X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=99498c4a753a434db415e87d5bf1ca9fc35ed4ad;hb=e053c8481ac962711348e466496e8a0c5a53f100;hp=3e004fdfc29aee5af00282504fc94dcdc38f04aa;hpb=3cb5ea93e29ba5baf6a5d1f170bbc003d7e2fe35;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index 3e004fd..99498c4 100644 --- a/scandir.cc +++ b/scandir.cc @@ -1,62 +1,215 @@ /* Recursively descend the filesystem and visit each file. */ -#include -#include -#include #include #include #include -#include +#include +#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 "sha1.h" +#include "statcache.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; + +/* Local database, which tracks objects written in this and previous + * invocations to help in creating incremental snapshots. */ +LocalDb *db; -static OutputStream *info_dump = NULL; +/* 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; + +/* 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); -/* Converts time to microseconds since the epoch. */ -int64_t encode_time(time_t time) +/* Selection of files to include/exclude in the snapshot. */ +std::list excludes; + +/* Ensure contents of metadata are flushed to an object. */ +void metadata_flush() +{ + string m = metadata.str(); + if (m.size() == 0) + return; + + /* Write current metadata information to a new object. */ + LbsObject *meta = new LbsObject; + meta->set_group("metadata"); + 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) { - return (int64_t)time * 1000000; + 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, const string &path) { struct stat stat_buf; fstat(fd, &stat_buf); int64_t size = 0; - - char buf[4096]; + 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. */ + 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); + + // 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; + + /* 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; } - printf(" bytes=%Ld\n", size); + 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. */ + 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("metadata"); + 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) @@ -66,6 +219,21 @@ void scanfile(const string& path) 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; + + // Check this file against the include/exclude list to see if it should be + // considered + for (list::iterator i = excludes.begin(); + i != excludes.end(); ++i) { + if (path == *i) { + printf("Excluding %s\n", path.c_str()); + return; + } + } dictionary file_info; @@ -73,12 +241,20 @@ void scanfile(const string& path) printf("%s\n", path.c_str()); - 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; @@ -104,16 +280,14 @@ 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"); } - file_info["contents"] = buf; - delete[] buf; break; case S_IFREG: @@ -129,22 +303,38 @@ void scanfile(const string& path) * - 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); - //dumpfile(fd); - file_info["size"] = encode_u64(stat_buf.st_size); + file_size = dumpfile(fd, file_info, path); + 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'; - scandir(path); + recurse = true; break; default: @@ -154,8 +344,18 @@ void scanfile(const string& path) file_info["type"] = string(1, inode_type); - info_dump->write_string(path); - info_dump->write_dictionary(file_info); + metadata << "name: " << uri_encode(path) << "\n"; + 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) @@ -163,7 +363,7 @@ void scandir(const string& path) DIR *dir = opendir(path.c_str()); if (dir == NULL) { - printf("Error: %m\n"); + fprintf(stderr, "Error: %m\n"); return; } @@ -181,22 +381,109 @@ void scandir(const string& path) for (vector::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); } +void usage(const char *program) +{ + 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[]) { - FILE *dump = fopen("fileinfo", "w"); - if (dump == NULL) { - fprintf(stderr, "Cannot open fileinfo: %m\n"); + string backup_source = "."; + string backup_dest = "."; + string localdb_dir = ""; + + 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}, + }; + + int long_index; + int c = getopt_long(argc, argv, "", long_options, &long_index); + + if (c == -1) + break; + + 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; } - FileOutputStream os(dump); - info_dump = &os; + 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); + + /* 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("."); @@ -204,5 +491,42 @@ int main(int argc, char *argv[]) fprintf(stderr, "IOException: %s\n", e.getError().c_str()); } + metadata_flush(); + const string md = metadata_root.str(); + + LbsObject *root = new LbsObject; + root->set_group("metadata"); + root->set_data(md.data(), md.size()); + root->write(tss); + root->checksum(); + segment_list.insert(root->get_ref().get_segment()); + + /* 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; + + descriptor << "Segments:\n"; + for (std::set::iterator i = segment_list.begin(); + i != segment_list.end(); ++i) { + descriptor << " " << *i << "\n"; + } + + db->Close(); + + statcache->Close(); + delete statcache; + + tss->sync(); + delete tss; + return 0; }