#include <dirent.h>
#include <errno.h>
#include <fcntl.h>
+#include <pwd.h>
+#include <grp.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <algorithm>
#include <string>
+#include <list>
#include <vector>
+#include <iostream>
+#include <fstream>
+#include <sstream>
+#include <set>
+#include "format.h"
+#include "localdb.h"
#include "store.h"
+#include "sha1.h"
+using std::list;
using std::string;
using std::vector;
+using std::ostream;
-static OutputStream *info_dump = NULL;
+static TarSegmentStore *tss = NULL;
+
+/* Buffer for holding a single block of data read from a file. */
+static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
+static char *block_buf;
+
+static const size_t LBS_METADATA_BLOCK_SIZE = 65536;
+
+/* Local database, which tracks objects written in this and previous
+ * invocations to help in creating incremental snapshots. */
+LocalDb *db;
+
+/* Contents of the root object. This will contain a set of indirect links to
+ * the metadata objects. */
+std::ostringstream metadata_root;
+
+/* Buffer for building up metadata. */
+std::ostringstream metadata;
+
+/* Keep track of all segments which are needed to reconstruct the snapshot. */
+std::set<string> segment_list;
void scandir(const string& path);
-/* Converts time to microseconds since the epoch. */
-int64_t encode_time(time_t time)
+/* Ensure contents of metadata are flushed to an object. */
+void metadata_flush()
{
- return (int64_t)time * 1000000;
+ string m = metadata.str();
+ if (m.size() == 0)
+ return;
+
+ /* Write current metadata information to a new object. */
+ LbsObject *meta = new LbsObject;
+ meta->set_group("root");
+ meta->set_data(m.data(), m.size());
+ meta->write(tss);
+ meta->checksum();
+
+ /* Write a reference to this block in the root. */
+ ObjectReference ref = meta->get_ref();
+ metadata_root << "@" << ref.to_string() << "\n";
+ segment_list.insert(ref.get_segment());
+
+ delete meta;
+
+ metadata.str("");
+}
+
+/* Read data from a file descriptor and return the amount of data read. A
+ * short read (less than the requested size) will only occur if end-of-file is
+ * hit. */
+size_t file_read(int fd, char *buf, size_t maxlen)
+{
+ size_t bytes_read = 0;
+
+ while (true) {
+ ssize_t res = read(fd, buf, maxlen);
+ if (res < 0) {
+ if (errno == EINTR)
+ continue;
+ throw IOException("file_read: error reading");
+ } else if (res == 0) {
+ break;
+ } else {
+ bytes_read += res;
+ buf += res;
+ maxlen -= res;
+ }
+ }
+
+ return bytes_read;
}
-void dumpfile(int fd)
+/* 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)
{
struct stat stat_buf;
fstat(fd, &stat_buf);
int64_t size = 0;
-
- char buf[4096];
+ list<string> object_list;
if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
- printf("file is no longer a regular file!\n");
- return;
+ fprintf(stderr, "file is no longer a regular file!\n");
+ return -1;
}
+ /* The index data consists of a sequence of pointers to the data blocks
+ * that actually comprise the file data. This level of indirection is used
+ * so that the same data block can be used in multiple files, or multiple
+ * versions of the same file. */
+ SHA1Checksum hash;
while (true) {
- ssize_t res = read(fd, buf, sizeof(buf));
- if (res < 0) {
- if (errno == EINTR)
- continue;
- printf("Error while reading: %m\n");
- return;
- } else if (res == 0) {
+ size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
+ if (bytes == 0)
break;
- } else {
- size += res;
+
+ hash.process(block_buf, bytes);
+
+ // tarstore processing
+ LbsObject *o = new LbsObject;
+ o->set_group("data");
+ o->set_data(block_buf, bytes);
+ o->write(tss);
+ object_list.push_back(o->get_name());
+ segment_list.insert(o->get_ref().get_segment());
+
+ // Index this block so it can be used by future snapshots
+ SHA1Checksum block_hash;
+ block_hash.process(block_buf, bytes);
+ db->StoreObject(o->get_ref(), block_hash.checksum_str(), bytes);
+
+ size += bytes;
+
+ delete o;
+ }
+
+ file_info["checksum"] = hash.checksum_str();
+
+ /* 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. */
+ if (object_list.size() < 8) {
+ string blocklist = "";
+ for (list<string>::iterator i = object_list.begin();
+ i != object_list.end(); ++i) {
+ if (i != object_list.begin())
+ blocklist += " ";
+ blocklist += *i;
}
+ file_info["data"] = blocklist;
+ } else {
+ string blocklist = "";
+ for (list<string>::iterator i = object_list.begin();
+ i != object_list.end(); ++i) {
+ blocklist += *i + "\n";
+ }
+
+ LbsObject *i = new LbsObject;
+ i->set_group("indirect");
+ i->set_data(blocklist.data(), blocklist.size());
+ i->write(tss);
+ file_info["data"] = "@" + i->get_name();
+ segment_list.insert(i->get_ref().get_segment());
+ delete i;
}
- printf(" bytes=%Ld\n", size);
+ return size;
}
void scanfile(const string& path)
struct stat stat_buf;
char *buf;
ssize_t len;
+ int64_t file_size;
+ list<string> refs;
+
+ // Set to true if the item is a directory and we should recursively scan
+ bool recurse = false;
dictionary file_info;
lstat(path.c_str(), &stat_buf);
- printf("%s:\n", path.c_str());
- printf(" ino=%Ld, perm=%04o, uid=%d, gid=%d, nlink=%d, blksize=%d, size=%Ld\n",
- (int64_t)stat_buf.st_ino, stat_buf.st_mode & 07777,
- stat_buf.st_uid, stat_buf.st_gid, stat_buf.st_nlink,
- (int)stat_buf.st_blksize, (int64_t)stat_buf.st_size);
+ printf("%s\n", path.c_str());
+
+ metadata << "name: " << uri_encode(path) << "\n";
- file_info["mode"] = encode_u16(stat_buf.st_mode & 07777);
- file_info["atime"] = encode_u64(encode_time(stat_buf.st_atime));
- file_info["ctime"] = encode_u64(encode_time(stat_buf.st_ctime));
- file_info["mtime"] = encode_u64(encode_time(stat_buf.st_mtime));
- file_info["user"] = encode_u32(stat_buf.st_uid);
- file_info["group"] = encode_u32(stat_buf.st_gid);
+ 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);
+ file_info["group"] = encode_int(stat_buf.st_gid);
+
+ struct passwd *pwd = getpwuid(stat_buf.st_uid);
+ if (pwd != NULL) {
+ file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")";
+ }
+
+ struct group *grp = getgrgid(stat_buf.st_gid);
+ if (pwd != NULL) {
+ file_info["group"] += " (" + uri_encode(grp->gr_name) + ")";
+ }
char inode_type;
switch (stat_buf.st_mode & S_IFMT) {
case S_IFIFO:
- printf(" FIFO\n");
inode_type = 'p';
break;
case S_IFSOCK:
- printf(" socket\n");
inode_type = 's';
break;
case S_IFCHR:
- printf(" character device\n");
inode_type = 'c';
break;
case S_IFBLK:
- printf(" block device\n");
inode_type = 'b';
break;
case S_IFLNK:
- printf(" symlink\n");
inode_type = 'l';
/* Use the reported file size to allocate a buffer large enough to read
buf = new char[stat_buf.st_size + 2];
len = readlink(path.c_str(), buf, stat_buf.st_size + 1);
if (len < 0) {
- printf("error reading symlink: %m\n");
+ fprintf(stderr, "error reading symlink: %m\n");
} else if (len <= stat_buf.st_size) {
buf[len] = '\0';
- printf(" contents=%s\n", buf);
+ file_info["contents"] = uri_encode(buf);
} else if (len > stat_buf.st_size) {
- printf("error reading symlink: name truncated\n");
+ fprintf(stderr, "error reading symlink: name truncated\n");
}
- file_info["contents"] = buf;
-
delete[] buf;
break;
case S_IFREG:
- printf(" regular file\n");
inode_type = '-';
/* Be paranoid when opening the file. We have no guarantee that the
flags = fcntl(fd, F_GETFL);
fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
- //dumpfile(fd);
- file_info["size"] = encode_u64(stat_buf.st_size);
+ file_size = dumpfile(fd, file_info);
+ file_info["size"] = encode_int(file_size);
close(fd);
+ if (file_size < 0)
+ return; // error occurred; do not dump file
+
+ if (file_size != stat_buf.st_size) {
+ fprintf(stderr, "Warning: Size of %s changed during reading\n",
+ path.c_str());
+ }
+
break;
case S_IFDIR:
- printf(" directory\n");
inode_type = 'd';
- scandir(path);
+ recurse = true;
break;
default:
file_info["type"] = string(1, inode_type);
- info_dump->write_string(path);
- info_dump->write_dictionary(file_info);
+ dict_output(metadata, file_info);
+ metadata << "\n";
+
+ // Break apart metadata listing if it becomes too large.
+ if (metadata.str().size() > LBS_METADATA_BLOCK_SIZE)
+ metadata_flush();
+
+ // If we hit a directory, now that we've written the directory itself,
+ // recursively scan the directory.
+ if (recurse)
+ scandir(path);
}
void scandir(const string& path)
DIR *dir = opendir(path.c_str());
if (dir == NULL) {
- printf("Error: %m\n");
+ fprintf(stderr, "Error: %m\n");
return;
}
int main(int argc, char *argv[])
{
- FILE *dump = fopen("fileinfo", "w");
- if (dump == NULL) {
- fprintf(stderr, "Cannot open fileinfo: %m\n");
- return 1;
- }
+ block_buf = new char[LBS_BLOCK_SIZE];
+
+ string backup_dest = ".";
+
+ if (argc > 1)
+ backup_dest = argv[1];
+
+ tss = new TarSegmentStore(backup_dest);
+
+ string database_path = backup_dest + "/localdb.sqlite";
+ db = new LocalDb;
+ db->Open(database_path.c_str());
- FileOutputStream os(dump);
- info_dump = &os;
+ /* 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. */
+ 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());
try {
- scandir(".");
+ scanfile(".");
} catch (IOException e) {
fprintf(stderr, "IOException: %s\n", e.getError().c_str());
}
+ metadata_flush();
+ const string md = metadata_root.str();
+
+ LbsObject *root = new LbsObject;
+ root->set_group("root");
+ 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";
+ strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
+ descriptor << "Date: " << desc_buf << "\n";
+
+ delete root;
+
+ descriptor << "Segments:\n";
+ for (std::set<string>::iterator i = segment_list.begin();
+ i != segment_list.end(); ++i) {
+ descriptor << " " << *i << "\n";
+ }
+
+ db->Close();
+
+ tss->sync();
+ delete tss;
+
return 0;
}
projects / cumulus.git / blobdiff