X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=223a99c7776b54952b9976389dbf92cb12e7696e;hb=51859528c5de1c90c553d1f174869005711f162a;hp=c8e43a8068776218b2f0da47354ab9a27ce58ecc;hpb=221ce05266860848c776e614b23216efb4daf573;p=cumulus.git diff --git a/scandir.cc b/scandir.cc index c8e43a8..223a99c 100644 --- a/scandir.cc +++ b/scandir.cc @@ -12,68 +12,170 @@ #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 int LBS_BLOCK_SIZE = 1024 * 1024; +static char *block_buf; + +void scandir(const string& path, std::ostream& metadata); -void scandir(const string& path); +/* Converts time to microseconds since the epoch. */ +int64_t encode_time(time_t time) +{ + return (int64_t)time * 1000000; +} -void dumpfile(int fd) +/* 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; +} + +/* 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) { struct stat stat_buf; fstat(fd, &stat_buf); int64_t size = 0; - - char buf[4096]; + list segment_list; if ((stat_buf.st_mode & S_IFMT) != S_IFREG) { printf("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) { - 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); + segment_list.push_back(o->get_name()); + delete o; + + size += bytes; } - printf(" bytes=%Ld\n", size); + 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 (segment_list.size() < 8) { + string blocklist = ""; + for (list::iterator i = segment_list.begin(); + i != segment_list.end(); ++i) { + if (i != segment_list.begin()) + blocklist += " "; + blocklist += *i; + } + file_info["data"] = blocklist; + } else { + string blocklist = ""; + for (list::iterator i = segment_list.begin(); + i != segment_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(); + delete i; + } } -void scanfile(const string& path) +void scanfile(const string& path, ostream &metadata) { int fd; long flags; 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 @@ -88,10 +190,14 @@ void scanfile(const string& path) } else if (len > stat_buf.st_size) { printf("error reading symlink: name truncated\n"); } + + file_info["contents"] = uri_encode(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 * 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 @@ -110,18 +216,34 @@ 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, metadata); 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); + metadata << "type: " << inode_type << "\n"; + + dict_output(metadata, file_info); + metadata << "\n"; + + // If we hit a directory, now that we've written the directory itself, + // recursively scan the directory. + if (recurse) + scandir(path, metadata); } -void scandir(const string& path) +void scandir(const string& path, ostream &metadata) { DIR *dir = opendir(path.c_str()); @@ -144,7 +266,7 @@ void scandir(const string& path) for (vector::iterator i = contents.begin(); i != contents.end(); ++i) { const string& filename = *i; - scanfile(path + "/" + filename); + scanfile(path + "/" + filename, metadata); } closedir(dir); @@ -152,7 +274,28 @@ void scandir(const string& path) int main(int argc, char *argv[]) { - scandir("."); + block_buf = new char[LBS_BLOCK_SIZE]; + + tss = new TarSegmentStore("."); + + std::ostringstream metadata; + + try { + scanfile(".", metadata); + } catch (IOException e) { + fprintf(stderr, "IOException: %s\n", e.getError().c_str()); + } + + const string md = metadata.str(); + + LbsObject *r = new LbsObject; + r->set_group("root"); + r->set_data(md.data(), md.size()); + r->write(tss); + delete r; + + tss->sync(); + delete tss; return 0; }