Per-VMA locking allows us to lock a struct vm_area_struct without
taking the process-wide mmap lock in read mode.
Consider a process workload where the mmap lock is taken constantly in
write mode. In this scenario, all zerocopy receives are periodically
blocked during that period of time - though in principle, the memory
ranges being used by TCP are not touched by the operations that need
the mmap write lock. This results in performance degradation.
Now consider another workload where the mmap lock is never taken in
write mode, but there are many TCP connections using receive zerocopy
that are concurrently receiving. These connections all take the mmap
lock in read mode, but this does induce a lot of contention and atomic
ops for this process-wide lock. This results in additional CPU
overhead caused by contending on the cache line for this lock.
However, with per-vma locking, both of these problems can be avoided.
As a test, I ran an RPC-style request/response workload with 4KB
payloads and receive zerocopy enabled, with 100 simultaneous TCP
connections. I measured perf cycles within the
find_tcp_vma/mmap_read_lock/mmap_read_unlock codepath, with and
without per-vma locking enabled.
When using process-wide mmap semaphore read locking, about 1% of
measured perf cycles were within this path. With per-VMA locking, this
value dropped to about 0.45%.
Signed-off-by: Arjun Roy <arjunroy@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Under certain circumstances, the tcp receive buffer memory limit
set by autotuning (sk_rcvbuf) is increased due to incoming data
packets as a result of the window not closing when it should be.
This can result in the receive buffer growing all the way up to
tcp_rmem[2], even for tcp sessions with a low BDP.
To reproduce: Connect a TCP session with the receiver doing
nothing and the sender sending small packets (an infinite loop
of socket send() with 4 bytes of payload with a sleep of 1 ms
in between each send()). This will cause the tcp receive buffer
to grow all the way up to tcp_rmem[2].
As a result, a host can have individual tcp sessions with receive
buffers of size tcp_rmem[2], and the host itself can reach tcp_mem
limits, causing the host to go into tcp memory pressure mode.
The fundamental issue is the relationship between the granularity
of the window scaling factor and the number of byte ACKed back
to the sender. This problem has previously been identified in
RFC 7323, appendix F [1].
The Linux kernel currently adheres to never shrinking the window.
In addition to the overallocation of memory mentioned above, the
current behavior is functionally incorrect, because once tcp_rmem[2]
is reached when no remediations remain (i.e. tcp collapse fails to
free up any more memory and there are no packets to prune from the
out-of-order queue), the receiver will drop in-window packets
resulting in retransmissions and an eventual timeout of the tcp
session. A receive buffer full condition should instead result
in a zero window and an indefinite wait.
In practice, this problem is largely hidden for most flows. It
is not applicable to mice flows. Elephant flows can send data
fast enough to "overrun" the sk_rcvbuf limit (in a single ACK),
triggering a zero window.
But this problem does show up for other types of flows. Examples
are websockets and other type of flows that send small amounts of
data spaced apart slightly in time. In these cases, we directly
encounter the problem described in [1].
RFC 7323, section 2.4 [2], says there are instances when a retracted
window can be offered, and that TCP implementations MUST ensure
that they handle a shrinking window, as specified in RFC 1122,
section 4.2.2.16 [3]. All prior RFCs on the topic of tcp window
management have made clear that sender must accept a shrunk window
from the receiver, including RFC 793 [4] and RFC 1323 [5].
This patch implements the functionality to shrink the tcp window
when necessary to keep the right edge within the memory limit by
autotuning (sk_rcvbuf). This new functionality is enabled with
the new sysctl: net.ipv4.tcp_shrink_window
Additional information can be found at:
https://blog.cloudflare.com/unbounded-memory-usage-by-tcp-for-receive-buffers-and-how-we-fixed-it/
[1] https://www.rfc-editor.org/rfc/rfc7323#appendix-F
[2] https://www.rfc-editor.org/rfc/rfc7323#section-2.4
[3] https://www.rfc-editor.org/rfc/rfc1122#page-91
[4] https://www.rfc-editor.org/rfc/rfc793
[5] https://www.rfc-editor.org/rfc/rfc1323
Signed-off-by: Mike Freemon <mfreemon@cloudflare.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Splicing to SOCK_RAW sockets may set MSG_SPLICE_PAGES, but in such a case,
__ip_append_data() will call skb_splice_from_iter() to access the 'from'
data, assuming it to point to a msghdr struct with an iter, instead of
using the provided getfrag function to access it.
In the case of raw_sendmsg(), however, this is not the case and 'from' will
point to a raw_frag_vec struct and raw_getfrag() will be the frag-getting
function. A similar issue may occur with rawv6_sendmsg().
Fix this by ignoring MSG_SPLICE_PAGES if getfrag != ip_generic_getfrag as
ip_generic_getfrag() expects "from" to be a msghdr*, but the other getfrags
don't. Note that this will prevent MSG_SPLICE_PAGES from being effective
for udplite.
This likely affects ping sockets too. udplite looks like it should be okay
as it expects "from" to be a msghdr.
Signed-off-by: David Howells <dhowells@redhat.com>
Reported-by: syzbot+d8486855ef44506fd675@syzkaller.appspotmail.com
Link: https://lore.kernel.org/r/000000000000ae4cbf05fdeb8349@google.com/
Fixes: 2dc334f1a6 ("splice, net: Use sendmsg(MSG_SPLICE_PAGES) rather than ->sendpage()")
Tested-by: syzbot+d8486855ef44506fd675@syzkaller.appspotmail.com
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/1410156.1686729856@warthog.procyon.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Most of the ioctls to net protocols operates directly on userspace
argument (arg). Usually doing get_user()/put_user() directly in the
ioctl callback. This is not flexible, because it is hard to reuse these
functions without passing userspace buffers.
Change the "struct proto" ioctls to avoid touching userspace memory and
operate on kernel buffers, i.e., all protocol's ioctl callbacks is
adapted to operate on a kernel memory other than on userspace (so, no
more {put,get}_user() and friends being called in the ioctl callback).
This changes the "struct proto" ioctl format in the following way:
int (*ioctl)(struct sock *sk, int cmd,
- unsigned long arg);
+ int *karg);
(Important to say that this patch does not touch the "struct proto_ops"
protocols)
So, the "karg" argument, which is passed to the ioctl callback, is a
pointer allocated to kernel space memory (inside a function wrapper).
This buffer (karg) may contain input argument (copied from userspace in
a prep function) and it might return a value/buffer, which is copied
back to userspace if necessary. There is not one-size-fits-all format
(that is I am using 'may' above), but basically, there are three type of
ioctls:
1) Do not read from userspace, returns a result to userspace
2) Read an input parameter from userspace, and does not return anything
to userspace
3) Read an input from userspace, and return a buffer to userspace.
The default case (1) (where no input parameter is given, and an "int" is
returned to userspace) encompasses more than 90% of the cases, but there
are two other exceptions. Here is a list of exceptions:
* Protocol RAW:
* cmd = SIOCGETVIFCNT:
* input and output = struct sioc_vif_req
* cmd = SIOCGETSGCNT
* input and output = struct sioc_sg_req
* Explanation: for the SIOCGETVIFCNT case, userspace passes the input
argument, which is struct sioc_vif_req. Then the callback populates
the struct, which is copied back to userspace.
* Protocol RAW6:
* cmd = SIOCGETMIFCNT_IN6
* input and output = struct sioc_mif_req6
* cmd = SIOCGETSGCNT_IN6
* input and output = struct sioc_sg_req6
* Protocol PHONET:
* cmd == SIOCPNADDRESOURCE | SIOCPNDELRESOURCE
* input int (4 bytes)
* Nothing is copied back to userspace.
For the exception cases, functions sock_sk_ioctl_inout() will
copy the userspace input, and copy it back to kernel space.
The wrapper that prepare the buffer and put the buffer back to user is
sk_ioctl(), so, instead of calling sk->sk_prot->ioctl(), the callee now
calls sk_ioctl(), which will handle all cases.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230609152800.830401-1-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Recently syzkaller reported a 7-year-old null-ptr-deref [0] that occurs
when a UDP-Lite socket tries to allocate a buffer under memory pressure.
Someone should have stumbled on the bug much earlier if UDP-Lite had been
used in a real app. Also, we do not always need a large UDP-Lite workload
to hit the bug since UDP and UDP-Lite share the same memory accounting
limit.
Removing UDP-Lite would simplify UDP code removing a bunch of conditionals
in fast path.
Let's add a deprecation notice when UDP-Lite socket is created and schedule
its removal to 2025.
Link: https://lore.kernel.org/netdev/20230523163305.66466-1-kuniyu@amazon.com/ [0]
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Make tcp_bpf_sendpage() a wrapper around tcp_bpf_sendmsg(MSG_SPLICE_PAGES)
rather than a loop calling tcp_sendpage(). sendpage() will be removed in
the future.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: John Fastabend <john.fastabend@gmail.com>
cc: Jakub Sitnicki <jakub@cloudflare.com>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Now all tcp_stream_alloc_skb() callers pass @size == 0, we can
remove this parameter.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now all skbs in write queue do not contain any payload in skb->head,
we can remove some dead code.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tcp_send_syn_data() is the last component in TCP transmit
path to put payload in skb->head.
Switch it to use page frags, so that we can remove dead
code later.
This allows to put more payload than previous implementation.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move declarations into include/net/gso.h and code into net/core/gso.c
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230608191738.3947077-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Allow splice to undo the effects of MSG_MORE after prematurely ending a
splice/sendfile due to getting an EOF condition (->splice_read() returned
0) after splice had called sendmsg() with MSG_MORE set when the user didn't
set MSG_MORE.
For UDP, a pending packet will not be emitted if the socket is closed
before it is flushed; with this change, it be flushed by ->splice_eof().
For TCP, it's not clear that MSG_MORE is actually effective.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://lore.kernel.org/r/CAHk-=wh=V579PDYvkpnTobCLGczbgxpMgGmmhqiTyE34Cpi5Gg@mail.gmail.com/
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Kuniyuki Iwashima <kuniyu@amazon.com>
cc: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_mtu_probe() is still copying payload from skbs in the write queue,
using skb_copy_bits(), ignoring potential errors.
Modern TCP stack wants to only deal with payload found in page frags,
as this is a prereq for TCPDirect (host stack might not have access
to the payload)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230607214113.1992947-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cross-merge networking fixes after downstream PR.
Conflicts:
net/sched/sch_taprio.c
d636fc5dd6 ("net: sched: add rcu annotations around qdisc->qdisc_sleeping")
dced11ef84 ("net/sched: taprio: don't overwrite "sch" variable in taprio_dump_class_stats()")
net/ipv4/sysctl_net_ipv4.c
e209fee411 ("net/ipv4: ping_group_range: allow GID from 2147483648 to 4294967294")
ccce324dab ("tcp: make the first N SYN RTO backoffs linear")
https://lore.kernel.org/all/20230605100816.08d41a7b@canb.auug.org.au/
No adjacent changes.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix incorrectly formatted tcp_syn_linear_timeouts sysctl in the
ipv4_net_table.
Fixes: ccce324dab ("tcp: make the first N SYN RTO backoffs linear")
Signed-off-by: David Morley <morleyd@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Tested-by: David Morley <morleyd@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RT_CONN_FLAGS(sk) overloads flowi4_tos with the RTO_ONLINK bit when
sk has the SOCK_LOCALROUTE flag set. This allows
ip_route_output_key_hash() to eventually adjust flowi4_scope.
Instead of relying on special handling of the RTO_ONLINK bit, we can
just set the route scope correctly. This will eventually allow to avoid
special interpretation of tos variables and to convert ->flowi4_tos to
dscp_t.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
RT_CONN_FLAGS(sk) overloads the tos parameter with the RTO_ONLINK bit
when sk has the SOCK_LOCALROUTE flag set. This is only useful for
ip_route_output_key_hash() to eventually adjust the route scope.
Let's drop RTO_ONLINK and set the correct scope directly to avoid this
special case in the future and to allow converting ->flowi4_tos to
dscp_t.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We missed that tcp_gso_segment() was assuming skb->len was smaller than 65535 :
oldlen = (u16)~skb->len;
This part came with commit 0718bcc09b ("[NET]: Fix CHECKSUM_HW GSO problems.")
This leads to wrong TCP checksum.
Adapt the code to accept arbitrary packet length.
v2:
- use two csum_add() instead of csum_fold() (Alexander Duyck)
- Change delta type to __wsum to reduce casts (Alexander Duyck)
Fixes: 09f3d1a3a5 ("ipv6/gso: remove temporary HBH/jumbo header")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230605161647.3624428-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
With this commit, all the GIDs ("0 4294967294") can be written to the
"net.ipv4.ping_group_range" sysctl.
Note that 4294967295 (0xffffffff) is an invalid GID (see gid_valid() in
include/linux/uidgid.h), and an attempt to register this number will cause
-EINVAL.
Prior to this commit, only up to GID 2147483647 could be covered.
Documentation/networking/ip-sysctl.rst had "0 4294967295" as an example
value, but this example was wrong and causing -EINVAL.
Fixes: c319b4d76b ("net: ipv4: add IPPROTO_ICMP socket kind")
Co-developed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
In this patch, we mainly try to handle sending a compressed ack
correctly if it's deferred.
Here are more details in the old logic:
When sack compression is triggered in the tcp_compressed_ack_kick(),
if the sock is owned by user, it will set TCP_DELACK_TIMER_DEFERRED
and then defer to the release cb phrase. Later once user releases
the sock, tcp_delack_timer_handler() should send a ack as expected,
which, however, cannot happen due to lack of ICSK_ACK_TIMER flag.
Therefore, the receiver would not sent an ack until the sender's
retransmission timeout. It definitely increases unnecessary latency.
Fixes: 5d9f4262b7 ("tcp: add SACK compression")
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: fuyuanli <fuyuanli@didiglobal.com>
Signed-off-by: Jason Xing <kerneljasonxing@gmail.com>
Link: https://lore.kernel.org/netdev/20230529113804.GA20300@didi-ThinkCentre-M920t-N000/
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230531080150.GA20424@didi-ThinkCentre-M920t-N000
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
tcp_gro_complete() function only updates the skb fields related to GRO
and it always returns zero. All the 3 drivers which are using it
do not check for the return value either.
Change it to return void instead which simplifies its callers as
error handing becomes unnecessary.
Signed-off-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQTFp0I1jqZrAX+hPRXbK58LschIgwUCZHEm+wAKCRDbK58LschI
gyIKAQCqO7B4sIu8hYVxBTwfHV2tIuXSMSCV4P9e78NUOPcO2QEAvLP/WVSjB0Bm
vpyTKKM22SpZvPe/jSp52j6t20N+qAc=
=HFxD
-----END PGP SIGNATURE-----
Merge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Daniel Borkmann says:
====================
pull-request: bpf-next 2023-05-26
We've added 54 non-merge commits during the last 10 day(s) which contain
a total of 76 files changed, 2729 insertions(+), 1003 deletions(-).
The main changes are:
1) Add the capability to destroy sockets in BPF through a new kfunc,
from Aditi Ghag.
2) Support O_PATH fds in BPF_OBJ_PIN and BPF_OBJ_GET commands,
from Andrii Nakryiko.
3) Add capability for libbpf to resize datasec maps when backed via mmap,
from JP Kobryn.
4) Move all the test kfuncs for CI out of the kernel and into bpf_testmod,
from Jiri Olsa.
5) Big batch of xsk selftest improvements to prep for multi-buffer testing,
from Magnus Karlsson.
6) Show the target_{obj,btf}_id in tracing link's fdinfo and dump it
via bpftool, from Yafang Shao.
7) Various misc BPF selftest improvements to work with upcoming LLVM 17,
from Yonghong Song.
8) Extend bpftool to specify netdevice for resolving XDP hints,
from Larysa Zaremba.
9) Document masking in shift operations for the insn set document,
from Dave Thaler.
10) Extend BPF selftests to check xdp_feature support for bond driver,
from Lorenzo Bianconi.
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (54 commits)
bpf: Fix bad unlock balance on freeze_mutex
libbpf: Ensure FD >= 3 during bpf_map__reuse_fd()
libbpf: Ensure libbpf always opens files with O_CLOEXEC
selftests/bpf: Check whether to run selftest
libbpf: Change var type in datasec resize func
bpf: drop unnecessary bpf_capable() check in BPF_MAP_FREEZE command
libbpf: Selftests for resizing datasec maps
libbpf: Add capability for resizing datasec maps
selftests/bpf: Add path_fd-based BPF_OBJ_PIN and BPF_OBJ_GET tests
libbpf: Add opts-based bpf_obj_pin() API and add support for path_fd
bpf: Support O_PATH FDs in BPF_OBJ_PIN and BPF_OBJ_GET commands
libbpf: Start v1.3 development cycle
bpf: Validate BPF object in BPF_OBJ_PIN before calling LSM
bpftool: Specify XDP Hints ifname when loading program
selftests/bpf: Add xdp_feature selftest for bond device
selftests/bpf: Test bpf_sock_destroy
selftests/bpf: Add helper to get port using getsockname
bpf: Add bpf_sock_destroy kfunc
bpf: Add kfunc filter function to 'struct btf_kfunc_id_set'
bpf: udp: Implement batching for sockets iterator
...
====================
Link: https://lore.kernel.org/r/20230526222747.17775-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
To keep things simple we used to include the uAPI header
in the kernel in the #include <linux/$family.h> format.
This works well enough, most of the genl families should
have headers in include/net/ so linux/$family.h ends up
referring to the uAPI header, anyway. And if it doesn't
no big deal, we'll just include more info than we need.
Unless that is there is a naming conflict. Someone recently
created include/linux/psp.h which will be a problem when
supporting the PSP protocol. (I'm talking about
work-in-progress patches, but it's just a proof that assuming
lack of name conflicts was overly optimistic.)
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When using IPv4/TCP, skb->hash comes from sk->sk_txhash except in
TIME_WAIT and SYN_RECV where it's not set in the reply skb from
ip_send_unicast_reply. Those packets will have a mismatched hash with
others from the same flow as their hashes will be 0. IPv6 does not have
the same issue as the hash is set from the socket txhash in those cases.
This commits sets the hash in the reply skb from ip_send_unicast_reply,
which makes the IPv4 code behaving like IPv6.
Signed-off-by: Antoine Tenart <atenart@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Commit c67b85558f ("ipv6: tcp: send consistent autoflowlabel in
TIME_WAIT state") made the socket txhash also available in TIME_WAIT
sockets but for IPv6 only. Make it available for IPv4 too as we'll use
it in later commits.
Signed-off-by: Antoine Tenart <atenart@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQTFp0I1jqZrAX+hPRXbK58LschIgwUCZG4AiAAKCRDbK58LschI
g+xlAQCmefGbDuwPckZLnomvt6gl4bkIjs7kc1ySbG9QBnaInwD/WyrJaQIPijuD
qziHPAyx+MEgPseFU1b7Le35SZ66IwM=
=s4R1
-----END PGP SIGNATURE-----
Merge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
Daniel Borkmann says:
====================
pull-request: bpf 2023-05-24
We've added 19 non-merge commits during the last 10 day(s) which contain
a total of 20 files changed, 738 insertions(+), 448 deletions(-).
The main changes are:
1) Batch of BPF sockmap fixes found when running against NGINX TCP tests,
from John Fastabend.
2) Fix a memleak in the LRU{,_PERCPU} hash map when bucket locking fails,
from Anton Protopopov.
3) Init the BPF offload table earlier than just late_initcall,
from Jakub Kicinski.
4) Fix ctx access mask generation for 32-bit narrow loads of 64-bit fields,
from Will Deacon.
5) Remove a now unsupported __fallthrough in BPF samples,
from Andrii Nakryiko.
6) Fix a typo in pkg-config call for building sign-file,
from Jeremy Sowden.
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf:
bpf, sockmap: Test progs verifier error with latest clang
bpf, sockmap: Test FIONREAD returns correct bytes in rx buffer with drops
bpf, sockmap: Test FIONREAD returns correct bytes in rx buffer
bpf, sockmap: Test shutdown() correctly exits epoll and recv()=0
bpf, sockmap: Build helper to create connected socket pair
bpf, sockmap: Pull socket helpers out of listen test for general use
bpf, sockmap: Incorrectly handling copied_seq
bpf, sockmap: Wake up polling after data copy
bpf, sockmap: TCP data stall on recv before accept
bpf, sockmap: Handle fin correctly
bpf, sockmap: Improved check for empty queue
bpf, sockmap: Reschedule is now done through backlog
bpf, sockmap: Convert schedule_work into delayed_work
bpf, sockmap: Pass skb ownership through read_skb
bpf: fix a memory leak in the LRU and LRU_PERCPU hash maps
bpf: Fix mask generation for 32-bit narrow loads of 64-bit fields
samples/bpf: Drop unnecessary fallthrough
bpf: netdev: init the offload table earlier
selftests/bpf: Fix pkg-config call building sign-file
====================
Link: https://lore.kernel.org/r/20230524170839.13905-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use ip_sendmsg_scope() to properly initialise the scope in
flowi4_init_output(), instead of overriding tos with the RTO_ONLINK
flag. The objective is to eventually remove RTO_ONLINK, which will
allow converting .flowi4_tos to dscp_t.
Now that the scope is determined by ip_sendmsg_scope(), we need to
check its result to set the 'connected' variable.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use ip_sendmsg_scope() to properly initialise the scope in
flowi4_init_output(), instead of overriding tos with the RTO_ONLINK
flag. The objective is to eventually remove RTO_ONLINK, which will
allow converting .flowi4_tos to dscp_t.
The MSG_DONTROUTE and SOCK_LOCALROUTE cases were already handled by
raw_sendmsg() (SOCK_LOCALROUTE was handled by the RT_CONN_FLAGS*()
macros called by get_rtconn_flags()). However, opt.is_strictroute
wasn't taken into account. Therefore, a side effect of this patch is to
now honour opt.is_strictroute, and thus align raw_sendmsg() with
ping_v4_sendmsg() and udp_sendmsg().
Since raw_sendmsg() was the only user of get_rtconn_flags(), we can now
remove this function.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define a new helper to figure out the correct route scope to use on TX,
depending on socket configuration, ancillary data and send flags.
Use this new helper to properly initialise the scope in
flowi4_init_output(), instead of overriding tos with the RTO_ONLINK
flag.
The objective is to eventually remove RTO_ONLINK, which will allow
converting .flowi4_tos to dscp_t.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
ip_append_page() is no longer used with the removal of udp_sendpage(), so
remove it.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Convert udp_sendpage() to use sendmsg() with MSG_SPLICE_PAGES rather than
directly splicing in the pages itself.
This allows ->sendpage() to be replaced by something that can handle
multiple multipage folios in a single transaction.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Make IP/UDP sendmsg() support MSG_SPLICE_PAGES. This causes pages to be
spliced from the source iterator.
This allows ->sendpage() to be replaced by something that can handle
multiple multipage folios in a single transaction.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fold do_tcp_sendpages() into its last remaining caller,
tcp_sendpage_locked().
Signed-off-by: David Howells <dhowells@redhat.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
do_tcp_sendpages() is now just a small wrapper around tcp_sendmsg_locked(),
so inline it. This is part of replacing ->sendpage() with a call to
sendmsg() with MSG_SPLICE_PAGES set.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: John Fastabend <john.fastabend@gmail.com>
cc: Jakub Sitnicki <jakub@cloudflare.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Convert do_tcp_sendpages() to use sendmsg() with MSG_SPLICE_PAGES rather
than directly splicing in the pages itself. do_tcp_sendpages() can then be
inlined in subsequent patches into its callers.
This allows ->sendpage() to be replaced by something that can handle
multiple multipage folios in a single transaction.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Make TCP's sendmsg() support MSG_SPLICE_PAGES. This causes pages to be
spliced or copied (if it cannot be spliced) from the source iterator.
This allows ->sendpage() to be replaced by something that can handle
multiple multipage folios in a single transaction.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pass the maximum number of fragments into skb_append_pagefrags() rather
than using MAX_SKB_FRAGS so that it can be used from code that wants to
specify sysctl_max_skb_frags.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: David Ahern <dsahern@kernel.org>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The read_skb() logic is incrementing the tcp->copied_seq which is used for
among other things calculating how many outstanding bytes can be read by
the application. This results in application errors, if the application
does an ioctl(FIONREAD) we return zero because this is calculated from
the copied_seq value.
To fix this we move tcp->copied_seq accounting into the recv handler so
that we update these when the recvmsg() hook is called and data is in
fact copied into user buffers. This gives an accurate FIONREAD value
as expected and improves ACK handling. Before we were calling the
tcp_rcv_space_adjust() which would update 'number of bytes copied to
user in last RTT' which is wrong for programs returning SK_PASS. The
bytes are only copied to the user when recvmsg is handled.
Doing the fix for recvmsg is straightforward, but fixing redirect and
SK_DROP pkts is a bit tricker. Build a tcp_psock_eat() helper and then
call this from skmsg handlers. This fixes another issue where a broken
socket with a BPF program doing a resubmit could hang the receiver. This
happened because although read_skb() consumed the skb through sock_drop()
it did not update the copied_seq. Now if a single reccv socket is
redirecting to many sockets (for example for lb) the receiver sk will be
hung even though we might expect it to continue. The hang comes from
not updating the copied_seq numbers and memory pressure resulting from
that.
We have a slight layer problem of calling tcp_eat_skb even if its not
a TCP socket. To fix we could refactor and create per type receiver
handlers. I decided this is more work than we want in the fix and we
already have some small tweaks depending on caller that use the
helper skb_bpf_strparser(). So we extend that a bit and always set
the strparser bit when it is in use and then we can gate the
seq_copied updates on this.
Fixes: 04919bed94 ("tcp: Introduce tcp_read_skb()")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20230523025618.113937-9-john.fastabend@gmail.com
A common mechanism to put a TCP socket into the sockmap is to hook the
BPF_SOCK_OPS_{ACTIVE_PASSIVE}_ESTABLISHED_CB event with a BPF program
that can map the socket info to the correct BPF verdict parser. When
the user adds the socket to the map the psock is created and the new
ops are assigned to ensure the verdict program will 'see' the sk_buffs
as they arrive.
Part of this process hooks the sk_data_ready op with a BPF specific
handler to wake up the BPF verdict program when data is ready to read.
The logic is simple enough (posted here for easy reading)
static void sk_psock_verdict_data_ready(struct sock *sk)
{
struct socket *sock = sk->sk_socket;
if (unlikely(!sock || !sock->ops || !sock->ops->read_skb))
return;
sock->ops->read_skb(sk, sk_psock_verdict_recv);
}
The oversight here is sk->sk_socket is not assigned until the application
accepts() the new socket. However, its entirely ok for the peer application
to do a connect() followed immediately by sends. The socket on the receiver
is sitting on the backlog queue of the listening socket until its accepted
and the data is queued up. If the peer never accepts the socket or is slow
it will eventually hit data limits and rate limit the session. But,
important for BPF sockmap hooks when this data is received TCP stack does
the sk_data_ready() call but the read_skb() for this data is never called
because sk_socket is missing. The data sits on the sk_receive_queue.
Then once the socket is accepted if we never receive more data from the
peer there will be no further sk_data_ready calls and all the data
is still on the sk_receive_queue(). Then user calls recvmsg after accept()
and for TCP sockets in sockmap we use the tcp_bpf_recvmsg_parser() handler.
The handler checks for data in the sk_msg ingress queue expecting that
the BPF program has already run from the sk_data_ready hook and enqueued
the data as needed. So we are stuck.
To fix do an unlikely check in recvmsg handler for data on the
sk_receive_queue and if it exists wake up data_ready. We have the sock
locked in both read_skb and recvmsg so should avoid having multiple
runners.
Fixes: 04919bed94 ("tcp: Introduce tcp_read_skb()")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20230523025618.113937-7-john.fastabend@gmail.com
The sockmap code is returning EAGAIN after a FIN packet is received and no
more data is on the receive queue. Correct behavior is to return 0 to the
user and the user can then close the socket. The EAGAIN causes many apps
to retry which masks the problem. Eventually the socket is evicted from
the sockmap because its released from sockmap sock free handling. The
issue creates a delay and can cause some errors on application side.
To fix this check on sk_msg_recvmsg side if length is zero and FIN flag
is set then set return to zero. A selftest will be added to check this
condition.
Fixes: 04919bed94 ("tcp: Introduce tcp_read_skb()")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: William Findlay <will@isovalent.com>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20230523025618.113937-6-john.fastabend@gmail.com
The read_skb hook calls consume_skb() now, but this means that if the
recv_actor program wants to use the skb it needs to inc the ref cnt
so that the consume_skb() doesn't kfree the sk_buff.
This is problematic because in some error cases under memory pressure
we may need to linearize the sk_buff from sk_psock_skb_ingress_enqueue().
Then we get this,
skb_linearize()
__pskb_pull_tail()
pskb_expand_head()
BUG_ON(skb_shared(skb))
Because we incremented users refcnt from sk_psock_verdict_recv() we
hit the bug on with refcnt > 1 and trip it.
To fix lets simply pass ownership of the sk_buff through the skb_read
call. Then we can drop the consume from read_skb handlers and assume
the verdict recv does any required kfree.
Bug found while testing in our CI which runs in VMs that hit memory
constraints rather regularly. William tested TCP read_skb handlers.
[ 106.536188] ------------[ cut here ]------------
[ 106.536197] kernel BUG at net/core/skbuff.c:1693!
[ 106.536479] invalid opcode: 0000 [#1] PREEMPT SMP PTI
[ 106.536726] CPU: 3 PID: 1495 Comm: curl Not tainted 5.19.0-rc5 #1
[ 106.537023] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ArchLinux 1.16.0-1 04/01/2014
[ 106.537467] RIP: 0010:pskb_expand_head+0x269/0x330
[ 106.538585] RSP: 0018:ffffc90000138b68 EFLAGS: 00010202
[ 106.538839] RAX: 000000000000003f RBX: ffff8881048940e8 RCX: 0000000000000a20
[ 106.539186] RDX: 0000000000000002 RSI: 0000000000000000 RDI: ffff8881048940e8
[ 106.539529] RBP: ffffc90000138be8 R08: 00000000e161fd1a R09: 0000000000000000
[ 106.539877] R10: 0000000000000018 R11: 0000000000000000 R12: ffff8881048940e8
[ 106.540222] R13: 0000000000000003 R14: 0000000000000000 R15: ffff8881048940e8
[ 106.540568] FS: 00007f277dde9f00(0000) GS:ffff88813bd80000(0000) knlGS:0000000000000000
[ 106.540954] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 106.541227] CR2: 00007f277eeede64 CR3: 000000000ad3e000 CR4: 00000000000006e0
[ 106.541569] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 106.541915] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 106.542255] Call Trace:
[ 106.542383] <IRQ>
[ 106.542487] __pskb_pull_tail+0x4b/0x3e0
[ 106.542681] skb_ensure_writable+0x85/0xa0
[ 106.542882] sk_skb_pull_data+0x18/0x20
[ 106.543084] bpf_prog_b517a65a242018b0_bpf_skskb_http_verdict+0x3a9/0x4aa9
[ 106.543536] ? migrate_disable+0x66/0x80
[ 106.543871] sk_psock_verdict_recv+0xe2/0x310
[ 106.544258] ? sk_psock_write_space+0x1f0/0x1f0
[ 106.544561] tcp_read_skb+0x7b/0x120
[ 106.544740] tcp_data_queue+0x904/0xee0
[ 106.544931] tcp_rcv_established+0x212/0x7c0
[ 106.545142] tcp_v4_do_rcv+0x174/0x2a0
[ 106.545326] tcp_v4_rcv+0xe70/0xf60
[ 106.545500] ip_protocol_deliver_rcu+0x48/0x290
[ 106.545744] ip_local_deliver_finish+0xa7/0x150
Fixes: 04919bed94 ("tcp: Introduce tcp_read_skb()")
Reported-by: William Findlay <will@isovalent.com>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: William Findlay <will@isovalent.com>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20230523025618.113937-2-john.fastabend@gmail.com
With a raw socket bound to IPPROTO_RAW (ie with hdrincl enabled), the
protocol field of the flow structure, build by raw_sendmsg() /
rawv6_sendmsg()), is set to IPPROTO_RAW. This breaks the ipsec policy
lookup when some policies are defined with a protocol in the selector.
For ipv6, the sin6_port field from 'struct sockaddr_in6' could be used to
specify the protocol. Just accept all values for IPPROTO_RAW socket.
For ipv4, the sin_port field of 'struct sockaddr_in' could not be used
without breaking backward compatibility (the value of this field was never
checked). Let's add a new kind of control message, so that the userland
could specify which protocol is used.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
CC: stable@vger.kernel.org
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Link: https://lore.kernel.org/r/20230522120820.1319391-1-nicolas.dichtel@6wind.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The new variable is only used when IPCONFIG_BOOTP is defined and otherwise
causes a warning:
net/ipv4/ipconfig.c:177:12: error: 'ic_nameservers_fallback' defined but not used [-Werror=unused-variable]
Move it next to the user.
Fixes: 81ac2722fa ("net: ipconfig: Allow DNS to be overwritten by DHCPACK")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The socket destroy kfunc is used to forcefully terminate sockets from
certain BPF contexts. We plan to use the capability in Cilium
load-balancing to terminate client sockets that continue to connect to
deleted backends. The other use case is on-the-fly policy enforcement
where existing socket connections prevented by policies need to be
forcefully terminated. The kfunc also allows terminating sockets that may
or may not be actively sending traffic.
The kfunc can currently be called only from BPF TCP and UDP iterators
where users can filter, and terminate selected sockets. More
specifically, it can only be called from BPF contexts that ensure
socket locking in order to allow synchronous execution of protocol
specific `diag_destroy` handlers. The previous commit that batches UDP
sockets during iteration facilitated a synchronous invocation of the UDP
destroy callback from BPF context by skipping socket locks in
`udp_abort`. TCP iterator already supported batching of sockets being
iterated. To that end, `tracing_iter_filter` callback filter is added so
that verifier can restrict the kfunc to programs with `BPF_TRACE_ITER`
attach type, and reject other programs.
The kfunc takes `sock_common` type argument, even though it expects, and
casts them to a `sock` pointer. This enables the verifier to allow the
sock_destroy kfunc to be called for TCP with `sock_common` and UDP with
`sock` structs. Furthermore, as `sock_common` only has a subset of
certain fields of `sock`, casting pointer to the latter type might not
always be safe for certain sockets like request sockets, but these have a
special handling in the diag_destroy handlers.
Additionally, the kfunc is defined with `KF_TRUSTED_ARGS` flag to avoid the
cases where a `PTR_TO_BTF_ID` sk is obtained by following another pointer.
eg. getting a sk pointer (may be even NULL) by following another sk
pointer. The pointer socket argument passed in TCP and UDP iterators is
tagged as `PTR_TRUSTED` in {tcp,udp}_reg_info. The TRUSTED arg changes
are contributed by Martin KaFai Lau <martin.lau@kernel.org>.
Signed-off-by: Aditi Ghag <aditi.ghag@isovalent.com>
Link: https://lore.kernel.org/r/20230519225157.760788-8-aditi.ghag@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
