Commit 8daa9925 authored by Wayne Davison's avatar Wayne Davison

More improvements.

parent e844be4e
......@@ -11,7 +11,7 @@ Changes since 2.6.3:
- The --stats output will contain file-list time-statistics if both
sides are 2.6.4, or if the local side is 2.6.4 and the files are
being pushed (since the stats come from the sending side).
(Requires protocol 29.)
(Requires protocol 29 for a pull.)
BUG FIXES:
......@@ -19,16 +19,15 @@ Changes since 2.6.3:
was only treating it as a special token in an rsync include/exclude
file).
- The combination of --verbose and --dry-run now mentions changes in
directories and it now includes the full update information that
would be output without --dry-run at higher levels of verbosity.
- The combination of --verbose and --dry-run now mentions the full list
of changes that would be output without --dry-run.
- Avoid a mkdir warning when removing a directory in the destination
that already exists in the --backup-dir.
- An OS that has a binary mode for its files (such as cygwin0 needed
setmode(fd, O_BINARY) called on the temp-file we opened with
mkstemp(). (Fix derived from the cygwin's rsync package.)
mkstemp(). (Fix derived from the Cygwin's 2.6.3 rsync package.)
- Fixed a potential hang when verbosity is high, the client side is
the sender, and the file-list is large.
......@@ -41,8 +40,8 @@ Changes since 2.6.3:
- One place in the code wasn't checking if fork() failed.
- The "ignore nonreadable" daemon parameter no longer affects symlinks
that are being copied, even if they point nowhere.
- The "ignore nonreadable" daemon parameter used to erroneously affect
symlinks that pointed to a non-existent file. This has been fixed.
- If the OS does not have lchown() and its chown() tries to set the
referent of a symlink (as it should), we no longer try to set the
......@@ -67,12 +66,13 @@ Changes since 2.6.3:
exited with an error for large files).
- If a daemon can't open the specified log file (i.e. syslog is not
being used), die without crashing. We also try to output an error
about the failure (which will only be seen if --no-detach was
being used), die without crashing. We also output an error about
the failure on stderr (which will only be seen if --no-detach was
specified).
- A local transfer no longer duplicates all its include/exclude options
by sending the forked process a copy of the list it already has.
(since the forked process already has a copy of the exclude list,
there's no need to send them a set of duplicates).
ENHANCEMENTS:
......@@ -81,13 +81,13 @@ Changes since 2.6.3:
- Added the --delete-during (--del) option which will delete files
from on the receiving side incrementally as each directory in the
transfer is being processed (which makes it more efficient than the
default, before-the-transfer behavior of --delete). Note that the
--del option is implemented as an internally-defined popt alias, so
an rsync daemon that refuses "delete" (which, for safety's sake,
really matches "delete*") will still refuse all delete options. The
default --delete behavior is also explicitly selectable via
--delete-before.
transfer is being processed. This makes it more efficient than the
default, before-the-transfer behavior, which is now available as
--delete-before (this is the default --delete-WHEN option that will
be chosen if --delete or --delete-excluded is specified without a
--delete-WHEN choice). All the --del* options infer --delete, so an
rsync daemon that refuses "delete" will still refuse to allow any
file-deleting options.
- All the --delete-WHEN options are now more memory efficient:
Previously an entire duplicate set of file-list objects was created
......@@ -106,7 +106,7 @@ Changes since 2.6.3:
- The daemon-mode options were separated from the normal rsync options
so that they can't be mixed together. This makes it impossible to
start a daemon that had improper default option values that could
cause problems (such as a hang or an abort) when a client connects.
cause problems (e.g. a hang or an abort) when a client connects.
- The --bwlimit option may now be used in combination with --daemon
to specify both a default value for the daemon side and a value
......@@ -129,11 +129,11 @@ Changes since 2.6.3:
- Added the --list-only option which is mainly a way for the client to
put the server into listing mode without needing to resort to any
option kluges (e.g. the age-old use of "-r --exclude="/*/*" for a
non-recursive listing). This option is used automatically when a
modern rsync speaks to a modern daemon, but may also be specified
manually if you want to force the use of the --list-only option over
a remote-shell connection.
internal option kluges (e.g. the age-old use of "-r --exclude="/*/*"
for a non-recursive listing). This option is used automatically
(behind the scenes) when a modern rsync speaks to a modern daemon,
but may also be specified manually if you want to force the use of
the --list-only option over a remote-shell connection.
- Added the --omit-dir-times (-O) option which will avoid updating the
modified time for directories when --times was specified. This
......@@ -150,7 +150,7 @@ Changes since 2.6.3:
include/exclude processing remained 100% compatible with older
versions. Protocol 29 is needed for full filter-rule support, but
backward-compatible rules work with earlier protocol versions.
(Promoted from the patches dir.)
(Promoted from the patches dir and enhanced.)
- Added the --delay-updates option that puts all updated files into
a temporary directory (by default ".~tmp~", but settable via the
......@@ -164,12 +164,13 @@ Changes since 2.6.3:
setting was added a couple releases ago, but left undocumented.)
- The sender and the generator now double-check the file-list index
they are given, and refuse to operate on a directory index (since
that would indicate that something had gone very wrong).
they are given, and refuse to try to do a file transfer on a
non-file index (since that would indicate that something had gone
very wrong).
- Added the --itemize-changes (-i) option that is a way to output the
list of files that got transferred and/or changed in any way, and how
they changed. The effect is the same as specifying a --log-format of
- Added the --itemize-changes (-i) option, which is a way to output a
more detailed list of what files changed in any way and how they
changed. The effect is the same as specifying a --log-format of
"%i %n%L" (see the rsyncd.conf manpage). Works with --dry-run too.
- Added the --fuzzy option, which attempts to find a basis file for a
......@@ -181,35 +182,37 @@ Changes since 2.6.3:
it needs the new file-sorting order. (Promoted from patches dir
and enhanced.)
SUPPORT FILES:
- Improved the option descriptions in the --help text.
- Added support/atomic-rsync -- a perl script that will transfer some
files using rsync, and then move the updated files into place all at
once at the end of the transfer. Only works when pulling, and uses
--link-dest and a parallel hierarchy of files to effect its update.
SUPPORT FILES:
- Added support/mnt-excl that takes the /proc/mounts file and
translates it into a set of excludes that will exclude all mount
points (even mapped mounts to the same disk). The excludes are made
relative to the specified source dir and properly anchored.
- Added atomic-rsync to the support dir: a perl script that will
transfer some files using rsync, and then move the updated files into
place all at once at the end of the transfer. Only works when
pulling, and uses --link-dest and a parallel hierarchy of files to
effect its update.
- Added support/savetransfer.c -- a C program that can make a copy of
all the data that flows over the wire. This lets you test for data
corruption (by saving the data on both the sending side and the
receiving side) or provides a way to help debug a protocol error.
- Added mnt-excl to the support dir: a perl script that takes the
/proc/mounts file and translates it into a set of excludes that will
exclude all mount points (even mapped mounts to the same disk). The
excludes are made relative to the specified source dir and properly
anchored.
- Added support/rrsync -- my version of Joe Smith's restricted rsync
perl script. This helps to ensure that only certain rsync commands
can be run by an ssh invocation.
- Added savetransfer.c to the support dir: a C program that can make
a copy of all the data that flows over the wire. This lets you test
for data corruption (by saving the data on both the sending side and
the receiving side) or provides a way to help debug a protocol error.
- Improved the option descriptions in the --help text.
- Added rrsync to the support dir: this is my version of Joe Smith's
restricted rsync perl script. This helps to ensure that only certain
rsync commands can be run by an ssh invocation.
INTERNAL:
- Added better checking of the checksum-header values that come over
the socket.
- Merged the various delete-file functions into a single function so
- Merged a variety of file-deleting functions into a single function so
that it is easier to maintain.
- Improved the type of some variables (particularly blocksize vars) for
......@@ -231,12 +234,13 @@ Changes since 2.6.3:
- A 16-bit flag-word is transmitted after every file-list index. This
indicates what is changing between the sender and the receiver. The
generator now transmits an index and a flag-word to indicate when
dirs and symlinks have changed (only outputting local change messages
for older protocols).
dirs and symlinks have changed (resorting to the old-style outputting
of local change-messages for older protocols).
- If --inplace is specified, the generator sends an extra byte after
the flag-word indicating what kind of basis file is being used for
the transfer (see the FNAMECMP_* defines).
the transfer (see the FNAMECMP_* defines). This information is used
to optimize the transfer when the basis file is not the destination.
- The sending of exclude names is done using filter-rule syntax. This
means that all names have a prefixed rule indicator, even excludes
......@@ -245,15 +249,16 @@ Changes since 2.6.3:
filter rules so it is positioned correctly (unlike in some older
transfer scenarios).
- Rsync sorts the filename list in a different way: it sorts the
subdir names after the non-subdir names for each dir's contents, and
it always puts a dir's contents immediately after the dir's name in
the list. (Previously an item named "foo.txt" would sort in between
- Rsync sorts the filename list in a different way: it sorts the subdir
names after the non-subdir names for each dir's contents, and it
always puts a dir's contents immediately after the dir's name in the
list. (Previously an item named "foo.txt" would sort in between
directory "foo/" and "foo/bar".)
- When talking to a protocol 29 rsync daemon, a list-only request
is able to note this before the options are sent over the wire, and
the new --list-only option is enabled.
the new --list-only option is encluded in the options sent over the
socket.
- When the --stats bytes are sent over the wire (or stored in a batch),
they now include two elapsed-time values: one for how long it took to
......@@ -261,11 +266,13 @@ Changes since 2.6.3:
wire (each expressed in thousandths of a second).
- When --delete-excluded is specified with some filter rules (AKA
excludes), a client sender will still initiate a send of the filter
rules to the receiver, but it only includes those rules that are
receiver-specific. Older protocols used to omit the sending of
excludes in this situation (since there were no receiver-specific
rules that survived --delete-excluded back then).
excludes), a client sender will now initiate a send of the filter
rules to the receiver (older protocols used to omit the sending of
excludes in this situation since there were no receiver-specific
rules that survived --delete-excluded back then). Note that, as with
all the filter-list sending, only items that are significant to the
other side will actually be sent over the wire, so the filter-rule
list is often empty in this scenario.
- A protocol-29 batch file includes a bit for the setting of the --dirs
option. Also, the shell script created by --write-batch will use the
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment