Skip to contentSkip to conversation

U-Boot environment management from userspace

DISCUSSION Vernon Mauery 7 participants 9 messages May 28, 2019 – Jun 12, 2019

lore
Share & tools
lore
#0 lore raw

Reading U-Boot environment variables from userspace is not difficult, but to do it in a standard way, (fw_printenv), it requires a fork and exec. We don't have any permissions problems because reading from the MTD partition is not restricted. It might be nice, however to have these variables exported on D-Bus so that a fork/exec is not necessary, just a property fetch.

But writing is a different story. That requires root privileges. To architect with a separation of privileges mechanism, this should probably be running as a daemon or service that could be spawned via D-Bus or something so that ipmid doesn't need root permission to set a U-Boot variable.

I see a couple of options:
1) Shoehorn U-Boot variables into the settings daemon so they just show up as settings. I am not sure on the details of how this would be done, but it might work.

2) Create yet another daemon that would provide a R/W interface (probably just using the D-Bus properties interface) that would act as a manager of U-Boot environment variables. It might even be able to place an inotify watch to get notified when an external process (fw_setenv) modifies the environment (like from a script or something) so the D-Bus properties could send out a PropertiesChanged notification.

3) Use a one-shot service that parses the 'instance' to extract a variable name and variable value. Then the variable could be activated by launching ubootenv@foo=bar.service. This would require some fancy parameter encoding to make it all work correctly to avoid string injections.

Am I the only one that has a need for this or is there a wider audience that would benefit?

Does anyone else already have a solution for this or an opinion on what path might be the best?

--Vernon

#1 lore raw

Vernon,

I just started a daemon for the same as i needed it for RestrictionMode (provisioning) . At this point of time, the daemon uses the fw_setenv & printenv internally, but atleast application will access the same in D-Bus

Regards,

Richard

Quoted 15 lines · Vernon Mauery

Reading U-Boot environment variables from userspace is not difficult, but to do it in a standard way, (fw_printenv), it requires a fork and exec. We don't have any permissions problems because reading from the MTD partition is not restricted. It might be nice, however to have these variables exported on D-Bus so that a fork/exec is not necessary, just a property fetch.

But writing is a different story. That requires root privileges. To architect with a separation of privileges mechanism, this should probably be running as a daemon or service that could be spawned via D-Bus or something so that ipmid doesn't need root permission to set a U-Boot variable.

I see a couple of options:
1) Shoehorn U-Boot variables into the settings daemon so they just show up as settings. I am not sure on the details of how this would be done, but it might work.

2) Create yet another daemon that would provide a R/W interface (probably just using the D-Bus properties interface) that would act as a manager of U-Boot environment variables. It might even be able to place an inotify watch to get notified when an external process (fw_setenv) modifies the environment (like from a script or something) so the D-Bus properties could send out a PropertiesChanged notification.

3) Use a one-shot service that parses the 'instance' to extract a variable name and variable value. Then the variable could be activated by launching ubootenv@foo=bar.service. This would require some fancy parameter encoding to make it all work correctly to avoid string injections. Am I the only one that has a need for this or is there a wider audience that would benefit?

Does anyone else already have a solution for this or an opinion on what path might be the best?

--Vernon

#2 lore raw
In reply to Vernon Mauery
Quoted 2 lines

Am I the only one that has a need for this or is there a wider audience that would benefit?

The software manager (phosphor-bmc-code-mgmt) relies on U-Boot environment variables for managing the images like for determining which image to boot from.

Quoted 5 lines

3) Use a one-shot service that parses the 'instance' to extract a variable name and variable value. Then the variable could be activated by launching ubootenv@foo=bar.service. This would require some fancy parameter encoding to make it all work correctly to avoid string injections.

Yeah, we went that route with an [email protected][1], with like you mentioned uses some 'fancy parameter encoding', ex:
"obmc-flash-bmc-setenv@" + entryId + "\\x3d" + std::to_string(value) + ".service";

This has worked so far but I'd be open on having a mapping of the env variables to D-Bus properties.

Quoted 3 lines

Reading U-Boot environment variables from userspace is not difficult, but to do it in a standard way, (fw_printenv), it requires a fork and exec.

We're actually reading the mtd device to get the values of the variables to avoid having a 'system' or 'fork/exec' call, then we put those values in D-Bus properties under the software/ path. Having some other app do that parsing would be nice especially if we're to expand the use of the U-Boot env vars.

---
[1] https://github.com/openbmc/phosphor-bmc-code-mgmt/blob/master/obmc-flash-bmc-setenv%40.service

#3 lore raw
Quoted 24 lines · Adriana Kobylak
Quoted 2 lines

Am I the only one that has a need for this or is there a wider audience that would benefit?

The software manager (phosphor-bmc-code-mgmt) relies on U-Boot environment variables for managing the images like for determining which image to boot from.

Quoted 5 lines

3) Use a one-shot service that parses the 'instance' to extract a variable name and variable value. Then the variable could be activated by launching ubootenv@foo=bar.service. This would require some fancy parameter encoding to make it all work correctly to avoid string injections.

Yeah, we went that route with an [email protected][1], with like you mentioned uses some 'fancy parameter encoding', ex:
"obmc-flash-bmc-setenv@" + entryId + "\\x3d" + std::to_string(value) + ".service";

This has worked so far but I'd be open on having a mapping of the env variables to D-Bus properties.

Quoted 3 lines

Reading U-Boot environment variables from userspace is not difficult, but to do it in a standard way, (fw_printenv), it requires a fork and exec.

Quoted 10 lines

We're actually reading the mtd device to get the values of the variables to avoid having a 'system' or 'fork/exec' call, then we put those values in D-Bus properties under the software/ path. Having some other app do that parsing would be nice especially if we're to expand the use of the U-Boot env vars.

---
[1] https://github.com/openbmc/phosphor-bmc-code-mgmt/blob/master/obmc-flash-bmc-setenv%40.service

Brad,

It sounds like Intel is not the only ones that might benefit from a service like this, so it might be a good time for a new project/repo.

Could you create a new repo for us: phosphor-u-boot-env-mgr

--Vernon

#4 lore raw

Hi Vernon

Quoted 7 lines

Brad,

It sounds like Intel is not the only ones that might benefit from a service like this, so it might be a good time for a new project/repo.

Could you create a new repo for us: phosphor-u-boot-env-mgr

Will do.

It sounds like you are working on something that needs a new repo, will generate a new dbus API and have applications providing and consuming that API. That sounds like a non-trivial enhancement to OpenBMC.

Please consider having someone submit a completed design template and opening a github issue. This enables the rest of the community to know what Intel is working on and when, and thus have input, possibly help, and avoid duplicate work. Please let me know if any of my thinking here is flawed.

thx - brad

#5 lore raw
Quoted 8 lines · Brad Bishop

Hi Vernon

Quoted 6 lines

Brad,

It sounds like Intel is not the only ones that might benefit from a service like this, so it might be a good time for a new project/repo.

Quoted 7 lines
Quoted 1 line

Could you create a new repo for us: phosphor-u-boot-env-mgr

Will do.

It sounds like you are working on something that needs a new repo, will generate a new dbus API and have applications providing and consuming that API. That sounds like a non-trivial enhancement to OpenBMC.

Yes. Non trivial changes require designs.

Quoted 5 lines

Please consider having someone submit a completed design template and opening a github issue. This enables the rest of the community to know what Intel is working on and when, and thus have input, possibly help, and avoid duplicate work. Please let me know if any of my thinking here is flawed.

Not flawed at all. It has been discussed and agreed upon by all participating companies. It also greatly reduces the review time for new features being contributed, because the community already knows what to expect.

Kurt Taylor (krtaylor)

Quoted 3 lines

thx - brad

#6 lore raw
In reply to Brad Bishop
Quoted 16 lines · Brad Bishop

Hi Vernon

Quoted 7 lines

Brad,

It sounds like Intel is not the only ones that might benefit from a service like this, so it might be a good time for a new project/repo.

Could you create a new repo for us: phosphor-u-boot-env-mgr

Will do.

It sounds like you are working on something that needs a new repo, will generate a new dbus API and have applications providing and consuming that API. That sounds like a non-trivial enhancement to OpenBMC.

Quoted 7 lines

Please consider having someone submit a completed design template and opening a github issue. This enables the rest of the community to know what Intel is working on and when, and thus have input, possibly help, and avoid duplicate work. Please let me know if any of my thinking here is flawed.

thx - brad

Hi Vernon

I created this today.

I would still love to hear about what Intel is cooking up that would make use of this.

Thanks! - brad

#7 lore raw
In reply to Vernon Mauery
Quoted 7 lines · Vernon Mauery

2) Create yet another daemon that would provide a R/W interface (probably just using the D-Bus properties interface) that would act as a manager of U-Boot environment variables. It might even be able to place an inotify watch to get notified when an external process (fw_setenv) modifies the environment (like from a script or something) so the D-Bus properties could send out a PropertiesChanged notification.

This would be my preference. It keeps the functionality self contained, and allows at least some security around adjusting parameters, given that said daemon could whitelist/blacklist certain parameters at some point in the future.

#8 lore raw
In reply to Vernon Mauery
Quoted 4 lines · Vernon Mauery

Reading U-Boot environment variables from userspace is not difficult, but to do it in a standard way, (fw_printenv), it requires a fork and exec.

I came across libubootenv the other day. Whatever solution you come up with to manage the variables could use this library. It's in tools/env/ in the u-boot sources.

I would caution against the u-boot environment being used as a general purpose settings store. We should use it for configuring u-boot, and keep other settings somewhere else in the system.

Cheers,

Joel