Kernel/linux
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git synced 2024-12-29 17:23:36 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
master
linux/net/tls/tls_strp.c

634 lines
14 KiB
C
Raw Permalink Normal View History

tls: rx: async: hold onto the input skb Async crypto currently benefits from the fact that we decrypt in place. When we allow input and output to be different skbs we will have to hang onto the input while we move to the next record. Clone the inputs and keep them on a list. Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-07-15 05:22:33 +00:00
// SPDX-License-Identifier: GPL-2.0-only
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
/* Copyright (c) 2016 Tom Herbert <tom@herbertland.com> */
tls: rx: async: hold onto the input skb Async crypto currently benefits from the fact that we decrypt in place. When we allow input and output to be different skbs we will have to hang onto the input while we move to the next record. Clone the inputs and keep them on a list. Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-07-15 05:22:33 +00:00
#include <linux/skbuff.h>
net: move skb ref helpers to new header Add a new header, linux/skbuff_ref.h, which contains all the skb_*_ref() helpers. Many of the consumers of skbuff.h do not actually use any of the skb ref helpers, and we can speed up compilation a bit by minimizing this header file. Additionally in the later patch in the series we add page_pool support to skb_frag_ref(), which requires some page_pool dependencies. We can now add these dependencies to skbuff_ref.h instead of a very ubiquitous skbuff.h Signed-off-by: Mina Almasry <almasrymina@google.com> Link: https://lore.kernel.org/r/20240410190505.1225848-2-almasrymina@google.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2024-04-10 19:05:01 +00:00
#include <linux/skbuff_ref.h>
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
#include <linux/workqueue.h>
#include <net/strparser.h>
#include <net/tcp.h>
#include <net/sock.h>
#include <net/tls.h>
tls: rx: async: hold onto the input skb Async crypto currently benefits from the fact that we decrypt in place. When we allow input and output to be different skbs we will have to hang onto the input while we move to the next record. Clone the inputs and keep them on a list. Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-07-15 05:22:33 +00:00
#include "tls.h"
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
static struct workqueue_struct *tls_strp_wq;
static void tls_strp_abort_strp(struct tls_strparser *strp, int err)
{
if (strp->stopped)
return;
strp->stopped = 1;
/* Report an error on the lower socket */
tls: improve lockless access safety of tls_err_abort() Most protos' poll() methods insert a memory barrier between writes to sk_err and sk_error_report(). This dates back to commit a4d258036ed9 ("tcp: Fix race in tcp_poll"). I guess we should do the same thing in TLS, tcp_poll() does not hold the socket lock. Fixes: 3c4d7559159b ("tls: kernel TLS support") 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>
2023-05-25 05:17:41 +00:00
WRITE_ONCE(strp->sk->sk_err, -err);
/* Paired with smp_rmb() in tcp_poll() */
smp_wmb();
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
sk_error_report(strp->sk);
}
static void tls_strp_anchor_free(struct tls_strparser *strp)
tls: rx: device: keep the zero copy status with offload The non-zero-copy path assumes a full skb with decrypted contents. This means the device offload would have to CoW the data. Try to keep the zero-copy status instead, copy the data to user space. Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:30 +00:00
{
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
struct skb_shared_info *shinfo = skb_shinfo(strp->anchor);
DEBUG_NET_WARN_ON_ONCE(atomic_read(&shinfo->dataref) != 1);
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
if (!strp->copy_mode)
shinfo->frag_list = NULL;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
consume_skb(strp->anchor);
strp->anchor = NULL;
}
tls: rx: strp: factor out copying skb data We'll need to copy input skbs individually in the next patch. Factor that code out (without assuming we're copying a full record). Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:40 +00:00
static struct sk_buff *
tls_strp_skb_copy(struct tls_strparser *strp, struct sk_buff *in_skb,
int offset, int len)
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
{
tls: rx: device: keep the zero copy status with offload The non-zero-copy path assumes a full skb with decrypted contents. This means the device offload would have to CoW the data. Try to keep the zero-copy status instead, copy the data to user space. Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:30 +00:00
struct sk_buff *skb;
tls: rx: strp: factor out copying skb data We'll need to copy input skbs individually in the next patch. Factor that code out (without assuming we're copying a full record). Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:40 +00:00
int i, err;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
tls: rx: strp: factor out copying skb data We'll need to copy input skbs individually in the next patch. Factor that code out (without assuming we're copying a full record). Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:40 +00:00
skb = alloc_skb_with_frags(0, len, TLS_PAGE_ORDER,
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
&err, strp->sk->sk_allocation);
if (!skb)
return NULL;
for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
tls: rx: device: keep the zero copy status with offload The non-zero-copy path assumes a full skb with decrypted contents. This means the device offload would have to CoW the data. Try to keep the zero-copy status instead, copy the data to user space. Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:30 +00:00
tls: rx: strp: factor out copying skb data We'll need to copy input skbs individually in the next patch. Factor that code out (without assuming we're copying a full record). Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:40 +00:00
WARN_ON_ONCE(skb_copy_bits(in_skb, offset,
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
skb_frag_address(frag),
skb_frag_size(frag)));
offset += skb_frag_size(frag);
}
tls: rx: strp: factor out copying skb data We'll need to copy input skbs individually in the next patch. Factor that code out (without assuming we're copying a full record). Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:40 +00:00
skb->len = len;
skb->data_len = len;
skb_copy_header(skb, in_skb);
return skb;
}
/* Create a new skb with the contents of input copied to its page frags */
static struct sk_buff *tls_strp_msg_make_copy(struct tls_strparser *strp)
{
struct strp_msg *rxm;
struct sk_buff *skb;
skb = tls_strp_skb_copy(strp, strp->anchor, strp->stm.offset,
strp->stm.full_len);
if (!skb)
return NULL;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
rxm = strp_msg(skb);
rxm->offset = 0;
tls: rx: device: keep the zero copy status with offload The non-zero-copy path assumes a full skb with decrypted contents. This means the device offload would have to CoW the data. Try to keep the zero-copy status instead, copy the data to user space. Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:30 +00:00
return skb;
}
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
/* Steal the input skb, input msg is invalid after calling this function */
struct sk_buff *tls_strp_msg_detach(struct tls_sw_context_rx *ctx)
{
struct tls_strparser *strp = &ctx->strp;
#ifdef CONFIG_TLS_DEVICE
DEBUG_NET_WARN_ON_ONCE(!strp->anchor->decrypted);
#else
/* This function turns an input into an output,
* that can only happen if we have offload.
*/
WARN_ON(1);
#endif
if (strp->copy_mode) {
struct sk_buff *skb;
/* Replace anchor with an empty skb, this is a little
* dangerous but __tls_cur_msg() warns on empty skbs
* so hopefully we'll catch abuses.
*/
skb = alloc_skb(0, strp->sk->sk_allocation);
if (!skb)
return NULL;
swap(strp->anchor, skb);
return skb;
}
return tls_strp_msg_make_copy(strp);
}
/* Force the input skb to be in copy mode. The data ownership remains
* with the input skb itself (meaning unpause will wipe it) but it can
* be modified.
*/
tls: rx: device: add input CoW helper Wrap the remaining skb_cow_data() into a helper, so it's easier to replace down the lane. The new version will change the skb so make sure relevant pointers get reloaded after the call. Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:32 +00:00
int tls_strp_msg_cow(struct tls_sw_context_rx *ctx)
{
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
struct tls_strparser *strp = &ctx->strp;
struct sk_buff *skb;
if (strp->copy_mode)
return 0;
skb = tls_strp_msg_make_copy(strp);
if (!skb)
return -ENOMEM;
tls_strp_anchor_free(strp);
strp->anchor = skb;
tcp_read_done(strp->sk, strp->stm.full_len);
strp->copy_mode = 1;
return 0;
}
/* Make a clone (in the skb sense) of the input msg to keep a reference
* to the underlying data. The reference-holding skbs get placed on
* @dst.
*/
int tls_strp_msg_hold(struct tls_strparser *strp, struct sk_buff_head *dst)
{
struct skb_shared_info *shinfo = skb_shinfo(strp->anchor);
if (strp->copy_mode) {
struct sk_buff *skb;
WARN_ON_ONCE(!shinfo->nr_frags);
/* We can't skb_clone() the anchor, it gets wiped by unpause */
skb = alloc_skb(0, strp->sk->sk_allocation);
if (!skb)
return -ENOMEM;
__skb_queue_tail(dst, strp->anchor);
strp->anchor = skb;
} else {
struct sk_buff *iter, *clone;
int chunk, len, offset;
offset = strp->stm.offset;
len = strp->stm.full_len;
iter = shinfo->frag_list;
while (len > 0) {
if (iter->len <= offset) {
offset -= iter->len;
goto next;
}
chunk = iter->len - offset;
offset = 0;
clone = skb_clone(iter, strp->sk->sk_allocation);
if (!clone)
return -ENOMEM;
__skb_queue_tail(dst, clone);
len -= chunk;
next:
iter = iter->next;
}
}
return 0;
}
static void tls_strp_flush_anchor_copy(struct tls_strparser *strp)
{
struct skb_shared_info *shinfo = skb_shinfo(strp->anchor);
int i;
DEBUG_NET_WARN_ON_ONCE(atomic_read(&shinfo->dataref) != 1);
for (i = 0; i < shinfo->nr_frags; i++)
__skb_frag_unref(&shinfo->frags[i], false);
shinfo->nr_frags = 0;
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
if (strp->copy_mode) {
kfree_skb_list(shinfo->frag_list);
shinfo->frag_list = NULL;
}
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
strp->copy_mode = 0;
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
strp->mixed_decrypted = 0;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
}
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
static int tls_strp_copyin_frag(struct tls_strparser *strp, struct sk_buff *skb,
struct sk_buff *in_skb, unsigned int offset,
size_t in_len)
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
{
tls: rx: Fix unsigned comparison with less than zero The return from the call to tls_rx_msg_size() is int, it can be a negative error code, however this is being assigned to an unsigned long variable 'sz', so making 'sz' an int. Eliminate the following coccicheck warning: ./net/tls/tls_strp.c:211:6-8: WARNING: Unsigned expression compared with zero: sz < 0 Reported-by: Abaci Robot <abaci@linux.alibaba.com> Signed-off-by: Yang Li <yang.lee@linux.alibaba.com> Link: https://lore.kernel.org/r/20220728031019.32838-1-yang.lee@linux.alibaba.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-28 03:10:19 +00:00
size_t len, chunk;
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
skb_frag_t *frag;
tls: rx: Fix unsigned comparison with less than zero The return from the call to tls_rx_msg_size() is int, it can be a negative error code, however this is being assigned to an unsigned long variable 'sz', so making 'sz' an int. Eliminate the following coccicheck warning: ./net/tls/tls_strp.c:211:6-8: WARNING: Unsigned expression compared with zero: sz < 0 Reported-by: Abaci Robot <abaci@linux.alibaba.com> Signed-off-by: Yang Li <yang.lee@linux.alibaba.com> Link: https://lore.kernel.org/r/20220728031019.32838-1-yang.lee@linux.alibaba.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-28 03:10:19 +00:00
int sz;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
frag = &skb_shinfo(skb)->frags[skb->len / PAGE_SIZE];
len = in_len;
/* First make sure we got the header */
if (!strp->stm.full_len) {
/* Assume one page is more than enough for headers */
chunk = min_t(size_t, len, PAGE_SIZE - skb_frag_size(frag));
WARN_ON_ONCE(skb_copy_bits(in_skb, offset,
skb_frag_address(frag) +
skb_frag_size(frag),
chunk));
tls: rx: strp: fix determining record length in copy mode We call tls_rx_msg_size(skb) before doing skb->len += chunk. So the tls_rx_msg_size() code will see old skb->len, most likely leading to an over-read. Worst case we will over read an entire record, next iteration will try to trim the skb but may end up turning frag len negative or discarding the subsequent record (since we already told TCP we've read it during previous read but now we'll trim it out of the skb). Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:39 +00:00
skb->len += chunk;
skb->data_len += chunk;
skb_frag_size_add(frag, chunk);
sz = tls_rx_msg_size(strp, skb);
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
if (sz < 0)
return sz;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
/* We may have over-read, sz == 0 is guaranteed under-read */
tls: rx: strp: fix determining record length in copy mode We call tls_rx_msg_size(skb) before doing skb->len += chunk. So the tls_rx_msg_size() code will see old skb->len, most likely leading to an over-read. Worst case we will over read an entire record, next iteration will try to trim the skb but may end up turning frag len negative or discarding the subsequent record (since we already told TCP we've read it during previous read but now we'll trim it out of the skb). Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:39 +00:00
if (unlikely(sz && sz < skb->len)) {
int over = skb->len - sz;
WARN_ON_ONCE(over > chunk);
skb->len -= over;
skb->data_len -= over;
skb_frag_size_add(frag, -over);
chunk -= over;
}
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
frag++;
len -= chunk;
offset += chunk;
strp->stm.full_len = sz;
if (!strp->stm.full_len)
goto read_done;
}
/* Load up more data */
while (len && strp->stm.full_len > skb->len) {
chunk = min_t(size_t, len, strp->stm.full_len - skb->len);
chunk = min_t(size_t, chunk, PAGE_SIZE - skb_frag_size(frag));
WARN_ON_ONCE(skb_copy_bits(in_skb, offset,
skb_frag_address(frag) +
skb_frag_size(frag),
chunk));
skb->len += chunk;
skb->data_len += chunk;
skb_frag_size_add(frag, chunk);
frag++;
len -= chunk;
offset += chunk;
}
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
read_done:
return in_len - len;
}
static int tls_strp_copyin_skb(struct tls_strparser *strp, struct sk_buff *skb,
struct sk_buff *in_skb, unsigned int offset,
size_t in_len)
{
struct sk_buff *nskb, *first, *last;
struct skb_shared_info *shinfo;
size_t chunk;
int sz;
if (strp->stm.full_len)
chunk = strp->stm.full_len - skb->len;
else
chunk = TLS_MAX_PAYLOAD_SIZE + PAGE_SIZE;
chunk = min(chunk, in_len);
nskb = tls_strp_skb_copy(strp, in_skb, offset, chunk);
if (!nskb)
return -ENOMEM;
shinfo = skb_shinfo(skb);
if (!shinfo->frag_list) {
shinfo->frag_list = nskb;
nskb->prev = nskb;
} else {
first = shinfo->frag_list;
last = first->prev;
last->next = nskb;
first->prev = nskb;
}
skb->len += chunk;
skb->data_len += chunk;
if (!strp->stm.full_len) {
sz = tls_rx_msg_size(strp, skb);
if (sz < 0)
return sz;
/* We may have over-read, sz == 0 is guaranteed under-read */
if (unlikely(sz && sz < skb->len)) {
int over = skb->len - sz;
WARN_ON_ONCE(over > chunk);
skb->len -= over;
skb->data_len -= over;
__pskb_trim(nskb, nskb->len - over);
chunk -= over;
}
strp->stm.full_len = sz;
}
return chunk;
}
static int tls_strp_copyin(read_descriptor_t *desc, struct sk_buff *in_skb,
unsigned int offset, size_t in_len)
{
struct tls_strparser *strp = (struct tls_strparser *)desc->arg.data;
struct sk_buff *skb;
int ret;
if (strp->msg_ready)
return 0;
skb = strp->anchor;
if (!skb->len)
skb_copy_decrypted(skb, in_skb);
else
strp->mixed_decrypted |= !!skb_cmp_decrypted(skb, in_skb);
if (IS_ENABLED(CONFIG_TLS_DEVICE) && strp->mixed_decrypted)
ret = tls_strp_copyin_skb(strp, skb, in_skb, offset, in_len);
else
ret = tls_strp_copyin_frag(strp, skb, in_skb, offset, in_len);
if (ret < 0) {
desc->error = ret;
ret = 0;
}
if (strp->stm.full_len && strp->stm.full_len == skb->len) {
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
desc->count = 0;
tls: fix lockless read of strp->msg_ready in ->poll tls_sk_poll is called without locking the socket, and needs to read strp->msg_ready (via tls_strp_msg_ready). Convert msg_ready to a bool and use READ_ONCE/WRITE_ONCE where needed. The remaining reads are only performed when the socket is locked. Fixes: 121dca784fc0 ("tls: suppress wakeups unless we have a full record") Signed-off-by: Sabrina Dubroca <sd@queasysnail.net> Link: https://lore.kernel.org/r/0b7ee062319037cf86af6b317b3d72f7bfcd2e97.1713797701.git.sd@queasysnail.net Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2024-04-24 10:25:47 +00:00
WRITE_ONCE(strp->msg_ready, 1);
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
tls_rx_msg_ready(strp);
}
tls: rx: strp: preserve decryption status of skbs when needed When receive buffer is small we try to copy out the data from TCP into a skb maintained by TLS to prevent connection from stalling. Unfortunately if a single record is made up of a mix of decrypted and non-decrypted skbs combining them into a single skb leads to loss of decryption status, resulting in decryption errors or data corruption. Similarly when trying to use TCP receive queue directly we need to make sure that all the skbs within the record have the same status. If we don't the mixed status will be detected correctly but we'll CoW the anchor, again collapsing it into a single paged skb without decrypted status preserved. So the "fixup" code will not know which parts of skb to re-encrypt. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:41 +00:00
return ret;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
}
static int tls_strp_read_copyin(struct tls_strparser *strp)
{
read_descriptor_t desc;
desc.arg.data = strp;
desc.error = 0;
desc.count = 1; /* give more than one skb per call */
/* sk should be locked here, so okay to do read_sock */
net/tls: Use tcp_read_sock() instead of ops->read_sock() TLS resets the protocol operations, so the read_sock() callback might be changed, too. In this case using sock->ops->readsock() in tls_strp_read_copyin() will enter an infinite recursion if the read_sock() callback is calling tls_rx_rec_wait() which will call into sock->ops->readsock() via tls_strp_read_copyin(). But as tls_strp_read_copyin() is supposed to produce data from the consumed socket and that socket is always a TCP socket we can call tcp_read_sock() directly without having to deal with callbacks. Signed-off-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Jakub Kicinski <kuba@kernel.org> Link: https://lore.kernel.org/r/20230726191556.41714-5-hare@suse.de Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-07-26 19:15:54 +00:00
tcp_read_sock(strp->sk, &desc, tls_strp_copyin);
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
return desc.error;
}
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
static int tls_strp_read_copy(struct tls_strparser *strp, bool qshort)
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
{
struct skb_shared_info *shinfo;
struct page *page;
int need_spc, len;
/* If the rbuf is small or rcv window has collapsed to 0 we need
* to read the data out. Otherwise the connection will stall.
* Without pressure threshold of INT_MAX will never be ready.
*/
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
if (likely(qshort && !tcp_epollin_ready(strp->sk, INT_MAX)))
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
return 0;
shinfo = skb_shinfo(strp->anchor);
shinfo->frag_list = NULL;
/* If we don't know the length go max plus page for cipher overhead */
need_spc = strp->stm.full_len ?: TLS_MAX_PAYLOAD_SIZE + PAGE_SIZE;
for (len = need_spc; len > 0; len -= PAGE_SIZE) {
page = alloc_page(strp->sk->sk_allocation);
if (!page) {
tls_strp_flush_anchor_copy(strp);
return -ENOMEM;
}
skb_fill_page_desc(strp->anchor, shinfo->nr_frags++,
page, 0, 0);
}
strp->copy_mode = 1;
strp->stm.offset = 0;
strp->anchor->len = 0;
strp->anchor->data_len = 0;
strp->anchor->truesize = round_up(need_spc, PAGE_SIZE);
tls_strp_read_copyin(strp);
return 0;
}
tls: rx: strp: force mixed decrypted records into copy mode If a record is partially decrypted we'll have to CoW it, anyway, so go into copy mode and allocate a writable skb right away. This will make subsequent fix simpler because we won't have to teach tls_strp_msg_make_copy() how to copy skbs while preserving decrypt status. Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:38 +00:00
static bool tls_strp_check_queue_ok(struct tls_strparser *strp)
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
{
unsigned int len = strp->stm.offset + strp->stm.full_len;
tls: rx: strp: force mixed decrypted records into copy mode If a record is partially decrypted we'll have to CoW it, anyway, so go into copy mode and allocate a writable skb right away. This will make subsequent fix simpler because we won't have to teach tls_strp_msg_make_copy() how to copy skbs while preserving decrypt status. Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:38 +00:00
struct sk_buff *first, *skb;
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
u32 seq;
tls: rx: strp: force mixed decrypted records into copy mode If a record is partially decrypted we'll have to CoW it, anyway, so go into copy mode and allocate a writable skb right away. This will make subsequent fix simpler because we won't have to teach tls_strp_msg_make_copy() how to copy skbs while preserving decrypt status. Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:38 +00:00
first = skb_shinfo(strp->anchor)->frag_list;
skb = first;
seq = TCP_SKB_CB(first)->seq;
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
tls: rx: strp: force mixed decrypted records into copy mode If a record is partially decrypted we'll have to CoW it, anyway, so go into copy mode and allocate a writable skb right away. This will make subsequent fix simpler because we won't have to teach tls_strp_msg_make_copy() how to copy skbs while preserving decrypt status. Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:38 +00:00
/* Make sure there's no duplicate data in the queue,
* and the decrypted status matches.
*/
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
while (skb->len < len) {
seq += skb->len;
len -= skb->len;
skb = skb->next;
if (TCP_SKB_CB(skb)->seq != seq)
return false;
tls: rx: strp: force mixed decrypted records into copy mode If a record is partially decrypted we'll have to CoW it, anyway, so go into copy mode and allocate a writable skb right away. This will make subsequent fix simpler because we won't have to teach tls_strp_msg_make_copy() how to copy skbs while preserving decrypt status. Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:38 +00:00
if (skb_cmp_decrypted(first, skb))
return false;
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
}
return true;
}
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
static void tls_strp_load_anchor_with_queue(struct tls_strparser *strp, int len)
{
struct tcp_sock *tp = tcp_sk(strp->sk);
struct sk_buff *first;
u32 offset;
first = tcp_recv_skb(strp->sk, tp->copied_seq, &offset);
if (WARN_ON_ONCE(!first))
return;
/* Bestow the state onto the anchor */
strp->anchor->len = offset + len;
strp->anchor->data_len = offset + len;
strp->anchor->truesize = offset + len;
skb_shinfo(strp->anchor)->frag_list = first;
skb_copy_header(strp->anchor, first);
strp->anchor->destructor = NULL;
strp->stm.offset = offset;
}
void tls_strp_msg_load(struct tls_strparser *strp, bool force_refresh)
{
struct strp_msg *rxm;
struct tls_msg *tlm;
DEBUG_NET_WARN_ON_ONCE(!strp->msg_ready);
DEBUG_NET_WARN_ON_ONCE(!strp->stm.full_len);
if (!strp->copy_mode && force_refresh) {
if (WARN_ON(tcp_inq(strp->sk) < strp->stm.full_len))
return;
tls_strp_load_anchor_with_queue(strp, strp->stm.full_len);
}
rxm = strp_msg(strp->anchor);
rxm->full_len = strp->stm.full_len;
rxm->offset = strp->stm.offset;
tlm = tls_msg(strp->anchor);
tlm->control = strp->mark;
}
/* Called with lock held on lower socket */
static int tls_strp_read_sock(struct tls_strparser *strp)
{
int sz, inq;
inq = tcp_inq(strp->sk);
if (inq < 1)
return 0;
if (unlikely(strp->copy_mode))
return tls_strp_read_copyin(strp);
if (inq < strp->stm.full_len)
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
return tls_strp_read_copy(strp, true);
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
if (!strp->stm.full_len) {
tls_strp_load_anchor_with_queue(strp, inq);
sz = tls_rx_msg_size(strp, strp->anchor);
if (sz < 0) {
tls_strp_abort_strp(strp, sz);
return sz;
}
strp->stm.full_len = sz;
if (!strp->stm.full_len || inq < strp->stm.full_len)
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
return tls_strp_read_copy(strp, true);
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
}
tls: rx: strp: force mixed decrypted records into copy mode If a record is partially decrypted we'll have to CoW it, anyway, so go into copy mode and allocate a writable skb right away. This will make subsequent fix simpler because we won't have to teach tls_strp_msg_make_copy() how to copy skbs while preserving decrypt status. Tested-by: Shai Amiram <samiram@nvidia.com> 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>
2023-05-17 01:50:38 +00:00
if (!tls_strp_check_queue_ok(strp))
tls: strp: make sure the TCP skbs do not have overlapping data TLS tries to get away with using the TCP input queue directly. This does not work if there is duplicated data (multiple skbs holding bytes for the same seq number range due to retransmits). Check for this condition and fall back to copy mode, it should be rare. Fixes: 84c61fe1a75b ("tls: rx: do not use the standard strparser") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-10-12 22:55:20 +00:00
return tls_strp_read_copy(strp, false);
tls: fix lockless read of strp->msg_ready in ->poll tls_sk_poll is called without locking the socket, and needs to read strp->msg_ready (via tls_strp_msg_ready). Convert msg_ready to a bool and use READ_ONCE/WRITE_ONCE where needed. The remaining reads are only performed when the socket is locked. Fixes: 121dca784fc0 ("tls: suppress wakeups unless we have a full record") Signed-off-by: Sabrina Dubroca <sd@queasysnail.net> Link: https://lore.kernel.org/r/0b7ee062319037cf86af6b317b3d72f7bfcd2e97.1713797701.git.sd@queasysnail.net Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2024-04-24 10:25:47 +00:00
WRITE_ONCE(strp->msg_ready, 1);
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
tls_rx_msg_ready(strp);
return 0;
}
void tls_strp_check_rcv(struct tls_strparser *strp)
{
if (unlikely(strp->stopped) || strp->msg_ready)
return;
if (tls_strp_read_sock(strp) == -ENOMEM)
queue_work(tls_strp_wq, &strp->work);
}
/* Lower sock lock held */
void tls_strp_data_ready(struct tls_strparser *strp)
{
/* This check is needed to synchronize with do_tls_strp_work.
* do_tls_strp_work acquires a process lock (lock_sock) whereas
* the lock held here is bh_lock_sock. The two locks can be
* held by different threads at the same time, but bh_lock_sock
* allows a thread in BH context to safely check if the process
* lock is held. In this case, if the lock is held, queue work.
*/
if (sock_owned_by_user_nocheck(strp->sk)) {
queue_work(tls_strp_wq, &strp->work);
return;
}
tls_strp_check_rcv(strp);
}
static void tls_strp_work(struct work_struct *w)
{
struct tls_strparser *strp =
container_of(w, struct tls_strparser, work);
lock_sock(strp->sk);
tls_strp_check_rcv(strp);
release_sock(strp->sk);
}
void tls_strp_msg_done(struct tls_strparser *strp)
{
WARN_ON(!strp->stm.full_len);
if (likely(!strp->copy_mode))
tcp_read_done(strp->sk, strp->stm.full_len);
else
tls_strp_flush_anchor_copy(strp);
tls: fix lockless read of strp->msg_ready in ->poll tls_sk_poll is called without locking the socket, and needs to read strp->msg_ready (via tls_strp_msg_ready). Convert msg_ready to a bool and use READ_ONCE/WRITE_ONCE where needed. The remaining reads are only performed when the socket is locked. Fixes: 121dca784fc0 ("tls: suppress wakeups unless we have a full record") Signed-off-by: Sabrina Dubroca <sd@queasysnail.net> Link: https://lore.kernel.org/r/0b7ee062319037cf86af6b317b3d72f7bfcd2e97.1713797701.git.sd@queasysnail.net Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2024-04-24 10:25:47 +00:00
WRITE_ONCE(strp->msg_ready, 0);
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
memset(&strp->stm, 0, sizeof(strp->stm));
tls_strp_check_rcv(strp);
}
void tls_strp_stop(struct tls_strparser *strp)
{
strp->stopped = 1;
}
int tls_strp_init(struct tls_strparser *strp, struct sock *sk)
{
memset(strp, 0, sizeof(*strp));
strp->sk = sk;
strp->anchor = alloc_skb(0, GFP_KERNEL);
if (!strp->anchor)
return -ENOMEM;
INIT_WORK(&strp->work, tls_strp_work);
tls: rx: device: add input CoW helper Wrap the remaining skb_cow_data() into a helper, so it's easier to replace down the lane. The new version will change the skb so make sure relevant pointers get reloaded after the call. Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:32 +00:00
return 0;
}
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
/* strp must already be stopped so that tls_strp_recv will no longer be called.
* Note that tls_strp_done is not called with the lower socket held.
*/
void tls_strp_done(struct tls_strparser *strp)
tls: rx: async: hold onto the input skb Async crypto currently benefits from the fact that we decrypt in place. When we allow input and output to be different skbs we will have to hang onto the input while we move to the next record. Clone the inputs and keep them on a list. Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-07-15 05:22:33 +00:00
{
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
WARN_ON(!strp->stopped);
tls: rx: async: hold onto the input skb Async crypto currently benefits from the fact that we decrypt in place. When we allow input and output to be different skbs we will have to hang onto the input while we move to the next record. Clone the inputs and keep them on a list. Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-07-15 05:22:33 +00:00
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
cancel_work_sync(&strp->work);
tls_strp_anchor_free(strp);
}
int __init tls_strp_dev_init(void)
{
tls: strp: rename and multithread the workqueue Paolo points out that there seems to be no strong reason strparser users a single threaded workqueue. Perhaps there were some performance or pinning considerations? Since we don't know (and it's the slow path) let's default to the most natural, multi-threaded choice. Also rename the workqueue to "tls-". Suggested-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-27 03:15:23 +00:00
tls_strp_wq = create_workqueue("tls-strp");
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
if (unlikely(!tls_strp_wq))
tls: rx: async: hold onto the input skb Async crypto currently benefits from the fact that we decrypt in place. When we allow input and output to be different skbs we will have to hang onto the input while we move to the next record. Clone the inputs and keep them on a list. Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-07-15 05:22:33 +00:00
return -ENOMEM;
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
tls: rx: async: hold onto the input skb Async crypto currently benefits from the fact that we decrypt in place. When we allow input and output to be different skbs we will have to hang onto the input while we move to the next record. Clone the inputs and keep them on a list. Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2022-07-15 05:22:33 +00:00
return 0;
}
tls: rx: do not use the standard strparser TLS is a relatively poor fit for strparser. We pause the input every time a message is received, wait for a read which will decrypt the message, start the parser, repeat. strparser is built to delineate the messages, wrap them in individual skbs and let them float off into the stack or a different socket. TLS wants the data pages and nothing else. There's no need for TLS to keep cloning (and occasionally skb_unclone()'ing) the TCP rx queue. This patch uses a pre-allocated skb and attaches the skbs from the TCP rx queue to it as frags. TLS is careful never to modify the input skb without CoW'ing / detaching it first. Since we call TCP rx queue cleanup directly we also get back the benefit of skb deferred free. Overall this results in a 6% gain in my benchmarks. Acked-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-07-22 23:50:33 +00:00
void tls_strp_dev_exit(void)
{
destroy_workqueue(tls_strp_wq);
}
Reference in New Issue Copy Permalink