The double free would occur if
1) some thread invoked Cancel() to queue destruction of a ProxyIostate,
   and
2) before that packet was processed, the ProxyIostate's handler ran and
   decided to destroy itself.

Bug: ZX-4060
Change-Id: I3b9c7275c4e0cd2dc3a6c6cd2d2376c947eb627a

This moves the ProxyIostate and ConnectionDestroyer types into their
own files, and feeds them async dispatchers via injection rather than
environmental globals.

Change-Id: Ifeadbed02e92a7e2a7af5346ddd253af74112abf

This reverts commit cf254ae8.

Reason for revert: Causes vim2 to fail to pave. See: https://fuchsia.atlassian.net/browse/DNO-519

Original change's description:
> [devhost] Wait for BindDriver in reply to Bind
> 
> Calls to fuchsia.device.Controller/Bind should only return once the
> operation completes, which is when
> fuchsia.device.manager.DeviceController/BindDriver is called.
> 
> ZX-3991 #done
> 
> Change-Id: I10607640072c8f779ac1d5be79d5a2da6de6288b

TBR=kulakowski@google.com,teisenbe@google.com,abdulla@google.com,brettw@google.com

Change-Id: I370d03abd9fa93a70a48f86f4df171f2cf87ee12
No-Presubmit: true
No-Tree-Checks: true
No-Try: true

Calls to fuchsia.device.Controller/Bind should only return once the
operation completes, which is when
fuchsia.device.manager.DeviceController/BindDriver is called.

ZX-3991 #done

Change-Id: I10607640072c8f779ac1d5be79d5a2da6de6288b

When devcoordinator comes online it prints an error message
if the root resource is not available, so it is unnecessary to
print that the resource is unavailable several tiems later on.

Change-Id: I0785840b53603a4495db286c1bf3495330e4b13f

Devfs should no longer call DirectoryOpen on devices except when opening
the device itself (i.e. with a path of ".").

This is a small change in behavior in that when the describe flag is set
devfs will now report ZX_ERR_NOT_FOUND when attempting to open a
nonexistant file under a device. Previously devmgr would simply close the
channel in this case.

Also remove the DirectoryOpen handler from devmgr.

Add an additional check to the devfs tests that opening a file under a
device path (i.e. /dev/<device>/<file>) fails with ZX_ERR_NOT_FOUND
unless the device has a child device (e.g. /dev/class/block/000).

Change-Id: If7f33ac306940da7e43ae8c5b53a15d907593042

The DirectoryOpen handler will be removed along with resulting changes
to devfs in a follow up CL.

Change-Id: I37df4c6bdf2aba84575a5ab6b692b2da5f362185

We suspect that something may be causing occasional deadlock around the
time that we attempt to bind the zxcrypt driver.  Add more logging, and
increase verbosity in a couple places to see if we can narrow down where
we're going awry.

DNO-492 #comment
TEST: no behavioral change, just more logging

Change-Id: Ib03af33da1d67533599c80e4453ef27dc160950e

Transitions callers of zx::channel::read and zx::channel::read_etc to a
corrected argument order. The first step happened in a previous commit
which temporarily added method names during the transition,
zx::channel::rea2 and zx::channel::rea2_etc to reflect the new argument
ordering.

This commit renames rea2 callers back to read, and changes the class method
definitions to the corrected order.

Test: CQ.

Bug: ZX-2812
Change-Id: I1ee9067d6a3507c18674d3e60b546882fcbd81f0

The coordinator now handles coordinating suspend on a device-by-device
basis, rather than just handling telling which devhosts to begin
suspending in which order.

Change-Id: I2dbd0370c93158f5a33db5a1f51f64200f13e6f5

Transitions callers of zx::channel::read and zx::channel::read_etc to a
corrected argument order and temporarily changes the method names during the
transition to zx::channel::rea2 and zx::channel::rea2_etc.

Test: CQ.

Bug: ZX-2812
Change-Id: If0446abdea9fbcb58568759ee1a0579a8c10cbc4

fbl::make_unique was functionally identical to std::make_unique. There
should be no behavior change in this CL.

The bulk of this CL was prepared via the following zsh script:

    for f in $(git grep -l 'fbl::make_unique<') ; \
    sed -i 's/fbl::make_unique</std::make_unique</' $f

Bug: ZX-3357 #comment remove make_unique
Bug: ZX-3837 #comment remove make_unique
Change-Id: Ic8bf489e4dbf7cb7d2d71ad95aa26018bf776e56

This type was previously being used for both devcoordinator to devhost
messages and devcoordinator to device messages.

Change-Id: Ife4928d0ddee0593923af27ab52213463565560f

The goal of this change is to make it more clear which messages are
supposed to be on which channels.  A subsequent change will make the
devhost code more clear.

Change-Id: I8816f18dc5ccd40bb3a8a5e7a000131b37e7f0ca

This enables line buffering, which folks were relying on for debugging.

MA-573 #done

Change-Id: I03c522eeab2cb5ceec8ff3459dc418d82b0a8c03

... and use to remove zx_thread_set_priority() call from usb-xhci.

devhost connects to scheduler.ProfileProvider, and allows drivers to
request a profile. This is an intermediate state until we decide if/how
we want to allow drivers to consume fidl services.

ZX-3703 #comment [syscalls][usb][xhci] Introduce scheduler profile to devhost
ZX-2914 #comment [syscalls][usb][xhci] Introduce scheduler profile to devhost

Test: CQ, manual instrumentation/logging
Change-Id: I23359f2062b77cf84e3bccef72682668817618ad

When all of a composite device's components are ready, the composite is
created.  In this implementation, we always proxy components, even if
they're in the same process.  This will be relaxed in a later patch.

Issue: ZX-2648
Change-Id: I97f5efb311ca18ec830f81158f022f230abc2d5d

This will let us construct composite devices in the devhost by
giving us names for which pieces to use.

Issue: ZX-2648
Change-Id: If02ef455358c616db0f88e19a2af3c99a28af793

The code was not written with the expectation that global dtors run, and
executing them currently results in ASAN violations.

Change-Id: Ib9e709483b30a8308af59009c0cbb22fec1fd2e1

This does not provide the actual implementation in devmgr, just the
plumbing to send the message from the driver to devmgr.

Issue: ZX-2648

Change-Id: I9c3160d86a35c19313b4cf7450a8e3e7ca4140f0

Change-Id: Iaa831631fa5229d638924c2b79e272043dd6b9a8

This header is deprecated and will be removed soon.

Change-Id: I9cbaf283dee8b43b14adb4ee35d67829d265ce2f

This is a very small step towards a better interface for bind and
create, internally (rather than exposing thread-local storage!).

This also makes it clear that Binding, at least, does not have the RPC
proxy case.

Change-Id: I8571e655e12002794980448cbd5869b63b95c98b

Driver tracing support is now always compiled in.

ZX-3317 #done

Tested: CQ
Change-Id: Ic5e9f952f3f93a9000404bb33bae62456ad4d6ae

Run `fx format-code` without changing "#pragma once" to header guards.
Also set the line limit to 100 columns in the Zircon clang-format
configuration.

Test: N/A
Change-Id: I7049fcec3ebe80a0ca910ed2be95bdbc0303a596

This patch adds a new method (device_get_metadata_size) to the DDK API, which allows
the retrieval of the size of a given metadata key on a device.

Test: runtests -t driver-test
Change-Id: Ib2535b17e3e2b6b9bc225835d6ac27e288aadd52

This allows a client to open a connection to the device
when adding a device, instead of having to wait for the device
appear in the filesystem and open it.

If the device is added with DEVICE_ADD_MUST_ISOLATE,
they will get connected to the proxy device instead.

If the device is added with DEVICE_ADD_INVISIBLE,
they will not be be connected until device_make_visible is called.

ZX-2956 #comment

TEST= I wrote a small device and proxy and verified the channel
worked.

Change-Id: I5e1eed1eedaa6be65d0f0053c4c68dee0ead3058

Test: no functional change; booted under arm qemu with asan
Change-Id: I5372d6ac791b678bf8536fb9120e803b73195e40

ZX-2468 #comment patch

Tested: Turn on verbosity in trace_manager, then
fx shell trace record trace --help
and verify all devhosts are participating in the trace.
Then boot with driver.tracing.enable=0, rerun, and verify devhosts are absent.

Change-Id: I5cd56133f12331aa604e6a857ffb5fef4eb40c0e

Longer term, we should express this relationship in FIDL.

ZX-3073 #comment static check

Test: Compiled
Change-Id: I04f38e3b2869c50eae4ff929bd9427283929b086

Method ordinals will soon be not guaranteed to be contiguous.

Test: Booted the system, verified devices were published.
Change-Id: Ic573c10adf847d34c7b8aeac788082322d80adde

Test: CQ
ZX-3091 #comment In Progress

Change-Id: Ib04c43b355cb99db9910434b33162c6118c611e6

Although there is still heavy reliance on ordinals as constants
outside generated FIDL code, this patch removes the filesystem
dependence on relative ordering of FIDL ordinals by instead
utilizing the "try_dispatch" family of functions.

Rather than routing to the appropriate dispatch function, the FIDL
handler will now "try_dispatch" to all possible interfaces,
checking against ERR_NOT_SUPPORTED until something else is returned
or the list is exhausted.

This is less efficient, but acts as a workaround for FTP-20.

ZX-3063 #done
Test: CQ

Change-Id: Iff1736757417e2b44765ae0083fc6d16022102fe

Test: clang lock annotation checking
Change-Id: I4f708e1449db34f0694ce1635eecd8d272f460f6

The one client now creates a custom zxio object.

Test: No behavior change
Change-Id: Ie3debcbe91908553b56033d9425fc32b6ffe5c03

Test: no functional change
Change-Id: Iea44b507d6a7cf13bb7011789a1dbb09a91b2dac

Test: Adding and binding devices works on arm qemu asan
Change-Id: I647a45aa07f894ec6ab2e6a845ff88ce6eacaa8b

Test: No functional change

Change-Id: Ie95b9428875f8434b085a9a93a17d883e8b1c45b

Test: no functional change
Change-Id: I7917af4cdc34d1a432e993c11d82fb67a298f276