X-Git-Url: http://git.vrable.net/?p=cumulus.git;a=blobdiff_plain;f=main.cc;h=c3f8c59ce57eba033fcef386a2d1bd45e3c0117f;hp=45a415428c2516b4f51574996ae932a308429027;hb=64bff41cb3ccdd60e767a5bb9ed8525d2dda1966;hpb=748d04c5c242a1dfcc932220eabd14f778e386c9 diff --git a/main.cc b/main.cc index 45a4154..c3f8c59 100644 --- a/main.cc +++ b/main.cc @@ -1,7 +1,6 @@ -/* Cumulus: Smart Filesystem Backup to Dumb Servers - * - * Copyright (C) 2006-2008 The Regents of the University of California - * Written by Michael Vrable +/* 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 @@ -41,20 +40,24 @@ #include #include #include +#include #include #include #include #include +#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; @@ -88,14 +91,7 @@ std::set segment_list; double snapshot_intent = 1.0; /* Selection of files to include/exclude in the snapshot. */ -std::list includes; // Paths in which files should be saved -std::list excludes; // Paths which will not be saved -std::list excluded_names; // Directories 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; +PathFilterList filter_rules; bool flag_rebuild_statcache = false; @@ -109,6 +105,61 @@ void add_segment(const string& segment) segment_list.insert(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) +{ + 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 * short read (less than the requested size) will only occur if end-of-file is * hit. */ @@ -192,7 +243,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, /* If the file is new or changed, we must read in the contents a block at a * time. */ if (!cached) { - SHA1Checksum hash; + Hash *hash = Hash::New(); Subfile subfile(db); subfile.load_old_blocks(old_blocks); @@ -206,7 +257,7 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, break; } - hash.process(block_buf, bytes); + hash->update(block_buf, bytes); // Sparse file processing: if we read a block of all zeroes, encode // that explicitly. @@ -291,7 +342,8 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, status = "old"; } - file_info["checksum"] = hash.checksum_str(); + file_info["checksum"] = hash->digest_str(); + delete hash; } // Sanity check: if we are rebuilding the statcache, but the file looks @@ -327,6 +379,39 @@ int64_t dumpfile(int fd, dictionary &file_info, const string &path, 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 user_cache; + map::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 group_cache; + map::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. */ @@ -348,24 +433,14 @@ void dump_inode(const string& path, // Path within snapshot 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); } @@ -438,123 +513,81 @@ void dump_inode(const string& path, // Path within snapshot 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 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::iterator i = includes.begin(); - i != includes.end(); ++i) { - if (path == *i) { - include = true; - } - } - - for (list::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::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(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::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 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; } @@ -571,55 +604,42 @@ void scanfile(const string& path, bool include) 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::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::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(); } } @@ -635,8 +655,10 @@ void usage(const char *program) " --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" @@ -661,6 +683,8 @@ void usage(const char *program) int main(int argc, char *argv[]) { + hash_init(); + string backup_dest = "", backup_script = ""; string localdb_dir = ""; string backup_scheme = ""; @@ -673,18 +697,19 @@ int main(int argc, char *argv[]) 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 + {"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}, @@ -701,46 +726,46 @@ int main(int argc, char *argv[]) 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 + case 6: // --intent snapshot_intent = atof(optarg); if (snapshot_intent <= 0) snapshot_intent = 1; 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"); @@ -763,10 +788,6 @@ int main(int argc, char *argv[]) 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"); @@ -833,7 +854,9 @@ int main(int argc, char *argv[]) metawriter = new MetadataWriter(tss, localdb_dir.c_str(), desc_buf, 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());