1* Introduction
   3The name "usbmon" in lowercase refers to a facility in kernel which is
   4used to collect traces of I/O on the USB bus. This function is analogous
   5to a packet socket used by network monitoring tools such as tcpdump(1)
   6or Ethereal. Similarly, it is expected that a tool such as usbdump or
   7USBMon (with uppercase letters) is used to examine raw traces produced
   8by usbmon.
  10The usbmon reports requests made by peripheral-specific drivers to Host
  11Controller Drivers (HCD). So, if HCD is buggy, the traces reported by
  12usbmon may not correspond to bus transactions precisely. This is the same
  13situation as with tcpdump.
  15* How to use usbmon to collect raw text traces
  17Unlike the packet socket, usbmon has an interface which provides traces
  18in a text format. This is used for two purposes. First, it serves as a
  19common trace exchange format for tools while more sophisticated formats
  20are finalized. Second, humans can read it in case tools are not available.
  22To collect a raw text trace, execute following steps.
  241. Prepare
  26Mount debugfs (it has to be enabled in your kernel configuration), and
  27load the usbmon module (if built as module). The second step is skipped
  28if usbmon is built into the kernel.
  30# mount -t debugfs none_debugs /sys/kernel/debug
  31# modprobe usbmon
  34Verify that bus sockets are present.
  36# ls /sys/kernel/debug/usbmon
  370s  0u  1s  1t  1u  2s  2t  2u  3s  3t  3u  4s  4t  4u
  40Now you can choose to either use the socket '0u' (to capture packets on all
  41buses), and skip to step #3, or find the bus used by your device with step #2.
  42This allows to filter away annoying devices that talk continuously.
  442. Find which bus connects to the desired device
  46Run "cat /proc/bus/usb/devices", and find the T-line which corresponds to
  47the device. Usually you do it by looking for the vendor string. If you have
  48many similar devices, unplug one and compare two /proc/bus/usb/devices outputs.
  49The T-line will have a bus number. Example:
  51T:  Bus=03 Lev=01 Prnt=01 Port=00 Cnt=01 Dev#=  2 Spd=12  MxCh= 0
  52D:  Ver= 1.10 Cls=00(>ifc ) Sub=00 Prot=00 MxPS= 8 #Cfgs=  1
  53P:  Vendor=0557 ProdID=2004 Rev= 1.00
  54S:  Manufacturer=ATEN
  55S:  Product=UC100KM V2.00
  57Bus=03 means it's bus 3.
  593. Start 'cat'
  61# cat /sys/kernel/debug/usbmon/3u > /tmp/1.mon.out
  63to listen on a single bus, otherwise, to listen on all buses, type:
  65# cat /sys/kernel/debug/usbmon/0u > /tmp/1.mon.out
  67This process will be reading until killed. Naturally, the output can be
  68redirected to a desirable location. This is preferred, because it is going
  69to be quite long.
  714. Perform the desired operation on the USB bus
  73This is where you do something that creates the traffic: plug in a flash key,
  74copy files, control a webcam, etc.
  765. Kill cat
  78Usually it's done with a keyboard interrupt (Control-C).
  80At this point the output file (/tmp/1.mon.out in this example) can be saved,
  81sent by e-mail, or inspected with a text editor. In the last case make sure
  82that the file size is not excessive for your favourite editor.
  84* Raw text data format
  86Two formats are supported currently: the original, or '1t' format, and
  87the '1u' format. The '1t' format is deprecated in kernel 2.6.21. The '1u'
  88format adds a few fields, such as ISO frame descriptors, interval, etc.
  89It produces slightly longer lines, but otherwise is a perfect superset
  90of '1t' format.
  92If it is desired to recognize one from the other in a program, look at the
  93"address" word (see below), where '1u' format adds a bus number. If 2 colons
  94are present, it's the '1t' format, otherwise '1u'.
  96Any text format data consists of a stream of events, such as URB submission,
  97URB callback, submission error. Every event is a text line, which consists
  98of whitespace separated words. The number or position of words may depend
  99on the event type, but there is a set of words, common for all types.
 101Here is the list of words, from left to right:
 103- URB Tag. This is used to identify URBs, and is normally an in-kernel address
 104  of the URB structure in hexadecimal, but can be a sequence number or any
 105  other unique string, within reason.
 107- Timestamp in microseconds, a decimal number. The timestamp's resolution
 108  depends on available clock, and so it can be much worse than a microsecond
 109  (if the implementation uses jiffies, for example).
 111- Event Type. This type refers to the format of the event, not URB type.
 112  Available types are: S - submission, C - callback, E - submission error.
 114- "Address" word (formerly a "pipe"). It consists of four fields, separated by
 115  colons: URB type and direction, Bus number, Device address, Endpoint number.
 116  Type and direction are encoded with two bytes in the following manner:
 117    Ci Co   Control input and output
 118    Zi Zo   Isochronous input and output
 119    Ii Io   Interrupt input and output
 120    Bi Bo   Bulk input and output
 121  Bus number, Device address, and Endpoint are decimal numbers, but they may
 122  have leading zeros, for the sake of human readers.
 124- URB Status word. This is either a letter, or several numbers separated
 125  by colons: URB status, interval, start frame, and error count. Unlike the
 126  "address" word, all fields save the status are optional. Interval is printed
 127  only for interrupt and isochronous URBs. Start frame is printed only for
 128  isochronous URBs. Error count is printed only for isochronous callback
 129  events.
 131  The status field is a decimal number, sometimes negative, which represents
 132  a "status" field of the URB. This field makes no sense for submissions, but
 133  is present anyway to help scripts with parsing. When an error occurs, the
 134  field contains the error code.
 136  In case of a submission of a Control packet, this field contains a Setup Tag
 137  instead of an group of numbers. It is easy to tell whether the Setup Tag is
 138  present because it is never a number. Thus if scripts find a set of numbers
 139  in this word, they proceed to read Data Length (except for isochronous URBs).
 140  If they find something else, like a letter, they read the setup packet before
 141  reading the Data Length or isochronous descriptors.
 143- Setup packet, if present, consists of 5 words: one of each for bmRequestType,
 144  bRequest, wValue, wIndex, wLength, as specified by the USB Specification 2.0.
 145  These words are safe to decode if Setup Tag was 's'. Otherwise, the setup
 146  packet was present, but not captured, and the fields contain filler.
 148- Number of isochronous frame descriptors and descriptors themselves.
 149  If an Isochronous transfer event has a set of descriptors, a total number
 150  of them in an URB is printed first, then a word per descriptor, up to a
 151  total of 5. The word consists of 3 colon-separated decimal numbers for
 152  status, offset, and length respectively. For submissions, initial length
 153  is reported. For callbacks, actual length is reported.
 155- Data Length. For submissions, this is the requested length. For callbacks,
 156  this is the actual length.
 158- Data tag. The usbmon may not always capture data, even if length is nonzero.
 159  The data words are present only if this tag is '='.
 161- Data words follow, in big endian hexadecimal format. Notice that they are
 162  not machine words, but really just a byte stream split into words to make
 163  it easier to read. Thus, the last word may contain from one to four bytes.
 164  The length of collected data is limited and can be less than the data length
 165  report in Data Length word.
 167Here is an example of code to read the data stream in a well known programming
 170class ParsedLine {
 171        int data_len;           /* Available length of data */
 172        byte data[];
 174        void parseData(StringTokenizer st) {
 175                int availwords = st.countTokens();
 176                data = new byte[availwords * 4];
 177                data_len = 0;
 178                while (st.hasMoreTokens()) {
 179                        String data_str = st.nextToken();
 180                        int len = data_str.length() / 2;
 181                        int i;
 182                        int b;  // byte is signed, apparently?! XXX
 183                        for (i = 0; i < len; i++) {
 184                                // data[data_len] = Byte.parseByte(
 185                                //     data_str.substring(i*2, i*2 + 2),
 186                                //     16);
 187                                b = Integer.parseInt(
 188                                     data_str.substring(i*2, i*2 + 2),
 189                                     16);
 190                                if (b >= 128)
 191                                        b *= -1;
 192                                data[data_len] = (byte) b;
 193                                data_len++;
 194                        }
 195                }
 196        }
 201An input control transfer to get a port status.
 203d5ea89a0 3575914555 S Ci:1:001:0 s a3 00 0000 0003 0004 4 <
 204d5ea89a0 3575914560 C Ci:1:001:0 0 4 = 01050000
 206An output bulk transfer to send a SCSI command 0x5E in a 31-byte Bulk wrapper
 207to a storage device at address 5:
 209dd65f0e8 4128379752 S Bo:1:005:2 -115 31 = 55534243 5e000000 00000000 00000600 00000000 00000000 00000000 000000
 210dd65f0e8 4128379808 C Bo:1:005:2 0 31 >
 212* Raw binary format and API
 214The overall architecture of the API is about the same as the one above,
 215only the events are delivered in binary format. Each event is sent in
 216the following structure (its name is made up, so that we can refer to it):
 218struct usbmon_packet {
 219        u64 id;                 /*  0: URB ID - from submission to callback */
 220        unsigned char type;     /*  8: Same as text; extensible. */
 221        unsigned char xfer_type; /*    ISO (0), Intr, Control, Bulk (3) */
 222        unsigned char epnum;    /*     Endpoint number and transfer direction */
 223        unsigned char devnum;   /*     Device address */
 224        u16 busnum;             /* 12: Bus number */
 225        char flag_setup;        /* 14: Same as text */
 226        char flag_data;         /* 15: Same as text; Binary zero is OK. */
 227        s64 ts_sec;             /* 16: gettimeofday */
 228        s32 ts_usec;            /* 24: gettimeofday */
 229        int status;             /* 28: */
 230        unsigned int length;    /* 32: Length of data (submitted or actual) */
 231        unsigned int len_cap;   /* 36: Delivered length */
 232        unsigned char setup[8]; /* 40: Only for Control 'S' */
 233};                              /* 48 bytes total */
 235These events can be received from a character device by reading with read(2),
 236with an ioctl(2), or by accessing the buffer with mmap.
 238The character device is usually called /dev/usbmonN, where N is the USB bus
 239number. Number zero (/dev/usbmon0) is special and means "all buses".
 240However, this feature is not implemented yet. Note that specific naming
 241policy is set by your Linux distribution.
 243If you create /dev/usbmon0 by hand, make sure that it is owned by root
 244and has mode 0600. Otherwise, unpriviledged users will be able to snoop
 245keyboard traffic.
 247The following ioctl calls are available, with MON_IOC_MAGIC 0x92:
 249 MON_IOCQ_URB_LEN, defined as _IO(MON_IOC_MAGIC, 1)
 251This call returns the length of data in the next event. Note that majority of
 252events contain no data, so if this call returns zero, it does not mean that
 253no events are available.
 255 MON_IOCG_STATS, defined as _IOR(MON_IOC_MAGIC, 3, struct mon_bin_stats)
 257The argument is a pointer to the following structure:
 259struct mon_bin_stats {
 260        u32 queued;
 261        u32 dropped;
 264The member "queued" refers to the number of events currently queued in the
 265buffer (and not to the number of events processed since the last reset).
 267The member "dropped" is the number of events lost since the last call
 270 MON_IOCT_RING_SIZE, defined as _IO(MON_IOC_MAGIC, 4)
 272This call sets the buffer size. The argument is the size in bytes.
 273The size may be rounded down to the next chunk (or page). If the requested
 274size is out of [unspecified] bounds for this kernel, the call fails with
 277 MON_IOCQ_RING_SIZE, defined as _IO(MON_IOC_MAGIC, 5)
 279This call returns the current size of the buffer in bytes.
 281 MON_IOCX_GET, defined as _IOW(MON_IOC_MAGIC, 6, struct mon_get_arg)
 283This call waits for events to arrive if none were in the kernel buffer,
 284then returns the first event. Its argument is a pointer to the following
 287struct mon_get_arg {
 288        struct usbmon_packet *hdr;
 289        void *data;
 290        size_t alloc;           /* Length of data (can be zero) */
 293Before the call, hdr, data, and alloc should be filled. Upon return, the area
 294pointed by hdr contains the next event structure, and the data buffer contains
 295the data, if any. The event is removed from the kernel buffer.
 297 MON_IOCX_MFETCH, defined as _IOWR(MON_IOC_MAGIC, 7, struct mon_mfetch_arg)
 299This ioctl is primarily used when the application accesses the buffer
 300with mmap(2). Its argument is a pointer to the following structure:
 302struct mon_mfetch_arg {
 303        uint32_t *offvec;       /* Vector of events fetched */
 304        uint32_t nfetch;        /* Number of events to fetch (out: fetched) */
 305        uint32_t nflush;        /* Number of events to flush */
 308The ioctl operates in 3 stages.
 310First, it removes and discards up to nflush events from the kernel buffer.
 311The actual number of events discarded is returned in nflush.
 313Second, it waits for an event to be present in the buffer, unless the pseudo-
 314device is open with O_NONBLOCK.
 316Third, it extracts up to nfetch offsets into the mmap buffer, and stores
 317them into the offvec. The actual number of event offsets is stored into
 318the nfetch.
 320 MON_IOCH_MFLUSH, defined as _IO(MON_IOC_MAGIC, 8)
 322This call removes a number of events from the kernel buffer. Its argument
 323is the number of events to remove. If the buffer contains fewer events
 324than requested, all events present are removed, and no error is reported.
 325This works when no events are available too.
 329The ioctl FIONBIO may be implemented in the future, if there's a need.
 331In addition to ioctl(2) and read(2), the special file of binary API can
 332be polled with select(2) and poll(2). But lseek(2) does not work.
 334* Memory-mapped access of the kernel buffer for the binary API
 336The basic idea is simple:
 338To prepare, map the buffer by getting the current size, then using mmap(2).
 339Then, execute a loop similar to the one written in pseudo-code below:
 341   struct mon_mfetch_arg fetch;
 342   struct usbmon_packet *hdr;
 343   int nflush = 0;
 344   for (;;) {
 345      fetch.offvec = vec; // Has N 32-bit words
 346      fetch.nfetch = N;   // Or less than N
 347      fetch.nflush = nflush;
 348      ioctl(fd, MON_IOCX_MFETCH, &fetch);   // Process errors, too
 349      nflush = fetch.nfetch;       // This many packets to flush when done
 350      for (i = 0; i < nflush; i++) {
 351         hdr = (struct ubsmon_packet *) &mmap_area[vec[i]];
 352         if (hdr->type == '@')     // Filler packet
 353            continue;
 354         caddr_t data = &mmap_area[vec[i]] + 64;
 355         process_packet(hdr, data);
 356      }
 357   }
 359Thus, the main idea is to execute only one ioctl per N events.
 361Although the buffer is circular, the returned headers and data do not cross
 362the end of the buffer, so the above pseudo-code does not need any gathering.