Re: [PATCH v4 10/10] user_events: Add documentation file

From: Jonathan Corbet
Date: Thu Nov 04 2021 - 15:06:01 EST


Beau Belgrave <beaub@xxxxxxxxxxxxxxxxxxx> writes:

> Add a documentation file about user_events with example code, etc.
> explaining how it may be used.

Yay documentation! Some nits for a slow moment...

> Signed-off-by: Beau Belgrave <beaub@xxxxxxxxxxxxxxxxxxx>
> ---
> Documentation/trace/user_events.rst | 298 ++++++++++++++++++++++++++++
> 1 file changed, 298 insertions(+)
> create mode 100644 Documentation/trace/user_events.rst

Actually, this isn't really a nit. When you add a new RST file, you
need to add it to the index.rst file too so that it gets pulled into the
docs build. Otherwise you'll get the warning you doubtless saw when you
tried building the docs after adding this file...:)

> diff --git a/Documentation/trace/user_events.rst b/Documentation/trace/user_events.rst
> new file mode 100644
> index 000000000000..d79c9f07d012
> --- /dev/null
> +++ b/Documentation/trace/user_events.rst
> @@ -0,0 +1,298 @@
> +=========================================
> +user_events: User-based Event Tracing
> +=========================================
> +
> +:Author: Beau Belgrave
> +
> +Overview
> +--------
> +User based trace events allow user processes to create events and trace data
> +that can be viewed via existing tools, such as ftrace, perf and eBPF.
> +To enable this feature, build your kernel with CONFIG_USER_EVENTS=y.
> +
> +Programs can view status of the events via
> +/sys/kernel/debug/tracing/user_events_status and can both register and write
> +data out via /sys/kernel/debug/tracing/user_events_data.
> +
> +Programs can also use /sys/kernel/debug/tracing/dynamic_events to register and
> +delete user based events via the u: prefix. The format of the command to
> +dynamic_events is the same as the ioctl with the u: prefix applied.
> +
> +Typically programs will register a set of events that they wish to expose to
> +tools that can read trace_events (such as ftrace and perf). The registration
> +process gives back two ints to the program for each event. The first int is the
> +status index. This index describes which byte in the
> +/sys/kernel/debug/tracing/user_events_status file represents this event. The
> +second int is the write index. This index describes the data when a write() or
> +writev() is called on the /sys/kernel/debug/tracing/user_events_data file.
> +
> +The structures referenced in this document are contained with the
> +/include/uap/linux/user_events.h file in the source tree.
> +
> +**NOTE:** *Both user_events_status and user_events_data are under the tracefs filesystem
> +and may be mounted at different paths than above.*

Best to stick with the 80-column guideline for docs, please.

This also won't render the way you expect and may add a warning.

> +Registering
> +-----------
> +Registering within a user process is done via ioctl() out to the
> +/sys/kernel/debug/tracing/user_events_data file. The command to issue is
> +DIAG_IOCSREG. This command takes a struct user_reg as an argument.
> +
> +The struct user_reg requires two values, the first is the size of the structure
> +to ensure forward and backward compatibility. The second is the command string
> +to issue for registering.
> +
> +User based events show up under tracefs like any other event under the subsystem
> +named "user_events". This means tools that wish to attach to the events need to
> +use /sys/kernel/debug/tracing/events/user_events/[name]/enable or perf record
> +-e user_events:[name] when attaching/recording.
> +
> +**NOTE:** *The write_index returned is only valid for the FD that was used*
> +
> +Command Format
> +^^^^^^^^^^^^^^
> +The command string format is as follows:
> +
> +::

You can express this more concisely (and readably) as

The command string format is as follows::

(there's a bunch of these in this file)

> + name[:FLAG1[,FLAG2...]] [Field1[;Field2...]]
> +
> +Supported Flags
> +^^^^^^^^^^^^^^^
> +**BPF_ITER** - EBPF programs attached to this event will get the raw iovec
> +struct instead of any data copies for max performance.
> +
> +Field Format
> +^^^^^^^^^^^^
> +
> +::
> +
> + type name [size]
> +
> +Basic types are supported (__data_loc, u32, u64, int, char, char[20]).
> +User programs are encouraged to use clearly sized types like u32.
> +
> +**NOTE:** *Long is not supported since size can vary between user and kernel.*
> +
> +The size is only valid for types that start with a struct prefix.
> +This allows user programs to describe custom structs out to tools, if required.
> +
> +For example, a struct in C that looks like this:
> +
> +::
> +
> + struct mytype {
> + char data[20];
> + };
> +
> +Would be represented by the following field:
> +
> +::
> +
> + struct mytype myname 20
> +
> +Status
> +------
> +When tools attach/record user based events the status of the event is updated
> +in realtime. This allows user programs to only incur the cost of the write() or
> +writev() calls when something is actively attached to the event.
> +
> +User programs call mmap() on /sys/kernel/debug/tracing/user_events_status to
> +check the status for each event that is registered. The byte to check in the
> +file is given back after the register ioctl() via user_reg.status_index.
> +Currently the size of user_events_status is a single page, however, custom
> +kernel configurations can change this size to allow more user based events. In
> +all cases the size of the file is a multiple of a page size.
> +
> +For example, if the register ioctl() gives back a status_index of 3 you would
> +check byte 3 of the returned mmap data to see if anything is attached to that
> +event.
> +
> +Administrators can easily check the status of all registered events by reading
> +the user_events_status file directly via a terminal. The output is as follows:
> +
> +::
> +
> + Byte:Name [# Comments]
> + ...
> +
> + Active: ActiveCount
> + Buisy: BusyCount
> + Max: MaxCount
> +
> +For example, on a system that has a single event the output looks like this:
> +
> +::
> +
> + 1:test
> +
> + Active: 1
> + Busy: 0
> + Max: 4096
> +
> +If a user enables the user event via ftrace, the output would change to this:
> +
> +::
> +
> + 1:test # Used by ftrace
> +
> + Active: 1
> + Busy: 1
> + Max: 4096
> +
> +**NOTE:** *A status index of 0 will never be returned. This allows user
> +programs to have an index that can be used on error cases.*
> +
> +Status Bits
> +^^^^^^^^^^^
> +The byte being checked will be non-zero if anything is attached. Programs can
> +check specific bits in the byte to see what mechanism has been attached.
> +
> +The following values are defined to aid in checking what has been attached:
> +**EVENT_STATUS_FTRACE** - Bit set if ftrace has been attached (Bit 0).
> +
> +**EVENT_STATUS_PERF** - Bit set if perf/eBPF has been attached (Bit 1).
> +
> +Writing Data
> +------------
> +After registering an event the same fd that was used to register can be used
> +to write an entry for that event. The write_index returned must be at the start
> +of the data, then the remaining data is treated as the payload of the event.
> +
> +For example, if write_index returned was 1 and I wanted to write out an int
> +payload of the event. Then the data would have to be 8 bytes (2 ints) long,
> +with the first 4 bytes being equal to 1 and the last 4 bytes being equal to the
> +value I want as the payload.
> +
> +In memory this would look like this:
> +
> +::
> +
> + int index;
> + int payload;
> +
> +User programs might have well known structs that they wish to use to emit out
> +as payloads. In those cases writev() can be used, with the first vector being
> +the index and the following vector(s) being the actual event payload.
> +
> +For example, if I have a struct like this:
> +
> +::
> +
> + struct payload {
> + int src;
> + int dst;
> + int flags;
> + };
> +
> +It's advised for user programs to do the following:
> +
> +::
> +
> + struct iovec io[2];
> + struct payload e;
> +
> + io[0].iov_base = &write_index;
> + io[0].iov_len = sizeof(write_index);
> + io[1].iov_base = &e;
> + io[1].iov_len = sizeof(e);
> +
> + writev(fd, (const struct iovec*)io, 2);
> +
> +**NOTE:** *The write_index is not emitted out into the trace being recorded.*
> +
> +EBPF
> +----
> +EBPF programs that attach to a user-based event tracepoint are given a pointer
> +to a struct user_bpf_context. The bpf context contains the data type (which can
> +be a user or kernel buffer, or can be a pointer to the iovec) and the data
> +length that was emitted (minus the write_index).
> +
> +Example Code
> +------------

Examples are great, but they are best placed under samples/ (or tools/
if they do something useful) where readers can build and run them.

> +::
> +
> + #include <errno.h>
> + #include <sys/ioctl.h>
> + #include <sys/mman.h>
> + #include <fcntl.h>
> + #include <stdio.h>
> + #include <unistd.h>
> + #include <linux/user_events.h>
> +
> + /* Assumes debugfs is mounted */
> + const char *data_file = "/sys/kernel/debug/tracing/user_events_data";
> + const char *status_file = "/sys/kernel/debug/tracing/user_events_status";
> +
> + static int event_status(char **status)
> + {
> + int fd = open(status_file, O_RDONLY);
> +
> + *status = mmap(NULL, sysconf(_SC_PAGESIZE), PROT_READ,
> + MAP_SHARED, fd, 0);
> +
> + close(fd);
> +
> + if (*status == MAP_FAILED)
> + return -1;
> +
> + return 0;
> + }
> +
> + static int event_reg(int fd, const char *command, int *status, int *write)
> + {
> + struct user_reg reg = {0};
> +
> + reg.size = sizeof(reg);
> + reg.name_args = (__u64)command;
> +
> + if (ioctl(fd, DIAG_IOCSREG, &reg) == -1)
> + return -1;
> +
> + *status = reg.status_index;
> + *write = reg.write_index;
> +
> + return 0;
> + }
> +
> + int main(int argc, char **argv)
> + {
> + int data_fd, status, write;
> + char *status_page;
> + struct iovec io[2];
> + __u32 count = 0;
> +
> + if (event_status(&status_page) == -1)
> + return errno;
> +
> + data_fd = open(data_file, O_RDWR);
> +
> + if (event_reg(data_fd, "test u32 count", &status, &write) == -1)
> + return errno;
> +
> + /* Setup iovec */
> + io[0].iov_base = &status;
> + io[0].iov_len = sizeof(status);
> + io[1].iov_base = &count;
> + io[1].iov_len = sizeof(count);
> +
> + ask:
> + printf("Press enter to check status...\n");
> + getchar();
> +
> + /* Check if anyone is listening */
> + if (status_page[status]) {
> + /* Yep, trace out our data */
> + writev(data_fd, (const struct iovec*)io, 2);
> +
> + /* Increase the count */
> + count++;
> +
> + printf("Something was attached, wrote data\n");
> + }
> +
> + goto ask;
> +
> + return 0;
> + }

Thanks,

jon