u-root / u-root

"""
.cirrus.yml#L8
you've probably meant template_task
"""

Signed-off-by: David Hu <xuehaohu@google.com>

Sometimes users need to add files to the kexec cpio. A common
request is to pass the kernel log buffer along.

Modify kexec so that initramfs can now specify multiple
cpio files, e.g.
kexec -i 'a.cpio b.cpio c.cpio'
The code uses strings.Fields, as that nicely handles the case
of superfluous seperators, i.e. '   a  b  c   ' returns [a b c].

Now, should someone desire to pass along, e.g., dmesg output, they can
do this via script or programatically:

mkdir /tmp/messageinabottle
dmesg > /tmp/messageinabottle/log
(cd /tmp && find messageinabottle -print | cpio -H newc -o > /tmp/m.cpio)
kexec -i 'inird.cpio /tmp/m.cpio'

A remaining question is how much of this work we ought to do in kexec.
It could, for example, take a switch called extrafiles:
kexec -extrafiles 'a b c' ...
and build an in-memory cpio archive to be appended to the initrd.

At some point, a line needs to be drawn, however, as to how much
we put into a single command like kexec.

Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>

Signed-off-by: David Hu <xuehaohu@google.com>

Signed-off-by: Rob Vandermeulen <rvandermeulen@google.com>

Calling term.Restore with oldstate == nil triggers a runtime error ( panic: runtime error: invalid memory address or nil pointer dereference ).

Getting into this state is possible when the term.MakeRaw call fails, which can happen if /dev/tty ends up pointing to a broken serial console.
With this change and a broken console, ShowMenuAndLoad will end up choosing the default entry instead of crashing itself.

Signed-off-by: Rob Vandermeulen <rvandermeulen@google.com>

In scripts, as used in, e.g., Harvey, it is important to
know what commands are built into a bb.

The glob passed to makebb will not always be the same
as the actual commands that get built (there might be a
match for a command that does not get built due to build
constraints).

The -c switch to bb will now list the commands that get built in.

Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>

Signed-off-by: Johnny Lin <johnny_lin@wiwynn.com>

…d 'delete'

By calling flashrom and vpd executables to set and delete vpd key-value.
The ideal long-term method could be employing Linux mtd driver for writing
flash VPD but this PR can provide a temporary workaround.

Move pkg/ipmi/ocp/set_vpd.go to pkg/vpd/flashromvpd.go

To set an existing vpd key to value:
vpdbootmanager set key value
If the key doesn't exist it would add it.

To delete a key:
vpdbootmanager delete key

Signed-off-by: Johnny Lin <johnny_lin@wiwynn.com>

This change solves the problem of readonly initramfs updating
the etc files. The init process copies the etc directory to
/tmp/etc and bind mount to /etc.

Tested:
```
~/# mount
none / rootfs rw 0 0
devtmpfs /dev devtmpfs rw,relatime 0 0
proc /proc proc rw,relatime 0 0
none /tmp tmpfs rw,relatime 0 0
devpts /dev/pts devpts rw,relatime,gid=0,mode=620,ptmxmode=666 0 0
none /dev/shm tmpfs rw,relatime 0 0
sysfs /sys sysfs rw,relatime 0 0
none /etc tmpfs rw,relatime 0 0
~/# echo hihi >/tmp/etc/hello
~/# cat /etc/hello
hihi
```

Signed-off-by: Medicine Yeh <medicinehy@gmail.com>

This allows filtering for block devices that have a partitions 1,2,4

Signed-off-by: Colin Mitchell <colinmitchell@google.com>
Change-Id: I9f59e197b7376b1ac7e357b1fa00c2d7e1180054

Signed-off-by: Christopher Meis <christopher.meis@9elements.com>

Having valid time is important for HTTPS boot to work, as it prevents
certificate validation failures due to invalid system time.

This PR adds NTP support, enabled by default, to query NTP servers
hard-coded in the image or returned by DHCPv4/v6.

Failure to set time is not fatal and boot proceeds regardless.
With no additional configuration and assumiong no NTP servers
returned by the DHCP server, it's a no-op.

Signed-off-by: Deomid "rojer" Ryabkov <rojer9@fb.com>

go get -u github.com/insomniacslk/dhcp
go mod vendor
go mod tidy

Signed-off-by: Deomid "rojer" Ryabkov <rojer9@fb.com>

This caused trouble for someone today, as they followed the instructions
in the order shown.

Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>

ntpdate(8) allows (in fact, requires) passing server(s) con the command line.
Make u-root version allow that too. Default behavior of parsing servers
from the config is preserved.

Add -rtc that will set hardware clock along with system clock.

Pull out the functionality into a library - this will be used in
fbnetboot in a future PR.

Signed-off-by: Deomid "rojer" Ryabkov <rojer9@fb.com>

Go's caching is mature and can be trusted to detect changes.

Signed-off-by: Deomid "rojer" Ryabkov <rojer9@fb.com>

* These empty files are necessary for `go test` to instrument the
  package for code coverage.
* Separately, our internal infra fails when a _test.go file contains no
  tests.

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

Our internal infra runs the subcommand under the wrong working
directory.

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

backoff will run a command, repeatedly, with exponential
backoff, until it succeeds or a timeout is passed.
The default timeout is zero, meaning "run until success".

The timeout is specified as a standard string duration,
e.g. 30s, 10m, and so on.

Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>

And refactored pkg/vpd to be testable.

The `get` command in vpdbootmanager allows to read VPD variables.
Example:
```
~/# vpdbootmanager get firmware_version
firmware_version(RO) => 0.0.0
```

If no arguments are passed to `get`, all the variables are read:
```
~/# vpdbootmanager get
firmware_version(RO) => 0.0.0
internal_versions(RO) => {
  "build_id": "6bc0645",
...
    }
  }
}
~/#
```

Signed-off-by: Andrea Barberio <insomniac@slackware.it>

The root cause was the wrong path of rmdir in (p *Pool) Mount().

Verified that this was caused by rmdir() with new error message format
```
2021/08/06 10:21:05 Failed in UnmountAll: (Rmdir) directory not empty
2021/08/06 10:21:05 executing command "/bin/defaultsh" with args []
u-root-mounts716997181
~/# ls /tmp/u-root-mounts716997181
sda
sdb1
sdb2
```

Signed-off-by: Medicine Yeh <medicinehy@gmail.com>

Removed the spawned function in ShowMenuAndLoad in favor of
file.SetDeadline(). Switched from Stdin to TTY to support this.

Previous tests for menu.Choose spawned a xterm shell and had to use manual sleeps prevent races.
The menu class used methods and member variables of xterm so a wrapper interface/class were made to support mocking+injection
This shaved ~20s off the test time

Signed-off-by: Colin Mitchell <colinmitchell@google.com>
Change-Id: I392d8e3d2629b8ae7cbf3b6fa47e9dbf2f28b015

Run gofmt.

Updates #2701.

Signed-off-by: Than McIntosh <thanm@google.com>

Add a unit test that invokes the main entry point in this package with
bogus value, just to make sure it doesn't crash. This apparently
needed so as to satify the CI coverage requirements.

Updates #2701.

Signed-off-by: Than McIntosh <thanm@google.com>

Fix bugs in the assembly wrappers added to collect trampoline
func addresses (missing return statements).

Updates #2701.

Signed-off-by: Than McIntosh <thanm@google.com>

Revises the recipe for obtaining the address (starting PC) of assembly
routines from Go, so as to have something that works with the new
(1.17+) register ABI.

Fixes #2701.

Signed-off-by: Than McIntosh <thanm@google.com>

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

The old format was deprecated see "go help build".

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

We had a "hidden" dependency with the %s print.

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>

This makes sure the u-root initramfs as well as the Go tests to run in
the VM are compiled with the same Go version as the test runner.

Previously, if one did (cd integration/gotests && go1.13 test), the test
runner and initramfs would be compiled with go1.13, but the Go VM tests
would be compiled with `go` found in $PATH.

Signed-off-by: Chris Koch <chrisko@google.com>

Signed-off-by: Ronald G Minnich <rminnich@gmail.com>

The big change is the switch to Go 1.15 and QEMU 6.0.0. This change
fixes issues arising from this migration.

This change also improves the logging of the gotest vmtest to help debug
issues with this migration.

Signed-off-by: Ryan O'Leary <ryanoleary@google.com>