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.
5 To use this patch, run these commands for a successful build:
7 patch -p1 <patches/preallocate.diff
12 diff --git a/compat.c b/compat.c
13 index 6e00072..c9590cc 100644
16 @@ -32,6 +32,7 @@ extern int inplace;
19 extern int allow_inc_recurse;
20 +extern int preallocate_files;
21 extern int append_mode;
22 extern int fuzzy_basis;
23 extern int read_batch;
24 @@ -186,6 +187,15 @@ void setup_protocol(int f_out,int f_in)
28 +#ifndef SUPPORT_PREALLOCATION
29 + if (preallocate_files && !am_sender) {
31 + "preallocation is not supported on this %s\n",
32 + am_server ? "server" : "client");
33 + exit_cleanup(RERR_SYNTAX);
37 if (protocol_version < 30) {
40 diff --git a/configure.in b/configure.in
41 index bc7d4a7..61f9b05 100644
44 @@ -553,13 +553,40 @@ AC_CHECK_FUNCS(waitpid wait4 getcwd strdup chown chmod lchmod mknod mkfifo \
45 setlocale setmode open64 lseek64 mkstemp64 mtrace va_copy __va_copy \
46 strerror putenv iconv_open locale_charset nl_langinfo getxattr \
47 extattr_get_link sigaction sigprocmask setattrlist getgrouplist \
49 + initgroups fallocate posix_fallocate)
51 dnl cygwin iconv.h defines iconv_open as libiconv_open
52 if test x"$ac_cv_func_iconv_open" != x"yes"; then
53 AC_CHECK_FUNC(libiconv_open, [ac_cv_func_iconv_open=yes; AC_DEFINE(HAVE_ICONV_OPEN, 1)])
56 +dnl Preallocation stuff (also fallocate, posix_fallocate function tests above):
58 +AC_CACHE_CHECK([for SYS_fallocate],rsync_cv_have_sys_fallocate,[
59 +AC_TRY_COMPILE([#include <sys/syscall.h>
60 +#include <sys/types.h>],
61 +[syscall(SYS_fallocate, 0, 0, (loff_t) 0, (loff_t) 0);],
62 +rsync_cv_have_sys_fallocate=yes,rsync_cv_have_sys_fallocate=no)])
63 +if test x"$rsync_cv_have_sys_fallocate" = x"yes"; then
64 + AC_DEFINE(HAVE_SYS_FALLOCATE, 1, [Define to 1 if you have the SYS_fallocate syscall number])
67 +if test x"$ac_cv_func_posix_fallocate" = x"yes"; then
68 + AC_MSG_CHECKING([whether posix_fallocate is efficient])
72 + AC_DEFINE(HAVE_EFFICIENT_POSIX_FALLOCATE, 1,
73 + [Define if posix_fallocate is efficient (Cygwin)])
81 +dnl End of preallocation stuff
83 AC_CHECK_FUNCS(getpgrp tcgetpgrp)
84 if test $ac_cv_func_getpgrp = yes; then
86 diff --git a/options.c b/options.c
87 index e7c6c61..6f848f1 100644
90 @@ -73,6 +73,7 @@ int remove_source_files = 0;
91 int one_file_system = 0;
92 int protocol_version = PROTOCOL_VERSION;
94 +int preallocate_files = 0;
95 int do_compression = 0;
96 int def_compress_level = Z_DEFAULT_COMPRESSION;
97 int am_root = 0; /* 0 = normal, 1 = root, 2 = --super, -1 = --fake-super */
98 @@ -566,6 +567,7 @@ static void print_rsync_version(enum logcode f)
99 char const *links = "no ";
100 char const *iconv = "no ";
101 char const *ipv6 = "no ";
102 + char const *preallocation = "no ";
103 STRUCT_STAT *dumstat;
105 #if SUBPROTOCOL_VERSION != 0
106 @@ -599,6 +601,9 @@ static void print_rsync_version(enum logcode f)
107 #if defined HAVE_LUTIMES && defined HAVE_UTIMES
110 +#ifdef SUPPORT_PREALLOCATION
111 + preallocation = "";
114 rprintf(f, "%s version %s protocol version %d%s\n",
115 RSYNC_NAME, RSYNC_VERSION, PROTOCOL_VERSION, subprotocol);
116 @@ -612,8 +617,8 @@ static void print_rsync_version(enum logcode f)
117 (int)(sizeof (int64) * 8));
118 rprintf(f, " %ssocketpairs, %shardlinks, %ssymlinks, %sIPv6, batchfiles, %sinplace,\n",
119 got_socketpair, hardlinks, links, ipv6, have_inplace);
120 - rprintf(f, " %sappend, %sACLs, %sxattrs, %siconv, %ssymtimes\n",
121 - have_inplace, acls, xattrs, iconv, symtimes);
122 + rprintf(f, " %sappend, %sACLs, %sxattrs, %siconv, %ssymtimes, %spreallocation\n",
123 + have_inplace, acls, xattrs, iconv, symtimes, preallocation);
125 #ifdef MAINTAINER_MODE
126 rprintf(f, "Panic Action: \"%s\"\n", get_panic_action());
127 @@ -703,6 +708,9 @@ void usage(enum logcode F)
128 rprintf(F," --fake-super store/recover privileged attrs using xattrs\n");
130 rprintf(F," -S, --sparse handle sparse files efficiently\n");
131 +#ifdef SUPPORT_PREALLOCATION
132 + rprintf(F," --preallocate allocate dest files before writing them\n");
134 rprintf(F," -n, --dry-run perform a trial run with no changes made\n");
135 rprintf(F," -W, --whole-file copy files whole (without delta-xfer algorithm)\n");
136 rprintf(F," -x, --one-file-system don't cross filesystem boundaries\n");
137 @@ -899,6 +907,7 @@ static struct poptOption long_options[] = {
138 {"sparse", 'S', POPT_ARG_VAL, &sparse_files, 1, 0, 0 },
139 {"no-sparse", 0, POPT_ARG_VAL, &sparse_files, 0, 0, 0 },
140 {"no-S", 0, POPT_ARG_VAL, &sparse_files, 0, 0, 0 },
141 + {"preallocate", 0, POPT_ARG_NONE, &preallocate_files, 0, 0, 0},
142 {"inplace", 0, POPT_ARG_VAL, &inplace, 1, 0, 0 },
143 {"no-inplace", 0, POPT_ARG_VAL, &inplace, 0, 0, 0 },
144 {"append", 0, POPT_ARG_NONE, 0, OPT_APPEND, 0, 0 },
145 @@ -2609,6 +2618,9 @@ void server_options(char **args, int *argc_p)
146 else if (remove_source_files)
147 args[ac++] = "--remove-sent-files";
149 + if (preallocate_files && am_sender)
150 + args[ac++] = "--preallocate";
152 if (ac > MAX_SERVER_ARGS) { /* Not possible... */
153 rprintf(FERROR, "argc overflow in server_options().\n");
154 exit_cleanup(RERR_MALLOC);
155 diff --git a/receiver.c b/receiver.c
156 index 4325e30..739a0ba 100644
159 @@ -44,6 +44,7 @@ extern int cleanup_got_literal;
160 extern int remove_source_files;
161 extern int append_mode;
162 extern int sparse_files;
163 +extern int preallocate_files;
164 extern int keep_partial;
165 extern int checksum_len;
166 extern int checksum_seed;
167 @@ -175,6 +176,18 @@ static int receive_data(int f_in, char *fname_r, int fd_r, OFF_T size_r,
171 +#ifdef SUPPORT_PREALLOCATION
172 + OFF_T preallocated_len = 0;
174 + if (preallocate_files && fd != -1 && total_size > 0) {
175 + /* Preallocate enough space for file's eventual length if
176 + * possible; seems to reduce fragmentation on Windows. */
177 + if (do_fallocate(fd, 0, total_size) == 0)
178 + preallocated_len = total_size;
180 + rsyserr(FWARNING, errno, "do_fallocate %s", full_fname(fname));
184 read_sum_head(f_in, &sum);
186 @@ -285,8 +298,14 @@ static int receive_data(int f_in, char *fname_r, int fd_r, OFF_T size_r,
187 goto report_write_error;
189 #ifdef HAVE_FTRUNCATE
190 - if (inplace && fd != -1
191 - && ftruncate(fd, offset) < 0) {
192 + /* inplace: New data could be shorter than old data.
193 + * preallocate_files: total_size could have been an overestimate.
194 + * Cut off any extra preallocated zeros from dest file. */
196 +#ifdef SUPPORT_PREALLOCATION
197 + || preallocated_len > offset
199 + ) && fd != -1 && ftruncate(fd, offset) < 0) {
200 rsyserr(FERROR_XFER, errno, "ftruncate failed on %s",
203 diff --git a/rsync.h b/rsync.h
204 index be7cf8a..0ad3075 100644
207 @@ -634,6 +634,13 @@ struct ht_int64_node {
208 #define ACLS_NEED_MASK 1
211 +#if defined HAVE_FTRUNCATE \
212 + && (defined HAVE_FALLOCATE \
213 + || defined HAVE_SYS_FALLOCATE \
214 + || defined HAVE_EFFICIENT_POSIX_FALLOCATE)
215 +#define SUPPORT_PREALLOCATION 1
221 diff --git a/rsync.yo b/rsync.yo
222 index 941f7a5..36a2077 100644
225 @@ -359,6 +359,7 @@ to the detailed description below for a complete description. verb(
226 --super receiver attempts super-user activities
227 --fake-super store/recover privileged attrs using xattrs
228 -S, --sparse handle sparse files efficiently
229 + --preallocate allocate dest files before writing
230 -n, --dry-run perform a trial run with no changes made
231 -W, --whole-file copy files whole (w/o delta-xfer algorithm)
232 -x, --one-file-system don't cross filesystem boundaries
233 @@ -1120,6 +1121,18 @@ NOTE: Don't use this option when the destination is a Solaris "tmpfs"
234 filesystem. It doesn't seem to handle seeks over null regions
235 correctly and ends up corrupting the files.
237 +dit(bf(--preallocate)) This tells the receiver to allocate each destination
238 +file to its eventual size before writing data to the file. Rsync will only use
239 +the real filesystem-level preallocation support provided by bf(fallocate)(2) or
240 +Cygwin's bf(posix_fallocate)(3), not the slow glibc implementation that writes
241 +a zero byte into each block. If the receiver is remote, this nonstandard
242 +option only works if the receiver also has the preallocation patch.
244 +Without this option on MS Windows, very large destination files tend to be
245 +broken into thousands of fragments; advising Windows ahead of time of the
246 +eventual file size using this option usually reduces the number of
247 +fragments to one. The usefulness of this option on Linux is yet to be tested.
249 dit(bf(-n, --dry-run)) This makes rsync perform a trial run that doesn't
250 make any changes (and produces mostly the same output as a real run). It
251 is most commonly used in combination with the bf(-v, --verbose) and/or
252 diff --git a/syscall.c b/syscall.c
253 index cfabc3e..81fb957 100644
257 #include <sys/attr.h>
260 +#if defined HAVE_SYS_FALLOCATE && !defined HAVE_FALLOCATE
261 +#include <sys/syscall.h>
266 extern int read_only;
267 @@ -282,3 +286,21 @@ OFF_T do_lseek(int fd, OFF_T offset, int whence)
268 return lseek(fd, offset, whence);
272 +#ifdef SUPPORT_PREALLOCATION
273 +int do_fallocate(int fd, OFF_T offset, OFF_T length)
275 + RETURN_ERROR_IF(dry_run, 0);
276 + RETURN_ERROR_IF_RO_OR_LO;
277 + /* TODO: Use FALLOC_FL_KEEP_SIZE to avoid the need to truncate. */
278 +#if defined HAVE_FALLOCATE
279 + return fallocate(fd, 0, offset, length);
280 +#elif defined HAVE_SYS_FALLOCATE
281 + return syscall(SYS_fallocate, fd, 0, (loff_t) offset, (loff_t) length);
282 +#elif defined HAVE_EFFICIENT_POSIX_FALLOCATE
283 + return posix_fallocate(fd, offset, length);
285 +#error coding error in SUPPORT_PREALLOCATION
289 diff --git a/t_stub.c b/t_stub.c
290 index 02cfa69..52a7f02 100644
296 int modify_window = 0;
297 +int preallocate_files = 0;
299 int relative_paths = 0;
300 int module_dirlen = 0;
301 diff --git a/util.c b/util.c
302 index 0cafed6..20dd0d3 100644
309 +extern int preallocate_files;
310 extern int module_id;
311 extern int modify_window;
312 extern int relative_paths;
313 @@ -276,6 +277,10 @@ int copy_file(const char *source, const char *dest, int ofd,
316 int len; /* Number of bytes read into `buf'. */
317 +#ifdef SUPPORT_PREALLOCATION
318 + OFF_T preallocated_len = 0;
322 if ((ifd = do_open(source, O_RDONLY, 0)) < 0) {
323 int save_errno = errno;
324 @@ -309,7 +314,27 @@ int copy_file(const char *source, const char *dest, int ofd,
328 +#ifdef SUPPORT_PREALLOCATION
329 + if (preallocate_files) {
330 + /* Preallocate enough space for file's eventual length if
331 + * possible; seems to reduce fragmentation on Windows. */
333 + if (do_fstat(ifd, &srcst) == 0) {
334 + if (srcst.st_size > 0) {
335 + if (do_fallocate(ofd, 0, srcst.st_size) == 0)
336 + preallocated_len = srcst.st_size;
338 + rsyserr(FWARNING, errno, "do_fallocate %s", full_fname(dest));
341 + rsyserr(FWARNING, errno, "fstat %s", full_fname(source));
345 while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
346 +#ifdef SUPPORT_PREALLOCATION
349 if (full_write(ofd, buf, len) < 0) {
350 int save_errno = errno;
351 rsyserr(FERROR_XFER, errno, "write %s", full_fname(dest));
352 @@ -334,6 +359,16 @@ int copy_file(const char *source, const char *dest, int ofd,
356 +#ifdef SUPPORT_PREALLOCATION
357 + /* Source file might have shrunk since we fstatted it.
358 + * Cut off any extra preallocated zeros from dest file. */
359 + if (preallocated_len > offset)
360 + if (ftruncate(ofd, offset) < 0)
361 + /* If we fail to truncate, the dest file may be wrong, so we
362 + * must trigger the "partial transfer" error. */
363 + rsyserr(FERROR_XFER, errno, "ftruncate %s", full_fname(dest));
366 if (close(ofd) < 0) {
367 int save_errno = errno;
368 rsyserr(FERROR_XFER, errno, "close failed on %s",