X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=0dfdbdba58518dd7e7f91ec9cd375a8c86110731;hb=0a76f30b578a2ace26090a4b9bbd74b04124a20e;hp=c1bedeb986be708c387cb340abe9d3720a22e6ab;hpb=c5e37e726177856fa06bcc2b4294fc63ae6fa1ef;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index c1bedeb..0dfdbdb 100644 --- a/scandir.cc +++ b/scandir.cc @@ -6,42 +6,77 @@ #include #include #include +#include +#include #include #include #include #include #include +#include #include #include +#include #include +#include #include "format.h" +#include "localdb.h" #include "store.h" -#include "tarstore.h" #include "sha1.h" +using std::list; using std::string; using std::vector; using std::ostream; -static SegmentStore *segment_store; -static OutputStream *info_dump = NULL; - static TarSegmentStore *tss = NULL; -static SegmentPartitioner *index_segment, *data_segment; - /* Buffer for holding a single block of data read from a file. */ -static const int LBS_BLOCK_SIZE = 1024 * 1024; +static const size_t LBS_BLOCK_SIZE = 1024 * 1024; static char *block_buf; -void scandir(const string& path, std::ostream& metadata); +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; -/* Converts time to microseconds since the epoch. */ -int64_t encode_time(time_t time) +/* Keep track of all segments which are needed to reconstruct the snapshot. */ +std::set segment_list; + +void scandir(const string& path); + +/* 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 @@ -70,67 +105,96 @@ size_t file_read(int fd, char *buf, size_t maxlen) } /* 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, ostream &metadata) + * 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; - string segment_list = "data:"; + list 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. */ - struct uuid segment_uuid; - int object_id; - OutputStream *index_data = index_segment->new_object(&segment_uuid, - &object_id, - "DREF"); - SHA1Checksum hash; while (true) { - struct uuid block_segment_uuid; - int block_object_id; - size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); if (bytes == 0) break; hash.process(block_buf, bytes); - OutputStream *block = data_segment->new_object(&block_segment_uuid, - &block_object_id, - "DATA"); - block->write(block_buf, bytes); - index_data->write_uuid(block_segment_uuid); - index_data->write_u32(block_object_id); - - // tarstore processing - string blockid = tss->write_object(block_buf, bytes, "data"); - segment_list += " " + blockid; + // Either find a copy of this block in an already-existing segment, or + // index it so it can be re-used in the future + SHA1Checksum block_hash; + block_hash.process(block_buf, bytes); + string block_csum = block_hash.checksum_str(); + ObjectReference ref = db->FindObject(block_csum, bytes); + + // 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"); + 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()); size += bytes; } - file_info["sha1"] = string((const char *)hash.checksum(), - hash.checksum_size()); - file_info["data"] = encode_objref(segment_uuid, object_id); + 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"; + } - metadata << segment_list << "\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; + } + + return size; } -void scanfile(const string& path, ostream &metadata) +void scanfile(const string& path) { int fd; long flags; struct stat stat_buf; char *buf; ssize_t len; + int64_t file_size; + list refs; // Set to true if the item is a directory and we should recursively scan bool recurse = false; @@ -142,19 +206,21 @@ void scanfile(const string& path, ostream &metadata) printf("%s\n", path.c_str()); metadata << "name: " << uri_encode(path) << "\n"; - metadata << "mode: " << (stat_buf.st_mode & 07777) << "\n"; - metadata << "atime: " << stat_buf.st_atime << "\n"; - metadata << "ctime: " << stat_buf.st_ctime << "\n"; - metadata << "mtime: " << stat_buf.st_mtime << "\n"; - metadata << "user: " << stat_buf.st_uid << "\n"; - metadata << "group: " << stat_buf.st_gid << "\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; @@ -180,16 +246,14 @@ void scanfile(const string& path, ostream &metadata) 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: @@ -213,10 +277,18 @@ void scanfile(const string& path, ostream &metadata) flags = fcntl(fd, F_GETFL); fcntl(fd, F_SETFL, flags & ~O_NONBLOCK); - file_info["size"] = encode_u64(stat_buf.st_size); - dumpfile(fd, file_info, metadata); + 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: inode_type = 'd'; @@ -229,25 +301,26 @@ void scanfile(const string& path, ostream &metadata) } file_info["type"] = string(1, inode_type); - metadata << "type: " << inode_type << "\n"; - - 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, metadata); + scandir(path); } -void scandir(const string& path, ostream &metadata) +void scandir(const string& path) { DIR *dir = opendir(path.c_str()); if (dir == NULL) { - printf("Error: %m\n"); + fprintf(stderr, "Error: %m\n"); return; } @@ -265,7 +338,7 @@ void scandir(const string& path, ostream &metadata) for (vector::iterator i = contents.begin(); i != contents.end(); ++i) { const string& filename = *i; - scanfile(path + "/" + filename, metadata); + scanfile(path + "/" + filename); } closedir(dir); @@ -275,36 +348,61 @@ int main(int argc, char *argv[]) { block_buf = new char[LBS_BLOCK_SIZE]; - tss = new TarSegmentStore("."); - segment_store = new SegmentStore("."); - SegmentWriter *sw = segment_store->new_segment(); - info_dump = sw->new_object(NULL, "ROOT"); + string backup_dest = "."; + + if (argc > 1) + backup_dest = argv[1]; - index_segment = new SegmentPartitioner(segment_store); - data_segment = new SegmentPartitioner(segment_store); + tss = new TarSegmentStore(backup_dest); - string uuid = SegmentWriter::format_uuid(sw->get_uuid()); - printf("Backup UUID: %s\n", uuid.c_str()); + string database_path = backup_dest + "/localdb.sqlite"; + db = new LocalDb; + db->Open(database_path.c_str()); - std::ostringstream metadata; + /* 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 { - scanfile(".", metadata); + scanfile("."); } catch (IOException e) { fprintf(stderr, "IOException: %s\n", e.getError().c_str()); } - const string md = metadata.str(); - string root = tss->write_object(md.data(), md.size(), "root"); + 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"; - fprintf(stderr, "Metadata root is at %s\n", root.c_str()); + delete root; + + descriptor << "Segments:\n"; + for (std::set::iterator i = segment_list.begin(); + i != segment_list.end(); ++i) { + descriptor << " " << *i << "\n"; + } + + db->Close(); tss->sync(); delete tss; - delete index_segment; - delete data_segment; - delete sw; - return 0; }