2 RFC: Common Trace Format (CTF) Proposal (v1.6)
4 Mathieu Desnoyers, EfficiOS Inc.
6 The goal of the present document is to propose a trace format that suits the
7 needs of the embedded, telecom, high-performance and kernel communities. It is
8 based on the Common Trace Format Requirements (v1.4) document. It is designed to
9 allow traces to be natively generated by the Linux kernel, Linux user-space
10 applications written in C/C++, and hardware components.
12 The latest version of this document can be found at:
14 git tree: git://git.efficios.com/ctf.git
15 gitweb: http://git.efficios.com/?p=ctf.git
17 A reference implementation of a library to read and write this trace format is
18 being implemented within the BabelTrace project, a converter between trace
19 formats. The development tree is available at:
21 git tree: git://git.efficios.com/babeltrace.git
22 gitweb: http://git.efficios.com/?p=babeltrace.git
25 1. Preliminary definitions
27 - Event Trace: An ordered sequence of events.
28 - Event Stream: An ordered sequence of events, containing a subset of the
30 - Event Packet: A sequence of physically contiguous events within an event
32 - Event: This is the basic entry in a trace. (aka: a trace record).
33 - An event identifier (ID) relates to the class (a type) of event within
35 e.g. event: irq_entry.
36 - An event (or event record) relates to a specific instance of an event
38 e.g. event: irq_entry, at time X, on CPU Y
39 - Source Architecture: Architecture writing the trace.
40 - Reader Architecture: Architecture reading the trace.
43 2. High-level representation of a trace
45 A trace is divided into multiple event streams. Each event stream contains a
46 subset of the trace event types.
48 The final output of the trace, after its generation and optional transport over
49 the network, is expected to be either on permanent or temporary storage in a
50 virtual file system. Because each event stream is appended to while a trace is
51 being recorded, each is associated with a separate file for output. Therefore,
52 a stored trace can be represented as a directory containing one file per stream.
54 A metadata event stream contains information on trace event types. It describes:
58 - Per-stream event header description.
59 - Per-stream event header selection.
60 - Per-stream event context fields.
62 - Event type to stream mapping.
63 - Event type to name mapping.
64 - Event type to ID mapping.
65 - Event fields description.
70 An event stream is divided in contiguous event packets of variable size. These
71 subdivisions have a variable size. An event packet can contain a certain amount
72 of padding at the end. The rationale for the event stream design choices is
73 explained in Appendix B. Stream Header Rationale.
75 An event stream is divided in contiguous event packets of variable size. These
76 subdivisions have a variable size. An event packet can contain a certain amount
77 of padding at the end. The stream header is repeated at the beginning of each
80 The event stream header will therefore be referred to as the "event packet
81 header" throughout the rest of this document.
88 A basic type is a scalar type, as described in this section.
90 4.1.1 Type inheritance
92 Type specifications can be inherited to allow deriving types from a
93 type class. For example, see the uint32_t named type derived from the "integer"
94 type class below ("Integers" section). Types have a precise binary
95 representation in the trace. A type class has methods to read and write these
96 types, but must be derived into a type to be usable in an event field.
100 We define "byte-packed" types as aligned on the byte size, namely 8-bit.
101 We define "bit-packed" types as following on the next bit, as defined by the
104 All basic types, except bitfields, are either aligned on an architecture-defined
105 specific alignment or byte-packed, depending on the architecture preference.
106 Architectures providing fast unaligned write byte-packed basic types to save
107 space, aligning each type on byte boundaries (8-bit). Architectures with slow
108 unaligned writes align types on specific alignment values. If no specific
109 alignment is declared for a type nor its parents, it is assumed to be bit-packed
110 for bitfields and byte-packed for other types.
112 Metadata attribute representation of a specific alignment:
114 align = value; /* value in bits */
118 By default, the native endianness of the source architecture the trace is used.
119 Byte order can be overridden for a basic type by specifying a "byte_order"
120 attribute. Typical use-case is to specify the network byte order (big endian:
121 "be") to save data captured from the network into the trace without conversion.
122 If not specified, the byte order is native.
124 Metadata representation:
126 byte_order = native OR network OR be OR le; /* network and be are aliases */
130 Type size, in bits, for integers and floats is that returned by "sizeof()" in C
131 multiplied by CHAR_BIT.
132 We require the size of "char" and "unsigned char" types (CHAR_BIT) to be fixed
133 to 8 bits for cross-endianness compatibility.
135 Metadata representation:
137 size = value; (value is in bits)
141 Signed integers are represented in two-complement. Integer alignment, size,
142 signedness and byte ordering are defined in the metadata. Integers aligned on
143 byte size (8-bit) and with length multiple of byte size (8-bit) correspond to
144 the C99 standard integers. In addition, integers with alignment and/or size that
145 are _not_ a multiple of the byte size are permitted; these correspond to the C99
146 standard bitfields, with the added specification that the CTF integer bitfields
147 have a fixed binary representation. A MIT-licensed reference implementation of
148 the CTF portable bitfields is available at:
150 http://git.efficios.com/?p=babeltrace.git;a=blob;f=include/babeltrace/bitfield.h
152 Binary representation of integers:
154 - On little and big endian:
155 - Within a byte, high bits correspond to an integer high bits, and low bits
156 correspond to low bits.
158 - Integer across multiple bytes are placed from the less significant to the
160 - Consecutive integers are placed from lower bits to higher bits (even within
163 - Integer across multiple bytes are placed from the most significant to the
165 - Consecutive integers are placed from higher bits to lower bits (even within
168 This binary representation is derived from the bitfield implementation in GCC
169 for little and big endian. However, contrary to what GCC does, integers can
170 cross units boundaries (no padding is required). Padding can be explicitely
171 added (see 4.1.6 GNU/C bitfields) to follow the GCC layout if needed.
173 Metadata representation:
176 signed = true OR false; /* default false */
177 byte_order = native OR network OR be OR le; /* default native */
178 size = value; /* value in bits, no default */
179 align = value; /* value in bits */
182 Example of type inheritance (creation of a uint32_t named type):
190 Definition of a named 5-bit signed bitfield:
198 4.1.6 GNU/C bitfields
200 The GNU/C bitfields follow closely the integer representation, with a
201 particularity on alignment: if a bitfield cannot fit in the current unit, the
202 unit is padded and the bitfield starts at the following unit. The unit size is
203 defined by the size of the type "unit_type".
205 Metadata representation. Either:
208 unit_type = integer {
214 Or bitfield within structures as specified by the C standard
218 As an example, the following structure declared in C compiled by GCC:
225 is equivalent to the following structure declaration, aligned on the largest
226 element (short). The second bitfield would be aligned on the next unit boundary,
227 because it would not fit in the current unit. The two declarations (C
228 declaration above or CTF declaration with "type gcc_bitfield") are strictly
244 The floating point values byte ordering is defined in the metadata.
246 Floating point values follow the IEEE 754-2008 standard interchange formats.
247 Description of the floating point values include the exponent and mantissa size
248 in bits. Some requirements are imposed on the floating point values:
250 - FLT_RADIX must be 2.
251 - mant_dig is the number of digits represented in the mantissa. It is specified
252 by the ISO C99 standard, section 5.2.4, as FLT_MANT_DIG, DBL_MANT_DIG and
253 LDBL_MANT_DIG as defined by <float.h>.
254 - exp_dig is the number of digits represented in the exponent. Given that
255 mant_dig is one bit more than its actual size in bits (leading 1 is not
256 needed) and also given that the sign bit always takes one bit, exp_dig can be
259 - sizeof(float) * CHAR_BIT - FLT_MANT_DIG
260 - sizeof(double) * CHAR_BIT - DBL_MANT_DIG
261 - sizeof(long double) * CHAR_BIT - LDBL_MANT_DIG
263 Metadata representation:
268 byte_order = native OR network OR be OR le;
271 Example of type inheritance:
273 typedef floating_point {
274 exp_dig = 8; /* sizeof(float) * CHAR_BIT - FLT_MANT_DIG */
275 mant_dig = 24; /* FLT_MANT_DIG */
279 TODO: define NaN, +inf, -inf behavior.
283 Enumerations are a mapping between an integer type and a table of strings. The
284 numerical representation of the enumeration follows the integer type specified
285 by the metadata. The enumeration mapping table is detailed in the enumeration
286 description within the metadata. The mapping table maps inclusive value ranges
287 (or single values) to strings. Instead of being limited to simple
288 "value -> string" mappings, these enumerations map
289 "[ start_value ... end_value ] -> string", which map inclusive ranges of
290 values to strings. An enumeration from the C language can be represented in
291 this format by having the same start_value and end_value for each element, which
292 is in fact a range of size 1. This single-value range is supported without
293 repeating the start and end values with the value = string declaration. If the
294 <integer_type> is omitted, the type chosen by the C compiler to hold the
295 enumeration is used. The <integer_type> specifier can only be omitted for
296 enumerations containing only simple "value -> string" mappings (compatible with
299 enum <integer_type> name {
300 string = start_value1 ... end_value1,
301 "other string" = start_value2 ... end_value2,
302 yet_another_string, /* will be assigned to end_value2 + 1 */
303 "some other string" = value,
307 If the values are omitted, the enumeration starts at 0 and increment of 1 for
318 Overlapping ranges within a single enumeration are implementation defined.
324 Structures are aligned on the largest alignment required by basic types
325 contained within the structure. (This follows the ISO/C standard for structures)
327 Metadata representation of a named structure:
330 field_type field_name;
331 field_type field_name;
338 integer { /* Nameless type */
343 uint64_t second_field_name; /* Named type declared in the metadata */
346 The fields are placed in a sequence next to each other. They each possess a
347 field name, which is a unique identifier within the structure.
349 A nameless structure can be declared as a field type:
357 Arrays are fixed-length. Their length is declared in the type declaration within
358 the metadata. They contain an array of "inner type" elements, which can refer to
359 any type not containing the type of the array being declared (no circular
360 dependency). The length is the number of elements in an array.
362 Metadata representation of a named array, either:
371 typedef elem_type name[length];
377 elem_type = uint32_t;
380 A nameless array can be declared as a field type, e.g.:
389 uint8_t field_name[10];
394 Sequences are dynamically-sized arrays. They start with an integer that specify
395 the length of the sequence, followed by an array of "inner type" elements.
396 The length is the number of elements in the sequence.
398 Metadata representation for a named sequence, either:
401 length_type = type; /* integer class */
407 typedef elem_type name[length_type];
409 A nameless sequence can be declared as a field type, e.g.:
418 long field_name[int];
420 The length type follows the integer types specifications, and the sequence
421 elements follow the "array" specifications.
425 Strings are an array of bytes of variable size and are terminated by a '\0'
426 "NULL" character. Their encoding is described in the metadata. In absence of
427 encoding attribute information, the default encoding is UTF-8.
429 Metadata representation of a named string type:
432 encoding = UTF8 OR ASCII;
435 A nameless string type can be declared as a field type:
437 string field_name; /* Use default UTF8 encoding */
439 5. Event Packet Header
441 The event packet header consists of two part: one is mandatory and have a fixed
442 layout. The second part, the "event packet context", has its layout described in
445 - Aligned on page size. Fixed size. Fields either aligned or packed (depending
446 on the architecture preference).
447 No padding at the end of the event packet header. Native architecture byte
450 Fixed layout (event packet header):
452 - Magic number (CTF magic numbers: 0xC1FC1FC1 and its reverse endianness
453 representation: 0xC11FFCC1) It needs to have a non-symmetric bytewise
454 representation. Used to distinguish between big and little endian traces (this
455 information is determined by knowing the endianness of the architecture
456 reading the trace and comparing the magic number against its value and the
457 reverse, 0xC11FFCC1). This magic number specifies that we use the CTF metadata
458 description language described in this document. Different magic numbers
459 should be used for other metadata description languages.
460 - Trace UUID, used to ensure the event packet match the metadata used.
461 (note: we cannot use a metadata checksum because metadata can be appended to
462 while tracing is active)
463 - Stream ID, used as reference to stream description in metadata.
465 Metadata-defined layout (event packet context):
467 - Event packet content size (in bytes).
468 - Event packet size (in bytes, includes padding).
469 - Event packet content checksum (optional). Checksum excludes the event packet
471 - Per-stream event packet sequence count (to deal with UDP packet loss). The
472 number of significant sequence counter bits should also be present, so
473 wrap-arounds are deal with correctly.
474 - Timestamp at the beginning and timestamp at the end of the event packet.
475 Both timestamps are written in the packet header, but sampled respectively
476 while (or before) writing the first event and while (or after) writing the
477 last event in the packet. The inclusive range between these timestamps should
478 include all event timestamps assigned to events contained within the packet.
479 - Events discarded count
480 - Snapshot of a per-stream free-running counter, counting the number of
481 events discarded that were supposed to be written in the stream prior to
482 the first event in the event packet.
483 * Note: producer-consumer buffer full condition should fill the current
484 event packet with padding so we know exactly where events have been
486 - Lossless compression scheme used for the event packet content. Applied
487 directly to raw data. New types of compression can be added in following
488 versions of the format.
489 0: no compression scheme
493 - Cypher used for the event packet content. Applied after compression.
496 - Checksum scheme used for the event packet content. Applied after encryption.
502 5.1 Event Packet Header Fixed Layout Description
504 struct event_packet_header {
506 uint8_t trace_uuid[16];
510 5.2 Event Packet Context Description
512 Event packet context example. These are declared within the stream declaration
513 in the metadata. All these fields are optional except for "content_size" and
514 "packet_size", which must be present in the context.
516 An example event packet context type:
518 struct event_packet_context {
519 uint64_t timestamp_begin;
520 uint64_t timestamp_end;
522 uint32_t stream_packet_count;
523 uint32_t events_discarded;
525 uint32_t/uint16_t content_size;
526 uint32_t/uint16_t packet_size;
527 uint8_t stream_packet_count_bits; /* Significant counter bits */
528 uint8_t compression_scheme;
529 uint8_t encryption_scheme;
535 The overall structure of an event is:
537 - Event Header (as specifed by the stream metadata)
538 - Extended Event Header (as specified by the event header)
539 - Event Context (as specified by the stream metadata)
540 - Event Payload (as specified by the event metadata)
545 One major factor can vary between streams: the number of event IDs assigned to
546 a stream. Luckily, this information tends to stay relatively constant (modulo
547 event registration while trace is being recorded), so we can specify different
548 representations for streams containing few event IDs and streams containing
549 many event IDs, so we end up representing the event ID and timestamp as densely
550 as possible in each case.
552 We therefore provide two types of events headers. Type 1 accommodates streams
553 with less than 31 event IDs. Type 2 accommodates streams with 31 or more event
556 The "extended headers" are used in the rare occasions where the information
557 cannot be represented in the ranges available in the event header. They are also
558 used in the rare occasions where the data required for a field could not be
559 collected: the flag corresponding to the missing field within the missing_fields
560 array is then set to 1.
562 Types uintX_t represent an X-bit unsigned integer.
565 6.1.1 Type 1 - Few event IDs
567 - Aligned on 32-bit (or 8-bit if byte-packed, depending on the architecture
569 - Fixed size: 32 bits.
570 - Native architecture byte ordering.
572 struct event_header_1 {
575 * id 31 is reserved to indicate a following
581 The end of a type 1 header is aligned on a 32-bit boundary (or packed).
584 6.1.2 Extended Type 1 Event Header
586 - Follows struct event_header_1, which is aligned on 32-bit, so no need to
588 - Variable size (depends on the number of fields per event).
589 - Native architecture byte ordering.
590 - NR_FIELDS is the number of fields within the event.
592 struct event_header_1_ext {
593 uint32_t id; /* 32-bit event IDs */
594 uint64_t timestamp; /* 64-bit timestamps */
595 uint1_t missing_fields[NR_FIELDS]; /* missing event fields bitmap */
599 6.1.3 Type 2 - Many event IDs
601 - Aligned on 32-bit (or 8-bit if byte-packed, depending on the architecture
603 - Fixed size: 48 bits.
604 - Native architecture byte ordering.
606 struct event_header_2 {
609 * id: range: 0 - 65534.
610 * id 65535 is reserved to indicate a following
615 The end of a type 2 header is aligned on a 16-bit boundary (or 8-bit if
619 6.1.4 Extended Type 2 Event Header
621 - Follows struct event_header_2, which alignment end on a 16-bit boundary, so
622 we need to align on 64-bit integer architecture alignment (or 8-bit if
624 - Variable size (depends on the number of fields per event).
625 - Native architecture byte ordering.
626 - NR_FIELDS is the number of fields within the event.
628 struct event_header_2_ext {
629 uint64_t timestamp; /* 64-bit timestamps */
630 uint32_t id; /* 32-bit event IDs */
631 uint1_t missing_fields[NR_FIELDS]; /* missing event fields bitmap */
637 The event context contains information relative to the current event. The choice
638 and meaning of this information is specified by the metadata "stream"
639 information. For this trace format, event context is usually empty, except when
640 the metadata "stream" information specifies otherwise by declaring a non-empty
641 structure for the event context. An example of event context is to save the
642 event payload size with each event, or to save the current PID with each event.
643 These are declared within the stream declaration within the metadata.
645 An example event context type:
647 struct event_context {
649 uint16_t payload_size;
655 An event payload contains fields specific to a given event type. The fields
656 belonging to an event type are described in the event-specific metadata
657 within a structure type.
661 No padding at the end of the event payload. This differs from the ISO/C standard
662 for structures, but follows the CTF standard for structures. In a trace, even
663 though it makes sense to align the beginning of a structure, it really makes no
664 sense to add padding at the end of the structure, because structures are usually
665 not followed by a structure of the same type.
667 This trick can be done by adding a zero-length "end" field at the end of the C
668 structures, and by using the offset of this field rather than using sizeof()
669 when calculating the size of a structure (see Appendix "A. Helper macros").
673 The event payload is aligned on the largest alignment required by types
674 contained within the payload. (This follows the ISO/C standard for structures)
680 The meta-data is located in a stream named "metadata". It is made of "event
681 packets", which each start with an event packet header. The event type within
682 the metadata stream have no event header nor event context. Each event only
683 contains a null-terminated "string" payload, which is a metadata description
684 entry. The events are packed one next to another. Each event packet start with
685 an event packet header, which contains, amongst other fields, the magic number
688 The metadata can be parsed by reading through the metadata strings, skipping
689 newlines and null-characters. Type names may contain spaces.
692 major = value; /* Trace format version */
694 uuid = value; /* Trace UUID */
701 /* Type 1 - Few event IDs; Type 2 - Many event IDs. See section 6.1. */
702 header_type = event_header_1 OR event_header_2;
704 * Extended event header type. Only present if specified in event header
705 * on a per-event basis.
707 header_type_ext = event_header_1_ext OR event_header_2_ext;
708 context_type = struct {
713 context_type = struct {
721 id = value; /* Numeric identifier within the stream */
728 /* More detail on types in section 4. Types */
731 typedef some existing type new_type;
745 /* Unnamed types, contained within compound type fields or type assignments. */
764 The two following macros keep track of the size of a GNU/C structure without
765 padding at the end by placing HEADER_END as the last field. A one byte end field
766 is used for C90 compatibility (C99 flexible arrays could be used here). Note
767 that this does not affect the effective structure size, which should always be
768 calculated with the header_sizeof() helper.
770 #define HEADER_END char end_field
771 #define header_sizeof(type) offsetof(typeof(type), end_field)
774 B. Stream Header Rationale
776 An event stream is divided in contiguous event packets of variable size. These
777 subdivisions allow the trace analyzer to perform a fast binary search by time
778 within the stream (typically requiring to index only the event packet headers)
779 without reading the whole stream. These subdivisions have a variable size to
780 eliminate the need to transfer the event packet padding when partially filled
781 event packets must be sent when streaming a trace for live viewing/analysis.
782 An event packet can contain a certain amount of padding at the end. Dividing
783 streams into event packets is also useful for network streaming over UDP and
784 flight recorder mode tracing (a whole event packet can be swapped out of the
785 buffer atomically for reading).
787 The stream header is repeated at the beginning of each event packet to allow
788 flexibility in terms of:
791 - allowing arbitrary buffers to be discarded without making the trace
793 - allow UDP packet loss handling by either dealing with missing event packet
794 or asking for re-transmission.
795 - transparently support flight recorder mode,
796 - transparently support crash dump.
798 The event stream header will therefore be referred to as the "event packet
799 header" throughout the rest of this document.