[2/4] tracing/user_events: Introduce multi-format events
Commit Message
Currently user_events supports 1 event with the same name and must have
the exact same format when referenced by multiple programs. This opens
an opportunity for malicous or poorly thought through programs to
create events that others use with different formats. Another scenario
is user programs wishing to use the same event name but add more fields
later when the software updates. Various versions of a program may be
running side-by-side, which is prevented by the current single format
requirement.
Add a new register flag (USER_EVENT_REG_MULTI_FORMAT) which indicates
the user program wishes to use the same user_event name, but may have
several different formats of the event in the future. When this flag is
used, create the underlying tracepoint backing the user_event with a
unique name per-version of the format. It's important that existing ABI
users do not get this logic automatically, even if one of the multi
format events matches the format. This ensures existing programs that
create events and assume the tracepoint name will match exactly continue
to work as expected. Add logic to only check multi-format events with
other multi-format events and single-format events to only check
single-format events during find.
Change system name of the multi-format event tracepoint to ensure that
multi-format events are isolated completely from single-format events.
Add a register_name (reg_name) to the user_event struct which allows for
split naming of events. We now have the name that was used to register
within user_events as well as the unique name for the tracepoint. Upon
registering events ensure matches based on first the reg_name, followed
by the fields and format of the event. This allows for multiple events
with the same registered name to have different formats. The underlying
tracepoint will have a unique name in the format of {reg_name}:[unique_id].
For example, if both "test u32 value" and "test u64 value" are used with
the USER_EVENT_REG_MULTI_FORMAT the system would have 2 unique
tracepoints. The dynamic_events file would then show the following:
u:test u64 count
u:test u32 count
The actual tracepoint names look like this:
test:[d5874fdac44]
test:[d5914662cd4]
Both would be under the new user_events_multi system name to prevent the
older ABI from being used to squat on multi-formatted events and block
their use.
Deleting events via "!u:test u64 count" would only delete the first
tracepoint that matched that format. When the delete ABI is used all
events with the same name will be attempted to be deleted. If
per-version deletion is required, user programs should either not use
persistent events or delete them via dynamic_events.
Signed-off-by: Beau Belgrave <beaub@linux.microsoft.com>
---
include/uapi/linux/user_events.h | 6 +-
kernel/trace/trace_events_user.c | 118 +++++++++++++++++++++++++++----
2 files changed, 111 insertions(+), 13 deletions(-)
Comments
On Tue, 23 Jan 2024 22:08:42 +0000
Beau Belgrave <beaub@linux.microsoft.com> wrote:
> Add a register_name (reg_name) to the user_event struct which allows for
> split naming of events. We now have the name that was used to register
> within user_events as well as the unique name for the tracepoint. Upon
> registering events ensure matches based on first the reg_name, followed
> by the fields and format of the event. This allows for multiple events
> with the same registered name to have different formats. The underlying
> tracepoint will have a unique name in the format of {reg_name}:[unique_id].
>
> For example, if both "test u32 value" and "test u64 value" are used with
> the USER_EVENT_REG_MULTI_FORMAT the system would have 2 unique
> tracepoints. The dynamic_events file would then show the following:
> u:test u64 count
> u:test u32 count
>
> The actual tracepoint names look like this:
> test:[d5874fdac44]
> test:[d5914662cd4]
>
> Both would be under the new user_events_multi system name to prevent the
> older ABI from being used to squat on multi-formatted events and block
> their use.
[...]
> @@ -1923,6 +1972,39 @@ static int user_event_trace_register(struct user_event *user)
> return ret;
> }
>
> +static int user_event_set_tp_name(struct user_event *user)
> +{
> + lockdep_assert_held(&user->group->reg_mutex);
> +
> + if (EVENT_MULTI_FORMAT(user->reg_flags)) {
> + char *multi_name;
> + int len;
> +
> + len = snprintf(NULL, 0, "%s:[%llx]", user->reg_name,
> + user->group->multi_id) + 1;
> +
> + multi_name = kzalloc(len, GFP_KERNEL_ACCOUNT);
> +
> + if (!multi_name)
> + return -ENOMEM;
> +
> + snprintf(multi_name, len, "%s:[%llx]", user->reg_name,
> + user->group->multi_id);
OK, so the each different event has suffixed name. But this will
introduce non C-variable name.
Steve, do you think your library can handle these symbols? It will
be something like "event:[1]" as the event name.
Personally I like "event.1" style. (of course we need to ensure the
user given event name is NOT including such suffix numbers)
Thank you.
On Sat, Jan 27, 2024 at 12:01:04AM +0900, Masami Hiramatsu wrote:
> On Tue, 23 Jan 2024 22:08:42 +0000
> Beau Belgrave <beaub@linux.microsoft.com> wrote:
>
> > Add a register_name (reg_name) to the user_event struct which allows for
> > split naming of events. We now have the name that was used to register
> > within user_events as well as the unique name for the tracepoint. Upon
> > registering events ensure matches based on first the reg_name, followed
> > by the fields and format of the event. This allows for multiple events
> > with the same registered name to have different formats. The underlying
> > tracepoint will have a unique name in the format of {reg_name}:[unique_id].
> >
> > For example, if both "test u32 value" and "test u64 value" are used with
> > the USER_EVENT_REG_MULTI_FORMAT the system would have 2 unique
> > tracepoints. The dynamic_events file would then show the following:
> > u:test u64 count
> > u:test u32 count
> >
> > The actual tracepoint names look like this:
> > test:[d5874fdac44]
> > test:[d5914662cd4]
> >
> > Both would be under the new user_events_multi system name to prevent the
> > older ABI from being used to squat on multi-formatted events and block
> > their use.
> [...]
> > @@ -1923,6 +1972,39 @@ static int user_event_trace_register(struct user_event *user)
> > return ret;
> > }
> >
> > +static int user_event_set_tp_name(struct user_event *user)
> > +{
> > + lockdep_assert_held(&user->group->reg_mutex);
> > +
> > + if (EVENT_MULTI_FORMAT(user->reg_flags)) {
> > + char *multi_name;
> > + int len;
> > +
> > + len = snprintf(NULL, 0, "%s:[%llx]", user->reg_name,
> > + user->group->multi_id) + 1;
> > +
> > + multi_name = kzalloc(len, GFP_KERNEL_ACCOUNT);
> > +
> > + if (!multi_name)
> > + return -ENOMEM;
> > +
> > + snprintf(multi_name, len, "%s:[%llx]", user->reg_name,
> > + user->group->multi_id);
>
> OK, so the each different event has suffixed name. But this will
> introduce non C-variable name.
>
> Steve, do you think your library can handle these symbols? It will
> be something like "event:[1]" as the event name.
> Personally I like "event.1" style. (of course we need to ensure the
> user given event name is NOT including such suffix numbers)
>
Just to clarify around events including a suffix number. This is why
multi-events use "user_events_multi" system name and the single-events
using just "user_events".
Even if a user program did include a suffix, the suffix would still get
appended. An example is "test" vs "test:[0]" using multi-format would
result in two tracepoints ("test:[0]" and "test:[0]:[1]" respectively
(assuming these are the first multi-events on the system).
I'm with you, we really don't want any spoofing or squatting possible.
By using different system names and always appending the suffix I
believe covers this.
Looking forward to hearing Steven's thoughts on this as well.
Thanks,
-Beau
> Thank you.
>
> --
> Masami Hiramatsu (Google) <mhiramat@kernel.org>
On Fri, 26 Jan 2024 11:10:07 -0800
Beau Belgrave <beaub@linux.microsoft.com> wrote:
> > OK, so the each different event has suffixed name. But this will
> > introduce non C-variable name.
> >
> > Steve, do you think your library can handle these symbols? It will
> > be something like "event:[1]" as the event name.
> > Personally I like "event.1" style. (of course we need to ensure the
> > user given event name is NOT including such suffix numbers)
> >
>
> Just to clarify around events including a suffix number. This is why
> multi-events use "user_events_multi" system name and the single-events
> using just "user_events".
>
> Even if a user program did include a suffix, the suffix would still get
> appended. An example is "test" vs "test:[0]" using multi-format would
> result in two tracepoints ("test:[0]" and "test:[0]:[1]" respectively
> (assuming these are the first multi-events on the system).
>
> I'm with you, we really don't want any spoofing or squatting possible.
> By using different system names and always appending the suffix I
> believe covers this.
>
> Looking forward to hearing Steven's thoughts on this as well.
I'm leaning towards Masami's suggestion to use dots, as that won't conflict
with special characters from bash, as '[' and ']' do.
-- Steve
On Fri, Jan 26, 2024 at 03:04:45PM -0500, Steven Rostedt wrote:
> On Fri, 26 Jan 2024 11:10:07 -0800
> Beau Belgrave <beaub@linux.microsoft.com> wrote:
>
> > > OK, so the each different event has suffixed name. But this will
> > > introduce non C-variable name.
> > >
> > > Steve, do you think your library can handle these symbols? It will
> > > be something like "event:[1]" as the event name.
> > > Personally I like "event.1" style. (of course we need to ensure the
> > > user given event name is NOT including such suffix numbers)
> > >
> >
> > Just to clarify around events including a suffix number. This is why
> > multi-events use "user_events_multi" system name and the single-events
> > using just "user_events".
> >
> > Even if a user program did include a suffix, the suffix would still get
> > appended. An example is "test" vs "test:[0]" using multi-format would
> > result in two tracepoints ("test:[0]" and "test:[0]:[1]" respectively
> > (assuming these are the first multi-events on the system).
> >
> > I'm with you, we really don't want any spoofing or squatting possible.
> > By using different system names and always appending the suffix I
> > believe covers this.
> >
> > Looking forward to hearing Steven's thoughts on this as well.
>
> I'm leaning towards Masami's suggestion to use dots, as that won't conflict
> with special characters from bash, as '[' and ']' do.
>
Thanks, yeah ideally we wouldn't use special characters.
I'm not picky about this. However, I did want something that clearly
allowed a glob pattern to find all versions of a given register name of
user_events by user programs that record. The dot notation will pull in
more than expected if dotted namespace style names are used.
An example is "Asserts" and "Asserts.Verbose" from different programs.
If we tried to find all versions of "Asserts" via glob of "Asserts.*" it
will pull in "Asserts.Verbose.1" in addition to "Asserts.0".
While a glob of "Asserts.[0-9]" works when the unique ID is 0-9, it
doesn't work if the number is higher, like 128. If we ever decide to
change the ID from an integer to say hex to save space, these globs
would break.
Is there some scheme that fits the C-variable name that addresses the
above scenarios? Brackets gave me a simple glob that seemed to prevent a
lot of this ("Asserts.\[*\]" in this case).
Are we confident that we always want to represent the ID as a base-10
integer vs a base-16 integer? The suffix will be ABI to ensure recording
programs can find their events easily.
Thanks,
-Beau
> -- Steve
On Mon, 29 Jan 2024 09:29:07 -0800
Beau Belgrave <beaub@linux.microsoft.com> wrote:
> Thanks, yeah ideally we wouldn't use special characters.
>
> I'm not picky about this. However, I did want something that clearly
> allowed a glob pattern to find all versions of a given register name of
> user_events by user programs that record. The dot notation will pull in
> more than expected if dotted namespace style names are used.
>
> An example is "Asserts" and "Asserts.Verbose" from different programs.
> If we tried to find all versions of "Asserts" via glob of "Asserts.*" it
> will pull in "Asserts.Verbose.1" in addition to "Asserts.0".
Do you prevent brackets in names?
>
> While a glob of "Asserts.[0-9]" works when the unique ID is 0-9, it
> doesn't work if the number is higher, like 128. If we ever decide to
> change the ID from an integer to say hex to save space, these globs
> would break.
>
> Is there some scheme that fits the C-variable name that addresses the
> above scenarios? Brackets gave me a simple glob that seemed to prevent a
> lot of this ("Asserts.\[*\]" in this case).
Prevent a lot of what? I'm not sure what your example here is.
>
> Are we confident that we always want to represent the ID as a base-10
> integer vs a base-16 integer? The suffix will be ABI to ensure recording
> programs can find their events easily.
Is there a difference to what we choose?
-- Steve
On Mon, 29 Jan 2024 09:29:07 -0800
Beau Belgrave <beaub@linux.microsoft.com> wrote:
> On Fri, Jan 26, 2024 at 03:04:45PM -0500, Steven Rostedt wrote:
> > On Fri, 26 Jan 2024 11:10:07 -0800
> > Beau Belgrave <beaub@linux.microsoft.com> wrote:
> >
> > > > OK, so the each different event has suffixed name. But this will
> > > > introduce non C-variable name.
> > > >
> > > > Steve, do you think your library can handle these symbols? It will
> > > > be something like "event:[1]" as the event name.
> > > > Personally I like "event.1" style. (of course we need to ensure the
> > > > user given event name is NOT including such suffix numbers)
> > > >
> > >
> > > Just to clarify around events including a suffix number. This is why
> > > multi-events use "user_events_multi" system name and the single-events
> > > using just "user_events".
> > >
> > > Even if a user program did include a suffix, the suffix would still get
> > > appended. An example is "test" vs "test:[0]" using multi-format would
> > > result in two tracepoints ("test:[0]" and "test:[0]:[1]" respectively
> > > (assuming these are the first multi-events on the system).
> > >
> > > I'm with you, we really don't want any spoofing or squatting possible.
> > > By using different system names and always appending the suffix I
> > > believe covers this.
> > >
> > > Looking forward to hearing Steven's thoughts on this as well.
> >
> > I'm leaning towards Masami's suggestion to use dots, as that won't conflict
> > with special characters from bash, as '[' and ']' do.
> >
>
> Thanks, yeah ideally we wouldn't use special characters.
>
> I'm not picky about this. However, I did want something that clearly
> allowed a glob pattern to find all versions of a given register name of
> user_events by user programs that record. The dot notation will pull in
> more than expected if dotted namespace style names are used.
>
> An example is "Asserts" and "Asserts.Verbose" from different programs.
> If we tried to find all versions of "Asserts" via glob of "Asserts.*" it
> will pull in "Asserts.Verbose.1" in addition to "Asserts.0".
If we use dot for the suffix number, we can prohibit user to use it
for their name. They still can use '_' (or change the group name?)
I just concerned that the name can be parsed by existing tools. Since
':' is used as a separator for group and event name in some case (e.g.
tracefs "set_event" is using, so trace-cmd and perf is using it.)
> While a glob of "Asserts.[0-9]" works when the unique ID is 0-9, it
> doesn't work if the number is higher, like 128. If we ever decide to
> change the ID from an integer to say hex to save space, these globs
> would break.
Hmm, why can't we use regexp?
And if we limits the number of events up to 1000 for each same-name event
we can use fixed numbers, like Assets.[0-9][0-9][0-9]
Thank you,
>
> Is there some scheme that fits the C-variable name that addresses the
> above scenarios? Brackets gave me a simple glob that seemed to prevent a
> lot of this ("Asserts.\[*\]" in this case).
>
> Are we confident that we always want to represent the ID as a base-10
> integer vs a base-16 integer? The suffix will be ABI to ensure recording
> programs can find their events easily.
>
> Thanks,
> -Beau
>
> > -- Steve
On Mon, Jan 29, 2024 at 09:24:07PM -0500, Steven Rostedt wrote:
> On Mon, 29 Jan 2024 09:29:07 -0800
> Beau Belgrave <beaub@linux.microsoft.com> wrote:
>
> > Thanks, yeah ideally we wouldn't use special characters.
> >
> > I'm not picky about this. However, I did want something that clearly
> > allowed a glob pattern to find all versions of a given register name of
> > user_events by user programs that record. The dot notation will pull in
> > more than expected if dotted namespace style names are used.
> >
> > An example is "Asserts" and "Asserts.Verbose" from different programs.
> > If we tried to find all versions of "Asserts" via glob of "Asserts.*" it
> > will pull in "Asserts.Verbose.1" in addition to "Asserts.0".
>
> Do you prevent brackets in names?
>
No. However, since brackets have a start and end token that are distinct
finding all versions of your event is trivial compared to a single dot.
Imagine two events:
Asserts
Asserts[MyCoolIndex]
Resolves to tracepoints of:
Asserts:[0]
Asserts[MyCoolIndex]:[1]
Regardless of brackets in the names, a simple glob of Asserts:\[*\] only
finds Asserts:[0]. This is because we have that end bracket in the glob
and the full event name including the start bracket.
If I register another "version" of Asserts, thne I'll have:
Asserts:[0]
Asserts[MyCoolIndex]:[1]
Asserts:[2]
The glob of Asserts:\[*\] will return both:
Asserts:[0]
Asserts:[2]
At this point the program can either record all versions or scan further
to find which version of Asserts is wanted.
> >
> > While a glob of "Asserts.[0-9]" works when the unique ID is 0-9, it
> > doesn't work if the number is higher, like 128. If we ever decide to
> > change the ID from an integer to say hex to save space, these globs
> > would break.
> >
> > Is there some scheme that fits the C-variable name that addresses the
> > above scenarios? Brackets gave me a simple glob that seemed to prevent a
> > lot of this ("Asserts.\[*\]" in this case).
>
> Prevent a lot of what? I'm not sure what your example here is.
>
I'll try again :)
We have 2 events registered via user_events:
Asserts
Asserts.Verbose
Using dot notation these would result in tracepoints of:
user_events_multi/Asserts.0
user_events_multi/Asserts.Verbose.1
Using bracket notation these would result in tracepoints of:
user_events_multi/Asserts:[0]
user_events_multi/Asserts.Verbose:[1]
A recording program only wants to enable the Asserts tracepoint. It does
not want to record the Asserts.Verbose tracepoint.
The program must find the right tracepoint by scanning tracefs under the
user_events_multi system.
A single dot suffix does not allow a simple glob to be used. The glob
Asserts.* will return both Asserts.0 and Asserts.Verbose.1.
A simple glob of Asserts:\[*\] will only find Asserts:[0], it will not
find Asserts.Verbose:[1].
We could just use brackets and not have the colon (Asserts[0] in this
case). But brackets are still special for bash.
> >
> > Are we confident that we always want to represent the ID as a base-10
> > integer vs a base-16 integer? The suffix will be ABI to ensure recording
> > programs can find their events easily.
>
> Is there a difference to what we choose?
>
If a simple glob of event_name:\[*\] cannot be used, then we must document
what the suffix format is, so an appropriate regex can be created. If we
start with base-10 then later move to base-16 we will break existing regex
patterns on the recording side.
I prefer, and have in this series, a base-16 output since it saves on
the tracepoint name size.
Either way we go, we need to define how recording programs should find
the events they care about. So we must be very clear, IMHO, about the
format of the tracepoint names in our documentation.
I personally think recording programs are likely to get this wrong
without proper guidance.
Thanks,
-Beau
> -- Steve
On Tue, Jan 30, 2024 at 11:12:22PM +0900, Masami Hiramatsu wrote:
> On Mon, 29 Jan 2024 09:29:07 -0800
> Beau Belgrave <beaub@linux.microsoft.com> wrote:
>
> > On Fri, Jan 26, 2024 at 03:04:45PM -0500, Steven Rostedt wrote:
> > > On Fri, 26 Jan 2024 11:10:07 -0800
> > > Beau Belgrave <beaub@linux.microsoft.com> wrote:
> > >
> > > > > OK, so the each different event has suffixed name. But this will
> > > > > introduce non C-variable name.
> > > > >
> > > > > Steve, do you think your library can handle these symbols? It will
> > > > > be something like "event:[1]" as the event name.
> > > > > Personally I like "event.1" style. (of course we need to ensure the
> > > > > user given event name is NOT including such suffix numbers)
> > > > >
> > > >
> > > > Just to clarify around events including a suffix number. This is why
> > > > multi-events use "user_events_multi" system name and the single-events
> > > > using just "user_events".
> > > >
> > > > Even if a user program did include a suffix, the suffix would still get
> > > > appended. An example is "test" vs "test:[0]" using multi-format would
> > > > result in two tracepoints ("test:[0]" and "test:[0]:[1]" respectively
> > > > (assuming these are the first multi-events on the system).
> > > >
> > > > I'm with you, we really don't want any spoofing or squatting possible.
> > > > By using different system names and always appending the suffix I
> > > > believe covers this.
> > > >
> > > > Looking forward to hearing Steven's thoughts on this as well.
> > >
> > > I'm leaning towards Masami's suggestion to use dots, as that won't conflict
> > > with special characters from bash, as '[' and ']' do.
> > >
> >
> > Thanks, yeah ideally we wouldn't use special characters.
> >
> > I'm not picky about this. However, I did want something that clearly
> > allowed a glob pattern to find all versions of a given register name of
> > user_events by user programs that record. The dot notation will pull in
> > more than expected if dotted namespace style names are used.
> >
> > An example is "Asserts" and "Asserts.Verbose" from different programs.
> > If we tried to find all versions of "Asserts" via glob of "Asserts.*" it
> > will pull in "Asserts.Verbose.1" in addition to "Asserts.0".
>
> If we use dot for the suffix number, we can prohibit user to use it
> for their name. They still can use '_' (or change the group name?)
We could, however, we have user_event integration in OpenTelemetry and
I'm unsure if we should really try to restrict names. We'll also at some
point have libside integration, which might not have the same
restrictions on the user-tracer side as the kernel-tracer side.
I'm trying to restrict the user_event group name from changing outside
of an eventual tracer namespace. I'd like for each container to inherit
a tracer namespace long-term which decides what the actual group name
will be instead of users self-selecting names to prevent squatting or
spoofing of events.
> I just concerned that the name can be parsed by existing tools. Since
> ':' is used as a separator for group and event name in some case (e.g.
> tracefs "set_event" is using, so trace-cmd and perf is using it.)
>
Good point.
What about just event_name[unique_id]? IE: Drop the colon.
Brackets are still special in bash, but it would prevent simple glob
patterns from matching to incorrect tracepoints under user_events_multi.
> > While a glob of "Asserts.[0-9]" works when the unique ID is 0-9, it
> > doesn't work if the number is higher, like 128. If we ever decide to
> > change the ID from an integer to say hex to save space, these globs
> > would break.
>
> Hmm, why can't we use regexp?
We can use regex, but we'll need to agree the suffix format. We won't be
able to change it after that point. I'd prefer a base-16/Hex suffix
either in brackets or a simple dot.
> And if we limits the number of events up to 1000 for each same-name event
> we can use fixed numbers, like Assets.[0-9][0-9][0-9]
>
I'm always wrong when I guess how many events programs will end up
using. Folks always surprise me. I'd rather have a solution that scales
to the max number of tracepoints allowed on the system (currently 16-bit
max value).
Thanks,
-Beau
> Thank you,
>
> >
> > Is there some scheme that fits the C-variable name that addresses the
> > above scenarios? Brackets gave me a simple glob that seemed to prevent a
> > lot of this ("Asserts.\[*\]" in this case).
> >
> > Are we confident that we always want to represent the ID as a base-10
> > integer vs a base-16 integer? The suffix will be ABI to ensure recording
> > programs can find their events easily.
> >
> > Thanks,
> > -Beau
> >
> > > -- Steve
>
>
> --
> Masami Hiramatsu (Google) <mhiramat@kernel.org>
On Tue, 30 Jan 2024 10:05:15 -0800
Beau Belgrave <beaub@linux.microsoft.com> wrote:
> On Mon, Jan 29, 2024 at 09:24:07PM -0500, Steven Rostedt wrote:
> > On Mon, 29 Jan 2024 09:29:07 -0800
> > Beau Belgrave <beaub@linux.microsoft.com> wrote:
> >
> > > Thanks, yeah ideally we wouldn't use special characters.
> > >
> > > I'm not picky about this. However, I did want something that clearly
> > > allowed a glob pattern to find all versions of a given register name of
> > > user_events by user programs that record. The dot notation will pull in
> > > more than expected if dotted namespace style names are used.
> > >
> > > An example is "Asserts" and "Asserts.Verbose" from different programs.
> > > If we tried to find all versions of "Asserts" via glob of "Asserts.*" it
> > > will pull in "Asserts.Verbose.1" in addition to "Asserts.0".
> >
> > Do you prevent brackets in names?
> >
>
> No. However, since brackets have a start and end token that are distinct
> finding all versions of your event is trivial compared to a single dot.
>
> Imagine two events:
> Asserts
> Asserts[MyCoolIndex]
>
> Resolves to tracepoints of:
> Asserts:[0]
> Asserts[MyCoolIndex]:[1]
>
> Regardless of brackets in the names, a simple glob of Asserts:\[*\] only
> finds Asserts:[0]. This is because we have that end bracket in the glob
> and the full event name including the start bracket.
>
> If I register another "version" of Asserts, thne I'll have:
> Asserts:[0]
> Asserts[MyCoolIndex]:[1]
> Asserts:[2]
>
> The glob of Asserts:\[*\] will return both:
> Asserts:[0]
> Asserts:[2]
But what if you had registered "Asserts:[MyCoolIndex]:[1]"
Do you prevent colons?
>
> At this point the program can either record all versions or scan further
> to find which version of Asserts is wanted.
>
> > >
> > > While a glob of "Asserts.[0-9]" works when the unique ID is 0-9, it
> > > doesn't work if the number is higher, like 128. If we ever decide to
> > > change the ID from an integer to say hex to save space, these globs
> > > would break.
> > >
> > > Is there some scheme that fits the C-variable name that addresses the
> > > above scenarios? Brackets gave me a simple glob that seemed to prevent a
> > > lot of this ("Asserts.\[*\]" in this case).
> >
> > Prevent a lot of what? I'm not sure what your example here is.
> >
>
> I'll try again :)
>
> We have 2 events registered via user_events:
> Asserts
> Asserts.Verbose
>
> Using dot notation these would result in tracepoints of:
> user_events_multi/Asserts.0
> user_events_multi/Asserts.Verbose.1
>
> Using bracket notation these would result in tracepoints of:
> user_events_multi/Asserts:[0]
> user_events_multi/Asserts.Verbose:[1]
>
> A recording program only wants to enable the Asserts tracepoint. It does
> not want to record the Asserts.Verbose tracepoint.
>
> The program must find the right tracepoint by scanning tracefs under the
> user_events_multi system.
>
> A single dot suffix does not allow a simple glob to be used. The glob
> Asserts.* will return both Asserts.0 and Asserts.Verbose.1.
>
> A simple glob of Asserts:\[*\] will only find Asserts:[0], it will not
> find Asserts.Verbose:[1].
>
> We could just use brackets and not have the colon (Asserts[0] in this
> case). But brackets are still special for bash.
Are these shell scripts or programs. I use regex in programs all the time.
And if you have shell scripts, use awk or something.
Unless you prevent something from being added, I don't see the protection.
>
> > >
> > > Are we confident that we always want to represent the ID as a base-10
> > > integer vs a base-16 integer? The suffix will be ABI to ensure recording
> > > programs can find their events easily.
> >
> > Is there a difference to what we choose?
> >
>
> If a simple glob of event_name:\[*\] cannot be used, then we must document
> what the suffix format is, so an appropriate regex can be created. If we
> start with base-10 then later move to base-16 we will break existing regex
> patterns on the recording side.
>
> I prefer, and have in this series, a base-16 output since it saves on
> the tracepoint name size.
I honestly don't care which base you use. So if you want to use base 16,
I'm fine with that.
>
> Either way we go, we need to define how recording programs should find
> the events they care about. So we must be very clear, IMHO, about the
> format of the tracepoint names in our documentation.
>
> I personally think recording programs are likely to get this wrong
> without proper guidance.
>
Agreed.
-- Steve
On Tue, Jan 30, 2024 at 01:52:30PM -0500, Steven Rostedt wrote:
> On Tue, 30 Jan 2024 10:05:15 -0800
> Beau Belgrave <beaub@linux.microsoft.com> wrote:
>
> > On Mon, Jan 29, 2024 at 09:24:07PM -0500, Steven Rostedt wrote:
> > > On Mon, 29 Jan 2024 09:29:07 -0800
> > > Beau Belgrave <beaub@linux.microsoft.com> wrote:
> > >
> > > > Thanks, yeah ideally we wouldn't use special characters.
> > > >
> > > > I'm not picky about this. However, I did want something that clearly
> > > > allowed a glob pattern to find all versions of a given register name of
> > > > user_events by user programs that record. The dot notation will pull in
> > > > more than expected if dotted namespace style names are used.
> > > >
> > > > An example is "Asserts" and "Asserts.Verbose" from different programs.
> > > > If we tried to find all versions of "Asserts" via glob of "Asserts.*" it
> > > > will pull in "Asserts.Verbose.1" in addition to "Asserts.0".
> > >
> > > Do you prevent brackets in names?
> > >
> >
> > No. However, since brackets have a start and end token that are distinct
> > finding all versions of your event is trivial compared to a single dot.
> >
> > Imagine two events:
> > Asserts
> > Asserts[MyCoolIndex]
> >
> > Resolves to tracepoints of:
> > Asserts:[0]
> > Asserts[MyCoolIndex]:[1]
> >
> > Regardless of brackets in the names, a simple glob of Asserts:\[*\] only
> > finds Asserts:[0]. This is because we have that end bracket in the glob
> > and the full event name including the start bracket.
> >
> > If I register another "version" of Asserts, thne I'll have:
> > Asserts:[0]
> > Asserts[MyCoolIndex]:[1]
> > Asserts:[2]
> >
> > The glob of Asserts:\[*\] will return both:
> > Asserts:[0]
> > Asserts:[2]
>
> But what if you had registered "Asserts:[MyCoolIndex]:[1]"
>
Good point, the above would still require a regex type pattern to not
get pulled in.
> Do you prevent colons?
>
No, nothing is prevented at this point.
It seems we could either prevent certain characters to make it easier or
define a good regex that we should document.
I'm leaning toward just doing a simple suffix and documenting the regex
well.
> >
> > At this point the program can either record all versions or scan further
> > to find which version of Asserts is wanted.
> >
> > > >
> > > > While a glob of "Asserts.[0-9]" works when the unique ID is 0-9, it
> > > > doesn't work if the number is higher, like 128. If we ever decide to
> > > > change the ID from an integer to say hex to save space, these globs
> > > > would break.
> > > >
> > > > Is there some scheme that fits the C-variable name that addresses the
> > > > above scenarios? Brackets gave me a simple glob that seemed to prevent a
> > > > lot of this ("Asserts.\[*\]" in this case).
> > >
> > > Prevent a lot of what? I'm not sure what your example here is.
> > >
> >
> > I'll try again :)
> >
> > We have 2 events registered via user_events:
> > Asserts
> > Asserts.Verbose
> >
> > Using dot notation these would result in tracepoints of:
> > user_events_multi/Asserts.0
> > user_events_multi/Asserts.Verbose.1
> >
> > Using bracket notation these would result in tracepoints of:
> > user_events_multi/Asserts:[0]
> > user_events_multi/Asserts.Verbose:[1]
> >
> > A recording program only wants to enable the Asserts tracepoint. It does
> > not want to record the Asserts.Verbose tracepoint.
> >
> > The program must find the right tracepoint by scanning tracefs under the
> > user_events_multi system.
> >
> > A single dot suffix does not allow a simple glob to be used. The glob
> > Asserts.* will return both Asserts.0 and Asserts.Verbose.1.
> >
> > A simple glob of Asserts:\[*\] will only find Asserts:[0], it will not
> > find Asserts.Verbose:[1].
> >
> > We could just use brackets and not have the colon (Asserts[0] in this
> > case). But brackets are still special for bash.
>
> Are these shell scripts or programs. I use regex in programs all the time.
> And if you have shell scripts, use awk or something.
>
They could be both. In our case, it is a program.
> Unless you prevent something from being added, I don't see the protection.
>
Yeah, it just makes it way less likely. Given that, I'm starting to lean
toward just documenting the regex well and not trying to get fancy.
> >
> > > >
> > > > Are we confident that we always want to represent the ID as a base-10
> > > > integer vs a base-16 integer? The suffix will be ABI to ensure recording
> > > > programs can find their events easily.
> > >
> > > Is there a difference to what we choose?
> > >
> >
> > If a simple glob of event_name:\[*\] cannot be used, then we must document
> > what the suffix format is, so an appropriate regex can be created. If we
> > start with base-10 then later move to base-16 we will break existing regex
> > patterns on the recording side.
> >
> > I prefer, and have in this series, a base-16 output since it saves on
> > the tracepoint name size.
>
> I honestly don't care which base you use. So if you want to use base 16,
> I'm fine with that.
>
> >
> > Either way we go, we need to define how recording programs should find
> > the events they care about. So we must be very clear, IMHO, about the
> > format of the tracepoint names in our documentation.
> >
> > I personally think recording programs are likely to get this wrong
> > without proper guidance.
> >
>
> Agreed.
>
> -- Steve
Thanks,
-Beau
@@ -12,6 +12,7 @@
#include <linux/ioctl.h>
#define USER_EVENTS_SYSTEM "user_events"
+#define USER_EVENTS_MULTI_SYSTEM "user_events_multi"
#define USER_EVENTS_PREFIX "u:"
/* Create dynamic location entry within a 32-bit value */
@@ -22,8 +23,11 @@ enum user_reg_flag {
/* Event will not delete upon last reference closing */
USER_EVENT_REG_PERSIST = 1U << 0,
+ /* Event will be allowed to have multiple formats */
+ USER_EVENT_REG_MULTI_FORMAT = 1U << 1,
+
/* This value or above is currently non-ABI */
- USER_EVENT_REG_MAX = 1U << 1,
+ USER_EVENT_REG_MAX = 1U << 2,
};
/*
@@ -34,7 +34,8 @@
/* Limit how long of an event name plus args within the subsystem. */
#define MAX_EVENT_DESC 512
-#define EVENT_NAME(user_event) ((user_event)->tracepoint.name)
+#define EVENT_NAME(user_event) ((user_event)->reg_name)
+#define EVENT_TP_NAME(user_event) ((user_event)->tracepoint.name)
#define MAX_FIELD_ARRAY_SIZE 1024
/*
@@ -54,10 +55,13 @@
* allows isolation for events by various means.
*/
struct user_event_group {
- char *system_name;
- struct hlist_node node;
- struct mutex reg_mutex;
+ char *system_name;
+ char *system_multi_name;
+ struct hlist_node node;
+ struct mutex reg_mutex;
DECLARE_HASHTABLE(register_table, 8);
+ /* ID that moves forward within the group for multi-event names */
+ u64 multi_id;
};
/* Group for init_user_ns mapping, top-most group */
@@ -78,6 +82,7 @@ static unsigned int current_user_events;
*/
struct user_event {
struct user_event_group *group;
+ char *reg_name;
struct tracepoint tracepoint;
struct trace_event_call call;
struct trace_event_class class;
@@ -127,6 +132,8 @@ struct user_event_enabler {
#define ENABLE_BIT(e) ((int)((e)->values & ENABLE_VAL_BIT_MASK))
+#define EVENT_MULTI_FORMAT(f) ((f) & USER_EVENT_REG_MULTI_FORMAT)
+
/* Used for asynchronous faulting in of pages */
struct user_event_enabler_fault {
struct work_struct work;
@@ -330,6 +337,7 @@ static void user_event_put(struct user_event *user, bool locked)
static void user_event_group_destroy(struct user_event_group *group)
{
kfree(group->system_name);
+ kfree(group->system_multi_name);
kfree(group);
}
@@ -348,6 +356,21 @@ static char *user_event_group_system_name(void)
return system_name;
}
+static char *user_event_group_system_multi_name(void)
+{
+ char *system_name;
+ int len = sizeof(USER_EVENTS_MULTI_SYSTEM) + 1;
+
+ system_name = kmalloc(len, GFP_KERNEL);
+
+ if (!system_name)
+ return NULL;
+
+ snprintf(system_name, len, "%s", USER_EVENTS_MULTI_SYSTEM);
+
+ return system_name;
+}
+
static struct user_event_group *current_user_event_group(void)
{
return init_group;
@@ -367,6 +390,11 @@ static struct user_event_group *user_event_group_create(void)
if (!group->system_name)
goto error;
+ group->system_multi_name = user_event_group_system_multi_name();
+
+ if (!group->system_multi_name)
+ goto error;
+
mutex_init(&group->reg_mutex);
hash_init(group->register_table);
@@ -1482,6 +1510,11 @@ static int destroy_user_event(struct user_event *user)
hash_del(&user->node);
user_event_destroy_validators(user);
+
+ /* If we have different names, both must be freed */
+ if (EVENT_NAME(user) != EVENT_TP_NAME(user))
+ kfree(EVENT_TP_NAME(user));
+
kfree(user->call.print_fmt);
kfree(EVENT_NAME(user));
kfree(user);
@@ -1504,12 +1537,24 @@ static struct user_event *find_user_event(struct user_event_group *group,
*outkey = key;
hash_for_each_possible(group->register_table, user, node, key) {
+ /*
+ * Single-format events shouldn't return multi-format
+ * events. Callers expect the underlying tracepoint to match
+ * the name exactly in these cases. Only check like-formats.
+ */
+ if (EVENT_MULTI_FORMAT(flags) != EVENT_MULTI_FORMAT(user->reg_flags))
+ continue;
+
if (strcmp(EVENT_NAME(user), name))
continue;
if (user_fields_match(user, argc, argv))
return user_event_get(user);
+ /* Scan others if this is a multi-format event */
+ if (EVENT_MULTI_FORMAT(flags))
+ continue;
+
return ERR_PTR(-EADDRINUSE);
}
@@ -1889,8 +1934,12 @@ static bool user_event_match(const char *system, const char *event,
struct user_event *user = container_of(ev, struct user_event, devent);
bool match;
- match = strcmp(EVENT_NAME(user), event) == 0 &&
- (!system || strcmp(system, USER_EVENTS_SYSTEM) == 0);
+ match = strcmp(EVENT_NAME(user), event) == 0;
+
+ if (match && system) {
+ match = strcmp(system, user->group->system_name) == 0 ||
+ strcmp(system, user->group->system_multi_name) == 0;
+ }
if (match)
match = user_fields_match(user, argc, argv);
@@ -1923,6 +1972,39 @@ static int user_event_trace_register(struct user_event *user)
return ret;
}
+static int user_event_set_tp_name(struct user_event *user)
+{
+ lockdep_assert_held(&user->group->reg_mutex);
+
+ if (EVENT_MULTI_FORMAT(user->reg_flags)) {
+ char *multi_name;
+ int len;
+
+ len = snprintf(NULL, 0, "%s:[%llx]", user->reg_name,
+ user->group->multi_id) + 1;
+
+ multi_name = kzalloc(len, GFP_KERNEL_ACCOUNT);
+
+ if (!multi_name)
+ return -ENOMEM;
+
+ snprintf(multi_name, len, "%s:[%llx]", user->reg_name,
+ user->group->multi_id);
+
+ user->call.name = multi_name;
+ user->tracepoint.name = multi_name;
+
+ /* Inc to ensure unique multi-event name next time */
+ user->group->multi_id++;
+ } else {
+ /* Non Multi-format uses register name */
+ user->call.name = user->reg_name;
+ user->tracepoint.name = user->reg_name;
+ }
+
+ return 0;
+}
+
/*
* Parses the event name, arguments and flags then registers if successful.
* The name buffer lifetime is owned by this method for success cases only.
@@ -1985,7 +2067,13 @@ static int user_event_parse(struct user_event_group *group, char *name,
INIT_LIST_HEAD(&user->validators);
user->group = group;
- user->tracepoint.name = name;
+ user->reg_name = name;
+ user->reg_flags = reg_flags;
+
+ ret = user_event_set_tp_name(user);
+
+ if (ret)
+ goto put_user;
ret = user_event_parse_fields(user, args);
@@ -1999,11 +2087,14 @@ static int user_event_parse(struct user_event_group *group, char *name,
user->call.data = user;
user->call.class = &user->class;
- user->call.name = name;
user->call.flags = TRACE_EVENT_FL_TRACEPOINT;
user->call.tp = &user->tracepoint;
user->call.event.funcs = &user_event_funcs;
- user->class.system = group->system_name;
+
+ if (EVENT_MULTI_FORMAT(user->reg_flags))
+ user->class.system = group->system_multi_name;
+ else
+ user->class.system = group->system_name;
user->class.fields_array = user_event_fields_array;
user->class.get_fields = user_event_get_fields;
@@ -2025,8 +2116,6 @@ static int user_event_parse(struct user_event_group *group, char *name,
if (ret)
goto put_user_lock;
- user->reg_flags = reg_flags;
-
if (user->reg_flags & USER_EVENT_REG_PERSIST) {
/* Ensure we track self ref and caller ref (2) */
refcount_set(&user->refcnt, 2);
@@ -2050,6 +2139,11 @@ static int user_event_parse(struct user_event_group *group, char *name,
user_event_destroy_fields(user);
user_event_destroy_validators(user);
kfree(user->call.print_fmt);
+
+ /* Caller frees reg_name on error, but not multi-name */
+ if (EVENT_NAME(user) != EVENT_TP_NAME(user))
+ kfree(EVENT_TP_NAME(user));
+
kfree(user);
return ret;
}
@@ -2640,7 +2734,7 @@ static int user_seq_show(struct seq_file *m, void *p)
hash_for_each(group->register_table, i, user, node) {
status = user->status;
- seq_printf(m, "%s", EVENT_NAME(user));
+ seq_printf(m, "%s", EVENT_TP_NAME(user));
if (status != 0)
seq_puts(m, " #");