Based on its name you would think that rqst_should_sleep() would be
read-only, not changing anything. But in fact it will clear
SP_TASK_PENDING if that was set. This is surprising, and it blurs the
line between "check for work to do" and "dequeue work to do".
So change the "test_and_clear" to simple "test" and clear the bit once
the thread has decided to wake up and return to the caller.
With this, it makes sense to *always* set SP_TASK_PENDING when asked,
rather than to set it only if no thread could be woken up.
[ cel: Previously TASK_PENDING indicated there is work waiting but no
idle threads were found to pick up that work. After this patch, it acts
as an XPT_BUSY flag for wake-ups that have no associated xprt. ]
Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Document the API contract and remove stale or obvious comments.
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
svc_xprt_enqueue() can be costly, since it involves selecting and
waking up a process.
More than one enqueue is done per incoming RPC. For example,
svc_data_ready() enqueues, and so does svc_xprt_receive(). Also, if
an RPC message requires more than one call to ->recvfrom() to
receive it fully, each one of those calls does an enqueue.
To get a sense of the average number of transport enqueue operations
needed to process an incoming RPC message, re-use the "packets" pool
stat. Track the number of complete RPC messages processed by each
thread pool.
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Refactor: Extract the loop that finds an idle service thread from
svc_xprt_enqueue() and svc_wake_up(). Both functions do just about
the same thing.
Note that svc_wake_up() currently does not hold the RCU read lock
while waking the target thread. It indeed should hold the lock, just
as svc_xprt_enqueue() does, to ensure the rqstp does not vanish
during the wake-up. This patch adds the RCU lock for svc_wake_up().
Note that shrinking the pool thread count is rare, and calls to
svc_wake_up() are also quite infrequent. In practice, this race is
very unlikely to be hit, so we are not marking the lock fix for
stable backport at this time.
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The xpt_flags field frequently changes between the time that
svc_xprt_ready() grabs a copy and execution flow arrives at the
tracepoint at the tail of svc_xprt_enqueue(). In fact, there's
usually a sleep/wake-up in there, so those flags are almost
guaranteed to be different.
It would be more useful to record the exact flags that were used to
decide whether the transport is ready, so move the tracepoint.
Moving it means the tracepoint can't pick up the waker's pid. That
can be added to struct svc_rqst if it turns out that is important.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
In addition to the benefits of using an enum rather than a set of
macros, we now have a named type that can improve static type
checking of function return values.
As part of this change, I removed a stale comment from svcauth.h;
the return values from current implementations of the
auth_ops::release method are all zero/negative errno, not the SVC_OK
enum values as the old comment suggested.
Suggested-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When a sequence of numbers are needed for internal-use only, an enum is
typically best. The sequence will inevitably need to be changed one
day, and having an enum means the developer doesn't need to think about
renumbering after insertion or deletion. Such patches will be easier
to review.
Suggested-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When a sequence of numbers are needed for internal-use only, an enum is
typically best. The sequence will inevitably need to be changed one
day, and having an enum means the developer doesn't need to think about
renumbering after insertion or deletion. Such patches will be easier
to review.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When a sequence of numbers are needed for internal-use only, an enum is
typically best. The sequence will inevitably need to be changed one
day, and having an enum means the developer doesn't need to think about
renumbering after insertion or deletion. Such patches will be easier
to review.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When a sequence of numbers are needed for internal-use only, an enum is
typically best. The sequence will inevitably need to be changed one
day, and having an enum means the developer doesn't need to think about
renumbering after insertion or deletion. Such patches will be easier
to review.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Most svc threads have no interest in a timeout.
nfsd sets it to 1 hour, but this is a wart of no significance.
lockd uses the timeout so that it can call nlmsvc_retry_blocked().
It also sometimes calls svc_wake_up() to ensure this is called.
So change lockd to be consistent and always use svc_wake_up() to trigger
nlmsvc_retry_blocked() - using a timer instead of a timeout to
svc_recv().
And change svc_recv() to not take a timeout arg.
This makes the sp_threads_timedout counter always zero.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
svc_recv() currently returns a 0 on success or one of two errors:
- -EAGAIN means no message was successfully received
- -EINTR means the thread has been told to stop
Previously nfsd would stop as the result of a signal as well as
following kthread_stop(). In that case the difference was useful: EINTR
means stop unconditionally. EAGAIN means stop if kthread_should_stop(),
continue otherwise.
Now threads only exit when kthread_should_stop() so we don't need the
distinction.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
All callers of svc_recv() go on to call svc_process() on success.
Simplify callers by having svc_recv() do that for them.
This loses one call to validate_process_creds() in nfsd. That was
debugging code added 14 years ago. I don't think we need to keep it.
Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Now that the last nfsd thread is stopped by an explicit act of calling
svc_set_num_threads() with a count of zero, we only have a limited
number of places that can happen, and don't need to call
nfsd_last_thread() in nfsd_put()
So separate that out and call it at the two places where the number of
threads is set to zero.
Move the clearing of ->nfsd_serv and the call to svc_xprt_destroy_all()
into nfsd_last_thread(), as they are really part of the same action.
nfsd_put() is now a thin wrapper around svc_put(), so make it a static
inline.
nfsd_put() cannot be called after nfsd_last_thread(), so in a couple of
places we have to use svc_put() instead.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Previously a thread could exit asynchronously (due to a signal) so some
care was needed to hold nfsd_mutex over the last svc_put() call. Now a
thread can only exit when svc_set_num_threads() is called, and this is
always called under nfsd_mutex. So no care is needed.
Not only is the mutex held when a thread exits now, but the svc refcount
is elevated, so the svc_put() in svc_exit_thread() will never be a final
put, so the mutex isn't even needed at this point in the code.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The original implementation of nfsd used signals to stop threads during
shutdown.
In Linux 2.3.46pre5 nfsd gained the ability to shutdown threads
internally it if was asked to run "0" threads. After this user-space
transitioned to using "rpc.nfsd 0" to stop nfsd and sending signals to
threads was no longer an important part of the API.
In commit 3ebdbe5203 ("SUNRPC: discard svo_setup and rename
svc_set_num_threads_sync()") (v5.17-rc1~75^2~41) we finally removed the
use of signals for stopping threads, using kthread_stop() instead.
This patch makes the "obvious" next step and removes the ability to
signal nfsd threads - or any svc threads. nfsd stops allowing signals
and we don't check for their delivery any more.
This will allow for some simplification in later patches.
A change worth noting is in nfsd4_ssc_setup_dul(). There was previously
a signal_pending() check which would only succeed when the thread was
being shut down. It should really have tested kthread_should_stop() as
well. Now it just does the latter, not the former.
Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
lockd allows SIGKILL and responds by dropping all locks and restarting
the grace period. This functionality has been present since 2.1.32 when
lockd was added to Linux.
This functionality is undocumented and most likely added as a useful
debug aid. When there is a need to drop locks, the better approach is
to use /proc/fs/nfsd/unlock_*.
This patch removes SIGKILL handling as part of preparation for removing
all signal handling from sunrpc service threads.
Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
clang's static analysis warning: fs/lockd/mon.c: line 293, column 2:
Null pointer passed as 2nd argument to memory copy function.
Assuming 'hostname' is NULL and calling 'nsm_create_handle()', this will
pass NULL as 2nd argument to memory copy function 'memcpy()'. So return
NULL if 'hostname' is invalid.
Fixes: 77a3ef33e2 ("NSM: More clean up of nsm_get_handle()")
Signed-off-by: Su Hui <suhui@nfschina.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Remove kernel-doc warning in exportfs:
fs/exportfs/expfs.c:395: warning: Function parameter or member 'parent'
not described in 'exportfs_encode_inode_fh'
Signed-off-by: Zhu Wang <wangzhu9@huawei.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
With large NFS WRITE requests on TCP, I measured 5-10 thread wake-
ups to receive each request. This is because the socket layer
calls ->sk_data_ready() frequently, and each call triggers a
thread wake-up. Each recvmsg() seems to pull in less than 100KB.
Have the socket layer hold ->sk_data_ready() calls until the full
incoming message has arrived to reduce the wake-up rate.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Flamegraph analysis showed that the cork/uncork calls consume
nearly a third of the CPU time spent in svc_tcp_sendto(). The
other two consumers are mutex lock/unlock and svc_tcp_sendmsg().
Now that svc_tcp_sendto() coalesces RPC messages properly, there
is no need to introduce artificial delays to prevent sending
partial messages.
After applying this change, I measured a 1.2K read IOPS increase
for 8KB random I/O (several percent) on 56Gb IP over IB.
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Commit da1661b93b ("SUNRPC: Teach server to use xprt_sock_sendmsg
for socket sends") modified svc_udp_sendto() to use xprt_sock_sendmsg()
because we originally believed xprt_sock_sendmsg() would be needed
for TLS support. That does not actually appear to be the case.
In addition, the linkage between the client and server send code has
been a bit of a maintenance headache because of the distinct ways
that the client and server handle memory allocation.
Going forward, eventually the XDR layer will deal with its buffers
in the form of bio_vec arrays, so convert this function accordingly.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
There is now enough infrastructure in place to combine the stream
record marker into the biovec array used to send each outgoing RPC
message on TCP. The whole message can be more efficiently sent with
a single call to sock_sendmsg() using a bio_vec iterator.
Note that this also helps with RPC-with-TLS: the TLS implementation
can now clearly see where the upper layer message boundaries are.
Before, it would send each component of the xdr_buf (record marker,
head, page payload, tail) in separate TLS records.
Suggested-by: David Howells <dhowells@redhat.com>
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Add a helper to convert a whole xdr_buf directly into an array of
bio_vecs, then send this array instead of iterating piecemeal over
the xdr_buf containing the outbound RPC message.
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
A well-formed NFSv4 ACL will always contain OWNER@/GROUP@/EVERYONE@
ACEs, but there is no requirement for inheritable entries for those
entities. POSIX ACLs must always have owner/group/other entries, even for a
default ACL.
nfsd builds the default ACL from inheritable ACEs, but the current code
just leaves any unspecified ACEs zeroed out. The result is that adding a
default user or group ACE to an inode can leave it with unwanted deny
entries.
For instance, a newly created directory with no acl will look something
like this:
# NFSv4 translation by server
A::OWNER@:rwaDxtTcCy
A::GROUP@:rxtcy
A::EVERYONE@:rxtcy
# POSIX ACL of underlying file
user::rwx
group::r-x
other::r-x
...if I then add new v4 ACE:
nfs4_setfacl -a A:fd:1000:rwx /mnt/local/test
...I end up with a result like this today:
user::rwx
user:1000:rwx
group::r-x
mask::rwx
other::r-x
default:user::---
default:user:1000:rwx
default:group::---
default😷:rwx
default:other::---
A::OWNER@:rwaDxtTcCy
A::1000:rwaDxtcy
A::GROUP@:rxtcy
A::EVERYONE@:rxtcy
D:fdi:OWNER@:rwaDx
A:fdi:OWNER@:tTcCy
A:fdi:1000:rwaDxtcy
A:fdi:GROUP@:tcy
A:fdi:EVERYONE@:tcy
...which is not at all expected. Adding a single inheritable allow ACE
should not result in everyone else losing access.
The setfacl command solves a silimar issue by copying owner/group/other
entries from the effective ACL when none of them are set:
"If a Default ACL entry is created, and the Default ACL contains no
owner, owning group, or others entry, a copy of the ACL owner,
owning group, or others entry is added to the Default ACL.
Having nfsd do the same provides a more sane result (with no deny ACEs
in the resulting set):
user::rwx
user:1000:rwx
group::r-x
mask::rwx
other::r-x
default:user::rwx
default:user:1000:rwx
default:group::r-x
default😷:rwx
default:other::r-x
A::OWNER@:rwaDxtTcCy
A::1000:rwaDxtcy
A::GROUP@:rxtcy
A::EVERYONE@:rxtcy
A:fdi:OWNER@:rwaDxtTcCy
A:fdi:1000:rwaDxtcy
A:fdi:GROUP@:rxtcy
A:fdi:EVERYONE@:rxtcy
Reported-by: Ondrej Valousek <ondrej.valousek@diasemi.com>
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=2136452
Suggested-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Since commit 49b28684fd ("nfsd: Remove deprecated nfsctl system call and related code.")
these declarations are unused, so can remove it.
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This patch fixes races when lockd accesses the global nlm_blocked list.
It was mostly safe to access the list because everything was accessed
from the lockd kernel thread context but there exist cases like
nlmsvc_grant_deferred() that could manipulate the nlm_blocked list and
it can be called from any context.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
In the event that we can't fetch post_op_attr attributes, we still need
to set a value for the after_change. The operation has already happened,
so we're not able to return an error at that point, but we do want to
ensure that the client knows that its cache should be invalidated.
If we weren't able to fetch post-op attrs, then just set the
after_change to before_change + 1. The atomic flag should already be
clear in this case.
Suggested-by: Neil Brown <neilb@suse.de>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
At one time, nfsd would scrape inode information directly out of struct
inode in order to populate the change_info4. At that time, the BUG_ON in
set_change_info made some sense, since having it unset meant a coding
error.
More recently, it calls vfs_getattr to get this information, which can
fail. If that fails, fh_pre_saved can end up not being set. While this
situation is unfortunate, we don't need to crash the box.
Move set_change_info to nfs4proc.c since all of the callers are there.
Revise the condition for setting "atomic" to also check for
fh_pre_saved. Drop the BUG_ON and just have it zero out both
change_attr4s when this occurs.
Reported-by: Boyang Xue <bxue@redhat.com>
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=2223560
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Collecting pre_op_attrs can fail, in which case it's probably best to
fail the whole operation.
Change fh_fill_pre_attrs and fh_fill_both_attrs to return __be32, and
have the callers check the return code and abort the operation if it's
not nfs_ok.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
I got this today from modpost:
WARNING: modpost: missing MODULE_DESCRIPTION() in fs/nfsd/nfsd.o
Add a module description.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The svc_ prefix is identified with the SunRPC layer. Although the
duplicate reply cache caches RPC replies, it is only for the NFS
protocol. Rename the struct to better reflect its purpose.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Over time I'd like to see NFS-specific fields moved out of struct
svc_rqst, which is an RPC layer object. These fields are layering
violations.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Avoid holding the bucket lock while freeing cache entries. This
change also caps the number of entries that are freed when the
shrinker calls to reduce the shrinker's impact on the cache's
effectiveness.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Enable nfsd_prune_bucket() to drop the bucket lock while calling
kfree(). Use the same pattern that Jeff recently introduced in the
NFSD filecache.
A few percpu operations are moved outside the lock since they
temporarily disable local IRQs which is expensive and does not
need to be done while the lock is held.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
To reduce contention on the bucket locks, we must avoid calling
kfree() while each bucket lock is held.
Start by refactoring nfsd_reply_cache_free_locked() into a helper
that removes an entry from the bucket (and must therefore run under
the lock) and a second helper that frees the entry (which does not
need to hold the lock).
For readability, rename the helpers nfsd_cacherep_<verb>.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
All supported encryption types now use the same context import
function.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This code is now always on, so the ifdef can be removed.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Make it impossible to enable support for the DES or DES3 Kerberos
encryption types in SunRPC. These enctypes were deprecated by RFCs
6649 and 8429 because they are known to be insecure.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This patch grants write delegations for OPEN with NFS4_SHARE_ACCESS_WRITE
if there is no conflict with other OPENs.
Write delegation conflicts with another OPEN, REMOVE, RENAME and SETATTR
are handled the same as read delegation using notify_change,
try_break_deleg.
The NFSv4.0 protocol does not enable a server to determine that a
conflicting GETATTR originated from the client holding the
delegation versus coming from some other client. With NFSv4.1 and
later, the SEQUENCE operation that begins each COMPOUND contains a
client ID, so delegation recall can be safely squelched in this case.
With NFSv4.0, however, the server must recall or send a CB_GETATTR
(per RFC 7530 Section 16.7.5) even when the GETATTR originates from
the client holding that delegation.
An NFSv4.0 client can trigger a pathological situation if it always
sends a DELEGRETURN preceded by a conflicting GETATTR in the same
COMPOUND. COMPOUND execution will always stop at the GETATTR and the
DELEGRETURN will never get executed. The server eventually revokes
the delegation, which can result in loss of open or lock state.
Tracepoint added to track whether read or write delegation is granted.
Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Replace the -1 (no limit) with a zero (no reserved space).
This prevents certain non-determinant client behavior, such as
silly-renaming a file when the only open reference is a write
delegation. Such a rename can leave unexpected .nfs files in a
directory that is otherwise supposed to be empty.
Note that other server implementations that support write delegation
also set this field to zero.
Suggested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
If the GETATTR request on a file that has write delegation in effect and
the request attributes include the change info and size attribute then
the write delegation is recalled. If the delegation is returned within
30ms then the GETATTR is serviced as normal otherwise the NFS4ERR_DELAY
error is returned for the GETATTR.
Add counter for write delegation recall due to conflict GETATTR. This is
used to evaluate the need to implement CB_GETATTR to adoid recalling the
delegation with conflit GETATTR.
Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Remove the check for F_WRLCK in generic_add_lease to allow file_lock
to be used for write delegation.
First consumer is NFSD.
Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
