Add HWPMC events to measure latency.
Provide sysctl to choose the number of outstanding events which
trigger HWPMC event.
Obtained from: Semihalf
Sponsored by: Stormshield
Differential revision: https://reviews.freebsd.org/D31283
Some system software expects to be able to read at least the number of
bytes returned by FIONREAD. When control messages are counted in this
return value, this assumption is violated. Follow Linux and OpenBSD
here (as well as our own kevent(EVFILT_READ)) and only return the number
of data bytes available.
Reported by: avg
MFC after: 2 weeks
As of commit 103b207536f9, the NFSv4.2 server will limit the size
of a Copy operation based upon a 1 second timeout. The Linux 5.2
kernel server also limits Copy operation size to 4Mbytes.
As such, the NFSv4.2 client can attempt a large Copy without
resulting in a long RPC RTT for these servers.
This patch changes vfs.nfs.maxcopyrange to 64bits and sets
the default to the maximum possible size of SSIZE_MAX, since
a larger size makes the Copy operation more efficient and
allows for copying to complete with fewer RPCs.
The sysctl may be need to be made smaller for other non-FreeBSD
NFSv4.2 servers.
MFC after: 2 weeks
With lio_listio(2), the opcode is specified by userspace rather than
being hard-coded by the system call (e.g., aio_readv() -> LIO_READV).
kern_lio_listio() calls aio_aqueue() with an opcode of LIO_NOP, which
gets fixed up when the aiocb is copied in.
When copying in a job request for vectored I/O, we need to dynamically
allocate a uio to wrap an iovec. So aiocb_copyin() needs to get the
opcode from the aiocb and then decide whether an allocation is required.
We failed to do this in the COMPAT_FREEBSD32 case. Fix it.
Reported by: syzbot+27eab6f2c2162f2885ee@syzkaller.appspotmail.com
Reviewed by: kib, asomers
Fixes: f30a1ae8d529 ("lio_listio(2): Allow LIO_READV and LIO_WRITEV.")
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31914
All callers of sctp_aloc_assoc() mark the PCB as connected after a
successful call (for one-to-one-style sockets). In all cases this is
done without the PCB lock, so the PCB's flags can be corrupted. We also
do not atomically check whether a one-to-one-style socket is a listening
socket, which violates various assumptions in solisten_proto().
We need to hold the PCB lock across all of sctp_aloc_assoc() to fix
this. In order to do that without introducing lock order reversals, we
have to hold the global info lock as well.
So:
- Convert sctp_aloc_assoc() so that the inp and info locks are
consistently held. It returns with the association lock held, as
before.
- Fix an apparent bug where we failed to remove an association from a
global hash if sctp_add_remote_addr() fails.
- sctp_select_a_tag() is called when initializing an association, and it
acquires the global info lock. To avoid lock recursion, push locking
into its callers.
- Introduce sctp_aloc_assoc_connected(), which atomically checks for a
listening socket and sets SCTP_PCB_FLAGS_CONNECTED.
There is still one edge case in sctp_process_cookie_new() where we do
not update PCB/socket state correctly.
Reviewed by: tuexen
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31908
This adds an option to detect if hole-punching is implemented by the
underlying file system. If this flag is set, and if the underlying file
system does not support hole-punching, md(4) fails BIO_DELETE requests
with EOPNOTSUPP.
Sponsored by: The FreeBSD Foundation
Reviewed by: markj
Differential Revision: https://reviews.freebsd.org/D31883
When a PDU with an error (bad padding, header digest, or data digest)
is received, log the error via ICL_WARN() and then reset the
connection via the ic_error callback.
While here, add per-rxq counters for errors.
Sponsored by: Chelsio Communications
soo_aio_queue() did not handle the possibility that the provided socket
is a listening socket. Up until recently, to fix this one would have to
acquire the socket lock first and check, since the socket buffer locks
were destroyed by listen(2).
Now that the socket buffer locks belong to the socket, simply check
SOLISTENING(so) after acquiring them, and make listen(2) return an error
if any AIO jobs are enqueued on the socket.
Add a couple of simple regression test cases.
Note that this fixes things only for the default AIO implementation;
cxgbe(4)'s TCP offload has a separate pru_aio_queue implementation which
requires its own solution.
Reported by: syzbot+c8aa122fa2c6a4e2a28b@syzkaller.appspotmail.com
Reported by: syzbot+39af117d43d4f0faf512@syzkaller.appspotmail.com
Reported by: syzbot+60cceb9569145a0b993b@syzkaller.appspotmail.com
Reported by: syzbot+2d522c5db87710277ca5@syzkaller.appspotmail.com
Reviewed by: tuexen, gallatin, jhb
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31901
Since commit c67f3b8b78e50c6df7c057d6cf108e4d6b4312d0 the sockbuf
mutexes belong to the containing socket. Sockbufs contain a pointer to
a mutex, which by default is initialized to the corresponding mutexes in
the socket. The SOCKBUF_LOCK() etc. macros operate on this pointer.
However, the pointer is clobbered by listen(2) so it's not safe to use
them unless one is sure that the socket is not a listening socket.
This change introduces a new set of macros which lock socket buffers
through the socket. This is a bit cheaper since it removes the pointer
indirection, and allows one to safely lock socket buffers and then check
for a listening socket.
For MFC, these macros should be reimplemented in terms of the existing
socket buffer layout.
Reviewed by: tuexen, gallatin, jhb
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31900
Apparently e62e4b85942 wasn't enough to close the race between
a queue being flushed by a packet and callout executing, because
the callouts used without a lock aren't 100% bulletproof. To close
the race use callout_init_mtx() for L2TP timers, and make sure that
all calls to ng_callout()/ng_uncallout() are done under the seq lock.
If used properly, a locked callout can be used transparently with
old netgraph KPI of ng_callout/ng_uncallout which predates locked
callouts.
While here, utilize ng_uncallout_drain() instead of ng_uncallout()
on the node shutdown.
PR: 241133
Reviewed by: mjg, markj
Differential Revision: https://reviews.freebsd.org/D31476
Cover few cases of access to seq without lock missed in 702f98951d5.
There are no known bugs fixed with this change, however. With INVARIANTS
embed ng_l2tp_seq_check() into lock/unlock macros. Slightly reduce number
of locks/unlocks per packet keeping the lock between functions.
Reviewed by: mjg, markj
Differential Revision: https://reviews.freebsd.org/D31476
Move shared code into ng_uncallout_internal(). While here add a comment
mentioning a problem with scheduled+executing callout.
Reviewed by: mjg, markj
Differential Revision: https://reviews.freebsd.org/D31476
ng_uncallout. Most of them do not check it anyway, so very little node
changes are required.
Reviewed by: mjg, markj
Differential Revision: https://reviews.freebsd.org/D31476
tag2name() returns a uint16_t, so we don't need to use uint32_t for the
qid (or pqid). This reduces the size of struct pf_kstate slightly. That
in turn buys us space to add extra fields for dummynet later.
Happily these fields are not exposed to user space (there are user space
versions of them, but they can just stay uint32_t), so there's no ABI
breakage in modifying this.
MFC after: 1 week
Sponsored by: Rubicon Communications, LLC ("Netgate")
Differential Revision: https://reviews.freebsd.org/D31873
Free memory before return from arprequest_internal(). In in_arpinput(),
if arp_fillheader() fails, it should use goto drop.
Reviewed by: melifaro, imp, markj
MFC after: 1 week
Pull Request: https://github.com/freebsd/freebsd-src/pull/534
When we're synproxy-ing a connection that's going to us (as opposed to a
forwarded one) we wound up trying to send out the pf-generated tcp
packets through pf_intr(), which called ip(6)_output(). That doesn't
work all that well for packets that are destined for us, so in that case
we must call ip(6)_input() instead.
MFC after: 1 week
Sponsored by: Modirum MDPay
Differential Revision: https://reviews.freebsd.org/D31853
key_allocsa() expects to handle only IPSec protocols and has an
assertion to this effect. However, ipsec4_ctlinput() has to handle
messages from ICMP unreachable packets and was not validating the
protocol number. In practice such a packet would simply fail to match
any SADB entries and would thus be ignored.
Reported by: syzbot+6a9ef6fcfadb9f3877fe@syzkaller.appspotmail.com
Reviewed by: ae
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31890
When traversing a list of interface addresses, we need to be in a net
epoch section, and protocol ctlinput routines need a stable reference to
the address.
Reported by: syzbot+3219af764ead146a3a4e@syzkaller.appspotmail.com
Reviewed by: kp, melifaro
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31889
Currently hkbd counts all key states to be "Up" at the start of
interrupt callback. That results in generation of "Key Up" event for
each key that has been downed before but is not listed in current
report while is still downed.
Fix that with clearing of temporary key data storage bits only for
keys contained in processed report.
Reported by: Greg V
Obtained from: sysutils/iichid
MFC after: 2 weeks
r123442 introduced solution for clamping of PS/2 mice jitter when using
a KVM. Solution is to buffer mouse packets for 0.050ms if mouse activity
has not been seen for more than 0.5 seconds. Then flush that data to driver
if no validation errors found or drop the entire queue otherwise.
While it works well with relative devices it has issues with absolute ones
Depending on history buffering may results in delaying of the touch front
edge for 0.050ms that affects gesture processing (tap detection).
As absolute touchpads usually are built-in devices we can safely disable
bufferization and KVM jitter clamping to avoid such a delays.
MFC after: 2 weeks
- Some configurations, e.g. HP EliteBook 840 G3, come with a dummy card
in the card slot which is detected as a valid SD card. This added long
timeout at boot time. To alleviate the problem, the default timeout is
reduced to one second during the setup phase. [1]
- Some configurations crash at boot if rtsx(4) is defined in the kernel
config. At boot time, without a card inserted, the driver found that
a card is present and just after that a "spontaneous" interrupt is
generated showing that no card is present. To solve this problem,
DELAY(9) is set to one quarter of a second before checking card presence
during driver attach.
- As advised by adrian, taskqueue and DMA are set up sooner during
the driver attach. A heuristic to try to detect configuration needing
inversion was added.
PR: 255130 [1]
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D30499
osd_register(9) may reallocate and expand the destructor array for a
given object type if no space is available for a new key. This happens
with the object lock held. Thus, when verifying that a given slot in
the array is occupied, we need to hold the object lock to avoid racing
with a reallocation.
Reported by: syzbot+69ce54c7d7d813315dd3@syzkaller.appspotmail.com
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Although it is not specified in the RFCs, the concept that
the NFSv4 server should reply to an RPC request within a
reasonable time is accepted practice within the NFSv4 community.
Without this patch, the NFSv4.2 server attempts to reply to
a Copy operation within 1 second by limiting the copy to
vfs.nfs.maxcopyrange bytes (default 10Mbytes). This is crude at
best, given the large variation in I/O subsystem performance.
This patch uses the COPY_FILE_RANGE_TIMEO1SEC flag added by
commit c5128c48df3c to limit the reply time for a Copy
operation to approximately 1 second.
MFC after: 2 weeks
When port reuse is enabled in a one-to-one-style socket, sctp_listen()
may call sctp_swap_inpcb_for_listen() to move the PCB out of the "TCP
pool". In so doing it will drop the PCB lock, yielding an LOR since we
now hold several socket locks. Reorder sctp_listen() so that it
performs this operation before beginning the conversion to a listening
socket. Also modify sctp_swap_inpcb_for_listen() to return with PCB
write-locked, since that's what sctp_listen() expects now.
Reviewed by: tuexen
Fixes: bd4a39cc93d9 ("socket: Properly interlock when transitioning to a listening socket")
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31879
After commit bd4a39cc93d9 we now hold the global inp info lock across
the call to sctp_swap_inpcb_for_listen(), which attempts to acquire it
again. Since sctp_swap_inpcb_for_listen()'s sole caller is
sctp_listen(), we can simply change it to not try to acquire the lock.
Reported by: syzbot+a76b19ea2f8e1190c451@syzkaller.appspotmail.com
Reported by: syzbot+a1b6cef257ad145b7187@syzkaller.appspotmail.com
Reviewed by: tuexen
Fixes: bd4a39cc93d9 ("socket: Properly interlock when transitioning to a listening socket")
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D31878
When running in a virtualized environment, TLB invalidations can only
be performed on process scope, as only the hypervisor is allowed to
invalidate a global scope, or else a Program Interrupt is triggered.
Since we are here, also make sure that the register process table
hypercall returns success.
Reviewed by: jhibbits
MFC after: 2 weeks
Sponsored by: Instituto de Pesquisas Eldorado (eldorado.org.br)
Differential Revision: https://reviews.freebsd.org/D31775
This disables testing the ACPI timer by default, forcing the use of
ACPI-fast rather than ACPI-safe. The broken-ACPI-timers workaround
can be re-enabled by setting the hw.acpi.timer_test_enabled=1 tunable.
This speeds up the FreeBSD boot process by 140 ms on an EC2 c5.xlarge
instance.
This change will not be MFCed.
Assuming no problems are reported, acpi_timer_test, the associated
tunable, and the ACPI-safe timecounter should be removed in FreeBSD 15.
Relnotes: The ACPI-safe timer is disabled in favour of ACPI-fast;
if timekeeping issues are observed, please test with
hw.acpi.timer_test_enabled=1 in loader.conf and report
if that fixes the problem.
When hw.acpi.timer_test_enabled is set to 0, this makes acpi_timer_test
return 1 without actually testing the ACPI timer; this results in the
ACPI-fast timecounter always being used rather than potentially using
ACPI-safe.
The ACPI timer testing was introduced in 2002 as a workaround for
errata in Pentium II and Pentium III chipsets, and is unlikely to be
needed in 2021.
While I'm here, add TSENTER/TSEXIT to make it easier to see the time
spent on the test (if it is enabled).
Reviewed by: allanjude, imp
MFC After: 1 week
Although it is not specified in the RFCs, the concept that
the NFSv4 server should reply to an RPC request within a
reasonable time is accepted practice within the NFSv4 community.
Without this patch, the NFSv4.2 server attempts to reply to
a Copy operation within 1second by limiting the copy to
vfs.nfs.maxcopyrange bytes (default 10Mbytes). This is crude at
best, given the large variation in I/O subsystem performance.
This patch adds a kernel only flag COPY_FILE_RANGE_TIMEO1SEC
that the NFSv4.2 can specify, which tells VOP_COPY_FILE_RANGE()
to return after approximately 1 second with a partial result and
implements this in vn_generic_copy_file_range(), used by
vop_stdcopyfilerange().
Modifying the NFSv4.2 server to set this flag will be done in
a separate patch. Also under consideration is exposing the
COPY_FILE_RANGE_TIMEO1SEC to userland for use on the FreeBSD
copy_file_range(2) syscall.
MFC after: 2 weeks
Reviewed by: khng
Differential Revision: https://reviews.freebsd.org/D31829
