We were previously statically linking libcutils into adbd for several
different reasons, which were addressed as follows:
socket functions: extracted to a statically linked libcutils_network
fs_config: wrapped with a shared library on /system
ATRACE: deleted the single use in adbd
Bug: http://b/150032044
Test: treehugger
Change-Id: I821fa174cfcbfa8e29a4be10de4016b817adbaf8
Merged-In: I821fa174cfcbfa8e29a4be10de4016b817adbaf8
(cherry picked from commit a9b62d5452)
This command will be sent by adbd to notify the client that the
connection will be over TLS.
When client connects, it will send the CNXN packet, as usual. If the
server connection has TLS enabled, it will send the A_STLS packet
(regardless of whether auth is required). At this point, the client's
only valid response is to send a A_STLS packet. Once both sides have
exchanged the A_STLS packet, both will start the TLS handshake.
If auth is required, then the client will receive a CertificateRequest
with a list of known public keys (SHA256 hash) that it can use in its
certificate. Otherwise, the list will be empty and the client can assume
that either any key will work, or none will work.
If the handshake was successful, the server will send the CNXN packet
and the usual adb protocol is resumed over TLS. If the handshake failed,
both sides will disconnect, as there's no point to retry because the
server's known keys have already been communicated.
Bug: 111434128
Test: WIP; will add to adb_test.py/adb_device.py.
Enable wireless debugging in the Settings, then 'adb connect
<ip>:<port>'. Connection should succeed if key is in keystore. Used
wireshark to check for packet encryption.
Change-Id: I3d60647491c6c6b92297e4f628707a6457fa9420
Try to not allocate as many blocks on the heap, and reuse
memory instead of copying it
Get rid of unique_ptr and shared_ptr where possible, move
the Block objects themselves
Overall this reduces the time spent in memcpy() from 30% to
15% of the whole 'adb push' command, and gets rid of about 5%
of the time spent in the malloc/free calls
Test: builds
Change-Id: I8995115274b6f08a4df13c58183c928ef384a767
Read until we receive FUNCTIONFS_BIND after every USB descriptor setup,
instead of only after initial creation of the control fd, so we properly
handle a rebind caused by a change of the USB configuration.
This reverts commit 55b31933f41b1c550b788fa2054010c51f56111b.
Bug: http://b/133132986
Bug: http://b/135155652
Test: manually enabled/disabled MTP with nonblocking_ffs enabled and disabled
Change-Id: Ibb3f46b8f84854ede47a669a85e07e16a5dd6008
Merged-In: Ibb3f46b8f84854ede47a669a85e07e16a5dd6008
(cherry picked from commit b00e66f754)
These CHECKs are expected to happen if the client does the wrong thing,
so we probably shouldn't be aborting in adbd.
Bug: http://b/133872605
Bug: http://b/134695864
Test: treehugger
Change-Id: I875b87a06ba9d399db26f8af1ad47ccf34e9445a
Merged-In: I875b87a06ba9d399db26f8af1ad47ccf34e9445a
(cherry picked from commit fc0f80ae49)
Failing to write to the endpoint results in subsequent reads on ep0 to
fail with EL2HLT, so do an empty write to fulfill the transfer.
Bug: http://b/131867920
Test: manually tested with modified auto client
Change-Id: If2eec162ca4b31a9974c7bd5802be51cee9e2708
It seems like we're blowing up when receiving a control transfer that's
intended for Android Auto, because we're not expecting to get the data
for the control transfer in a subsequent read.
Bug: http://b/131867920
Test: none
Change-Id: Icfd642e6dfc02d2ccbdd60c39f89e534298c944d
Kernel does not accept non-zero value of iocb->aio_reserved2.
Bug: 132803232
Test: initialize malloc() memory to non-zero pattern and see what breaks
Change-Id: I65a7e89e3a2c1ba79df1dc2d011d6c76c41afb81
(cherry picked from commit fe7eca7b8f)
Also, improve the logging in the cases where we do abort.
Bug: http://b/131867920
Test: treehugger
Change-Id: If8ec9f4614ce146e6dbd21cc77587ea81658199b
(cherry picked from commit 2916e148d9)
A previous patch intended to make receiving multiple FUNCTIONFS_ENABLEs
non-fatal, but failed to do so because we would try to spawn another
worker thread and hit a different assertion.
Bug: http://b/130638368
Test: echo mem | adb shell "su 0 sh -c 'cat > /sys/power/state'"
Change-Id: I53456112244d8b4f7d26df6ec6961389fca70498
(cherry picked from commit 910ce0ff08)
Each operation we submit results in a contiguous allocation in the
kernel which can trigger OOM-killing. Tune the queue depths to reduce
the chances of this happening, with a minimal impact on performance.
Bug: http://b/126582877
Test: ./benchmark_device.py with blueline
Change-Id: I4b0f83d54a9a9a69f0ab48322d78fb137c94dd35
(cherry picked from commit 28293f16a8)
Some USB controllers only support 16k writes, so drop down to that.
Bug: http://b/130622010
Test: treehugger
Change-Id: I836d5b2d0d6fcae05f290af11a6a19c6e8d7c1ba
(cherry picked from commit 770a6a4d46)
It seems like we don't actually always get these events in order, so
demote them to a log and restart the connection instead of aborting.
Bug: http://b/129464137
Test: mma
Change-Id: I7dffbf62f9dea665cf9f9f9e1bd18d444b6905ec
Occasionally, after we open the functionfs endpoint and write our
descriptors to it, we end up in a state where we never receive any
events from ep0, presumably due to a race against some other part of
the system, since adding sleeps reduces the likelihood. Add an
(arbitrarily chosen) timeout for the initial FUNCTIONFS_BIND we read,
and retry on failure.
Bug: http://b/128679787
Test: unplug and replug a blueline 14,000 times with a usb switch board
Change-Id: I4b3d5d552179ee7f3fefcc15f9575f793926f784
Monitor for FUNCTIONFS_UNBIND as well, so that in the case where we get
FUNCTIONFS_BIND, FUNCTIONFS_UNBIND, FUNCTIONFS_BIND, we don't trigger an
assertion failure from seeing two FUNCTIONFS_BINDs.
Bug: http://b/129134256
Test: manual
Change-Id: I80af5f4b833513e932262638b9f8d76bbcb35504
If we fail to submit writes for some reason (e.g. the USB cable was
unplugged), another thread that's waiting on the write mutex can enter
SubmitWrites and attempt to resubmit the writes that we already failed
to submit, leading to a failed assertion of !IoBlock::pending.
Increment writes_submitted_ before actually calling io_submit, so we
skip over these writes and fall through to exit.
Bug: http://b/129134256
Test: manually unplugged a blueline
Change-Id: I2fd0034e45db22c8f637c81039ce686b7aa6a03b
Allow disabling ro.adb.nonblocking_ffs via an ro system property, which
gets overridden by the currently supported persist.adb.nonblocking_ffs.
Bug: http://b/128679787
Test: mma
Change-Id: I28891d2da129e9654bf63ab8a4c330c4d2f5b336
Previously, we were initializing IoBlock::pending when submitting a read,
but leaving it uninitialized when doing a write, which would sometimes
result in an assertion firing when we received a previously allocated
block of memory that happened to have a nonzero value in pending.
Test: treehugger
Change-Id: I434c65f554ceed313fea9cfcc98788908f3ca8bc
If a device is unplugged, a BIND will happen without an ENABLE until the
device is plugged in, which would previously result in adbd crashing
every 5 seconds due to an attempt to join a thread that hadn't been
started.
Bug: http://b/128255355
Test: manually observed adbd via serial cable while unplugged
Change-Id: I90d6384e7e829df1ffaf73a95f2418a2ff97fb03
Reduce the operation sizes to sizes that are small enough that the
kernel will run the OOM killer to try to make them succeed. Compensate
for the smaller operation sizes by increasing the queue depth.
Bug: http://b/126582877
Test: benchmark_device.py
Change-Id: I4d351f56c8f83fea882614238cc3fec8ba1e9bf9
The kernel does a kmalloc of the write size for each write during
io_submit, which can lead to failure if the kernel's heap is fragmented
such that it can't allocate a contiguous 1MB chunk. Partition writes
into 16 page chunks which are much more likely to succeed.
Bug: http://b/126582877
Test: test_device.py
Change-Id: Ib2d48561594e1d81f1a2f62e34eaa40265ac47c4
Switch to the nonblocking FFS implementation that fixes several race
conditions on connect/disconnect.
Test: treehugger
Change-Id: I91fdb9b27152078cb38e610861a05f1fa95aa3a2
Detect kernels that don't support AIO on functionfs endpoints and fall
back to the old legacy implementation.
Test: treehugger
Change-Id: Ice28dcd40c9730ac2dd5e468d065b41c0e076146
Implement a Connection that implements a nonblocking interface to
functionfs, to replace the existing implementation that uses two
threads that loop and call read and write respectively. The existing
implementation is vulnerable to a race condition that can occur when a
connection is terminated, where one thread can notice failure and
complete reinitialization of the USB endpoints before the other thread
noticed anything went wrong, resulting in either the first packet
coming from the other end disappearing in to the void, or the other end
getting a packet of garbage.
As a side benefit, this improves performance on walleye from:
push 100MiB: 10 runs: median 49.48 MiB/s, mean 50.00 MiB/s, stddev: 2.77 MiB/s
pull 100MiB: 10 runs: median 75.82 MiB/s, mean 76.18 MiB/s, stddev: 6.60 MiB/s
to:
push 100MiB: 10 runs: median 73.90 MiB/s, mean 73.51 MiB/s, stddev: 5.26 MiB/s
pull 100MiB: 10 runs: median 105.90 MiB/s, mean 107.19 MiB/s, stddev: 6.10 MiB/s
Test: python test_device.py
Change-Id: I9b77c1057965edfef739ed9736e5d76613adf60a
Extract common functionality that will be useful for implementing a
nonblocking Connection for functionfs.
Test: mma
Change-Id: Iae4f3d55b2f56568d4a5138db9a2af5622eb83ad
Fastboot protocol doesn't include zero packets,
so make it possible to configure that.
Allow fastbootd to see how many bytes the handle
can read/write at once.
Test: adb works
Bug: 78793464
Change-Id: I31e444f384d9d0cdf1ea936439b2028f8151c3b8
Previously, read and write would return 0
on success. Now it will return the number
of bytes read/write. This is more consistent
with other usb handles and is needed in order
to handle partial packets (for fastbootd).
Update usb_write in other usb handles
to return amount written.
Change transport_usb accordingly.
Test: adb works
Bug: 78793464
Change-Id: If07ff05fbc8120343f20661475d34f4e5ff805de
Fastbootd will reuses adb's functionfs transport
implementation.
Move it to daemon/include as well so it can be accessed
with "adbd/usb.h". Otherwise usb.h will conflict with
other imports.
Test: adb builds and works
Bug: 78793464
Change-Id: If3ba190d5c74b5f3633411f0484195e5a2a34d44
Remove some redundancy by moving io_setup
into aio_block_init. The io_contexts will
now live till the end of the program, similar
to the rest of usb_handle
Test: adb works
Bug: 78793464
Change-Id: I910a39d4395645f1ff231f265c285a72893a0ac6
io_submit sleeps waiting for the gadget
to be enabled. Currently if that sleep
is interrupted it will shut down adb,
causing it to have to start back up again.
Rather than return EINTR if interrupted,
io_submit actually completes and the EINTR
is found later when looking through events.
Since an io that is interrupted will be
small anyway, add a loop to retry small ios.
Also upgrade aio logs in accordance with
their importance.
Fixes: 75981904
Test: adb works, logs show successful interrupt handling
Change-Id: I35973fce130ee849ce59fef80d15b65afb816ba4
(cherry picked from commit 6e9a327504)
All devices that previously used ENDPOINT_ALLOC
are on the new async io routines. None of the devices
using aio_compat have ENDPOINT_ALLOC so remove the
code to stop logging failures.
Fixes: 74213465
Test: adb works
Change-Id: I0c903eb76b006b6bcce48cec24f5353fa47cc128
(cherry picked from commit 9a1be3ad0e)
This change opens the control, bulk-in, and bulk-out FunctionFS
endpoints with O_RDONLY or O_WRONLY.
Bug: 70349025
Test: Android on Chrome OS can still use ADB
Test: aosp_sailfish-userdebug can still use ADB
Change-Id: I3895b626cf7cec8437c36bdc6c6e6055aaf3777c
This method works around the downsides of
ENDPOINT_ALLOC, namely that it is not affected
by memory fragmentation and it uses an upstream
interface.
Also add libasyncio to provide the necessary syscalls
to both adb and mtp.
Add some small optimizations to file_sync.
Bug: 37916658
Test: run adb push/pull
Change-Id: If3b3be02b5e2d4f9cffec1b8ddc02a5768a51a1f
Josh Gao