| Message ID | 20210524121220.1577321-3-maximmi@nvidia.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Fix use-after-free after the TLS device goes down and up | expand |
On 2021-05-24 19:15, Jakub Kicinski wrote: > On Mon, 24 May 2021 15:12:20 +0300 Maxim Mikityanskiy wrote: >> @@ -1290,6 +1304,26 @@ static int tls_device_down(struct net_device *netdev) >> spin_unlock_irqrestore(&tls_device_lock, flags); >> >> list_for_each_entry_safe(ctx, tmp, &list, list) { >> + /* Stop offloaded TX and switch to the fallback. >> + * tls_is_sk_tx_device_offloaded will return false. >> + */ >> + WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw); >> + >> + /* Stop the RX and TX resync. >> + * tls_dev_resync must not be called after tls_dev_del. >> + */ >> + WRITE_ONCE(ctx->netdev, NULL); >> + >> + /* Start skipping the RX resync logic completely. */ >> + set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags); >> + >> + /* Sync with inflight packets. After this point: >> + * TX: no non-encrypted packets will be passed to the driver. >> + * RX: resync requests from the driver will be ignored. >> + */ >> + synchronize_net(); >> + >> + /* Release the offload context on the driver side. */ >> if (ctx->tx_conf == TLS_HW) >> netdev->tlsdev_ops->tls_dev_del(netdev, ctx, >> TLS_OFFLOAD_CTX_DIR_TX); > > Can we have the Rx resync take the device_offload_lock for read instead? > Like Tx already does? I believe you previously made this attempt in commit 38030d7cb779 ("net/tls: avoid NULL-deref on resync during device removal"), and this approach turned out to be problematic, as explained in commit e52972c11d6b ("net/tls: replace the sleeping lock around RX resync with a bit lock"): "RX resync may get called from soft IRQ, so we can't use the rwsem". > >> +EXPORT_SYMBOL_GPL(tls_validate_xmit_skb_sw); > > Why the export? Because tls_validate_xmit_skb was also exported. Now I see it's needed for tls_validate_xmit_skb, because tls_is_sk_tx_device_offloaded needs its address and can be called from the drivers. There is no similar public function for tls_validate_xmit_skb_sw, so you are probably right that we don't need to export it. >
On Tue, 25 May 2021 11:52:31 +0300 Maxim Mikityanskiy wrote: > > Can we have the Rx resync take the device_offload_lock for read instead? > > Like Tx already does? > > I believe you previously made this attempt in commit 38030d7cb779 > ("net/tls: avoid NULL-deref on resync during device removal"), and this > approach turned out to be problematic, as explained in commit > e52972c11d6b ("net/tls: replace the sleeping lock around RX resync with > a bit lock"): "RX resync may get called from soft IRQ, so we can't use > the rwsem". If only my memory wasn't this shit.. :) Let me take a look at the patch again.
On Mon, 24 May 2021 15:12:20 +0300 Maxim Mikityanskiy wrote: > When a netdev with active TLS offload goes down, tls_device_down is > called to stop the offload and tear down the TLS context. However, the > socket stays alive, and it still points to the TLS context, which is now > deallocated. If a netdev goes up, while the connection is still active, > and the data flow resumes after a number of TCP retransmissions, it will > lead to a use-after-free of the TLS context. > > This commit addresses this bug by keeping the context alive until its > normal destruction, and implements the necessary fallbacks, so that the > connection can resume in software (non-offloaded) kTLS mode. > > On the TX side tls_sw_fallback is used to encrypt all packets. The RX > side already has all the necessary fallbacks, because receiving > non-decrypted packets is supported. The thing needed on the RX side is > to block resync requests, which are normally produced after receiving > non-decrypted packets. > > The necessary synchronization is implemented for a graceful teardown: > first the fallbacks are deployed, then the driver resources are released > (it used to be possible to have a tls_dev_resync after tls_dev_del). > > A new flag called TLS_RX_DEV_DEGRADED is added to indicate the fallback > mode. It's used to skip the RX resync logic completely, as it becomes > useless, and some objects may be released (for example, resync_async, > which is allocated and freed by the driver). > > Fixes: e8f69799810c ("net/tls: Add generic NIC offload infrastructure") > Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com> > Reviewed-by: Tariq Toukan <tariqt@nvidia.com> > @@ -961,6 +964,17 @@ int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx, > > ctx->sw.decrypted |= is_decrypted; > > + if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) { Why not put the check in tls_device_core_ctrl_rx_resync()? Would be less code, right? > + if (likely(is_encrypted || is_decrypted)) > + return 0; > + > + /* After tls_device_down disables the offload, the next SKB will > + * likely have initial fragments decrypted, and final ones not > + * decrypted. We need to reencrypt that single SKB. > + */ > + return tls_device_reencrypt(sk, skb); > + } > + > /* Return immediately if the record is either entirely plaintext or > * entirely ciphertext. Otherwise handle reencrypt partially decrypted > * record.
On 2021-05-25 20:39, Jakub Kicinski wrote: > On Mon, 24 May 2021 15:12:20 +0300 Maxim Mikityanskiy wrote: >> When a netdev with active TLS offload goes down, tls_device_down is >> called to stop the offload and tear down the TLS context. However, the >> socket stays alive, and it still points to the TLS context, which is now >> deallocated. If a netdev goes up, while the connection is still active, >> and the data flow resumes after a number of TCP retransmissions, it will >> lead to a use-after-free of the TLS context. >> >> This commit addresses this bug by keeping the context alive until its >> normal destruction, and implements the necessary fallbacks, so that the >> connection can resume in software (non-offloaded) kTLS mode. >> >> On the TX side tls_sw_fallback is used to encrypt all packets. The RX >> side already has all the necessary fallbacks, because receiving >> non-decrypted packets is supported. The thing needed on the RX side is >> to block resync requests, which are normally produced after receiving >> non-decrypted packets. >> >> The necessary synchronization is implemented for a graceful teardown: >> first the fallbacks are deployed, then the driver resources are released >> (it used to be possible to have a tls_dev_resync after tls_dev_del). >> >> A new flag called TLS_RX_DEV_DEGRADED is added to indicate the fallback >> mode. It's used to skip the RX resync logic completely, as it becomes >> useless, and some objects may be released (for example, resync_async, >> which is allocated and freed by the driver). >> >> Fixes: e8f69799810c ("net/tls: Add generic NIC offload infrastructure") >> Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com> >> Reviewed-by: Tariq Toukan <tariqt@nvidia.com> > >> @@ -961,6 +964,17 @@ int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx, >> >> ctx->sw.decrypted |= is_decrypted; >> >> + if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) { > > Why not put the check in tls_device_core_ctrl_rx_resync()? > Would be less code, right? I see what you mean, and I considered this option, but I think my option has better readability and is more future-proof. By doing an early return, I skip all code irrelevant to the degraded mode, and even though changing ctx->resync_nh_reset won't have effect in the degraded mode, it will be easier for readers to understand that this part of code is not relevant. Furthermore, if someone decides to add more code to !is_encrypted branches in the future, there is a chance that the degraded mode will be missed from consideration. With the early return there is not problem, but if I follow your suggestion and do the check only under is_encrypted, a future contributor unfamiliar with this "degraded flow" might fail to add that check where it will be needed. This was the reason I implemented it this way. What do you think? >> + if (likely(is_encrypted || is_decrypted)) >> + return 0; >> + >> + /* After tls_device_down disables the offload, the next SKB will >> + * likely have initial fragments decrypted, and final ones not >> + * decrypted. We need to reencrypt that single SKB. >> + */ >> + return tls_device_reencrypt(sk, skb); >> + } >> + >> /* Return immediately if the record is either entirely plaintext or >> * entirely ciphertext. Otherwise handle reencrypt partially decrypted >> * record. > >
On Fri, 28 May 2021 15:40:38 +0300 Maxim Mikityanskiy wrote: > >> @@ -961,6 +964,17 @@ int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx, > >> > >> ctx->sw.decrypted |= is_decrypted; > >> > >> + if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) { > > > > Why not put the check in tls_device_core_ctrl_rx_resync()? > > Would be less code, right? > > I see what you mean, and I considered this option, but I think my option > has better readability and is more future-proof. By doing an early > return, I skip all code irrelevant to the degraded mode, and even though > changing ctx->resync_nh_reset won't have effect in the degraded mode, it > will be easier for readers to understand that this part of code is not > relevant. Furthermore, if someone decides to add more code to > !is_encrypted branches in the future, there is a chance that the > degraded mode will be missed from consideration. With the early return > there is not problem, but if I follow your suggestion and do the check > only under is_encrypted, a future contributor unfamiliar with this > "degraded flow" might fail to add that check where it will be needed. > > This was the reason I implemented it this way. What do you think? In general "someone may miss this in the future" is better served by adding a test case than code duplication. But we don't have infra to fake-offload TLS so I don't feel strongly. You can keep as is if that's your preference.
On 2021-05-28 22:44, Jakub Kicinski wrote: > On Fri, 28 May 2021 15:40:38 +0300 Maxim Mikityanskiy wrote: >>>> @@ -961,6 +964,17 @@ int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx, >>>> >>>> ctx->sw.decrypted |= is_decrypted; >>>> >>>> + if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) { >>> >>> Why not put the check in tls_device_core_ctrl_rx_resync()? >>> Would be less code, right? >> >> I see what you mean, and I considered this option, but I think my option >> has better readability and is more future-proof. By doing an early >> return, I skip all code irrelevant to the degraded mode, and even though >> changing ctx->resync_nh_reset won't have effect in the degraded mode, it >> will be easier for readers to understand that this part of code is not >> relevant. Furthermore, if someone decides to add more code to >> !is_encrypted branches in the future, there is a chance that the >> degraded mode will be missed from consideration. With the early return >> there is not problem, but if I follow your suggestion and do the check >> only under is_encrypted, a future contributor unfamiliar with this >> "degraded flow" might fail to add that check where it will be needed. >> >> This was the reason I implemented it this way. What do you think? > > In general "someone may miss this in the future" is better served by > adding a test case than code duplication. But we don't have infra to > fake-offload TLS so I don't feel strongly. You can keep as is if that's > your preference. Yeah, I agree that we would benefit from having unit tests for such flows. But as we don't have the needed infrastructure, and you are fine with the current implementation, I'll keep it. The only thing I need to fix when resubmitting is the unneeded EXPORT_SYMBOL_GPL, right? Thanks for reviewing. >
On Mon, 31 May 2021 14:09:56 +0300 Maxim Mikityanskiy wrote: > On 2021-05-28 22:44, Jakub Kicinski wrote: > > In general "someone may miss this in the future" is better served by > > adding a test case than code duplication. But we don't have infra to > > fake-offload TLS so I don't feel strongly. You can keep as is if that's > > your preference. > > Yeah, I agree that we would benefit from having unit tests for such > flows. But as we don't have the needed infrastructure, and you are fine > with the current implementation, I'll keep it. The only thing I need to > fix when resubmitting is the unneeded EXPORT_SYMBOL_GPL, right? Yup!
diff --git a/include/net/tls.h b/include/net/tls.h index 6531ace2a68b..8341a8d1e807 100644 --- a/include/net/tls.h +++ b/include/net/tls.h @@ -193,6 +193,11 @@ struct tls_offload_context_tx { (sizeof(struct tls_offload_context_tx) + TLS_DRIVER_STATE_SIZE_TX) enum tls_context_flags { + /* tls_device_down was called after the netdev went down, device state + * was released, and kTLS works in software, even though rx_conf is + * still TLS_HW (needed for transition). + */ + TLS_RX_DEV_DEGRADED = 0, /* Unlike RX where resync is driven entirely by the core in TX only * the driver knows when things went out of sync, so we need the flag * to be atomic. @@ -265,6 +270,7 @@ struct tls_context { /* cache cold stuff */ struct proto *sk_proto; + struct sock *sk; void (*sk_destruct)(struct sock *sk); @@ -447,6 +453,9 @@ static inline u16 tls_user_config(struct tls_context *ctx, bool tx) struct sk_buff * tls_validate_xmit_skb(struct sock *sk, struct net_device *dev, struct sk_buff *skb); +struct sk_buff * +tls_validate_xmit_skb_sw(struct sock *sk, struct net_device *dev, + struct sk_buff *skb); static inline bool tls_is_sk_tx_device_offloaded(struct sock *sk) { diff --git a/net/tls/tls_device.c b/net/tls/tls_device.c index 171752cd6910..bd9f1567aa39 100644 --- a/net/tls/tls_device.c +++ b/net/tls/tls_device.c @@ -50,6 +50,7 @@ static void tls_device_gc_task(struct work_struct *work); static DECLARE_WORK(tls_device_gc_work, tls_device_gc_task); static LIST_HEAD(tls_device_gc_list); static LIST_HEAD(tls_device_list); +static LIST_HEAD(tls_device_down_list); static DEFINE_SPINLOCK(tls_device_lock); static void tls_device_free_ctx(struct tls_context *ctx) @@ -759,6 +760,8 @@ void tls_device_rx_resync_new_rec(struct sock *sk, u32 rcd_len, u32 seq) if (tls_ctx->rx_conf != TLS_HW) return; + if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) + return; prot = &tls_ctx->prot_info; rx_ctx = tls_offload_ctx_rx(tls_ctx); @@ -961,6 +964,17 @@ int tls_device_decrypted(struct sock *sk, struct tls_context *tls_ctx, ctx->sw.decrypted |= is_decrypted; + if (unlikely(test_bit(TLS_RX_DEV_DEGRADED, &tls_ctx->flags))) { + if (likely(is_encrypted || is_decrypted)) + return 0; + + /* After tls_device_down disables the offload, the next SKB will + * likely have initial fragments decrypted, and final ones not + * decrypted. We need to reencrypt that single SKB. + */ + return tls_device_reencrypt(sk, skb); + } + /* Return immediately if the record is either entirely plaintext or * entirely ciphertext. Otherwise handle reencrypt partially decrypted * record. @@ -1290,6 +1304,26 @@ static int tls_device_down(struct net_device *netdev) spin_unlock_irqrestore(&tls_device_lock, flags); list_for_each_entry_safe(ctx, tmp, &list, list) { + /* Stop offloaded TX and switch to the fallback. + * tls_is_sk_tx_device_offloaded will return false. + */ + WRITE_ONCE(ctx->sk->sk_validate_xmit_skb, tls_validate_xmit_skb_sw); + + /* Stop the RX and TX resync. + * tls_dev_resync must not be called after tls_dev_del. + */ + WRITE_ONCE(ctx->netdev, NULL); + + /* Start skipping the RX resync logic completely. */ + set_bit(TLS_RX_DEV_DEGRADED, &ctx->flags); + + /* Sync with inflight packets. After this point: + * TX: no non-encrypted packets will be passed to the driver. + * RX: resync requests from the driver will be ignored. + */ + synchronize_net(); + + /* Release the offload context on the driver side. */ if (ctx->tx_conf == TLS_HW) netdev->tlsdev_ops->tls_dev_del(netdev, ctx, TLS_OFFLOAD_CTX_DIR_TX); @@ -1297,13 +1331,21 @@ static int tls_device_down(struct net_device *netdev) !test_bit(TLS_RX_DEV_CLOSED, &ctx->flags)) netdev->tlsdev_ops->tls_dev_del(netdev, ctx, TLS_OFFLOAD_CTX_DIR_RX); - WRITE_ONCE(ctx->netdev, NULL); - synchronize_net(); + dev_put(netdev); - list_del_init(&ctx->list); - if (refcount_dec_and_test(&ctx->refcount)) - tls_device_free_ctx(ctx); + /* Move the context to a separate list for two reasons: + * 1. When the context is deallocated, list_del is called. + * 2. It's no longer an offloaded context, so we don't want to + * run offload-specific code on this context. + */ + spin_lock_irqsave(&tls_device_lock, flags); + list_move_tail(&ctx->list, &tls_device_down_list); + spin_unlock_irqrestore(&tls_device_lock, flags); + + /* Device contexts for RX and TX will be freed in on sk_destruct + * by tls_device_free_ctx. rx_conf and tx_conf stay in TLS_HW. + */ } up_write(&device_offload_lock); diff --git a/net/tls/tls_device_fallback.c b/net/tls/tls_device_fallback.c index cacf040872c7..a72c89b48a59 100644 --- a/net/tls/tls_device_fallback.c +++ b/net/tls/tls_device_fallback.c @@ -431,6 +431,14 @@ struct sk_buff *tls_validate_xmit_skb(struct sock *sk, } EXPORT_SYMBOL_GPL(tls_validate_xmit_skb); +struct sk_buff *tls_validate_xmit_skb_sw(struct sock *sk, + struct net_device *dev, + struct sk_buff *skb) +{ + return tls_sw_fallback(sk, skb); +} +EXPORT_SYMBOL_GPL(tls_validate_xmit_skb_sw); + struct sk_buff *tls_encrypt_skb(struct sk_buff *skb) { return tls_sw_fallback(skb->sk, skb); diff --git a/net/tls/tls_main.c b/net/tls/tls_main.c index 47b7c5334c34..fde56ff49163 100644 --- a/net/tls/tls_main.c +++ b/net/tls/tls_main.c @@ -636,6 +636,7 @@ struct tls_context *tls_ctx_create(struct sock *sk) mutex_init(&ctx->tx_lock); rcu_assign_pointer(icsk->icsk_ulp_data, ctx); ctx->sk_proto = READ_ONCE(sk->sk_prot); + ctx->sk = sk; return ctx; }
When a netdev with active TLS offload goes down, tls_device_down is called to stop the offload and tear down the TLS context. However, the socket stays alive, and it still points to the TLS context, which is now deallocated. If a netdev goes up, while the connection is still active, and the data flow resumes after a number of TCP retransmissions, it will lead to a use-after-free of the TLS context. This commit addresses this bug by keeping the context alive until its normal destruction, and implements the necessary fallbacks, so that the connection can resume in software (non-offloaded) kTLS mode. On the TX side tls_sw_fallback is used to encrypt all packets. The RX side already has all the necessary fallbacks, because receiving non-decrypted packets is supported. The thing needed on the RX side is to block resync requests, which are normally produced after receiving non-decrypted packets. The necessary synchronization is implemented for a graceful teardown: first the fallbacks are deployed, then the driver resources are released (it used to be possible to have a tls_dev_resync after tls_dev_del). A new flag called TLS_RX_DEV_DEGRADED is added to indicate the fallback mode. It's used to skip the RX resync logic completely, as it becomes useless, and some objects may be released (for example, resync_async, which is allocated and freed by the driver). Fixes: e8f69799810c ("net/tls: Add generic NIC offload infrastructure") Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com> Reviewed-by: Tariq Toukan <tariqt@nvidia.com> --- include/net/tls.h | 9 ++++++ net/tls/tls_device.c | 52 +++++++++++++++++++++++++++++++---- net/tls/tls_device_fallback.c | 8 ++++++ net/tls/tls_main.c | 1 + 4 files changed, 65 insertions(+), 5 deletions(-)