We still have protocol.h because the zxs-test needs the protocol
definition. Once NET-1865 is fixed, we'll be able to remove protocol.h
as well.

Also, removed a number of unused declarations from these headers.

Test: No behavior change
Change-Id: Ifeee4e3bfc3e0781e3cef28fb327836d9642904e

This CL adds a simple fuzzer which fuzzes the Edid::Init method.

SEC-44 #comment Another fuzzer to test the framework with

Tests: Manually tested with uapp/fuzz
Change-Id: Ib802490a7a1131b9ebc65b652e11ad56fa5906a0

This is a relanding of 5e1b40e9 which
was reverted in 63d3696c for causing
failures on the roller.  The failures were caused by the unrelated
bug ZX-2961.

This captures the ownership model of DcIostate's memory and its
associated IPC channel.

Test: Booted on hardware and tested interactions
Change-Id: I709a7f8679e8831529e5f4a5a554cbed462fe7d1

This is a relanding of 57372092 which was
reverted in e278a58e for causing
failures on the roller.  The only change between this patch and that are
naming changes.

This will get us things like future-scheduled events, which
would be useful for implementing backoff on driver binding attempts.

Test: Booted in qemu with ASAN enabled.  Booted on hardware, verified
system comes up and is responsive to hardware events.

Change-Id: Ia2fbc2c7fe1347184c1e3645daefdc1dec6f3ea0

Previously multiple threads were accessing the dev->proxy_ios and the
proxy_ios->ph.handle fields concurrently.

ZX-2691 #done

Test: Do not yet have a reliable test for this. ZX-2987 tracks adding
a test later.

Change-Id: Iaae98c30f9fc17efaf7ad1f84668503adb52754f

Test: none

Change-Id: I9eb2c9c193ca7fd0e7e692e7fc0c5297c232655e

Print a warning when we fail to print all frames in a
backtrace. Hopefully, this warning will make it more obvious when a
crash is caused by unbounded recursion.

Old style:

[00024.028] 01103.01117> bt#48: pc 0x5bde7df5e670 sp 0x1e75cb9c73d0 (app:/boot/test/sys/time-test,0x670)
[00024.028] 01103.01117> bt#49: pc 0x5bde7df5e670 sp 0x1e75cb9c7480 (app:/boot/test/sys/time-test,0x670)
[00024.028] 01103.01117> bt#50: end
[00024.028] 01103.01117> warning: backtrace frame limit exceeded; backtrace may be truncated

New style:

[00024.031] 01103.01117> {{{bt:48:0x5bde7df5e670}}}
[00024.031] 01103.01117> {{{bt:49:0x5bde7df5e670}}}
[00024.031] 01103.01117> warning: backtrace frame limit exceeded; backtrace may be truncated

Test: ran a program that recurses infinitely, saw the new warning

ZX-2989 #done

Change-Id: I0434a86986e63391f647db9c09be328cac23f8a3

This function is the last function to be implemented in zxs.

Test: Covered by loopback tests in //garnet
Change-Id: I95e1c18f5672e91c3f49089d68ebb37ad6a967a5

device_remove() may be called from a different thread than the async
loop.  Because of that, we need to avoid mutating the async connection
data structure from device_remove().  This patch reworks the shutdown
logic to avoid unsafe accesses.

ZX-2961 #comment Fixes the non-proxy datarace

Test: Ran /pkgfs/packages/wlan-hw-sim-tests/0/test/wlan-hw-sim_bin_test_rustc
repeatedly.  Previously would trigger a devhost crash roughly 1 in 5
times.  Did not see anymore devhost crashes with this patch, except for
the discovery of a bug in the wlan driver (NET-1860).

Change-Id: Ib48d058b0f5f4d5b1958d6ea688c67ae1e53b6d1

DevcoordinatorConnection is now used solely for the devhost<->devcoordinator
communications channel.

DevfsConnection has been introduced to track application <-> device
connections (which happen via devfs).

Test: Build and ran with ASAN in qemu
Change-Id: Ibb7d557a8b6422dc0ffbbe6c57d0b62ebfb4cdba

DevhostIostate is used to track two different types of async
connections:
1) The devhost <-> devcoordinator connection for a device
2) The application <-> device (devfs) connections for a device

This diff refactors away the one point at which the types were being
merged.

Test: Build and ran under ASAN
Change-Id: Icd87c8ebd2b622d63f47803e34d1747fcb5d92c0

This CL handles the case that Unbind is called simultaneouesly with
Bind.  It prevents DdkRemove from being called twice, which would lead
to a use-after-free.

ZX-2988 #done

Tests: zxcrypt unit tests
Change-Id: Ib3da97b1b6b0841e94b8538b5a575391596b8b1e

Ported from previously removed iotxn tests to usb-request.
Will be adding more tests in future when modifying phys_iter to
to account for scatter gather.

ZX-936 #comment

TEST= runtests /system/test/ddk-test

Change-Id: I5170726383ab949626dd60e0c9db187b20be25d3

This is part2 of my previous changeset, https://fuchsia-review.googlesource.com/c/zircon/+/218979
I just broke this down into 2 changesets, so as not to change too many
clients in 1 change.

Test:
vim2: Plugged in a usb mass storage device
      and read using dd on it.
Pixelbook: Netbooted pixelbook. Plugged in a usb mass storage
           device and read some blocks using dd.
           audio tone
Astro: Netbooted. Verified that the usb ethernet adapter showed up.

Change-Id: I6136b937988958a460b369b368c84a60199cb528

Convert launchpad tests to C++, changing as little as possible.

Test: /boot/test/sys/launchpad-test

Change-Id: Ib2ebdc74880aac70598d5a051fa520f5a40190a2

This reverts commit 663d78ff.

Reason for revert: Incompatible change with a driver in astro

Original change's description:
> [block][ddk][ddktl][fvm] changed signature of get_stats for block device
> 
> changed signature of get_stats for block device to reveal less
> information. This also required changing the code in a few places
> that were autogenerated from the old signature.
> 
> tests: runtests
> Change-Id: I57483dc6d0fdf06e209e0183150d2f5e6353e8c3

TBR=smklein@google.com,rvargas@google.com,gevalentino@google.com,kmerrick@google.com

Change-Id: I756a88eb753483beefbed7ec815eea2e5ba7223e
No-Presubmit: true
No-Tree-Checks: true
No-Try: true

Test: boot and run "set", observe new path
Change-Id: Iea7c0c2ebbf17997ff3aa1d5ec9d9e73aad62d1e

Test: iochk
ZX-2783

Change-Id: I3d543ded18a0d14b0828ff122612ee51b3f89e1f

Test: N/A
Change-Id: Id99eacb07c1c8a46ee390982cf60736abf262dc9

DdkRemove() will drop the driver's reference to the device.  Since the
driver only has one reference, calling it twice is unsafe.

Test: Ran the blobfs-tests with new devhost changes that caused this
bug to trigger frequently.

Change-Id: Id322cd131c4fefe125b6aa2df79c8c49694ccccc

changed signature of get_stats for block device to reveal less
information. This also required changing the code in a few places
that were autogenerated from the old signature.

tests: runtests
Change-Id: I57483dc6d0fdf06e209e0183150d2f5e6353e8c3

Fix missing angular bracket and make it clear we are comparing to
std::optional.

TEST: no test needed since only comments are changed.

Change-Id: Ib2557bfc65d668b7b878a2bd88dfced0a1c603ca

No longer need to define max_histograms/max_counters.

Use a wrapper class to dereference a pointer to the vector
and an index in the vector, ElementView.

Histogram and Counter used to have a pointer to their respective
internal implementations(Remote and local in the future). This meant
that the vector that contained them was not allowed to reallocated,
which is why we requested the amount of each before allocation.

Instead now we prevent Collector from being moved, and provide only
heap allocated Collector. This allows to have a pointer to the vector
and index within the vector, fetching on demand.

Due to lack of support for immutable vector mutable elements, or
vecotr view (std::span) we provide an equivalent of for a single
element, eq of span of size 1, ElementView.

TEST=cobalt-client-test

Change-Id: Ie18af1b3c6927cfe41282d8ac951fce2c1dff181

Koids with the topmost bit set are for use by software for
artificially created objects, e.g., "virtual threads" in tracing.

Tested: CQ
Change-Id: I55ecb7fe01a9b20256b04d733643cc48e0726f01

This CL adds a simple fuzzer that executes the Parse method.

SEC-44 #comment Add a fuzzer to test with

Tests: Manual tested using uapp/fuzz

Change-Id: I1825d1c298f5455067dfa8f5b95367708fd837fe

Motivation:usb_request_t struct needs to be broken into public and private regions
that each driver in usb driver stack can access privately. This is done to avoid
drivers touching each other's internal context data in the
usb_request_t. The overall request size will be sizeof(public region -> usb_request_t)
plus all the private contexts of each layer in the stack.
This changeset allocates the size required to accommodate contexts of
usb_bus and controller layers. In a later changeset, xhci and usb-device
will use the extra space allocated without changing the request struct.
The duplicated fields in the usb_request_t will be removed at the end.

I have fixed the issue that caused network outage previously, in this
changeset. The issue was that I was using the dev->req_size before
initializing it, in usb-device.c

Test:fx full-build. Netbooted vim2. Plugged in a usb mass storage device
and read using dd on it. Netbooted pixelbook. Plugged in a usb mass storage
device and read some blocks using dd. Netbooted astro and did a lsusb.
Verified that the network adapter plugged-in showed up in lsusb.
runtests /boot/test/ddk

Change-Id: I3937113260f19d2b30e3d9e6109a9e9b724b95b3

After this CL, all the zxsio transactions are issued by the zxs library.

Test: ioctl_test
Change-Id: I91d265529f5c65d68e50ff7fdb78c87825c860c1

This causes more things to take references to zx_device, which should
help us avoid some use-after-free risks, especially as we try to address
the data race described in ZX-2961.

Test: Booted full system and ran tests in qemu with ASAN.  Booted full
system on hardware, confirmed peripherals functioned.

Change-Id: I8d7683a689a1bd066c0f31ef7b67202e6c2d8c01

Future CLs will actually use this and pass it to phys_iter.

ZX-936 #comment

TEST= runtests /boot/test/ddk

Change-Id: I9afe02121a5d2bccc2858193654c292e2df32edd

Test: CQ
Change-Id: If589566ae6ed9579ccdb8a7c069fb8065d853e18

When a client connects to an ethernet device, set the status signal on
the rx fifo to encourage them to read the status. This is normally
triggered by the start() method in the ethernet protocol, except that
the new client isn't added to the active set at the time so it misses
the signal.

Reland of https://fuchsia-review.googlesource.com/c/zircon/+/181072

NET-713
Test: - manual testing in QEMU with the 'set_link' monitor command
      - /system/test/sys/ethernet-test
      - manual testing with a Pixelbook and a USB ethernet adapter

Change-Id: I36410293f3ca5225e90fea01468fa7a284548af1

Test: no functional change
Change-Id: Ifc72cefec2cba2b32e7a5c3ac0be3f1c0cc12ae2

This CL moves the data transport code for sockets from FDIO to ZXS.
The FDIO code now uses ZXS to transport data for sockets.

Test: No behavior change
Change-Id: Ibf08374fba75f56a11e29cee546f70f7af4ec38d

Tested: doc change, eyeballed it
Change-Id: I823ff3d80c505344aa39080c6657a8fe795b6679

... and reorganize things a little

Tested: eyeballed it in gitiles
Change-Id: Ibcf817a3775b48f988b9813db8b4f9f5550ac65f

GPT partitions have their own sizes stored in info,
which is not necessarily equal to the entire size of the
underlying device.

Test: Able to run `fvm-check` on an FVM within GPT successfully.
ZX-2978 #comment In progress

Change-Id: I30411ecfe9f36b133c525adbe2922c5b9ad8ac33

There's no guarantee the bootstrap stack has enough space for all the
string pointers.  It's better to have those arrays on the user (unsafe)
stack anyway.

Bug: ZX-2926 #comment libc unpacks strings on user stack
Test: CQ
Change-Id: I381c5f2b091d8ca329b16e951cc88d90ef358199

"type" field in these constants are supposed to be used with has type
uint8.  This is a follow up to

  7f36f25d
  [fdio][fidl] Use unaligned dirents for readdir, add ino

where the type of the field was changed but the type of the constants
was not.

Test:
  $ /system/test/fs/fs-test -f memfs
  CASES:  108     SUCCESS:  108     FAILED:  0
  $ /system/test/fs/fs-test -f minfs
  CASES:  108     SUCCESS:  108     FAILED:  0

Change-Id: I3076a3b295c64b4cb11a3e4d9301d19590d756e1

Previously I had the assumption that we would know what we are logging
a priori, which is not always true. For example, when Operation A
has multiple branches and different branches map to different event
codes, we dont know which branch we are gonna pick until it is picked.
At this point binding the timer to a specific histogram, is less
useful than simply providing a fs::Ticker equivalent.

For this reason, removed the histogram as part of the timer,
and we can do something like:
Timer timer(is_collecting);

.....
    hitogram.Add(timer.End());
TESTS= cobalt-client-test

Change-Id: I93731f983112b0a061130c8f6f7bab9d7f98cffc

A previous iteration of devfs and ulib/fs filesystems transmitted FIDL
messages which, although parsed by the corresponding client bindings in
the fdio C library, were technically invalid FIDL. This bug was due to
the fact that an optional field from the OnOpen event was always
transmitted, even when it was "FIDL_ALLOC_ABSENT". This arose partially
because the hand-rolled FIDL bindings have historical origins in a
non-FIDL protocol, called "RemoteIO", and some artifacts of this
conversion still exist.

This patch narrows the line between FIDL server implementations and
their respective auto-generated bindings, removing some leftovers from
RemoteIO:

- Vnode::GetHandles has been replaced with a version which operates
on "fuchsia_io_NodeInfo" directly, rather than an opaque "handles,
type, and zxrio_node_info_t".
- zxrio_node_info_t has been deprecated in favor of using
fuchsia_io_NodeInfo directly. This allows us to reomve static assertions
that the two structures are identical.
- Some headers have been reshuffled to avoid host-side dependencies
on FIDL bindings.
- A test has been added for receiving this open event, on either
success or failure, to the devfs test suite. This patch also tests
the ulib/fs filesystems by opening "/boot/lib".

Test: Expanded system/utest/devfs with "TestFidlOpen".
ZX-2927 #done

Change-Id: I37528a57567ce6c7cb53a76ed28142be9f5dc8fb