diff mbox series

[v2,net-next,2/2] tcp: fix possible freeze in tx path under memory pressure

Message ID 20220614171734.1103875-3-eric.dumazet@gmail.com (mailing list archive)
State Accepted
Delegated to: Netdev Maintainers
Headers show
Series tcp: final (?) round of mem pressure fixes | expand

Checks

Context Check Description
netdev/tree_selection success Clearly marked for net-next
netdev/fixes_present success Fixes tag not required for -next series
netdev/subject_prefix success Link
netdev/cover_letter success Series has a cover letter
netdev/patch_count success Link
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 2 this patch: 2
netdev/cc_maintainers warning 2 maintainers not CCed: yoshfuji@linux-ipv6.org dsahern@kernel.org
netdev/build_clang success Errors and warnings before: 6 this patch: 6
netdev/module_param success Was 0 now: 0
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success Fixes tag looks correct
netdev/build_allmodconfig_warn success Errors and warnings before: 2 this patch: 2
netdev/checkpatch success total: 0 errors, 0 warnings, 0 checks, 57 lines checked
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0

Commit Message

Eric Dumazet June 14, 2022, 5:17 p.m. UTC
From: Eric Dumazet <edumazet@google.com>

Blamed commit only dealt with applications issuing small writes.

Issue here is that we allow to force memory schedule for the sk_buff
allocation, but we have no guarantee that sendmsg() is able to
copy some payload in it.

In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.

For example, if we consider tcp_wmem[0] = 4096 (default on x86),
and initial skb->truesize being 1280, tcp_sendmsg() is able to
copy up to 2816 bytes under memory pressure.

Before this patch a sendmsg() sending more than 2816 bytes
would either block forever (if persistent memory pressure),
or return -EAGAIN.

For bigger MTU networks, it is advised to increase tcp_wmem[0]
to avoid sending too small packets.

v2: deal with zero copy paths.

Fixes: 8e4d980ac215 ("tcp: fix behavior for epoll edge trigger")
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/tcp.c | 33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

Comments

Soheil Hassas Yeganeh June 14, 2022, 5:42 p.m. UTC | #1
On Tue, Jun 14, 2022 at 1:17 PM Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> From: Eric Dumazet <edumazet@google.com>
>
> Blamed commit only dealt with applications issuing small writes.
>
> Issue here is that we allow to force memory schedule for the sk_buff
> allocation, but we have no guarantee that sendmsg() is able to
> copy some payload in it.
>
> In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
>
> For example, if we consider tcp_wmem[0] = 4096 (default on x86),
> and initial skb->truesize being 1280, tcp_sendmsg() is able to
> copy up to 2816 bytes under memory pressure.
>
> Before this patch a sendmsg() sending more than 2816 bytes
> would either block forever (if persistent memory pressure),
> or return -EAGAIN.
>
> For bigger MTU networks, it is advised to increase tcp_wmem[0]
> to avoid sending too small packets.
>
> v2: deal with zero copy paths.
>
> Fixes: 8e4d980ac215 ("tcp: fix behavior for epoll edge trigger")
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Acked-by: Soheil Hassas Yeganeh <soheil@google.com>

Very nice find! Thank you!

> ---
>  net/ipv4/tcp.c | 33 +++++++++++++++++++++++++++++----
>  1 file changed, 29 insertions(+), 4 deletions(-)
>
> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> index 14ebb4ec4a51f3c55501aa53423ce897599e8637..56083c2497f0b695c660256aa43f8a743d481697 100644
> --- a/net/ipv4/tcp.c
> +++ b/net/ipv4/tcp.c
> @@ -951,6 +951,23 @@ static int tcp_downgrade_zcopy_pure(struct sock *sk, struct sk_buff *skb)
>         return 0;
>  }
>
> +static int tcp_wmem_schedule(struct sock *sk, int copy)
> +{
> +       int left;
> +
> +       if (likely(sk_wmem_schedule(sk, copy)))
> +               return copy;
> +
> +       /* We could be in trouble if we have nothing queued.
> +        * Use whatever is left in sk->sk_forward_alloc and tcp_wmem[0]
> +        * to guarantee some progress.
> +        */
> +       left = sock_net(sk)->ipv4.sysctl_tcp_wmem[0] - sk->sk_wmem_queued;
> +       if (left > 0)
> +               sk_forced_mem_schedule(sk, min(left, copy));
> +       return min(copy, sk->sk_forward_alloc);
> +}
> +
>  static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
>                                       struct page *page, int offset, size_t *size)
>  {
> @@ -986,7 +1003,11 @@ static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
>                 tcp_mark_push(tp, skb);
>                 goto new_segment;
>         }
> -       if (tcp_downgrade_zcopy_pure(sk, skb) || !sk_wmem_schedule(sk, copy))
> +       if (tcp_downgrade_zcopy_pure(sk, skb))
> +               return NULL;
> +
> +       copy = tcp_wmem_schedule(sk, copy);
> +       if (!copy)
>                 return NULL;
>
>         if (can_coalesce) {
> @@ -1334,8 +1355,11 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
>
>                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
>
> -                       if (tcp_downgrade_zcopy_pure(sk, skb) ||
> -                           !sk_wmem_schedule(sk, copy))
> +                       if (tcp_downgrade_zcopy_pure(sk, skb))
> +                               goto wait_for_space;
> +
> +                       copy = tcp_wmem_schedule(sk, copy);
> +                       if (!copy)
>                                 goto wait_for_space;
>
>                         err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
> @@ -1362,7 +1386,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
>                                 skb_shinfo(skb)->flags |= SKBFL_PURE_ZEROCOPY;
>
>                         if (!skb_zcopy_pure(skb)) {
> -                               if (!sk_wmem_schedule(sk, copy))
> +                               copy = tcp_wmem_schedule(sk, copy);
> +                               if (!copy)
>                                         goto wait_for_space;
>                         }
>
> --
> 2.36.1.476.g0c4daa206d-goog
>
Wei Wang June 14, 2022, 6:41 p.m. UTC | #2
On Tue, Jun 14, 2022 at 10:42 AM Soheil Hassas Yeganeh
<soheil@google.com> wrote:
>
> On Tue, Jun 14, 2022 at 1:17 PM Eric Dumazet <eric.dumazet@gmail.com> wrote:
> >
> > From: Eric Dumazet <edumazet@google.com>
> >
> > Blamed commit only dealt with applications issuing small writes.
> >
> > Issue here is that we allow to force memory schedule for the sk_buff
> > allocation, but we have no guarantee that sendmsg() is able to
> > copy some payload in it.
> >
> > In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
> >
> > For example, if we consider tcp_wmem[0] = 4096 (default on x86),
> > and initial skb->truesize being 1280, tcp_sendmsg() is able to
> > copy up to 2816 bytes under memory pressure.
> >
> > Before this patch a sendmsg() sending more than 2816 bytes
> > would either block forever (if persistent memory pressure),
> > or return -EAGAIN.
> >
> > For bigger MTU networks, it is advised to increase tcp_wmem[0]
> > to avoid sending too small packets.
> >
> > v2: deal with zero copy paths.
> >
> > Fixes: 8e4d980ac215 ("tcp: fix behavior for epoll edge trigger")
> > Signed-off-by: Eric Dumazet <edumazet@google.com>
>
> Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
>
> Very nice find! Thank you!
>
> > ---
> >  net/ipv4/tcp.c | 33 +++++++++++++++++++++++++++++----
> >  1 file changed, 29 insertions(+), 4 deletions(-)
> >
> > diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> > index 14ebb4ec4a51f3c55501aa53423ce897599e8637..56083c2497f0b695c660256aa43f8a743d481697 100644
> > --- a/net/ipv4/tcp.c
> > +++ b/net/ipv4/tcp.c
> > @@ -951,6 +951,23 @@ static int tcp_downgrade_zcopy_pure(struct sock *sk, struct sk_buff *skb)
> >         return 0;
> >  }
> >
> > +static int tcp_wmem_schedule(struct sock *sk, int copy)
> > +{
> > +       int left;
> > +
> > +       if (likely(sk_wmem_schedule(sk, copy)))
> > +               return copy;
> > +
> > +       /* We could be in trouble if we have nothing queued.
> > +        * Use whatever is left in sk->sk_forward_alloc and tcp_wmem[0]
> > +        * to guarantee some progress.
> > +        */
> > +       left = sock_net(sk)->ipv4.sysctl_tcp_wmem[0] - sk->sk_wmem_queued;
> > +       if (left > 0)
> > +               sk_forced_mem_schedule(sk, min(left, copy));
> > +       return min(copy, sk->sk_forward_alloc);
> > +}
> > +
> >  static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
> >                                       struct page *page, int offset, size_t *size)
> >  {
> > @@ -986,7 +1003,11 @@ static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
> >                 tcp_mark_push(tp, skb);
> >                 goto new_segment;
> >         }
> > -       if (tcp_downgrade_zcopy_pure(sk, skb) || !sk_wmem_schedule(sk, copy))
> > +       if (tcp_downgrade_zcopy_pure(sk, skb))
> > +               return NULL;

Do we need to take care of the call to sk_wmem_schedule() inside
tcp_downgrade_zcopy_pure()?

> > +
> > +       copy = tcp_wmem_schedule(sk, copy);
> > +       if (!copy)
> >                 return NULL;
> >
> >         if (can_coalesce) {
> > @@ -1334,8 +1355,11 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
> >
> >                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
> >
> > -                       if (tcp_downgrade_zcopy_pure(sk, skb) ||
> > -                           !sk_wmem_schedule(sk, copy))
> > +                       if (tcp_downgrade_zcopy_pure(sk, skb))
> > +                               goto wait_for_space;
> > +
> > +                       copy = tcp_wmem_schedule(sk, copy);
> > +                       if (!copy)
> >                                 goto wait_for_space;
> >
> >                         err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
> > @@ -1362,7 +1386,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
> >                                 skb_shinfo(skb)->flags |= SKBFL_PURE_ZEROCOPY;
> >
> >                         if (!skb_zcopy_pure(skb)) {
> > -                               if (!sk_wmem_schedule(sk, copy))
> > +                               copy = tcp_wmem_schedule(sk, copy);
> > +                               if (!copy)
> >                                         goto wait_for_space;
> >                         }
> >
> > --
> > 2.36.1.476.g0c4daa206d-goog
> >
Wei Wang June 14, 2022, 8:17 p.m. UTC | #3
On Tue, Jun 14, 2022 at 12:10 PM Eric Dumazet <edumazet@google.com> wrote:
>
>
>
> On Tue, Jun 14, 2022, 11:41 AM Wei Wang <weiwan@google.com> wrote:
>>
>> On Tue, Jun 14, 2022 at 10:42 AM Soheil Hassas Yeganeh
>> <soheil@google.com> wrote:
>> >
>> > On Tue, Jun 14, 2022 at 1:17 PM Eric Dumazet <eric.dumazet@gmail.com> wrote:
>> > >
>> > > From: Eric Dumazet <edumazet@google.com>
>> > >
>> > > Blamed commit only dealt with applications issuing small writes.
>> > >
>> > > Issue here is that we allow to force memory schedule for the sk_buff
>> > > allocation, but we have no guarantee that sendmsg() is able to
>> > > copy some payload in it.
>> > >
>> > > In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
>> > >
>> > > For example, if we consider tcp_wmem[0] = 4096 (default on x86),
>> > > and initial skb->truesize being 1280, tcp_sendmsg() is able to
>> > > copy up to 2816 bytes under memory pressure.
>> > >
>> > > Before this patch a sendmsg() sending more than 2816 bytes
>> > > would either block forever (if persistent memory pressure),
>> > > or return -EAGAIN.
>> > >
>> > > For bigger MTU networks, it is advised to increase tcp_wmem[0]
>> > > to avoid sending too small packets.
>> > >
>> > > v2: deal with zero copy paths.
>> > >
>> > > Fixes: 8e4d980ac215 ("tcp: fix behavior for epoll edge trigger")
>> > > Signed-off-by: Eric Dumazet <edumazet@google.com>

Reviewed-by: Wei Wang <weiwan@google.com>

>> >
>> > Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
>> >
>> > Very nice find! Thank you!
>> >
>> > > ---
>> > >  net/ipv4/tcp.c | 33 +++++++++++++++++++++++++++++----
>> > >  1 file changed, 29 insertions(+), 4 deletions(-)
>> > >
>> > > diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
>> > > index 14ebb4ec4a51f3c55501aa53423ce897599e8637..56083c2497f0b695c660256aa43f8a743d481697 100644
>> > > --- a/net/ipv4/tcp.c
>> > > +++ b/net/ipv4/tcp.c
>> > > @@ -951,6 +951,23 @@ static int tcp_downgrade_zcopy_pure(struct sock *sk, struct sk_buff *skb)
>> > >         return 0;
>> > >  }
>> > >
>> > > +static int tcp_wmem_schedule(struct sock *sk, int copy)
>> > > +{
>> > > +       int left;
>> > > +
>> > > +       if (likely(sk_wmem_schedule(sk, copy)))
>> > > +               return copy;
>> > > +
>> > > +       /* We could be in trouble if we have nothing queued.
>> > > +        * Use whatever is left in sk->sk_forward_alloc and tcp_wmem[0]
>> > > +        * to guarantee some progress.
>> > > +        */
>> > > +       left = sock_net(sk)->ipv4.sysctl_tcp_wmem[0] - sk->sk_wmem_queued;
>> > > +       if (left > 0)
>> > > +               sk_forced_mem_schedule(sk, min(left, copy));
>> > > +       return min(copy, sk->sk_forward_alloc);
>> > > +}
>> > > +
>> > >  static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
>> > >                                       struct page *page, int offset, size_t *size)
>> > >  {
>> > > @@ -986,7 +1003,11 @@ static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
>> > >                 tcp_mark_push(tp, skb);
>> > >                 goto new_segment;
>> > >         }
>> > > -       if (tcp_downgrade_zcopy_pure(sk, skb) || !sk_wmem_schedule(sk, copy))
>> > > +       if (tcp_downgrade_zcopy_pure(sk, skb))
>> > > +               return NULL;
>>
>> Do we need to take care of the call to sk_wmem_schedule() inside
>> tcp_downgrade_zcopy_pure()?
>
>
> We can not. Payload has been added already, and will be sent eventually.

Ack. Thanks for the explanation.


>>
>>
>> > > +
>> > > +       copy = tcp_wmem_schedule(sk, copy);
>> > > +       if (!copy)
>> > >                 return NULL;
>> > >
>> > >         if (can_coalesce) {
>> > > @@ -1334,8 +1355,11 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
>> > >
>> > >                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
>> > >
>> > > -                       if (tcp_downgrade_zcopy_pure(sk, skb) ||
>> > > -                           !sk_wmem_schedule(sk, copy))
>> > > +                       if (tcp_downgrade_zcopy_pure(sk, skb))
>> > > +                               goto wait_for_space;
>> > > +
>> > > +                       copy = tcp_wmem_schedule(sk, copy);
>> > > +                       if (!copy)
>> > >                                 goto wait_for_space;
>> > >
>> > >                         err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
>> > > @@ -1362,7 +1386,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
>> > >                                 skb_shinfo(skb)->flags |= SKBFL_PURE_ZEROCOPY;
>> > >
>> > >                         if (!skb_zcopy_pure(skb)) {
>> > > -                               if (!sk_wmem_schedule(sk, copy))
>> > > +                               copy = tcp_wmem_schedule(sk, copy);
>> > > +                               if (!copy)
>> > >                                         goto wait_for_space;
>> > >                         }
>> > >
>> > > --
>> > > 2.36.1.476.g0c4daa206d-goog
>> > >
Shakeel Butt June 14, 2022, 8:31 p.m. UTC | #4
On Tue, Jun 14, 2022 at 10:17 AM Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> From: Eric Dumazet <edumazet@google.com>
>
> Blamed commit only dealt with applications issuing small writes.
>
> Issue here is that we allow to force memory schedule for the sk_buff
> allocation, but we have no guarantee that sendmsg() is able to
> copy some payload in it.
>
> In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
>
> For example, if we consider tcp_wmem[0] = 4096 (default on x86),
> and initial skb->truesize being 1280, tcp_sendmsg() is able to
> copy up to 2816 bytes under memory pressure.
>
> Before this patch a sendmsg() sending more than 2816 bytes
> would either block forever (if persistent memory pressure),
> or return -EAGAIN.
>
> For bigger MTU networks, it is advised to increase tcp_wmem[0]
> to avoid sending too small packets.
>
> v2: deal with zero copy paths.
>
> Fixes: 8e4d980ac215 ("tcp: fix behavior for epoll edge trigger")
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Reviewed-by: Shakeel Butt <shakeelb@google.com>
Jason A. Donenfeld June 17, 2022, 10:08 a.m. UTC | #5
Hi,

On Tue, Jun 14, 2022 at 10:17:34AM -0700, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> Blamed commit only dealt with applications issuing small writes.
> 
> Issue here is that we allow to force memory schedule for the sk_buff
> allocation, but we have no guarantee that sendmsg() is able to
> copy some payload in it.
> 
> In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
> 
> For example, if we consider tcp_wmem[0] = 4096 (default on x86),
> and initial skb->truesize being 1280, tcp_sendmsg() is able to
> copy up to 2816 bytes under memory pressure.
> 
> Before this patch a sendmsg() sending more than 2816 bytes
> would either block forever (if persistent memory pressure),
> or return -EAGAIN.
> 
> For bigger MTU networks, it is advised to increase tcp_wmem[0]
> to avoid sending too small packets.
> 
> v2: deal with zero copy paths.

I think this might have gotten double applied:

https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=849b425cd091e1804af964b771761cfbefbafb43
https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=f54755f6a11accb2db5ef17f8f75aad0875aefdc

and now net-next builds are broken:

../../../../../../../../net/ipv4/tcp.c:971:12: error: redefinition of ‘tcp_wmem_schedule’
  971 | static int tcp_wmem_schedule(struct sock *sk, int copy)                                |            ^~~~~~~~~~~~~~~~~
../../../../../../../../net/ipv4/tcp.c:954:12: note: previous definition of ‘tcp_wmem_schedule’ with type ‘int(struct sock *, int)’
  954 | static int tcp_wmem_schedule(struct sock *sk, int copy)                                |            ^~~~~~~~~~~~~~~~~
../../../../../../../../net/ipv4/tcp.c:954:12: warning: ‘tcp_wmem_schedule’ defined but not used [-Wunused-function]                                                              make[5]: *** [/home/wgci/tmp/2813390.12234/tmp.0PMBO65tGf/scripts/Makefile.build:249: net
/ipv4/tcp.o] Error 1                                                                     make[4]: *** [/home/wgci/tmp/2813390.12234/tmp.0PMBO65tGf/scripts/Makefile.build:466: net/ipv4] Error 2
make[4]: *** Waiting for unfinished jobs....

Jason
Jason A. Donenfeld June 17, 2022, 10:13 a.m. UTC | #6
On Fri, Jun 17, 2022 at 12:08:14PM +0200, Jason A. Donenfeld wrote:
> Hi,
> 
> On Tue, Jun 14, 2022 at 10:17:34AM -0700, Eric Dumazet wrote:
> > From: Eric Dumazet <edumazet@google.com>
> > 
> > Blamed commit only dealt with applications issuing small writes.
> > 
> > Issue here is that we allow to force memory schedule for the sk_buff
> > allocation, but we have no guarantee that sendmsg() is able to
> > copy some payload in it.
> > 
> > In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
> > 
> > For example, if we consider tcp_wmem[0] = 4096 (default on x86),
> > and initial skb->truesize being 1280, tcp_sendmsg() is able to
> > copy up to 2816 bytes under memory pressure.
> > 
> > Before this patch a sendmsg() sending more than 2816 bytes
> > would either block forever (if persistent memory pressure),
> > or return -EAGAIN.
> > 
> > For bigger MTU networks, it is advised to increase tcp_wmem[0]
> > to avoid sending too small packets.
> > 
> > v2: deal with zero copy paths.
> 
> I think this might have gotten double applied:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=849b425cd091e1804af964b771761cfbefbafb43
> https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=f54755f6a11accb2db5ef17f8f75aad0875aefdc
> 
> and now net-next builds are broken:
> 
> ../../../../../../../../net/ipv4/tcp.c:971:12: error: redefinition of ‘tcp_wmem_schedule’
>   971 | static int tcp_wmem_schedule(struct sock *sk, int copy)                                |            ^~~~~~~~~~~~~~~~~
> ../../../../../../../../net/ipv4/tcp.c:954:12: note: previous definition of ‘tcp_wmem_schedule’ with type ‘int(struct sock *, int)’
>   954 | static int tcp_wmem_schedule(struct sock *sk, int copy)                                |            ^~~~~~~~~~~~~~~~~
> ../../../../../../../../net/ipv4/tcp.c:954:12: warning: ‘tcp_wmem_schedule’ defined but not used [-Wunused-function]                                                              make[5]: *** [/home/wgci/tmp/2813390.12234/tmp.0PMBO65tGf/scripts/Makefile.build:249: net
> /ipv4/tcp.o] Error 1                                                                     make[4]: *** [/home/wgci/tmp/2813390.12234/tmp.0PMBO65tGf/scripts/Makefile.build:466: net/ipv4] Error 2
> make[4]: *** Waiting for unfinished jobs....

Ah, fixed already:
https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=fd8b330ce1bb79ba00047435cc213b14f886bf1f

> 
> Jason
diff mbox series

Patch

diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 14ebb4ec4a51f3c55501aa53423ce897599e8637..56083c2497f0b695c660256aa43f8a743d481697 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -951,6 +951,23 @@  static int tcp_downgrade_zcopy_pure(struct sock *sk, struct sk_buff *skb)
 	return 0;
 }
 
+static int tcp_wmem_schedule(struct sock *sk, int copy)
+{
+	int left;
+
+	if (likely(sk_wmem_schedule(sk, copy)))
+		return copy;
+
+	/* We could be in trouble if we have nothing queued.
+	 * Use whatever is left in sk->sk_forward_alloc and tcp_wmem[0]
+	 * to guarantee some progress.
+	 */
+	left = sock_net(sk)->ipv4.sysctl_tcp_wmem[0] - sk->sk_wmem_queued;
+	if (left > 0)
+		sk_forced_mem_schedule(sk, min(left, copy));
+	return min(copy, sk->sk_forward_alloc);
+}
+
 static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
 				      struct page *page, int offset, size_t *size)
 {
@@ -986,7 +1003,11 @@  static struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
 		tcp_mark_push(tp, skb);
 		goto new_segment;
 	}
-	if (tcp_downgrade_zcopy_pure(sk, skb) || !sk_wmem_schedule(sk, copy))
+	if (tcp_downgrade_zcopy_pure(sk, skb))
+		return NULL;
+
+	copy = tcp_wmem_schedule(sk, copy);
+	if (!copy)
 		return NULL;
 
 	if (can_coalesce) {
@@ -1334,8 +1355,11 @@  int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 
 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
 
-			if (tcp_downgrade_zcopy_pure(sk, skb) ||
-			    !sk_wmem_schedule(sk, copy))
+			if (tcp_downgrade_zcopy_pure(sk, skb))
+				goto wait_for_space;
+
+			copy = tcp_wmem_schedule(sk, copy);
+			if (!copy)
 				goto wait_for_space;
 
 			err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
@@ -1362,7 +1386,8 @@  int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 				skb_shinfo(skb)->flags |= SKBFL_PURE_ZEROCOPY;
 
 			if (!skb_zcopy_pure(skb)) {
-				if (!sk_wmem_schedule(sk, copy))
+				copy = tcp_wmem_schedule(sk, copy);
+				if (!copy)
 					goto wait_for_space;
 			}