lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date:   Thu, 20 Apr 2023 16:33:02 -0700
From:   Jakub Kicinski <kuba@...nel.org>
To:     davem@...emloft.net
Cc:     netdev@...r.kernel.org, edumazet@...gle.com, pabeni@...hat.com,
        mkubecek@...e.cz, Jakub Kicinski <kuba@...nel.org>, corbet@....net,
        dnlplm@...il.com, linux-doc@...r.kernel.org, saeedm@...dia.com,
        tariqt@...dia.com, leonro@...dia.com
Subject: [PATCH net-next] net: ethtool: coalesce: try to make user settings stick twice

SET_COALESCE may change operation mode and parameters in one call.
Changing operation mode may cause the driver to reset the parameter
values to what is a reasonable default for new operation mode.

Since driver does not know which parameters come from user and which
are echoed back from ->get, driver may ignore the parameters when
switching operation modes.

This used to be inevitable for ioctl() but in netlink we know which
parameters are actually specified by the user.

We could inform which parameters were set by the user but this would
lead to a lot of code duplication in the drivers. Instead try to call
the drivers twice if both mode and params are changed. The set method
already checks if any params need updating so in case the driver did
the right thing the first time around - there will be no second call
to it's ->set method (only an extra call to ->get()).

For mlx5 for example before this patch we'd see:

 # ethtool -C eth0 adaptive-rx on  adaptive-tx on
 # ethtool -C eth0 adaptive-rx off adaptive-tx off \
		   tx-usecs 123 rx-usecs 123
 Adaptive RX: off  TX: off
 rx-usecs: 3
 rx-frames: 32
 tx-usecs: 16
 tx-frames: 32
 [...]

After the change:

 # ethtool -C eth0 adaptive-rx on  adaptive-tx on
 # ethtool -C eth0 adaptive-rx off adaptive-tx off \
		   tx-usecs 123 rx-usecs 123
 Adaptive RX: off  TX: off
 rx-usecs: 123
 rx-frames: 32
 tx-usecs: 123
 tx-frames: 32
 [...]

This only works for netlink, so it's a small discrepancy between
netlink and ioctl(). Since we anticipate most users to move to
netlink I believe it's worth making their lives easier.

Signed-off-by: Jakub Kicinski <kuba@...nel.org>
---
mlx5 folks, LMK if you have a good idea of fixing this in the driver
instead, that may be cleaner, but I can't think of a good way :(

CC: corbet@....net
CC: dnlplm@...il.com
CC: linux-doc@...r.kernel.org

CC: saeedm@...dia.com
CC: tariqt@...dia.com
CC: leonro@...dia.com
---
 Documentation/networking/ethtool-netlink.rst |  4 ++
 net/ethtool/coalesce.c                       | 54 ++++++++++++++++----
 2 files changed, 47 insertions(+), 11 deletions(-)

diff --git a/Documentation/networking/ethtool-netlink.rst b/Documentation/networking/ethtool-netlink.rst
index cd0973d4ba01..2540c70952ff 100644
--- a/Documentation/networking/ethtool-netlink.rst
+++ b/Documentation/networking/ethtool-netlink.rst
@@ -1099,6 +1099,10 @@ such that the corresponding bit in ``ethtool_ops::supported_coalesce_params``
 is not set), regardless of their values. Driver may impose additional
 constraints on coalescing parameters and their values.
 
+Compared to requests issued via the ``ioctl()`` netlink version of this request
+will try harder to make sure that values specified by the user have been applied
+and may call the driver twice.
+
 
 PAUSE_GET
 =========
diff --git a/net/ethtool/coalesce.c b/net/ethtool/coalesce.c
index 443e7e642c96..01a59ce211c8 100644
--- a/net/ethtool/coalesce.c
+++ b/net/ethtool/coalesce.c
@@ -254,13 +254,14 @@ ethnl_set_coalesce_validate(struct ethnl_req_info *req_info,
 }
 
 static int
-ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info)
+__ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info,
+		     bool *dual_change)
 {
 	struct kernel_ethtool_coalesce kernel_coalesce = {};
 	struct net_device *dev = req_info->dev;
 	struct ethtool_coalesce coalesce = {};
+	bool mod_mode = false, mod = false;
 	struct nlattr **tb = info->attrs;
-	bool mod = false;
 	int ret;
 
 	ret = dev->ethtool_ops->get_coalesce(dev, &coalesce, &kernel_coalesce,
@@ -268,6 +269,7 @@ ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info)
 	if (ret < 0)
 		return ret;
 
+	/* Update values */
 	ethnl_update_u32(&coalesce.rx_coalesce_usecs,
 			 tb[ETHTOOL_A_COALESCE_RX_USECS], &mod);
 	ethnl_update_u32(&coalesce.rx_max_coalesced_frames,
@@ -286,10 +288,6 @@ ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info)
 			 tb[ETHTOOL_A_COALESCE_TX_MAX_FRAMES_IRQ], &mod);
 	ethnl_update_u32(&coalesce.stats_block_coalesce_usecs,
 			 tb[ETHTOOL_A_COALESCE_STATS_BLOCK_USECS], &mod);
-	ethnl_update_bool32(&coalesce.use_adaptive_rx_coalesce,
-			    tb[ETHTOOL_A_COALESCE_USE_ADAPTIVE_RX], &mod);
-	ethnl_update_bool32(&coalesce.use_adaptive_tx_coalesce,
-			    tb[ETHTOOL_A_COALESCE_USE_ADAPTIVE_TX], &mod);
 	ethnl_update_u32(&coalesce.pkt_rate_low,
 			 tb[ETHTOOL_A_COALESCE_PKT_RATE_LOW], &mod);
 	ethnl_update_u32(&coalesce.rx_coalesce_usecs_low,
@@ -312,17 +310,25 @@ ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info)
 			 tb[ETHTOOL_A_COALESCE_TX_MAX_FRAMES_HIGH], &mod);
 	ethnl_update_u32(&coalesce.rate_sample_interval,
 			 tb[ETHTOOL_A_COALESCE_RATE_SAMPLE_INTERVAL], &mod);
-	ethnl_update_u8(&kernel_coalesce.use_cqe_mode_tx,
-			tb[ETHTOOL_A_COALESCE_USE_CQE_MODE_TX], &mod);
-	ethnl_update_u8(&kernel_coalesce.use_cqe_mode_rx,
-			tb[ETHTOOL_A_COALESCE_USE_CQE_MODE_RX], &mod);
 	ethnl_update_u32(&kernel_coalesce.tx_aggr_max_bytes,
 			 tb[ETHTOOL_A_COALESCE_TX_AGGR_MAX_BYTES], &mod);
 	ethnl_update_u32(&kernel_coalesce.tx_aggr_max_frames,
 			 tb[ETHTOOL_A_COALESCE_TX_AGGR_MAX_FRAMES], &mod);
 	ethnl_update_u32(&kernel_coalesce.tx_aggr_time_usecs,
 			 tb[ETHTOOL_A_COALESCE_TX_AGGR_TIME_USECS], &mod);
-	if (!mod)
+
+	/* Update operation modes */
+	ethnl_update_bool32(&coalesce.use_adaptive_rx_coalesce,
+			    tb[ETHTOOL_A_COALESCE_USE_ADAPTIVE_RX], &mod_mode);
+	ethnl_update_bool32(&coalesce.use_adaptive_tx_coalesce,
+			    tb[ETHTOOL_A_COALESCE_USE_ADAPTIVE_TX], &mod_mode);
+	ethnl_update_u8(&kernel_coalesce.use_cqe_mode_tx,
+			tb[ETHTOOL_A_COALESCE_USE_CQE_MODE_TX], &mod_mode);
+	ethnl_update_u8(&kernel_coalesce.use_cqe_mode_rx,
+			tb[ETHTOOL_A_COALESCE_USE_CQE_MODE_RX], &mod_mode);
+
+	*dual_change = mod && mod_mode;
+	if (!mod && !mod_mode)
 		return 0;
 
 	ret = dev->ethtool_ops->set_coalesce(dev, &coalesce, &kernel_coalesce,
@@ -330,6 +336,32 @@ ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info)
 	return ret < 0 ? ret : 1;
 }
 
+static int
+ethnl_set_coalesce(struct ethnl_req_info *req_info, struct genl_info *info)
+{
+	bool dual_change;
+	int err, ret;
+
+	/* SET_COALESCE may change operation mode and parameters in one call.
+	 * Changing operation mode may cause the driver to reset the parameter
+	 * values, and therefore ignore user input (driver does not know which
+	 * parameters come from user and which are echoed back from ->get).
+	 * To not complicate the drivers if user tries to change both the mode
+	 * and parameters at once - call the driver twice.
+	 */
+	err = __ethnl_set_coalesce(req_info, info, &dual_change);
+	if (err < 0)
+		return err;
+	ret = err;
+
+	if (ret && dual_change) {
+		err = __ethnl_set_coalesce(req_info, info, &dual_change);
+		if (err < 0)
+			return err;
+	}
+	return ret;
+}
+
 const struct ethnl_request_ops ethnl_coalesce_request_ops = {
 	.request_cmd		= ETHTOOL_MSG_COALESCE_GET,
 	.reply_cmd		= ETHTOOL_MSG_COALESCE_GET_REPLY,
-- 
2.39.2

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ