mlx5 fixes 2022-11-24
Subsystems (5)
From: Saeed Mahameed <[email protected]>
This series provides bug fixes to mlx5 driver.
Focusing on error handling and proper memory management in mlx5, in general and in the newly added macsec module.
I still have few fixes left in my queue and I hope those will be the last ones for mlx5 for this cycle.
Please pull and let me know if there is any problem.
Happy thanksgiving.
Thanks,
Saeed.
The following changes since commit cd07eadd5147ffdae11b6fd28b77a3872f2a2484:
octeontx2-pf: Add check for devm_kcalloc (2022-11-24 08:34:45 +0100)
are available in the Git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git tags/mlx5-fixes-2022-11-24
for you to fetch changes up to 9034b29251818ab7b4b38bf6ca860cee45d2726a:
net/mlx5e: MACsec, block offload requests with encrypt off (2022-11-24 00:03:23 -0800)
----------------------------------------------------------------
mlx5-fixes-2022-11-24
----------------------------------------------------------------
Chris Mi (2):
net/mlx5: E-switch, Destroy legacy fdb table when needed net/mlx5: E-switch, Fix duplicate lag creation
Dan Carpenter (1):
net/mlx5e: Fix a couple error codes
Emeel Hakim (3):
net/mlx5e: MACsec, fix add Rx security association (SA) rule memory leak net/mlx5e: MACsec, remove replay window size limitation in offload path net/mlx5e: MACsec, block offload requests with encrypt off
Raed Salem (5):
net/mlx5e: MACsec, fix RX data path 16 RX security channel limit net/mlx5e: MACsec, fix memory leak when MACsec device is deleted net/mlx5e: MACsec, fix update Rx secure channel active field net/mlx5e: MACsec, fix mlx5e_macsec_update_rxsa bail condition and functionality net/mlx5e: MACsec, fix Tx SA active field update
Roi Dayan (1):
net/mlx5e: Fix use-after-free when reverting termination table
YueHaibing (3):
net/mlx5: DR, Fix uninitialized var warning net/mlx5: Fix uninitialized variable bug in outlen_write() net/mlx5e: Use kvfree() in mlx5e_accel_fs_tcp_create()
drivers/net/ethernet/mellanox/mlx5/core/cmd.c | 4 +- .../ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c | 4 +- .../ethernet/mellanox/mlx5/core/en_accel/macsec.c | 138 +++++++++------------ .../ethernet/mellanox/mlx5/core/en_accel/macsec.h | 6 +- .../mellanox/mlx5/core/en_accel/macsec_fs.c | 17 ++- drivers/net/ethernet/mellanox/mlx5/core/eswitch.c | 3 + drivers/net/ethernet/mellanox/mlx5/core/eswitch.h | 8 ++ .../ethernet/mellanox/mlx5/core/eswitch_offloads.c | 7 ++ .../mellanox/mlx5/core/eswitch_offloads_termtbl.c | 2 + drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c | 5 +- .../mellanox/mlx5/core/steering/dr_table.c | 5 +- include/linux/mlx5/mlx5_ifc.h | 7 -- 12 files changed, 105 insertions(+), 101 deletions(-)
PATCH 1/15 net/mlx5: DR, Fix uninitialized var warning
From: YueHaibing <[email protected]>
Smatch warns this:
drivers/net/ethernet/mellanox/mlx5/core/steering/dr_table.c:81
mlx5dr_table_set_miss_action() error: uninitialized symbol 'ret'.
Initializing ret with -EOPNOTSUPP and fix missing action case.
--- drivers/net/ethernet/mellanox/mlx5/core/steering/dr_table.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/steering/dr_table.c +4 -1
@@ -46,7 +46,7 @@ static int dr_table_set_miss_action_nic(struct mlx5dr_domain *dmn,int mlx5dr_table_set_miss_action (struct mlx5dr_table *tbl,struct mlx5dr_action *action) { -int ret; +int ret = -EOPNOTSUPP ;if (action && action->action_type !=DR_ACTION_TYP_FT )return -EOPNOTSUPP ;
@@ -67,6 +67,9 @@ int mlx5dr_table_set_miss_action(struct mlx5dr_table *tbl, goto out; } +if (ret) + goto out; +/* Release old action */ if (tbl->miss_action)refcount_dec (&tbl->miss_action->refcount);
Signature · 1 line
2.38.1
PATCH 2/15 net/mlx5: E-switch, Destroy legacy fdb table when needed
In reply to Saeed MahameedFrom: Chris Mi <[email protected]>
The cited commit removes eswitch mode none. But when disabling sriov in legacy mode or changing from switchdev to legacy mode without sriov enabled, the legacy fdb table is not destroyed.
It is not the right behavior. Destroy legacy fdb table in above two caes.
--- drivers/net/ethernet/mellanox/mlx5/core/eswitch.c | 3 +++ drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c | 7 +++++++ 2 files changed, 10 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/eswitch.c +3 -0
@@ -1362,6 +1362,9 @@ void mlx5_eswitch_disable_sriov(struct mlx5_eswitch *esw, bool clear_vf)devl_rate_nodes_destroy (devlink); } +/* Destroy legacy fdb when disabling sriov in legacy mode. */ +if (esw->mode ==MLX5_ESWITCH_LEGACY ) +mlx5_eswitch_disable_locked (esw); esw->esw_funcs.num_vfs =0 ;
drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c +7 -0
@@ -3387,6 +3387,13 @@ static int esw_offloads_stop(struct mlx5_eswitch *esw,int err; esw->mode =MLX5_ESWITCH_LEGACY ; + +/* If changing from switchdev to legacy mode without sriov enabled, +* no need to create legacy fdb. +*/ +if (!mlx5_sriov_is_enabled (esw->dev)) +return 0 ; + err =mlx5_eswitch_enable_locked (esw,MLX5_ESWITCH_IGNORE_NUM_VFS );if (err)NL_SET_ERR_MSG_MOD (extack,"Failed setting eswitch to legacy" );
Signature · 1 line
2.38.1
PATCH 4/15 net/mlx5: Fix uninitialized variable bug in outlen_write()
In reply to Saeed MahameedFrom: YueHaibing <[email protected]>
If sscanf() return 0, outlen is uninitialized and used in kzalloc(), this is unexpected. We should return -EINVAL if the string is invalid.
--- drivers/net/ethernet/mellanox/mlx5/core/cmd.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/cmd.c +2 -2
@@ -1497,8 +1497,8 @@ static ssize_t outlen_write(struct file *filp, const char __user *buf,return -EFAULT ; err =sscanf (outlen_str,"%d" , &outlen); -if (err <0 ) -return err; +if (err !=1 ) +return -EINVAL ; ptr =kzalloc (outlen,GFP_KERNEL );if (!ptr)
Signature · 1 line
2.38.1
PATCH 3/15 net/mlx5: E-switch, Fix duplicate lag creation
In reply to Saeed MahameedFrom: Chris Mi <[email protected]>
If creating bond first and then enabling sriov in switchdev mode, will hit the following syndrome:
mlx5_core 0000:08:00.0: mlx5_cmd_out_err:778:(pid 25543): CREATE_LAG(0x840) op_mod(0x0) failed, status bad parameter(0x3), syndrome (0x7d49cb), err(-22)
The reason is because the offending patch removes eswitch mode none. In vf lag, the checking of eswitch mode none is replaced by checking if sriov is enabled. But when driver enables sriov, it triggers the bond workqueue task first and then setting sriov number in pci_enable_sriov(). So the check fails.
Fix it by checking if sriov is enabled using eswitch internal counter that is set before triggering the bond workqueue task.
--- drivers/net/ethernet/mellanox/mlx5/core/eswitch.h | 8 ++++++++ drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c | 5 +++-- 2 files changed, 11 insertions(+), 2 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/eswitch.h +8 -0
@@ -736,6 +736,14 @@ void mlx5_eswitch_offloads_destroy_single_fdb(struct mlx5_eswitch *master_esw, struct mlx5_eswitch *slave_esw);int mlx5_eswitch_reload_reps (struct mlx5_eswitch *esw); +static inline int mlx5_eswitch_num_vfs (struct mlx5_eswitch *esw) +{ +if (mlx5_esw_allowed (esw)) +return esw->esw_funcs.num_vfs; + +return 0 ; +} +#else /* CONFIG_MLX5_ESWITCH */ /* eswitch API stubs */ static inline int mlx5_eswitch_init (struct mlx5_core_dev *dev) {return 0 ; }
drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c +3 -2
@@ -701,8 +701,9 @@ static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev)#ifdef CONFIG_MLX5_ESWITCH dev = ldev->pf[MLX5_LAG_P1 ].dev; -if ((mlx5_sriov_is_enabled (dev)) && !is_mdev_switchdev_mode (dev)) -return false; +for (i =0 ; i < ldev->ports; i++) +if (mlx5_eswitch_num_vfs (dev->priv.eswitch) && !is_mdev_switchdev_mode (dev)) +return false; mode =mlx5_eswitch_mode (dev);for (i =0 ; i < ldev->ports; i++)
Signature · 1 line
2.38.1
PATCH 5/15 net/mlx5e: Fix use-after-free when reverting termination table
In reply to Saeed MahameedFrom: Roi Dayan <[email protected]>
When having multiple dests with termination tables and second one or afterwards fails the driver reverts usage of term tables but doesn't reset the assignment in attr->dests[num_vport_dests].termtbl which case a use-after-free when releasing the rule.
Fix by resetting the assignment of termtbl to null.
--- .../net/ethernet/mellanox/mlx5/core/eswitch_offloads_termtbl.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads_termtbl.c +2 -0
@@ -312,6 +312,8 @@ mlx5_eswitch_add_termtbl_rule(struct mlx5_eswitch *esw,for (curr_dest =0 ; curr_dest < num_vport_dests; curr_dest++) {struct mlx5_termtbl_handle *tt = attr->dests[curr_dest].termtbl; + attr->dests[curr_dest].termtbl =NULL ; +/* search for the destination associated with the * current term table */
Signature · 1 line
2.38.1
PATCH 6/15 net/mlx5e: Fix a couple error codes
In reply to Saeed MahameedFrom: Dan Carpenter <[email protected]>
If kvzalloc() fails then return -ENOMEM. Don't return success.
--- .../ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c +8 -4
@@ -250,7 +250,7 @@ static int macsec_fs_tx_create(struct mlx5e_macsec_fs *macsec_fs)struct mlx5_flow_handle *rule;struct mlx5_flow_spec *spec;u32 *flow_group_in; -int err =0 ; +int err; ns =mlx5_get_flow_namespace (macsec_fs->mdev,MLX5_FLOW_NAMESPACE_EGRESS_MACSEC );if (!ns)
@@ -261,8 +261,10 @@ static int macsec_fs_tx_create(struct mlx5e_macsec_fs *macsec_fs)return -ENOMEM ; flow_group_in =kvzalloc (inlen,GFP_KERNEL ); -if (!flow_group_in) +if (!flow_group_in) { + err = -ENOMEM ; goto out_spec; + } tx_tables = &tx_fs->tables; ft_crypto = &tx_tables->ft_crypto;
@@ -898,7 +900,7 @@ static int macsec_fs_rx_create(struct mlx5e_macsec_fs *macsec_fs)struct mlx5_flow_handle *rule;struct mlx5_flow_spec *spec;u32 *flow_group_in; -int err =0 ; +int err; ns =mlx5_get_flow_namespace (macsec_fs->mdev,MLX5_FLOW_NAMESPACE_KERNEL_RX_MACSEC );if (!ns)
@@ -909,8 +911,10 @@ static int macsec_fs_rx_create(struct mlx5e_macsec_fs *macsec_fs)return -ENOMEM ; flow_group_in =kvzalloc (inlen,GFP_KERNEL ); -if (!flow_group_in) +if (!flow_group_in) { + err = -ENOMEM ; goto free_spec; + } rx_tables = &rx_fs->tables; ft_crypto = &rx_tables->ft_crypto;
Signature · 1 line
2.38.1
PATCH 8/15 net/mlx5e: MACsec, fix RX data path 16 RX security channel limit
In reply to Saeed MahameedFrom: Raed Salem <[email protected]>
Currently the data path metadata flow id mask wrongly limits the number of different RX security channels (SC) to 16, whereas in adding RX SC the limit is "2^16 - 1" this cause an overlap in metadata flow id once more than 16 RX SCs is added, this corrupts MACsec RX offloaded flow handling.
Fix by using the correct mask, while at it improve code to use this mask when adding the Rx rule and improve visibility of such errors by adding debug massage.
--- .../net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 11 ++++++++--- .../net/ethernet/mellanox/mlx5/core/en_accel/macsec.h | 6 ++++-- .../ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c | 4 ++-- 3 files changed, 14 insertions(+), 7 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +8 -3
@@ -736,9 +736,14 @@ static int mlx5e_macsec_add_rxsc(struct macsec_context *ctx) sc_xarray_element->rx_sc = rx_sc; err =xa_alloc (&macsec->sc_xarray, &sc_xarray_element->fs_id, sc_xarray_element, -XA_LIMIT (1 ,USHRT_MAX ),GFP_KERNEL ); -if (err) +XA_LIMIT (1 ,MLX5_MACEC_RX_FS_ID_MAX ),GFP_KERNEL ); +if (err) { +if (err == -EBUSY ) +netdev_err (ctx->netdev, +"MACsec offload: unable to create entry for RX SC (%d Rx SCs already allocated)\n" , +MLX5_MACEC_RX_FS_ID_MAX ); goto destroy_sc_xarray_elemenet; + } rx_sc->md_dst =metadata_dst_alloc (0 ,METADATA_MACSEC ,GFP_KERNEL );if (!rx_sc->md_dst) {
@@ -1748,7 +1753,7 @@ void mlx5e_macsec_offload_handle_rx_skb(struct net_device *netdev,if (!macsec)return ; - fs_id =MLX5_MACSEC_METADATA_HANDLE (macsec_meta_data); + fs_id =MLX5_MACSEC_RX_METADAT_HANDLE (macsec_meta_data);rcu_read_lock (); sc_xarray_element =xa_load (&macsec->sc_xarray, fs_id);
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.h +4 -2
@@ -10,9 +10,11 @@#include <net/macsec.h> #include <net/dst_metadata.h> -/* Bit31 - 30: MACsec marker, Bit3-0: MACsec id */ +/* Bit31 - 30: MACsec marker, Bit15-0: MACsec id */ +#define MLX5_MACEC_RX_FS_ID_MAX USHRT_MAX/* Must be power of two */ +#define MLX5_MACSEC_RX_FS_ID_MASK MLX5_MACEC_RX_FS_ID_MAX#define MLX5_MACSEC_METADATA_MARKER (metadata) ((((metadata) >> 30) & 0x3) == 0x1) -#define MLX5_MACSEC_METADATA_HANDLE (metadata) ((metadata) & GENMASK(3, 0)) +#define MLX5_MACSEC_RX_METADAT_HANDLE (metadata) ((metadata) & MLX5_MACSEC_RX_FS_ID_MASK)struct mlx5e_priv ;struct mlx5e_macsec ;
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c +2 -2
@@ -1146,10 +1146,10 @@ macsec_fs_rx_add_rule(struct mlx5e_macsec_fs *macsec_fs, ft_crypto = &rx_tables->ft_crypto;/* Set bit[31 - 30] macsec marker - 0x01 */ -/* Set bit[3-0] fs id */ +/* Set bit[15-0] fs id */ MLX5_SET (set_action_in, action, action_type,MLX5_ACTION_TYPE_SET );MLX5_SET (set_action_in, action, field,MLX5_ACTION_IN_FIELD_METADATA_REG_B ); -MLX5_SET (set_action_in, action, data, fs_id |BIT (30 )); +MLX5_SET (set_action_in, action, data,MLX5_MACSEC_RX_METADAT_HANDLE (fs_id) |BIT (30 ));MLX5_SET (set_action_in, action, offset,0 );MLX5_SET (set_action_in, action, length,32 );
Signature · 1 line
2.38.1
PATCH 7/15 net/mlx5e: Use kvfree() in mlx5e_accel_fs_tcp_create()
In reply to Saeed MahameedFrom: YueHaibing <[email protected]>
'accel_tcp' is allocated by kvzalloc(), which should freed by kvfree().
--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/fs_tcp.c +2 -2
@@ -365,7 +365,7 @@ void mlx5e_accel_fs_tcp_destroy(struct mlx5e_flow_steering *fs)for (i =0 ; i <ACCEL_FS_TCP_NUM_TYPES ; i++)accel_fs_tcp_destroy_table (fs, i); -kfree (accel_tcp); +kvfree (accel_tcp);mlx5e_fs_set_accel_tcp (fs,NULL ); }
@@ -397,7 +397,7 @@ int mlx5e_accel_fs_tcp_create(struct mlx5e_flow_steering *fs) err_destroy_tables:while (--i >=0 )accel_fs_tcp_destroy_table (fs, i); -kfree (accel_tcp); +kvfree (accel_tcp);mlx5e_fs_set_accel_tcp (fs,NULL );return err; }
Signature · 1 line
2.38.1
PATCH 9/15 net/mlx5e: MACsec, fix memory leak when MACsec device is deleted
In reply to Saeed MahameedFrom: Raed Salem <[email protected]>
When the MACsec netdevice is deleted, all related Rx/Tx HW/SW states should be released/deallocated, however currently part of the Rx security channel association data is not cleaned properly, hence the memory leaks.
Fix by make sure all related Rx Sc resources are cleaned/freed, while at it improve code by grouping release SC context in a function so it can be used in both delete MACsec device and delete Rx SC operations.
--- .../mellanox/mlx5/core/en_accel/macsec.c | 77 ++++++++----------- 1 file changed, 33 insertions(+), 44 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +33 -44
@@ -823,16 +823,43 @@ static int mlx5e_macsec_upd_rxsc(struct macsec_context *ctx)return err; } +static void macsec_del_rxsc_ctx (struct mlx5e_macsec *macsec,struct mlx5e_macsec_rx_sc *rx_sc) +{ +struct mlx5e_macsec_sa *rx_sa; +int i; + +for (i =0 ; i <MACSEC_NUM_AN ; ++i) { + rx_sa = rx_sc->rx_sa[i]; +if (!rx_sa) +continue ; + +mlx5e_macsec_cleanup_sa (macsec, rx_sa, false); +mlx5_destroy_encryption_key (macsec->mdev, rx_sa->enc_key_id); + +kfree (rx_sa); + rx_sc->rx_sa[i] =NULL ; + } + +/* At this point the relevant MACsec offload Rx rule already removed at +* mlx5e_macsec_cleanup_sa need to wait for datapath to finish current +* Rx related data propagating using xa_erase which uses rcu to sync, +* once fs_id is erased then this rx_sc is hidden from datapath. +*/ +list_del_rcu (&rx_sc->rx_sc_list_element); +xa_erase (&macsec->sc_xarray, rx_sc->sc_xarray_element->fs_id); +metadata_dst_free (rx_sc->md_dst); +kfree (rx_sc->sc_xarray_element); +kfree_rcu (rx_sc); +} +static int mlx5e_macsec_del_rxsc (struct macsec_context *ctx) {struct mlx5e_priv *priv =netdev_priv (ctx->netdev);struct mlx5e_macsec_device *macsec_device;struct mlx5e_macsec_rx_sc *rx_sc; -struct mlx5e_macsec_sa *rx_sa;struct mlx5e_macsec *macsec;struct list_head *list;int err =0 ; -int i;mutex_lock (&priv->macsec->lock);
@@ -854,31 +881,7 @@ static int mlx5e_macsec_del_rxsc(struct macsec_context *ctx) goto out; } -for (i =0 ; i <MACSEC_NUM_AN ; ++i) { - rx_sa = rx_sc->rx_sa[i]; -if (!rx_sa) -continue ; - -mlx5e_macsec_cleanup_sa (macsec, rx_sa, false); -mlx5_destroy_encryption_key (macsec->mdev, rx_sa->enc_key_id); - -kfree (rx_sa); - rx_sc->rx_sa[i] =NULL ; - } - -/* -* At this point the relevant MACsec offload Rx rule already removed at -* mlx5e_macsec_cleanup_sa need to wait for datapath to finish current -* Rx related data propagating using xa_erase which uses rcu to sync, -* once fs_id is erased then this rx_sc is hidden from datapath. -*/ -list_del_rcu (&rx_sc->rx_sc_list_element); -xa_erase (&macsec->sc_xarray, rx_sc->sc_xarray_element->fs_id); -metadata_dst_free (rx_sc->md_dst); -kfree (rx_sc->sc_xarray_element); - -kfree_rcu (rx_sc); - +macsec_del_rxsc_ctx (macsec, rx_sc); out:mutex_unlock (&macsec->lock);
@@ -1239,7 +1242,6 @@ static int mlx5e_macsec_del_secy(struct macsec_context *ctx)struct mlx5e_priv *priv =netdev_priv (ctx->netdev);struct mlx5e_macsec_device *macsec_device;struct mlx5e_macsec_rx_sc *rx_sc, *tmp; -struct mlx5e_macsec_sa *rx_sa;struct mlx5e_macsec_sa *tx_sa;struct mlx5e_macsec *macsec;struct list_head *list;
@@ -1268,28 +1270,15 @@ static int mlx5e_macsec_del_secy(struct macsec_context *ctx) } list = &macsec_device->macsec_rx_sc_list_head; -list_for_each_entry_safe (rx_sc, tmp, list, rx_sc_list_element) { -for (i =0 ; i <MACSEC_NUM_AN ; ++i) { - rx_sa = rx_sc->rx_sa[i]; -if (!rx_sa) -continue ; - -mlx5e_macsec_cleanup_sa (macsec, rx_sa, false); -mlx5_destroy_encryption_key (macsec->mdev, rx_sa->enc_key_id); -kfree (rx_sa); - rx_sc->rx_sa[i] =NULL ; - } - -list_del_rcu (&rx_sc->rx_sc_list_element); - -kfree_rcu (rx_sc); - } +list_for_each_entry_safe (rx_sc, tmp, list, rx_sc_list_element) +macsec_del_rxsc_ctx (macsec, rx_sc);kfree (macsec_device->dev_addr); macsec_device->dev_addr =NULL ;list_del_rcu (&macsec_device->macsec_device_list_element); --macsec->num_of_devices; +kfree (macsec_device); out:mutex_unlock (&macsec->lock);
Signature · 1 line
2.38.1
PATCH 10/15 net/mlx5e: MACsec, fix update Rx secure channel active field
In reply to Saeed MahameedFrom: Raed Salem <[email protected]>
The main functionality for this operation is to update the active state of the Rx security channel (SC) if the new active setting is different from the current active state of this Rx SC, however the relevant active state check is done post updating the current active state to match the new active state, effectively blocks any offload state update for the Rx SC in question.
Fix by delay the assignment to be post the relevant check.
--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +1 -1
@@ -803,10 +803,10 @@ static int mlx5e_macsec_upd_rxsc(struct macsec_context *ctx) goto out; } - rx_sc->active = ctx_rx_sc->active;if (rx_sc->active == ctx_rx_sc->active) goto out; + rx_sc->active = ctx_rx_sc->active;for (i =0 ; i <MACSEC_NUM_AN ; ++i) { rx_sa = rx_sc->rx_sa[i];if (!rx_sa)
Signature · 1 line
2.38.1
PATCH 11/15 net/mlx5e: MACsec, fix mlx5e_macsec_update_rxsa bail condition and functionality
In reply to Saeed MahameedFrom: Raed Salem <[email protected]>
Fix update Rx SA wrong bail condition, naturally update functionality needs to check that something changed otherwise bailout currently the active state check does just the opposite, furthermore unlike deactivate path which remove the macsec rules to deactivate the offload, the activation path does not include the counter part installation of the macsec rules.
Fix by using correct bailout condition and when Rx SA changes state to active then add the relevant macsec rules.
While at it, refine function name to reflect more precisely its role.
--- .../mellanox/mlx5/core/en_accel/macsec.c | 24 +++++++++---------- 1 file changed, 11 insertions(+), 13 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +11 -13
@@ -427,15 +427,15 @@ mlx5e_macsec_get_rx_sc_from_sc_list(const struct list_head *list, sci_t sci)return NULL ; } -static int mlx5e_macsec_update_rx_sa (struct mlx5e_macsec *macsec, -struct mlx5e_macsec_sa *rx_sa, -bool active) +static int macsec_rx_sa_active_update (struct macsec_context *ctx, +struct mlx5e_macsec_sa *rx_sa, +bool active) { -struct mlx5_core_dev *mdev = macsec->mdev; -struct mlx5_macsec_obj_attrs attrs = {}; +struct mlx5e_priv *priv =netdev_priv (ctx->netdev); +struct mlx5e_macsec *macsec = priv->macsec;int err =0 ; -if (rx_sa->active != active) +if (rx_sa->active == active)return 0 ; rx_sa->active = active;
@@ -444,13 +444,11 @@ static int mlx5e_macsec_update_rx_sa(struct mlx5e_macsec *macsec,return 0 ; } - attrs.sci =cpu_to_be64 ((__force u64 )rx_sa->sci); - attrs.enc_key_id = rx_sa->enc_key_id; - err =mlx5e_macsec_create_object (mdev, &attrs, false, &rx_sa->macsec_obj_id); + err =mlx5e_macsec_init_sa (ctx, rx_sa, true, false);if (err) -return err; + rx_sa->active = false; -return 0 ; +return err; }static bool mlx5e_macsec_secy_features_validate (struct macsec_context *ctx)
@@ -812,7 +810,7 @@ static int mlx5e_macsec_upd_rxsc(struct macsec_context *ctx)if (!rx_sa)continue ; - err =mlx5e_macsec_update_rx_sa (macsec, rx_sa, rx_sa->active && ctx_rx_sc->active); + err =macsec_rx_sa_active_update (ctx, rx_sa, rx_sa->active && ctx_rx_sc->active);if (err) goto out; }
@@ -1023,7 +1021,7 @@ static int mlx5e_macsec_upd_rxsa(struct macsec_context *ctx) goto out; } - err =mlx5e_macsec_update_rx_sa (macsec, rx_sa, ctx_rx_sa->active); + err =macsec_rx_sa_active_update (ctx, rx_sa, ctx_rx_sa->active); out:mutex_unlock (&macsec->lock);
Signature · 1 line
2.38.1
PATCH 12/15 net/mlx5e: MACsec, fix add Rx security association (SA) rule memory leak
In reply to Saeed MahameedFrom: Emeel Hakim <[email protected]>
Currently MACsec's add Rx SA flow steering (fs) rule routine uses a spec object which is dynamically allocated and do not free it upon leaving. The above led to a memory leak.
Fix by freeing dynamically allocated objects.
--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c | 1 + 1 file changed, 1 insertion(+)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c +1 -0
@@ -1209,6 +1209,7 @@ macsec_fs_rx_add_rule(struct mlx5e_macsec_fs *macsec_fs, rx_rule->rule[1 ] = rule; } +kvfree (spec);return macsec_rule; err:
Signature · 1 line
2.38.1
PATCH 14/15 net/mlx5e: MACsec, fix Tx SA active field update
In reply to Saeed MahameedFrom: Raed Salem <[email protected]>
Currently during update Tx security association (SA) flow, the Tx SA active state is updated only if the Tx SA in question is the same SA that the MACsec interface is using for Tx,in consequence when the MACsec interface chose to work with this Tx SA later, where this SA for example should have been updated to active state and it was not, the relevant Tx SA HW context won't be installed, hence the MACSec flow won't be offloaded.
Fix by update Tx SA active state as part of update flow regardless whether the SA in question is the same Tx SA used by the MACsec interface.
--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +1 -2
@@ -602,6 +602,7 @@ static int mlx5e_macsec_upd_txsa(struct macsec_context *ctx)if (tx_sa->active == ctx_tx_sa->active) goto out; + tx_sa->active = ctx_tx_sa->active;if (tx_sa->assoc_num != tx_sc->encoding_sa) goto out;
@@ -617,8 +618,6 @@ static int mlx5e_macsec_upd_txsa(struct macsec_context *ctx)mlx5e_macsec_cleanup_sa (macsec, tx_sa, true); } - - tx_sa->active = ctx_tx_sa->active; out:mutex_unlock (&macsec->lock);
Signature · 1 line
2.38.1
PATCH 13/15 net/mlx5e: MACsec, remove replay window size limitation in offload path
In reply to Saeed MahameedFrom: Emeel Hakim <[email protected]>
Currently offload path limits replay window size to 32/64/128/256 bits, such a limitation should not exist since software allows it. Remove such limitation.
--- .../mellanox/mlx5/core/en_accel/macsec.c | 16 ---------------- include/linux/mlx5/mlx5_ifc.h | 7 ------- 2 files changed, 23 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0 -16
@@ -229,22 +229,6 @@ static int macsec_set_replay_protection(struct mlx5_macsec_obj_attrs *attrs, voiif (!attrs->replay_protect)return 0 ; -switch (attrs->replay_window) { -case 256 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_256BIT ; -break ; -case 128 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_128BIT ; -break ; -case 64 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_64BIT ; -break ; -case 32 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_32BIT ; -break ; -default : -return -EINVAL ; - }MLX5_SET (macsec_aso, aso_ctx, window_size, window_sz);MLX5_SET (macsec_aso, aso_ctx, mode,MLX5_MACSEC_ASO_REPLAY_PROTECTION );
include/linux/mlx5/mlx5_ifc.h +0 -7
@@ -11611,13 +11611,6 @@ enum { MLX5_MACSEC_ASO_REPLAY_PROTECTION = 0x1, }; -enum { -MLX5_MACSEC_ASO_REPLAY_WIN_32BIT =0x0 , -MLX5_MACSEC_ASO_REPLAY_WIN_64BIT =0x1 , -MLX5_MACSEC_ASO_REPLAY_WIN_128BIT =0x2 , -MLX5_MACSEC_ASO_REPLAY_WIN_256BIT =0x3 , -}; -#define MLX5_MACSEC_ASO_INC_SN 0x2#define MLX5_MACSEC_ASO_REG_C_4_5 0x2
Signature · 1 line
2.38.1
PATCH 15/15 net/mlx5e: MACsec, block offload requests with encrypt off
In reply to Saeed MahameedFrom: Emeel Hakim <[email protected]>
Currently offloading MACsec with authentication only (encrypt property set to off) is not supported, block such requests when adding/updating a macsec device.
--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 5 +++++ 1 file changed, 5 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +5 -0
@@ -458,6 +458,11 @@ static bool mlx5e_macsec_secy_features_validate(struct macsec_context *ctx)return false; } +if (!ctx->secy->tx_sc.encrypt) { +netdev_err (netdev,"MACsec offload: encrypt off isn't supported\n" ); +return false; + } +return true; }
Signature · 1 line
2.38.1
Re: [net 07/15] net/mlx5e: Use kvfree() in mlx5e_accel_fs_tcp_create()
In reply to Saeed MahameedQuoted 9 lines · Saeed Mahameed
From: YueHaibing <[email protected]>
'
accel_tcp' is allocated bykvzalloc(), which should freed bykvfree().Fixes f52f2faee581 ("net/mlx5e: Introduce flow steering API")Signed-off-by YueHaibing <[email protected]>Reviewed-by Tariq Toukan <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>---
Hi Saeed,
There was a v3 of this, that changes the alloc side instead.
Re: [net 07/15] net/mlx5e: Use kvfree() in mlx5e_accel_fs_tcp_create()
Quoted 7 lines · Tariq Toukan
Quoted 4 lines · Saeed Mahameed
From: YueHaibing <[email protected]>
'
accel_tcp' is allocated bykvzalloc(), which should freed bykvfree().
Quoted 9 lines
Quoted 5 lines
Fixes f52f2faee581 ("net/mlx5e: Introduce flow steering API")Signed-off-by YueHaibing <[email protected]>Reviewed-by Tariq Toukan <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>---
Hi Saeed,
There was a v3 of this, that changes the alloc side instead.
Thanks Tariq, that patch was Marked for -next for some reason, and it's in my net-next queue, i think it's ok if this went to net and the other to net-next.
Re: [net 03/15] net/mlx5: E-switch, Fix duplicate lag creation
In reply to Saeed MahameedQuoted 51 lines · Saeed Mahameed
From: Chris Mi <[email protected]>
If creating bond first and then enabling sriov in switchdev mode, will hit the following syndrome:
mlx5_core0000:08:00.0: mlx5_cmd_out_err:778:(pid 25543):CREATE_LAG(0x840)op_mod(0x0)failed, status bad parameter(0x3), syndrome (0x7d49cb), err(-22)The reason is because the offending patch removes eswitch mode none. In vf lag, the checking of eswitch mode none is replaced by checking if sriov is enabled. But when driver enables sriov, it triggers the bond workqueue task first and then setting sriov number in
pci_enable_sriov(). So the check fails.Fix it by checking if sriov is enabled using eswitch internal counter that is set before triggering the bond workqueue task.
Fixes f019679ea5f2 ("net/mlx5: E-switch, Remove dependency between sriov and eswitch mode")Signed-off-by Chris Mi <[email protected]>Reviewed-by Roi Dayan <[email protected]>Reviewed-by Mark Bloch <[email protected]>Reviewed-by Vlad Buslov <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- drivers/net/ethernet/mellanox/mlx5/core/eswitch.h | 8 ++++++++ drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c | 5 +++-- 2 files changed, 11 insertions(+), 2 deletions(-)drivers/net/ethernet/mellanox/mlx5/core/eswitch.h +8
-0@@ -736,6 +736,14 @@ void mlx5_eswitch_offloads_destroy_single_fdb(struct mlx5_eswitch *master_esw, struct mlx5_eswitch *slave_esw);int mlx5_eswitch_reload_reps (struct mlx5_eswitch *esw); +static inline int mlx5_eswitch_num_vfs (struct mlx5_eswitch *esw) +{ +if (mlx5_esw_allowed (esw)) +return esw->esw_funcs.num_vfs; + +return 0 ; +} +#else /* CONFIG_MLX5_ESWITCH */ /* eswitch API stubs */ static inline int mlx5_eswitch_init (struct mlx5_core_dev *dev) {return 0 ; }drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c +0
-0
drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev) · line 707
Quoted 9 lines
drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c +3
-2@@ -701,8 +701,9 @@ static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev)#ifdef CONFIG_MLX5_ESWITCH dev = ldev->pf[MLX5_LAG_P1 ].dev; -if ((mlx5_sriov_is_enabled (dev)) && !is_mdev_switchdev_mode (dev)) -return false; +for (i =0 ; i < ldev->ports; i++) +if (mlx5_eswitch_num_vfs (dev->priv.eswitch) && !is_mdev_switchdev_mode (dev)) +return false;
Am I missing something? whats with the for loop iterator here? i isn't used or passed into these functions?
Do you need to check multiple times or do these functions have some side effect? But looking at their implementation neither of them appear to have side effects?
What am I missing?
Shouldn't this just be:
Quoted 1 line
Review
Quoted 1 line
(quoted diff) +0
-1- if ((mlx5_sriov_is_enabled(dev)) && !is_mdev_switchdev_mode(dev) >> + if (mlx5_eswitch_num_vfs(dev->priv.eswitch) &&
!is_mdev_switchdev_mode(dev))
Quoted 1 line
Quoted 1 line
return false;
Re: [net 05/15] net/mlx5e: Fix use-after-free when reverting termination table
In reply to Saeed MahameedQuoted 8 lines · Saeed Mahameed
From: Roi Dayan <[email protected]>
When having multiple dests with termination tables and second one or afterwards fails the driver reverts usage of term tables but doesn't reset the assignment in attr->dests[num_vport_dests].termtbl which case a use-after-free when releasing the rule.
Fix by resetting the assignment of termtbl to null.
Quoted 21 lines
Fixes 10caabdaad5a ("net/mlx5e: Use termination table for VLAN push actions")Signed-off-by Roi Dayan <[email protected]>Reviewed-by Maor Dickman <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- .../net/ethernet/mellanox/mlx5/core/eswitch_offloads_termtbl.c | 2 ++ 1 file changed, 2 insertions(+)drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads_termtbl.c +2
-0@@ -312,6 +312,8 @@ mlx5_eswitch_add_termtbl_rule(struct mlx5_eswitch *esw,for (curr_dest =0 ; curr_dest < num_vport_dests; curr_dest++) {struct mlx5_termtbl_handle *tt = attr->dests[curr_dest].termtbl; + attr->dests[curr_dest].termtbl =NULL ; +/* search for the destination associated with the * current term table */
Re: [net 06/15] net/mlx5e: Fix a couple error codes
In reply to Saeed MahameedQuoted 4 lines · Saeed Mahameed
From: Dan Carpenter <[email protected]>
If
kvzalloc()fails then return-ENOMEM. Don't return success.
Makes sense.
Quoted 54 lines
Fixes 3b20949cb21b ("net/mlx5e: Add MACsec RX steering rules")Fixes e467b283ffd5 ("net/mlx5e: Add MACsec TX steering rules")Signed-off-by Dan Carpenter <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- .../ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c +8
-4@@ -250,7 +250,7 @@ static int macsec_fs_tx_create(struct mlx5e_macsec_fs *macsec_fs)struct mlx5_flow_handle *rule;struct mlx5_flow_spec *spec;u32 *flow_group_in; -int err =0 ; +int err; ns =mlx5_get_flow_namespace (macsec_fs->mdev,MLX5_FLOW_NAMESPACE_EGRESS_MACSEC );if (!ns)@@ -261,8 +261,10 @@ static int macsec_fs_tx_create(struct mlx5e_macsec_fs *macsec_fs)return -ENOMEM ; flow_group_in =kvzalloc (inlen,GFP_KERNEL ); -if (!flow_group_in) +if (!flow_group_in) { + err = -ENOMEM ; goto out_spec; + } tx_tables = &tx_fs->tables; ft_crypto = &tx_tables->ft_crypto;@@ -898,7 +900,7 @@ static int macsec_fs_rx_create(struct mlx5e_macsec_fs *macsec_fs)struct mlx5_flow_handle *rule;struct mlx5_flow_spec *spec;u32 *flow_group_in; -int err =0 ; +int err; ns =mlx5_get_flow_namespace (macsec_fs->mdev,MLX5_FLOW_NAMESPACE_KERNEL_RX_MACSEC );if (!ns)@@ -909,8 +911,10 @@ static int macsec_fs_rx_create(struct mlx5e_macsec_fs *macsec_fs)return -ENOMEM ; flow_group_in =kvzalloc (inlen,GFP_KERNEL ); -if (!flow_group_in) +if (!flow_group_in) { + err = -ENOMEM ; goto free_spec; + } rx_tables = &rx_fs->tables; ft_crypto = &rx_tables->ft_crypto;
Re: [net 07/15] net/mlx5e: Use kvfree() in mlx5e_accel_fs_tcp_create()
In reply to Saeed MahameedQuoted 14 lines · Saeed Mahameed
Quoted 13 lines · Tariq Toukan
Quoted 9 lines · Saeed Mahameed
From: YueHaibing <[email protected]>
'
accel_tcp' is allocated bykvzalloc(), which should freed bykvfree().Fixes f52f2faee581 ("net/mlx5e: Introduce flow steering API")Signed-off-by YueHaibing <[email protected]>Reviewed-by Tariq Toukan <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>---
Quoted 6 lines
Quoted 3 lines
Hi Saeed,
There was a v3 of this, that changes the alloc side instead.Thanks Tariq, that patch was Marked for -next for some reason, and it's in my net-next queue, i think it's ok if this went to net and the other to net-next.
That route would make sense to me.
Thanks,
Jake
Re: [net 09/15] net/mlx5e: MACsec, fix memory leak when MACsec device is deleted
In reply to Saeed MahameedQuoted 25 lines · Saeed Mahameed
From: Raed Salem <[email protected]>
When the MACsec netdevice is deleted, all related Rx/Tx HW/SW states should be released/deallocated, however currently part of the Rx security channel association data is not cleaned properly, hence the memory leaks.
Fix by make sure all related Rx Sc resources are cleaned/freed, while at it improve code by grouping release SC context in a function so it can be used in both delete MACsec device and delete Rx SC operations.
Fixes 5a39816a75e5 ("net/mlx5e: Add MACsec offload SecY support")Signed-off-by Raed Salem <[email protected]>Reviewed-by Emeel Hakim <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- .../mellanox/mlx5/core/en_accel/macsec.c | 77 ++++++++----------- 1 file changed, 33 insertions(+), 44 deletions(-)drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-0
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c static int mlx5e_macsec_upd_rxsc(struct macsec_context *ctx) · line 825
Quoted 3 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-0@@ -823,16 +823,43 @@ static int mlx5e_macsec_upd_rxsc(struct macsec_context *ctx)return err; }
Quoted 115 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +33
-44+static void macsec_del_rxsc_ctx (struct mlx5e_macsec *macsec,struct mlx5e_macsec_rx_sc *rx_sc) +{ +struct mlx5e_macsec_sa *rx_sa; +int i; + +for (i =0 ; i <MACSEC_NUM_AN ; ++i) { + rx_sa = rx_sc->rx_sa[i]; +if (!rx_sa) +continue ; + +mlx5e_macsec_cleanup_sa (macsec, rx_sa, false); +mlx5_destroy_encryption_key (macsec->mdev, rx_sa->enc_key_id); + +kfree (rx_sa); + rx_sc->rx_sa[i] =NULL ; + } + +/* At this point the relevant MACsec offload Rx rule already removed at +* mlx5e_macsec_cleanup_sa need to wait for datapath to finish current +* Rx related data propagating using xa_erase which uses rcu to sync, +* once fs_id is erased then this rx_sc is hidden from datapath. +*/ +list_del_rcu (&rx_sc->rx_sc_list_element); +xa_erase (&macsec->sc_xarray, rx_sc->sc_xarray_element->fs_id); +metadata_dst_free (rx_sc->md_dst); +kfree (rx_sc->sc_xarray_element); +kfree_rcu (rx_sc); +} +static int mlx5e_macsec_del_rxsc (struct macsec_context *ctx) {struct mlx5e_priv *priv =netdev_priv (ctx->netdev);struct mlx5e_macsec_device *macsec_device;struct mlx5e_macsec_rx_sc *rx_sc; -struct mlx5e_macsec_sa *rx_sa;struct mlx5e_macsec *macsec;struct list_head *list;int err =0 ; -int i;mutex_lock (&priv->macsec->lock);@@ -854,31 +881,7 @@ static int mlx5e_macsec_del_rxsc(struct macsec_context *ctx) goto out; } -for (i =0 ; i <MACSEC_NUM_AN ; ++i) { - rx_sa = rx_sc->rx_sa[i]; -if (!rx_sa) -continue ; - -mlx5e_macsec_cleanup_sa (macsec, rx_sa, false); -mlx5_destroy_encryption_key (macsec->mdev, rx_sa->enc_key_id); - -kfree (rx_sa); - rx_sc->rx_sa[i] =NULL ; - } - -/* -* At this point the relevant MACsec offload Rx rule already removed at -* mlx5e_macsec_cleanup_sa need to wait for datapath to finish current -* Rx related data propagating using xa_erase which uses rcu to sync, -* once fs_id is erased then this rx_sc is hidden from datapath. -*/ -list_del_rcu (&rx_sc->rx_sc_list_element); -xa_erase (&macsec->sc_xarray, rx_sc->sc_xarray_element->fs_id); -metadata_dst_free (rx_sc->md_dst); -kfree (rx_sc->sc_xarray_element); - -kfree_rcu (rx_sc); - +macsec_del_rxsc_ctx (macsec, rx_sc); out:mutex_unlock (&macsec->lock);@@ -1239,7 +1242,6 @@ static int mlx5e_macsec_del_secy(struct macsec_context *ctx)struct mlx5e_priv *priv =netdev_priv (ctx->netdev);struct mlx5e_macsec_device *macsec_device;struct mlx5e_macsec_rx_sc *rx_sc, *tmp; -struct mlx5e_macsec_sa *rx_sa;struct mlx5e_macsec_sa *tx_sa;struct mlx5e_macsec *macsec;struct list_head *list;@@ -1268,28 +1270,15 @@ static int mlx5e_macsec_del_secy(struct macsec_context *ctx) } list = &macsec_device->macsec_rx_sc_list_head; -list_for_each_entry_safe (rx_sc, tmp, list, rx_sc_list_element) { -for (i =0 ; i <MACSEC_NUM_AN ; ++i) { - rx_sa = rx_sc->rx_sa[i]; -if (!rx_sa) -continue ; - -mlx5e_macsec_cleanup_sa (macsec, rx_sa, false); -mlx5_destroy_encryption_key (macsec->mdev, rx_sa->enc_key_id); -kfree (rx_sa); - rx_sc->rx_sa[i] =NULL ; - } - -list_del_rcu (&rx_sc->rx_sc_list_element); - -kfree_rcu (rx_sc); - } +list_for_each_entry_safe (rx_sc, tmp, list, rx_sc_list_element) +macsec_del_rxsc_ctx (macsec, rx_sc);kfree (macsec_device->dev_addr); macsec_device->dev_addr =NULL ;list_del_rcu (&macsec_device->macsec_device_list_element); --macsec->num_of_devices; +kfree (macsec_device); out:mutex_unlock (&macsec->lock);
Re: [net 10/15] net/mlx5e: MACsec, fix update Rx secure channel active field
In reply to Saeed MahameedQuoted 10 lines · Saeed Mahameed
From: Raed Salem <[email protected]>
The main functionality for this operation is to update the active state of the Rx security channel (SC) if the new active setting is different from the current active state of this Rx SC, however the relevant active state check is done post updating the current active state to match the new active state, effectively blocks any offload state update for the Rx SC in question.
Quoted 2 lines
Fix by delay the assignment to be post the relevant check.
Makes sense.
Quoted 24 lines
Fixes aae3454e4d4c ("net/mlx5e: Add MACsec offload Rx command support")Signed-off-by Raed Salem <[email protected]>Reviewed-by Emeel Hakim <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +1
-1@@ -803,10 +803,10 @@ static int mlx5e_macsec_upd_rxsc(struct macsec_context *ctx) goto out; } - rx_sc->active = ctx_rx_sc->active;if (rx_sc->active == ctx_rx_sc->active) goto out; + rx_sc->active = ctx_rx_sc->active;for (i =0 ; i <MACSEC_NUM_AN ; ++i) { rx_sa = rx_sc->rx_sa[i];if (!rx_sa)
Re: [net 12/15] net/mlx5e: MACsec, fix add Rx security association (SA) rule memory leak
In reply to Saeed MahameedQuoted 6 lines · Saeed Mahameed
From: Emeel Hakim <[email protected]>
Currently MACsec's add Rx SA flow steering (fs) rule routine uses a spec object which is dynamically allocated and do not free it upon leaving. The above led to a memory leak.
Quoted 6 lines
Fix by freeing dynamically allocated objects.
Fixes 3b20949cb21b ("net/mlx5e: Add MACsec RX steering rules")Signed-off-by Emeel Hakim <[email protected]>Reviewed-by Raed Salem <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>
Quoted 16 lines
--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c | 1 + 1 file changed, 1 insertion(+)drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec_fs.c +1
-0@@ -1209,6 +1209,7 @@ macsec_fs_rx_add_rule(struct mlx5e_macsec_fs *macsec_fs, rx_rule->rule[1 ] = rule; } +kvfree (spec);return macsec_rule; err:
Re: [net 13/15] net/mlx5e: MACsec, remove replay window size limitation in offload path
In reply to Saeed MahameedQuoted 29 lines · Saeed Mahameed
From: Emeel Hakim <[email protected]>
Currently offload path limits replay window size to 32/64/128/256 bits, such a limitation should not exist since software allows it. Remove such limitation.
Fixes eb43846b43c3 ("net/mlx5e: Support MACsec offload replay window")Signed-off-by Emeel Hakim <[email protected]>Reviewed-by Raed Salem <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- .../mellanox/mlx5/core/en_accel/macsec.c | 16 ---------------- include/linux/mlx5/mlx5_ifc.h | 7 ------- 2 files changed, 23 deletions(-)drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-6@@ -229,22 +229,6 @@ static int macsec_set_replay_protection(struct mlx5_macsec_obj_attrs *attrs, voiif (!attrs->replay_protect)return 0 ; -switch (attrs->replay_window) { -case 256 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_256BIT ; -break ; -case 128 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_128BIT ;
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c static int macsec_set_replay_protection(struct mlx5_macsec_obj_attrs *attrs, voi · line 247
Quoted 10 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-10-break ; -case 64 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_64BIT ; -break ; -case 32 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_32BIT ; -break ; -default : -return -EINVAL ; - }
What sets window_sz now? Looking at the current code wouldn't this leave window_sz uninitialized and this undefined behavior of MLX5_SET? Either you should just forward in attrs->replay_window and remove window_sz local or drop the MLX5_SET call for setting window size?
Quoted 3 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-0MLX5_SET (macsec_aso, aso_ctx, window_size, window_sz);MLX5_SET (macsec_aso, aso_ctx, mode,MLX5_MACSEC_ASO_REPLAY_PROTECTION );Review
Quoted 1 line
(quoted diff) +0
-0
b/include/linux/mlx5/mlx5_ifc.h
Quoted 17 lines
index
5a4e914e2a6f..981fc7dfa408100644include/linux/mlx5/mlx5_ifc.h +0
-7@@ -11611,13 +11611,6 @@ enum { MLX5_MACSEC_ASO_REPLAY_PROTECTION = 0x1, }; -enum { -MLX5_MACSEC_ASO_REPLAY_WIN_32BIT =0x0 , -MLX5_MACSEC_ASO_REPLAY_WIN_64BIT =0x1 , -MLX5_MACSEC_ASO_REPLAY_WIN_128BIT =0x2 , -MLX5_MACSEC_ASO_REPLAY_WIN_256BIT =0x3 , -}; -#define MLX5_MACSEC_ASO_INC_SN 0x2#define MLX5_MACSEC_ASO_REG_C_4_5 0x2
Re: [net 14/15] net/mlx5e: MACsec, fix Tx SA active field update
In reply to Saeed MahameedQuoted 10 lines · Saeed Mahameed
From: Raed Salem <[email protected]>
Currently during update Tx security association (SA) flow, the Tx SA active state is updated only if the Tx SA in question is the same SA that the MACsec interface is using for Tx,in consequence when the MACsec interface chose to work with this Tx SA later, where this SA for example should have been updated to active state and it was not, the relevant Tx SA HW context won't be installed, hence the MACSec flow won't be offloaded.
Quoted 3 lines
Fix by update Tx SA active state as part of update flow regardless whether the SA in question is the same Tx SA used by the MACsec interface.
Quoted 30 lines
Fixes 8ff0ac5be144 ("net/mlx5: Add MACsec offload Tx command support")Signed-off-by Raed Salem <[email protected]>Reviewed-by Emeel Hakim <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +1
-2@@ -602,6 +602,7 @@ static int mlx5e_macsec_upd_txsa(struct macsec_context *ctx)if (tx_sa->active == ctx_tx_sa->active) goto out; + tx_sa->active = ctx_tx_sa->active;if (tx_sa->assoc_num != tx_sc->encoding_sa) goto out;@@ -617,8 +618,6 @@ static int mlx5e_macsec_upd_txsa(struct macsec_context *ctx)mlx5e_macsec_cleanup_sa (macsec, tx_sa, true); } - - tx_sa->active = ctx_tx_sa->active; out:mutex_unlock (&macsec->lock);
Re: [net 15/15] net/mlx5e: MACsec, block offload requests with encrypt off
In reply to Saeed MahameedQuoted 6 lines · Saeed Mahameed
From: Emeel Hakim <[email protected]>
Currently offloading MACsec with authentication only (encrypt property set to off) is not supported, block such requests when adding/updating a macsec device.
Makes sense.
Quoted 24 lines
Fixes 8ff0ac5be144 ("net/mlx5: Add MACsec offload Tx command support")Signed-off-by Emeel Hakim <[email protected]>Reviewed-by Raed Salem <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 5 +++++ 1 file changed, 5 insertions(+)drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +5
-0@@ -458,6 +458,11 @@ static bool mlx5e_macsec_secy_features_validate(struct macsec_context *ctx)return false; } +if (!ctx->secy->tx_sc.encrypt) { +netdev_err (netdev,"MACsec offload: encrypt off isn't supported\n" ); +return false; + } +return true; }
Jakub Kicinski
maintainerRe: [net 13/15] net/mlx5e: MACsec, remove replay window size limitation in offload path
In reply to Jacob KellerQuoted 20 lines · On Mon, 28 Nov 2022 15:42:19 -0800 Jacob Keller wrote:
Quoted 20 lines
index
c19581f1f733..72f8be65fa90100644drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-13@@ -229,22 +229,6 @@ static int macsec_set_replay_protection(struct mlx5_macsec_obj_attrs *attrs, voiif (!attrs->replay_protect)return 0 ; -switch (attrs->replay_window) { -case 256 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_256BIT ; -break ; -case 128 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_128BIT ; -break ; -case 64 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_64BIT ; -break ; -case 32 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_32BIT ; -break ;
Quoted 8 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c static int macsec_set_replay_protection(struct mlx5_macsec_obj_attrs *attrs, voi · line 247
Quoted 3 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-3-default : -return -EINVAL ; - }What sets
window_sznow? Looking at the current code wouldn't this leavewindow_szuninitialized and this undefined behavior ofMLX5_SET? Either you should just forward inattrs->replay_windowand removewindow_szlocal or drop theMLX5_SETcall for setting window size?
Damn it, this is a clang warning, I need to rescind the PR :/
Re: [net 13/15] net/mlx5e: MACsec, remove replay window size limitation in offload path
Quoted 30 lines · Jakub Kicinski
Quoted 20 lines · On Mon, 28 Nov 2022 15:42:19 -0800 Jacob Keller wrote:
Quoted 20 lines
index
c19581f1f733..72f8be65fa90100644drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-13@@ -229,22 +229,6 @@ static int macsec_set_replay_protection(struct mlx5_macsec_obj_attrs *attrs, voiif (!attrs->replay_protect)return 0 ; -switch (attrs->replay_window) { -case 256 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_256BIT ; -break ; -case 128 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_128BIT ; -break ; -case 64 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_64BIT ; -break ; -case 32 : - window_sz =MLX5_MACSEC_ASO_REPLAY_WIN_32BIT ; -break ;Quoted 8 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c static int macsec_set_replay_protection(struct mlx5_macsec_obj_attrs *attrs, voi · line 247
Quoted 3 lines
drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +0
-3-default : -return -EINVAL ; - }What sets
window_sznow? Looking at the current code wouldn't this leavewindow_szuninitialized and this undefined behavior ofMLX5_SET? Either you should just forward inattrs->replay_windowand removewindow_szlocal or drop theMLX5_SETcall for setting window size?
Quoted 1 line
Damn it, this is a clang warning, I need to rescind the PR :/
Make sense, Jacob found two real issues and this one is critical, but I don't know how that works for PRs, let me know when you do it so I will add his reviewed-by tags and address the two issues when is send v2.
another option is that i am currently working on my next PR for net, I can address those two issues there, but we won't have Jacobs reviewed-by for the work he already done on this series :/ ..
Thanks.
Re: [net 03/15] net/mlx5: E-switch, Fix duplicate lag creation
In reply to Jacob KellerQuoted 30 lines · Jacob Keller
Quoted 27 lines · Saeed Mahameed
From: Chris Mi <[email protected]>
If creating bond first and then enabling sriov in switchdev mode, will hit the following syndrome:
mlx5_core0000:08:00.0: mlx5_cmd_out_err:778:(pid 25543):CREATE_LAG(0x840)op_mod(0x0)failed, status bad parameter(0x3), syndrome (0x7d49cb), err(-22)The reason is because the offending patch removes eswitch mode none. In vf lag, the checking of eswitch mode none is replaced by checking if sriov is enabled. But when driver enables sriov, it triggers the bond workqueue task first and then setting sriov number in
pci_enable_sriov(). So the check fails.Fix it by checking if sriov is enabled using eswitch internal counter that is set before triggering the bond workqueue task.
Fixes f019679ea5f2 ("net/mlx5: E-switch, Remove dependency between sriov and eswitch mode")Signed-off-by Chris Mi <[email protected]>Reviewed-by Roi Dayan <[email protected]>Reviewed-by Mark Bloch <[email protected]>Reviewed-by Vlad Buslov <[email protected]>Signed-off-by Saeed Mahameed <[email protected]>--- drivers/net/ethernet/mellanox/mlx5/core/eswitch.h | 8 ++++++++ drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c | 5 +++-- 2 files changed, 11 insertions(+), 2 deletions(-)
[...]
Quoted 15 lines
drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev) · line 705
Quoted 12 lines
drivers/net/ethernet/mellanox/mlx5/core/lag/lag.c +3
-2@@ -701,8 +701,9 @@ static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev)#ifdef CONFIG_MLX5_ESWITCH dev = ldev->pf[MLX5_LAG_P1 ].dev; -if ((mlx5_sriov_is_enabled (dev)) && !is_mdev_switchdev_mode (dev)) -return false; +for (i =0 ; i < ldev->ports; i++) +if (mlx5_eswitch_num_vfs (dev->priv.eswitch) && !is_mdev_switchdev_mode (dev)) +return false;Am I missing something? whats with the for loop iterator here? i isn't used or passed into these functions?
Quoted 3 lines
Do you need to check multiple times or do these functions have some side effect? But looking at their implementation neither of them appear to have side effects?
What am I missing?
Great catch! it's a copy/paste bug, here we need to grab each port's eswitch on every iteration.
something like:
for (i = 0; i < ldev->ports; i++) {
+ dev = ldev->pf[i].dev;
if (mlx5_eswitch_num_vfs(dev->priv.eswitch) && !is_mdev_switchdev_mode(dev))
return false;
}Re: [net 13/15] net/mlx5e: MACsec, remove replay window size limitation in offload path
In reply to Saeed MahameedQuoted 7 lines · Saeed Mahameed
Quoted 1 line · Jakub Kicinski
Damn it, this is a clang warning, I need to rescind the PR :/
Make sense, Jacob found two real issues and this one is critical, but I don't know how that works for PRs, let me know when you do it so I will add his reviewed-by tags and address the two issues when is send v2.
hmm, I thought you were planing to revert my PR :).. anyways, i am will send the two fixes soon, so you could release your pr to linus, sorry about the mess.
Re: [net 13/15] net/mlx5e: MACsec, remove replay window size limitation in offload path
Quoted 4 lines · Saeed Mahameed
Quoted 3 lines · Saeed Mahameed
Quoted 1 line · Jakub Kicinski
Damn it, this is a clang warning, I need to rescind the PR :/
Quoted 9 lines
Quoted 4 lines
Make sense, Jacob found two real issues and this one is critical, but I don't know how that works for PRs, let me know when you do it so I will add his reviewed-by tags and address the two issues when is send v2.
hmm, I thought you were planing to revert my PR :).. anyways, i am will send the two fixes soon, so you could release your pr to linus, sorry about the mess.
The fixes look fine to me! Glad things got caught and fixed before getting spread widely.
Thanks,
Jake