Linux has had an "nconnect" NFS mount option for some time.
It specifies that N (up to 16) TCP connections are to created for a mount,
instead of just one TCP connection.
A discussion on freebsd-net@ indicated that this could improve
client<-->server network bandwidth, if either the client or server
have one of the following:
- multiple network ports aggregated to-gether with lagg/lacp.
- a fast NIC that is using multiple queues
It does result in using more IP port#s and might increase server
peak load for a client.
One difference from the Linux implementation is that this implementation
uses the first TCP connection for all RPCs composed of small messages
and uses the additional TCP connections for RPCs that normally have
large messages (Read/Readdir/Write). The Linux implementation spreads
all RPCs across all TCP connections in a round robin fashion, whereas
this implementation spreads Read/Readdir/Write across the additional
TCP connections in a round robin fashion.
(cherry picked from commit 1e0a518d65)
hst will be nul-terminated but the remaining space in the buffer is left
uninitialized. Avoid copying the entire buffer to ensure that
uninitialized bytes are not leaked via statfs(2).
Reported by: KMSAN
Reviewed by: rmacklem
Sponsored by: The FreeBSD Foundation
(cherry picked from commit 44de1834b5)
Commit 844aa31c6d added cache_enter_time_flags(), specifically
so that the NFS client could specify that cache enter replace
any stale entry for the same name. Doing so avoids a KASSERT()
panic() in cache_enter_time(), as reported by the PR.
This patch uses cache_enter_time_flags() for Readdirplus, to
avoid the panic(), since it is impossible for the NFS client
to know if another client (or a local process on the NFS server)
has replaced a file with another file of the same name.
This patch only affects NFS mounts that use the "rdirplus"
mount option.
There may be other places in the NFS client where this needs
to be done, but no panic() has been observed during testing.
PR: 257043
(cherry picked from commit 7f5508fe78)
When NFSv4.1 support was added to the client, the implementation was
still experimental and, as such, the default minor version was set to 0.
Since the NFSv4.1 client implementation is now believed to be solid
and the NFSv4.1/4.2 protocol is significantly better than NFSv4.0,
I beieve that NFSv4.1/4.2 should be used where possible.
This patch changes the default minor version for NFSv4 to be the highest
minor version supported by the NFSv4 server. If a specific minor version
is desired, the "minorversion" mount option can be used to override
this default. This is compatible with the Linux NFSv4 client behaviour.
This was discussed on freebsd-current@ in mid-May 2021 under
the subject "changing the default NFSv4 minor version" and
the consensus seemed to be support for this change.
It also appeared that changing this for FreeBSD 13.1 was
not considered a POLA violation, so long as UPDATING
and RELNOTES entries were made for it.
(cherry picked from commit a145cf3f73)
When the NFSv4.0 client was implemented, acquisition of a clientid
via SetClientID/SetClientIDConfirm was done upon the first Open,
since that was when it was needed. NFSv4.1/4.2 acquires the clientid
during mount (via ExchangeID/CreateSession), since the associated
session is required during mount.
This patch modifies the NFSv4.0 mount so that it acquires the
clientid during mount. This simplifies the code and makes it
easy to implement "find the highest minor version supported by
the NFSv4 server", which will be done for the default minorversion
in a future commit.
The "start_renewthread" argument for nfscl_getcl() is replaced
by "tryminvers", which will be used by the aforementioned
future commit.
(cherry picked from commit aed98fa5ac)
A problem was reported via email, where a large (130000+) accumulation
of NFSv4 opens on an NFSv4 mount caused significant lock contention
on the mutex used to protect the client mount's open/lock state.
Although the root cause for the accumulation of opens was not
resolved, it is obvious that the NFSv4 client is not designed to
handle 100000+ opens efficiently.
For a common case where delegations are not being issued by the
NFSv4 server, the code acquires the mutex lock for open/lock state,
finds the delegation list empty and just unlocks the mutex and returns.
This patch adds an NFS mount point flag that is set when a delegation
is issued for the mount. Then the patched code checks for this flag
before acquiring the open/lock mutex, avoiding the need to acquire
the lock for the case where delegations are not being issued by the
NFSv4 server.
This change appears to be performance neutral for a small number
of opens, but should reduce lock contention for a large number of opens
for the common case where server is not issuing delegations.
This commit should not affect the high level semantics of delegation
handling.
(cherry picked from commit 5e5ca4c8fc)
Pre-r318997 the code looked like:
if (vp->v_mount->mnt_stat.f_fsid.val[0] != (uint32_t)np->n_vattr.na_filesid[0])
vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
Doing this assignment got lost by r318997 and, as such, NFSv4 mounts
of servers with trees of file systems on the server is broken, due to duplicate
fileno values for the same st_dev/va_fsid.
Although I could have re-introduced the assignment, since the value of
na_filesid[0] is not guaranteed to be unique across the server file systems,
I felt it was better to always do the hash for na_filesid[0,1].
Since dev_t (st_dev/va_fsid) is now 64bits, I switched to a 64bit hash.
There is a slight chance of a hash conflict where 2 different na_filesid
values map to same va_fsid, which will be documented in the BUGS
section of the man page for mount_nfs(8). Using a table to keep track
of mappings to catch conflicts would not easily scale to 10,000+ server file
systems and, when the conflict occurs, it only results in fts(3) reporting
a "directory cycle" under certain circumstances.
(cherry picked from commit 03c81af249)
A problem was reported via email, where a large (130000+) accumulation
of NFSv4 opens on an NFSv4 mount caused significant lock contention
on the mutex used to protect the client mount's open/lock state.
Although the root cause for the accumulation of opens was not
resolved, it is obvious that the NFSv4 client is not designed to
handle 100000+ opens efficiently. When searching for an open,
usually for a match by file handle, a linear search of all opens
is done.
Commit 3f7e14ad93 added a hash table of lists hashed on file handle
for the opens. This patch uses the hash lists for searching for
a matching open based of file handle instead of an exhaustive
linear search of all opens.
This change appears to be performance neutral for a small number
of opens, but should improve expected performance for a large
number of opens.
This commit should not affect the high level semantics of open
handling.
(cherry picked from commit 96b40b8967)
A problem was reported via email, where a large (130000+) accumulation
of NFSv4 opens on an NFSv4 mount caused significant lock contention
on the mutex used to protect the client mount's open/lock state.
Although the root cause for the accumulation of opens was not
resolved, it is obvious that the NFSv4 client is not designed to
handle 100000+ opens efficiently. When searching for an open,
usually for a match by file handle, a linear search of all opens
is done.
Commit 3f7e14ad93 added a hash table of lists hashed on file handle
for the opens. This patch uses the hash lists for searching for
a matching open based of file handle instead of an exhaustive
linear search of all opens.
This change appears to be performance neutral for a small number
of opens, but should improve expected performance for a large
number of opens. This patch also moves any found match to the front
of the hash list, to try and maintain the hash lists in recently
used ordering (least recently used at the end of the list).
This commit should not affect the high level semantics of open
handling.
(cherry picked from commit 724072ab1d)
A problem was reported via email, where a large (130000+) accumulation
of NFSv4 opens on an NFSv4 mount caused significant lock contention
on the mutex used to protect the client mount's open/lock state.
Although the root cause for the accumulation of opens was not
resolved, it is obvious that the NFSv4 client is not designed to
handle 100000+ opens efficiently. When searching for an open,
usually for a match by file handle, a linear search of all opens
is done.
This patch adds a table of hash lists for the opens, hashed on
file handle. This table will be used by future commits to
search for an open based on file handle more efficiently.
(cherry picked from commit 3f7e14ad93)
The most difficult NFSv4 client recovery case happens when the
lease has expired on the server. For NFSv4.0, the client will
receive a NFSERR_EXPIRED reply from the server to indicate this
has happened.
For NFSv4.1/4.2, most RPCs have a Sequence operation and, as such,
the client will receive a NFSERR_BADSESSION reply when the lease
has expired for these RPCs. The client will then call nfscl_recover()
to handle the NFSERR_BADSESSION reply. However, for the expired lease
case, the first reclaim Open will fail with NFSERR_NOGRACE.
This patch recognizes this case and calls nfscl_expireclient()
to handle the recovery from an expired lease.
This patch only affects NFSv4.1/4.2 mounts when the lease
expires on the server, due to a network partitioning that
exceeds the lease duration or similar.
(cherry picked from commit c28cb257dd)
There is a NFSv4 file attribute called TimeCreate
that can be used for va_birthtime.
r362175 added some support for use of TimeCreate.
This patch completes support of va_birthtime by adding
support for setting this attribute to the server.
It also eanbles the client to
acquire and set the attribute for a NFSv4
server that supports the attribute.
(cherry picked from commit dd02d9d605)
Commit aad780464f added a function called nfscl_delegreturnvp()
to return delegations during the NFS VOP_RECLAIM().
The function erroneously assumed that nm_clp would
be non-NULL. It will be NULL for NFSV4.0 mounts until
a regular file is opened. It will also be NULL during
vflush() in nfs_unmount() for a forced dismount.
This patch adds a check for clp == NULL to fix this.
Also, since it makes no sense to call nfscl_delegreturnvp()
during a forced dismount, the patch adds a check for that
case and does not do the call during forced dismounts.
PR: 255436
(cherry picked from commit f6fec55fe3)
After a vnode is recycled it can no longer be
acquired via vfs_hash_get() and, as such,
a delegation for the vnode cannot be recalled.
In the unlikely event that a delegation still
exists when the vnode is being recycled, return
the delegation since it will no longer be
recallable.
Until you have this patch in your NFSv4 client,
you should consider avoiding the use of delegations.
(cherry picked from commit aad780464f)
When loading attributes from the cache, the NFS client is careful to
copy only the fields that it initialized. After fetching attributes
from the server, however, it would copy the entire vattr structure
initialized from the RPC response, so uninitialized stack bytes would
end up being copied to userspace. In particular, va_birthtime (v2 and
v3) and va_gen (v3) had this problem.
Use a common subroutine to copy fields provided by the NFS client, and
ensure that we provide a dummy va_gen for the v3 case.
Reviewed by: rmacklem
Reported by: KMSAN
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D30090
(cherry picked from commit 8bde6d15d1)
Without this patch, if a NFSv4 server recalled a
delegation when the file is not open, the renew
thread would block in the NFS VOP_INACTIVE()
trying to acquire the client state lock that it
already holds.
This patch fixes the problem by delaying the
vrele() call until after the client state
lock is released.
This bug has been in the NFSv4 client for
a long time, but since it only affects
delegation when recalled due to another
client opening the file, it got missed
during previous testing.
Until you have this patch in your client,
you should avoid the use of delegations.
(cherry picked from commit 02695ea890)
During a recent testing event, it was reported that the NFSv4.1/4.2
server erroneously bound the back channel to a new TCP connection.
RFC5661 specifies that the fore channel is implicitly bound to a
new TCP connection when an RPC with Sequence (almost any of them)
is done on it. For the back channel to be bound to the new TCP
connection, an explicit BindConnectionToSession must be done as
the first RPC on the new connection.
Since new TCP connections are created by the "reconnect" layer
(sys/rpc/clnt_rc.c) of the krpc, this patch adds an optional
upcall done by the krpc whenever a new connection is created.
The patch also adds the specific upcall function that does a
BindConnectionToSession and configures the krpc to call it
when required.
This is necessary for correct interoperability with NFSv4.1/NFSv4.2
servers when the nfscbd daemon is running.
If doing NFSv4.1/NFSv4.2 mounts without this patch, it is
recommended that the nfscbd daemon not be running and that
the "pnfs" mount option not be specified.
PR: 254840
(cherry picked from commit 7763814fc9)
Commit fdc9b2d50f replaced a couple of while loops with LIST_FOREACH()
loops. This patch factors the body of that loop out into a separate
function called nfscl_checkown().
This prepares the code for future changes to use a hash table of
lists for open searches via file handle.
This patch should not result in a semantics change.
(cherry picked from commit 4e6c2a1ee9)
This patch replaces a couple of while() loops with LIST_FOREACH() loops.
While here, declare a couple of variables "bool".
I think LIST_FOREACH() is preferred and makes the code more readable.
This also prepares the code for future changes to use a hash table of
lists for open searches via file handle.
This patch should not result in a semantics change.
(cherry picked from commit fdc9b2d50f)
If a delegation for a file has been acquired, the "oneopenown" option
was ignored when the local open was issued. This could result in multiple
openowners/opens for a file, that would be transferred to the server
when the delegation was recalled.
This would not be serious, but could result in more than one openowner.
Since the Amazon/EFS does not issue delegations, this probably never
occurs in practice.
Spotted during code inspection.
This small patch fixes the code so that it checks for "oneopenown"
when doing client local opens on a delegation.
(cherry picked from commit e61b29ab5d)
During a recent NFSv4 testing event a test server caused a hang
where "umount -N" failed. The renew thread was sleeping on "nfsv4lck"
and the "umount" was sleeping, waiting for the renew thread to
terminate.
This is the second of two patches that is hoped to fix the renew thread
so that it will terminate when "umount -N" is done on the mount.
This patch adds a 5second timeout on the msleep()s and checks for
the forced dismount flag so that the renew thread will
wake up and see the forced dismount flag. Normally a wakeup()
will occur in less than 5seconds, but if a premature return from
msleep() does occur, it will simply loop around and msleep() again.
The patch also adds the "mp" argument to nfsv4_lock() so that it
will return when the forced dismount flag is set.
While here, replace the nfsmsleep() wrapper that was used for portability
with the actual msleep() call.
(cherry picked from commit 82ee386c2a)
During a recent NFSv4 testing event a test server was replying
NFSERR_OLDSTATEID for layout stateids presented to the server
for LayoutReturn operations. Upon rereading RFC5661, it was
apparent that the FreeBSD NFSv4.1/4.2 pNFS client did not
maintain the seqid field of the layout stateid correctly.
This patch is believed to correct the problem. Tested against
a FreeBSD pNFS server with diagnostics added to check the stateid's
seqid did not indicate problems. Unfortunately, testing aginst
this server will not happen in the near future, so the fix may
not be correct yet.
(cherry picked from commit fd232a21bb)
During a recent virtual NFSv4 testing event, a bug in the FreeBSD client
was detected when doing I/O DS operations on a Flexible File Layout pNFS
server. For an NFSv3 DS, the Read/Write/Commit nfsstats were incremented
instead of the ReadDS/WriteDS/CommitDS counts.
This patch fixes this.
Only the RPC counts reported by nfsstat(1) were affected by this bug,
the I/O operations were performed correctly.
(cherry picked from commit c04199affe)
During a recent virtual NFSv4 testing event, a bug in the FreeBSD client
was detected when doing a File Layout pNFS DS I/O operation.
The size of the I/O operation was smaller than expected.
The I/O size is specified as a stripe unit size in bits 6->31 of nflh_util
in the layout. I had misinterpreted RFC5661 and had shifted the value
right by 6 bits. The correct interpretation is to use the value as
presented (it is always an exact multiple of 64), clearing bits 0->5.
This patch fixes this.
Without the patch, I/O through the DSs work, but the I/O size is 1/64th
of what is optimal.
(cherry picked from commit 94f2e42f5e)
During code inspection I noticed that the n_direofoffset field
of the NFS node was being manipulated without any lock being
held to make it SMP safe.
This patch adds locking of the NFS node's mutex around
handling of n_direofoffset to make it SMP safe.
I have not seen any failure that could be attributed to n_direofoffset
being manipulated concurrently by multiple processors, but I think this
is possible, since directories are read with shared vnode
locking, plus locks only on individual buffer cache blocks.
However, there have been as yet unexplained issues w.r.t reading
large directories over NFS that could have conceivably been caused
by concurrent manipulation of n_direofoffset.
(cherry picked from commit 15bed8c46b)
Commit 3fe2c68ba2 dealt with a panic in cache_enter_time() where
the vnode referred to the directory argument.
It would also be possible to get these panics if a broken
NFS server were to return the directory as an new object being
created within the directory or in a Lookup reply.
This patch adds checks to avoid the panics and logs
messages to indicate that the server is broken for the
file object creation cases.
(cherry picked from commit 3e04ab36ba)
in6_selectsrc() may call fib6_lookup() in some cases, which requires
epoch. Wrap in6_selectsrc* calls into epoch inside its users.
Mark it as requiring epoch by adding NET_EPOCH_ASSERT().
Differential Revision: https://reviews.freebsd.org/D28647
(cherry picked from commit 605284b894)
Juraj Lutter (otis@) reported a panic "dvp != vp not true" in
cache_enter_time() called from the NFS client's nfsrpc_readdirplus()
function.
This is specific to an NFSv3 mount with the "rdirplus" mount
option. Unlike NFSv4, NFSv3 replies to ReaddirPlus
includes entries for the current directory.
This trivial patch avoids doing a cache_enter_time()
call for the current directory to avoid the panic.
(cherry picked from commit 3fe2c68ba2)
When using NFS-over-TLS, an NFS client can optionally provide an X.509
certificate to the server during the TLS handshake. For some situations,
such as different NFS servers or different certificates being mapped
to different user credentials on the NFS server, there may be a need
for different mounts to provide different certificates.
This new mount option called "tlscertname" may be used to specify a
non-default certificate be provided. This alernate certificate will
be stored in /etc/rpc.tlsclntd in a file with a name based on what is
provided by this mount option.
successful RPC.
Without this patch, the NFSv4.2 VOP_COPY_FILE_RANGE() client call would
loop until the copy "len" was completed. The problem with doing this is
that it might take a considerable time to complete for a large "len".
By returning after a single successful Copy RPC that copied some of the
data, the application that did the copy_file_range(2) syscall will be
more responsive to signal delivery for large "len" copies.
The KERN_TLS only supports TCP, so use of the "tls" option with "udp" will
not work. This patch adds a test for this case, so that the mount is not
attempted when both "tls" and "udp" are specified.
An Internet Draft titled "Towards Remote Procedure Call Encryption By Default"
(soon to be an RFC I think) describes how Sun RPC is to use TLS with NFS
as a specific application case.
Various commits prepared the NFS code to use KERN_TLS, mainly enabling use
of ext_pgs mbufs for large RPC messages.
r364475 added TLS support to the kernel RPC.
This commit (which is the final one for kernel changes required to do
NFS over TLS) adds support for three export flags:
MNT_EXTLS - Requires a TLS connection.
MNT_EXTLSCERT - Requires a TLS connection where the client presents a valid
X.509 certificate during TLS handshake.
MNT_EXTLSCERTUSER - Requires a TLS connection where the client presents a
valid X.509 certificate with "user@domain" in the otherName
field of the SubjectAltName during TLS handshake.
Without these export options, clients are permitted, but not required, to
use TLS.
For the client, a new nmount(2) option called "tls" makes the client do
a STARTTLS Null RPC and TLS handshake for all TCP connections used for the
mount. The CLSET_TLS client control option is used to indicate to the kernel RPC
that this should be done.
Unless the above export flags or "tls" option is used, semantics should
not change for the NFS client nor server.
For NFS over TLS to work, the userspace daemons rpctlscd(8) { for client }
or rpctlssd(8) daemon { for server } must be running.
This is a partial revert of r363210, since the "use_ext" argument added
by that commit is not actually useful.
This patch should not result in any semantics change.
r363001 added support for ext_pgs mbufs to nfsm_uiombuf().
By inspection, I noticed that "mlen" was not set non-zero and, as such, there
would be an iteration of the loop that did nothing.
This patch sets it.
This bug would have no effect on the system, since the ext_pgs mbuf code
is not yet enabled.
For NFSv4.0, the server creates a server->client TCP connection for callbacks.
If the client mount on the server is using TLS, enable TLS for this callback
TCP connection.
TLS connections from clients will not be supported until the kernel RPC
changes are committed.
Since this changes the internal ABI between the NFS kernel modules that
will require a version bump, delete newnfs_trimtrailing(), which is no
longer used.
Since LCL_TLSCB is not yet set, these changes should not have any semantic
affect at this time.
This patch uses a slightly different algorithm for nfsm_uiombuflist() for
the non-ext_pgs case, where a variable called "mcp" is maintained, pointing to
the current location that mbuf data can be filled into. This avoids use of
mtod(mp, char *) + mp->m_len to calculate the location, since this does
not work for ext_pgs mbufs and I think it makes the algorithm more readable.
This change should not result in semantic changes for the non-ext_pgs case.
The patch also deletes come unneeded code.
It also adds support for anonymous page ext_pgs mbufs to nfsm_split().
This is another in the series of commits that add support to the NFS client
and server for building RPC messages in ext_pgs mbufs with anonymous pages.
This is useful so that the entire mbuf list does not need to be
copied before calling sosend() when NFS over TLS is enabled.
At this time for this case, use of ext_pgs mbufs cannot be enabled, since
ktls_encrypt() replaces the unencrypted data with encrypted data in place.
Until such time as this can be enabled, there should be no semantic change.
Also, note that this code is only used by the NFS client for a mirrored pNFS
server.
This patch modifies writing to mirrored pNFS DSs slightly so that there is
only one m_copym() call for a mirrored pair instead of two of them.
This call replaces the custom nfsm_copym() call, which is no longer needed
and deleted by this patch. The patch does introduce a new nfsm_split()
function that only calls m_split() for the non-ext_pgs case.
The semantics of nfsm_uiombuflist() is changed to include code that nul
pads the generated mbuf list. This was done by nfsm_copym() prior to this patch.
The main reason for this change is that it allows the data to be a list
of ext_pgs mbufs, since the m_copym() is for the entire mbuf list.
This support will be added in a future commit.
This patch only affects writing to mirrored flexible file layout pNFS servers.
Remove all variations of rtrequest <rtrequest1_fib, rtrequest_fib,
in6_rtrequest, rtrequest_fib> and their uses and switch to
to rib_action(). This is part of the new routing KPI.
Submitted by: Neel Chauhan <neel AT neelc DOT org>
Differential Revision: https://reviews.freebsd.org/D25546
Remove all variations of rtrequest <rtrequest1_fib, rtrequest_fib,
in6_rtrequest, rtrequest_fib> and their uses and switch to
to rib_action(). This is part of the new routing KPI.
Submitted by: Neel Chauhan <neel AT neelc DOT org>
Differential Revision: https://reviews.freebsd.org/D25546
The code in nfscl_dofflayout() loops when a flexible file layout server
provides a small write data limit (no extant server is known to do this).
If/when it looped, it erroneously reused the "drpc" argument for the
mirror worker thread, corrupting it.
This patch fixes the problem by only using the calling thread after the
first loop iteration.
Found during testing by simulating a server with a small write size.
Since no extant pNFS server is known to provide a small write size,
this fix it not needed in practice at this time.
MFC after: 2 weeks
