X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=30b8d4572aa6eab67a9796018f8ba8c59310e6a4;hb=6ac0cb32a08c81ddb87cd3a1373e8e64e8757534;hp=c8e43a8068776218b2f0da47354ab9a27ce58ecc;hpb=221ce05266860848c776e614b23216efb4daf573;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index c8e43a8..30b8d45 100644 --- a/scandir.cc +++ b/scandir.cc @@ -12,41 +12,158 @@ #include #include +#include #include +#include +#include +#include +#include +#include "format.h" +#include "store.h" +#include "sha1.h" + +using std::list; using std::string; using std::vector; +using std::ostream; + +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; + +/* 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 segment_list; void scandir(const string& path); -void dumpfile(int fd) +/* Ensure contents of metadata are flushed to an object. */ +void metadata_flush() { - struct stat stat_buf; - fstat(fd, &stat_buf); - int64_t size = 0; + string m = metadata.str(); + if (m.size() == 0) + return; - char buf[4096]; + /* 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(); - if ((stat_buf.st_mode & S_IFMT) != S_IFREG) { - printf("file is no longer a regular file!\n"); - return; - } + /* 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, sizeof(buf)); + ssize_t res = read(fd, buf, maxlen); if (res < 0) { if (errno == EINTR) continue; - printf("Error while reading: %m\n"); - return; + throw IOException("file_read: error reading"); } else if (res == 0) { break; } else { - size += res; + bytes_read += res; + buf += res; + maxlen -= res; } } - printf(" bytes=%Ld\n", size); + return bytes_read; +} + +/* Read the contents of a file (specified by an open file descriptor) and copy + * the data to the store. */ +void dumpfile(int fd, dictionary &file_info) +{ + struct stat stat_buf; + fstat(fd, &stat_buf); + int64_t size = 0; + list object_list; + + if ((stat_buf.st_mode & S_IFMT) != S_IFREG) { + fprintf(stderr, "file is no longer a regular file!\n"); + return; + } + + /* 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) { + size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); + if (bytes == 0) + break; + + 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()); + delete o; + + size += bytes; + } + + 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::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::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; + } } void scanfile(const string& path) @@ -56,24 +173,43 @@ void scanfile(const string& path) struct stat stat_buf; char *buf; ssize_t len; + list 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_int(stat_buf.st_mode & 07777); + file_info["atime"] = encode_int(stat_buf.st_atime); + file_info["ctime"] = encode_int(stat_buf.st_ctime); + 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); + + char inode_type; switch (stat_buf.st_mode & S_IFMT) { case S_IFIFO: + inode_type = 'p'; + break; case S_IFSOCK: + inode_type = 's'; + break; case S_IFCHR: + inode_type = 'c'; + break; case S_IFBLK: - printf(" special file\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 * the symlink. Allocate slightly more space, so that we ask for more @@ -81,17 +217,19 @@ void scanfile(const string& path) 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"); } + delete[] buf; break; case S_IFREG: - printf(" regular file\n"); + inode_type = '-'; + /* Be paranoid when opening the file. We have no guarantee that the * file was not replaced between the stat() call above and the open() * call below, so we might not even be opening a regular file. That @@ -110,15 +248,34 @@ void scanfile(const string& path) flags = fcntl(fd, F_GETFL); fcntl(fd, F_SETFL, flags & ~O_NONBLOCK); - dumpfile(fd); + file_info["size"] = encode_int(stat_buf.st_size); + dumpfile(fd, file_info); close(fd); break; case S_IFDIR: - printf(" directory\n"); - scandir(path); + inode_type = 'd'; + recurse = true; break; + + default: + fprintf(stderr, "Unknown inode type: mode=%x\n", stat_buf.st_mode); + return; } + + file_info["type"] = string(1, inode_type); + + 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) @@ -126,7 +283,7 @@ void scandir(const string& path) DIR *dir = opendir(path.c_str()); if (dir == NULL) { - printf("Error: %m\n"); + fprintf(stderr, "Error: %m\n"); return; } @@ -152,7 +309,46 @@ void scandir(const string& path) int main(int argc, char *argv[]) { - scandir("."); + block_buf = new char[LBS_BLOCK_SIZE]; + + string backup_dest = "."; + + if (argc > 1) + backup_dest = argv[1]; + + tss = new TarSegmentStore(backup_dest); + + string desc_filename = backup_dest + "/snapshot.lbs"; + std::ofstream descriptor(desc_filename.c_str()); + + try { + 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\n"; + + delete root; + + descriptor << "segments:\n"; + for (std::set::iterator i = segment_list.begin(); + i != segment_list.end(); ++i) { + descriptor << " " << *i << "\n"; + } + + tss->sync(); + delete tss; return 0; }