart into a few objects, store
* the list of objects directly. For larger files, store the data
* out-of-line and provide a pointer to the indrect object. */
}
LbsObject *i = new LbsObject;
- i->set_group("indirect");
+ i->set_group("metadata");
i->set_data(blocklist.data(), blocklist.size());
i->write(tss);
file_info["data"] = "@" + i->get_name();
// Set to true if the item is a directory and we should recursively scan
bool recurse = false;
+ // Check this file against the include/exclude list to see if it should be
+ // considered
+ for (list<string>::iterator i = excludes.begin();
+ i != excludes.end(); ++i) {
+ if (path == *i) {
+ printf("Excluding %s\n", path.c_str());
+ return;
+ }
+ }
+
dictionary file_info;
lstat(path.c_str(), &stat_buf);
printf("%s\n", path.c_str());
- metadata << "name: " << uri_encode(path) << "\n";
-
file_info["mode"] = encode_int(stat_buf.st_mode & 07777);
file_info["mtime"] = encode_int(stat_buf.st_mtime);
file_info["user"] = encode_int(stat_buf.st_uid);
* - O_NONBLOCK: prevents open() from blocking if the file was
* replaced by a fifo
* We also add in O_NOATIME, since this may reduce disk writes (for
- * inode updates). */
+ * inode updates). However, O_NOATIME may result in EPERM, so if the
+ * initial open fails, try again without O_NOATIME. */
fd = open(path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
+ if (fd < 0) {
+ fd = open(path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK);
+ }
+ if (fd < 0) {
+ fprintf(stderr, "Unable to open file %s: %m\n", path.c_str());
+ return;
+ }
/* Drop the use of the O_NONBLOCK flag; we only wanted that for file
* open. */
flags = fcntl(fd, F_GETFL);
fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
- file_size = dumpfile(fd, file_info);
+ file_size = dumpfile(fd, file_info, path);
file_info["size"] = encode_int(file_size);
close(fd);
file_info["type"] = string(1, inode_type);
+ metadata << "name: " << uri_encode(path) << "\n";
dict_output(metadata, file_info);
metadata << "\n";
for (vector<string>::iterator i = contents.begin();
i != contents.end(); ++i) {
const string& filename = *i;
- scanfile(path + "/" + filename);
+ if (path == ".")
+ scanfile(filename);
+ else
+ scanfile(path + "/" + filename);
}
closedir(dir);
}
-int main(int argc, char *argv[])
+void usage(const char *program)
{
- block_buf = new char[LBS_BLOCK_SIZE];
+ fprintf(stderr,
+ "Usage: %s [OPTION]... SOURCE DEST\n"
+ "Produce backup snapshot of files in SOURCE and store to DEST.\n"
+ "\n"
+ "Options:\n"
+ " --exclude=PATH exclude files in PATH from snapshot\n"
+ " --localdb=PATH local backup metadata is stored in PATH\n",
+ program);
+}
+int main(int argc, char *argv[])
+{
+ string backup_source = ".";
string backup_dest = ".";
+ string localdb_dir = "";
- if (argc > 1)
- backup_dest = argv[1];
+ while (1) {
+ static struct option long_options[] = {
+ {"localdb", 1, 0, 0}, // 0
+ {"exclude", 1, 0, 0}, // 1
+ {"filter", 1, 0, 0}, // 2
+ {"filter-extension", 1, 0, 0}, // 3
+ {NULL, 0, 0, 0},
+ };
- tss = new TarSegmentStore(backup_dest);
+ int long_index;
+ int c = getopt_long(argc, argv, "", long_options, &long_index);
- string database_path = backup_dest + "/localdb.sqlite";
- db = new LocalDb;
- db->Open(database_path.c_str());
+ if (c == -1)
+ break;
- /* Write a backup descriptor file, which says which segments are needed and
- * where to start to restore this snapshot. The filename is based on the
- * current time. */
+ if (c == 0) {
+ switch (long_index) {
+ case 0: // --localdb
+ localdb_dir = optarg;
+ break;
+ case 1: // --exclude
+ excludes.push_back(optarg);
+ break;
+ case 2: // --filter
+ filter_program = optarg;
+ break;
+ case 3: // --filter-extension
+ filter_extension = optarg;
+ break;
+ default:
+ fprintf(stderr, "Unhandled long option!\n");
+ return 1;
+ }
+ } else {
+ usage(argv[0]);
+ return 1;
+ }
+ }
+
+ if (argc < optind + 2) {
+ usage(argv[0]);
+ return 1;
+ }
+
+ backup_source = argv[optind];
+ backup_dest = argv[argc - 1];
+
+ if (localdb_dir == "") {
+ localdb_dir = backup_dest;
+ }
+
+ printf("Source: %s\nDest: %s\nDatabase: %s\n\n",
+ backup_source.c_str(), backup_dest.c_str(), localdb_dir.c_str());
+
+ tss = new TarSegmentStore(backup_dest);
+ block_buf = new char[LBS_BLOCK_SIZE];
+
+ /* Store the time when the backup started, so it can be included in the
+ * snapshot name. */
time_t now;
struct tm time_buf;
char desc_buf[256];
time(&now);
localtime_r(&now, &time_buf);
strftime(desc_buf, sizeof(desc_buf), "%Y%m%dT%H%M%S", &time_buf);
- string desc_filename = backup_dest + "/" + desc_buf + ".lbs";
- std::ofstream descriptor(desc_filename.c_str());
+
+ /* Open the local database which tracks all objects that are stored
+ * remotely, for efficient incrementals. Provide it with the name of this
+ * snapshot. */
+ string database_path = localdb_dir + "/localdb.sqlite";
+ db = new LocalDb;
+ db->Open(database_path.c_str(), desc_buf);
+
+ /* Initialize the stat cache, for skipping over unchanged files. */
+ statcache = new StatCache;
+ statcache->Open(localdb_dir.c_str(), desc_buf);
try {
scanfile(".");
const string md = metadata_root.str();
LbsObject *root = new LbsObject;
- root->set_group("root");
+ root->set_group("metadata");
root->set_data(md.data(), md.size());
root->write(tss);
root->checksum();
-
segment_list.insert(root->get_ref().get_segment());
- descriptor << "Root: " << root->get_ref().to_string() << "\n";
+
+ /* Write a backup descriptor file, which says which segments are needed and
+ * where to start to restore this snapshot. The filename is based on the
+ * current time. */
+ string desc_filename = backup_dest + "/snapshot-" + desc_buf + ".lbs";
+ std::ofstream descriptor(desc_filename.c_str());
+
+ descriptor << "Format: LBS Snapshot v0.1\n";
strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
descriptor << "Date: " << desc_buf << "\n";
+ descriptor << "Root: " << root->get_ref().to_string() << "\n";
delete root;
db->Close();
+ statcache->Close();
+ delete statcache;
+
tss->sync();
delete tss;