| Message ID | 20210628144908.881499-1-phind.uet@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series | tcp: Do not reset the icsk_ca_initialized in tcp_init_transfer. | expand |
On June 29, 2021 1:20:19 AM GMT+08:00, Neal Cardwell <ncardwell@google.com> wrote: >) > >On Mon, Jun 28, 2021 at 1:15 PM Phi Nguyen <phind.uet@gmail.com> wrote: >> >> On 6/29/2021 12:24 AM, Neal Cardwell wrote: >> >> > Thanks. >> > >> > Can you also please provide a summary of the event sequence that >> > triggers the bug? Based on your Reported-by tag, I guess this is >based >> > on the syzbot reproducer: >> > >> > >https://groups.google.com/g/syzkaller-bugs/c/VbHoSsBz0hk/m/cOxOoTgPCAAJ >> > >> > but perhaps you can give a summary of the event sequence that >causes >> > the bug? Is it that the call: >> > >> > setsockopt$inet_tcp_TCP_CONGESTION(r0, 0x6, 0xd, >> > &(0x7f0000000000)='cdg\x00', 0x4) >> > >> > initializes the CC and happens before the connection is >established, >> > and then when the connection is established, the line that sets: >> > icsk->icsk_ca_initialized = 0; >> > is incorrect, causing the CC to be initialized again without first >> > calling the cleanup code that deallocates the CDG-allocated memory? >> > >> > thanks, >> > neal >> > >> >> Hi Neal, >> >> The gdb stack trace that lead to init_transfer_input() is as bellow, >the >> current sock state is TCP_SYN_RECV. > >Thanks. That makes sense as a snapshot of time for >tcp_init_transfer(), but I think what would be more useful would be a >description of the sequence of events, including when the CC was >initialized previous to that point (as noted above, was it that the >setsockopt(TCP_CONGESTION) completed before that point?). > >thanks, >neal I resend my message because I accidently used html format in last one. I am very sorry for the inconvenience caused. --- Yes, the CC had been initialized by the setsockopt, after that, it was initialized again in function tcp_init_transfer() because of setting isck_ca_initialized to 0. Regards, Phi.
On June 29, 2021 4:21:59 PM GMT+08:00, Eric Dumazet <edumazet@google.com> wrote: >On Tue, Jun 29, 2021 at 9:17 AM Nguyen Dinh Phi <phind.uet@gmail.com> >wrote: >> >> On June 29, 2021 1:20:19 AM GMT+08:00, Neal Cardwell ><ncardwell@google.com> wrote: >> >) >> > >> >On Mon, Jun 28, 2021 at 1:15 PM Phi Nguyen <phind.uet@gmail.com> >wrote: >> >> >> >> On 6/29/2021 12:24 AM, Neal Cardwell wrote: >> >> >> >> > Thanks. >> >> > >> >> > Can you also please provide a summary of the event sequence that >> >> > triggers the bug? Based on your Reported-by tag, I guess this is >> >based >> >> > on the syzbot reproducer: >> >> > >> >> > >> >>https://groups.google.com/g/syzkaller-bugs/c/VbHoSsBz0hk/m/cOxOoTgPCAAJ >> >> > >> >> > but perhaps you can give a summary of the event sequence that >> >causes >> >> > the bug? Is it that the call: >> >> > >> >> > setsockopt$inet_tcp_TCP_CONGESTION(r0, 0x6, 0xd, >> >> > &(0x7f0000000000)='cdg\x00', 0x4) >> >> > >> >> > initializes the CC and happens before the connection is >> >established, >> >> > and then when the connection is established, the line that sets: >> >> > icsk->icsk_ca_initialized = 0; >> >> > is incorrect, causing the CC to be initialized again without >first >> >> > calling the cleanup code that deallocates the CDG-allocated >memory? >> >> > >> >> > thanks, >> >> > neal >> >> > >> >> >> >> Hi Neal, >> >> >> >> The gdb stack trace that lead to init_transfer_input() is as >bellow, >> >the >> >> current sock state is TCP_SYN_RECV. >> > >> >Thanks. That makes sense as a snapshot of time for >> >tcp_init_transfer(), but I think what would be more useful would be >a >> >description of the sequence of events, including when the CC was >> >initialized previous to that point (as noted above, was it that the >> >setsockopt(TCP_CONGESTION) completed before that point?). >> > >> >thanks, >> >neal >> >> I resend my message because I accidently used html format in last >one. I am very sorry for the inconvenience caused. >> --- >> Yes, the CC had been initialized by the setsockopt, after that, it >was initialized again in function tcp_init_transfer() because of >setting isck_ca_initialized to 0. > >"the setsockopt" is rather vague, sorry. > > >The hard part is that all scenarios have to be considered. > >TCP flows can either be passive and active. > >CC can be set : > >1) Before the connect() or accept() >2) After the connect() or accept() >3) after the connect() but before 3WHS is completed. > >So we need to make sure all cases will still work with any combination >of CDG CC (before/after) in the picture. > >Note that a memory leak for a restricted CC (CDG can only be used by >CAP_NET_ADMIN privileged user) > is a small problem compared to more serious bug that could be added >by an incomplete fix. > >I also note that if icsk_ca_priv] was increased from 104 to 120 bytes, >tcp_cdg would no longer need a dynamic memory allocation. > >Thank you. Hi, I will try to see whether I am able to get the full sequence. I am also affraid of making a change that could affect big part of the kernel. About CDG, how we can get rid of dynamic allocation by increasing icsk_priv_data to 120? because I see that the window size is a module parameter, so I guess it is not a fixed value. Because the problem only happens with CDG, is adding check in its tcp_cdg_init() function Ok? And about icsk_ca_initialized, Could I expect it to be 0 in CC's init functions? Thank you.
On Tue, Jun 29, 2021 at 8:58 AM Eric Dumazet <edumazet@google.com> wrote: > > Because the problem only happens with CDG, is adding check in its tcp_cdg_init() function Ok? And about icsk_ca_initialized, Could I expect it to be 0 in CC's init functions? > > I think icsk_ca_initialized lost its strong meaning when CDG was > introduced (since this is the only CC allocating memory) > > The bug really is that before clearing icsk_ca_initialized we should > call cc->release() > > Maybe we missed this cleanup in commit > 8919a9b31eb4fb4c0a93e5fb350a626924302aa6 ("tcp: Only init congestion > control if not initialized already")
On 6/29/2021 11:59 PM, Neal Cardwell wrote: > On Tue, Jun 29, 2021 at 8:58 AM Eric Dumazet <edumazet@google.com> wrote: > From my perspective, the bug was introduced when that 8919a9b31eb4 > commit introduced icsk_ca_initialized and set icsk_ca_initialized to 0 > in tcp_init_transfer(), missing the possibility that a process could > call setsockopt(TCP_CONGESTION) in state TCP_SYN_SENT (i.e. after the > connect() or TFO open sendmsg()), which would call > tcp_init_congestion_control(). The 8919a9b31eb4 commit did not intend > to reset any initialization that the user had already explicitly made; > it just missed the possibility of that particular sequence (which > syzkaller managed to find!). > >> Although I am not sure what happens at accept() time when the listener >> socket is cloned. > > It seems that for listener sockets, they cannot initialize their CC > module state, because there is no way for them to reach > tcp_init_congestion_control(), since: > > (a) tcp_set_congestion_control() -> tcp_reinit_congestion_control() > will not call tcp_init_congestion_control() on a socket in CLOSE or > LISTEN > > (b) tcp_init_transfer() -> tcp_init_congestion_control() can only > happen for established sockets and successful TFO SYN_RECV sockets Is this what was mentioned in this commit ce69e563b325(tcp: make sure listeners don't initialize congestion-control state) > -- > [PATCH] tcp: fix tcp_init_transfer() to not reset icsk_ca_initialized > > This commit fixes a bug (found by syzkaller) that could cause spurious > double-initializations for congestion control modules, which could cause memory > leaks orother problems for congestion control modules (like CDG) that allocate > memory in their init functions. > > The buggy scenario constructed by syzkaller was something like: > > (1) create a TCP socket > (2) initiate a TFO connect via sendto() > (3) while socket is in TCP_SYN_SENT, call setsockopt(TCP_CONGESTION), > which calls: > tcp_set_congestion_control() -> > tcp_reinit_congestion_control() -> > tcp_init_congestion_control() > (4) receive ACK, connection is established, call tcp_init_transfer(), > set icsk_ca_initialized=0 (without first calling cc->release()), > call tcp_init_congestion_control() again. > > Note that in this sequence tcp_init_congestion_control() is called twice > without a cc->release() call in between. Thus, for CC modules that allocate > memory in their init() function, e.g, CDG, a memory leak may occur. The > syzkaller tool managed to find a reproducer that triggered such a leak in CDG. > > The bug was introduced when that 8919a9b31eb4 commit introduced > icsk_ca_initialized and set icsk_ca_initialized to 0 in tcp_init_transfer(), > missing the possibility for a sequence like the one above, where a process > could call setsockopt(TCP_CONGESTION) in state TCP_SYN_SENT (i.e. after the > connect() or TFO open sendmsg()), which would call > tcp_init_congestion_control(). The 8919a9b31eb4 commit did not intend to reset > any initialization that the user had already explicitly made; it just missed > the possibility of that particular sequence (which syzkaller managed to find). Could I use your commit message when I resubmit patch? Thank you.
On Wed, Jun 30, 2021 at 2:25 PM Phi Nguyen <phind.uet@gmail.com> wrote: > > On 6/29/2021 11:59 PM, Neal Cardwell wrote: > > On Tue, Jun 29, 2021 at 8:58 AM Eric Dumazet <edumazet@google.com> wrote: > > From my perspective, the bug was introduced when that 8919a9b31eb4 > > commit introduced icsk_ca_initialized and set icsk_ca_initialized to 0 > > in tcp_init_transfer(), missing the possibility that a process could > > call setsockopt(TCP_CONGESTION) in state TCP_SYN_SENT (i.e. after the > > connect() or TFO open sendmsg()), which would call > > tcp_init_congestion_control(). The 8919a9b31eb4 commit did not intend > > to reset any initialization that the user had already explicitly made; > > it just missed the possibility of that particular sequence (which > > syzkaller managed to find!). > > > >> Although I am not sure what happens at accept() time when the listener > >> socket is cloned. > > > > It seems that for listener sockets, they cannot initialize their CC > > module state, because there is no way for them to reach > > tcp_init_congestion_control(), since: > > > > (a) tcp_set_congestion_control() -> tcp_reinit_congestion_control() > > will not call tcp_init_congestion_control() on a socket in CLOSE or > > LISTEN > > > > (b) tcp_init_transfer() -> tcp_init_congestion_control() can only > > happen for established sockets and successful TFO SYN_RECV sockets > Is this what was mentioned in this commit ce69e563b325(tcp: make sure > listeners don't initialize congestion-control state) Yes, exactly. > > -- > > [PATCH] tcp: fix tcp_init_transfer() to not reset icsk_ca_initialized > > > > This commit fixes a bug (found by syzkaller) that could cause spurious > > double-initializations for congestion control modules, which could cause memory > > leaks orother problems for congestion control modules (like CDG) that allocate > > memory in their init functions. > > > > The buggy scenario constructed by syzkaller was something like: > > > > (1) create a TCP socket > > (2) initiate a TFO connect via sendto() > > (3) while socket is in TCP_SYN_SENT, call setsockopt(TCP_CONGESTION), > > which calls: > > tcp_set_congestion_control() -> > > tcp_reinit_congestion_control() -> > > tcp_init_congestion_control() > > (4) receive ACK, connection is established, call tcp_init_transfer(), > > set icsk_ca_initialized=0 (without first calling cc->release()), > > call tcp_init_congestion_control() again. > > > > Note that in this sequence tcp_init_congestion_control() is called twice > > without a cc->release() call in between. Thus, for CC modules that allocate > > memory in their init() function, e.g, CDG, a memory leak may occur. The > > syzkaller tool managed to find a reproducer that triggered such a leak in CDG. > > > > The bug was introduced when that 8919a9b31eb4 commit introduced > > icsk_ca_initialized and set icsk_ca_initialized to 0 in tcp_init_transfer(), > > missing the possibility for a sequence like the one above, where a process > > could call setsockopt(TCP_CONGESTION) in state TCP_SYN_SENT (i.e. after the > > connect() or TFO open sendmsg()), which would call > > tcp_init_congestion_control(). The 8919a9b31eb4 commit did not intend to reset > > any initialization that the user had already explicitly made; it just missed > > the possibility of that particular sequence (which syzkaller managed to find). > > Could I use your commit message when I resubmit patch? Yes, feel free to use that commit message verbatim or modified. thanks, neal
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 7d5e59f688de..855ada2be25e 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -5922,7 +5922,6 @@ void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb) tp->snd_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk)); tp->snd_cwnd_stamp = tcp_jiffies32; - icsk->icsk_ca_initialized = 0; bpf_skops_established(sk, bpf_op, skb); if (!icsk->icsk_ca_initialized) tcp_init_congestion_control(sk);
icsk_ca_initialized be always set to zero before we examine it in if block, this makes the congestion control module's initialization be called even if the CC module was initialized already. In case the CC module allocates and setups its dynamically allocated private data in its init() function, e.g, CDG, the memory leak may occur. Reported-by: syzbot+f1e24a0594d4e3a895d3@syzkaller.appspotmail.com Signed-off-by: Nguyen Dinh Phi <phind.uet@gmail.com> --- net/ipv4/tcp_input.c | 1 - 1 file changed, 1 deletion(-) -- 2.25.1