In commit 6be8ac2f ("[NET]: uninline skb_pull, de-bloats a lot")
we uninlined skb_pull.
But in some critical paths it makes sense to inline this thing
and it helps performance significantly.
Create an skb_pull_inline() so that we can do this in a way that
serves also as annotation.
Based upon a patch by Eric Dumazet.
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no need to export skb_under_panic() and skb_over_panic() in
skbuff.c, since these methods are used only in skbuff.c ; this patch
removes these two exports. It also marks these functions as 'static'
and removeS the extern declarations of them from
include/linux/skbuff.h
Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements software receive side packet steering (RPS). RPS
distributes the load of received packet processing across multiple CPUs.
Problem statement: Protocol processing done in the NAPI context for received
packets is serialized per device queue and becomes a bottleneck under high
packet load. This substantially limits pps that can be achieved on a single
queue NIC and provides no scaling with multiple cores.
This solution queues packets early on in the receive path on the backlog queues
of other CPUs. This allows protocol processing (e.g. IP and TCP) to be
performed on packets in parallel. For each device (or each receive queue in
a multi-queue device) a mask of CPUs is set to indicate the CPUs that can
process packets. A CPU is selected on a per packet basis by hashing contents
of the packet header (e.g. the TCP or UDP 4-tuple) and using the result to index
into the CPU mask. The IPI mechanism is used to raise networking receive
softirqs between CPUs. This effectively emulates in software what a multi-queue
NIC can provide, but is generic requiring no device support.
Many devices now provide a hash over the 4-tuple on a per packet basis
(e.g. the Toeplitz hash). This patch allow drivers to set the HW reported hash
in an skb field, and that value in turn is used to index into the RPS maps.
Using the HW generated hash can avoid cache misses on the packet when
steering it to a remote CPU.
The CPU mask is set on a per device and per queue basis in the sysfs variable
/sys/class/net/<device>/queues/rx-<n>/rps_cpus. This is a set of canonical
bit maps for receive queues in the device (numbered by <n>). If a device
does not support multi-queue, a single variable is used for the device (rx-0).
Generally, we have found this technique increases pps capabilities of a single
queue device with good CPU utilization. Optimal settings for the CPU mask
seem to depend on architectures and cache hierarcy. Below are some results
running 500 instances of netperf TCP_RR test with 1 byte req. and resp.
Results show cumulative transaction rate and system CPU utilization.
e1000e on 8 core Intel
Without RPS: 108K tps at 33% CPU
With RPS: 311K tps at 64% CPU
forcedeth on 16 core AMD
Without RPS: 156K tps at 15% CPU
With RPS: 404K tps at 49% CPU
bnx2x on 16 core AMD
Without RPS 567K tps at 61% CPU (4 HW RX queues)
Without RPS 738K tps at 96% CPU (8 HW RX queues)
With RPS: 854K tps at 76% CPU (4 HW RX queues)
Caveats:
- The benefits of this patch are dependent on architecture and cache hierarchy.
Tuning the masks to get best performance is probably necessary.
- This patch adds overhead in the path for processing a single packet. In
a lightly loaded server this overhead may eliminate the advantages of
increased parallelism, and possibly cause some relative performance degradation.
We have found that masks that are cache aware (share same caches with
the interrupting CPU) mitigate much of this.
- The RPS masks can be changed dynamically, however whenever the mask is changed
this introduces the possibility of generating out of order packets. It's
probably best not change the masks too frequently.
Signed-off-by: Tom Herbert <therbert@google.com>
include/linux/netdevice.h | 32 ++++-
include/linux/skbuff.h | 3 +
net/core/dev.c | 335 +++++++++++++++++++++++++++++++++++++--------
net/core/net-sysfs.c | 225 ++++++++++++++++++++++++++++++-
net/core/skbuff.c | 2 +
5 files changed, 538 insertions(+), 59 deletions(-)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we've merged skb's with page frags, and subsequently receive
a trailer skb (< MSS) that is not completely non-linear (this can
occur on Intel NICs if the packet size falls below the threshold),
GRO ends up producing an illegal GSO skb with a frag_list.
This is harmless unless the skb is then forwarded through an
interface that requires software GSO, whereupon the GSO code
will BUG.
This patch detects this case in GRO and avoids merging the
trailer skb.
Reported-by: Mark Wagner <mwagner@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
NAPI drivers try to recycle SKBs in their polling routine, but we
generally don't know the context in which the polling will be called,
and the skb recycling itself may require IRQs to be enabled.
This patch adds irqs_disabled() test to the skb_recycle_check()
routine, so that we'll not let the drivers hit the skb recycling
path with IRQs disabled.
As a side effect, this patch actually disables skb recycling for some
[broken] drivers. E.g. gianfar driver grabs an irqsave spinlock during
TX ring processing, and then tries to recycle an skb, and that caused
the following badness:
nf_conntrack version 0.5.0 (1008 buckets, 4032 max)
------------[ cut here ]------------
Badness at kernel/softirq.c:143
NIP: c003e3c4 LR: c423a528 CTR: c003e344
...
NIP [c003e3c4] local_bh_enable+0x80/0xc4
LR [c423a528] destroy_conntrack+0xd4/0x13c [nf_conntrack]
Call Trace:
[c15d1b60] [c003e32c] local_bh_disable+0x1c/0x34 (unreliable)
[c15d1b70] [c423a528] destroy_conntrack+0xd4/0x13c [nf_conntrack]
[c15d1b80] [c02c6370] nf_conntrack_destroy+0x3c/0x70
Signed-off-by: David S. Miller <davem@davemloft.net>
mac80211 required this due to the master netdev, but now
it can put all information into skb->cb and this can go.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The skb mac_header field is sometimes NULL (or ~0u) as a sentinel
value. The places where skb is expanded add an offset which would
change this flag into an invalid pointer (or offset).
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Looking at the crash in log_martians(), one suspect is that the check for
mac header being set is not correct. The value of mac_header defaults to
0 on allocation, therefore skb_mac_header_was_set will always be true on
platforms using NET_SKBUFF_USES_OFFSET.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to handle powersave frames properly we had needed
to pass these out to the device queues again, and introduce
the skb->requeue bit. This, however, also has unnecessary
overhead by needing to 'clean up' already tried frames, and
this clean-up code is also buggy when software encryption
is used.
Instead of sending the frames via the master netdev queue
again, simply put them into the pending queue. This also
fixes a problem where frames for that particular station
could be reordered when some were still on the software
queues and older ones are re-injected into the software
queue after them.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
On Thu, Jun 04, 2009 at 09:06:00PM +1000, Herbert Xu wrote:
>
> tun: Optimise handling of bogus gso->hdr_len
>
> As all current versions of virtio_net generate a value for the
> header length that's too small, we should optimise this so that
> we don't copy it twice. This can be done by ensuring that it is
> at least as large as the place where we'll write the checksum.
>
> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
With this applied we can strengthen the partial checksum check:
In skb_partial_csum_set we check to see if the checksum offset
is within the packet. However, we really should check that it
is within the skb head as that's the only bit we can modify
without copying.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define three accessors to get/set dst attached to a skb
struct dst_entry *skb_dst(const struct sk_buff *skb)
void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
void skb_dst_drop(struct sk_buff *skb)
This one should replace occurrences of :
dst_release(skb->dst)
skb->dst = NULL;
Delete skb->dst field
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch stores the two shinfo pointers in local variables
because they're used over and over again in skb_gro_receive.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch reverses the direction of the frags array copy in
skb_gro_receive in order simplify the loop conditional. It
also avoids touching the first element of the original frags
array.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch stores the offset/headlen in local variables as they're
used repeatedly in skb_gro_offset.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
gcc does a poor job at generating code for the memcpy of the frags
array in skb_gro_receive, which is the primary purpose of that
function when merging frags. In particular, it can't utilise the
alignment information of the source and destination. This patch
open-codes the copy so we process words instead of bytes.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Hi:
skbuff: Copy csum instead of csum_start/csum_offset
It's easier to copy the u32 csum instead of its two u16
constituents.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Cheers,
Signed-off-by: David S. Miller <davem@davemloft.net>
Hi:
skbuff: Move new __skb_clone code into __copy_skb_header
It seems that people just keep on adding stuff to __skb_clone
instead __copy_skb_header. This is wrong as it means your brand-new
attributes won't always get copied as you intended.
This patch moves them to the right place, and adds a comment to
prevent this from happening again.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Thanks,
Signed-off-by: David S. Miller <davem@davemloft.net>
When called with a consumed value that is less than skb_headlen(skb)
bytes into a page frag, skb_seq_read() incorrectly returns an
offset/length relative to skb->data. Ensure that data which should come
from a page frag does.
Signed-off-by: Thomas Chenault <thomas_chenault@dell.com>
Tested-by: Shyam Iyer <shyam_iyer@dell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit ac45f602ee ("net: infrastructure
for hardware time stamping") added two skb initialization actions to
__alloc_skb(), which need to be added to skb_recycle_check() as well.
Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Lennert Buytenhek wrote:
> Since 4fb6699481 ("net: Optimize memory
> usage when splicing from sockets.") I'm seeing this oops (e.g. in
> 2.6.30-rc3) when splicing from a TCP socket to /dev/null on a driver
> (mv643xx_eth) that uses LRO in the skb mode (lro_receive_skb) rather
> than the frag mode:
My patch incorrectly assumed skb->sk was always valid, but for
"frag_listed" skbs we can only use skb->sk of their parent.
Reported-by: Lennert Buytenhek <buytenh@wantstofly.org>
Debugged-by: Lennert Buytenhek <buytenh@wantstofly.org>
Tested-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Impact: clean up
Create a sub directory in include/trace called events to keep the
trace point headers in their own separate directory. Only headers that
declare trace points should be defined in this directory.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When GRO/frag_list support was added to GSO, I made an error
which broke the support for segmenting linear GSO packets (GSO
packets are normally non-linear in the payload).
These days most of these packets are constructed by the tun
driver, which prefers to allocate linear memory if possible.
This is fixed in the latest kernel, but for 2.6.29 and earlier
it is still the norm.
Therefore this bug causes failures with GSO when used with tun
in 2.6.29.
Reported-by: James Huang <jamesclhuang@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove some pointless conditionals before kfree_skb().
Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A long time ago we had bugs, primarily in TCP, where we would modify
skb->truesize (for TSO queue collapsing) in ways which would corrupt
the socket memory accounting.
skb_truesize_check() was added in order to try and catch this error
more systematically.
However this debugging check has morphed into a Frankenstein of sorts
and these days it does nothing other than catch false-positives.
Signed-off-by: David S. Miller <davem@davemloft.net>
The additional per-packet information (16 bytes for time stamps, 1
byte for flags) is stored for all packets in the skb_shared_info
struct. This implementation detail is hidden from users of that
information via skb_* accessor functions. A separate struct resp.
union is used for the additional information so that it can be
stored/copied easily outside of skb_shared_info.
Compared to previous implementations (reusing the tstamp field
depending on the context, optional additional structures) this
is the simplest solution. It does not extend sk_buff itself.
TX time stamping is implemented in software if the device driver
doesn't support hardware time stamping.
The new semantic for hardware/software time stamping around
ndo_start_xmit() is based on two assumptions about existing
network device drivers which don't support hardware time
stamping and know nothing about it:
- they leave the new skb_shared_tx unmodified
- the keep the connection to the originating socket in skb->sk
alive, i.e., don't call skb_orphan()
Given that skb_shared_tx is new, the first assumption is safe.
The second is only true for some drivers. As a result, software
TX time stamping currently works with the bnx2 driver, but not
with the unmodified igb driver (the two drivers this patch series
was tested with).
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
struct page walking should be done with proper accessor functions, not
directly.
With doubts from David S. Miller and Herbert Xu.
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/skbuff.c is a hodge-podge of symbol export placement.
Some of the exports are right after the definition of the
symbol being exported, others are clumped together into a big
group at the end of the file.
Make things consistent.
Signed-off-by: David S. Miller <davem@davemloft.net>
The previous fix ad0f990444 (gro:
Fix handling of imprecisely split packets) only fixed the case
of frags merging, frag_list merging in the same circumstances
were still broken.
In particular, the packet headers end up in the data stream.
This patch fixes this plus another issue where an imprecisely
split packet header may be read incorrectly (this is mostly
harmless since it'll simply cause the packet to not match and
be rejected for GRO).
Thanks to Emil Tantilov and Jeff Kirsher for helping to track
this down.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The recent fix of data corruption when splicing from sockets uses
memory very inefficiently allocating a new page to copy each chunk of
linear part of skb. This patch uses the same page until it's full
(almost) by caching the page in sk_sndmsg_page field.
With changes from David S. Miller <davem@davemloft.net>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
gro: Do not merge paged packets into frag_list
Bigger is not always better :)
It was easy to continue to merged packets into frag_list after the
page array is full. However, this turns out to be worse than LRO
because frag_list is a much less efficient form of storage than the
page array. So we're better off stopping the merge and starting
a new entry with an empty page array.
In future we can optimise this further by doing frag_list merging
but making sure that we continue to fill in the page array.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Unfortunately simplicity isn't always the best. The fraginfo
interface turned out to be suboptimal. The problem was quite
obvious. For every packet, we have to copy the headers from
the frags structure into skb->head, even though for 99% of the
packets this part is immediately thrown away after the merge.
LRO didn't have this problem because it directly read the headers
from the frags structure.
This patch attempts to address this by creating an interface
that allows GRO to access the headers in the first frag without
having to copy it. Because all drivers that use frags place the
headers in the first frag this optimisation should be enough.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
It oopsd for me in skb_seq_read. addr2line said it was
linux-2.6/net/core/skbuff.c:2228, which is this line:
while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
I added some printks in there and it looks like we hit this:
} else if (st->root_skb == st->cur_skb &&
skb_shinfo(st->root_skb)->frag_list) {
st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
st->frag_idx = 0;
goto next_skb;
}
Actually I did some testing and added a few printks and found that the
st->cur_skb->data was 0 and hence the ptr used by iscsi_tcp was null.
This caused the kernel panic.
if (abs_offset < block_limit) {
- *data = st->cur_skb->data + abs_offset;
+ *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
I enabled the debug_tcp and with a few printks found that the code did
not go to the next_skb label and could find that the sequence being
followed was this -
It hit this if condition -
if (st->cur_skb->next) {
st->cur_skb = st->cur_skb->next;
st->frag_idx = 0;
goto next_skb;
And so, now the st pointer is shifted to the next skb whereas actually
it should have hit the second else if first since the data is in the
frag_list.
else if (st->root_skb == st->cur_skb &&
skb_shinfo(st->root_skb)->frag_list) {
st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
goto next_skb;
}
Reversing the two conditions the attached patch fixes the issue for me
on top of Herbert's patches.
Signed-off-by: Shyam Iyer <shyam_iyer@dell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The frag_list handling was broken in skb_seq_read:
1) We didn't add the stepped offset when looking at the head
are of fragments other than the first.
2) We didn't take the stepped offset away when setting the data
pointer in the head area.
3) The frag index wasn't reset.
This patch fixes both issues.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The previous fix to paged packets broke the merging because it
reset the skb->len before we added it to the merged packet. This
wasn't detected because it simply resulted in the truncation of
the packet while the missing bit is subsequently retransmitted.
The fix is to store skb->len before we clobber it.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The trick in socket splicing where we try to convert the skb->data
into a page based reference using virt_to_page() does not work so
well.
The idea is to pass the virt_to_page() reference via the pipe
buffer, and refcount the buffer using a SKB reference.
But if we are splicing from a socket to a socket (via sendpage)
this doesn't work.
The from side processing will grab the page (and SKB) references.
The sendpage() calls will grab page references only, return, and
then the from side processing completes and drops the SKB ref.
The page based reference to skb->data is not enough to keep the
kmalloc() buffer backing it from being reused. Yet, that is
all that the socket send side has at this point.
This leads to data corruption if the skb->data buffer is reused
by SLAB before the send side socket actually gets the TX packet
out to the device.
The fix employed here is to simply allocate a page and copy the
skb->data bytes into that page.
This will hurt performance, but there is no clear way to fix this
properly without a copy at the present time, and it is important
to get rid of the data corruption.
With fixes from Herbert Xu.
Tested-by: Willy Tarreau <w@1wt.eu>
Foreseen-by: Changli Gao <xiaosuo@gmail.com>
Diagnosed-by: Willy Tarreau <w@1wt.eu>
Reported-by: Willy Tarreau <w@1wt.eu>
Fixed-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When an skb with page frags is merged into an existing one, we
cannibalise its reference count. This is OK when the skb is
reused because we set nr_frags to zero in that case. However,
for the case where the skb is freed through kfree_skb, we didn't
clear nr_frags which causes the page to be freed prematurely.
This is fixed by moving the skb resetting into skb_gro_receive.
Reported-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch allows GRO to merge page frags (skb_shinfo(skb)->frags)
in one skb, rather than using the less efficient frag_list.
It also adds a new interface, napi_gro_frags to allow drivers
to inject page frags directly into the stack without allocating
an skb. This is intended to be the GRO equivalent for LRO's
lro_receive_frags interface.
The existing GSO interface can already handle page frags with
or without an appended frag_list so nothing needs to be changed
there.
The merging itself is rather simple. We store any new frag entries
after the last existing entry, without checking whether the first
new entry can be merged with the last existing entry. Making this
check would actually be easy but since no existing driver can
produce contiguous frags anyway it would just be mental masturbation.
If the total number of entries would exceed the capacity of a
single skb, we simply resort to using frag_list as we do now.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to allow GRO packets without frag_list at all, we need to
store the MSS in the packet itself. The obvious place is gso_size.
The only thing to watch out for is if the packet ends up not being
GRO then we need to clear gso_size before pushing the packet into
the stack.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the helper skb_gro_receive to merge packets for
GRO. The current method is to allocate a new header skb and then
chain the original packets to its frag_list. This is done to
make it easier to integrate into the existing GSO framework.
In future as GSO is moved into the drivers, we can undo this and
simply chain the original packets together.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds limited support for handling frag_list packets in
skb_segment. The intention is to support GRO (Generic Receive Offload)
packets which will be constructed by chaining normal packets using
frag_list.
As such we require all frag_list members terminate on exact MSS
boundaries. This is checked using BUG_ON.
As there should only be one producer in the kernel of such packets,
namely GRO, this requirement should not be difficult to maintain.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The truesize message check is important enough to make it print "BUG"
to the user console... lets also make it important enough to spit a
backtrace/module list etc so that kerneloops.org can track them.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since pskb_expand_head creates copy of the shared area we
cannot keep any frag ptr past de-cloning. This fixes the
tcpdump recvfrom -EFAULT problem.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
This caused me to get repeatably:
tcpdump: pcap_loop: recvfrom: Bad address
Happens occassionally when I tcpdump my for-looped test xfers:
while [ : ]; do echo -n "$(date '+%s.%N') "; ./sendfile; sleep 20; done
Rest of the relevant commands:
ethtool -K eth0 tso off
tc qdisc add dev eth0 root netem drop 4%
tcpdump -n -s0 -i eth0 -w sacklog.all
Running net-next under kvm, connection goes to the same host
(basically just out of kvm). The connection itself works ok
and data gets sent without corruption even with a large
number of tests while tcpdump fails usually within less than
5 tests.
Whether it only happens because of this change or not, I
don't know for sure but it's the only thing with which
I've seen that error. The non-cloned variant works w/o it
for much longer time. I'm yet to debug where the error
actually comes from.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since skb_reset_tail_pointer() reads skb->data, we need to set
skb->data before calling skb_reset_tail_pointer(). This was causing
spurious skb_over_panic()s from skb_put() being called on a recycled
skb that had its skb->tail set to beyond where it should have been.
Bug report from Peter van Valderen <linux@ddcrew.com>.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller