-/* Cumulus: Smart Filesystem Backup to Dumb Servers
- *
- * Copyright (C) 2006-2008 The Regents of the University of California
- * Written by Michael Vrable <mvrable@cs.ucsd.edu>
+/* Cumulus: Efficient Filesystem Backup to the Cloud
+ * Copyright (C) 2006-2009, 2012 The Cumulus Developers
+ * See the AUTHORS file for a list of contributors.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
-/* Main entry point for LBS. Contains logic for traversing the filesystem and
- * constructing a backup. */
+/* Main entry point for Cumulus. Contains logic for traversing the filesystem
+ * and constructing a backup. */
#include <dirent.h>
#include <errno.h>
#include <fstream>
#include <iostream>
#include <list>
+#include <map>
+#include <memory>
#include <set>
#include <sstream>
#include <string>
#include <vector>
+#include "exclude.h"
+#include "hash.h"
#include "localdb.h"
#include "metadata.h"
#include "remote.h"
#include "store.h"
-#include "sha1.h"
#include "subfile.h"
#include "util.h"
+#include "third_party/sha1.h"
using std::list;
+using std::map;
using std::string;
using std::vector;
using std::ostream;
+using std::unique_ptr;
/* Version information. This will be filled in by the Makefile. */
#ifndef CUMULUS_VERSION
* invocations to help in creating incremental snapshots. */
LocalDb *db;
-/* Keep track of all segments which are needed to reconstruct the snapshot. */
-std::set<string> segment_list;
-
-/* Snapshot intent: 1=daily, 7=weekly, etc. This is not used directly, but is
- * stored in the local database and can help guide segment cleaning and
- * snapshot expiration policies. */
-double snapshot_intent = 1.0;
-
/* Selection of files to include/exclude in the snapshot. */
-std::list<string> includes; // Paths in which files should be saved
-std::list<string> excludes; // Paths which will not be saved
-std::list<string> excluded_names; // Directories which will not be saved
-std::list<string> searches; // Directories we don't want to save, but
- // do want to descend searching for data
- // in included paths
-
-bool relative_paths = true;
+PathFilterList filter_rules;
bool flag_rebuild_statcache = false;
/* Whether verbose output is enabled. */
bool verbose = false;
-/* Ensure that the given segment is listed as a dependency of the current
- * snapshot. */
-void add_segment(const string& segment)
+/* Attempts to open a regular file read-only, but with safety checks for files
+ * that might not be fully trusted. */
+int safe_open(const string& path, struct stat *stat_buf)
{
- segment_list.insert(segment);
+ int fd;
+
+ /* 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. We
+ * supply flags to open to to guard against various conditions before
+ * we can perform an lstat to check that the file is still a regular
+ * file:
+ * - O_NOFOLLOW: in the event the file was replaced by a symlink
+ * - 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). 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 -1;
+ }
+
+ /* Drop the use of the O_NONBLOCK flag; we only wanted that for file
+ * open. */
+ long flags = fcntl(fd, F_GETFL);
+ fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
+
+ /* Re-check file attributes, storing them into stat_buf if that is
+ * non-NULL. */
+ struct stat internal_stat_buf;
+ if (stat_buf == NULL)
+ stat_buf = &internal_stat_buf;
+
+ /* Perform the stat call again, and check that we still have a regular
+ * file. */
+ if (fstat(fd, stat_buf) < 0) {
+ fprintf(stderr, "fstat: %m\n");
+ close(fd);
+ return -1;
+ }
+
+ if ((stat_buf->st_mode & S_IFMT) != S_IFREG) {
+ fprintf(stderr, "file is no longer a regular file!\n");
+ close(fd);
+ return -1;
+ }
+
+ return fd;
}
/* Read data from a file descriptor and return the amount of data read. A
i != old_blocks.end(); ++i) {
const ObjectReference &ref = *i;
object_list.push_back(ref.to_string());
- if (ref.is_normal())
- add_segment(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) {
- SHA1Checksum hash;
+ unique_ptr<Hash> file_hash(Hash::New());
Subfile subfile(db);
subfile.load_old_blocks(old_blocks);
break;
}
- hash.process(block_buf, bytes);
+ file_hash->update(block_buf, bytes);
// Sparse file processing: if we read a block of all zeroes, encode
// that explicitly.
double block_age = 0.0;
ObjectReference ref;
- SHA1Checksum block_hash;
- block_hash.process(block_buf, bytes);
- string block_csum = block_hash.checksum_str();
+ unique_ptr<Hash> block_hash(Hash::New());
+ block_hash->update(block_buf, bytes);
+ string block_csum = block_hash->digest_str();
if (all_zero) {
ref = ObjectReference(ObjectReference::REF_ZERO);
if (object_group == 0) {
o->set_group("data");
} else {
- char group[32];
- sprintf(group, "compacted-%d", object_group);
- o->set_group(group);
+ o->set_group(string_printf("compacted-%d",
+ object_group));
}
if (status == NULL)
status = "partial";
while (!refs.empty()) {
ref = refs.front(); refs.pop_front();
+
+ // The file-level checksum guarantees integrity of the data.
+ // To reduce the metadata log size, do not include checksums on
+ // individual objects.
+ ref.clear_checksum();
+
object_list.push_back(ref.to_string());
- if (ref.is_normal())
- add_segment(ref.get_segment());
db->UseObject(ref);
}
size += bytes;
status = "old";
}
- file_info["checksum"] = hash.checksum_str();
+ file_info["checksum"] = file_hash->digest_str();
}
- // Sanity check: if we are rebuilding the statcache, but the file looks
- // like it hasn't changed, then the newly-computed checksum should match
- // the checksum in the statcache. If not, we have possible disk corruption
- // and report a warning.
- if (flag_rebuild_statcache) {
- if (found
- && metawriter->is_unchanged(&stat_buf)
- && file_info["checksum"] != metawriter->get_checksum()) {
- fprintf(stderr,
- "Warning: Checksum for %s does not match expected value\n"
- " expected: %s\n"
- " actual: %s\n",
- path.c_str(),
- metawriter->get_checksum().c_str(),
- file_info["checksum"].c_str());
- }
+ // Sanity check: if the file looks like it hasn't changed, then the
+ // newly-computed checksum should match the checksum in the statcache. If
+ // not, we have possible disk corruption and report a warning.
+ if (found
+ && metawriter->is_unchanged(&stat_buf)
+ && file_info["checksum"] != metawriter->get_checksum()) {
+ fprintf(stderr,
+ "Warning: Checksum for %s does not match expected value\n"
+ " expected: %s\n"
+ " actual: %s\n",
+ path.c_str(),
+ metawriter->get_checksum().c_str(),
+ file_info["checksum"].c_str());
}
if (verbose && status != NULL)
return size;
}
+/* Look up a user/group and convert it to string form (either strictly numeric
+ * or numeric plus symbolic). Caches the results of the call to
+ * getpwuid/getgrgid. */
+string user_to_string(uid_t uid) {
+ static map<uid_t, string> user_cache;
+ map<uid_t, string>::const_iterator i = user_cache.find(uid);
+ if (i != user_cache.end())
+ return i->second;
+
+ string result = encode_int(uid);
+ struct passwd *pwd = getpwuid(uid);
+ if (pwd != NULL && pwd->pw_name != NULL) {
+ result += " (" + uri_encode(pwd->pw_name) + ")";
+ }
+ user_cache[uid] = result;
+ return result;
+}
+
+string group_to_string(gid_t gid) {
+ static map<gid_t, string> group_cache;
+ map<gid_t, string>::const_iterator i = group_cache.find(gid);
+ if (i != group_cache.end())
+ return i->second;
+
+ string result = encode_int(gid);
+ struct group *grp = getgrgid(gid);
+ if (grp != NULL && grp->gr_name != NULL) {
+ result += " (" + uri_encode(grp->gr_name) + ")";
+ }
+ group_cache[gid] = result;
+ return result;
+}
+
/* Dump a specified filesystem object (file, directory, etc.) based on its
* inode information. If the object is a regular file, an open filehandle is
* provided. */
file_info["mode"] = encode_int(stat_buf.st_mode & 07777, 8);
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);
+ file_info["user"] = user_to_string(stat_buf.st_uid);
+ file_info["group"] = group_to_string(stat_buf.st_gid);
time_t now = time(NULL);
if (now - stat_buf.st_ctime < 30 || now - stat_buf.st_mtime < 30)
if ((stat_buf.st_mode & S_IFMT) != S_IFDIR)
file_info["volatile"] = "1";
- struct passwd *pwd = getpwuid(stat_buf.st_uid);
- if (pwd != NULL && pwd->pw_name != NULL) {
- file_info["user"] += " (" + uri_encode(pwd->pw_name) + ")";
- }
-
- struct group *grp = getgrgid(stat_buf.st_gid);
- if (grp != NULL && grp->gr_name != NULL) {
- file_info["group"] += " (" + uri_encode(grp->gr_name) + ")";
- }
-
if (stat_buf.st_nlink > 1 && (stat_buf.st_mode & S_IFMT) != S_IFDIR) {
file_info["links"] = encode_int(stat_buf.st_nlink);
}
metawriter->add(file_info);
}
-void scanfile(const string& path, bool include)
+/* Converts a path to the normalized form used in the metadata log. Paths are
+ * written as relative (without any leading slashes). The root directory is
+ * referred to as ".". */
+string metafile_path(const string& path)
{
- int fd = -1;
- long flags;
- struct stat stat_buf;
- list<string> refs;
-
- string true_path;
- if (relative_paths)
- true_path = path;
- else
- true_path = "/" + path;
-
- // 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<string>::iterator i = includes.begin();
- i != includes.end(); ++i) {
- if (path == *i) {
- include = true;
- }
- }
-
- for (list<string>::iterator i = excludes.begin();
- i != excludes.end(); ++i) {
- if (path == *i) {
- include = false;
- }
- }
+ const char *newpath = path.c_str();
+ if (*newpath == '/')
+ newpath++;
+ if (*newpath == '\0')
+ newpath = ".";
+ return newpath;
+}
- if (excluded_names.size() > 0) {
- std::string name = path;
- std::string::size_type last_slash = name.rfind('/');
- if (last_slash != std::string::npos) {
- name.replace(0, last_slash + 1, "");
- }
+void try_merge_filter(const string& path, const string& basedir)
+{
+ struct stat stat_buf;
+ if (lstat(path.c_str(), &stat_buf) < 0)
+ return;
+ if ((stat_buf.st_mode & S_IFMT) != S_IFREG)
+ return;
+ int fd = safe_open(path, NULL);
+ if (fd < 0)
+ return;
- for (list<string>::iterator i = excluded_names.begin();
- i != excluded_names.end(); ++i) {
- if (name == *i) {
- include = false;
- }
- }
+ /* As a very crude limit on the complexity of merge rules, only read up to
+ * one block (1 MB) worth of data. If the file doesn't seems like it might
+ * be larger than that, don't parse the rules in it. */
+ ssize_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
+ close(fd);
+ if (bytes < 0 || bytes >= static_cast<ssize_t>(LBS_BLOCK_SIZE - 1)) {
+ /* TODO: Add more strict resource limits on merge files? */
+ fprintf(stderr,
+ "Unable to read filter merge file (possibly size too large\n");
+ return;
}
+ filter_rules.merge_patterns(metafile_path(path), basedir,
+ string(block_buf, bytes));
+}
- for (list<string>::iterator i = searches.begin();
- i != searches.end(); ++i) {
- if (path == *i) {
- scan_only = true;
- }
- }
+void scanfile(const string& path)
+{
+ int fd = -1;
+ struct stat stat_buf;
+ list<string> refs;
- if (!include && !scan_only)
- return;
+ string output_path = metafile_path(path);
- if (lstat(true_path.c_str(), &stat_buf) < 0) {
+ if (lstat(path.c_str(), &stat_buf) < 0) {
fprintf(stderr, "lstat(%s): %m\n", path.c_str());
return;
}
- if ((stat_buf.st_mode & S_IFMT) == S_IFREG) {
- /* 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. We
- * supply flags to open to to guard against various conditions before
- * we can perform an lstat to check that the file is still a regular
- * file:
- * - O_NOFOLLOW: in the event the file was replaced by a symlink
- * - 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). However, O_NOATIME may result in EPERM, so if the
- * initial open fails, try again without O_NOATIME. */
- fd = open(true_path.c_str(), O_RDONLY|O_NOATIME|O_NOFOLLOW|O_NONBLOCK);
- if (fd < 0) {
- 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());
- 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);
-
- /* Perform the stat call again, and check that we still have a regular
- * file. */
- if (fstat(fd, &stat_buf) < 0) {
- fprintf(stderr, "fstat: %m\n");
- close(fd);
- return;
- }
+ bool is_directory = ((stat_buf.st_mode & S_IFMT) == S_IFDIR);
+ if (!filter_rules.is_included(output_path, is_directory))
+ return;
- if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
- fprintf(stderr, "file is no longer a regular file!\n");
- close(fd);
+ if ((stat_buf.st_mode & S_IFMT) == S_IFREG) {
+ fd = safe_open(path, &stat_buf);
+ if (fd < 0)
return;
- }
}
- dump_inode(path, true_path, stat_buf, fd);
+ dump_inode(output_path, path, stat_buf, fd);
if (fd >= 0)
close(fd);
- // If we hit a directory, now that we've written the directory itself,
- // recursively scan the directory.
- if ((stat_buf.st_mode & S_IFMT) == S_IFDIR) {
- DIR *dir = opendir(true_path.c_str());
+ /* If we hit a directory, now that we've written the directory itself,
+ * recursively scan the directory. */
+ if (is_directory) {
+ DIR *dir = opendir(path.c_str());
if (dir == NULL) {
- fprintf(stderr, "Error: %m\n");
+ fprintf(stderr, "Error reading directory %s: %m\n",
+ path.c_str());
return;
}
sort(contents.begin(), contents.end());
+ filter_rules.save();
+
+ /* First pass through the directory items: look for any filter rules to
+ * merge and do so. */
for (vector<string>::iterator i = contents.begin();
i != contents.end(); ++i) {
- const string& filename = *i;
+ string filename;
if (path == ".")
- scanfile(filename, include);
+ filename = *i;
+ else if (path == "/")
+ filename = "/" + *i;
else
- scanfile(path + "/" + filename, include);
+ filename = path + "/" + *i;
+ if (filter_rules.is_mergefile(metafile_path(filename))) {
+ if (verbose) {
+ printf("Merging directory filter rules %s\n",
+ filename.c_str());
+ }
+ try_merge_filter(filename, output_path);
+ }
}
- }
-}
-/* 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)
-{
- /* 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);
+ /* Second pass: recursively scan all items in the directory for backup;
+ * scanfile() will check if the item should be included or not. */
+ for (vector<string>::iterator i = contents.begin();
+ i != contents.end(); ++i) {
+ const string& filename = *i;
+ if (path == ".")
+ scanfile(filename);
+ else if (path == "/")
+ scanfile("/" + filename);
+ else
+ scanfile(path + "/" + filename);
}
- 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);
+ filter_rules.restore();
}
}
" --dest=PATH path where backup is to be written\n"
" --upload-script=COMMAND\n"
" program to invoke for each backup file generated\n"
- " --exclude=PATH exclude files in PATH from snapshot\n"
- " --exclude-name=NAME exclude files called NAME from snapshot\n"
+ " --exclude=PATTERN exclude files matching PATTERN from snapshot\n"
+ " --include=PATTERN include files matching PATTERN in snapshot\n"
+ " --dir-merge=PATTERN parse files matching PATTERN to read additional\n"
+ " subtree-specific include/exclude rules during backup\n"
" --localdb=PATH local backup metadata is stored in PATH\n"
" --tmpdir=PATH path for temporarily storing backup files\n"
" (defaults to TMPDIR environment variable or /tmp)\n"
" --signature-filter=COMMAND\n"
" program though which to filter descriptor\n"
" --scheme=NAME optional name for this snapshot\n"
- " --intent=FLOAT intended backup type: 1=daily, 7=weekly, ...\n"
- " (defaults to \"1\")\n"
+ " --intent=FLOAT DEPRECATED: ignored, and will be removed soon\n"
" --full-metadata do not re-use metadata from previous backups\n"
" --rebuild-statcache re-read all file data to verify statcache\n"
" -v --verbose list files as they are backed up\n"
int main(int argc, char *argv[])
{
+ hash_init();
+
string backup_dest = "", backup_script = "";
string localdb_dir = "";
string backup_scheme = "";
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
- {"dest", 1, 0, 0}, // 4
- {"scheme", 1, 0, 0}, // 5
- {"signature-filter", 1, 0, 0}, // 6
- {"intent", 1, 0, 0}, // 7
- {"full-metadata", 0, 0, 0}, // 8
- {"tmpdir", 1, 0, 0}, // 9
- {"upload-script", 1, 0, 0}, // 10
- {"rebuild-statcache", 0, 0, 0}, // 11
- {"exclude-name", 1, 0, 0}, // 12
+ {"filter", 1, 0, 0}, // 1
+ {"filter-extension", 1, 0, 0}, // 2
+ {"dest", 1, 0, 0}, // 3
+ {"scheme", 1, 0, 0}, // 4
+ {"signature-filter", 1, 0, 0}, // 5
+ {"intent", 1, 0, 0}, // 6, DEPRECATED
+ {"full-metadata", 0, 0, 0}, // 7
+ {"tmpdir", 1, 0, 0}, // 8
+ {"upload-script", 1, 0, 0}, // 9
+ {"rebuild-statcache", 0, 0, 0}, // 10
+ {"include", 1, 0, 0}, // 11
+ {"exclude", 1, 0, 0}, // 12
+ {"dir-merge", 1, 0, 0}, // 13
// Aliases for short options
{"verbose", 0, 0, 'v'},
{NULL, 0, 0, 0},
case 0: // --localdb
localdb_dir = optarg;
break;
- case 1: // --exclude
- if (optarg[0] != '/')
- excludes.push_back(optarg);
- else
- excludes.push_back(optarg + 1);
- break;
- case 2: // --filter
+ case 1: // --filter
filter_program = optarg;
break;
- case 3: // --filter-extension
+ case 2: // --filter-extension
filter_extension = optarg;
break;
- case 4: // --dest
+ case 3: // --dest
backup_dest = optarg;
break;
- case 5: // --scheme
+ case 4: // --scheme
backup_scheme = optarg;
break;
- case 6: // --signature-filter
+ case 5: // --signature-filter
signature_filter = optarg;
break;
- case 7: // --intent
- snapshot_intent = atof(optarg);
- if (snapshot_intent <= 0)
- snapshot_intent = 1;
+ case 6: // --intent
+ fprintf(stderr,
+ "Warning: The --intent= option is deprecated and will "
+ "be removed in the future.\n");
break;
- case 8: // --full-metadata
+ case 7: // --full-metadata
flag_full_metadata = true;
break;
- case 9: // --tmpdir
+ case 8: // --tmpdir
tmp_dir = optarg;
break;
- case 10: // --upload-script
+ case 9: // --upload-script
backup_script = optarg;
break;
- case 11: // --rebuild-statcache
+ case 10: // --rebuild-statcache
flag_rebuild_statcache = true;
break;
- case 12: // --exclude-name
- excluded_names.push_back(optarg);
+ case 11: // --include
+ filter_rules.add_pattern(PathFilterList::INCLUDE, optarg, "");
+ break;
+ case 12: // --exclude
+ filter_rules.add_pattern(PathFilterList::EXCLUDE, optarg, "");
+ break;
+ case 13: // --dir-merge
+ filter_rules.add_pattern(PathFilterList::DIRMERGE, optarg, "");
break;
default:
fprintf(stderr, "Unhandled long option!\n");
return 1;
}
- searches.push_back(".");
- for (int i = optind; i < argc; i++)
- add_include(argv[i]);
-
if (backup_dest == "" && backup_script == "") {
fprintf(stderr,
"Error: Backup destination must be specified using --dest= or --upload-script=\n");
* a temporary directory for staging files. Otherwise, write backups
* directly to the destination directory. */
if (backup_script != "") {
- tmp_dir = tmp_dir + "/lbs." + generate_uuid();
+ tmp_dir = tmp_dir + "/cumulus." + generate_uuid();
if (mkdir(tmp_dir.c_str(), 0700) < 0) {
fprintf(stderr, "Cannot create temporary directory %s: %m\n",
tmp_dir.c_str());
/* 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 timestamp
+ = TimeFormat::format(now, TimeFormat::FORMAT_FILENAME, true);
/* 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, backup_scheme.c_str(),
- snapshot_intent);
+ db->Open(database_path.c_str(), timestamp.c_str(), backup_scheme.c_str());
tss = new TarSegmentStore(remote, db);
/* Initialize the stat cache, for skipping over unchanged files. */
- metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf,
+ metawriter = new MetadataWriter(tss, localdb_dir.c_str(), timestamp.c_str(),
backup_scheme.c_str());
- scanfile(".", false);
+ for (int i = optind; i < argc; i++) {
+ scanfile(argv[i]);
+ }
ObjectReference root_ref = metawriter->close();
- add_segment(root_ref.get_segment());
string backup_root = root_ref.to_string();
delete metawriter;
tss->dump_stats();
delete tss;
- /* Write out a checksums file which lists the checksums for all the
- * segments included in this snapshot. The format is designed so that it
- * may be easily verified using the sha1sums command. */
- const char csum_type[] = "sha1";
- string checksum_filename = "snapshot-";
+ /* Write out a summary file with metadata for all the segments in this
+ * snapshot (can be used to reconstruct database contents if needed), and
+ * contains hash values for the segments for quick integrity checks. */
+ string dbmeta_filename = "snapshot-";
if (backup_scheme.size() > 0)
- checksum_filename += backup_scheme + "-";
- checksum_filename = checksum_filename + desc_buf + "." + csum_type + "sums";
- RemoteFile *checksum_file = remote->alloc_file(checksum_filename,
- "checksums");
- FILE *checksums = fdopen(checksum_file->get_fd(), "w");
+ dbmeta_filename += backup_scheme + "-";
+ dbmeta_filename += timestamp + ".meta" + filter_extension;
+ RemoteFile *dbmeta_file = remote->alloc_file(dbmeta_filename, "meta");
+ unique_ptr<FileFilter> dbmeta_filter(FileFilter::New(dbmeta_file->get_fd(),
+ filter_program));
+ if (dbmeta_filter == NULL) {
+ fprintf(stderr, "Unable to open descriptor output file: %m\n");
+ return 1;
+ }
+ FILE *dbmeta = fdopen(dbmeta_filter->get_wrapped_fd(), "w");
+ std::set<string> segment_list = db->GetUsedSegments();
for (std::set<string>::iterator i = segment_list.begin();
i != segment_list.end(); ++i) {
- string seg_path, seg_csum;
- if (db->GetSegmentChecksum(*i, &seg_path, &seg_csum)) {
- const char *raw_checksum = NULL;
- if (strncmp(seg_csum.c_str(), csum_type,
- strlen(csum_type)) == 0) {
- raw_checksum = seg_csum.c_str() + strlen(csum_type);
- if (*raw_checksum == '=')
- raw_checksum++;
- else
- raw_checksum = NULL;
+ map<string, string> segment_metadata = db->GetSegmentMetadata(*i);
+ if (segment_metadata.size() > 0) {
+ map<string, string>::const_iterator j;
+ for (j = segment_metadata.begin();
+ j != segment_metadata.end(); ++j)
+ {
+ fprintf(dbmeta, "%s: %s\n",
+ j->first.c_str(), j->second.c_str());
}
-
- if (raw_checksum != NULL)
- fprintf(checksums, "%s *%s\n",
- raw_checksum, seg_path.c_str());
+ fprintf(dbmeta, "\n");
}
}
- fclose(checksums);
+ fclose(dbmeta);
+ dbmeta_filter->wait();
- SHA1Checksum checksum_csum;
- string csum;
- checksum_filename = checksum_file->get_local_path();
- if (checksum_csum.process_file(checksum_filename.c_str())) {
- csum = checksum_csum.checksum_str();
- }
-
- checksum_file->send();
+ string dbmeta_csum
+ = Hash::hash_file(dbmeta_file->get_local_path().c_str());
+ dbmeta_file->send();
db->Close();
string desc_filename = "snapshot-";
if (backup_scheme.size() > 0)
desc_filename += backup_scheme + "-";
- desc_filename = desc_filename + desc_buf + ".lbs";
+ desc_filename = desc_filename + timestamp + ".cumulus";
RemoteFile *descriptor_file = remote->alloc_file(desc_filename,
"snapshots");
- int descriptor_fd = descriptor_file->get_fd();
- if (descriptor_fd < 0) {
+ unique_ptr<FileFilter> descriptor_filter(
+ FileFilter::New(descriptor_file->get_fd(), signature_filter.c_str()));
+ if (descriptor_filter == NULL) {
fprintf(stderr, "Unable to open descriptor output file: %m\n");
return 1;
}
- pid_t signature_pid = 0;
- if (signature_filter.size() > 0) {
- int new_fd = spawn_filter(descriptor_fd, signature_filter.c_str(),
- &signature_pid);
- close(descriptor_fd);
- descriptor_fd = new_fd;
- }
- FILE *descriptor = fdopen(descriptor_fd, "w");
+ FILE *descriptor = fdopen(descriptor_filter->get_wrapped_fd(), "w");
- fprintf(descriptor, "Format: LBS Snapshot v0.8\n");
+ fprintf(descriptor, "Format: Cumulus Snapshot v0.11\n");
fprintf(descriptor, "Producer: Cumulus %s\n", cumulus_version);
- strftime(desc_buf, sizeof(desc_buf), "%Y-%m-%d %H:%M:%S %z", &time_buf);
- fprintf(descriptor, "Date: %s\n", desc_buf);
+ string timestamp_local
+ = TimeFormat::format(now, TimeFormat::FORMAT_LOCALTIME, false);
+ fprintf(descriptor, "Date: %s\n", timestamp_local.c_str());
if (backup_scheme.size() > 0)
fprintf(descriptor, "Scheme: %s\n", backup_scheme.c_str());
- fprintf(descriptor, "Backup-Intent: %g\n", snapshot_intent);
fprintf(descriptor, "Root: %s\n", backup_root.c_str());
- if (csum.size() > 0) {
- fprintf(descriptor, "Checksums: %s\n", csum.c_str());
+ if (dbmeta_csum.size() > 0) {
+ fprintf(descriptor, "Segment-metadata: %s\n", dbmeta_csum.c_str());
}
fprintf(descriptor, "Segments:\n");
}
fclose(descriptor);
-
- if (signature_pid) {
- int status;
- waitpid(signature_pid, &status, 0);
-
- if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
- fatal("Signature filter process error");
- }
+ if (descriptor_filter->wait() < 0) {
+ fatal("Signature filter process error");
}
descriptor_file->send();