OSDN Git Service

tomoyo/tomoyo-test1.git
4 years agoio_uring: honor original task RLIMIT_FSIZE
Jens Axboe [Fri, 20 Mar 2020 17:23:41 +0000 (11:23 -0600)]
io_uring: honor original task RLIMIT_FSIZE

With the previous fixes for number of files open checking, I added some
debug code to see if we had other spots where we're checking rlimit()
against the async io-wq workers. The only one I found was file size
checking, which we should also honor.

During write and fallocate prep, store the max file size and override
that for the current ask if we're in io-wq worker context.

Cc: stable@vger.kernel.org # 5.1+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: hash dependent work
Pavel Begunkov [Fri, 13 Mar 2020 21:31:05 +0000 (00:31 +0300)]
io-wq: hash dependent work

Enable io-wq hashing stuff for dependent works simply by re-enqueueing
such requests.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: split hashing and enqueueing
Pavel Begunkov [Fri, 13 Mar 2020 21:31:04 +0000 (00:31 +0300)]
io-wq: split hashing and enqueueing

It's a preparation patch removing io_wq_enqueue_hashed(), which
now should be done by io_wq_hash_work() + io_wq_enqueue().

Also, set hash value for dependant works, and do it as late as possible,
because req->file can be unavailable before. This hash will be ignored
by io-wq.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: don't resched if there is no work
Pavel Begunkov [Fri, 13 Mar 2020 21:31:03 +0000 (00:31 +0300)]
io-wq: don't resched if there is no work

This little tweak restores the behaviour that was before the recent
io_worker_handle_work() optimisation patches. It makes the function do
cond_resched() and flush_signals() only if there is an actual work to
execute.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: remove duplicated cancel code
Pavel Begunkov [Fri, 6 Mar 2020 22:15:39 +0000 (01:15 +0300)]
io-wq: remove duplicated cancel code

Deduplicate cancellation parts, as many of them looks the same, as do
e.g.
- io_wqe_cancel_cb_work() and io_wqe_cancel_work()
- io_wq_worker_cancel() and io_work_cancel()

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: fix truncated async read/readv and write/writev retry
Jens Axboe [Wed, 11 Mar 2020 18:27:04 +0000 (12:27 -0600)]
io_uring: fix truncated async read/readv and write/writev retry

Ensure we keep the truncated value, if we did truncate it. If not, we
might read/write more than the registered buffer size.

Also for retry, ensure that we return the truncated mapped value for
the vectorized versions of the read/write commands.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: dual license io_uring.h uapi header
Jens Axboe [Wed, 11 Mar 2020 13:45:46 +0000 (07:45 -0600)]
io_uring: dual license io_uring.h uapi header

This just syncs the header it with the liburing version, so there's no
confusion on the license of the header parts.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: io_uring_enter(2) don't poll while SETUP_IOPOLL|SETUP_SQPOLL enabled
Xiaoguang Wang [Wed, 11 Mar 2020 01:26:09 +0000 (09:26 +0800)]
io_uring: io_uring_enter(2) don't poll while SETUP_IOPOLL|SETUP_SQPOLL enabled

When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, applications don't need
to do io completion events polling again, they can rely on io_sq_thread to do
polling work, which can reduce cpu usage and uring_lock contention.

I modify fio io_uring engine codes a bit to evaluate the performance:
static int fio_ioring_getevents(struct thread_data *td, unsigned int min,
                        continue;
                }

-               if (!o->sqpoll_thread) {
+               if (o->sqpoll_thread && o->hipri) {
                        r = io_uring_enter(ld, 0, actual_min,
                                                IORING_ENTER_GETEVENTS);
                        if (r < 0) {

and use "fio  -name=fiotest -filename=/dev/nvme0n1 -iodepth=$depth -thread
-rw=read -ioengine=io_uring  -hipri=1 -sqthread_poll=1  -direct=1 -bs=4k
-size=10G -numjobs=1  -time_based -runtime=120"

original codes
--------------------------------------------------------------------
iodepth       |        4 |        8 |       16 |       32 |       64
bw            | 1133MB/s | 1519MB/s | 2090MB/s | 2710MB/s | 3012MB/s
fio cpu usage |     100% |     100% |     100% |     100% |     100%
--------------------------------------------------------------------

with patch
--------------------------------------------------------------------
iodepth       |        4 |        8 |       16 |       32 |       64
bw            | 1196MB/s | 1721MB/s | 2351MB/s | 2977MB/s | 3357MB/s
fio cpu usage |    63.8% |   74.4%% |    81.1% |    83.7% |    82.4%
--------------------------------------------------------------------
bw improve    |     5.5% |    13.2% |    12.3% |     9.8% |    11.5%
--------------------------------------------------------------------

From above test results, we can see that bw has above 5.5%~13%
improvement, and fio process's cpu usage also drops much. Note this
won't improve io_sq_thread's cpu usage when SETUP_IOPOLL|SETUP_SQPOLL
are both enabled, in this case, io_sq_thread always has 100% cpu usage.
I think this patch will be friendly to applications which will often use
io_uring_wait_cqe() or similar from liburing.

Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: Fix unused function warnings
YueHaibing [Wed, 4 Mar 2020 07:53:52 +0000 (15:53 +0800)]
io_uring: Fix unused function warnings

If CONFIG_NET is not set, gcc warns:

fs/io_uring.c:3110:12: warning: io_setup_async_msg defined but not used [-Wunused-function]
 static int io_setup_async_msg(struct io_kiocb *req,
            ^~~~~~~~~~~~~~~~~~

There are many funcions wraped by CONFIG_NET, move them
together to simplify code, also fix this warning.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Minor tweaks.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add end-of-bits marker and build time verify it
Jens Axboe [Tue, 3 Mar 2020 22:28:17 +0000 (15:28 -0700)]
io_uring: add end-of-bits marker and build time verify it

Not easy to tell if we're going over the size of bits we can shove
in req->flags, so add an end-of-bits marker and a BUILD_BUG_ON()
check for it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: provide means of removing buffers
Jens Axboe [Mon, 2 Mar 2020 23:32:28 +0000 (16:32 -0700)]
io_uring: provide means of removing buffers

We have IORING_OP_PROVIDE_BUFFERS, but the only way to remove buffers
is to trigger IO on them. The usual case of shrinking a buffer pool
would be to just not replenish the buffers when IO completes, and
instead just free it. But it may be nice to have a way to manually
remove a number of buffers from a given group, and
IORING_OP_REMOVE_BUFFERS provides that functionality.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add IOSQE_BUFFER_SELECT support for IORING_OP_RECVMSG
Jens Axboe [Thu, 27 Feb 2020 17:15:42 +0000 (10:15 -0700)]
io_uring: add IOSQE_BUFFER_SELECT support for IORING_OP_RECVMSG

Like IORING_OP_READV, this is limited to supporting just a single
segment in the iovec passed in.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agonet: abstract out normal and compat msghdr import
Jens Axboe [Thu, 27 Feb 2020 15:11:20 +0000 (08:11 -0700)]
net: abstract out normal and compat msghdr import

This splits it into two parts, one that imports the message, and one
that imports the iovec. This allows a caller to only do the first part,
and import the iovec manually afterwards.

No functional changes in this patch.

Acked-by: David Miller <davem@davemloft.net>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add IOSQE_BUFFER_SELECT support for IORING_OP_READV
Jens Axboe [Thu, 27 Feb 2020 14:31:19 +0000 (07:31 -0700)]
io_uring: add IOSQE_BUFFER_SELECT support for IORING_OP_READV

This adds support for the vectored read. This is limited to supporting
just 1 segment in the iov, and is provided just for convenience for
applications that use IORING_OP_READV already.

The iov helpers will be used for IORING_OP_RECVMSG as well.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: support buffer selection for OP_READ and OP_RECV
Jens Axboe [Sun, 23 Feb 2020 23:42:51 +0000 (16:42 -0700)]
io_uring: support buffer selection for OP_READ and OP_RECV

If a server process has tons of pending socket connections, generally
it uses epoll to wait for activity. When the socket is ready for reading
(or writing), the task can select a buffer and issue a recv/send on the
given fd.

Now that we have fast (non-async thread) support, a task can have tons
of pending reads or writes pending. But that means they need buffers to
back that data, and if the number of connections is high enough, having
them preallocated for all possible connections is unfeasible.

With IORING_OP_PROVIDE_BUFFERS, an application can register buffers to
use for any request. The request then sets IOSQE_BUFFER_SELECT in the
sqe, and a given group ID in sqe->buf_group. When the fd becomes ready,
a free buffer from the specified group is selected. If none are
available, the request is terminated with -ENOBUFS. If successful, the
CQE on completion will contain the buffer ID chosen in the cqe->flags
member, encoded as:

(buffer_id << IORING_CQE_BUFFER_SHIFT) | IORING_CQE_F_BUFFER;

Once a buffer has been consumed by a request, it is no longer available
and must be registered again with IORING_OP_PROVIDE_BUFFERS.

Requests need to support this feature. For now, IORING_OP_READ and
IORING_OP_RECV support it. This is checked on SQE submission, a CQE with
res == -EOPNOTSUPP will be posted if attempted on unsupported requests.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add IORING_OP_PROVIDE_BUFFERS
Jens Axboe [Sun, 23 Feb 2020 23:41:33 +0000 (16:41 -0700)]
io_uring: add IORING_OP_PROVIDE_BUFFERS

IORING_OP_PROVIDE_BUFFERS uses the buffer registration infrastructure to
support passing in an addr/len that is associated with a buffer ID and
buffer group ID. The group ID is used to index and lookup the buffers,
while the buffer ID can be used to notify the application which buffer
in the group was used. The addr passed in is the starting buffer address,
and length is each buffer length. A number of buffers to add with can be
specified, in which case addr is incremented by length for each addition,
and each buffer increments the buffer ID specified.

No validation is done of the buffer ID. If the application provides
buffers within the same group with identical buffer IDs, then it'll have
a hard time telling which buffer ID was used. The only restriction is
that the buffer ID can be a max of 16-bits in size, so USHRT_MAX is the
maximum ID that can be used.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: buffer registration infrastructure
Jens Axboe [Sun, 23 Feb 2020 23:23:11 +0000 (16:23 -0700)]
io_uring: buffer registration infrastructure

This just prepares the ring for having lists of buffers associated with
it, that the application can provide for SQEs to consume instead of
providing their own.

The buffers are organized by group ID.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring/io-wq: forward submission ref to async
Pavel Begunkov [Wed, 4 Mar 2020 13:14:12 +0000 (16:14 +0300)]
io_uring/io-wq: forward submission ref to async

First it changes io-wq interfaces. It replaces {get,put}_work() with
free_work(), which guaranteed to be called exactly once. It also enforces
free_work() callback to be non-NULL.

io_uring follows the changes and instead of putting a submission reference
in io_put_req_async_completion(), it will be done in io_free_work(). As
removes io_get_work() with corresponding refcount_inc(), the ref balance
is maintained.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: optimise out *next_work() double lock
Pavel Begunkov [Wed, 4 Mar 2020 13:14:11 +0000 (16:14 +0300)]
io-wq: optimise out *next_work() double lock

When executing non-linked hashed work, io_worker_handle_work()
will lock-unlock wqe->lock to update hash, and then immediately
lock-unlock to get next work. Optimise this case and do
lock/unlock only once.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: optimise locking in io_worker_handle_work()
Pavel Begunkov [Wed, 4 Mar 2020 13:14:10 +0000 (16:14 +0300)]
io-wq: optimise locking in io_worker_handle_work()

There are 2 optimisations:
- Now, io_worker_handler_work() do io_assign_current_work() twice per
request, and each one adds lock/unlock(worker->lock) pair. The first is
to reset worker->cur_work to NULL, and the second to set a real work
shortly after. If there is a dependant work, set it immediately, that
effectively removes the extra NULL'ing.

- And there is no use in taking wqe->lock for linked works, as they are
not hashed now. Optimise it out.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: shuffle io_worker_handle_work() code
Pavel Begunkov [Wed, 4 Mar 2020 13:14:09 +0000 (16:14 +0300)]
io-wq: shuffle io_worker_handle_work() code

This is a preparation patch, it adds some helpers and makes
the next patches cleaner.

- extract io_impersonate_work() and io_assign_current_work()
- replace @next label with nested do-while
- move put_work() right after NULL'ing cur_work.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: get next work with submission ref drop
Pavel Begunkov [Tue, 3 Mar 2020 18:33:13 +0000 (21:33 +0300)]
io_uring: get next work with submission ref drop

If after dropping the submission reference req->refs == 1, the request
is done, because this one is for io_put_work() and will be dropped
synchronously shortly after. In this case it's safe to steal a next
work from the request.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove @nxt from handlers
Pavel Begunkov [Tue, 3 Mar 2020 18:33:12 +0000 (21:33 +0300)]
io_uring: remove @nxt from handlers

There will be no use for @nxt in the handlers, and it's doesn't work
anyway, so purge it

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: make submission ref putting consistent
Pavel Begunkov [Tue, 3 Mar 2020 18:33:11 +0000 (21:33 +0300)]
io_uring: make submission ref putting consistent

The rule is simple, any async handler gets a submission ref and should
put it at the end. Make them all follow it, and so more consistent.

This is a preparation patch, and as io_wq_assign_next() currently won't
ever work, this doesn't care to use io_put_req_find_next() instead of
io_put_req().

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
refcount_inc_not_zero() -> refcount_inc() fix.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: clean up io_close
Pavel Begunkov [Mon, 2 Mar 2020 20:45:16 +0000 (23:45 +0300)]
io_uring: clean up io_close

Don't abuse labels for plain and straightworward code.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: Ensure mask is initialized in io_arm_poll_handler
Nathan Chancellor [Mon, 2 Mar 2020 23:01:19 +0000 (16:01 -0700)]
io_uring: Ensure mask is initialized in io_arm_poll_handler

Clang warns:

fs/io_uring.c:4178:6: warning: variable 'mask' is used uninitialized
whenever 'if' condition is false [-Wsometimes-uninitialized]
        if (def->pollin)
            ^~~~~~~~~~~
fs/io_uring.c:4182:2: note: uninitialized use occurs here
        mask |= POLLERR | POLLPRI;
        ^~~~
fs/io_uring.c:4178:2: note: remove the 'if' if its condition is always
true
        if (def->pollin)
        ^~~~~~~~~~~~~~~~
fs/io_uring.c:4154:15: note: initialize the variable 'mask' to silence
this warning
        __poll_t mask, ret;
                     ^
                      = 0
1 warning generated.

io_op_defs has many definitions where pollin is not set so mask indeed
might be uninitialized. Initialize it to zero and change the next
assignment to |=, in case further masks are added in the future to avoid
missing changing the assignment then.

Fixes: d7718a9d25a6 ("io_uring: use poll driven retry for files that support it")
Link: https://github.com/ClangBuiltLinux/linux/issues/916
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove io_prep_next_work()
Pavel Begunkov [Sat, 29 Feb 2020 19:56:10 +0000 (22:56 +0300)]
io_uring: remove io_prep_next_work()

io-wq cares about IO_WQ_WORK_UNBOUND flag only while enqueueing, so
it's useless setting it for a next req of a link. Thus, removed it
from io_prep_linked_timeout(), and inline the function.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove extra nxt check after punt
Pavel Begunkov [Sat, 29 Feb 2020 19:48:24 +0000 (22:48 +0300)]
io_uring: remove extra nxt check after punt

After __io_queue_sqe() ended up in io_queue_async_work(), it's already
known that there is no @nxt req, so skip the check and return from the
function.

Also, @nxt initialisation now can be done just before
io_put_req_find_next(), as there is no jumping until it's checked.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: use poll driven retry for files that support it
Jens Axboe [Sat, 15 Feb 2020 05:23:12 +0000 (22:23 -0700)]
io_uring: use poll driven retry for files that support it

Currently io_uring tries any request in a non-blocking manner, if it can,
and then retries from a worker thread if we get -EAGAIN. Now that we have
a new and fancy poll based retry backend, use that to retry requests if
the file supports it.

This means that, for example, an IORING_OP_RECVMSG on a socket no longer
requires an async thread to complete the IO. If we get -EAGAIN reading
from the socket in a non-blocking manner, we arm a poll handler for
notification on when the socket becomes readable. When it does, the
pending read is executed directly by the task again, through the io_uring
task work handlers. Not only is this faster and more efficient, it also
means we're not generating potentially tons of async threads that just
sit and block, waiting for the IO to complete.

The feature is marked with IORING_FEAT_FAST_POLL, meaning that async
pollable IO is fast, and that poll<link>other_op is fast as well.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: mark requests that we can do poll async in io_op_defs
Jens Axboe [Thu, 20 Feb 2020 16:59:44 +0000 (09:59 -0700)]
io_uring: mark requests that we can do poll async in io_op_defs

Add a pollin/pollout field to the request table, and have commands that
we can safely poll for properly marked.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add per-task callback handler
Jens Axboe [Mon, 17 Feb 2020 16:52:41 +0000 (09:52 -0700)]
io_uring: add per-task callback handler

For poll requests, it's not uncommon to link a read (or write) after
the poll to execute immediately after the file is marked as ready.
Since the poll completion is called inside the waitqueue wake up handler,
we have to punt that linked request to async context. This slows down
the processing, and actually means it's faster to not use a link for this
use case.

We also run into problems if the completion_lock is contended, as we're
doing a different lock ordering than the issue side is. Hence we have
to do trylock for completion, and if that fails, go async. Poll removal
needs to go async as well, for the same reason.

eventfd notification needs special case as well, to avoid stack blowing
recursion or deadlocks.

These are all deficiencies that were inherited from the aio poll
implementation, but I think we can do better. When a poll completes,
simply queue it up in the task poll list. When the task completes the
list, we can run dependent links inline as well. This means we never
have to go async, and we can remove a bunch of code associated with
that, and optimizations to try and make that run faster. The diffstat
speaks for itself.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: store io_kiocb in wait->private
Jens Axboe [Mon, 10 Feb 2020 16:07:05 +0000 (09:07 -0700)]
io_uring: store io_kiocb in wait->private

Store the io_kiocb in the private field instead of the poll entry, this
is in preparation for allowing multiple waitqueues.

No functional changes in this patch.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agotask_work_run: don't take ->pi_lock unconditionally
Oleg Nesterov [Tue, 18 Feb 2020 15:50:18 +0000 (16:50 +0100)]
task_work_run: don't take ->pi_lock unconditionally

As Peter pointed out, task_work() can avoid ->pi_lock and cmpxchg()
if task->task_works == NULL && !PF_EXITING.

And in fact the only reason why task_work_run() needs ->pi_lock is
the possible race with task_work_cancel(), we can optimize this code
and make the locking more clear.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: use BIT for ulong hash
Pavel Begunkov [Fri, 28 Feb 2020 07:36:39 +0000 (10:36 +0300)]
io-wq: use BIT for ulong hash

@hash_map is unsigned long, but BIT_ULL() is used for manipulations.
BIT() is a better match as it returns exactly unsigned long value.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove IO_WQ_WORK_CB
Pavel Begunkov [Fri, 28 Feb 2020 07:36:38 +0000 (10:36 +0300)]
io_uring: remove IO_WQ_WORK_CB

IO_WQ_WORK_CB is used only for linked timeouts, which will be armed
before the work setup (i.e. mm, override creds, etc). The setup
shouldn't take long, so it's ok to arm it a bit later and get rid
of IO_WQ_WORK_CB.

Make io-wq call work->func() only once, callbacks will handle the rest.
i.e. the linked timeout handler will do the actual issue. And as a
bonus, it removes an extra indirect call.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: remove unused IO_WQ_WORK_HAS_MM
Pavel Begunkov [Fri, 28 Feb 2020 07:36:37 +0000 (10:36 +0300)]
io-wq: remove unused IO_WQ_WORK_HAS_MM

IO_WQ_WORK_HAS_MM is set but never used, remove it.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: extract kmsg copy helper
Pavel Begunkov [Fri, 28 Feb 2020 07:36:36 +0000 (10:36 +0300)]
io_uring: extract kmsg copy helper

io_recvmsg() and io_sendmsg() duplicate nonblock -EAGAIN finilising
part, so add helper for that.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: clean io_poll_complete
Pavel Begunkov [Fri, 28 Feb 2020 07:36:35 +0000 (10:36 +0300)]
io_uring: clean io_poll_complete

Deduplicate call to io_cqring_fill_event(), plain and easy

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add splice(2) support
Pavel Begunkov [Mon, 24 Feb 2020 08:32:45 +0000 (11:32 +0300)]
io_uring: add splice(2) support

Add support for splice(2).

- output file is specified as sqe->fd, so it's handled by generic code
- hash_reg_file handled by generic code as well
- len is 32bit, but should be fine
- the fd_in is registered file, when SPLICE_F_FD_IN_FIXED is set, which
is a splice flag (i.e. sqe->splice_flags).

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: add interface for getting files
Pavel Begunkov [Mon, 24 Feb 2020 08:32:44 +0000 (11:32 +0300)]
io_uring: add interface for getting files

Preparation without functional changes. Adds io_get_file(), that allows
to grab files not only into req->file.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agosplice: make do_splice public
Pavel Begunkov [Mon, 24 Feb 2020 08:32:43 +0000 (11:32 +0300)]
splice: make do_splice public

Make do_splice(), so other kernel parts can reuse it

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: remove req->in_async
Pavel Begunkov [Mon, 24 Feb 2020 08:30:18 +0000 (11:30 +0300)]
io_uring: remove req->in_async

req->in_async is not really needed, it only prevents propagation of
@nxt for fast not-blocked submissions. Remove it.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: don't do full *prep_worker() from io-wq
Pavel Begunkov [Mon, 24 Feb 2020 08:30:17 +0000 (11:30 +0300)]
io_uring: don't do full *prep_worker() from io-wq

io_prep_async_worker() called io_wq_assign_next() do many useless checks:
io_req_work_grab_env() was already called during prep, and @do_hashed
is not ever used. Add io_prep_next_work() -- simplified version, that
can be called io-wq.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: don't call work.func from sync ctx
Pavel Begunkov [Mon, 24 Feb 2020 08:30:16 +0000 (11:30 +0300)]
io_uring: don't call work.func from sync ctx

Many operations define custom work.func before getting into an io-wq.
There are several points against:
- it calls io_wq_assign_next() from outside io-wq, that may be confusing
- sync context would go unnecessary through io_req_cancelled()
- prototypes are quite different, so work!=old_work looks strange
- makes async/sync responsibilities fuzzy
- adds extra overhead

Don't call generic path and io-wq handlers from each other, but use
helpers instead

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: io_accept() should hold on to submit reference on retry
Jens Axboe [Thu, 20 Feb 2020 16:59:02 +0000 (09:59 -0700)]
io_uring: io_accept() should hold on to submit reference on retry

Don't drop an early reference, hang on to it and let the caller drop
it. This makes it behave more like "regular" requests.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio_uring: consider any io_read/write -EAGAIN as final
Jens Axboe [Thu, 20 Feb 2020 16:56:08 +0000 (09:56 -0700)]
io_uring: consider any io_read/write -EAGAIN as final

If the -EAGAIN happens because of a static condition, then a poll
or later retry won't fix it. We must call it again from blocking
condition. Play it safe and ensure that any -EAGAIN condition from read
or write must retry from async context.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: remove io_wq_flush and IO_WQ_WORK_INTERNAL
Pavel Begunkov [Mon, 2 Mar 2020 20:46:10 +0000 (23:46 +0300)]
io-wq: remove io_wq_flush and IO_WQ_WORK_INTERNAL

io_wq_flush() is buggy, during cancelation of a flush, the associated
work may be passed to the caller's (i.e. io_uring) @match callback. That
callback is expecting it to be embedded in struct io_kiocb. Cancelation
of internal work probably doesn't make a lot of sense to begin with.

As the flush helper is no longer used, just delete it and the associated
work flag.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoio-wq: fix IO_WQ_WORK_NO_CANCEL cancellation
Pavel Begunkov [Sun, 1 Mar 2020 16:18:19 +0000 (19:18 +0300)]
io-wq: fix IO_WQ_WORK_NO_CANCEL cancellation

To cancel a work, io-wq sets IO_WQ_WORK_CANCEL and executes the
callback. However, IO_WQ_WORK_NO_CANCEL works will just execute and may
return next work, which will be ignored and lost.

Cancel the whole link.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agoLinux 5.6-rc4 v5.6-rc4
Linus Torvalds [Sun, 1 Mar 2020 22:38:46 +0000 (16:38 -0600)]
Linux 5.6-rc4

4 years agoMerge tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 1 Mar 2020 22:35:08 +0000 (16:35 -0600)]
Merge tag 'ext4_for_linus_stable' of git://git./linux/kernel/git/tytso/ext4

Pull ext4 fixes from Ted Ts'o:
 "Two more bug fixes (including a regression) for 5.6"

* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: potential crash on allocation error in ext4_alloc_flex_bg_array()
  jbd2: fix data races at struct journal_head

4 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sun, 1 Mar 2020 21:16:35 +0000 (15:16 -0600)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "More bugfixes, including a few remaining "make W=1" issues such as too
  large frame sizes on some configurations.

  On the ARM side, the compiler was messing up shadow stacks between EL1
  and EL2 code, which is easily fixed with __always_inline"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: VMX: check descriptor table exits on instruction emulation
  kvm: x86: Limit the number of "kvm: disabled by bios" messages
  KVM: x86: avoid useless copy of cpufreq policy
  KVM: allow disabling -Werror
  KVM: x86: allow compiling as non-module with W=1
  KVM: Pre-allocate 1 cpumask variable per cpu for both pv tlb and pv ipis
  KVM: Introduce pv check helpers
  KVM: let declaration of kvm_get_running_vcpus match implementation
  KVM: SVM: allocate AVIC data structures based on kvm_amd module parameter
  arm64: Ask the compiler to __always_inline functions used by KVM at HYP
  KVM: arm64: Define our own swab32() to avoid a uapi static inline
  KVM: arm64: Ask the compiler to __always_inline functions used at HYP
  kvm: arm/arm64: Fold VHE entry/exit work into kvm_vcpu_run_vhe()
  KVM: arm/arm64: Fix up includes for trace.h

4 years agoKVM: VMX: check descriptor table exits on instruction emulation
Oliver Upton [Sat, 29 Feb 2020 19:30:14 +0000 (11:30 -0800)]
KVM: VMX: check descriptor table exits on instruction emulation

KVM emulates UMIP on hardware that doesn't support it by setting the
'descriptor table exiting' VM-execution control and performing
instruction emulation. When running nested, this emulation is broken as
KVM refuses to emulate L2 instructions by default.

Correct this regression by allowing the emulation of descriptor table
instructions if L1 hasn't requested 'descriptor table exiting'.

Fixes: 07721feee46b ("KVM: nVMX: Don't emulate instructions in guest mode")
Reported-by: Jan Kiszka <jan.kiszka@web.de>
Cc: stable@vger.kernel.org
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Jim Mattson <jmattson@google.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoMerge branch 'i2c/for-current-fixed' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 1 Mar 2020 01:16:46 +0000 (19:16 -0600)]
Merge branch 'i2c/for-current-fixed' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "I2C has three driver bugfixes for you. We agreed on the Mac regression
  to go in via I2C"

* 'i2c/for-current-fixed' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  macintosh: therm_windtunnel: fix regression when instantiating devices
  i2c: altera: Fix potential integer overflow
  i2c: jz4780: silence log flood on txabrt

4 years agoext4: potential crash on allocation error in ext4_alloc_flex_bg_array()
Dan Carpenter [Fri, 28 Feb 2020 09:22:56 +0000 (12:22 +0300)]
ext4: potential crash on allocation error in ext4_alloc_flex_bg_array()

If sbi->s_flex_groups_allocated is zero and the first allocation fails
then this code will crash.  The problem is that "i--" will set "i" to
-1 but when we compare "i >= sbi->s_flex_groups_allocated" then the -1
is type promoted to unsigned and becomes UINT_MAX.  Since UINT_MAX
is more than zero, the condition is true so we call kvfree(new_groups[-1]).
The loop will carry on freeing invalid memory until it crashes.

Fixes: 7c990728b99e ("ext4: fix potential race between s_flex_groups online resizing and access")
Reviewed-by: Suraj Jitindar Singh <surajjs@amazon.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20200228092142.7irbc44yaz3by7nb@kili.mountain
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agomacintosh: therm_windtunnel: fix regression when instantiating devices
Wolfram Sang [Tue, 25 Feb 2020 14:12:29 +0000 (15:12 +0100)]
macintosh: therm_windtunnel: fix regression when instantiating devices

Removing attach_adapter from this driver caused a regression for at
least some machines. Those machines had the sensors described in their
DT, too, so they didn't need manual creation of the sensor devices. The
old code worked, though, because manual creation came first. Creation of
DT devices then failed later and caused error logs, but the sensors
worked nonetheless because of the manually created devices.

When removing attach_adaper, manual creation now comes later and loses
the race. The sensor devices were already registered via DT, yet with
another binding, so the driver could not be bound to it.

This fix refactors the code to remove the race and only manually creates
devices if there are no DT nodes present. Also, the DT binding is updated
to match both, the DT and manually created devices. Because we don't
know which device creation will be used at runtime, the code to start
the kthread is moved to do_probe() which will be called by both methods.

Fixes: 3e7bed52719d ("macintosh: therm_windtunnel: drop using attach_adapter")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=201723
Reported-by: Erhard Furtner <erhard_f@mailbox.org>
Tested-by: Erhard Furtner <erhard_f@mailbox.org>
Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Cc: stable@kernel.org # v4.19+
4 years agojbd2: fix data races at struct journal_head
Qian Cai [Sat, 22 Feb 2020 04:31:11 +0000 (23:31 -0500)]
jbd2: fix data races at struct journal_head

journal_head::b_transaction and journal_head::b_next_transaction could
be accessed concurrently as noticed by KCSAN,

 LTP: starting fsync04
 /dev/zero: Can't open blockdev
 EXT4-fs (loop0): mounting ext3 file system using the ext4 subsystem
 EXT4-fs (loop0): mounted filesystem with ordered data mode. Opts: (null)
 ==================================================================
 BUG: KCSAN: data-race in __jbd2_journal_refile_buffer [jbd2] / jbd2_write_access_granted [jbd2]

 write to 0xffff99f9b1bd0e30 of 8 bytes by task 25721 on cpu 70:
  __jbd2_journal_refile_buffer+0xdd/0x210 [jbd2]
  __jbd2_journal_refile_buffer at fs/jbd2/transaction.c:2569
  jbd2_journal_commit_transaction+0x2d15/0x3f20 [jbd2]
  (inlined by) jbd2_journal_commit_transaction at fs/jbd2/commit.c:1034
  kjournald2+0x13b/0x450 [jbd2]
  kthread+0x1cd/0x1f0
  ret_from_fork+0x27/0x50

 read to 0xffff99f9b1bd0e30 of 8 bytes by task 25724 on cpu 68:
  jbd2_write_access_granted+0x1b2/0x250 [jbd2]
  jbd2_write_access_granted at fs/jbd2/transaction.c:1155
  jbd2_journal_get_write_access+0x2c/0x60 [jbd2]
  __ext4_journal_get_write_access+0x50/0x90 [ext4]
  ext4_mb_mark_diskspace_used+0x158/0x620 [ext4]
  ext4_mb_new_blocks+0x54f/0xca0 [ext4]
  ext4_ind_map_blocks+0xc79/0x1b40 [ext4]
  ext4_map_blocks+0x3b4/0x950 [ext4]
  _ext4_get_block+0xfc/0x270 [ext4]
  ext4_get_block+0x3b/0x50 [ext4]
  __block_write_begin_int+0x22e/0xae0
  __block_write_begin+0x39/0x50
  ext4_write_begin+0x388/0xb50 [ext4]
  generic_perform_write+0x15d/0x290
  ext4_buffered_write_iter+0x11f/0x210 [ext4]
  ext4_file_write_iter+0xce/0x9e0 [ext4]
  new_sync_write+0x29c/0x3b0
  __vfs_write+0x92/0xa0
  vfs_write+0x103/0x260
  ksys_write+0x9d/0x130
  __x64_sys_write+0x4c/0x60
  do_syscall_64+0x91/0xb05
  entry_SYSCALL_64_after_hwframe+0x49/0xbe

 5 locks held by fsync04/25724:
  #0: ffff99f9911093f8 (sb_writers#13){.+.+}, at: vfs_write+0x21c/0x260
  #1: ffff99f9db4c0348 (&sb->s_type->i_mutex_key#15){+.+.}, at: ext4_buffered_write_iter+0x65/0x210 [ext4]
  #2: ffff99f5e7dfcf58 (jbd2_handle){++++}, at: start_this_handle+0x1c1/0x9d0 [jbd2]
  #3: ffff99f9db4c0168 (&ei->i_data_sem){++++}, at: ext4_map_blocks+0x176/0x950 [ext4]
  #4: ffffffff99086b40 (rcu_read_lock){....}, at: jbd2_write_access_granted+0x4e/0x250 [jbd2]
 irq event stamp: 1407125
 hardirqs last  enabled at (1407125): [<ffffffff980da9b7>] __find_get_block+0x107/0x790
 hardirqs last disabled at (1407124): [<ffffffff980da8f9>] __find_get_block+0x49/0x790
 softirqs last  enabled at (1405528): [<ffffffff98a0034c>] __do_softirq+0x34c/0x57c
 softirqs last disabled at (1405521): [<ffffffff97cc67a2>] irq_exit+0xa2/0xc0

 Reported by Kernel Concurrency Sanitizer on:
 CPU: 68 PID: 25724 Comm: fsync04 Tainted: G L 5.6.0-rc2-next-20200221+ #7
 Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 07/10/2019

The plain reads are outside of jh->b_state_lock critical section which result
in data races. Fix them by adding pairs of READ|WRITE_ONCE().

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Qian Cai <cai@lca.pw>
Link: https://lore.kernel.org/r/20200222043111.2227-1-cai@lca.pw
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
4 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 29 Feb 2020 15:58:47 +0000 (09:58 -0600)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Four small fixes.

  Three are in drivers for fairly obvious bugs. The fourth is a set of
  regressions introduced by the compat_ioctl changes because some of the
  compat updates wrongly replaced .ioctl instead of .compat_ioctl"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: compat_ioctl: cdrom: Replace .ioctl with .compat_ioctl in four appropriate places
  scsi: zfcp: fix wrong data and display format of SFP+ temperature
  scsi: sd_sbc: Fix sd_zbc_report_zones()
  scsi: libfc: free response frame from GPN_ID

4 years agoMerge tag 'pci-v5.6-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
Linus Torvalds [Fri, 28 Feb 2020 19:51:53 +0000 (11:51 -0800)]
Merge tag 'pci-v5.6-fixes-2' of git://git./linux/kernel/git/helgaas/pci

Pull PCI fixes from Bjorn Helgaas:

 - Fix build issue on 32-bit ARM with old compilers (Marek Szyprowski)

 - Update MAINTAINERS for recent Cadence driver file move (Lukas
   Bulwahn)

* tag 'pci-v5.6-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  MAINTAINERS: Correct Cadence PCI driver path
  PCI: brcmstb: Fix build on 32bit ARM platforms with older compilers

4 years agoMerge tag 'block-5.6-2020-02-28' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 28 Feb 2020 19:43:30 +0000 (11:43 -0800)]
Merge tag 'block-5.6-2020-02-28' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:

 - Passthrough insertion fix (Ming)

 - Kill off some unused arguments (John)

 - blktrace RCU fix (Jan)

 - Dead fields removal for null_blk (Dongli)

 - NVMe polled IO fix (Bijan)

* tag 'block-5.6-2020-02-28' of git://git.kernel.dk/linux-block:
  nvme-pci: Hold cq_poll_lock while completing CQEs
  blk-mq: Remove some unused function arguments
  null_blk: remove unused fields in 'nullb_cmd'
  blktrace: Protect q->blk_trace with RCU
  blk-mq: insert passthrough request into hctx->dispatch directly

4 years agoMerge tag 'io_uring-5.6-2020-02-28' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 28 Feb 2020 19:39:14 +0000 (11:39 -0800)]
Merge tag 'io_uring-5.6-2020-02-28' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:

 - Fix for a race with IOPOLL used with SQPOLL (Xiaoguang)

 - Only show ->fdinfo if procfs is enabled (Tobias)

 - Fix for a chain with multiple personalities in the SQEs

 - Fix for a missing free of personality idr on exit

 - Removal of the spin-for-work optimization

 - Fix for next work lookup on request completion

 - Fix for non-vec read/write result progation in case of links

 - Fix for a fileset references on switch

 - Fix for a recvmsg/sendmsg 32-bit compatability mode

* tag 'io_uring-5.6-2020-02-28' of git://git.kernel.dk/linux-block:
  io_uring: fix 32-bit compatability with sendmsg/recvmsg
  io_uring: define and set show_fdinfo only if procfs is enabled
  io_uring: drop file set ref put/get on switch
  io_uring: import_single_range() returns 0/-ERROR
  io_uring: pick up link work on submit reference drop
  io-wq: ensure work->task_pid is cleared on init
  io-wq: remove spin-for-work optimization
  io_uring: fix poll_list race for SETUP_IOPOLL|SETUP_SQPOLL
  io_uring: fix personality idr leak
  io_uring: handle multiple personalities in link chains

4 years agoMerge branch 'nvme-5.6-rc4' of git://git.infradead.org/nvme into block-5.6
Jens Axboe [Fri, 28 Feb 2020 17:02:36 +0000 (10:02 -0700)]
Merge branch 'nvme-5.6-rc4' of git://git.infradead.org/nvme into block-5.6

Pull NVMe fix from Keith.

* 'nvme-5.6-rc4' of git://git.infradead.org/nvme:
  nvme-pci: Hold cq_poll_lock while completing CQEs

4 years agoMerge tag 'acpi-5.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Fri, 28 Feb 2020 17:02:18 +0000 (09:02 -0800)]
Merge tag 'acpi-5.6-rc4' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI fixes from Rafael Wysocki:
 "Fix a couple of configuration issues in the ACPI watchdog (WDAT)
  driver (Mika Westerberg) and make it possible to disable that driver
  at boot time in case it still does not work as expected (Jean
  Delvare)"

* tag 'acpi-5.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: watchdog: Set default timeout in probe
  ACPI: watchdog: Fix gas->access_width usage
  ACPICA: Introduce ACPI_ACCESS_BYTE_WIDTH() macro
  ACPI: watchdog: Allow disabling WDAT at boot

4 years agoMerge tag 'pm-5.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Fri, 28 Feb 2020 16:49:52 +0000 (08:49 -0800)]
Merge tag 'pm-5.6-rc4' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "Fix a recent cpufreq initialization regression (Rafael Wysocki),
  revert a devfreq commit that made incompatible changes and broke user
  land on some systems (Orson Zhai), drop a stale reference to a
  document that has gone away recently (Jonathan NeuschƤfer), and fix a
  typo in a hibernation code comment (Alexandre Belloni)"

* tag 'pm-5.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq: Fix policy initialization for internal governor drivers
  Revert "PM / devfreq: Modify the device name as devfreq(X) for sysfs"
  PM / hibernate: fix typo "reserverd_size" -> "reserved_size"
  Documentation: power: Drop reference to interface.rst

4 years agoMerge tag 'zonefs-5.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/dlemoal...
Linus Torvalds [Fri, 28 Feb 2020 16:34:47 +0000 (08:34 -0800)]
Merge tag 'zonefs-5.6-rc4' of git://git./linux/kernel/git/dlemoal/zonefs

Pull zonefs fixes from Damien Le Moal:
 "Two fixes in here:

   - Revert the initial decision to silently ignore IOCB_NOWAIT for
     asynchronous direct IOs to sequential zone files. Instead, return
     an error to the user to signal that the feature is not supported
     (from Christoph)

   - A fix to zonefs Kconfig to select FS_IOMAP to avoid build failures
     if no other file system already selected this option (from
     Johannes)"

* tag 'zonefs-5.6-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/dlemoal/zonefs:
  zonefs: select FS_IOMAP
  zonefs: fix IOCB_NOWAIT handling

4 years agoMerge tag 'kvmarm-fixes-5.6-1' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmar...
Paolo Bonzini [Fri, 28 Feb 2020 10:46:59 +0000 (11:46 +0100)]
Merge tag 'kvmarm-fixes-5.6-1' of git://git./linux/kernel/git/kvmarm/kvmarm into HEAD

KVM/arm fixes for 5.6, take #1

- Fix compilation on 32bit
- Move  VHE guest entry/exit into the VHE-specific entry code
- Make sure all functions called by the non-VHE HYP code is tagged as __always_inline

4 years agokvm: x86: Limit the number of "kvm: disabled by bios" messages
Erwan Velu [Thu, 27 Feb 2020 18:00:46 +0000 (19:00 +0100)]
kvm: x86: Limit the number of "kvm: disabled by bios" messages

In older version of systemd(219), at boot time, udevadm is called with :
/usr/bin/udevadm trigger --type=devices --action=add"

This program generates an echo "add" in /sys/devices/system/cpu/cpu<x>/uevent,
leading to the "kvm: disabled by bios" message in case of your Bios disabled
the virtualization extensions.

On a modern system running up to 256 CPU threads, this pollutes the Kernel logs.

This patch offers to ratelimit this message to avoid any userspace program triggering
this uevent printing this message too often.

This patch is only a workaround but greatly reduce the pollution without
breaking the current behavior of printing a message if some try to instantiate
KVM on a system that doesn't support it.

Note that recent versions of systemd (>239) do not have trigger this behavior.

This patch will be useful at least for some using older systemd with recent Kernels.

Signed-off-by: Erwan Velu <e.velu@criteo.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoMerge branches 'pm-sleep' and 'pm-devfreq'
Rafael J. Wysocki [Fri, 28 Feb 2020 10:00:50 +0000 (11:00 +0100)]
Merge branches 'pm-sleep' and 'pm-devfreq'

* pm-sleep:
  PM / hibernate: fix typo "reserverd_size" -> "reserved_size"
  Documentation: power: Drop reference to interface.rst

* pm-devfreq:
  Revert "PM / devfreq: Modify the device name as devfreq(X) for sysfs"

4 years agoKVM: x86: avoid useless copy of cpufreq policy
Paolo Bonzini [Fri, 28 Feb 2020 09:49:10 +0000 (10:49 +0100)]
KVM: x86: avoid useless copy of cpufreq policy

struct cpufreq_policy is quite big and it is not a good idea
to allocate one on the stack.  Just use cpufreq_cpu_get and
cpufreq_cpu_put which is even simpler.

Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: allow disabling -Werror
Paolo Bonzini [Fri, 28 Feb 2020 09:42:31 +0000 (10:42 +0100)]
KVM: allow disabling -Werror

Restrict -Werror to well-tested configurations and allow disabling it
via Kconfig.

Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: x86: allow compiling as non-module with W=1
Valdis Klētnieks [Fri, 28 Feb 2020 02:49:52 +0000 (21:49 -0500)]
KVM: x86: allow compiling as non-module with W=1

Compile error with CONFIG_KVM_INTEL=y and W=1:

  CC      arch/x86/kvm/vmx/vmx.o
arch/x86/kvm/vmx/vmx.c:68:32: error: 'vmx_cpu_id' defined but not used [-Werror=unused-const-variable=]
   68 | static const struct x86_cpu_id vmx_cpu_id[] = {
      |                                ^~~~~~~~~~
cc1: all warnings being treated as errors

When building with =y, the MODULE_DEVICE_TABLE macro doesn't generate a
reference to the structure (or any code at all).  This makes W=1 compiles
unhappy.

Wrap both in a #ifdef to avoid the issue.

Signed-off-by: Valdis Kletnieks <valdis.kletnieks@vt.edu>
[Do the same for CONFIG_KVM_AMD. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: Pre-allocate 1 cpumask variable per cpu for both pv tlb and pv ipis
Wanpeng Li [Tue, 18 Feb 2020 01:08:24 +0000 (09:08 +0800)]
KVM: Pre-allocate 1 cpumask variable per cpu for both pv tlb and pv ipis

Nick Desaulniers Reported:

  When building with:
  $ make CC=clang arch/x86/ CFLAGS=-Wframe-larger-than=1000
  The following warning is observed:
  arch/x86/kernel/kvm.c:494:13: warning: stack frame size of 1064 bytes in
  function 'kvm_send_ipi_mask_allbutself' [-Wframe-larger-than=]
  static void kvm_send_ipi_mask_allbutself(const struct cpumask *mask, int
  vector)
              ^
  Debugging with:
  https://github.com/ClangBuiltLinux/frame-larger-than
  via:
  $ python3 frame_larger_than.py arch/x86/kernel/kvm.o \
    kvm_send_ipi_mask_allbutself
  points to the stack allocated `struct cpumask newmask` in
  `kvm_send_ipi_mask_allbutself`. The size of a `struct cpumask` is
  potentially large, as it's CONFIG_NR_CPUS divided by BITS_PER_LONG for
  the target architecture. CONFIG_NR_CPUS for X86_64 can be as high as
  8192, making a single instance of a `struct cpumask` 1024 B.

This patch fixes it by pre-allocate 1 cpumask variable per cpu and use it for
both pv tlb and pv ipis..

Reported-by: Nick Desaulniers <ndesaulniers@google.com>
Acked-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: Introduce pv check helpers
Wanpeng Li [Tue, 18 Feb 2020 01:08:23 +0000 (09:08 +0800)]
KVM: Introduce pv check helpers

Introduce some pv check helpers for consistency.

Suggested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: let declaration of kvm_get_running_vcpus match implementation
Christian Borntraeger [Fri, 28 Feb 2020 08:49:41 +0000 (09:49 +0100)]
KVM: let declaration of kvm_get_running_vcpus match implementation

Sparse notices that declaration and implementation do not match:
arch/s390/kvm/../../../virt/kvm/kvm_main.c:4435:17: warning: incorrect type in return expression (different address spaces)
arch/s390/kvm/../../../virt/kvm/kvm_main.c:4435:17:    expected struct kvm_vcpu [noderef] <asn:3> **
arch/s390/kvm/../../../virt/kvm/kvm_main.c:4435:17:    got struct kvm_vcpu *[noderef] <asn:3> *

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoKVM: SVM: allocate AVIC data structures based on kvm_amd module parameter
Paolo Bonzini [Tue, 25 Feb 2020 07:54:26 +0000 (08:54 +0100)]
KVM: SVM: allocate AVIC data structures based on kvm_amd module parameter

Even if APICv is disabled at startup, the backing page and ir_list need
to be initialized in case they are needed later.  The only case in
which this can be skipped is for userspace irqchip, and that must be
done because avic_init_backing_page dereferences vcpu->arch.apic
(which is NULL for userspace irqchip).

Tested-by: rmuncrief@humanavance.com
Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=206579
Reviewed-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
4 years agoMerge tag 'drm-fixes-2020-02-28' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 28 Feb 2020 05:52:18 +0000 (21:52 -0800)]
Merge tag 'drm-fixes-2020-02-28' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Just some fixes for this week: amdgpu, radeon and i915.

  The main i915 one is a regression Gen7 (Ivybridge/Haswell), this moves
  them back from trying to use the full-ppgtt support to the aliasing
  version it used to use due to gpu hangs. Otherwise it's pretty quiet.

  amdgpu:
   - Drop DRIVER_USE_AGP
   - Fix memory leak in GPU reset
   - Resume fix for raven

  radeon:
   - Drop DRIVER_USE_AGP

  i915:
   - downgrade gen7 back to aliasing-ppgtt to avoid GPU hangs
   - shrinker fix
   - pmu leak and double free fixes
   - gvt user after free and virtual display reset fixes
   - randconfig build fix"

* tag 'drm-fixes-2020-02-28' of git://anongit.freedesktop.org/drm/drm:
  drm/radeon: Inline drm_get_pci_dev
  drm/amdgpu: Drop DRIVER_USE_AGP
  drm/i915: Avoid recursing onto active vma from the shrinker
  drm/i915/pmu: Avoid using globals for PMU events
  drm/i915/pmu: Avoid using globals for CPU hotplug state
  drm/i915/gtt: Downgrade gen7 (ivb, byt, hsw) back to aliasing-ppgtt
  drm/i915: fix header test with GCOV
  amdgpu/gmc_v9: save/restore sdpif regs during S3
  drm/amdgpu: fix memory leak during TDR test(v2)
  drm/i915/gvt: Fix orphan vgpu dmabuf_objs' lifetime
  drm/i915/gvt: Separate display reset from ALL_ENGINES reset

4 years agoMerge tag 'drm-intel-fixes-2020-02-27' of git://anongit.freedesktop.org/drm/drm-intel...
Dave Airlie [Fri, 28 Feb 2020 02:40:41 +0000 (12:40 +1000)]
Merge tag 'drm-intel-fixes-2020-02-27' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

drm/i915 fixes for v5.6-rc4:
- downgrade gen7 back to aliasing-ppgtt to avoid GPU hangs
- shrinker fix
- pmu leak and double free fixes
- gvt user after free and virtual display reset fixes
- randconfig build fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/874kvcsh00.fsf@intel.com
4 years agoMerge tag 'amd-drm-fixes-5.6-2020-02-26' of git://people.freedesktop.org/~agd5f/linux...
Dave Airlie [Fri, 28 Feb 2020 02:30:18 +0000 (12:30 +1000)]
Merge tag 'amd-drm-fixes-5.6-2020-02-26' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

amd-drm-fixes-5.6-2020-02-26:

amdgpu:
- Drop DRIVER_USE_AGP
- Fix memory leak in GPU reset
- Resume fix for raven

radeon:
- Drop DRIVER_USE_AGP

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200227034106.3912-1-alexander.deucher@amd.com
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Fri, 28 Feb 2020 00:34:41 +0000 (16:34 -0800)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:

 1) Fix leak in nl80211 AP start where we leak the ACL memory, from
    Johannes Berg.

 2) Fix double mutex unlock in mac80211, from Andrei Otcheretianski.

 3) Fix RCU stall in ipset, from Jozsef Kadlecsik.

 4) Fix devlink locking in devlink_dpipe_table_register, from Madhuparna
    Bhowmik.

 5) Fix race causing TX hang in ll_temac, from Esben Haabendal.

 6) Stale eth hdr pointer in br_dev_xmit(), from Nikolay Aleksandrov.

 7) Fix TX hash calculation bounds checking wrt. tc rules, from Amritha
    Nambiar.

 8) Size netlink responses properly in schedule action code to take into
    consideration TCA_ACT_FLAGS. From Jiri Pirko.

 9) Fix firmware paths for mscc PHY driver, from Antoine Tenart.

10) Don't register stmmac notifier multiple times, from Aaro Koskinen.

11) Various rmnet bug fixes, from Taehee Yoo.

12) Fix vsock deadlock in vsock transport release, from Stefano
    Garzarella.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (61 commits)
  net: dsa: mv88e6xxx: Fix masking of egress port
  mlxsw: pci: Wait longer before accessing the device after reset
  sfc: fix timestamp reconstruction at 16-bit rollover points
  vsock: fix potential deadlock in transport->release()
  unix: It's CONFIG_PROC_FS not CONFIG_PROCFS
  net: rmnet: fix packet forwarding in rmnet bridge mode
  net: rmnet: fix bridge mode bugs
  net: rmnet: use upper/lower device infrastructure
  net: rmnet: do not allow to change mux id if mux id is duplicated
  net: rmnet: remove rcu_read_lock in rmnet_force_unassociate_device()
  net: rmnet: fix suspicious RCU usage
  net: rmnet: fix NULL pointer dereference in rmnet_changelink()
  net: rmnet: fix NULL pointer dereference in rmnet_newlink()
  net: phy: marvell: don't interpret PHY status unless resolved
  mlx5: register lag notifier for init network namespace only
  unix: define and set show_fdinfo only if procfs is enabled
  hinic: fix a bug of rss configuration
  hinic: fix a bug of setting hw_ioctxt
  hinic: fix a irq affinity bug
  net/smc: check for valid ib_client_data
  ...

4 years agoMAINTAINERS: Correct Cadence PCI driver path
Lukas Bulwahn [Fri, 21 Feb 2020 18:54:02 +0000 (19:54 +0100)]
MAINTAINERS: Correct Cadence PCI driver path

de80f95ccb9c ("PCI: cadence: Move all files to per-device cadence
directory") moved files of the PCI cadence drivers, but did not update the
MAINTAINERS entry.

Since then, ./scripts/get_maintainer.pl --self-test complains:

  warning: no file matches F: drivers/pci/controller/pcie-cadence*

Repair the MAINTAINERS entry.

Link: https://lore.kernel.org/r/20200221185402.4703-1-lukas.bulwahn@gmail.com
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
4 years agoio_uring: fix 32-bit compatability with sendmsg/recvmsg
Jens Axboe [Thu, 27 Feb 2020 21:17:49 +0000 (14:17 -0700)]
io_uring: fix 32-bit compatability with sendmsg/recvmsg

We must set MSG_CMSG_COMPAT if we're in compatability mode, otherwise
the iovec import for these commands will not do the right thing and fail
the command with -EINVAL.

Found by running the test suite compiled as 32-bit.

Cc: stable@vger.kernel.org
Fixes: aa1fa28fc73e ("io_uring: add support for recvmsg()")
Fixes: 0fa03c624d8f ("io_uring: add support for sendmsg()")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 years agonet: dsa: mv88e6xxx: Fix masking of egress port
Andrew Lunn [Thu, 27 Feb 2020 20:20:49 +0000 (21:20 +0100)]
net: dsa: mv88e6xxx: Fix masking of egress port

Add missing ~ to the usage of the mask.

Reported-by: Kevin Benson <Kevin.Benson@zii.aero>
Reported-by: Chris Healy <Chris.Healy@zii.aero>
Fixes: 5c74c54ce6ff ("net: dsa: mv88e6xxx: Split monitor port configuration")
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: pci: Wait longer before accessing the device after reset
Amit Cohen [Thu, 27 Feb 2020 20:07:53 +0000 (21:07 +0100)]
mlxsw: pci: Wait longer before accessing the device after reset

During initialization the driver issues a reset to the device and waits
for 100ms before checking if the firmware is ready. The waiting is
necessary because before that the device is irresponsive and the first
read can result in a completion timeout.

While 100ms is sufficient for Spectrum-1 and Spectrum-2, it is
insufficient for Spectrum-3.

Fix this by increasing the timeout to 200ms.

Fixes: da382875c616 ("mlxsw: spectrum: Extend to support Spectrum-3 ASIC")
Signed-off-by: Amit Cohen <amitc@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosfc: fix timestamp reconstruction at 16-bit rollover points
Alex Maftei (amaftei) [Wed, 26 Feb 2020 17:33:19 +0000 (17:33 +0000)]
sfc: fix timestamp reconstruction at 16-bit rollover points

We can't just use the top bits of the last sync event as they could be
off-by-one every 65,536 seconds, giving an error in reconstruction of
65,536 seconds.

This patch uses the difference in the bottom 16 bits (mod 2^16) to
calculate an offset that needs to be applied to the last sync event to
get to the current time.

Signed-off-by: Alexandru-Mihai Maftei <amaftei@solarflare.com>
Acked-by: Martin Habets <mhabets@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovsock: fix potential deadlock in transport->release()
Stefano Garzarella [Wed, 26 Feb 2020 10:58:18 +0000 (11:58 +0100)]
vsock: fix potential deadlock in transport->release()

Some transports (hyperv, virtio) acquire the sock lock during the
.release() callback.

In the vsock_stream_connect() we call vsock_assign_transport(); if
the socket was previously assigned to another transport, the
vsk->transport->release() is called, but the sock lock is already
held in the vsock_stream_connect(), causing a deadlock reported by
syzbot:

    INFO: task syz-executor280:9768 blocked for more than 143 seconds.
      Not tainted 5.6.0-rc1-syzkaller #0
    "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
    syz-executor280 D27912  9768   9766 0x00000000
    Call Trace:
     context_switch kernel/sched/core.c:3386 [inline]
     __schedule+0x934/0x1f90 kernel/sched/core.c:4082
     schedule+0xdc/0x2b0 kernel/sched/core.c:4156
     __lock_sock+0x165/0x290 net/core/sock.c:2413
     lock_sock_nested+0xfe/0x120 net/core/sock.c:2938
     virtio_transport_release+0xc4/0xd60 net/vmw_vsock/virtio_transport_common.c:832
     vsock_assign_transport+0xf3/0x3b0 net/vmw_vsock/af_vsock.c:454
     vsock_stream_connect+0x2b3/0xc70 net/vmw_vsock/af_vsock.c:1288
     __sys_connect_file+0x161/0x1c0 net/socket.c:1857
     __sys_connect+0x174/0x1b0 net/socket.c:1874
     __do_sys_connect net/socket.c:1885 [inline]
     __se_sys_connect net/socket.c:1882 [inline]
     __x64_sys_connect+0x73/0xb0 net/socket.c:1882
     do_syscall_64+0xfa/0x790 arch/x86/entry/common.c:294
     entry_SYSCALL_64_after_hwframe+0x49/0xbe

To avoid this issue, this patch remove the lock acquiring in the
.release() callback of hyperv and virtio transports, and it holds
the lock when we call vsk->transport->release() in the vsock core.

Reported-by: syzbot+731710996d79d0d58fbc@syzkaller.appspotmail.com
Fixes: 408624af4c89 ("vsock: use local transport when it is loaded")
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agounix: It's CONFIG_PROC_FS not CONFIG_PROCFS
David S. Miller [Thu, 27 Feb 2020 19:52:35 +0000 (11:52 -0800)]
unix: It's CONFIG_PROC_FS not CONFIG_PROCFS

Fixes: 3a12500ed5dd ("unix: define and set show_fdinfo only if procfs is enabled")
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'net-rmnet-fix-several-bugs'
David S. Miller [Thu, 27 Feb 2020 19:45:07 +0000 (11:45 -0800)]
Merge branch 'net-rmnet-fix-several-bugs'

Taehee Yoo says:

====================
net: rmnet: fix several bugs

This patchset is to fix several bugs in RMNET module.

1. The first patch fixes NULL-ptr-deref in rmnet_newlink().
When rmnet interface is being created, it uses IFLA_LINK
without checking NULL.
So, if userspace doesn't set IFLA_LINK, panic will occur.
In this patch, checking NULL pointer code is added.

2. The second patch fixes NULL-ptr-deref in rmnet_changelink().
To get real device in rmnet_changelink(), it uses IFLA_LINK.
But, IFLA_LINK should not be used in rmnet_changelink().

3. The third patch fixes suspicious RCU usage in rmnet_get_port().
rmnet_get_port() uses rcu_dereference_rtnl().
But, rmnet_get_port() is used by datapath.
So, rcu_dereference_bh() should be used instead of rcu_dereference_rtnl().

4. The fourth patch fixes suspicious RCU usage in
rmnet_force_unassociate_device().
RCU critical section should not be scheduled.
But, unregister_netdevice_queue() in the rmnet_force_unassociate_device()
would be scheduled.
So, the RCU warning occurs.
In this patch, the rcu_read_lock() in the rmnet_force_unassociate_device()
is removed because it's unnecessary.

5. The fifth patch fixes duplicate MUX ID case.
RMNET MUX ID is unique.
So, rmnet interface isn't allowed to be created, which have
a duplicate MUX ID.
But, only rmnet_newlink() checks this condition, rmnet_changelink()
doesn't check this.
So, duplicate MUX ID case would happen.

6. The sixth patch fixes upper/lower interface relationship problems.
When IFLA_LINK is used, the upper/lower infrastructure should be used.
Because it checks the maximum depth of upper/lower interfaces and it also
checks circular interface relationship, etc.
In this patch, netdev_upper_dev_link() is used.

7. The seventh patch fixes bridge related problems.
a) ->ndo_del_slave() doesn't work.
b) It couldn't detect circular upper/lower interface relationship.
c) It couldn't prevent stack overflow because of too deep depth
of upper/lower interface
d) It doesn't check the number of lower interfaces.
e) Panics because of several reasons.
These problems are actually the same problem.
So, this patch fixes these problems.

8. The eighth patch fixes packet forwarding issue in bridge mode
Packet forwarding is not working in rmnet bridge mode.
Because when a packet is forwarded, skb_push() for an ethernet header
is needed. But it doesn't call skb_push().
So, the ethernet header will be lost.

Change log:
 - update commit logs.
 - drop two patches in this patchset because of wrong target branch.
   - ("net: rmnet: add missing module alias")
   - ("net: rmnet: print error message when command fails")
 - remove unneessary rcu_read_lock() in the third patch.
 - use rcu_dereference_bh() instead of rcu_dereference in third patch.
 - do not allow to add a bridge device if rmnet interface is already
   bridge mode in the seventh patch.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: fix packet forwarding in rmnet bridge mode
Taehee Yoo [Thu, 27 Feb 2020 12:26:15 +0000 (12:26 +0000)]
net: rmnet: fix packet forwarding in rmnet bridge mode

Packet forwarding is not working in rmnet bridge mode.
Because when a packet is forwarded, skb_push() for an ethernet header
is needed. But it doesn't call skb_push().
So, the ethernet header will be lost.

Test commands:
    modprobe rmnet
    ip netns add nst
    ip netns add nst2
    ip link add veth0 type veth peer name veth1
    ip link add veth2 type veth peer name veth3
    ip link set veth1 netns nst
    ip link set veth3 netns nst2

    ip link add rmnet0 link veth0 type rmnet mux_id 1
    ip link set veth2 master rmnet0
    ip link set veth0 up
    ip link set veth2 up
    ip link set rmnet0 up
    ip a a 192.168.100.1/24 dev rmnet0

    ip netns exec nst ip link set veth1 up
    ip netns exec nst ip a a 192.168.100.2/24 dev veth1
    ip netns exec nst2 ip link set veth3 up
    ip netns exec nst2 ip a a 192.168.100.3/24 dev veth3
    ip netns exec nst2 ping 192.168.100.2

Fixes: 60d58f971c10 ("net: qualcomm: rmnet: Implement bridge mode")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: fix bridge mode bugs
Taehee Yoo [Thu, 27 Feb 2020 12:26:02 +0000 (12:26 +0000)]
net: rmnet: fix bridge mode bugs

In order to attach a bridge interface to the rmnet interface,
"master" operation is used.
(e.g. ip link set dummy1 master rmnet0)
But, in the rmnet_add_bridge(), which is a callback of ->ndo_add_slave()
doesn't register lower interface.
So, ->ndo_del_slave() doesn't work.
There are other problems too.
1. It couldn't detect circular upper/lower interface relationship.
2. It couldn't prevent stack overflow because of too deep depth
of upper/lower interface
3. It doesn't check the number of lower interfaces.
4. Panics because of several reasons.

The root problem of these issues is actually the same.
So, in this patch, these all problems will be fixed.

Test commands:
    modprobe rmnet
    ip link add dummy0 type dummy
    ip link add rmnet0 link dummy0 type rmnet mux_id 1
    ip link add dummy1 master rmnet0 type dummy
    ip link add dummy2 master rmnet0 type dummy
    ip link del rmnet0
    ip link del dummy2
    ip link del dummy1

Splat looks like:
[   41.867595][ T1164] general protection fault, probably for non-canonical address 0xdffffc0000000101I
[   41.869993][ T1164] KASAN: null-ptr-deref in range [0x0000000000000808-0x000000000000080f]
[   41.872950][ T1164] CPU: 0 PID: 1164 Comm: ip Not tainted 5.6.0-rc1+ #447
[   41.873915][ T1164] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   41.875161][ T1164] RIP: 0010:rmnet_unregister_bridge.isra.6+0x71/0xf0 [rmnet]
[   41.876178][ T1164] Code: 48 89 ef 48 89 c6 5b 5d e9 fc fe ff ff e8 f7 f3 ff ff 48 8d b8 08 08 00 00 48 ba 00 7
[   41.878925][ T1164] RSP: 0018:ffff8880c4d0f188 EFLAGS: 00010202
[   41.879774][ T1164] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000101
[   41.887689][ T1164] RDX: dffffc0000000000 RSI: ffffffffb8cf64f0 RDI: 0000000000000808
[   41.888727][ T1164] RBP: ffff8880c40e4000 R08: ffffed101b3c0e3c R09: 0000000000000001
[   41.889749][ T1164] R10: 0000000000000001 R11: ffffed101b3c0e3b R12: 1ffff110189a1e3c
[   41.890783][ T1164] R13: ffff8880c4d0f200 R14: ffffffffb8d56160 R15: ffff8880ccc2c000
[   41.891794][ T1164] FS:  00007f4300edc0c0(0000) GS:ffff8880d9c00000(0000) knlGS:0000000000000000
[   41.892953][ T1164] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   41.893800][ T1164] CR2: 00007f43003bc8c0 CR3: 00000000ca53e001 CR4: 00000000000606f0
[   41.894824][ T1164] Call Trace:
[   41.895274][ T1164]  ? rcu_is_watching+0x2c/0x80
[   41.895895][ T1164]  rmnet_config_notify_cb+0x1f7/0x590 [rmnet]
[   41.896687][ T1164]  ? rmnet_unregister_bridge.isra.6+0xf0/0xf0 [rmnet]
[   41.897611][ T1164]  ? rmnet_unregister_bridge.isra.6+0xf0/0xf0 [rmnet]
[   41.898508][ T1164]  ? __module_text_address+0x13/0x140
[   41.899162][ T1164]  notifier_call_chain+0x90/0x160
[   41.899814][ T1164]  rollback_registered_many+0x660/0xcf0
[   41.900544][ T1164]  ? netif_set_real_num_tx_queues+0x780/0x780
[   41.901316][ T1164]  ? __lock_acquire+0xdfe/0x3de0
[   41.901958][ T1164]  ? memset+0x1f/0x40
[   41.902468][ T1164]  ? __nla_validate_parse+0x98/0x1ab0
[   41.903166][ T1164]  unregister_netdevice_many.part.133+0x13/0x1b0
[   41.903988][ T1164]  rtnl_delete_link+0xbc/0x100
[ ... ]

Fixes: 60d58f971c10 ("net: qualcomm: rmnet: Implement bridge mode")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: use upper/lower device infrastructure
Taehee Yoo [Thu, 27 Feb 2020 12:25:43 +0000 (12:25 +0000)]
net: rmnet: use upper/lower device infrastructure

netdev_upper_dev_link() is useful to manage lower/upper interfaces.
And this function internally validates looping, maximum depth.
All or most virtual interfaces that could have a real interface
(e.g. macsec, macvlan, ipvlan etc.) use lower/upper infrastructure.

Test commands:
    modprobe rmnet
    ip link add dummy0 type dummy
    ip link add rmnet1 link dummy0 type rmnet mux_id 1
    for i in {2..100}
    do
        let A=$i-1
        ip link add rmnet$i link rmnet$A type rmnet mux_id $i
    done
    ip link del dummy0

The purpose of the test commands is to make stack overflow.

Splat looks like:
[   52.411438][ T1395] BUG: KASAN: slab-out-of-bounds in find_busiest_group+0x27e/0x2c00
[   52.413218][ T1395] Write of size 64 at addr ffff8880c774bde0 by task ip/1395
[   52.414841][ T1395]
[   52.430720][ T1395] CPU: 1 PID: 1395 Comm: ip Not tainted 5.6.0-rc1+ #447
[   52.496511][ T1395] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   52.513597][ T1395] Call Trace:
[   52.546516][ T1395]
[   52.558773][ T1395] Allocated by task 3171537984:
[   52.588290][ T1395] BUG: unable to handle page fault for address: ffffffffb999e260
[   52.589311][ T1395] #PF: supervisor read access in kernel mode
[   52.590529][ T1395] #PF: error_code(0x0000) - not-present page
[   52.591374][ T1395] PGD d6818067 P4D d6818067 PUD d6819063 PMD 0
[   52.592288][ T1395] Thread overran stack, or stack corrupted
[   52.604980][ T1395] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI
[   52.605856][ T1395] CPU: 1 PID: 1395 Comm: ip Not tainted 5.6.0-rc1+ #447
[   52.611764][ T1395] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   52.621520][ T1395] RIP: 0010:stack_depot_fetch+0x10/0x30
[   52.622296][ T1395] Code: ff e9 f9 fe ff ff 48 89 df e8 9c 1d 91 ff e9 ca fe ff ff cc cc cc cc cc cc cc 89 f8 0
[   52.627887][ T1395] RSP: 0018:ffff8880c774bb60 EFLAGS: 00010006
[   52.628735][ T1395] RAX: 00000000001f8880 RBX: ffff8880c774d140 RCX: 0000000000000000
[   52.631773][ T1395] RDX: 000000000000001d RSI: ffff8880c774bb68 RDI: 0000000000003ff0
[   52.649584][ T1395] RBP: ffffea00031dd200 R08: ffffed101b43e403 R09: ffffed101b43e403
[   52.674857][ T1395] R10: 0000000000000001 R11: ffffed101b43e402 R12: ffff8880d900e5c0
[   52.678257][ T1395] R13: ffff8880c774c000 R14: 0000000000000000 R15: dffffc0000000000
[   52.694541][ T1395] FS:  00007fe867f6e0c0(0000) GS:ffff8880da000000(0000) knlGS:0000000000000000
[   52.764039][ T1395] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   52.815008][ T1395] CR2: ffffffffb999e260 CR3: 00000000c26aa005 CR4: 00000000000606e0
[   52.862312][ T1395] Call Trace:
[   52.887133][ T1395] Modules linked in: dummy rmnet veth openvswitch nsh nf_conncount nf_nat nf_conntrack nf_dex
[   52.936749][ T1395] CR2: ffffffffb999e260
[   52.965695][ T1395] ---[ end trace 7e32ca99482dbb31 ]---
[   52.966556][ T1395] RIP: 0010:stack_depot_fetch+0x10/0x30
[   52.971083][ T1395] Code: ff e9 f9 fe ff ff 48 89 df e8 9c 1d 91 ff e9 ca fe ff ff cc cc cc cc cc cc cc 89 f8 0
[   53.003650][ T1395] RSP: 0018:ffff8880c774bb60 EFLAGS: 00010006
[   53.043183][ T1395] RAX: 00000000001f8880 RBX: ffff8880c774d140 RCX: 0000000000000000
[   53.076480][ T1395] RDX: 000000000000001d RSI: ffff8880c774bb68 RDI: 0000000000003ff0
[   53.093858][ T1395] RBP: ffffea00031dd200 R08: ffffed101b43e403 R09: ffffed101b43e403
[   53.112795][ T1395] R10: 0000000000000001 R11: ffffed101b43e402 R12: ffff8880d900e5c0
[   53.139837][ T1395] R13: ffff8880c774c000 R14: 0000000000000000 R15: dffffc0000000000
[   53.141500][ T1395] FS:  00007fe867f6e0c0(0000) GS:ffff8880da000000(0000) knlGS:0000000000000000
[   53.143343][ T1395] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   53.152007][ T1395] CR2: ffffffffb999e260 CR3: 00000000c26aa005 CR4: 00000000000606e0
[   53.156459][ T1395] Kernel panic - not syncing: Fatal exception
[   54.213570][ T1395] Shutting down cpus with NMI
[   54.354112][ T1395] Kernel Offset: 0x33000000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0x)
[   54.355687][ T1395] Rebooting in 5 seconds..

Fixes: b37f78f234bf ("net: qualcomm: rmnet: Fix crash on real dev unregistration")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: do not allow to change mux id if mux id is duplicated
Taehee Yoo [Thu, 27 Feb 2020 12:25:19 +0000 (12:25 +0000)]
net: rmnet: do not allow to change mux id if mux id is duplicated

Basically, duplicate mux id isn't be allowed.
So, the creation of rmnet will be failed if there is duplicate mux id
is existing.
But, changelink routine doesn't check duplicate mux id.

Test commands:
    modprobe rmnet
    ip link add dummy0 type dummy
    ip link add rmnet0 link dummy0 type rmnet mux_id 1
    ip link add rmnet1 link dummy0 type rmnet mux_id 2
    ip link set rmnet1 type rmnet mux_id 1

Fixes: 23790ef12082 ("net: qualcomm: rmnet: Allow to configure flags for existing devices")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: remove rcu_read_lock in rmnet_force_unassociate_device()
Taehee Yoo [Thu, 27 Feb 2020 12:25:05 +0000 (12:25 +0000)]
net: rmnet: remove rcu_read_lock in rmnet_force_unassociate_device()

The notifier_call() of the slave interface removes rmnet interface with
unregister_netdevice_queue().
But, before calling unregister_netdevice_queue(), it acquires
rcu readlock.
In the RCU critical section, sleeping isn't be allowed.
But, unregister_netdevice_queue() internally calls synchronize_net(),
which would sleep.
So, suspicious RCU usage warning occurs.

Test commands:
    modprobe rmnet
    ip link add dummy0 type dummy
    ip link add dummy1 type dummy
    ip link add rmnet0 link dummy0 type rmnet mux_id 1
    ip link set dummy1 master rmnet0
    ip link del dummy0

Splat looks like:
[   79.639245][ T1195] =============================
[   79.640134][ T1195] WARNING: suspicious RCU usage
[   79.640852][ T1195] 5.6.0-rc1+ #447 Not tainted
[   79.641657][ T1195] -----------------------------
[   79.642472][ T1195] ./include/linux/rcupdate.h:273 Illegal context switch in RCU read-side critical section!
[   79.644043][ T1195]
[   79.644043][ T1195] other info that might help us debug this:
[   79.644043][ T1195]
[   79.645682][ T1195]
[   79.645682][ T1195] rcu_scheduler_active = 2, debug_locks = 1
[   79.646980][ T1195] 2 locks held by ip/1195:
[   79.647629][ T1195]  #0: ffffffffa3cf64f0 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x457/0x890
[   79.649312][ T1195]  #1: ffffffffa39256c0 (rcu_read_lock){....}, at: rmnet_config_notify_cb+0xf0/0x590 [rmnet]
[   79.651717][ T1195]
[   79.651717][ T1195] stack backtrace:
[   79.652650][ T1195] CPU: 3 PID: 1195 Comm: ip Not tainted 5.6.0-rc1+ #447
[   79.653702][ T1195] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   79.655037][ T1195] Call Trace:
[   79.655560][ T1195]  dump_stack+0x96/0xdb
[   79.656252][ T1195]  ___might_sleep+0x345/0x440
[   79.656994][ T1195]  synchronize_net+0x18/0x30
[   79.661132][ T1195]  netdev_rx_handler_unregister+0x40/0xb0
[   79.666266][ T1195]  rmnet_unregister_real_device+0x42/0xb0 [rmnet]
[   79.667211][ T1195]  rmnet_config_notify_cb+0x1f7/0x590 [rmnet]
[   79.668121][ T1195]  ? rmnet_unregister_bridge.isra.6+0xf0/0xf0 [rmnet]
[   79.669166][ T1195]  ? rmnet_unregister_bridge.isra.6+0xf0/0xf0 [rmnet]
[   79.670286][ T1195]  ? __module_text_address+0x13/0x140
[   79.671139][ T1195]  notifier_call_chain+0x90/0x160
[   79.671973][ T1195]  rollback_registered_many+0x660/0xcf0
[   79.672893][ T1195]  ? netif_set_real_num_tx_queues+0x780/0x780
[   79.675091][ T1195]  ? __lock_acquire+0xdfe/0x3de0
[   79.675825][ T1195]  ? memset+0x1f/0x40
[   79.676367][ T1195]  ? __nla_validate_parse+0x98/0x1ab0
[   79.677290][ T1195]  unregister_netdevice_many.part.133+0x13/0x1b0
[   79.678163][ T1195]  rtnl_delete_link+0xbc/0x100
[ ... ]

Fixes: ceed73a2cf4a ("drivers: net: ethernet: qualcomm: rmnet: Initial implementation")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: fix suspicious RCU usage
Taehee Yoo [Thu, 27 Feb 2020 12:24:45 +0000 (12:24 +0000)]
net: rmnet: fix suspicious RCU usage

rmnet_get_port() internally calls rcu_dereference_rtnl(),
which checks RTNL.
But rmnet_get_port() could be called by packet path.
The packet path is not protected by RTNL.
So, the suspicious RCU usage problem occurs.

Test commands:
    modprobe rmnet
    ip netns add nst
    ip link add veth0 type veth peer name veth1
    ip link set veth1 netns nst
    ip link add rmnet0 link veth0 type rmnet mux_id 1
    ip netns exec nst ip link add rmnet1 link veth1 type rmnet mux_id 1
    ip netns exec nst ip link set veth1 up
    ip netns exec nst ip link set rmnet1 up
    ip netns exec nst ip a a 192.168.100.2/24 dev rmnet1
    ip link set veth0 up
    ip link set rmnet0 up
    ip a a 192.168.100.1/24 dev rmnet0
    ping 192.168.100.2

Splat looks like:
[  146.630958][ T1174] WARNING: suspicious RCU usage
[  146.631735][ T1174] 5.6.0-rc1+ #447 Not tainted
[  146.632387][ T1174] -----------------------------
[  146.633151][ T1174] drivers/net/ethernet/qualcomm/rmnet/rmnet_config.c:386 suspicious rcu_dereference_check() !
[  146.634742][ T1174]
[  146.634742][ T1174] other info that might help us debug this:
[  146.634742][ T1174]
[  146.645992][ T1174]
[  146.645992][ T1174] rcu_scheduler_active = 2, debug_locks = 1
[  146.646937][ T1174] 5 locks held by ping/1174:
[  146.647609][ T1174]  #0: ffff8880c31dea70 (sk_lock-AF_INET){+.+.}, at: raw_sendmsg+0xab8/0x2980
[  146.662463][ T1174]  #1: ffffffff93925660 (rcu_read_lock_bh){....}, at: ip_finish_output2+0x243/0x2150
[  146.671696][ T1174]  #2: ffffffff93925660 (rcu_read_lock_bh){....}, at: __dev_queue_xmit+0x213/0x2940
[  146.673064][ T1174]  #3: ffff8880c19ecd58 (&dev->qdisc_running_key#7){+...}, at: ip_finish_output2+0x714/0x2150
[  146.690358][ T1174]  #4: ffff8880c5796898 (&dev->qdisc_xmit_lock_key#3){+.-.}, at: sch_direct_xmit+0x1e2/0x1020
[  146.699875][ T1174]
[  146.699875][ T1174] stack backtrace:
[  146.701091][ T1174] CPU: 0 PID: 1174 Comm: ping Not tainted 5.6.0-rc1+ #447
[  146.705215][ T1174] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[  146.706565][ T1174] Call Trace:
[  146.707102][ T1174]  dump_stack+0x96/0xdb
[  146.708007][ T1174]  rmnet_get_port.part.9+0x76/0x80 [rmnet]
[  146.709233][ T1174]  rmnet_egress_handler+0x107/0x420 [rmnet]
[  146.710492][ T1174]  ? sch_direct_xmit+0x1e2/0x1020
[  146.716193][ T1174]  rmnet_vnd_start_xmit+0x3d/0xa0 [rmnet]
[  146.717012][ T1174]  dev_hard_start_xmit+0x160/0x740
[  146.717854][ T1174]  sch_direct_xmit+0x265/0x1020
[  146.718577][ T1174]  ? register_lock_class+0x14d0/0x14d0
[  146.719429][ T1174]  ? dev_watchdog+0xac0/0xac0
[  146.723738][ T1174]  ? __dev_queue_xmit+0x15fd/0x2940
[  146.724469][ T1174]  ? lock_acquire+0x164/0x3b0
[  146.725172][ T1174]  __dev_queue_xmit+0x20c7/0x2940
[ ... ]

Fixes: ceed73a2cf4a ("drivers: net: ethernet: qualcomm: rmnet: Initial implementation")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: fix NULL pointer dereference in rmnet_changelink()
Taehee Yoo [Thu, 27 Feb 2020 12:24:26 +0000 (12:24 +0000)]
net: rmnet: fix NULL pointer dereference in rmnet_changelink()

In the rmnet_changelink(), it uses IFLA_LINK without checking
NULL pointer.
tb[IFLA_LINK] could be NULL pointer.
So, NULL-ptr-deref could occur.

rmnet already has a lower interface (real_dev).
So, after this patch, rmnet_changelink() does not use IFLA_LINK anymore.

Test commands:
    modprobe rmnet
    ip link add dummy0 type dummy
    ip link add rmnet0 link dummy0 type rmnet mux_id 1
    ip link set rmnet0 type rmnet mux_id 2

Splat looks like:
[   90.578726][ T1131] general protection fault, probably for non-canonical address 0xdffffc0000000000I
[   90.581121][ T1131] KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
[   90.582380][ T1131] CPU: 2 PID: 1131 Comm: ip Not tainted 5.6.0-rc1+ #447
[   90.584285][ T1131] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   90.587506][ T1131] RIP: 0010:rmnet_changelink+0x5a/0x8a0 [rmnet]
[   90.588546][ T1131] Code: 83 ec 20 48 c1 ea 03 80 3c 02 00 0f 85 6f 07 00 00 48 8b 5e 28 48 b8 00 00 00 00 00 0
[   90.591447][ T1131] RSP: 0018:ffff8880ce78f1b8 EFLAGS: 00010247
[   90.592329][ T1131] RAX: dffffc0000000000 RBX: 0000000000000000 RCX: ffff8880ce78f8b0
[   90.593253][ T1131] RDX: 0000000000000000 RSI: ffff8880ce78f4a0 RDI: 0000000000000004
[   90.594058][ T1131] RBP: ffff8880cf543e00 R08: 0000000000000002 R09: 0000000000000002
[   90.594859][ T1131] R10: ffffffffc0586a40 R11: 0000000000000000 R12: ffff8880ca47c000
[   90.595690][ T1131] R13: ffff8880ca47c000 R14: ffff8880cf545000 R15: 0000000000000000
[   90.596553][ T1131] FS:  00007f21f6c7e0c0(0000) GS:ffff8880da400000(0000) knlGS:0000000000000000
[   90.597504][ T1131] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   90.599418][ T1131] CR2: 0000556e413db458 CR3: 00000000c917a002 CR4: 00000000000606e0
[   90.600289][ T1131] Call Trace:
[   90.600631][ T1131]  __rtnl_newlink+0x922/0x1270
[   90.601194][ T1131]  ? lock_downgrade+0x6e0/0x6e0
[   90.601724][ T1131]  ? rtnl_link_unregister+0x220/0x220
[   90.602309][ T1131]  ? lock_acquire+0x164/0x3b0
[   90.602784][ T1131]  ? is_bpf_image_address+0xff/0x1d0
[   90.603331][ T1131]  ? rtnl_newlink+0x4c/0x90
[   90.603810][ T1131]  ? kernel_text_address+0x111/0x140
[   90.604419][ T1131]  ? __kernel_text_address+0xe/0x30
[   90.604981][ T1131]  ? unwind_get_return_address+0x5f/0xa0
[   90.605616][ T1131]  ? create_prof_cpu_mask+0x20/0x20
[   90.606304][ T1131]  ? arch_stack_walk+0x83/0xb0
[   90.606985][ T1131]  ? stack_trace_save+0x82/0xb0
[   90.607656][ T1131]  ? stack_trace_consume_entry+0x160/0x160
[   90.608503][ T1131]  ? deactivate_slab.isra.78+0x2c5/0x800
[   90.609336][ T1131]  ? kasan_unpoison_shadow+0x30/0x40
[   90.610096][ T1131]  ? kmem_cache_alloc_trace+0x135/0x350
[   90.610889][ T1131]  ? rtnl_newlink+0x4c/0x90
[   90.611512][ T1131]  rtnl_newlink+0x65/0x90
[ ... ]

Fixes: 23790ef12082 ("net: qualcomm: rmnet: Allow to configure flags for existing devices")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: rmnet: fix NULL pointer dereference in rmnet_newlink()
Taehee Yoo [Thu, 27 Feb 2020 12:23:52 +0000 (12:23 +0000)]
net: rmnet: fix NULL pointer dereference in rmnet_newlink()

rmnet registers IFLA_LINK interface as a lower interface.
But, IFLA_LINK could be NULL.
In the current code, rmnet doesn't check IFLA_LINK.
So, panic would occur.

Test commands:
    modprobe rmnet
    ip link add rmnet0 type rmnet mux_id 1

Splat looks like:
[   36.826109][ T1115] general protection fault, probably for non-canonical address 0xdffffc0000000000I
[   36.838817][ T1115] KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
[   36.839908][ T1115] CPU: 1 PID: 1115 Comm: ip Not tainted 5.6.0-rc1+ #447
[   36.840569][ T1115] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[   36.841408][ T1115] RIP: 0010:rmnet_newlink+0x54/0x510 [rmnet]
[   36.841986][ T1115] Code: 83 ec 18 48 c1 e9 03 80 3c 01 00 0f 85 d4 03 00 00 48 8b 6a 28 48 b8 00 00 00 00 00 c
[   36.843923][ T1115] RSP: 0018:ffff8880b7e0f1c0 EFLAGS: 00010247
[   36.844756][ T1115] RAX: dffffc0000000000 RBX: ffff8880d14cca00 RCX: 1ffff11016fc1e99
[   36.845859][ T1115] RDX: 0000000000000000 RSI: ffff8880c3d04000 RDI: 0000000000000004
[   36.846961][ T1115] RBP: 0000000000000000 R08: ffff8880b7e0f8b0 R09: ffff8880b6ac2d90
[   36.848020][ T1115] R10: ffffffffc0589a40 R11: ffffed1016d585b7 R12: ffffffff88ceaf80
[   36.848788][ T1115] R13: ffff8880c3d04000 R14: ffff8880b7e0f8b0 R15: ffff8880c3d04000
[   36.849546][ T1115] FS:  00007f50ab3360c0(0000) GS:ffff8880da000000(0000) knlGS:0000000000000000
[   36.851784][ T1115] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   36.852422][ T1115] CR2: 000055871afe5ab0 CR3: 00000000ae246001 CR4: 00000000000606e0
[   36.853181][ T1115] Call Trace:
[   36.853514][ T1115]  __rtnl_newlink+0xbdb/0x1270
[   36.853967][ T1115]  ? lock_downgrade+0x6e0/0x6e0
[   36.854420][ T1115]  ? rtnl_link_unregister+0x220/0x220
[   36.854936][ T1115]  ? lock_acquire+0x164/0x3b0
[   36.855376][ T1115]  ? is_bpf_image_address+0xff/0x1d0
[   36.855884][ T1115]  ? rtnl_newlink+0x4c/0x90
[   36.856304][ T1115]  ? kernel_text_address+0x111/0x140
[   36.856857][ T1115]  ? __kernel_text_address+0xe/0x30
[   36.857440][ T1115]  ? unwind_get_return_address+0x5f/0xa0
[   36.858063][ T1115]  ? create_prof_cpu_mask+0x20/0x20
[   36.858644][ T1115]  ? arch_stack_walk+0x83/0xb0
[   36.859171][ T1115]  ? stack_trace_save+0x82/0xb0
[   36.859710][ T1115]  ? stack_trace_consume_entry+0x160/0x160
[   36.860357][ T1115]  ? deactivate_slab.isra.78+0x2c5/0x800
[   36.860928][ T1115]  ? kasan_unpoison_shadow+0x30/0x40
[   36.861520][ T1115]  ? kmem_cache_alloc_trace+0x135/0x350
[   36.862125][ T1115]  ? rtnl_newlink+0x4c/0x90
[   36.864073][ T1115]  rtnl_newlink+0x65/0x90
[ ... ]

Fixes: ceed73a2cf4a ("drivers: net: ethernet: qualcomm: rmnet: Initial implementation")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge tag 'kbuild-fixes-v5.6-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 27 Feb 2020 19:26:33 +0000 (11:26 -0800)]
Merge tag 'kbuild-fixes-v5.6-2' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - fix missed rebuild of DT schema check

 - add some phony targets to PHONY

 - fix comments and documents

* tag 'kbuild-fixes-v5.6-2' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  kbuild: get rid of trailing slash from subdir- example
  kbuild: add dt_binding_check to PHONY in a correct place
  kbuild: add dtbs_check to PHONY
  kbuild: remove unneeded semicolon at the end of cmd_dtb_check
  kbuild: fix DT binding schema rule to detect command line changes
  kbuild: remove wrong documentation about mandatory-y
  kbuild: add comment for V=2 mode

4 years agonet: phy: marvell: don't interpret PHY status unless resolved
Russell King [Thu, 27 Feb 2020 09:44:49 +0000 (09:44 +0000)]
net: phy: marvell: don't interpret PHY status unless resolved

Don't attempt to interpret the PHY specific status register unless
the PHY is indicating that the resolution is valid.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlx5: register lag notifier for init network namespace only
Jiri Pirko [Thu, 27 Feb 2020 07:22:10 +0000 (08:22 +0100)]
mlx5: register lag notifier for init network namespace only

The current code causes problems when the unregistering netdevice could
be different then the registering one.

Since the check in mlx5_lag_netdev_event() does not allow any other
network namespace anyway, fix this by registerting the lag notifier
per init network namespace only.

Fixes: d48834f9d4b4 ("mlx5: Use dev_net netdevice notifier registrations")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Tested-by: Aya Levin <ayal@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid
Linus Torvalds [Thu, 27 Feb 2020 19:13:27 +0000 (11:13 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/hid/hid

Pull HID subsystem fixes from Jiri Kosina:

 - syzkaller-reported error handling fixes in various drivers, from
   various people

 - increase of HID report buffer size to 8K, which is apparently needed
   by certain modern devices

 - a few new device-ID-specific fixes / quirks

 - battery charging status reporting fix in logitech-hidpp, from Filipe
   LaĆ­ns

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid:
  HID: hid-bigbenff: fix race condition for scheduled work during removal
  HID: hid-bigbenff: call hid_hw_stop() in case of error
  HID: hid-bigbenff: fix general protection fault caused by double kfree
  HID: i2c-hid: add Trekstor Surfbook E11B to descriptor override
  HID: alps: Fix an error handling path in 'alps_input_configured()'
  HID: hiddev: Fix race in in hiddev_disconnect()
  HID: core: increase HID report buffer size to 8KiB
  HID: core: fix off-by-one memset in hid_report_raw_event()
  HID: apple: Add support for recent firmware on Magic Keyboards
  HID: ite: Only bind to keyboard USB interface on Acer SW5-012 keyboard dock
  HID: logitech-hidpp: BatteryVoltage: only read chargeStatus if extPower is active

4 years agounix: define and set show_fdinfo only if procfs is enabled
Tobias Klauser [Wed, 26 Feb 2020 17:29:53 +0000 (18:29 +0100)]
unix: define and set show_fdinfo only if procfs is enabled

Follow the pattern used with other *_show_fdinfo functions and only
define unix_show_fdinfo and set it in proto_ops if CONFIG_PROCFS
is set.

Fixes: 3c32da19a858 ("unix: Show number of pending scm files of receive queue in fdinfo")
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Reviewed-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'hinic-BugFixes'
David S. Miller [Thu, 27 Feb 2020 19:08:01 +0000 (11:08 -0800)]
Merge branch 'hinic-BugFixes'

Luo bin says:

====================
hinic: BugFixes

the bug fixed in patch #2 has been present since the first commit.
the bugs fixed in patch #1 and patch #3 have been present since the
following commits:
patch #1: 352f58b0d9f2 ("net-next/hinic: Set Rxq irq to specific cpu for NUMA")
patch #3: 421e9526288b ("hinic: add rss support")
====================

Signed-off-by: David S. Miller <davem@davemloft.net>