batman-adv: tp_meter: error reporting and role mixups
Sven Eckelmann
maintainerv2 cover batman-adv: tp_meter: error reporting and role mixups
batadv_tp_sender_shutdown() previously used two separate variables to track session state: sending (an atomic flag indicating whether the session was active) and reason (a plain enum storing the stop reason). This introduced a race window between the two writes: after sending was cleared to 0, batadv_tp_send() could observe the stopped state and call batadv_tp_sender_end() before reason was written, causing the wrong stop reason to be reported to the caller.
The state + reason handling has to be reworked to only modify a single variable
At the same, time we had some tp_vars role confusion and overly large tp_vars for receiving. Just split them up. Making sure that the search functions only search for the correct role and at the same time only use the required memory.
The last patch should not enter stable/net.git but go through main/net-next.git. It is just added here because it depends on the first patch.
---
Changes in v2:
- fix send_result kernel-doc
- rebase patch
- add patch to avoid role mixup in search functions
- add patch to split sender and receiver sessions
- Link to v1: https://patch.msgid.link/[email protected]
---
Sven Eckelmann (3):
batman-adv: tp_meter: fix race condition in send error reporting
batman-adv: tp_meter: avoid role confusion in tp_list
batman-adv: tp_meter: split vars into sender and receiver types
net/batman-adv/main.c | 3 +- net/batman-adv/tp_meter.c | 453 +++++++++++++++++++++++++++------------------- net/batman-adv/types.h | 109 +++++------ 3 files changed, 325 insertions(+), 240 deletions(-)
---
base-commit: ff182fc0c666754e3f35f7b63c9ccf0dff239e52
change-id: 20260513-tp-reason-missing-f1aa51f3f374
Best regards,
Signature · 1 line
Sven Eckelmann <[email protected]>
Sven Eckelmann
maintainerPATCH v2 1/3 batman-adv: tp_meter: fix race condition in send error reporting
batadv_tp_sender_shutdown() previously used two separate variables to track session state: sending (an atomic flag indicating whether the session was active) and reason (a plain enum storing the stop reason). This introduced a race window between the two writes: after sending was cleared to 0, batadv_tp_send() could observe the stopped state and call batadv_tp_sender_end() before reason was written, causing the wrong stop reason to be reported to the caller.
Fix this by consolidating both variables into a single atomic send_result, which holds 0 while the session is running and the stop reason once it ends.
--- net/batman-adv/tp_meter.c | 40 +++++++++++++++++++++++++--------------- net/batman-adv/types.h | 10 +++++----- 2 files changed, 30 insertions(+), 20 deletions(-)
net/batman-adv/tp_meter.c +25 -15
@@ -413,11 +413,14 @@ static void batadv_tp_sender_cleanup(struct batadv_tp_vars *tp_vars)static void batadv_tp_sender_end (struct batadv_priv *bat_priv,struct batadv_tp_vars *tp_vars) { +enum batadv_tp_meter_reason reason;u32 session_cookie; + reason =atomic_read (&tp_vars->send_result); +batadv_dbg (BATADV_DBG_TP_METER , bat_priv,"Test towards %pM finished..shutting down (reason=%d)\n" , - tp_vars->other_end, tp_vars->reason); + tp_vars->other_end, reason); batadv_dbg(BATADV_DBG_TP_METER, bat_priv, "Last timing stats: SRTT=%ums RTTVAR=%ums RTO=%ums\n",
@@ -430,7 +433,7 @@ static void batadv_tp_sender_end(struct batadv_priv *bat_priv, session_cookie =batadv_tp_session_cookie (tp_vars->session, tp_vars->icmp_uid); - batadv_tp_batctl_notify(tp_vars->reason, + batadv_tp_batctl_notify(reason, tp_vars->other_end, bat_priv, tp_vars->start_time,
@@ -446,10 +449,18 @@ static void batadv_tp_sender_end(struct batadv_priv *bat_priv,static void batadv_tp_sender_shutdown (struct batadv_tp_vars *tp_vars,enum batadv_tp_meter_reason reason) { -if (atomic_xchg (&tp_vars->sending,0 ) !=1 ) -return ; +atomic_cmpxchg (&tp_vars->send_result,0 , reason); +} - tp_vars->reason = reason; +/** +* batadv_tp_sender_stopped() - check if tp session was stopped with reason +* @tp_vars: the private data of the current TP meter session +* +* Return: whether stop reason was found +*/ +static bool batadv_tp_sender_stopped (struct batadv_tp_vars *tp_vars) +{ +return atomic_read (&tp_vars->send_result) !=0 ; }/**
@@ -479,7 +490,7 @@ static void batadv_tp_reset_sender_timer(struct batadv_tp_vars *tp_vars)/* most of the time this function is invoked while normal packet * reception... */ -if (unlikely (atomic_read (&tp_vars->sending) ==0 )) +if (unlikely (batadv_tp_sender_stopped (tp_vars)))/* timer ref will be dropped in batadv_tp_sender_cleanup */ return ;
@@ -499,7 +510,7 @@ static void batadv_tp_sender_timeout(struct timer_list *t)struct batadv_tp_vars *tp_vars =timer_container_of (tp_vars, t, timer);struct batadv_priv *bat_priv = tp_vars->bat_priv; -if (atomic_read (&tp_vars->sending) ==0 ) +if (batadv_tp_sender_stopped (tp_vars))return ;/* if the user waited long enough...shutdown the test */
@@ -661,7 +672,7 @@ static void batadv_tp_recv_ack(struct batadv_priv *bat_priv,if (unlikely (tp_vars->role !=BATADV_TP_SENDER )) goto out; -if (unlikely (atomic_read (&tp_vars->sending) ==0 )) +if (unlikely (batadv_tp_sender_stopped (tp_vars))) goto out;/* old ACK? silently drop it.. */
@@ -827,21 +838,21 @@ static int batadv_tp_send(void *arg)if (unlikely (tp_vars->role !=BATADV_TP_SENDER )) { err =BATADV_TP_REASON_DST_UNREACHABLE ; - tp_vars->reason = err; +batadv_tp_sender_shutdown (tp_vars, err); goto out; } orig_node =batadv_orig_hash_find (bat_priv, tp_vars->other_end);if (unlikely (!orig_node)) { err =BATADV_TP_REASON_DST_UNREACHABLE ; - tp_vars->reason = err; +batadv_tp_sender_shutdown (tp_vars, err); goto out; } primary_if =batadv_primary_if_get_selected (bat_priv);if (unlikely (!primary_if)) { err =BATADV_TP_REASON_DST_UNREACHABLE ; - tp_vars->reason = err; +batadv_tp_sender_shutdown (tp_vars, err); goto out; }
@@ -860,7 +871,7 @@ static int batadv_tp_send(void *arg)queue_delayed_work (batadv_event_workqueue, &tp_vars->finish_work,msecs_to_jiffies (tp_vars->test_length)); -while (atomic_read (&tp_vars->sending) !=0 ) { +while (!batadv_tp_sender_stopped (tp_vars)) {if (unlikely (!batadv_tp_avail (tp_vars, payload_len))) {batadv_tp_wait_available (tp_vars, payload_len);continue ;
@@ -883,8 +894,7 @@ static int batadv_tp_send(void *arg)"Meter: %s() cannot send packets (%d)\n" , __func__, err);/* ensure nobody else tries to stop the thread now */ -if (atomic_xchg (&tp_vars->sending,0 ) ==1 ) - tp_vars->reason = err; +batadv_tp_sender_shutdown (tp_vars, err);break ; }
@@ -1006,7 +1016,7 @@ void batadv_tp_start(struct batadv_priv *bat_priv, const u8 *dst,ether_addr_copy (tp_vars->other_end, dst);kref_init (&tp_vars->refcount); tp_vars->role =BATADV_TP_SENDER ; -atomic_set (&tp_vars->sending,1 ); +atomic_set (&tp_vars->send_result,0 );memcpy (tp_vars->session, session_id,sizeof (session_id)); tp_vars->icmp_uid = icmp_uid;
net/batman-adv/types.h +5 -5
@@ -1320,15 +1320,15 @@ struct batadv_tp_vars {/** @role: receiver/sender modi */ enum batadv_tp_meter_role role; -/** @sending: sending binary semaphore: 1 if sending, 0 is not */ -atomic_t sending; +/** +* @send_result: 0 when sending is ongoing and otherwise +* enum batadv_tp_meter_reason +*/ +atomic_t send_result;/** @receiving: receiving binary semaphore: 1 if receiving, 0 is not */ atomic_t receiving; -/** @reason: reason for a stopped session */ -enum batadv_tp_meter_reason reason; -/** @finish_work: work item for the finishing procedure */ struct delayed_work finish_work;
Signature · 1 line
2.47.3
Sven Eckelmann
maintainerPATCH v2 2/3 batman-adv: tp_meter: avoid role confusion in tp_list
In reply to Sven EckelmannSession lookups in tp_list matched only on destination address (and optionally session ID), leaving role validation to the caller. If two sessions with the same other_end coexisted (one as sender, one as receiver) a lookup could silently return the wrong one, causing the caller's role to bail out early, potentially skipping necessary cleanup.
Move the role check into the lookup functions themselves so the correct entry is always returned, or none at all. Since batadv_tp_start() legitimately needs to detect any active session to a destination regardless of role, introduce a dedicated helper for that case rather than bending the existing lookup semantics.
--- net/batman-adv/tp_meter.c | 59 +++++++++++++++++++++++++++++------------------ 1 file changed, 36 insertions(+), 23 deletions(-)
net/batman-adv/tp_meter.c +36 -23
@@ -255,6 +255,7 @@ static void batadv_tp_batctl_error_notify(enum batadv_tp_meter_reason reason,* batadv_tp_list_find() - find a tp_vars object in the global list * @bat_priv: the bat priv with all the mesh interface information * @dst: the other endpoint MAC address to look for +* @role: role of the session * * Look for a tp_vars object matching dst as end_point and return it after * having increment the refcounter. Return NULL is not found
@@ -262,7 +263,8 @@ static void batadv_tp_batctl_error_notify(enum batadv_tp_meter_reason reason,* Return: matching tp_vars or NULL when no tp_vars with @dst was found */ static struct batadv_tp_vars *batadv_tp_list_find (struct batadv_priv *bat_priv, -const u8 *dst) +const u8 *dst, +enum batadv_tp_meter_role role) {struct batadv_tp_vars *pos, *tp_vars =NULL ;
@@ -271,6 +273,9 @@ static struct batadv_tp_vars *batadv_tp_list_find(struct batadv_priv *bat_priv,if (!batadv_compare_eth (pos->other_end, dst))continue ; +if (pos->role != role) +continue ; +/* most of the time this function is invoked during the normal * process..it makes sens to pay more when the session is * finished and to speed the process up during the measurement
@@ -286,12 +291,33 @@ static struct batadv_tp_vars *batadv_tp_list_find(struct batadv_priv *bat_priv,return tp_vars; } +/** +* batadv_tp_list_active() - check if session from/to destination is ongoing +* @bat_priv: the bat priv with all the mesh interface information +* @dst: the other endpoint MAC address to look for +* +* Return: if matching session with @dst was found +*/ +static bool batadv_tp_list_active (struct batadv_priv *bat_priv,const u8 *dst) + __must_hold(&bat_priv->tp_list_lock) +{ +struct batadv_tp_vars *tp_vars; + +hlist_for_each_entry_rcu (tp_vars, &bat_priv->tp_list, list) { +if (batadv_compare_eth (tp_vars->other_end, dst)) +return true; + } + +return false; +} +/** * batadv_tp_list_find_session() - find tp_vars session object in the global * list * @bat_priv: the bat priv with all the mesh interface information * @dst: the other endpoint MAC address to look for * @session: session identifier +* @role: role of the session * * Look for a tp_vars object matching dst as end_point, session as tp meter * session and return it after having increment the refcounter. Return NULL
@@ -301,7 +327,7 @@ static struct batadv_tp_vars *batadv_tp_list_find(struct batadv_priv *bat_priv,*/ static struct batadv_tp_vars *batadv_tp_list_find_session (struct batadv_priv *bat_priv,const u8 *dst, -const u8 *session) +const u8 *session,enum batadv_tp_meter_role role) {struct batadv_tp_vars *pos, *tp_vars =NULL ;
@@ -313,6 +339,9 @@ batadv_tp_list_find_session(struct batadv_priv *bat_priv, const u8 *dst,if (memcmp (pos->session, session,sizeof (pos->session)) !=0 )continue ; +if (pos->role != role) +continue ; +/* most of the time this function is invoked during the normal * process..it makes sense to pay more when the session is * finished and to speed the process up during the measurement
@@ -665,13 +694,10 @@ static void batadv_tp_recv_ack(struct batadv_priv *bat_priv,/* find the tp_vars */ tp_vars =batadv_tp_list_find_session (bat_priv, icmp->orig, - icmp->session); + icmp->session,BATADV_TP_SENDER );if (unlikely (!tp_vars))return ; -if (unlikely (tp_vars->role !=BATADV_TP_SENDER )) - goto out; -if (unlikely (batadv_tp_sender_stopped (tp_vars))) goto out;
@@ -980,10 +1006,8 @@ void batadv_tp_start(struct batadv_priv *bat_priv, const u8 *dst,return ; } - tp_vars =batadv_tp_list_find (bat_priv, dst); -if (tp_vars) { +if (batadv_tp_list_active (bat_priv, dst)) {spin_unlock_bh (&bat_priv->tp_list_lock); -batadv_tp_vars_put (tp_vars);batadv_dbg (BATADV_DBG_TP_METER , bat_priv,"Meter: test to or from the same node already ongoing, aborting\n" ); batadv_tp_batctl_error_notify(BATADV_TP_REASON_ALREADY_ONGOING,
@@ -1104,18 +1128,14 @@ void batadv_tp_stop(struct batadv_priv *bat_priv, const u8 *dst,if (!orig_node)return ; - tp_vars =batadv_tp_list_find (bat_priv, orig_node->orig); + tp_vars =batadv_tp_list_find (bat_priv, orig_node->orig,BATADV_TP_SENDER );if (!tp_vars) {batadv_dbg (BATADV_DBG_TP_METER , bat_priv,"Meter: trying to interrupt an already over connection\n" ); goto out_put_orig_node; } -if (unlikely (tp_vars->role !=BATADV_TP_SENDER )) - goto out_put_tp_vars; -batadv_tp_sender_shutdown (tp_vars, return_value); -out_put_tp_vars:batadv_tp_vars_put (tp_vars); out_put_orig_node:batadv_orig_node_put (orig_node);
@@ -1371,7 +1391,7 @@ batadv_tp_init_recv(struct batadv_priv *bat_priv, goto out_unlock; tp_vars =batadv_tp_list_find_session (bat_priv, icmp->orig, - icmp->session); + icmp->session,BATADV_TP_RECEIVER );if (tp_vars) goto out_unlock;
@@ -1442,7 +1462,7 @@ static void batadv_tp_recv_msg(struct batadv_priv *bat_priv, } }else { tp_vars =batadv_tp_list_find_session (bat_priv, icmp->orig, - icmp->session); + icmp->session,BATADV_TP_RECEIVER );if (!tp_vars) { batadv_dbg(BATADV_DBG_TP_METER, bat_priv, "Unexpected packet from %pM!\n",
@@ -1451,13 +1471,6 @@ static void batadv_tp_recv_msg(struct batadv_priv *bat_priv, } } -if (unlikely (tp_vars->role !=BATADV_TP_RECEIVER )) { -batadv_dbg (BATADV_DBG_TP_METER , bat_priv, -"Meter: dropping packet: not expected (role=%u)\n" , - tp_vars->role); - goto out; - } - tp_vars->last_recv_time = jiffies;/* if the packet is a duplicate, it may be the case that an ACK has been
Signature · 1 line
2.47.3
Sven Eckelmann
maintainerPATCH v2 3/3 batman-adv: tp_meter: split vars into sender and receiver types
In reply to Sven EckelmannThe monolithic batadv_tp_vars struct holds fields for both sender and receiver roles, distinguished only by a runtime enum role. This makes it easy to accidentally access a field intended for the opposite role, since neither the compiler nor the type system provide any guard against such mistakes. The role check also adds unnecessary branching in several code paths.
Introduce batadv_tp_vars_common to hold fields shared across both roles, then derive two separate types (sender/receiver) from it. The functions can operate on them without any ambiguity about the available fields. This also reduces the memory footprint of receiver sessions, which no longer carry the substantial sender-only fields.
Care must be taken to prevent concurrent TP sessions between the same two peers in opposite directions, since sender and receiver sessions are now tracked in separate lists and a lookup in one list no longer detects a session in the other.
--- net/batman-adv/main.c | 3 +- net/batman-adv/tp_meter.c | 396 ++++++++++++++++++++++++++-------------------- net/batman-adv/types.h | 101 ++++++------ 3 files changed, 281 insertions(+), 219 deletions(-)
net/batman-adv/main.c +2 -1
@@ -181,7 +181,8 @@ int batadv_mesh_init(struct net_device *mesh_iface)INIT_HLIST_HEAD (&bat_priv->tvlv.container_list);INIT_HLIST_HEAD (&bat_priv->tvlv.handler_list);INIT_HLIST_HEAD (&bat_priv->meshif_vlan_list); -INIT_HLIST_HEAD (&bat_priv->tp_list); +INIT_HLIST_HEAD (&bat_priv->tp_sender_list); +INIT_HLIST_HEAD (&bat_priv->tp_receiver_list); bat_priv->gw.generation =0 ;
net/batman-adv/tp_meter.c +228 -168
859 lines not on this page · raw message
net/batman-adv/types.h +51 -50
163 lines not on this page · raw message
Signature · 1 line
2.47.3