From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from chinatelecom.cn (prt-mail.chinatelecom.cn [42.123.76.228]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 3E6D272 for ; Mon, 12 Jul 2021 09:14:46 +0000 (UTC) HMM_SOURCE_IP:172.18.0.48:43174.539687320 HMM_ATTACHE_NUM:0000 HMM_SOURCE_TYPE:SMTP Received: from clientip-36.111.140.26?logid-a97f149ddbdd45d9b7fcd70e0d914cc9 (unknown [172.18.0.48]) by chinatelecom.cn (HERMES) with SMTP id 22A21280117; Mon, 12 Jul 2021 17:14:19 +0800 (CST) X-189-SAVE-TO-SEND: liyonglong@chinatelecom.cn Received: from ([172.18.0.48]) by app0024 with ESMTP id a97f149ddbdd45d9b7fcd70e0d914cc9 for mptcp@lists.linux.dev; Mon Jul 12 17:14:38 2021 X-Transaction-ID: a97f149ddbdd45d9b7fcd70e0d914cc9 X-filter-score: X-Real-From: liyonglong@chinatelecom.cn X-Receive-IP: 172.18.0.48 X-MEDUSA-Status: 0 Sender: liyonglong@chinatelecom.cn Subject: Re: [MPTCP][PATCH mptcp-next] Squash to "mptcp: build ADD_ADDR/echo-ADD_ADDR option according pm.add_signal" To: Geliang Tang Cc: mptcp@lists.linux.dev References: <9365b79b245b8a87af18ca458c67820d47de2515.1626016228.git.geliangtang@gmail.com> <80fa33a249c2ecc7edd9d0047dd84f163307cee7.1626016292.git.geliangtang@gmail.com> <347f6214-cc6b-2af2-c1e7-d9ac7f77f87e@chinatelecom.cn> <4f790fa7-f99e-401e-d266-8fd85921204a@chinatelecom.cn> From: Yonglong Li Message-ID: <3f3c3065-8a08-fc0c-da8b-0e210097b136@chinatelecom.cn> Date: Mon, 12 Jul 2021 17:14:14 +0800 User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101 Thunderbird/78.9.0 Precedence: bulk X-Mailing-List: mptcp@lists.linux.dev List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 8bit On 2021/7/12 16:44, Geliang Tang wrote: > Yonglong Li 于2021年7月12日周一 下午4:07写道： >> >> >> >> On 2021/7/12 15:33, Geliang Tang wrote: >>> Hi Yonglong, >>> >>> Yonglong Li 于2021年7月12日周一 上午9:34写道： >>>> >>>> >>>> >>>> On 2021/7/11 23:15, Geliang Tang wrote: >>>>> I think there're still some issues in v8: >>>>> >>>>> The remaining value is incorrect since "remaining += opt_size;" in the >>>>> "drop other suboptions" checks has been called twice in >>>>> mptcp_pm_add_addr_signal and mptcp_established_options_add_addr. >>>>> >>>> I think "remaining" in mptcp_pm_add_addr_signal does not touch "remaining" in >>>> mptcp_established_options_add_addr. >>>> >>>>> opts->local and opts->remote in mptcp_pm_add_addr_signal need be >>>>> populate after the length chech, not before the check.] >>>>> >>>>> The squash-to patch keeped the more orignal code unchanged, and just do >>>>> the least, necessary modifications. >>>>> >>>> Agree opts->local and opts->remote should be asigned after the length check. >>>> But if keep the length check out of mptcp_pm_add_addr_signal (out of pm lock ) >>>> as orignal code, there is a race that: >>>> >>>> ==> a add addr event (pm.addr_signal == MPTCP_ADD_ADDR_SIGNAL) >>>> ==> call mptcp_pm_add_addr_signal >>>> ==> a echo add addr event trigger (pm.addr_signal == MPTCP_ADD_ADDR_ECHO|MPTCP_ADD_ADDR_SIGNAL) >>>> ==> at this time opts->remote is empty and the length is incorrect. >>>> >>> >>> What will happen in v8 when this race occurs? How dose v8 deal with the >>> race? >> Hi Geliang, thinks for your patience. >> >> I think v8 doesn't have this issue: >> ==> a add addr event (pm.addr_signal == MPTCP_ADD_ADDR_SIGNAL) >> ==> call mptcp_pm_add_addr_signal, save pm.addr_signal to add_addr and save addr in opts under pm.lock >> ==> a echo add addr event trigger (pm.addr_signal == MPTCP_ADD_ADDR_ECHO), but add_addr doesn't changed. >> ==> use add_addr and opts to check length. >> ==> next send ack process will deal with MPTCP_ADD_ADDR_ECHO event. > > Thanks for your explanation. > > I think this squash-to patch did the same thing: > > ==> an add addr event (pm.addr_signal == MPTCP_ADD_ADDR_SIGNAL) > ==> call mptcp_pm_add_addr_signal, save echo bit of pm.addr_signal to > 'echo' (echo = false), save the port number to 'port', and save addr > in opts under pm.lock > ==> an echo add addr event trigger (pm.addr_signal == > MPTCP_ADD_ADDR_ECHO), but 'echo' and 'port' don't changed. > ==> use 'echo' to get the address family, use 'family', 'echo' and > 'port' to check length. > ==> next send ack process will deal with MPTCP_ADD_ADDR_ECHO event. > > Do you think so? yep. In this case the squash-to patch is ok. But I think between "drop other suboptions" checks and mptcp_pm_add_addr_signal the race still exist. ==> an add addr event (pm.addr_signal == MPTCP_ADD_ADDR_SIGNAL) ==> "drop other suboptions" checks will use MPTCP_ADD_ADDR_SIGNAL to check ==> an echo add addr event trigger (pm.addr_signal == MPTCP_ADD_ADDR_SIGNAL | MPTCP_ADD_ADDR_ECHO ) ==> call mptcp_pm_add_addr_signal, MPTCP_ADD_ADDR_ECHO will be clear in pm.addr_signal ==> process MPTCP_ADD_ADDR_ECHO event. WDYT? > >> >>> >>>> So I think the orignal code is incorrect. WDYT? >>>> >>>>> Drop the "drop other suboptions" check in mptcp_pm_add_addr_signal. >>>>> >>>>> Change arguments of mptcp_pm_add_addr_signal. >>>>> >>>>> Keep mptcp_add_addr_len unchanged. >>>>> >>>>> Signed-off-by: Geliang Tang >>>>> --- >>>>> net/mptcp/options.c | 35 +++++++++++++++++------------------ >>>>> net/mptcp/pm.c | 23 +++++++++-------------- >>>>> net/mptcp/protocol.h | 27 +++++++++------------------ >>>>> 3 files changed, 35 insertions(+), 50 deletions(-) >>>>> >>>>> diff --git a/net/mptcp/options.c b/net/mptcp/options.c >>>>> index 5c0ad9b90866..93ad7b134f74 100644 >>>>> --- a/net/mptcp/options.c >>>>> +++ b/net/mptcp/options.c >>>>> @@ -663,16 +663,14 @@ static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff * >>>>> struct mptcp_sock *msk = mptcp_sk(subflow->conn); >>>>> bool drop_other_suboptions = false; >>>>> unsigned int opt_size = *size; >>>>> - u8 add_addr; >>>>> + bool echo; >>>>> + bool port; >>>>> + u8 family; >>>>> int len; >>>>> >>>>> - if (!mptcp_pm_should_add_signal(msk) || >>>>> - !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, opts, &add_addr)) >>>>> - return false; >>>>> - >>>>> - if (((add_addr & BIT(MPTCP_ADD_ADDR_ECHO)) || >>>>> - ((add_addr & BIT(MPTCP_ADD_ADDR_SIGNAL)) && >>>>> - (opts->local.family == AF_INET6 || opts->local.port))) && >>>>> + if ((mptcp_pm_should_add_signal_echo(msk) || >>>>> + (mptcp_pm_should_add_signal_addr(msk) && >>>>> + (msk->pm.local.family == AF_INET6 || msk->pm.local.port))) && >>>>> skb && skb_is_tcp_pure_ack(skb)) { >>>>> pr_debug("drop other suboptions"); >>>>> opts->suboptions = 0; >>>>> @@ -682,7 +680,12 @@ static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff * >>>>> drop_other_suboptions = true; >>>>> } >>>>> >>>>> - len = mptcp_add_addr_len(opts, add_addr); >>>>> + if (!mptcp_pm_should_add_signal(msk) || >>>>> + !mptcp_pm_add_addr_signal(msk, remaining, &opts->local, &opts->remote, &echo, &port)) >>>>> + return false; >>>>> + >>>>> + family = echo ? opts->remote.family : opts->local.family; >>>>> + len = mptcp_add_addr_len(family, echo, port); >>>>> if (remaining < len) >>>>> return false; >>>>> >>>>> @@ -690,15 +693,14 @@ static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff * >>>>> if (drop_other_suboptions) >>>>> *size -= opt_size; >>>>> opts->suboptions |= OPTION_MPTCP_ADD_ADDR; >>>>> - if (!(add_addr & BIT(MPTCP_ADD_ADDR_ECHO)) && >>>>> - (add_addr & BIT(MPTCP_ADD_ADDR_SIGNAL))) { >>>>> + if (!echo) { >>>>> opts->ahmac = add_addr_generate_hmac(msk->local_key, >>>>> msk->remote_key, >>>>> &opts->local); >>>>> } >>>>> - pr_debug("addr_signal:%x, echo=%d, local_addr_id=%d, ahmac=%llu, local_port=%d, remote_addr_id=%d, remote_port=%d", >>>>> - add_addr, (opts->ahmac == 0), opts->local.id, opts->ahmac, >>>>> - ntohs(opts->local.port), opts->remote.id, ntohs(opts->remote.port)); >>>>> + pr_debug("local_id=%d, local_port=%d, remote_id=%d, remote_port=%d, ahmac=%llu, echo=%d", >>>>> + opts->local.id, ntohs(opts->local.port), opts->remote.id, >>>>> + ntohs(opts->remote.port), opts->ahmac, echo); >>>>> >>>>> return true; >>>>> } >>>>> @@ -1253,13 +1255,10 @@ void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, >>>>> >>>>> mp_capable_done: >>>>> if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { >>>>> - struct mptcp_addr_info *addr = &opts->remote; >>>>> + struct mptcp_addr_info *addr = opts->ahmac ? &opts->local : &opts->remote; >>>>> u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; >>>>> u8 echo = MPTCP_ADDR_ECHO; >>>>> >>>>> - if (opts->ahmac) >>>>> - addr = &opts->local; >>>>> - >>>>> #if IS_ENABLED(CONFIG_MPTCP_IPV6) >>>>> if (addr->family == AF_INET6) >>>>> len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; >>>>> diff --git a/net/mptcp/pm.c b/net/mptcp/pm.c >>>>> index 264f522af530..399b59cb7563 100644 >>>>> --- a/net/mptcp/pm.c >>>>> +++ b/net/mptcp/pm.c >>>>> @@ -253,12 +253,13 @@ void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup) >>>>> >>>>> /* path manager helpers */ >>>>> >>>>> -bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, struct sk_buff *skb, >>>>> - unsigned int opt_size, unsigned int remaining, >>>>> - struct mptcp_out_options *opts, u8 *add_addr) >>>>> +bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, unsigned int remaining, >>>>> + struct mptcp_addr_info *saddr, struct mptcp_addr_info *daddr, >>>>> + bool *echo, bool *port) >>>>> { >>>>> int ret = false; >>>>> u8 add_addr; >>>>> + u8 family; >>>>> >>>>> spin_lock_bh(&msk->pm.lock); >>>>> >>>>> @@ -266,21 +267,15 @@ bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, struct sk_buff *skb, >>>>> if (!mptcp_pm_should_add_signal(msk)) >>>>> goto out_unlock; >>>>> >>>>> - opts->local = msk->pm.local; >>>>> - opts->remote = msk->pm.remote; >>>>> - *add_addr = msk->pm.addr_signal; >>>>> + *echo = mptcp_pm_should_add_signal_echo(msk); >>>>> + *port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port); >>>>> >>>>> - if (((msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_ECHO)) || >>>>> - ((msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) && >>>>> - (msk->pm.local.family == AF_INET6 || msk->pm.local.port))) && >>>>> - skb && skb_is_tcp_pure_ack(skb)) { >>>>> - remaining += opt_size; >>>>> - } >>>>> - >>>>> - if (remaining < mptcp_add_addr_len(opts, *add_addr)) >>>>> + family = *echo ? msk->pm.remote.family : msk->pm.local.family; >>>>> + if (remaining < mptcp_add_addr_len(family, *echo, *port)) >>>>> goto out_unlock; >>>>> >>>>> *saddr = msk->pm.local; >>>>> + *daddr = msk->pm.remote; >>>>> add_addr = READ_ONCE(msk->pm.addr_signal); >>>>> if (mptcp_pm_should_add_signal_echo(msk)) >>>>> add_addr &= ~BIT(MPTCP_ADD_ADDR_ECHO); >>>>> diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h >>>>> index 937e0309e340..4b63cc6079fa 100644 >>>>> --- a/net/mptcp/protocol.h >>>>> +++ b/net/mptcp/protocol.h >>>>> @@ -766,25 +766,16 @@ static inline bool mptcp_pm_should_rm_signal(struct mptcp_sock *msk) >>>>> return READ_ONCE(msk->pm.addr_signal) & BIT(MPTCP_RM_ADDR_SIGNAL); >>>>> } >>>>> >>>>> -static inline unsigned int mptcp_add_addr_len(struct mptcp_out_options *opts, >>>>> - u8 add_addr) >>>>> +static inline unsigned int mptcp_add_addr_len(int family, bool echo, bool port) >>>>> { >>>>> - struct mptcp_addr_info *addr = &opts->remote; >>>>> - u8 len = 0; >>>>> + u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; >>>>> >>>>> - if (!(add_addr & BIT(MPTCP_ADD_ADDR_ECHO)) && >>>>> - (add_addr & BIT(MPTCP_ADD_ADDR_SIGNAL))) { >>>>> - addr = &opts->local; >>>>> + if (family == AF_INET6) >>>>> + len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; >>>>> + if (!echo) >>>>> len += MPTCPOPT_THMAC_LEN; >>>>> - } >>>>> - >>>>> - if (addr->family == AF_INET6) >>>>> - len += TCPOLEN_MPTCP_ADD_ADDR6_BASE; >>>>> - else >>>>> - len += TCPOLEN_MPTCP_ADD_ADDR_BASE; >>>>> - >>>>> /* account for 2 trailing 'nop' options */ >>>>> - if (addr->port) >>>>> + if (port) >>>>> len += TCPOLEN_MPTCP_PORT_LEN + TCPOLEN_MPTCP_PORT_ALIGN; >>>>> >>>>> return len; >>>>> @@ -798,9 +789,9 @@ static inline int mptcp_rm_addr_len(const struct mptcp_rm_list *rm_list) >>>>> return TCPOLEN_MPTCP_RM_ADDR_BASE + roundup(rm_list->nr - 1, 4) + 1; >>>>> } >>>>> >>>>> -bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, struct sk_buff *skb, >>>>> - unsigned int opt_size, unsigned int remaining, >>>>> - struct mptcp_out_options *opts, u8 *add_addr); >>>>> +bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, unsigned int remaining, >>>>> + struct mptcp_addr_info *saddr, struct mptcp_addr_info *daddr, >>>>> + bool *echo, bool *port); >>>>> bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, >>>>> struct mptcp_rm_list *rm_list); >>>>> int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc); >>>>> >>>> >>>> -- >>>> Li YongLong >>> >> >> -- >> Li YongLong > > -- Li YongLong