o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
A long long time ago the register keyword told the compiler to store
the corresponding variable in a CPU register, but it is not relevant
for any compiler used in the FreeBSD world today.
ANSIfy related prototypes while here.
Reviewed by: cem, jhb
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D10193
When sosend() replies ERESTART in the client side krpc, it indicates that
the RPC message hasn't yet been sent and that the send queue is full or
locked while a signal is posted for the process.
Without this patch, this would result in a RPC_CANTSEND reply from
clnt_vc_call(), which would cause clnt_reconnect_call() to create a new
TCP transport connection. For most NFS servers, this wasn't a serious problem,
although it did imply retries of outstanding RPCs, which could possibly
have missed the DRC.
For an NFSv4.1 mount to AmazonEFS, this caused a serious problem, since
AmazonEFS often didn't retain the NFSv4.1 session and would reply with
NFS4ERR_BAD_SESSION. This implies to the client a crash/reboot which
requires open/lock state recovery.
Three options were considered to fix this:
- Return the ERESTART all the way up to the system call boundary and then
have the system call redone. This is fraught with risk, due to convoluted
code paths, asynchronous I/O RPCs etc. cperciva@ worked on this, but it
is still a work in prgress and may not be feasible.
- Set SB_NOINTR for the socket buffer. This fixes the problem, but makes
the sosend() completely non interruptible, which kib@ considered
inappropriate. It also would break forced dismount when a thread
was blocked in sosend().
- Modify the retry loop in clnt_vc_call(), so that it loops for this case
for up to 15sec. Testing showed that the sosend() usually succeeded by
the 2nd retry. The extreme case observed was 111 loop iterations, or
about 100msec of delay.
This third alternative is what is implemented in this patch, since the
change is:
- localized
- straightforward
- forced dismount is not broken by it.
This patch has been tested by cperciva@ extensively against AmazonEFS.
Reported by: cperciva
Tested by: cperciva
MFC after: 2 weeks
Larry Rosenman reported a crash on freebsd-current@ which was caused by
a premature release of the krpc backchannel socket structure.
I believe this was caused by a race between the SVC_RELEASE() in clnt_vc.c
and the xprt_unregister() in the higher layer (clnt_rc.c), which tried
to lock the mutex in the xprt structure and crashed.
This patch fixes this by removing the xprt_unregister() in the clnt_vc
layer and allowing this to always be done by the clnt_rc (higher reconnect
layer).
Reported by: ler@lerctr.org
Tested by: ler@letctr.org
MFC after: 2 weeks
Renumber cluase 4 to 3, per what everybody else did when BSD granted
them permission to remove clause 3. My insistance on keeping the same
numbering for legal reasons is too pedantic, so give up on that point.
Submitted by: Jan Schaumann <jschauma@stevens.edu>
Pull Request: https://github.com/freebsd/freebsd/pull/96
This patch adds a new function to the server krpc called
svcpool_close(). It is similar to svcpool_destroy(), but does not free
the data structures, so that the pool can be used again.
This function is then used instead of svcpool_destroy(),
svcpool_create() when the nfsd threads are killed.
PR: 204340
Reported by: Panzura
Approved by: rmacklem
Obtained from: rmacklem
MFC after: 1 week
and getboottimebin(9) KPI. Change consumers of boottime to use the
KPI. The variables were renamed to avoid shadowing issues with local
variables of the same name.
Issue is that boottime* should be adjusted from tc_windup(), which
requires them to be members of the timehands structure. As a
preparation, this commit only introduces the interface.
Some uses of boottime were found doubtful, e.g. NLM uses boottime to
identify the system boot instance. Arguably the identity should not
change on the leap second adjustment, but the commit is about the
timekeeping code and the consumers were kept bug-to-bug compatible.
Tested by: pho (as part of the bigger patch)
Reviewed by: jhb (same)
Discussed with: bde
Sponsored by: The FreeBSD Foundation
MFC after: 1 month
X-Differential revision: https://reviews.freebsd.org/D7302
Similar to r300836, r301800, and r302550, cl and ct will always
be non-NULL as they're allocated using the mem_alloc routines,
which always use `malloc(..., M_WAITOK)`.
MFC after: 1 week
Reported by: Coverity
CID: 1007342
Sponsored by: EMC / Isilon Storage Division
Similar to r300836 and r301800, cl and cu will always be non-NULL as they're
allocated using the mem_alloc routines, which always use
`malloc(..., M_WAITOK)`.
Deobfuscating the cleanup path fixes a leak where if cl was NULL and
cu was not, cu would not be free'd, and also removes a duplicate test for
cl not being NULL.
MFC after: 1 week
Reported by: Coverity
CID: 1007033, 1007344
Sponsored by: EMC / Isilon Storage Division
Similar to r300836, cl and ct will always be non-NULL as they're allocated
using the mem_alloc routines, which always use `malloc(..., M_WAITOK)`.
Deobfuscating the cleanup path fixes a leak where if cl was NULL and
ct was not, ct would not be free'd, and also removes a duplicate test for
cl not being NULL.
Approved by: re (gjb)
Differential Revision: https://reviews.freebsd.org/D6801
MFC after: 1 week
Reported by: Coverity
CID: 1229999
Reviewed by: cem
Sponsored by: EMC / Isilon Storage Division
Both cd and xprt will be non-NULL after their respective malloc(9) wrappers are
called (mem_alloc and svc_xprt_alloc, which calls mem_alloc) as mem_alloc
always gets called with M_WAITOK|M_ZERO today. Thus, testing for them being
non-NULL is incorrect -- it misleads Coverity and it misleads the reader.
Remove some unnecessary NULL initializations as a follow up to help solidify
the fact that these pointers will be initialized properly in sys/rpc/.. with
the interfaces the way they are currently.
Differential Revision: https://reviews.freebsd.org/D6572
MFC after: 2 weeks
Reported by: Coverity
CID: 1007338, 1007339, 1007340
Reviewed by: markj, truckman
Sponsored by: EMC / Isilon Storage Division
The mem_alloc macro calls calloc (userspace) / malloc(.., M_WAITOK|M_ZERO)
under the covers, so zeroing out memory is already handled by the underlying
calls
MFC after: 1 week
Sponsored by: EMC / Isilon Storage Division
'buf.value' was previously treated as a nul-terminated string, but only
allocated with strlen() space. Rectify this.
Reported by: Coverity
CID: 1007639
Sponsored by: EMC / Isilon Storage Division
sosend(). The only release on the xp_snt_cnt is done after sosend(),
with an intent to synchronize with load_acq in svc_vc_ack().
Reviewed by: alc
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
implementation.
The kernel RPC code, which is responsible for the low-level scheduling
of incoming NFS requests, contains a throttling mechanism that
prevents too much kernel memory from being tied up by NFS requests
that are being serviced. When the throttle is engaged, the RPC layer
stops servicing incoming NFS sockets, resulting ultimately in
backpressure on the clients (if they're using TCP). However, this is
a very heavy-handed mechanism as it prevents all clients from making
any requests, regardless of how heavy or light they are. (Thus, when
engaged, the throttle often prevents clients from even mounting the
filesystem.) The throttle mechanism applies specifically to requests
that have been received by the RPC layer (from a TCP or UDP socket)
and are queued waiting to be serviced by one of the nfsd threads; it
does not limit the amount of backlog in the socket buffers.
The original implementation limited the total bytes of queued requests
to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB.
The former limit seems reasonable, since requests queued in the socket
buffers and replies being constructed to the requests in progress will
all require some amount of network memory, but the 45 MiB limit is
plainly ridiculous for modern memory sizes: when running 256 service
threads on a busy server, 45 MiB would result in just a single
maximum-sized NFS3PROC_WRITE queued per thread before throttling.
Removing this limit exposed integer-overflow bugs in the original
computation, and related bugs in the routines that actually account
for the amount of traffic enqueued for service threads. The old
implementation also attempted to reduce accounting overhead by
batching updates until each queue is fully drained, but this is prone
to livelock, resulting in repeated accumulate-throttle-drain cycles on
a busy server. Various data types are changed to long or unsigned
long; explicit 64-bit types are not used due to the unavailability of
64-bit atomics on many 32-bit platforms, but those platforms also
cannot support nmbclusters large enough to cause overflow.
This code (in a 10.1 kernel) is presently running on production NFS
servers at CSAIL.
Summary of this revision:
* Removes 45 MiB limit on requests queued for nfsd service threads
* Fixes integer-overflow and signedness bugs
* Avoids unnecessary throttling by not deferring accounting for
completed requests
Differential Revision: https://reviews.freebsd.org/D2165
Reviewed by: rmacklem, mav
MFC after: 30 days
Relnotes: yes
Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
feature is to quisce the system before suspend.
Stop is implemented by reusing the thread_single(9) with the special
mode SINGLE_ALLPROC. SINGLE_ALLPROC differs from the existing
single-threading modes by allowing (requiring) caller to operate on
other process. Interruptible sleeps for !TDF_SBDRY threads are
suspended like SIGSTOP does it, instead of aborting the sleep, like
SINGLE_NO_EXIT, to avoid spurious EINTRs on resume.
Provide debugging sysctl debug.stop_all_proc, which causes total stop
and suspends syncer, while waiting for variable reset for resume. It
is used for debugging; should be removed after the real use of the
interface is added.
In collaboration with: pho
Discussed with: avg
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
This is not correct at least for the stop requests. Check for stop
conditions and suspend threads if requested.
Reported and tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
sb_cc member of struct sockbuf to a couple of inline functions:
sbavail() and sbused()
Right now they are equal, but once notion of "not ready socket buffer data",
will be checked in, they are going to be different.
Sponsored by: Netflix
Sponsored by: Nginx, Inc.
into head. The code is not believed to have any effect
on the semantics of non-NFSv4.1 server behaviour.
It is a rather large merge, but I am hoping that there will
not be any regressions for the NFS server.
MFC after: 1 month
Old design with unified thread pool was good from the point of thread
utilization. But single pool-wide mutex became huge congestion point
for systems with many CPUs. To reduce the congestion create several
thread groups within a pool (one group for every 6 CPUs and 12 threads),
each group with own mutex. Each connection during its registration is
assigned to one of the groups in round-robin fashion. File affinify
code may still move requests between the groups, but otherwise groups
are self-contained.
MFC after: 2 weeks
Sponsored by: iXsystems, Inc.
This allows to slightly simplify svc_run_internal() code: if we processed
all the requests in a queue, then we know that new one will not appear.
MFC after: 2 weeks
New algorithm does not create additional lock congestion, while some races
it includes should not be a problem. Those races may keep requests in DRC
cache for some more time by returning ACK position smaller then actual,
but it still should be able to drop thems when proper ACK finally read.
Races of the original algorithm based on TCP seq number were worse because
they happened when reply sequence number were recorded. After that even
correctly read ACKs could not clean DRC sometimes.
- Introduce additional hash to group requests by hash of sockref. This
allows to process TCP acknowledgements without looping though all the cache,
and as result allows to do it every time.
- Indroduce additional callbacks to notify application layer about sockets
disconnection. Without this last few requests processed just before socket
disconnection never processed their ACKs and stuck in cache for many hours.
- Implement transport-specific method for tracking reply acknowledgements.
New implementation does not cross multiple stack layers to get the data and
does not have race conditions that previously made some requests stuck
in cache. This could be done more efficiently at sockbuf layer, but that
would broke some KBIs, while I don't know other consumers for it aside NFS.
- Instead of traversing all DRC twice per request, run cleaning only once
per request, and except in some conditions traverse only single hash slot
at a time.
Together this limits NFS DRC growth only to situations of real connectivity
problems. If network is working well, and so all replies are acknowledged,
cache remains almost empty even after hours of heavy load. Without this
change on the same test cache was growing to many thousand requests even
with perfectly working local network.
As another result this reduces CPU time spent on the DRC handling during
SPEC NFS benchmark from about 10% to 0.5%.
Sponsored by: iXsystems, Inc.
