This changes also introduces:
- `SocketOptionsHandler` interface which can be implemented by endpoints to
handle endpoint specific behavior on SetSockOpt. This is analogous to what
Linux does.
- `DefaultSocketOptionsHandler` which is a default implementation of the above.
This is embedded in all endpoints so that we don't have to uselessly
implement empty functions. Endpoints with specific behavior can override the
embedded method by manually defining its own implementation.
PiperOrigin-RevId: 343158301
This change also makes the following fixes:
- Make SocketOptions use atomic operations instead of having to acquire/drop
locks upon each get/set option.
- Make documentation more consistent.
- Remove tcpip.SocketOptions from socketOpsCommon because it already exists
in transport.Endpoint.
- Refactors get/set socket options tests to be easily extendable.
PiperOrigin-RevId: 343103780
In UDP endpoint.Write() sendUDP is called with e.mu Rlocked. But if this happens
to send a datagram over loopback which ends up generating an ICMP response of
say ErrNoPortReachable, the handling of the response in HandleControlPacket also
acquires e.mu using RLock. This is mostly fine unless there is a competing
caller trying to acquire e.mu in exclusive mode using Lock(). This will deadlock
as a caller waiting in Lock() disallows an new RLocks() to ensure it can
actually acquire the Lock.
This is documented here https://golang.org/pkg/sync/#RWMutex.
This change releases the endpoint mutex before calling sendUDP to resolve the
possibility of the deadlock.
Reported-by: syzbot+537989797548c66e8ee3@syzkaller.appspotmail.com
Reported-by: syzbot+eb0b73b4ab486f7673ba@syzkaller.appspotmail.com
PiperOrigin-RevId: 342894148
Fixes the behaviour of SO_ERROR for tcp sockets where in linux it returns
sk->sk_err and if sk->sk_err is 0 then it returns sk->sk_soft_err. In gVisor TCP
we endpoint.HardError is the equivalent of sk->sk_err and endpoint.LastError
holds soft errors. This change brings this into alignment with Linux such that
both hard/soft errors are cleared when retrieved using getsockopt(.. SO_ERROR)
is called on a socket.
Fixes#3812
PiperOrigin-RevId: 342868552
As part of this, change Task.interrupted() to not drain Task.interruptChan, and
do so explicitly using new function Task.unsetInterrupted() instead.
PiperOrigin-RevId: 342768365
Store all the socket level options in a struct and call {Get/Set}SockOpt on
this struct. This will avoid implementing socket level options on all
endpoints. This CL contains implementing one socket level option for tcp and
udp endpoints.
PiperOrigin-RevId: 342203981
This Notify was added as part of cl/279106406; but notifying `EventHUp`
in `FIN_WAIT2` is incorrect, as we want to only notify later on
`TIME_WAIT` or a reset. However, we do need to notify any blocked
waiters of an activity on the endpoint with `EventIn`|`EventOut`.
PiperOrigin-RevId: 341490913
cl/340002915 modified the code to return EADDRNOTAVAIL if connect
is called for a localhost address which isn't set.
But actually, Linux returns EADDRNOTAVAIL for ipv6 addresses and ENETUNREACH
for ipv4 addresses.
Updates #4735
PiperOrigin-RevId: 341479129
Writes to pipes of size < PIPE_BUF are guaranteed to be atomic, so writes
larger than that will return EAGAIN if the pipe has capacity < PIPE_BUF.
Writes to eventfds will return EAGAIN if the write would cause the eventfd
value to go over the max.
In both such cases, calling Ready() on the FD will return true (because it is
possible to write), but specific kinds of writes will in fact return EAGAIN.
This CL fixes an infinite loop in splice and sendfile (VFS1 and VFS2) by
forcing skipping the readiness check for the outfile in send, splice, and tee.
PiperOrigin-RevId: 341102260
- Disable saving in tests that wait for EINTR.
- Do not execute async-signal-unsafe code after fork() (see fork(2)'s manpage,
"After a fork in a multithreaded program ...")
- Check for errors returned by semctl(GETZCNT).
PiperOrigin-RevId: 340901353
The active_closefd has to be shutdown only for write,
otherwise the second poll will always return immediately.
The second poll should not be called from a separate thread.
PiperOrigin-RevId: 340319071
In the docker container, the ipv6 loopback address is not set,
and connect("::1") has to return ENEADDRNOTAVAIL in this case.
Without this fix, it returns EHOSTUNREACH.
PiperOrigin-RevId: 340002915
kernel.copyContext{t} cannot be used outside of t's task goroutine, for three
reasons:
- t.CopyScratchBuffer() is task-goroutine-local.
- Calling t.MemoryManager() without running on t's task goroutine or locking
t.mu violates t.MemoryManager()'s preconditions.
- kernel.copyContext passes t as context.Context to MM IO methods, which is
illegal outside of t's task goroutine (cf. kernel.Task.Value()).
Fix this by splitting AsCopyContext() into CopyContext() (which takes an
explicit context.Context and is usable outside of the task goroutine) and
OwnCopyContext() (which uses t as context.Context, but is only usable by t's
task goroutine).
PiperOrigin-RevId: 339933809
This change wakes up any waiters when we receive an ICMP port unreachable
control packet on an UDP socket as well as sets waiter.EventErr in
the result returned by Readiness() when e.lastError is not nil.
The latter is required where an epoll()/poll() is done after the error
is already handled since we will never notify again in such cases.
PiperOrigin-RevId: 339370469
This PR implements /proc/[pid]/mem for `pkg/sentry/fs` (refer to #2716) and `pkg/sentry/fsimpl`.
@majek
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/4060 from lnsp:proc-pid-mem 2caf9021254646f441be618a9bb5528610e44d43
PiperOrigin-RevId: 339369629
Inode number consistency checks are now skipped in save/restore tests for
reasons described in greatest detail in StatTest.StateDoesntChangeAfterRename.
They pass in VFS1 due to the bug described in new test case
SimpleStatTest.DifferentFilesHaveDifferentDeviceInodeNumberPairs.
Fixes#1663
PiperOrigin-RevId: 338776148
Handle "Resource temporarily unavailable" EAGAIN errors with a select
call before calling recvmsg.
Also rename similar helper call from "RecvMsgTimeout" to "RecvTimeout",
because it calls "recv".
PiperOrigin-RevId: 338761695
The SO_ACCEPTCONN option is used only on getsockopt(). When this option is
specified, getsockopt() indicates whether socket listening is enabled for
the socket. A value of zero indicates that socket listening is disabled;
non-zero that it is enabled.
PiperOrigin-RevId: 338703206
Earlier the count was dropped only after calling e.deliverAccepted. This lead to
an issue where there were no connections in SYN-RCVD state for the listening
endpoint but e.synRcvdCount would not be zero because it was being reduced only
when handleSynSegment returned after deliverAccepted returned.
This issue is seen when the Nth SYN for a listen backlog of size N which would
cause the listen backlog to be full gets dropped occasionally. This happens when
the new SYN comes at when the previous completed endpoint has been delivered to
the accept queue but the synRcvdCount hasn't yet been decremented because the
goroutine running handleSynSegment has not yet completed.
PiperOrigin-RevId: 338690646
Our current reference leak checker uses finalizers to verify whether an object
has reached zero references before it is garbage collected. There are multiple
problems with this mechanism, so a rewrite is in order.
With finalizers, there is no way to guarantee that a finalizer will run before
the program exits. When an unreachable object with a finalizer is garbage
collected, its finalizer will be added to a queue and run asynchronously. The
best we can do is run garbage collection upon sandbox exit to make sure that
all finalizers are enqueued.
Furthermore, if there is a chain of finalized objects, e.g. A points to B
points to C, garbage collection needs to run multiple times before all of the
finalizers are enqueued. The first GC run will register the finalizer for A but
not free it. It takes another GC run to free A, at which point B's finalizer
can be registered. As a result, we need to run GC as many times as the length
of the longest such chain to have a somewhat reliable leak checker.
Finally, a cyclical chain of structs pointing to one another will never be
garbage collected if a finalizer is set. This is a well-known issue with Go
finalizers (https://github.com/golang/go/issues/7358). Using leak checking on
filesystem objects that produce cycles will not work and even result in memory
leaks.
The new leak checker stores reference counted objects in a global map when
leak check is enabled and removes them once they are destroyed. At sandbox
exit, any remaining objects in the map are considered as leaked. This provides
a deterministic way of detecting leaks without relying on the complexities of
finalizers and garbage collection.
This approach has several benefits over the former, including:
- Always detects leaks of objects that should be destroyed very close to
sandbox exit. The old checker very rarely detected these leaks, because it
relied on garbage collection to be run in a short window of time.
- Panics if we forgot to enable leak check on a ref-counted object (we will try
to remove it from the map when it is destroyed, but it will never have been
added).
- Can store extra logging information in the map values without adding to the
size of the ref count struct itself. With the size of just an int64, the ref
count object remains compact, meaning frequent operations like IncRef/DecRef
are more cache-efficient.
- Can aggregate leak results in a single report after the sandbox exits.
Instead of having warnings littered in the log, which were
non-deterministically triggered by garbage collection, we can print all
warning messages at once. Note that this could also be a limitation--the
sandbox must exit properly for leaks to be detected.
Some basic benchmarking indicates that this change does not significantly
affect performance when leak checking is enabled, which is understandable
since registering/unregistering is only done once for each filesystem object.
Updates #1486.
PiperOrigin-RevId: 338685972
- Check the sticky bit in overlay.filesystem.UnlinkAt(). Fixes
StickyTest.StickyBitPermDenied.
- When configuring a VFS2 overlay in runsc, copy the lower layer's root
owner/group/mode to the upper layer's root (as in the VFS1 equivalent,
boot.addOverlay()). This makes the overlay root owned by UID/GID 65534 with
mode 0755 rather than owned by UID/GID 0 with mode 01777. Fixes
CreateTest.CreateFailsOnUnpermittedDir, which assumes that the test cannot
create files in /.
- MknodTest.UnimplementedTypesReturnError assumes that the creation of device
special files is not supported. However, while the VFS2 gofer client still
doesn't support device special files, VFS2 tmpfs does, and in the overlay
test dimension mknod() targets a tmpfs upper layer. The test initially has
all capabilities, including CAP_MKNOD, so its creation of these files
succeeds. Constrain these tests to VFS1.
- Rename overlay.nonDirectoryFD to overlay.regularFileFD and only use it for
regular files, using the original FD for pipes and device special files. This
is more consistent with Linux (which gets the original inode_operations, and
therefore file_operations, for these file types from ovl_fill_inode() =>
init_special_inode()) and fixes remaining mknod and pipe tests.
- Read/write 1KB at a time in PipeTest.Streaming, rather than 4 bytes. This
isn't strictly necessary, but it makes the test less obnoxiously slow on
ptrace.
Fixes#4407
PiperOrigin-RevId: 337971042
ualarm(2) is obsolete. Move IntervalTimer into a test util, where it can be
used by flock tests.
These tests were flaky with TSAN, probably because it slowed the tests down
enough that the alarm was expiring before flock() was called. Use an interval
timer so that even if we miss the first alarm (or more), flock() is still
guaranteed to be interrupted.
PiperOrigin-RevId: 337578751
The fix in commit 028e045da9 was incorrect as
it can cause the right edge of the window to shrink when we announce
a zero window due to receive buffer being full as its done before the check
for seeing if the window is being shrunk because of the selected window.
Further the window was calculated purely on available space but in cases where
we are getting full sized segments it makes more sense to use the actual bytes
being held. This CL changes to use the lower of the total available space vs
the available space in the maximal window we could advertise minus the actual
payload bytes being held.
This change also cleans up the code so that the window selection logic is
not duplicated between getSendParams() and windowCrossedACKThresholdLocked.
PiperOrigin-RevId: 336404827
- When the KCOV_ENABLE_TRACE ioctl is called with the trace kind KCOV_TRACE_PC,
the kcov mode should be set to KCOV_*MODE*_TRACE_PC.
- When the owning task of kcov exits, the memory mapping should not be cleared
so it can be used by other tasks.
- Add more tests (also tested on native Linux kcov).
PiperOrigin-RevId: 335202585
Adds support for the IPv6-compatible redirect target. Redirection is a limited
form of DNAT, where the destination is always the localhost.
Updates #3549.
PiperOrigin-RevId: 334698344
There is a new random issue on some Arm64 machines.
This scene can be summarized as following:
Sometimes, the content of the func() pointer is still 0 opcode.
The probability of this kind of issue is very low,
currently only available on some machines.
After inserting a simple memory barrier, this issue was gone.
The code to directly use the memory barrier is as follows:
memcpy(reinterpret_cast<void*>(addr), machine_code, sizeof(machine_code));
isb()
func = reinterpret_cast<uint32_t (*)(void)>(addr);
Signed-off-by: Bin Lu <bin.lu@arm.com>
Like matchers, targets should use a module-like register/lookup system. This
replaces the brittle switch statements we had before.
The only behavior change is supporing IPT_GET_REVISION_TARGET. This makes it
much easier to add IPv6 redirect in the next change.
Updates #3549.
PiperOrigin-RevId: 334469418
Previously, we did not check the kcov mode when performing task work. As a
result, disabling kcov did not do anything.
Also avoid expensive atomic RMW when consuming coverage data. We don't need the
swap if the value is already zero (which is most of the time), and it is ok if
there are slight inconsistencies due to a race between coverage data generation
(incrementing the value) and consumption (reading a nonzero value and writing
zero).
PiperOrigin-RevId: 334049207
Fabricio Voznika
Ayush Ranjan
Bhasker Hariharan
Jamie Liu
Mithun Iyer
Nayana Bidari
Andrei Vagin
Jing Chen
Nicolas Lacasse
Ghanan Gowripalan
Ian Lewis
gVisor bot
Lennart
Dean Deng
Zach Koopmans
Kevin Krakauer
Bin Lu