From 961f96833ae8ad26ad29eb7570058b9fcbb081a1 Mon Sep 17 00:00:00 2001
From: Axel Burri <axel@tty0.ch>
Date: Thu, 25 Aug 2016 12:45:13 +0200
Subject: [PATCH] btrbk: use "dd" for raw target output, with configurable
 block size (raw_target_block_size)

We use "dd" instead of shell redirections, as it is common to have
special filesystems (like NFS, SMB, FUSE) mounted on the raw target
path. By using "dd" we make sure to write in reasonably large blocks
(default=128K), which is not always the case when using redirections
(e.g. "gpg > outfile" writes in 8K blocks).

Another approach would be to always pipe through "cat", which uses
st_blksize from fstat(2) (with a minimum of 128K) to determine the
block size.
---
 ChangeLog        |  1 +
 btrbk            | 14 +++++++++++++-
 doc/btrbk.conf.5 |  2 ++
 3 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/ChangeLog b/ChangeLog
index 23b2e35..26015b9 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -8,6 +8,7 @@ btrbk-current
   * Perform extra metadata check on target subvolume after "btrfs
     receive" (adds an additional call to "btrfs subvolume show").
   * Bugfix: Replace "realpath" with "readlink" in ssh_filter_btrbk.sh
+  * Add "raw_target_block_size" configuration option (close #105).
 
 btrbk-0.23.3
 
diff --git a/btrbk b/btrbk
index b481305..bd32a0d 100755
--- a/btrbk
+++ b/btrbk
@@ -118,6 +118,7 @@ my %config_options = (
   raw_target_compress_level   => { default => "default", accept => [ "default" ], accept_numeric => 1 },
   raw_target_compress_threads => { default => "default", accept => [ "default" ], accept_numeric => 1 },
   raw_target_encrypt          => { default => undef,     accept => [ "no", "gpg" ] },
+  raw_target_block_size       => { default => "128K",    accept_regexp => qr/^[0-9]+(kB|k|K|KiB|MB|M|MiB)?$/ },
   gpg_keyring                 => { default => undef,     accept_file => { absolute => 1 } },
   gpg_recipient               => { default => undef,     accept_regexp => qr/^[0-9a-zA-Z_@\+\-\.]+$/ },
 
@@ -1279,7 +1280,18 @@ sub btrfs_send_to_file($$$$;@)
     };
   }
   push @cmd_pipe, {
-    redirect_to_file => { unsafe => "${target_path}/${target_filename}.part" },
+    # NOTE: We use "dd" instead of shell redirections here, as it is
+    # common to have special filesystems (like NFS, SMB, FUSE) mounted
+    # on $target_path. By using "dd" we make sure to write in
+    # reasonably large blocks (default=128K), which is not always the
+    # case when using redirections (e.g. "gpg > outfile" writes in 8K
+    # blocks).
+    # Another approach would be to always pipe through "cat", which
+    # uses st_blksize from fstat(2) (with a minimum of 128K) to
+    # determine the block size.
+    cmd => [ 'dd', 'status=none', 'bs=' . config_key($target, "raw_target_block_size"), "of=${target_path}/${target_filename}.part" ],
+    check_unsafe => [ { unsafe => "${target_path}/${target_filename}.part" } ],
+    #redirect_to_file => { unsafe => "${target_path}/${target_filename}.part" },  # alternative (use shell redirection), less overhead on local filesystems (barely measurable):
     rsh => vinfo_rsh($target, disable_compression => $opts{compress} || config_compress_hash($target, "stream_compress")),
     rsh_compress_in => $opts{compress} || config_compress_hash($target, "stream_compress"),
     compressed_ok => ($opts{compress} ? 1 : 0),
diff --git a/doc/btrbk.conf.5 b/doc/btrbk.conf.5
index d0693de..59716e7 100644
--- a/doc/btrbk.conf.5
+++ b/doc/btrbk.conf.5
@@ -380,6 +380,8 @@ raw_target_compress_threads  default|<number>
 .PP
 raw_target_encrypt  gpg|no
 .PP
+raw_target_block_size  <number>  (defaults to 128K)
+.PP
 gpg_keyring  <file>
 .PP
 gpg_recipient  <name>