We've now increased the size of the duplicate reply cache by quite a
bit, but the number of hash buckets has not changed. So, we've gone from
an average hash chain length of 16 in the old code to 4096 when the
cache is its largest. Change the code to scale out the number of buckets
with the max size of the cache.
At the same time, we also need to fix the hash function since the
existing one isn't really suitable when there are more than 256 buckets.
Move instead to use the stock hash_32 function for this. Testing on a
machine that had 2048 buckets showed that this gave a smaller
longest:average ratio than the existing hash function:
The formula here is longest hash bucket searched divided by average
number of entries per bucket at the time that we saw that longest
bucket:
old hash: 68/(39258/2048) == 3.547404
hash_32: 45/(33773/2048) == 2.728807
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
The typical case with the DRC is a cache miss, so if we keep track of
the max number of entries that we've ever walked over in a search, then
we should have a reasonable estimate of the longest hash chain that
we've ever seen.
With that, we'll also keep track of the total size of the cache when we
see the longest chain. In the case of a tie, we prefer to track the
smallest total cache size in order to properly gauge the worst-case
ratio of max vs. avg chain length.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
For presenting statistics relating to duplicate reply cache.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Break out the function that compares the rqstp and checksum against a
reply cache entry. While we're at it, track the efficacy of the checksum
over the NFS data by tracking the cases where we would have incorrectly
matched a DRC entry if we had not tracked it or the length.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
The most common case is to do a search of the cache, followed by an
insert. In the case where we have to allocate an entry off the slab,
then we end up having to redo the search, which is wasteful.
Better optimize the code for the common case by eliminating the initial
search of the cache and always preallocating an entry. In the case of a
cache hit, we'll end up just freeing that entry but that's preferable to
an extra search.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
If we end up doing "goto out_nomem" in this function, we'll call
nfsd_reply_cache_shutdown. That will attempt to walk the LRU list and
free entries, but that list may not be initialized yet if the server is
starting up for the first time. It's also possible for the shrinker to
kick in before we've initialized the LRU list.
Rearrange the initialization so that the LRU list_head and cache size
are initialized before doing any of the allocations that might fail.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
It's not safe to call hlist_del() on a newly initialized hlist_node.
That leads to a NULL pointer dereference. Only do that if the entry
is hashed.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Pull nfsd changes from J Bruce Fields:
"Miscellaneous bugfixes, plus:
- An overhaul of the DRC cache by Jeff Layton. The main effect is
just to make it larger. This decreases the chances of intermittent
errors especially in the UDP case. But we'll need to watch for any
reports of performance regressions.
- Containerized nfsd: with some limitations, we now support
per-container nfs-service, thanks to extensive work from Stanislav
Kinsbursky over the last year."
Some notes about conflicts, since there were *two* non-data semantic
conflicts here:
- idr_remove_all() had been added by a memory leak fix, but has since
become deprecated since idr_destroy() does it for us now.
- xs_local_connect() had been added by this branch to make AF_LOCAL
connections be synchronous, but in the meantime Trond had changed the
calling convention in order to avoid a RCU dereference.
There were a couple of more obvious actual source-level conflicts due to
the hlist traversal changes and one just due to code changes next to
each other, but those were trivial.
* 'for-3.9' of git://linux-nfs.org/~bfields/linux: (49 commits)
SUNRPC: make AF_LOCAL connect synchronous
nfsd: fix compiler warning about ambiguous types in nfsd_cache_csum
svcrpc: fix rpc server shutdown races
svcrpc: make svc_age_temp_xprts enqueue under sv_lock
lockd: nlmclnt_reclaim(): avoid stack overflow
nfsd: enable NFSv4 state in containers
nfsd: disable usermode helper client tracker in container
nfsd: use proper net while reading "exports" file
nfsd: containerize NFSd filesystem
nfsd: fix comments on nfsd_cache_lookup
SUNRPC: move cache_detail->cache_request callback call to cache_read()
SUNRPC: remove "cache_request" argument in sunrpc_cache_pipe_upcall() function
SUNRPC: rework cache upcall logic
SUNRPC: introduce cache_detail->cache_request callback
NFS: simplify and clean cache library
NFS: use SUNRPC cache creation and destruction helper for DNS cache
nfsd4: free_stid can be static
nfsd: keep a checksum of the first 256 bytes of request
sunrpc: trim off trailing checksum before returning decrypted or integrity authenticated buffer
sunrpc: fix comment in struct xdr_buf definition
...
I'm not sure why, but the hlist for each entry iterators were conceived
list_for_each_entry(pos, head, member)
The hlist ones were greedy and wanted an extra parameter:
hlist_for_each_entry(tpos, pos, head, member)
Why did they need an extra pos parameter? I'm not quite sure. Not only
they don't really need it, it also prevents the iterator from looking
exactly like the list iterator, which is unfortunate.
Besides the semantic patch, there was some manual work required:
- Fix up the actual hlist iterators in linux/list.h
- Fix up the declaration of other iterators based on the hlist ones.
- A very small amount of places were using the 'node' parameter, this
was modified to use 'obj->member' instead.
- Coccinelle didn't handle the hlist_for_each_entry_safe iterator
properly, so those had to be fixed up manually.
The semantic patch which is mostly the work of Peter Senna Tschudin is here:
@@
iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host;
type T;
expression a,c,d,e;
identifier b;
statement S;
@@
-T b;
<+... when != b
(
hlist_for_each_entry(a,
- b,
c, d) S
|
hlist_for_each_entry_continue(a,
- b,
c) S
|
hlist_for_each_entry_from(a,
- b,
c) S
|
hlist_for_each_entry_rcu(a,
- b,
c, d) S
|
hlist_for_each_entry_rcu_bh(a,
- b,
c, d) S
|
hlist_for_each_entry_continue_rcu_bh(a,
- b,
c) S
|
for_each_busy_worker(a, c,
- b,
d) S
|
ax25_uid_for_each(a,
- b,
c) S
|
ax25_for_each(a,
- b,
c) S
|
inet_bind_bucket_for_each(a,
- b,
c) S
|
sctp_for_each_hentry(a,
- b,
c) S
|
sk_for_each(a,
- b,
c) S
|
sk_for_each_rcu(a,
- b,
c) S
|
sk_for_each_from
-(a, b)
+(a)
S
+ sk_for_each_from(a) S
|
sk_for_each_safe(a,
- b,
c, d) S
|
sk_for_each_bound(a,
- b,
c) S
|
hlist_for_each_entry_safe(a,
- b,
c, d, e) S
|
hlist_for_each_entry_continue_rcu(a,
- b,
c) S
|
nr_neigh_for_each(a,
- b,
c) S
|
nr_neigh_for_each_safe(a,
- b,
c, d) S
|
nr_node_for_each(a,
- b,
c) S
|
nr_node_for_each_safe(a,
- b,
c, d) S
|
- for_each_gfn_sp(a, c, d, b) S
+ for_each_gfn_sp(a, c, d) S
|
- for_each_gfn_indirect_valid_sp(a, c, d, b) S
+ for_each_gfn_indirect_valid_sp(a, c, d) S
|
for_each_host(a,
- b,
c) S
|
for_each_host_safe(a,
- b,
c, d) S
|
for_each_mesh_entry(a,
- b,
c, d) S
)
...+>
[akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c]
[akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c]
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: fix warnings]
[akpm@linux-foudnation.org: redo intrusive kvm changes]
Tested-by: Peter Senna Tschudin <peter.senna@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Now that we're allowing more DRC entries, it becomes a lot easier to hit
problems with XID collisions. In order to mitigate those, calculate a
checksum of up to the first 256 bytes of each request coming in and store
that in the cache entry, along with the total length of the request.
This initially used crc32, but Chuck Lever and Jim Rees pointed out that
crc32 is probably more heavyweight than we really need for generating
these checksums, and recommended looking at using the same routines that
are used to generate checksums for IP packets.
On an x86_64 KVM guest measurements with ftrace showed ~800ns to use
csum_partial vs ~1750ns for crc32. The difference probably isn't
terribly significant, but for now we may as well use csum_partial.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Stones-thrown-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
These routines are used by server and client code, so having them in a
separate header would be best.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Since we dynamically allocate them now, allow the system to call us up
to release them if it gets low on memory. Since these entries aren't
replaceable, only free ones that are expired or that are over the cap.
The the seeks value is set to '1' however to indicate that freeing the
these entries is low-cost.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
It's not sufficient to only clean the cache when requests come in. What
if we have a flurry of activity and then the server goes idle? Add a
workqueue job that will clean the cache every RC_EXPIRE period.
Care is taken to only run this when we expect to have entries expiring.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
There's no need to keep entries around that we're declaring RC_NOCACHE.
Ditto if there's a problem with the entry.
With this change too, there's no need to test for RC_UNUSED in the
search function. If the entry's in the hash table then it's either
INPROG or DONE.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
With the change to dynamically allocate entries, the cache is never
disabled on the fly. Remove this flag.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
The existing code keeps a fixed-size cache of 1024 entries. This is much
too small for a busy server, and wastes memory on an idle one. This
patch changes the code to dynamically allocate and free these cache
entries.
A cap on the number of entries is retained, but it's much larger than
the existing value and now scales with the amount of low memory in the
machine.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
...otherwise, we end up with the list ordering wrong. Currently, it's
not a problem since we skip RC_INPROG entries, but keeping the ordering
strict will be necessary for a later patch that adds a cache cleaner.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Later, we'll need more than one call site for this, so break it out
into a new function.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Add a preprocessor constant for the expiry time of cache entries, and
move the test for an expired entry into a function. Note that the current
code does not test for RC_INPROG. It just assumes that it won't take more
than 2 minutes to fill out an in-progress entry.
I'm not sure how valid that assumption is though, so let's just ensure
that we never consider an RC_INPROG entry to be expired.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Entries can only get a c_type of RC_REPLBUFF iff they are
RC_DONE. Therefore the test for RC_DONE isn't necessary here.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Currently we use kmalloc() which wastes a little bit of memory on each
allocation since it's a power of 2 allocator. Since we're allocating a
1024 of these now, and may need even more later, let's create a new
slabcache for them.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
The locking rules for cache entries say that locking the cache_lock
isn't needed if you're just touching the current entry. Earlier
in this function we set rp->c_state to RC_UNUSED without any locking,
so I believe it's ok to do the same here.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Currently, it only stores the first 16 bytes of any address. struct
sockaddr_in6 is 28 bytes however, so we're currently ignoring the last
12 bytes of the address.
Expand the c_addr field to a sockaddr_in6, and cast it to a sockaddr_in
as necessary. Also fix the comparitor to use the existing RPC
helpers for this.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
It's sort of ridiculous that we've never had a working reply cache for
NFSv4.
On the other hand, we may still not: our current reply cache is likely
not very good, especially in the TCP case (which is the only case that
matters for v4). What we really need here is some serious testing.
Anyway, here's a start.
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
The new .h files have paths at the top that are now out of date. While
we're here, just remove all of those from fs/nfsd; they never served any
purpose.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Lots of include/linux/nfsd/* headers are only used by
nfsd module. Move them to the source directory
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Now that the headers are fixed and carry their own wait, all fs/nfsd/
source files can include a minimal set of headers. and still compile just
fine.
This patch should improve the compilation speed of the nfsd module.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Fix a regression in the reply cache introduced when the code was
converted to use proper Linux lists. When a new entry needs to be
inserted, the case where all the entries are currently being used
by threads is not correctly detected. This can result in memory
corruption and a crash. In the current code this is an extremely
unlikely corner case; it would require the machine to have 1024
nfsd threads and all of them to be busy at the same time. However,
upcoming reply cache changes make this more likely; a crash due to
this problem was actually observed in field.
Signed-off-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Make REQHASH() an inline function. Rename hash_list to cache_hash.
Fix an obsolete comment.
Signed-off-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
If the reply cache initialization fails due to a kmalloc failure,
currently we try to soldier on with a reduced (or nonexistant) reply
cache.
Better to just fail immediately: the failure is then much easier to
understand and debug, and it could save us complexity in some later
code. (But actually, it doesn't help currently because the cache is
also turned off in some odd failure cases; we should probably find a
better way to handle those failure cases some day.)
Fix some minor style problems while we're at it, and rename
nfsd_cache_init() to remove the need for a comment describing it.
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Expand the rq_addr field to allow it to contain larger addresses.
Specifically, we replace a 'sockaddr_in' with a 'sockaddr_storage', then
everywhere the 'sockaddr_in' was referenced, we use instead an accessor
function (svc_addr_in) which safely casts the _storage to _in.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Cc: Aurelien Charbon <aurelien.charbon@ext.bull.net>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Replace kmalloc+memset with kcalloc and simplify
Signed-off-by: Yan Burman <burman.yan@gmail.com>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Trond Myklebust <trond.myklebust@fys.uio.no>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch converts the combination of list_del(A) and list_add(A, B) to
list_move(A, B) under fs/.
Cc: Ian Kent <raven@themaw.net>
Acked-by: Joel Becker <joel.becker@oracle.com>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: Hans Reiser <reiserfs-dev@namesys.com>
Cc: Urban Widmark <urban@teststation.com>
Acked-by: David Howells <dhowells@redhat.com>
Acked-by: Mark Fasheh <mark.fasheh@oracle.com>
Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This is the fs/ part of the big kfree cleanup patch.
Remove pointless checks for NULL prior to calling kfree() in fs/.
Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
