| 1 | This patch adds the --preallocate option that asks rsync to preallocate the |
| 2 | copied files. This slows down the copy, but should reduce fragmentation on |
| 3 | systems that need that. |
| 4 | |
| 5 | To use this patch, run these commands for a successful build: |
| 6 | |
| 7 | patch -p1 <patches/preallocate.diff |
| 8 | ./prepare-source |
| 9 | ./configure |
| 10 | make |
| 11 | |
| 12 | diff --git a/configure.in b/configure.in |
| 13 | --- a/configure.in |
| 14 | +++ b/configure.in |
| 15 | @@ -552,7 +552,7 @@ AC_CHECK_FUNCS(waitpid wait4 getcwd strdup chown chmod lchmod mknod mkfifo \ |
| 16 | strlcat strlcpy strtol mallinfo getgroups setgroups geteuid getegid \ |
| 17 | setlocale setmode open64 lseek64 mkstemp64 mtrace va_copy __va_copy \ |
| 18 | strerror putenv iconv_open locale_charset nl_langinfo getxattr \ |
| 19 | - extattr_get_link sigaction sigprocmask setattrlist) |
| 20 | + extattr_get_link sigaction sigprocmask setattrlist posix_fallocate) |
| 21 | |
| 22 | dnl cygwin iconv.h defines iconv_open as libiconv_open |
| 23 | if test x"$ac_cv_func_iconv_open" != x"yes"; then |
| 24 | diff --git a/options.c b/options.c |
| 25 | --- a/options.c |
| 26 | +++ b/options.c |
| 27 | @@ -72,6 +72,7 @@ int remove_source_files = 0; |
| 28 | int one_file_system = 0; |
| 29 | int protocol_version = PROTOCOL_VERSION; |
| 30 | int sparse_files = 0; |
| 31 | +int preallocate_files = 0; |
| 32 | int do_compression = 0; |
| 33 | int def_compress_level = Z_DEFAULT_COMPRESSION; |
| 34 | int am_root = 0; /* 0 = normal, 1 = root, 2 = --super, -1 = --fake-super */ |
| 35 | @@ -223,6 +224,7 @@ static void print_rsync_version(enum logcode f) |
| 36 | char const *links = "no "; |
| 37 | char const *iconv = "no "; |
| 38 | char const *ipv6 = "no "; |
| 39 | + char const *preallocation = "no "; |
| 40 | STRUCT_STAT *dumstat; |
| 41 | |
| 42 | #if SUBPROTOCOL_VERSION != 0 |
| 43 | @@ -252,6 +254,9 @@ static void print_rsync_version(enum logcode f) |
| 44 | #ifdef ICONV_OPTION |
| 45 | iconv = ""; |
| 46 | #endif |
| 47 | +#ifdef SUPPORT_PREALLOCATION |
| 48 | + preallocation = ""; |
| 49 | +#endif |
| 50 | |
| 51 | rprintf(f, "%s version %s protocol version %d%s\n", |
| 52 | RSYNC_NAME, RSYNC_VERSION, PROTOCOL_VERSION, subprotocol); |
| 53 | @@ -265,8 +270,8 @@ static void print_rsync_version(enum logcode f) |
| 54 | (int)(sizeof (int64) * 8)); |
| 55 | rprintf(f, " %ssocketpairs, %shardlinks, %ssymlinks, %sIPv6, batchfiles, %sinplace,\n", |
| 56 | got_socketpair, hardlinks, links, ipv6, have_inplace); |
| 57 | - rprintf(f, " %sappend, %sACLs, %sxattrs, %siconv\n", |
| 58 | - have_inplace, acls, xattrs, iconv); |
| 59 | + rprintf(f, " %sappend, %sACLs, %sxattrs, %siconv, %spreallocation\n", |
| 60 | + have_inplace, acls, xattrs, iconv, preallocation); |
| 61 | |
| 62 | #ifdef MAINTAINER_MODE |
| 63 | rprintf(f, "Panic Action: \"%s\"\n", get_panic_action()); |
| 64 | @@ -353,6 +358,9 @@ void usage(enum logcode F) |
| 65 | rprintf(F," --fake-super store/recover privileged attrs using xattrs\n"); |
| 66 | #endif |
| 67 | rprintf(F," -S, --sparse handle sparse files efficiently\n"); |
| 68 | +#ifdef SUPPORT_PREALLOCATION |
| 69 | + rprintf(F," --preallocate posix_fallocate dest files before writing them\n"); |
| 70 | +#endif |
| 71 | rprintf(F," -n, --dry-run perform a trial run with no changes made\n"); |
| 72 | rprintf(F," -W, --whole-file copy files whole (without delta-xfer algorithm)\n"); |
| 73 | rprintf(F," -x, --one-file-system don't cross filesystem boundaries\n"); |
| 74 | @@ -531,6 +539,7 @@ static struct poptOption long_options[] = { |
| 75 | {"max-size", 0, POPT_ARG_STRING, &max_size_arg, OPT_MAX_SIZE, 0, 0 }, |
| 76 | {"min-size", 0, POPT_ARG_STRING, &min_size_arg, OPT_MIN_SIZE, 0, 0 }, |
| 77 | {"sparse", 'S', POPT_ARG_NONE, &sparse_files, 0, 0, 0 }, |
| 78 | + {"preallocate", 0, POPT_ARG_NONE, &preallocate_files, 0, 0, 0}, |
| 79 | {"inplace", 0, POPT_ARG_NONE, &inplace, 0, 0, 0 }, |
| 80 | {"append", 0, POPT_ARG_NONE, 0, OPT_APPEND, 0, 0 }, |
| 81 | {"append-verify", 0, POPT_ARG_VAL, &append_mode, 2, 0, 0 }, |
| 82 | @@ -1289,6 +1298,15 @@ int parse_arguments(int *argc_p, const char ***argv_p, int frommain) |
| 83 | } |
| 84 | #endif |
| 85 | |
| 86 | +#ifndef SUPPORT_PREALLOCATION |
| 87 | + if (preallocate_files && !am_sender) { |
| 88 | + snprintf(err_buf, sizeof err_buf, |
| 89 | + "preallocation is not supported on this %s\n", |
| 90 | + am_server ? "server" : "client"); |
| 91 | + return 0; |
| 92 | + } |
| 93 | +#endif |
| 94 | + |
| 95 | if (write_batch && read_batch) { |
| 96 | snprintf(err_buf, sizeof err_buf, |
| 97 | "--write-batch and --read-batch can not be used together\n"); |
| 98 | @@ -1999,6 +2017,9 @@ void server_options(char **args, int *argc_p) |
| 99 | else if (remove_source_files) |
| 100 | args[ac++] = "--remove-sent-files"; |
| 101 | |
| 102 | + if (preallocate_files && am_sender) |
| 103 | + args[ac++] = "--preallocate"; |
| 104 | + |
| 105 | *argc_p = ac; |
| 106 | return; |
| 107 | |
| 108 | diff --git a/receiver.c b/receiver.c |
| 109 | --- a/receiver.c |
| 110 | +++ b/receiver.c |
| 111 | @@ -45,6 +45,7 @@ extern int cleanup_got_literal; |
| 112 | extern int remove_source_files; |
| 113 | extern int append_mode; |
| 114 | extern int sparse_files; |
| 115 | +extern int preallocate_files; |
| 116 | extern int keep_partial; |
| 117 | extern int checksum_seed; |
| 118 | extern int inplace; |
| 119 | @@ -175,6 +176,19 @@ static int receive_data(int f_in, char *fname_r, int fd_r, OFF_T size_r, |
| 120 | int32 i; |
| 121 | char *map = NULL; |
| 122 | |
| 123 | +#ifdef SUPPORT_PREALLOCATION |
| 124 | + OFF_T preallocated_len = 0; |
| 125 | + |
| 126 | + if (preallocate_files && fd != -1 && total_size > 0) { |
| 127 | + /* Preallocate enough space for file's eventual length if |
| 128 | + * possible; seems to reduce fragmentation on Windows. */ |
| 129 | + if (posix_fallocate(fd, 0, total_size) == 0) |
| 130 | + preallocated_len = total_size; |
| 131 | + else |
| 132 | + rsyserr(FINFO, errno, "preallocate %s", full_fname(fname)); |
| 133 | + } |
| 134 | +#endif |
| 135 | + |
| 136 | read_sum_head(f_in, &sum); |
| 137 | |
| 138 | if (fd_r >= 0 && size_r > 0) { |
| 139 | @@ -284,8 +298,18 @@ static int receive_data(int f_in, char *fname_r, int fd_r, OFF_T size_r, |
| 140 | goto report_write_error; |
| 141 | |
| 142 | #ifdef HAVE_FTRUNCATE |
| 143 | - if (inplace && fd != -1) |
| 144 | - ftruncate(fd, offset); |
| 145 | + /* inplace: New data could be shorter than old data. |
| 146 | + * preallocate_files: total_size could have been an overestimate. |
| 147 | + * Cut off any extra preallocated zeros from dest file. */ |
| 148 | + if ((inplace |
| 149 | +#ifdef SUPPORT_PREALLOCATION |
| 150 | + || preallocated_len > offset |
| 151 | +#endif |
| 152 | + ) && fd != -1) |
| 153 | + if (ftruncate(fd, offset) < 0) |
| 154 | + /* If we fail to truncate, the dest file may be wrong, so we |
| 155 | + * must trigger the "partial transfer" error. */ |
| 156 | + rsyserr(FERROR_XFER, errno, "ftruncate %s", full_fname(fname)); |
| 157 | #endif |
| 158 | |
| 159 | if (do_progress) |
| 160 | diff --git a/rsync.h b/rsync.h |
| 161 | --- a/rsync.h |
| 162 | +++ b/rsync.h |
| 163 | @@ -600,6 +600,10 @@ struct ht_int64_node { |
| 164 | #define ACLS_NEED_MASK 1 |
| 165 | #endif |
| 166 | |
| 167 | +#if defined HAVE_FTRUNCATE && defined HAVE_POSIX_FALLOCATE |
| 168 | +#define SUPPORT_PREALLOCATION 1 |
| 169 | +#endif |
| 170 | + |
| 171 | union file_extras { |
| 172 | int32 num; |
| 173 | uint32 unum; |
| 174 | diff --git a/rsync.yo b/rsync.yo |
| 175 | --- a/rsync.yo |
| 176 | +++ b/rsync.yo |
| 177 | @@ -352,6 +352,7 @@ to the detailed description below for a complete description. verb( |
| 178 | --super receiver attempts super-user activities |
| 179 | --fake-super store/recover privileged attrs using xattrs |
| 180 | -S, --sparse handle sparse files efficiently |
| 181 | + --preallocate posix_fallocate dest files before writing |
| 182 | -n, --dry-run perform a trial run with no changes made |
| 183 | -W, --whole-file copy files whole (w/o delta-xfer algorithm) |
| 184 | -x, --one-file-system don't cross filesystem boundaries |
| 185 | @@ -1023,6 +1024,19 @@ NOTE: Don't use this option when the destination is a Solaris "tmpfs" |
| 186 | filesystem. It doesn't seem to handle seeks over null regions |
| 187 | correctly and ends up corrupting the files. |
| 188 | |
| 189 | +dit(bf(--preallocate)) This tells the receiver to allocate each destination |
| 190 | +file to its eventual size using bf(posix_fallocate)(3) before writing data |
| 191 | +to the file. If the receiver is remote, this nonstandard option only works |
| 192 | +if the receiver also has the preallocation patch. Furthermore, this option |
| 193 | +only works if the receiver found the bf(posix_fallocate)(3) call at |
| 194 | +configure time. |
| 195 | + |
| 196 | +Without this option on MS Windows, very large destination files tend to be |
| 197 | +broken into thousands of fragments; advising Windows ahead of time of the |
| 198 | +eventual file size using this option usually reduces the number of |
| 199 | +fragments to one. However, on Linux, this option appears to just waste |
| 200 | +disk I/O. |
| 201 | + |
| 202 | dit(bf(-n, --dry-run)) This makes rsync perform a trial run that doesn't |
| 203 | make any changes (and produces mostly the same output as a real run). It |
| 204 | is most commonly used in combination with the bf(-v, --verbose) and/or |
| 205 | diff --git a/t_stub.c b/t_stub.c |
| 206 | --- a/t_stub.c |
| 207 | +++ b/t_stub.c |
| 208 | @@ -22,6 +22,7 @@ |
| 209 | #include "rsync.h" |
| 210 | |
| 211 | int modify_window = 0; |
| 212 | +int preallocate_files = 0; |
| 213 | int module_id = -1; |
| 214 | int relative_paths = 0; |
| 215 | int human_readable = 0; |
| 216 | diff --git a/util.c b/util.c |
| 217 | --- a/util.c |
| 218 | +++ b/util.c |
| 219 | @@ -25,6 +25,7 @@ |
| 220 | |
| 221 | extern int verbose; |
| 222 | extern int dry_run; |
| 223 | +extern int preallocate_files; |
| 224 | extern int module_id; |
| 225 | extern int modify_window; |
| 226 | extern int relative_paths; |
| 227 | @@ -273,6 +274,10 @@ int copy_file(const char *source, const char *dest, int ofd, |
| 228 | int ifd; |
| 229 | char buf[1024 * 8]; |
| 230 | int len; /* Number of bytes read into `buf'. */ |
| 231 | +#ifdef SUPPORT_PREALLOCATION |
| 232 | + OFF_T preallocated_len = 0; |
| 233 | + OFF_T offset = 0; |
| 234 | +#endif |
| 235 | |
| 236 | if ((ifd = do_open(source, O_RDONLY, 0)) < 0) { |
| 237 | rsyserr(FERROR_XFER, errno, "open %s", full_fname(source)); |
| 238 | @@ -294,7 +299,27 @@ int copy_file(const char *source, const char *dest, int ofd, |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | +#ifdef SUPPORT_PREALLOCATION |
| 243 | + if (preallocate_files) { |
| 244 | + /* Preallocate enough space for file's eventual length if |
| 245 | + * possible; seems to reduce fragmentation on Windows. */ |
| 246 | + STRUCT_STAT srcst; |
| 247 | + if (do_fstat(ifd, &srcst) == 0) { |
| 248 | + if (srcst.st_size > 0) { |
| 249 | + if (posix_fallocate(ofd, 0, srcst.st_size) == 0) |
| 250 | + preallocated_len = srcst.st_size; |
| 251 | + else |
| 252 | + rsyserr(FINFO, errno, "posix_fallocate %s", full_fname(dest)); |
| 253 | + } |
| 254 | + } else |
| 255 | + rsyserr(FINFO, errno, "fstat %s", full_fname(source)); |
| 256 | + } |
| 257 | +#endif |
| 258 | + |
| 259 | while ((len = safe_read(ifd, buf, sizeof buf)) > 0) { |
| 260 | +#ifdef SUPPORT_PREALLOCATION |
| 261 | + offset += len; |
| 262 | +#endif |
| 263 | if (full_write(ofd, buf, len) < 0) { |
| 264 | rsyserr(FERROR_XFER, errno, "write %s", full_fname(dest)); |
| 265 | close(ifd); |
| 266 | @@ -315,6 +340,16 @@ int copy_file(const char *source, const char *dest, int ofd, |
| 267 | full_fname(source)); |
| 268 | } |
| 269 | |
| 270 | +#ifdef SUPPORT_PREALLOCATION |
| 271 | + /* Source file might have shrunk since we fstatted it. |
| 272 | + * Cut off any extra preallocated zeros from dest file. */ |
| 273 | + if (preallocated_len > offset) |
| 274 | + if (ftruncate(ofd, offset) < 0) |
| 275 | + /* If we fail to truncate, the dest file may be wrong, so we |
| 276 | + * must trigger the "partial transfer" error. */ |
| 277 | + rsyserr(FERROR_XFER, errno, "ftruncate %s", full_fname(dest)); |
| 278 | +#endif |
| 279 | + |
| 280 | if (close(ofd) < 0) { |
| 281 | rsyserr(FERROR_XFER, errno, "close failed on %s", |
| 282 | full_fname(dest)); |