4 The interfaces for receiving network packages timestamps are:
7 Generates a timestamp for each incoming packet in (not necessarily
8 monotonic) system time. Reports the timestamp via recvmsg() in a
9 control message as struct timeval (usec resolution).
12 Same timestamping mechanism as SO_TIMESTAMP, but reports the
13 timestamp as struct timespec (nsec resolution).
15 * IP_MULTICAST_LOOP + SO_TIMESTAMP[NS]
16 Only for multicast:approximate transmit timestamp obtained by
17 reading the looped packet receive timestamp.
20 Generates timestamps on reception, transmission or both. Supports
21 multiple timestamp sources, including hardware. Supports generating
22 timestamps for stream sockets.
27 This socket option enables timestamping of datagrams on the reception
28 path. Because the destination socket, if any, is not known early in
29 the network stack, the feature has to be enabled for all packets. The
30 same is true for all early receive timestamp options.
32 For interface details, see `man 7 socket`.
37 This option is identical to SO_TIMESTAMP except for the returned data type.
38 Its struct timespec allows for higher resolution (ns) timestamps than the
39 timeval of SO_TIMESTAMP (ms).
44 Supports multiple types of timestamp requests. As a result, this
45 socket option takes a bitmap of flags, not a boolean. In
47 err = setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPING, (void *) val, &val);
49 val is an integer with any of the following bits set. Setting other
50 bit returns EINVAL and does not change the current state.
53 1.3.1 Timestamp Generation
55 Some bits are requests to the stack to try to generate timestamps. Any
56 combination of them is valid. Changes to these bits apply to newly
57 created packets, not to packets already in the stack. As a result, it
58 is possible to selectively request timestamps for a subset of packets
59 (e.g., for sampling) by embedding an send() call within two setsockopt
60 calls, one to enable timestamp generation and one to disable it.
61 Timestamps may also be generated for reasons other than being
62 requested by a particular socket, such as when receive timestamping is
63 enabled system wide, as explained earlier.
65 SOF_TIMESTAMPING_RX_HARDWARE:
66 Request rx timestamps generated by the network adapter.
68 SOF_TIMESTAMPING_RX_SOFTWARE:
69 Request rx timestamps when data enters the kernel. These timestamps
70 are generated just after a device driver hands a packet to the
73 SOF_TIMESTAMPING_TX_HARDWARE:
74 Request tx timestamps generated by the network adapter.
76 SOF_TIMESTAMPING_TX_SOFTWARE:
77 Request tx timestamps when data leaves the kernel. These timestamps
78 are generated in the device driver as close as possible, but always
79 prior to, passing the packet to the network interface. Hence, they
80 require driver support and may not be available for all devices.
82 SOF_TIMESTAMPING_TX_SCHED:
83 Request tx timestamps prior to entering the packet scheduler. Kernel
84 transmit latency is, if long, often dominated by queuing delay. The
85 difference between this timestamp and one taken at
86 SOF_TIMESTAMPING_TX_SOFTWARE will expose this latency independent
87 of protocol processing. The latency incurred in protocol
88 processing, if any, can be computed by subtracting a userspace
89 timestamp taken immediately before send() from this timestamp. On
90 machines with virtual devices where a transmitted packet travels
91 through multiple devices and, hence, multiple packet schedulers,
92 a timestamp is generated at each layer. This allows for fine
93 grained measurement of queuing delay.
95 SOF_TIMESTAMPING_TX_ACK:
96 Request tx timestamps when all data in the send buffer has been
97 acknowledged. This only makes sense for reliable protocols. It is
98 currently only implemented for TCP. For that protocol, it may
99 over-report measurement, because the timestamp is generated when all
100 data up to and including the buffer at send() was acknowledged: the
101 cumulative acknowledgment. The mechanism ignores SACK and FACK.
104 1.3.2 Timestamp Reporting
106 The other three bits control which timestamps will be reported in a
107 generated control message. Changes to the bits take immediate
108 effect at the timestamp reporting locations in the stack. Timestamps
109 are only reported for packets that also have the relevant timestamp
110 generation request set.
112 SOF_TIMESTAMPING_SOFTWARE:
113 Report any software timestamps when available.
115 SOF_TIMESTAMPING_SYS_HARDWARE:
116 This option is deprecated and ignored.
118 SOF_TIMESTAMPING_RAW_HARDWARE:
119 Report hardware timestamps as generated by
120 SOF_TIMESTAMPING_TX_HARDWARE when available.
123 1.3.3 Timestamp Options
125 The interface supports the options
127 SOF_TIMESTAMPING_OPT_ID:
129 Generate a unique identifier along with each packet. A process can
130 have multiple concurrent timestamping requests outstanding. Packets
131 can be reordered in the transmit path, for instance in the packet
132 scheduler. In that case timestamps will be queued onto the error
133 queue out of order from the original send() calls. It is not always
134 possible to uniquely match timestamps to the original send() calls
135 based on timestamp order or payload inspection alone, then.
137 This option associates each packet at send() with a unique
138 identifier and returns that along with the timestamp. The identifier
139 is derived from a per-socket u32 counter (that wraps). For datagram
140 sockets, the counter increments with each sent packet. For stream
141 sockets, it increments with every byte.
143 The counter starts at zero. It is initialized the first time that
144 the socket option is enabled. It is reset each time the option is
145 enabled after having been disabled. Resetting the counter does not
146 change the identifiers of existing packets in the system.
148 This option is implemented only for transmit timestamps. There, the
149 timestamp is always looped along with a struct sock_extended_err.
150 The option modifies field ee_data to pass an id that is unique
151 among all possibly concurrently outstanding timestamp requests for
155 SOF_TIMESTAMPING_OPT_CMSG:
157 Support recv() cmsg for all timestamped packets. Control messages
158 are already supported unconditionally on all packets with receive
159 timestamps and on IPv6 packets with transmit timestamp. This option
160 extends them to IPv4 packets with transmit timestamp. One use case
161 is to correlate packets with their egress device, by enabling socket
162 option IP_PKTINFO simultaneously.
165 SOF_TIMESTAMPING_OPT_TSONLY:
167 Applies to transmit timestamps only. Makes the kernel return the
168 timestamp as a cmsg alongside an empty packet, as opposed to
169 alongside the original packet. This reduces the amount of memory
170 charged to the socket's receive budget (SO_RCVBUF) and delivers
171 the timestamp even if sysctl net.core.tstamp_allow_data is 0.
172 This option disables SOF_TIMESTAMPING_OPT_CMSG.
175 New applications are encouraged to pass SOF_TIMESTAMPING_OPT_ID to
176 disambiguate timestamps and SOF_TIMESTAMPING_OPT_TSONLY to operate
177 regardless of the setting of sysctl net.core.tstamp_allow_data.
179 An exception is when a process needs additional cmsg data, for
180 instance SOL_IP/IP_PKTINFO to detect the egress network interface.
181 Then pass option SOF_TIMESTAMPING_OPT_CMSG. This option depends on
182 having access to the contents of the original packet, so cannot be
183 combined with SOF_TIMESTAMPING_OPT_TSONLY.
186 1.4 Bytestream Timestamps
188 The SO_TIMESTAMPING interface supports timestamping of bytes in a
189 bytestream. Each request is interpreted as a request for when the
190 entire contents of the buffer has passed a timestamping point. That
191 is, for streams option SOF_TIMESTAMPING_TX_SOFTWARE will record
192 when all bytes have reached the device driver, regardless of how
193 many packets the data has been converted into.
195 In general, bytestreams have no natural delimiters and therefore
196 correlating a timestamp with data is non-trivial. A range of bytes
197 may be split across segments, any segments may be merged (possibly
198 coalescing sections of previously segmented buffers associated with
199 independent send() calls). Segments can be reordered and the same
200 byte range can coexist in multiple segments for protocols that
201 implement retransmissions.
203 It is essential that all timestamps implement the same semantics,
204 regardless of these possible transformations, as otherwise they are
205 incomparable. Handling "rare" corner cases differently from the
206 simple case (a 1:1 mapping from buffer to skb) is insufficient
207 because performance debugging often needs to focus on such outliers.
209 In practice, timestamps can be correlated with segments of a
210 bytestream consistently, if both semantics of the timestamp and the
211 timing of measurement are chosen correctly. This challenge is no
212 different from deciding on a strategy for IP fragmentation. There, the
213 definition is that only the first fragment is timestamped. For
214 bytestreams, we chose that a timestamp is generated only when all
215 bytes have passed a point. SOF_TIMESTAMPING_TX_ACK as defined is easy to
216 implement and reason about. An implementation that has to take into
217 account SACK would be more complex due to possible transmission holes
218 and out of order arrival.
220 On the host, TCP can also break the simple 1:1 mapping from buffer to
221 skbuff as a result of Nagle, cork, autocork, segmentation and GSO. The
222 implementation ensures correctness in all cases by tracking the
223 individual last byte passed to send(), even if it is no longer the
224 last byte after an skbuff extend or merge operation. It stores the
225 relevant sequence number in skb_shinfo(skb)->tskey. Because an skbuff
226 has only one such field, only one timestamp can be generated.
228 In rare cases, a timestamp request can be missed if two requests are
229 collapsed onto the same skb. A process can detect this situation by
230 enabling SOF_TIMESTAMPING_OPT_ID and comparing the byte offset at
231 send time with the value returned for each timestamp. It can prevent
232 the situation by always flushing the TCP stack in between requests,
233 for instance by enabling TCP_NODELAY and disabling TCP_CORK and
236 These precautions ensure that the timestamp is generated only when all
237 bytes have passed a timestamp point, assuming that the network stack
238 itself does not reorder the segments. The stack indeed tries to avoid
239 reordering. The one exception is under administrator control: it is
240 possible to construct a packet scheduler configuration that delays
241 segments from the same stream differently. Such a setup would be
247 Timestamps are read using the ancillary data feature of recvmsg().
248 See `man 3 cmsg` for details of this interface. The socket manual
249 page (`man 7 socket`) describes how timestamps generated with
250 SO_TIMESTAMP and SO_TIMESTAMPNS records can be retrieved.
253 2.1 SCM_TIMESTAMPING records
255 These timestamps are returned in a control message with cmsg_level
256 SOL_SOCKET, cmsg_type SCM_TIMESTAMPING, and payload of type
258 struct scm_timestamping {
259 struct timespec ts[3];
262 The structure can return up to three timestamps. This is a legacy
263 feature. Only one field is non-zero at any time. Most timestamps
264 are passed in ts[0]. Hardware timestamps are passed in ts[2].
266 ts[1] used to hold hardware timestamps converted to system time.
267 Instead, expose the hardware clock device on the NIC directly as
268 a HW PTP clock source, to allow time conversion in userspace and
269 optionally synchronize system time with a userspace PTP stack such
270 as linuxptp. For the PTP clock API, see Documentation/ptp/ptp.txt.
272 2.1.1 Transmit timestamps with MSG_ERRQUEUE
274 For transmit timestamps the outgoing packet is looped back to the
275 socket's error queue with the send timestamp(s) attached. A process
276 receives the timestamps by calling recvmsg() with flag MSG_ERRQUEUE
277 set and with a msg_control buffer sufficiently large to receive the
278 relevant metadata structures. The recvmsg call returns the original
279 outgoing data packet with two ancillary messages attached.
281 A message of cm_level SOL_IP(V6) and cm_type IP(V6)_RECVERR
282 embeds a struct sock_extended_err. This defines the error type. For
283 timestamps, the ee_errno field is ENOMSG. The other ancillary message
284 will have cm_level SOL_SOCKET and cm_type SCM_TIMESTAMPING. This
285 embeds the struct scm_timestamping.
288 2.1.1.2 Timestamp types
290 The semantics of the three struct timespec are defined by field
291 ee_info in the extended error structure. It contains a value of
292 type SCM_TSTAMP_* to define the actual timestamp passed in
295 The SCM_TSTAMP_* types are 1:1 matches to the SOF_TIMESTAMPING_*
296 control fields discussed previously, with one exception. For legacy
297 reasons, SCM_TSTAMP_SND is equal to zero and can be set for both
298 SOF_TIMESTAMPING_TX_HARDWARE and SOF_TIMESTAMPING_TX_SOFTWARE. It
299 is the first if ts[2] is non-zero, the second otherwise, in which
300 case the timestamp is stored in ts[0].
303 2.1.1.3 Fragmentation
305 Fragmentation of outgoing datagrams is rare, but is possible, e.g., by
306 explicitly disabling PMTU discovery. If an outgoing packet is fragmented,
307 then only the first fragment is timestamped and returned to the sending
311 2.1.1.4 Packet Payload
313 The calling application is often not interested in receiving the whole
314 packet payload that it passed to the stack originally: the socket
315 error queue mechanism is just a method to piggyback the timestamp on.
316 In this case, the application can choose to read datagrams with a
317 smaller buffer, possibly even of length 0. The payload is truncated
318 accordingly. Until the process calls recvmsg() on the error queue,
319 however, the full packet is queued, taking up budget from SO_RCVBUF.
322 2.1.1.5 Blocking Read
324 Reading from the error queue is always a non-blocking operation. To
325 block waiting on a timestamp, use poll or select. poll() will return
326 POLLERR in pollfd.revents if any data is ready on the error queue.
327 There is no need to pass this flag in pollfd.events. This flag is
328 ignored on request. See also `man 2 poll`.
331 2.1.2 Receive timestamps
333 On reception, there is no reason to read from the socket error queue.
334 The SCM_TIMESTAMPING ancillary data is sent along with the packet data
335 on a normal recvmsg(). Since this is not a socket error, it is not
336 accompanied by a message SOL_IP(V6)/IP(V6)_RECVERROR. In this case,
337 the meaning of the three fields in struct scm_timestamping is
338 implicitly defined. ts[0] holds a software timestamp if set, ts[1]
339 is again deprecated and ts[2] holds a hardware timestamp if set.
342 3. Hardware Timestamping configuration: SIOCSHWTSTAMP and SIOCGHWTSTAMP
344 Hardware time stamping must also be initialized for each device driver
345 that is expected to do hardware time stamping. The parameter is defined in
346 /include/linux/net_tstamp.h as:
348 struct hwtstamp_config {
349 int flags; /* no flags defined right now, must be zero */
350 int tx_type; /* HWTSTAMP_TX_* */
351 int rx_filter; /* HWTSTAMP_FILTER_* */
354 Desired behavior is passed into the kernel and to a specific device by
355 calling ioctl(SIOCSHWTSTAMP) with a pointer to a struct ifreq whose
356 ifr_data points to a struct hwtstamp_config. The tx_type and
357 rx_filter are hints to the driver what it is expected to do. If
358 the requested fine-grained filtering for incoming packets is not
359 supported, the driver may time stamp more than just the requested types
362 A driver which supports hardware time stamping shall update the struct
363 with the actual, possibly more permissive configuration. If the
364 requested packets cannot be time stamped, then nothing should be
365 changed and ERANGE shall be returned (in contrast to EINVAL, which
366 indicates that SIOCSHWTSTAMP is not supported at all).
368 Only a processes with admin rights may change the configuration. User
369 space is responsible to ensure that multiple processes don't interfere
370 with each other and that the settings are reset.
372 Any process can read the actual configuration by passing this
373 structure to ioctl(SIOCGHWTSTAMP) in the same way. However, this has
374 not been implemented in all drivers.
376 /* possible values for hwtstamp_config->tx_type */
379 * no outgoing packet will need hardware time stamping;
380 * should a packet arrive which asks for it, no hardware
381 * time stamping will be done
386 * enables hardware time stamping for outgoing packets;
387 * the sender of the packet decides which are to be
388 * time stamped by setting SOF_TIMESTAMPING_TX_SOFTWARE
389 * before sending the packet
394 /* possible values for hwtstamp_config->rx_filter */
396 /* time stamp no incoming packet at all */
397 HWTSTAMP_FILTER_NONE,
399 /* time stamp any incoming packet */
402 /* return value: time stamp all packets requested plus some others */
403 HWTSTAMP_FILTER_SOME,
405 /* PTP v1, UDP, any kind of event packet */
406 HWTSTAMP_FILTER_PTP_V1_L4_EVENT,
408 /* for the complete list of values, please check
409 * the include file /include/linux/net_tstamp.h
413 3.1 Hardware Timestamping Implementation: Device Drivers
415 A driver which supports hardware time stamping must support the
416 SIOCSHWTSTAMP ioctl and update the supplied struct hwtstamp_config with
417 the actual values as described in the section on SIOCSHWTSTAMP. It
418 should also support SIOCGHWTSTAMP.
420 Time stamps for received packets must be stored in the skb. To get a pointer
421 to the shared time stamp structure of the skb call skb_hwtstamps(). Then
422 set the time stamps in the structure:
424 struct skb_shared_hwtstamps {
425 /* hardware time stamp transformed into duration
426 * since arbitrary point in time
431 Time stamps for outgoing packets are to be generated as follows:
432 - In hard_start_xmit(), check if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)
433 is set no-zero. If yes, then the driver is expected to do hardware time
435 - If this is possible for the skb and requested, then declare
436 that the driver is doing the time stamping by setting the flag
437 SKBTX_IN_PROGRESS in skb_shinfo(skb)->tx_flags , e.g. with
439 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
441 You might want to keep a pointer to the associated skb for the next step
442 and not free the skb. A driver not supporting hardware time stamping doesn't
443 do that. A driver must never touch sk_buff::tstamp! It is used to store
444 software generated time stamps by the network subsystem.
445 - Driver should call skb_tx_timestamp() as close to passing sk_buff to hardware
446 as possible. skb_tx_timestamp() provides a software time stamp if requested
447 and hardware timestamping is not possible (SKBTX_IN_PROGRESS not set).
448 - As soon as the driver has sent the packet and/or obtained a
449 hardware time stamp for it, it passes the time stamp back by
450 calling skb_hwtstamp_tx() with the original skb, the raw
451 hardware time stamp. skb_hwtstamp_tx() clones the original skb and
452 adds the timestamps, therefore the original skb has to be freed now.
453 If obtaining the hardware time stamp somehow fails, then the driver
454 should not fall back to software time stamping. The rationale is that
455 this would occur at a later time in the processing pipeline than other
456 software time stamping and therefore could lead to unexpected deltas