When a redirect failure happens we release the buffers in-flight
without calling a sk_mem_uncharge(), the uncharge is called before
dropping the sock lock for the redirecte, however we missed updating
the ring start index. When no apply actions are in progress this
is OK because we uncharge the entire buffer before the redirect.
But, when we have apply logic running its possible that only a
portion of the buffer is being redirected. In this case we only
do memory accounting for the buffer slice being redirected and
expect to be able to loop over the BPF program again and/or if
a sock is closed uncharge the memory at sock destruct time.
With an invalid start index however the program logic looks at
the start pointer index, checks the length, and when seeing the
length is zero (from the initial release and failure to update
the pointer) aborts without uncharging/releasing the remaining
memory.
The fix for this is simply to update the start index. To avoid
fixing this error in two locations we do a small refactor and
remove one case where it is open-coded. Then fix it in the
single function.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When an error occurs during a redirect we have two cases that need
to be handled (i) we have a cork'ed buffer (ii) we have a normal
sendmsg buffer.
In the cork'ed buffer case we don't currently support recovering from
errors in a redirect action. So the buffer is released and the error
should _not_ be pushed back to the caller of sendmsg/sendpage. The
rationale here is the user will get an error that relates to old
data that may have been sent by some arbitrary thread on that sock.
Instead we simple consume the data and tell the user that the data
has been consumed. We may add proper error recovery in the future.
However, this patch fixes a bug where the bytes outstanding counter
sg_size was not zeroed. This could result in a case where if the user
has both a cork'ed action and apply action in progress we may
incorrectly call into the BPF program when the user expected an
old verdict to be applied via the apply action. I don't have a use
case where using apply and cork at the same time is valid but we
never explicitly reject it because it should work fine. This patch
ensures the sg_size is zeroed so we don't have this case.
In the normal sendmsg buffer case (no cork data) we also do not
zero sg_size. Again this can confuse the apply logic when the logic
calls into the BPF program when the BPF programmer expected the old
verdict to remain. So ensure we set sg_size to zero here as well. And
additionally to keep the psock state in-sync with the sk_msg_buff
release all the memory as well. Previously we did this before
returning to the user but this left a gap where psock and sk_msg_buff
states were out of sync which seems fragile. No additional overhead
is taken here except for a call to check the length and realize its
already been freed. This is in the error path as well so in my
opinion lets have robust code over optimized error paths.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When the call to do_tcp_sendpage() fails to send the complete block
requested we either retry if only a partial send was completed or
abort if we receive a error less than or equal to zero. Before
returning though we must update the scatterlist length/offset to
account for any partial send completed.
Before this patch we did this at the end of the retry loop, but
this was buggy when used while applying a verdict to fewer bytes
than in the scatterlist. When the scatterlist length was being set
we forgot to account for the apply logic reducing the size variable.
So the result was we chopped off some bytes in the scatterlist without
doing proper cleanup on them. This results in a WARNING when the
sock is tore down because the bytes have previously been charged to
the socket but are never uncharged.
The simple fix is to simply do the accounting inside the retry loop
subtracting from the absolute scatterlist values rather than trying
to accumulate the totals and subtract at the end.
Reported-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In the case where the socket memory boundary is hit the redirect
path returns an ENOMEM error. However, before checking for this
condition the redirect scatterlist buffer is setup with a valid
page and length. This is never unwound so when the buffers are
released latter in the error path we do a put_page() and clear
the scatterlist fields. But, because the initial error happens
before completing the scatterlist buffer we end up with both the
original buffer and the redirect buffer pointing to the same page
resulting in duplicate put_page() calls.
To fix this simply move the initial configuration of the redirect
scatterlist buffer below the sock memory check.
Found this while running TCP_STREAM test with netperf using Cilium.
Fixes: fa246693a1 ("bpf: sockmap, BPF_F_INGRESS flag for BPF_SK_SKB_STREAM_VERDICT")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
In the recvmsg handler we need to add a wait event to support the
blocking use cases. Without this we return zero and may confuse
user applications. In the wait event any data received on the
sk either via sk_receive_queue or the psock ingress list will
wake up the sock.
Fixes: fa246693a1 ("bpf: sockmap, BPF_F_INGRESS flag for BPF_SK_SKB_STREAM_VERDICT")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Relying on map_release hook to decrement the reference counts when a
map is removed only works if the map is not being pinned. In the
pinned case the ref is decremented immediately and the BPF programs
released. After this BPF programs may not be in-use which is not
what the user would expect.
This patch moves the release logic into bpf_map_put_uref() and brings
sockmap in-line with how a similar case is handled in prog array maps.
Fixes: 3d9e952697 ("bpf: sockmap, fix leaking maps with attached but not detached progs")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Remove dead code that bails on `attr->value_size > KMALLOC_MAX_SIZE` - the
previous check already bails on `attr->value_size != 4`.
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
It is possible to have multiple ULP tcp_release call paths in flight
if a sock is closed and simultaneously being removed from the sockmap
control path. The result would be setting the sk_prot to the saved
values on the first iteration and then on the second iteration setting
the value to NULL.
This patch resolves this by ensuring we only reset the sk_prot pointer
if we have a valid saved state to set.
Fixes: 4f738adba3 ("bpf: create tcp_bpf_ulp allowing BPF to monitor socket TX/RX data")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
If a socket with pending cork data is closed we do not return the
memory to the socket until the garbage collector free's the psock
structure. The garbage collector though can run after the sock has
completed its close operation. If this ordering happens the sock code
will through a WARN_ON because there is still outstanding memory
accounted to the sock.
To resolve this ensure we return memory to the sock when a socket
is closed.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Fixes: 91843d540a ("bpf: sockmap, add msg_cork_bytes() helper")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
When CONFIG_DEBUG_SG is set, sg->sg_magic is initialized in
sg_init_table() and it is verified in sg api while navigating. We hit
BUG_ON when magic check is failed.
In functions sg_tcp_sendpage and sg_tcp_sendmsg, the struct containing
the scatterlist is already zeroed out. So to avoid extra memset, we
use sg_init_marker() to initialize sg_magic.
Fixed following things:
- In bpf_tcp_sendpage: initialize sg using sg_init_marker
- In bpf_tcp_sendmsg: Replace sg_init_table with sg_init_marker
- In bpf_tcp_push: Replace memset with sg_init_table where consumed
sg entry needs to be re-initialized.
Signed-off-by: Prashant Bhole <bhole_prashant_q7@lab.ntt.co.jp>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Add support for the BPF_F_INGRESS flag in skb redirect helper. To
do this convert skb into a scatterlist and push into ingress queue.
This is the same logic that is used in the sk_msg redirect helper
so it should feel familiar.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Add support for the BPF_F_INGRESS flag in sk_msg redirect helper.
To do this add a scatterlist ring for receiving socks to check
before calling into regular recvmsg call path. Additionally, because
the poll wakeup logic only checked the skb recv queue we need to
add a hook in TCP stack (similar to write side) so that we have
a way to wake up polling socks when a scatterlist is redirected
to that sock.
After this all that is needed is for the redirect helper to
push the scatterlist into the psock receive queue.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This implements a BPF ULP layer to allow policy enforcement and
monitoring at the socket layer. In order to support this a new
program type BPF_PROG_TYPE_SK_MSG is used to run the policy at
the sendmsg/sendpage hook. To attach the policy to sockets a
sockmap is used with a new program attach type BPF_SK_MSG_VERDICT.
Similar to previous sockmap usages when a sock is added to a
sockmap, via a map update, if the map contains a BPF_SK_MSG_VERDICT
program type attached then the BPF ULP layer is created on the
socket and the attached BPF_PROG_TYPE_SK_MSG program is run for
every msg in sendmsg case and page/offset in sendpage case.
BPF_PROG_TYPE_SK_MSG Semantics/API:
BPF_PROG_TYPE_SK_MSG supports only two return codes SK_PASS and
SK_DROP. Returning SK_DROP free's the copied data in the sendmsg
case and in the sendpage case leaves the data untouched. Both cases
return -EACESS to the user. Returning SK_PASS will allow the msg to
be sent.
In the sendmsg case data is copied into kernel space buffers before
running the BPF program. The kernel space buffers are stored in a
scatterlist object where each element is a kernel memory buffer.
Some effort is made to coalesce data from the sendmsg call here.
For example a sendmsg call with many one byte iov entries will
likely be pushed into a single entry. The BPF program is run with
data pointers (start/end) pointing to the first sg element.
In the sendpage case data is not copied. We opt not to copy the
data by default here, because the BPF infrastructure does not
know what bytes will be needed nor when they will be needed. So
copying all bytes may be wasteful. Because of this the initial
start/end data pointers are (0,0). Meaning no data can be read or
written. This avoids reading data that may be modified by the
user. A new helper is added later in this series if reading and
writing the data is needed. The helper call will do a copy by
default so that the page is exclusively owned by the BPF call.
The verdict from the BPF_PROG_TYPE_SK_MSG applies to the entire msg
in the sendmsg() case and the entire page/offset in the sendpage case.
This avoids ambiguity on how to handle mixed return codes in the
sendmsg case. Again a helper is added later in the series if
a verdict needs to apply to multiple system calls and/or only
a subpart of the currently being processed message.
The helper msg_redirect_map() can be used to select the socket to
send the data on. This is used similar to existing redirect use
cases. This allows policy to redirect msgs.
Pseudo code simple example:
The basic logic to attach a program to a socket is as follows,
// load the programs
bpf_prog_load(SOCKMAP_TCP_MSG_PROG, BPF_PROG_TYPE_SK_MSG,
&obj, &msg_prog);
// lookup the sockmap
bpf_map_msg = bpf_object__find_map_by_name(obj, "my_sock_map");
// get fd for sockmap
map_fd_msg = bpf_map__fd(bpf_map_msg);
// attach program to sockmap
bpf_prog_attach(msg_prog, map_fd_msg, BPF_SK_MSG_VERDICT, 0);
Adding sockets to the map is done in the normal way,
// Add a socket 'fd' to sockmap at location 'i'
bpf_map_update_elem(map_fd_msg, &i, fd, BPF_ANY);
After the above any socket attached to "my_sock_map", in this case
'fd', will run the BPF msg verdict program (msg_prog) on every
sendmsg and sendpage system call.
For a complete example see BPF selftests or sockmap samples.
Implementation notes:
It seemed the simplest, to me at least, to use a refcnt to ensure
psock is not lost across the sendmsg copy into the sg, the bpf program
running on the data in sg_data, and the final pass to the TCP stack.
Some performance testing may show a better method to do this and avoid
the refcnt cost, but for now use the simpler method.
Another item that will come after basic support is in place is
supporting MSG_MORE flag. At the moment we call sendpages even if
the MSG_MORE flag is set. An enhancement would be to collect the
pages into a larger scatterlist and pass down the stack. Notice that
bpf_tcp_sendmsg() could support this with some additional state saved
across sendmsg calls. I built the code to support this without having
to do refactoring work. Other features TBD include ZEROCOPY and the
TCP_RECV_QUEUE/TCP_NO_QUEUE support. This will follow initial series
shortly.
Future work could improve size limits on the scatterlist rings used
here. Currently, we use MAX_SKB_FRAGS simply because this was being
used already in the TLS case. Future work could extend the kernel sk
APIs to tune this depending on workload. This is a trade-off
between memory usage and throughput performance.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
The sockmap refcnt up until now has been wrapped in the
sk_callback_lock(). So its not actually needed any locking of its
own. The counter itself tracks the lifetime of the psock object.
Sockets in a sockmap have a lifetime that is independent of the
map they are part of. This is possible because a single socket may
be in multiple maps. When this happens we can only release the
psock data associated with the socket when the refcnt reaches
zero. There are three possible delete sock reference decrement
paths first through the normal sockmap process, the user deletes
the socket from the map. Second the map is removed and all sockets
in the map are removed, delete path is similar to case 1. The third
case is an asyncronous socket event such as a closing the socket. The
last case handles removing sockets that are no longer available.
For completeness, although inc does not pose any problems in this
patch series, the inc case only happens when a psock is added to a
map.
Next we plan to add another socket prog type to handle policy and
monitoring on the TX path. When we do this however we will need to
keep a reference count open across the sendmsg/sendpage call and
holding the sk_callback_lock() here (on every send) seems less than
ideal, also it may sleep in cases where we hit memory pressure.
Instead of dealing with these issues in some clever way simply make
the reference counting a refcnt_t type and do proper atomic ops.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
In case user program provides silly parameters, we want
a map_alloc() handler to return an error, not a NULL pointer,
otherwise we crash later in find_and_alloc_map()
Fixes: 1aa12bdf1b ("bpf: sockmap, add sock close() hook to remove socks")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When a program is attached to a map we increment the program refcnt
to ensure that the program is not removed while it is potentially
being referenced from sockmap side. However, if this same program
also references the map (this is a reasonably common pattern in
my programs) then the verifier will also increment the maps refcnt
from the verifier. This is to ensure the map doesn't get garbage
collected while the program has a reference to it.
So we are left in a state where the map holds the refcnt on the
program stopping it from being removed and releasing the map refcnt.
And vice versa the program holds a refcnt on the map stopping it
from releasing the refcnt on the prog.
All this is fine as long as users detach the program while the
map fd is still around. But, if the user omits this detach command
we are left with a dangling map we can no longer release.
To resolve this when the map fd is released decrement the program
references and remove any reference from the map to the program.
This fixes the issue with possibly dangling map and creates a
user side API constraint. That is, the map fd must be held open
for programs to be attached to a map.
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
The selftests test_maps program was leaving dangling BPF sockmap
programs around because not all psock elements were removed from
the map. The elements in turn hold a reference on the BPF program
they are attached to causing BPF programs to stay open even after
test_maps has completed.
The original intent was that sk_state_change() would be called
when TCP socks went through TCP_CLOSE state. However, because
socks may be in SOCK_DEAD state or the sock may be a listening
socket the event is not always triggered.
To resolve this use the ULP infrastructure and register our own
proto close() handler. This fixes the above case.
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Reported-by: Prashant Bhole <bhole_prashant_q7@lab.ntt.co.jp>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
All map types reimplement the field-by-field copy of union bpf_attr
members into struct bpf_map. Add a helper to perform this operation.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
BPF alignment tests got a conflict because the registers
are output as Rn_w instead of just Rn in net-next, and
in net a fixup for a testcase prohibits logical operations
on pointers before using them.
Also, we should attempt to patch BPF call args if JIT always on is
enabled. Instead, if we fail to JIT the subprogs we should pass
an error back up and fail immediately.
Signed-off-by: David S. Miller <davem@davemloft.net>
Add psock NULL check to handle a racing sock event that can get the
sk_callback_lock before this case but after xchg happens causing the
refcnt to hit zero and sock user data (psock) to be null and queued
for garbage collection.
Also add a comment in the code because this is a bit subtle and
not obvious in my opinion.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This was added for some work that was eventually factored out but the
helper call was missed. Remove it now and add it back later if needed.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Smooth Cong Wang's bug fix into 'net-next'. Basically put
the bulk of the tcf_block_put() logic from 'net' into
tcf_block_put_ext(), but after the offload unbind.
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that SK_REDIRECT is no longer a valid return code. Remove it
from the UAPI completely. Then do a namespace remapping internal
to sockmap so SK_REDIRECT is no longer externally visible.
Patchs primary change is to do a namechange from SK_REDIRECT to
__SK_REDIRECT
Reported-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Several conflicts here.
NFP driver bug fix adding nfp_netdev_is_nfp_repr() check to
nfp_fl_output() needed some adjustments because the code block is in
an else block now.
Parallel additions to net/pkt_cls.h and net/sch_generic.h
A bug fix in __tcp_retransmit_skb() conflicted with some of
the rbtree changes in net-next.
The tc action RCU callback fixes in 'net' had some overlap with some
of the recent tcf_block reworking.
Signed-off-by: David S. Miller <davem@davemloft.net>
Recent additions to support multiple programs in cgroups impose
a strict requirement, "all yes is yes, any no is no". To enforce
this the infrastructure requires the 'no' return code, SK_DROP in
this case, to be 0.
To apply these rules to SK_SKB program types the sk_actions return
codes need to be adjusted.
This fix adds SK_PASS and makes 'SK_DROP = 0'. Finally, remove
SK_ABORTED to remove any chance that the API may allow aborted
program flows to be passed up the stack. This would be incorrect
behavior and allow programs to break existing policies.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
SK_SKB program types use bpf_compute_data to store the end of the
packet data. However, bpf_compute_data assumes the cb is stored in the
qdisc layer format. But, for SK_SKB this is the wrong layer of the
stack for this type.
It happens to work (sort of!) because in most cases nothing happens
to be overwritten today. This is very fragile and error prone.
Fortunately, we have another hole in tcp_skb_cb we can use so lets
put the data_end value there.
Note, SK_SKB program types do not use data_meta, they are failed by
sk_skb_is_valid_access().
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
There were quite a few overlapping sets of changes here.
Daniel's bug fix for off-by-ones in the new BPF branch instructions,
along with the added allowances for "data_end > ptr + x" forms
collided with the metadata additions.
Along with those three changes came veritifer test cases, which in
their final form I tried to group together properly. If I had just
trimmed GIT's conflict tags as-is, this would have split up the
meta tests unnecessarily.
In the socketmap code, a set of preemption disabling changes
overlapped with the rename of bpf_compute_data_end() to
bpf_compute_data_pointers().
Changes were made to the mv88e6060.c driver set addr method
which got removed in net-next.
The hyperv transport socket layer had a locking change in 'net'
which overlapped with a change of socket state macro usage
in 'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce the map read/write flags to the eBPF syscalls that returns the
map fd. The flags is used to set up the file mode when construct a new
file descriptor for bpf maps. To not break the backward capability, the
f_flags is set to O_RDWR if the flag passed by syscall is 0. Otherwise
it should be O_RDONLY or O_WRONLY. When the userspace want to modify or
read the map content, it will check the file mode to see if it is
allowed to make the change.
Signed-off-by: Chenbo Feng <fengc@google.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Restrict sockmap to CAP_NET_ADMIN.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
SK_SKB BPF programs are run from the socket/tcp context but early in
the stack before much of the TCP metadata is needed in tcp_skb_cb. So
we can use some unused fields to place BPF metadata needed for SK_SKB
programs when implementing the redirect function.
This allows us to drop the preempt disable logic. It does however
require an API change so sk_redirect_map() has been updated to
additionally provide ctx_ptr to skb. Note, we do however continue to
disable/enable preemption around actual BPF program running to account
for map updates.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Only TCP sockets have been tested and at the moment the state change
callback only handles TCP sockets. This adds a check to ensure that
sockets actually being added are TCP sockets.
For net-next we can consider UDP support.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Just do the rename into bpf_compute_data_pointers() as we'll add
one more pointer here to recompute.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The bpf map sockmap supports adding programs via attach commands. This
patch adds the detach command to keep the API symmetric and allow
users to remove previously added programs. Otherwise the user would
have to delete the map and re-add it to get in this state.
This also adds a series of additional tests to capture detach operation
and also attaching/detaching invalid prog types.
API note: socks will run (or not run) programs depending on the state
of the map at the time the sock is added. We do not for example walk
the map and remove programs from previously attached socks.
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of tracking wmem_queued and sk_mem_charge by incrementing
in the verdict SK_REDIRECT paths and decrementing in the tx work
path use skb_set_owner_w and sock_writeable helpers. This solves
a few issues with the current code. First, in SK_REDIRECT inc on
sk_wmem_queued and sk_mem_charge were being done without the peers
sock lock being held. Under stress this can result in accounting
errors when tx work and/or multiple verdict decisions are working
on the peer psock.
Additionally, this cleans up the code because we can rely on the
default destructor to decrement memory accounting on kfree_skb. Also
this will trigger sk_write_space when space becomes available on
kfree_skb() which wasn't happening before and prevent __sk_free
from being called until all in-flight packets are completed.
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
"err" is set to zero if bpf_map_area_alloc() fails so it means we return
ERR_PTR(0) which is NULL. The caller, find_and_alloc_map(), is not
expecting NULL returns and will oops.
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After userspace pushes sockets into a sockmap it may not be receiving
data (assuming stream_{parser|verdict} programs are attached). But, it
may still want to manage the socks. A common pattern is to poll/select
for a POLLRDHUP event so we can close the sock.
This patch adds the logic to wake up these listeners.
Also add TCP_SYN_SENT to the list of events to handle. We don't want
to break the connection just because we happen to be in this state.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When attaching a program to sockmap we need to check map type
is correct.
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
References to psock must be done inside RCU critical section.
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The addition of map_flags BPF_SOCKMAP_STRPARSER flags was to handle a
specific use case where we want to have BPF parse program disabled on
an entry in a sockmap.
However, Alexei found the API a bit cumbersome and I agreed. Lets
remove the STRPARSER flag and support the use case by allowing socks
to be in multiple maps. This allows users to create two maps one with
programs attached and one without. When socks are added to maps they
now inherit any programs attached to the map. This is a nice
generalization and IMO improves the API.
The API rules are less ambiguous and do not need a flag:
- When a sock is added to a sockmap we have two cases,
i. The sock map does not have any attached programs so
we can add sock to map without inheriting bpf programs.
The sock may exist in 0 or more other maps.
ii. The sock map has an attached BPF program. To avoid duplicate
bpf programs we only add the sock entry if it does not have
an existing strparser/verdict attached, returning -EBUSY if
a program is already attached. Otherwise attach the program
and inherit strparser/verdict programs from the sock map.
This allows for socks to be in a multiple maps for redirects and
inherit a BPF program from a single map.
Also this patch simplifies the logic around BPF_{EXIST|NOEXIST|ANY}
flags. In the original patch I tried to be extra clever and only
update map entries when necessary. Now I've decided the complexity
is not worth it. If users constantly update an entry with the same
sock for no reason (i.e. update an entry without actually changing
any parameters on map or sock) we still do an alloc/release. Using
this and allowing multiple entries of a sock to exist in a map the
logic becomes much simpler.
Note: Now that multiple maps are supported the "maps" pointer called
when a socket is closed becomes a list of maps to remove the sock from.
To keep the map up to date when a sock is added to the sockmap we must
add the map/elem in the list. Likewise when it is removed we must
remove it from the list. This results in searching the per psock list
on delete operation. On TCP_CLOSE events we walk the list and remove
the psock from all map/entry locations. I don't see any perf
implications in this because at most I have a psock in two maps. If
a psock were to be in many maps its possibly this might be noticeable
on delete but I can't think of a reason to dup a psock in many maps.
The sk_callback_lock is used to protect read/writes to the list. This
was convenient because in all locations we were taking the lock
anyways just after working on the list. Also the lock is per sock so
in normal cases we shouldn't see any contention.
Suggested-by: Alexei Starovoitov <ast@kernel.org>
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the initial sockmap API we provided strparser and verdict programs
using a single attach command by extending the attach API with a the
attach_bpf_fd2 field.
However, if we add other programs in the future we will be adding a
field for every new possible type, attach_bpf_fd(3,4,..). This
seems a bit clumsy for an API. So lets push the programs using two
new type fields.
BPF_SK_SKB_STREAM_PARSER
BPF_SK_SKB_STREAM_VERDICT
This has the advantage of having a readable name and can easily be
extended in the future.
Updates to samples and sockmap included here also generalize tests
slightly to support upcoming patch for multiple map support.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Suggested-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current map creation API does not allow to provide the numa-node
preference. The memory usually comes from where the map-creation-process
is running. The performance is not ideal if the bpf_prog is known to
always run in a numa node different from the map-creation-process.
One of the use case is sharding on CPU to different LRU maps (i.e.
an array of LRU maps). Here is the test result of map_perf_test on
the INNER_LRU_HASH_PREALLOC test if we force the lru map used by
CPU0 to be allocated from a remote numa node:
[ The machine has 20 cores. CPU0-9 at node 0. CPU10-19 at node 1 ]
># taskset -c 10 ./map_perf_test 512 8 1260000 8000000
5:inner_lru_hash_map_perf pre-alloc 1628380 events per sec
4:inner_lru_hash_map_perf pre-alloc 1626396 events per sec
3:inner_lru_hash_map_perf pre-alloc 1626144 events per sec
6:inner_lru_hash_map_perf pre-alloc 1621657 events per sec
2:inner_lru_hash_map_perf pre-alloc 1621534 events per sec
1:inner_lru_hash_map_perf pre-alloc 1620292 events per sec
7:inner_lru_hash_map_perf pre-alloc 1613305 events per sec
0:inner_lru_hash_map_perf pre-alloc 1239150 events per sec #<<<
After specifying numa node:
># taskset -c 10 ./map_perf_test 512 8 1260000 8000000
5:inner_lru_hash_map_perf pre-alloc 1629627 events per sec
3:inner_lru_hash_map_perf pre-alloc 1628057 events per sec
1:inner_lru_hash_map_perf pre-alloc 1623054 events per sec
6:inner_lru_hash_map_perf pre-alloc 1616033 events per sec
2:inner_lru_hash_map_perf pre-alloc 1614630 events per sec
4:inner_lru_hash_map_perf pre-alloc 1612651 events per sec
7:inner_lru_hash_map_perf pre-alloc 1609337 events per sec
0:inner_lru_hash_map_perf pre-alloc 1619340 events per sec #<<<
This patch adds one field, numa_node, to the bpf_attr. Since numa node 0
is a valid node, a new flag BPF_F_NUMA_NODE is also added. The numa_node
field is honored if and only if the BPF_F_NUMA_NODE flag is set.
Numa node selection is not supported for percpu map.
This patch does not change all the kmalloc. F.e.
'htab = kzalloc()' is not changed since the object
is small enough to stay in the cache.
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In smap_do_verdict(), the fall-through branch leads to call
preempt_enable() twice for the SK_REDIRECT, which creates an
imbalance. Only enable it for all remaining cases again.
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Reported-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
psock will uninitialized in default case we need to do the same psock lookup
and check as in other branch. Fixes compile warning below.
kernel/bpf/sockmap.c: In function ‘smap_state_change’:
kernel/bpf/sockmap.c:156:21: warning: ‘psock’ may be used uninitialized in this function [-Wmaybe-uninitialized]
struct smap_psock *psock;
Fixes: 174a79ff95 ("bpf: sockmap with sk redirect support")
Reported-by: David Miller <davem@davemloft.net>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Recently we added a new map type called dev map used to forward XDP
packets between ports (6093ec2dc3). This patches introduces a
similar notion for sockets.
A sockmap allows users to add participating sockets to a map. When
sockets are added to the map enough context is stored with the
map entry to use the entry with a new helper
bpf_sk_redirect_map(map, key, flags)
This helper (analogous to bpf_redirect_map in XDP) is given the map
and an entry in the map. When called from a sockmap program, discussed
below, the skb will be sent on the socket using skb_send_sock().
With the above we need a bpf program to call the helper from that will
then implement the send logic. The initial site implemented in this
series is the recv_sock hook. For this to work we implemented a map
attach command to add attributes to a map. In sockmap we add two
programs a parse program and a verdict program. The parse program
uses strparser to build messages and pass them to the verdict program.
The parse programs use the normal strparser semantics. The verdict
program is of type SK_SKB.
The verdict program returns a verdict SK_DROP, or SK_REDIRECT for
now. Additional actions may be added later. When SK_REDIRECT is
returned, expected when bpf program uses bpf_sk_redirect_map(), the
sockmap logic will consult per cpu variables set by the helper routine
and pull the sock entry out of the sock map. This pattern follows the
existing redirect logic in cls and xdp programs.
This gives the flow,
recv_sock -> str_parser (parse_prog) -> verdict_prog -> skb_send_sock
\
-> kfree_skb
As an example use case a message based load balancer may use specific
logic in the verdict program to select the sock to send on.
Sample programs are provided in future patches that hopefully illustrate
the user interfaces. Also selftests are in follow-on patches.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
