X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=d398eeebac34229bd255d6aea449e9c61d218035;hb=4bda3653b0b3a270acdf643cf4918a97dbf147c5;hp=3d4dcb5c46e2f944fb4bd84f83cfa325b4516ef2;hpb=715efebee9dff8df72feda19e77ee516798444be;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index 3d4dcb5..d398eee 100644 --- a/scandir.cc +++ b/scandir.cc @@ -10,6 +10,7 @@ #include #include #include +#include #include #include @@ -22,16 +23,25 @@ #include #include -#include "format.h" #include "localdb.h" #include "store.h" #include "sha1.h" +#include "statcache.h" +#include "util.h" using std::list; using std::string; using std::vector; using std::ostream; +/* Version information. This will be filled in by the Makefile. */ +#ifndef LBS_VERSION +#define LBS_VERSION Unknown +#endif +#define LBS_STRINGIFY(s) LBS_STRINGIFY2(s) +#define LBS_STRINGIFY2(s) #s +static const char lbs_version[] = LBS_STRINGIFY(LBS_VERSION); + static TarSegmentStore *tss = NULL; /* Buffer for holding a single block of data read from a file. */ @@ -44,6 +54,10 @@ static const size_t LBS_METADATA_BLOCK_SIZE = 65536; * 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; @@ -54,10 +68,16 @@ 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 scandir(const string& path, bool include); /* Selection of files to include/exclude in the snapshot. */ -std::list excludes; +std::list includes; // Paths in which files should be saved +std::list excludes; // Paths which will not be saved +std::list searches; // Directories we don't want to save, but + // do want to descend searching for data + // in included paths + +bool relative_paths = true; /* Ensure contents of metadata are flushed to an object. */ void metadata_flush() @@ -111,7 +131,7 @@ 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. 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); @@ -123,56 +143,98 @@ int64_t dumpfile(int fd, dictionary &file_info) 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) { - size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); - if (bytes == 0) - break; + /* Look up this file in the old stat cache, if we can. If the stat + * information indicates that the file has not changed, do not bother + * re-reading the entire contents. */ + bool cached = false; + + if (statcache->Find(path, &stat_buf)) { + cached = true; + const list &blocks = statcache->get_blocks(); + + /* If any of the blocks in the object have been expired, then we should + * fall back to fully reading in the file. */ + for (list::const_iterator i = blocks.begin(); + i != blocks.end(); ++i) { + const ObjectReference &ref = *i; + if (!db->IsAvailable(ref)) { + cached = false; + break; + } + } - 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; + /* If everything looks okay, use the cached information */ + if (cached) { + file_info["checksum"] = statcache->get_checksum(); + for (list::const_iterator i = blocks.begin(); + i != blocks.end(); ++i) { + const ObjectReference &ref = *i; + object_list.push_back(ref.to_string()); + segment_list.insert(ref.get_segment()); + db->UseObject(ref); + } + size = stat_buf.st_size; } + } + + /* If the file is new or changed, we must read in the contents a block at a + * time. */ + if (!cached) { + printf(" [new]\n"); - object_list.push_back(ref.to_string()); - segment_list.insert(ref.get_segment()); - db->UseObject(ref); - size += bytes; + SHA1Checksum hash; + while (true) { + size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE); + if (bytes == 0) + break; + + 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 + double block_age = 0.0; + 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. + * Additionally, keep track of the age of the data by looking + * up the age of the block which was expired and using that + * instead of the current time. */ + if (db->IsOldObject(block_csum, bytes, &block_age)) + 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, block_age); + 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(); } - 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 @@ -205,7 +267,7 @@ int64_t dumpfile(int fd, dictionary &file_info) return size; } -void scanfile(const string& path) +void scanfile(const string& path, bool include) { int fd; long flags; @@ -215,22 +277,51 @@ void scanfile(const string& path) int64_t file_size; list refs; + string true_path; + if (relative_paths) + true_path = path; + else + true_path = "/" + path; + // Set to true if the item is a directory and we should recursively scan bool recurse = false; + // Set to true if we should scan through the contents of this directory, + // but not actually back files up + bool scan_only = false; + // Check this file against the include/exclude list to see if it should be // considered + for (list::iterator i = includes.begin(); + i != includes.end(); ++i) { + if (path == *i) { + printf("Including %s\n", path.c_str()); + include = true; + } + } + for (list::iterator i = excludes.begin(); i != excludes.end(); ++i) { if (path == *i) { printf("Excluding %s\n", path.c_str()); - return; + include = false; + } + } + + for (list::iterator i = searches.begin(); + i != searches.end(); ++i) { + if (path == *i) { + printf("Scanning %s\n", path.c_str()); + scan_only = true; } } + if (!include && !scan_only) + return; + dictionary file_info; - lstat(path.c_str(), &stat_buf); + lstat(true_path.c_str(), &stat_buf); printf("%s\n", path.c_str()); @@ -258,11 +349,11 @@ void scanfile(const string& path) case S_IFSOCK: inode_type = 's'; break; - case S_IFCHR: - inode_type = 'c'; - break; case S_IFBLK: - inode_type = 'b'; + case S_IFCHR: + inode_type = ((stat_buf.st_mode & S_IFMT) == S_IFBLK) ? 'b' : 'c'; + file_info["device"] = encode_int(major(stat_buf.st_rdev)) + + "/" + encode_int(minor(stat_buf.st_rdev)); break; case S_IFLNK: inode_type = 'l'; @@ -271,7 +362,7 @@ void scanfile(const string& path) * the symlink. Allocate slightly more space, so that we ask for more * bytes than we expect and so check for truncation. */ buf = new char[stat_buf.st_size + 2]; - len = readlink(path.c_str(), buf, stat_buf.st_size + 1); + len = readlink(true_path.c_str(), buf, stat_buf.st_size + 1); if (len < 0) { fprintf(stderr, "error reading symlink: %m\n"); } else if (len <= stat_buf.st_size) { @@ -298,9 +389,9 @@ void scanfile(const string& path) * We also add in O_NOATIME, since this may reduce disk writes (for * 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); + fd = open(true_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); + fd = open(true_path.c_str(), O_RDONLY|O_NOFOLLOW|O_NONBLOCK); } if (fd < 0) { fprintf(stderr, "Unable to open file %s: %m\n", path.c_str()); @@ -312,7 +403,7 @@ void scanfile(const string& path) 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); @@ -348,12 +439,18 @@ void scanfile(const string& path) // If we hit a directory, now that we've written the directory itself, // recursively scan the directory. if (recurse) - scandir(path); + scandir(path, include); } -void scandir(const string& path) +void scandir(const string& path, bool include) { - DIR *dir = opendir(path.c_str()); + string true_path; + if (relative_paths) + true_path = path; + else + true_path = "/" + path; + + DIR *dir = opendir(true_path.c_str()); if (dir == NULL) { fprintf(stderr, "Error: %m\n"); @@ -375,30 +472,80 @@ void scandir(const string& path) i != contents.end(); ++i) { const string& filename = *i; if (path == ".") - scanfile(filename); + scanfile(filename, include); else - scanfile(path + "/" + filename); + scanfile(path + "/" + filename, include); } closedir(dir); } +/* Include the specified file path in the backups. Append the path to the + * includes list, and to ensure that we actually see the path when scanning the + * directory tree, add all the parent directories to the search list, which + * means we will scan through the directory listing even if the files + * themselves are excluded from being backed up. */ +void add_include(const char *path) +{ + printf("Add: %s\n", path); + /* Was an absolute path specified? If so, we'll need to start scanning + * from the root directory. Make sure that the user was consistent in + * providing either all relative paths or all absolute paths. */ + if (path[0] == '/') { + if (includes.size() > 0 && relative_paths == true) { + fprintf(stderr, + "Error: Cannot mix relative and absolute paths!\n"); + exit(1); + } + + relative_paths = false; + + // Skip over leading '/' + path++; + } else if (relative_paths == false && path[0] != '/') { + fprintf(stderr, "Error: Cannot mix relative and absolute paths!\n"); + exit(1); + } + + includes.push_back(path); + + /* Split the specified path into directory components, and ensure that we + * descend into all the directories along the path. */ + const char *slash = path; + + if (path[0] == '\0') + return; + + while ((slash = strchr(slash + 1, '/')) != NULL) { + string component(path, slash - path); + searches.push_back(component); + } +} + 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); + fprintf( + stderr, + "Usage: %s [OPTION]... --dest=DEST PATHS...\n" + "Produce backup snapshot of files in SOURCE and store to DEST.\n" + "\n" + "Options:\n" + " --dest=PATH path where backup is to be written [REQUIRED]\n" + " --exclude=PATH exclude files in PATH from snapshot\n" + " --localdb=PATH local backup metadata is stored in PATH\n" + " --filter=COMMAND program through which to filter segment data\n" + " (defaults to \"bzip2 -c\")\n" + " --filter-extension=EXT\n" + " string to append to segment files\n" + " (defaults to \".bz2\")\n", + program + ); } int main(int argc, char *argv[]) { string backup_source = "."; - string backup_dest = "."; + string backup_dest = ""; string localdb_dir = ""; while (1) { @@ -407,6 +554,7 @@ int main(int argc, char *argv[]) {"exclude", 1, 0, 0}, // 1 {"filter", 1, 0, 0}, // 2 {"filter-extension", 1, 0, 0}, // 3 + {"dest", 1, 0, 0}, // 4 {NULL, 0, 0, 0}, }; @@ -422,7 +570,10 @@ int main(int argc, char *argv[]) localdb_dir = optarg; break; case 1: // --exclude - excludes.push_back(optarg); + if (optarg[0] != '/') + excludes.push_back(optarg); + else + excludes.push_back(optarg + 1); break; case 2: // --filter filter_program = optarg; @@ -430,6 +581,9 @@ int main(int argc, char *argv[]) case 3: // --filter-extension filter_extension = optarg; break; + case 4: // --dest + backup_dest = optarg; + break; default: fprintf(stderr, "Unhandled long option!\n"); return 1; @@ -445,30 +599,61 @@ int main(int argc, char *argv[]) return 1; } + searches.push_back("."); + if (optind == argc) { + add_include("."); + } else { + for (int i = optind; i < argc; i++) + add_include(argv[i]); + } + backup_source = argv[optind]; - backup_dest = argv[argc - 1]; + if (backup_dest == "") { + fprintf(stderr, + "Error: Backup destination must be specified with --dest=\n"); + usage(argv[0]); + return 1; + } + + // Default for --localdb is the same as --dest 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()); + // Dump paths for debugging/informational purposes + { + list::const_iterator i; + + printf("LBS Version: %s\n", lbs_version); + + printf("--dest=%s\n--localdb=%s\n\n", + backup_dest.c_str(), localdb_dir.c_str()); + + printf("Includes:\n"); + for (i = includes.begin(); i != includes.end(); ++i) + printf(" %s\n", i->c_str()); + + printf("Excludes:\n"); + for (i = excludes.begin(); i != excludes.end(); ++i) + printf(" %s\n", i->c_str()); + + printf("Searching:\n"); + for (i = searches.begin(); i != searches.end(); ++i) + printf(" %s\n", i->c_str()); + } tss = new TarSegmentStore(backup_dest); block_buf = new char[LBS_BLOCK_SIZE]; - /* 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. */ + /* 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 @@ -477,11 +662,11 @@ int main(int argc, char *argv[]) db = new LocalDb; db->Open(database_path.c_str(), desc_buf); - try { - scanfile("."); - } catch (IOException e) { - fprintf(stderr, "IOException: %s\n", e.getError().c_str()); - } + /* Initialize the stat cache, for skipping over unchanged files. */ + statcache = new StatCache; + statcache->Open(localdb_dir.c_str(), desc_buf); + + scanfile(".", false); metadata_flush(); const string md = metadata_root.str(); @@ -491,14 +676,31 @@ int main(int argc, char *argv[]) root->set_data(md.data(), md.size()); root->write(tss); root->checksum(); - segment_list.insert(root->get_ref().get_segment()); + + string backup_root = root->get_ref().to_string(); + delete root; + + db->Close(); + + statcache->Close(); + delete statcache; + + tss->sync(); + tss->dump_stats(); + delete tss; + + /* 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"; + descriptor << "Producer: " << lbs_version << "\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 << "Root: " << backup_root << "\n"; descriptor << "Segments:\n"; for (std::set::iterator i = segment_list.begin(); @@ -506,10 +708,5 @@ int main(int argc, char *argv[]) descriptor << " " << *i << "\n"; } - db->Close(); - - tss->sync(); - delete tss; - return 0; }