1nbdcopy(1) LIBNBD nbdcopy(1)
2
6 nbdcopy - copy to and from an NBD server
7
9 nbdcopy [--allocated] [-C N|--connections=N]
10 [--destination-is-zero|--target-is-zero] [--flush]
11 [--no-extents] [-p|--progress|--progress=FD]
12 [--queue-size=N] [--request-size=N] [-R N|--requests=N]
13 [-S N|--sparse=N] [--synchronous] [-T N|--threads=N]
14 [-v|--verbose]
15 SOURCE DESTINATION
16 SOURCE, DESTINATION := - | FILE | DEVICE | NBD-URI | [ CMD ARGS ... ]
18 DESTINATION += null:
19 nbdcopy --help
21 nbdcopy -V|--version
23
25 nbdcopy nbd://example.com local.img
26 This copies everything from the NBD server at "example.com" to a local
27 file called local.img.
28 nbdcopy nbd://example.com - | file -
30 This streams the first part of the disk on the NBD server at
31 "example.com" into the file(1) command. Note here that "-" means to
32 stream to stdout (and therefore into the pipe to the file command).
33 nbdinfo(1) is another way to detect the content on an NBD server.
35 nbdcopy -p /dev/sdX "nbd+unix:///?socket=/tmp/unixsock"
37 Copy the full local hard disk "/dev/sdX" to the NBD server listening on
38 the Unix domain socket /tmp/unixsock. Because of the -p option this
39 will print a progress bar.
40 nbdcopy nbd://server1 nbd://server2
42 Copy a full disk from one NBD server to another.
43 nbdcopy -- [ qemu-nbd -r -f qcow2 https://example.com/disk.qcow2 ] -
45 Run qemu-nbd(8) as a subprocess to open URL
46 "https://example.com/disk.qcow2", which is then streamed to stdout
47 ("-"), converted to blocks of raw format data. Note "--" to prevent
48 qemu-nbd flags from being interpreted as nbdcopy flags.
49 cat disk1 disk2 | nbdcopy -- - [ qemu-nbd -f qcow2 output.qcow2 ]
51 Concatenate two raw-format disk images into the qcow2 file
52 output.qcow2. The output file has to be precreated.
53 nbdcopy nbd://server null:
55 Read the contents of the NBD server as fast as possible and throw it
56 away (write it to a virtual null device). This is useful for
57 benchmarking NBD servers and/or nbdcopy.
58
60 nbdcopy copies to and from an NBD server. It can upload a local file
61 to an NBD server, or download the contents of an NBD server to a local
62 file, device or stdin/stdout. It can also copy between NBD servers.
63 The local file can be a file, a block device (eg. "/dev/cdrom"), or "-"
65 which means stream in from stdin or stream out to stdout.
66 The NBD server can be specified using an NBD URI (like
68 "nbd://localhost"). The NBD server can be local or remote, and
69 encryption can be used if libnbd was built with encryption support.
70 Alternately you can use square brackets around a qemu-nbd(8) or
71 nbdkit(1) command to run the NBD server as a subprocess of nbdcopy.
72 The destination may be the special word "null:" to throw away the
74 output.
75 For more complex copying operations including converting between disk
77 formats use "qemu-img convert", see qemu-img(1).
78
80 --help
81 Display brief command line help and exit.
82 --allocated
84 Normally nbdcopy tries to create sparse output (with holes) if the
85 destination supports that. It does this in two ways: either using
86 extent information from the source to copy holes (see
87 --no-extents), or by detecting runs of zeroes (see -S). If you use
88 --allocated then nbdcopy creates a fully allocated, non-sparse
89 output on the destination.
90 -C N
92 --connections=N
93 Set the maximum number of NBD connections ("multi-conn"). By
94 default nbdcopy will try to use multi-conn with up to 4 connections
95 if the NBD server supports it. If copying between NBD servers then
96 nbdcopy cannot use multi-conn if either of the servers does not
97 support it.
98 --destination-is-zero
100 --target-is-zero
101 Assume the destination is already zeroed. This allows nbdcopy to
102 skip copying blocks of zeroes from the source to the destination.
103 This is not safe unless the destination device is already zeroed.
104 (--target-is-zero is provided for compatibility with qemu-img(1).)
105 --flush
107 Flush writes to ensure that everything is written to persistent
108 storage before nbdcopy exits.
109 --no-extents
111 Normally nbdcopy uses extent metadata to skip over parts of the
112 source disk which contain holes. If you use this flag, nbdcopy
113 ignores extent information and reads everything, which is usually
114 slower. You might use this flag in two situations: the source NBD
115 server has incorrect metadata information; or the source has very
116 slow extent querying so it's faster to simply read all of the data.
117 -p
119 --progress
120 Display a progress bar.
121 --progress=FD
123 Write a progress bar to the file descriptor "FD" (a number) in a
124 format which is easily parsable by other programs. nbdcopy will
125 periodically write the string "N/100\n" (where N is an integer
126 between 0 and 100) to the file descriptor.
127 To get nbdcopy to write the progress bar to a file you can use the
129 following shell commands:
130 exec 3>/tmp/progress
132 nbdcopy --progress=3 ...
133 exec 3>&-
134 --queue-size=N
136 Set the maximum number of bytes to queue for in flight requests.
137 The default value is 16 MiB, allowing up to 64 256k requests per
138 NBD connection. If you use larger --request-size you may want to
139 increase this value.
140 --request-size=N
142 Set the maximum request size in bytes. The maximum value is 32 MiB,
143 specified by the NBD protocol.
144 -R N
146 --requests=N
147 Set the maximum number of requests in flight per NBD connection.
148 -S N
150 --sparse=N
151 Detect all zero blocks of size N (bytes) and make them sparse on
152 the output. You can also turn off sparse detection using -S 0.
153 The default is 4096 bytes, or the destination preferred block size,
154 whichever is larger.
155 --synchronous
157 Force synchronous copying using the libnbd(3) synchronous ("high
158 level") API. This is slow but may be necessary for some broken NBD
159 servers which cannot handle multiple requests in flight. This mode
160 is also used when streaming to and from stdio, pipes and sockets.
161 -T N
163 --threads=N
164 Use up to N threads for copying. By default this is set to the
165 number of processor cores available.
166 Note --threads=0 means autodetect and --threads=1 means use a
168 single thread.
169 -v
171 --verbose
172 Verbose output. This enables debug in libnbd (see
173 nbd_set_debug(3)) as well as printing other useful information.
174 -V
176 --version
177 Display the package name and version and exit.
178
180 The three options --connections, --threads and --requests are related
181 and control the amount of parallelism available. The defaults should
182 ensure a reasonable amount of parallelism if possible and you don’t
183 need to adjust them, but this section tries to describe what is going
184 on.
185 Firstly if either side of the copy is streaming to or from stdio, a
187 pipe, or a socket, or if you use the --synchronous option, then nbdcopy
188 works in synchronous mode with no parallelism, and nothing else in this
189 section applies.
190 The --connections=N option controls NBD multi-conn (see "Multi-conn" in
192 libnbd(3)), opening up to N connections to the NBD server (or to both
193 NBD servers if copying between NBD servers). This defaults to 4. The
194 NBD servers must support and advertise multi-conn. For nbdkit(1)
195 availability of multi-conn can depend on the plugin. You can use
196 nbdinfo(1) to find out if a particular NBD server is advertising multi-
197 conn. If the NBD server doesn’t advertise multi-conn then only one
198 connection will be opened regardless of the --connections flag.
199 When copying between two NBD servers, the number of connections is
201 limited to the minimum multi-conn supported on both sides. For the
202 purposes of this calculation, you can consider local files and block
203 devices as supporting infinite multi-conn.
204 When you run an NBD server as a subprocess (using the "[ ... ]" syntax)
206 multi-conn cannot be used.
207 The --threads=N option allows nbdcopy to start up to N threads
209 (defaulting to the number of cores). However nbdcopy cannot use more
210 threads than the number of NBD connections.
211 The --requests=N option controls the maximum number of requests in
213 flight on each NBD connection. This enables the NBD server to process
214 requests in parallel even when multi-conn isn’t available or when using
215 a single thread. The default is chosen to allow a reasonable amount of
216 parallelism without using too much memory.
217 Because of this parallelism, nbdcopy does not read or write blocks in
219 order. If for some reason you require that blocks are copied in strict
220 order then you must use --synchronous.
221
223 Instead of connecting to an already running server using an NBD URI,
224 you can run an NBD server as a subprocess using:
225 nbdcopy -- [ CMD ARGS ... ] ...
227 This requires the server to support systemd socket activation, which
229 both qemu-nbd(8) and nbdkit(1) support (see also
230 nbd_connect_systemd_socket_activation(3)).
231 "[" and "]" must be separate command line parameters. You will usually
233 need to use "--" to stop nbdcopy from misinterpreting NBD server flags
234 as nbdcopy flags. Both the source and destination may be subprocesses.
235 nbdcopy cleans up the subprocess on exit.
236 Some examples follow.
238 nbdcopy -- [ qemu-nbd -f qcow2 disk.qcow2 ] - | hexdump -C
240 In this example, qemu-nbd(8) is run as a subprocess. The subprocess
241 opens disk.qcow2 and exposes it as NBD to nbdcopy. nbdcopy streams
242 this to stdout ("-") into the pipe which is read by hexdump(1). (See
243 also nbddump(1))
244 nbdcopy -- [ qemu-nbd -f qcow2 disk.qcow2 ] [ nbdkit memory 1G ]
246 Two subprocesses are created, qemu-nbd(8) as the source and nbdkit(1)
247 as the destination. The qcow2 file is converted to raw and stored
248 temporarily in the RAM disk (nbdkit-memory-plugin(1)).
249 When nbdcopy exits both servers are killed and the RAM disk goes away,
251 so this command has no overall effect, but is useful for testing.
252
254 libnbd(3), nbddump(1), nbdfuse(1), nbdinfo(1), nbdsh(1), nbdkit(1),
255 qemu-img(1).
256
258 Richard W.M. Jones
259
261 Copyright (C) 2020-2021 Red Hat Inc.
262
264 This library is free software; you can redistribute it and/or modify it
265 under the terms of the GNU Lesser General Public License as published
266 by the Free Software Foundation; either version 2 of the License, or
267 (at your option) any later version.
268 This library is distributed in the hope that it will be useful, but
270 WITHOUT ANY WARRANTY; without even the implied warranty of
271 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
272 Lesser General Public License for more details.
273 You should have received a copy of the GNU Lesser General Public
275 License along with this library; if not, write to the Free Software
276 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
277 02110-1301 USA
278libnbd-1.14.2 2023-01-03 nbdcopy(1)