ipmr: remove cache_resolve_queue_len
Subsystems (3)
PATCH 1/1 ipmr: remove cache_resolve_queue_len
This is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source addresses(cache_resolve_queue_len) to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.
To resolve this issue, we need either add a sysctl entry to make the cache_resolve_queue_len configurable, or just remove cache_resolve_queue_len directly, as we already have the socket receive queue limits of mrouted socket, pointed by David.
From my side, I'd perfer to remove the cache_resolve_queue_len instead of creating two more(IPv4 and IPv6 version) sysctl entry.
[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343
--- include/linux/mroute_base.h | 2 -- net/ipv4/ipmr.c | 27 ++++++++++++++++++--------- net/ipv6/ip6mr.c | 10 +++------- 3 files changed, 21 insertions(+), 18 deletions(-)
include/linux/mroute_base.h +0 -2
@@ -235,7 +235,6 @@ struct mr_table_ops {* @mfc_hash: Hash table of all resolved routes for easy lookup * @mfc_cache_list: list of resovled routes for possible traversal * @maxvif: Identifier of highest value vif currently in use -* @cache_resolve_queue_len: current size of unresolved queue * @mroute_do_assert: Whether to inform userspace on wrong ingress * @mroute_do_pim: Whether to receive IGMP PIMv1 * @mroute_reg_vif_num: PIM-device vif index
@@ -252,7 +251,6 @@ struct mr_table {struct rhltable mfc_hash;struct list_head mfc_cache_list;int maxvif; -atomic_t cache_resolve_queue_len;bool mroute_do_assert;bool mroute_do_pim;bool mroute_do_wrvifwhole;
net/ipv4/ipmr.c +18 -9
@@ -744,8 +744,6 @@ static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)struct sk_buff *skb;struct nlmsgerr *e; -atomic_dec (&mrt->cache_resolve_queue_len); -while ((skb =skb_dequeue (&c->_c.mfc_un.unres.unresolved))) {if (ip_hdr (skb)->version ==0 ) { struct nlmsghdr *nlh = skb_pull(skb,
@@ -1133,9 +1131,11 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, }if (!found) { +bool was_empty; +/* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ipmr_cache_alloc_unres ()) ==NULL ) { + c =ipmr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
@@ -1161,11 +1161,11 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,return err; } -atomic_inc (&mrt->cache_resolve_queue_len); + was_empty =list_empty (&mrt->mfc_unres_queue);list_add (&c->_c.list, &mrt->mfc_unres_queue);mroute_netlink_event (mrt, c,RTM_NEWROUTE ); -if (atomic_read (&mrt->cache_resolve_queue_len) ==1 ) +if (was_empty)mod_timer (&mrt->ipmr_expire_timer, c->_c.mfc_un.unres.expires); }
@@ -1272,7 +1272,6 @@ static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,if (uc->mfc_origin == c->mfc_origin && uc->mfc_mcastgrp == c->mfc_mcastgrp) {list_del (&_uc->list); -atomic_dec (&mrt->cache_resolve_queue_len); found = true;break ; }
@@ -1328,7 +1327,7 @@ static void mroute_clean_tables(struct mr_table *mrt, int flags) }if (flags &MRT_FLUSH_MFC ) { -if (atomic_read (&mrt->cache_resolve_queue_len) !=0 ) { +if (!list_empty (&mrt->mfc_unres_queue)) {spin_lock_bh (&mfc_unres_lock);list_for_each_entry_safe (c, tmp, &mrt->mfc_unres_queue, list) {list_del (&c->list);
@@ -2750,9 +2749,19 @@ static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,return ipmr_mfc_delete (tbl, &mfcc, parent); } +static int queue_count (struct mr_table *mrt) +{ +struct list_head *pos; +int count =0 ; + +list_for_each (pos, &mrt->mfc_unres_queue) + count++; +return count; +} +static bool ipmr_fill_table (struct mr_table *mrt,struct sk_buff *skb) { -u32 queue_len =atomic_read (&mrt->cache_resolve_queue_len); +u32 queue_len =queue_count (mrt); if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) || nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
net/ipv6/ip6mr.c +3 -7
@@ -768,8 +768,6 @@ static void ip6mr_destroy_unres(struct mr_table *mrt, struct mfc6_cache *c)struct net *net =read_pnet (&mrt->net);struct sk_buff *skb; -atomic_dec (&mrt->cache_resolve_queue_len); -while ((skb =skb_dequeue (&c->_c.mfc_un.unres.unresolved)) !=NULL ) {if (ipv6_hdr (skb)->version ==0 ) { struct nlmsghdr *nlh = skb_pull(skb,
@@ -1148,8 +1146,8 @@ static int ip6mr_cache_unresolved(struct mr_table *mrt, mifi_t mifi,* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ip6mr_cache_alloc_unres ()) ==NULL ) { + c =ip6mr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
@@ -1176,7 +1174,6 @@ static int ip6mr_cache_unresolved(struct mr_table *mrt, mifi_t mifi,return err; } -atomic_inc (&mrt->cache_resolve_queue_len);list_add (&c->_c.list, &mrt->mfc_unres_queue);mr6_netlink_event (mrt, c,RTM_NEWROUTE );
@@ -1468,7 +1465,6 @@ static int ip6mr_mfc_add(struct net *net, struct mr_table *mrt,if (ipv6_addr_equal (&uc->mf6c_origin, &c->mf6c_origin) &&ipv6_addr_equal (&uc->mf6c_mcastgrp, &c->mf6c_mcastgrp)) {list_del (&_uc->list); -atomic_dec (&mrt->cache_resolve_queue_len); found = true;break ; }
@@ -1526,7 +1522,7 @@ static void mroute_clean_tables(struct mr_table *mrt, int flags) }if (flags &MRT6_FLUSH_MFC ) { -if (atomic_read (&mrt->cache_resolve_queue_len) !=0 ) { +if (!list_empty (&mrt->mfc_unres_queue)) {spin_lock_bh (&mfc_unres_lock);list_for_each_entry_safe (c, tmp, &mrt->mfc_unres_queue, list) {list_del (&c->list);
Signature · 1 line
2.19.2
Quoted 22 lines · Hangbin Liu
This is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source
addresses(cache_resolve_queue_len)to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.To resolve this issue, we need either add a sysctl entry to make the
cache_resolve_queue_lenconfigurable, or just removecache_resolve_queue_lendirectly, as we already have the socket receive queue limits of mrouted socket, pointed by David.From my side, I'd perfer to remove the
cache_resolve_queue_leninstead of creating two more(IPv4 and IPv6 version) sysctl entry.[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343
Quoted 8 lines
Reported-by Phil Karn <[email protected]>Signed-off-by Hangbin Liu <[email protected]>--- include/linux/mroute_base.h | 2 -- net/ipv4/ipmr.c | 27 ++++++++++++++++++--------- net/ipv6/ip6mr.c | 10 +++------- 3 files changed, 21 insertions(+), 18 deletions(-)
Hi,
IMO this is definitely net-next material. A few more comments below.
Note that hosts will automatically have this limit lifted to about 270 entries with current defaults, some might be surprised if they have higher limits set and could be left with queues allowing for thousands of entries in a linked list.
Quoted 81 lines
include/linux/mroute_base.h +0
-2@@ -235,7 +235,6 @@ struct mr_table_ops {* @mfc_hash: Hash table of all resolved routes for easy lookup * @mfc_cache_list: list of resovled routes for possible traversal * @maxvif: Identifier of highest value vif currently in use -* @cache_resolve_queue_len: current size of unresolved queue * @mroute_do_assert: Whether to inform userspace on wrong ingress * @mroute_do_pim: Whether to receive IGMP PIMv1 * @mroute_reg_vif_num: PIM-device vif index @@ -252,7 +251,6 @@ struct mr_table {struct rhltable mfc_hash;struct list_head mfc_cache_list;int maxvif; -atomic_t cache_resolve_queue_len;bool mroute_do_assert;bool mroute_do_pim;bool mroute_do_wrvifwhole;net/ipv4/ipmr.c +7
-8@@ -744,8 +744,6 @@ static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)struct sk_buff *skb;struct nlmsgerr *e; -atomic_dec (&mrt->cache_resolve_queue_len); -while ((skb =skb_dequeue (&c->_c.mfc_un.unres.unresolved))) {if (ip_hdr (skb)->version ==0 ) { struct nlmsghdr *nlh = skb_pull(skb,@@ -1133,9 +1131,11 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, }if (!found) { +bool was_empty; +/* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ipmr_cache_alloc_unres ()) ==NULL ) { + c =ipmr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);@@ -1161,11 +1161,11 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,return err; } -atomic_inc (&mrt->cache_resolve_queue_len); + was_empty =list_empty (&mrt->mfc_unres_queue);list_add (&c->_c.list, &mrt->mfc_unres_queue);mroute_netlink_event (mrt, c,RTM_NEWROUTE ); -if (atomic_read (&mrt->cache_resolve_queue_len) ==1 ) +if (was_empty)mod_timer (&mrt->ipmr_expire_timer, c->_c.mfc_un.unres.expires); }@@ -1272,7 +1272,6 @@ static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,if (uc->mfc_origin == c->mfc_origin && uc->mfc_mcastgrp == c->mfc_mcastgrp) {list_del (&_uc->list); -atomic_dec (&mrt->cache_resolve_queue_len); found = true;break ; }@@ -1328,7 +1327,7 @@ static void mroute_clean_tables(struct mr_table *mrt, int flags) }if (flags &MRT_FLUSH_MFC ) { -if (atomic_read (&mrt->cache_resolve_queue_len) !=0 ) { +if (!list_empty (&mrt->mfc_unres_queue)) {spin_lock_bh (&mfc_unres_lock);list_for_each_entry_safe (c, tmp, &mrt->mfc_unres_queue, list) {list_del (&c->list);@@ -2750,9 +2749,19 @@ static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,return ipmr_mfc_delete (tbl, &mfcc, parent); }
net/ipv4/ipmr.c static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh, · line 2760
Quoted 10 lines
net/ipv4/ipmr.c +9
-0+static int queue_count (struct mr_table *mrt) +{ +struct list_head *pos; +int count =0 ; + +list_for_each (pos, &mrt->mfc_unres_queue) + count++; +return count; +}
I don't think we hold the mfc_unres_lock here while walking the unresolved list below in ipmr_fill_table().
Quoted 58 lines
net/ipv4/ipmr.c +2
-1+static bool ipmr_fill_table (struct mr_table *mrt,struct sk_buff *skb) { -u32 queue_len =atomic_read (&mrt->cache_resolve_queue_len); +u32 queue_len =queue_count (mrt); if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) || nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||net/ipv6/ip6mr.c +3
-7@@ -768,8 +768,6 @@ static void ip6mr_destroy_unres(struct mr_table *mrt, struct mfc6_cache *c)struct net *net =read_pnet (&mrt->net);struct sk_buff *skb; -atomic_dec (&mrt->cache_resolve_queue_len); -while ((skb =skb_dequeue (&c->_c.mfc_un.unres.unresolved)) !=NULL ) {if (ipv6_hdr (skb)->version ==0 ) { struct nlmsghdr *nlh = skb_pull(skb,@@ -1148,8 +1146,8 @@ static int ip6mr_cache_unresolved(struct mr_table *mrt, mifi_t mifi,* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ip6mr_cache_alloc_unres ()) ==NULL ) { + c =ip6mr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);@@ -1176,7 +1174,6 @@ static int ip6mr_cache_unresolved(struct mr_table *mrt, mifi_t mifi,return err; } -atomic_inc (&mrt->cache_resolve_queue_len);list_add (&c->_c.list, &mrt->mfc_unres_queue);mr6_netlink_event (mrt, c,RTM_NEWROUTE );@@ -1468,7 +1465,6 @@ static int ip6mr_mfc_add(struct net *net, struct mr_table *mrt,if (ipv6_addr_equal (&uc->mf6c_origin, &c->mf6c_origin) &&ipv6_addr_equal (&uc->mf6c_mcastgrp, &c->mf6c_mcastgrp)) {list_del (&_uc->list); -atomic_dec (&mrt->cache_resolve_queue_len); found = true;break ; }@@ -1526,7 +1522,7 @@ static void mroute_clean_tables(struct mr_table *mrt, int flags) }if (flags &MRT6_FLUSH_MFC ) { -if (atomic_read (&mrt->cache_resolve_queue_len) !=0 ) { +if (!list_empty (&mrt->mfc_unres_queue)) {spin_lock_bh (&mfc_unres_lock);list_for_each_entry_safe (c, tmp, &mrt->mfc_unres_queue, list) {list_del (&c->list);
Hi Nikolay,
Thanks for the feedback, see comments below.
Quoted 32 lines · Nikolay Aleksandrov
Quoted 22 lines · Hangbin Liu
This is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source
addresses(cache_resolve_queue_len)to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.To resolve this issue, we need either add a sysctl entry to make the
cache_resolve_queue_lenconfigurable, or just removecache_resolve_queue_lendirectly, as we already have the socket receive queue limits of mrouted socket, pointed by David.From my side, I'd perfer to remove the
cache_resolve_queue_leninstead of creating two more(IPv4 and IPv6 version) sysctl entry.[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343Quoted 8 lines
Reported-by Phil Karn <[email protected]>Signed-off-by Hangbin Liu <[email protected]>--- include/linux/mroute_base.h | 2 -- net/ipv4/ipmr.c | 27 ++++++++++++++++++--------- net/ipv6/ip6mr.c | 10 +++------- 3 files changed, 21 insertions(+), 18 deletions(-)
Quoted 2 lines
Hi,
IMO this is definitely net-next material. A few more comments below.
I thoug this is a bug fix. But it looks more suitable to net-next as you said.
Quoted 5 lines
Note that hosts will automatically have this limit lifted to about 270 entries with current defaults, some might be surprised if they have higher limits set and could be left with queues allowing for thousands of entries in a linked list.
I think this is just a cache list and should be expired soon. The cache creation would also failed if there is no buffer.
But if you like, I can write a patch with sysctl parameter.
Quoted 13 lines
net/ipv4/ipmr.c static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh, · line 2760
Quoted 10 lines
net/ipv4/ipmr.c +9
-0+static int queue_count (struct mr_table *mrt) +{ +struct list_head *pos; +int count =0 ; + +list_for_each (pos, &mrt->mfc_unres_queue) + count++; +return count; +}I don't think we hold the
mfc_unres_lockhere while walking the unresolved list below inipmr_fill_table().
ah, yes, I will fix this.
Thanks
Hangbin
Quoted 47 lines · Hangbin Liu
Hi Nikolay,
Thanks for the feedback, see comments below.
Quoted 32 lines · Nikolay Aleksandrov
Quoted 22 lines · Hangbin Liu
This is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source
addresses(cache_resolve_queue_len)to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.To resolve this issue, we need either add a sysctl entry to make the
cache_resolve_queue_lenconfigurable, or just removecache_resolve_queue_lendirectly, as we already have the socket receive queue limits of mrouted socket, pointed by David.From my side, I'd perfer to remove the
cache_resolve_queue_leninstead of creating two more(IPv4 and IPv6 version) sysctl entry.[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343Quoted 8 lines
Reported-by Phil Karn <[email protected]>Signed-off-by Hangbin Liu <[email protected]>--- include/linux/mroute_base.h | 2 -- net/ipv4/ipmr.c | 27 ++++++++++++++++++--------- net/ipv6/ip6mr.c | 10 +++------- 3 files changed, 21 insertions(+), 18 deletions(-)Quoted 2 lines
Hi,
IMO this is definitely net-next material. A few more comments below.I thoug this is a bug fix. But it looks more suitable to net-next as you said.
Quoted 5 lines
Note that hosts will automatically have this limit lifted to about 270 entries with current defaults, some might be surprised if they have higher limits set and could be left with queues allowing for thousands of entries in a linked list.
Quoted 4 lines
I think this is just a cache list and should be expired soon. The cache creation would also failed if there is no buffer.
But if you like, I can write a patch with sysctl parameter.
I wasn't suggesting the sysctl, I'm actually against it as I've said before when commenting these patches. Just wanted to note that some setups will end up with the possibility of having thousands of entries in a linked list after updating. Unless we change the data structure I guess we just have to live with it, I wouldn't want to add a sysctl or another hard limit that will have to be removed after a while. In the long run we might update the algorithm used for unres entries.
So to make it clear: I'm fine with this patch as it is, definitely an improvement IMO.
Quoted 19 lines
Quoted 13 lines
net/ipv4/ipmr.c static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh, · line 2760
Quoted 10 lines
net/ipv4/ipmr.c +9
-0+static int queue_count (struct mr_table *mrt) +{ +struct list_head *pos; +int count =0 ; + +list_for_each (pos, &mrt->mfc_unres_queue) + count++; +return count; +}I don't think we hold the
mfc_unres_lockhere while walking the unresolved list below inipmr_fill_table().ah, yes, I will fix this.
Thanks
Hangbin
PATCH v2 1/1 ipmr: remove cache_resolve_queue_len
In reply to Hangbin LiuThis is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source addresses(cache_resolve_queue_len) to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.
To resolve this issue, we need either add a sysctl entry to make the cache_resolve_queue_len configurable, or just remove cache_resolve_queue_len directly, as we already have the socket receive queue limits of mrouted socket, pointed by David.
From my side, I'd perfer to remove the cache_resolve_queue_len instead of creating two more(IPv4 and IPv6 version) sysctl entry.
[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343
v2: hold the mfc_unres_lock while walking the unresolved list in queue_count(), as Nikolay Aleksandrov remind.
--- include/linux/mroute_base.h | 2 -- net/ipv4/ipmr.c | 30 +++++++++++++++++++++--------- net/ipv6/ip6mr.c | 10 +++------- 3 files changed, 24 insertions(+), 18 deletions(-)
include/linux/mroute_base.h +0 -2
@@ -235,7 +235,6 @@ struct mr_table_ops {* @mfc_hash: Hash table of all resolved routes for easy lookup * @mfc_cache_list: list of resovled routes for possible traversal * @maxvif: Identifier of highest value vif currently in use -* @cache_resolve_queue_len: current size of unresolved queue * @mroute_do_assert: Whether to inform userspace on wrong ingress * @mroute_do_pim: Whether to receive IGMP PIMv1 * @mroute_reg_vif_num: PIM-device vif index
@@ -252,7 +251,6 @@ struct mr_table {struct rhltable mfc_hash;struct list_head mfc_cache_list;int maxvif; -atomic_t cache_resolve_queue_len;bool mroute_do_assert;bool mroute_do_pim;bool mroute_do_wrvifwhole;
net/ipv4/ipmr.c +21 -9
@@ -744,8 +744,6 @@ static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)struct sk_buff *skb;struct nlmsgerr *e; -atomic_dec (&mrt->cache_resolve_queue_len); -while ((skb =skb_dequeue (&c->_c.mfc_un.unres.unresolved))) {if (ip_hdr (skb)->version ==0 ) { struct nlmsghdr *nlh = skb_pull(skb,
@@ -1133,9 +1131,11 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, }if (!found) { +bool was_empty; +/* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ipmr_cache_alloc_unres ()) ==NULL ) { + c =ipmr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
@@ -1161,11 +1161,11 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,return err; } -atomic_inc (&mrt->cache_resolve_queue_len); + was_empty =list_empty (&mrt->mfc_unres_queue);list_add (&c->_c.list, &mrt->mfc_unres_queue);mroute_netlink_event (mrt, c,RTM_NEWROUTE ); -if (atomic_read (&mrt->cache_resolve_queue_len) ==1 ) +if (was_empty)mod_timer (&mrt->ipmr_expire_timer, c->_c.mfc_un.unres.expires); }
@@ -1272,7 +1272,6 @@ static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,if (uc->mfc_origin == c->mfc_origin && uc->mfc_mcastgrp == c->mfc_mcastgrp) {list_del (&_uc->list); -atomic_dec (&mrt->cache_resolve_queue_len); found = true;break ; }
@@ -1328,7 +1327,7 @@ static void mroute_clean_tables(struct mr_table *mrt, int flags) }if (flags &MRT_FLUSH_MFC ) { -if (atomic_read (&mrt->cache_resolve_queue_len) !=0 ) { +if (!list_empty (&mrt->mfc_unres_queue)) {spin_lock_bh (&mfc_unres_lock);list_for_each_entry_safe (c, tmp, &mrt->mfc_unres_queue, list) {list_del (&c->list);
@@ -2750,9 +2749,22 @@ static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,return ipmr_mfc_delete (tbl, &mfcc, parent); } +static int queue_count (struct mr_table *mrt) +{ +struct list_head *pos; +int count =0 ; + +spin_lock_bh (&mfc_unres_lock); +list_for_each (pos, &mrt->mfc_unres_queue) + count++; +spin_unlock_bh (&mfc_unres_lock); + +return count; +} +static bool ipmr_fill_table (struct mr_table *mrt,struct sk_buff *skb) { -u32 queue_len =atomic_read (&mrt->cache_resolve_queue_len); +u32 queue_len =queue_count (mrt); if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) || nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
net/ipv6/ip6mr.c +3 -7
@@ -768,8 +768,6 @@ static void ip6mr_destroy_unres(struct mr_table *mrt, struct mfc6_cache *c)struct net *net =read_pnet (&mrt->net);struct sk_buff *skb; -atomic_dec (&mrt->cache_resolve_queue_len); -while ((skb =skb_dequeue (&c->_c.mfc_un.unres.unresolved)) !=NULL ) {if (ipv6_hdr (skb)->version ==0 ) { struct nlmsghdr *nlh = skb_pull(skb,
@@ -1148,8 +1146,8 @@ static int ip6mr_cache_unresolved(struct mr_table *mrt, mifi_t mifi,* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ip6mr_cache_alloc_unres ()) ==NULL ) { + c =ip6mr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
@@ -1176,7 +1174,6 @@ static int ip6mr_cache_unresolved(struct mr_table *mrt, mifi_t mifi,return err; } -atomic_inc (&mrt->cache_resolve_queue_len);list_add (&c->_c.list, &mrt->mfc_unres_queue);mr6_netlink_event (mrt, c,RTM_NEWROUTE );
@@ -1468,7 +1465,6 @@ static int ip6mr_mfc_add(struct net *net, struct mr_table *mrt,if (ipv6_addr_equal (&uc->mf6c_origin, &c->mf6c_origin) &&ipv6_addr_equal (&uc->mf6c_mcastgrp, &c->mf6c_mcastgrp)) {list_del (&_uc->list); -atomic_dec (&mrt->cache_resolve_queue_len); found = true;break ; }
@@ -1526,7 +1522,7 @@ static void mroute_clean_tables(struct mr_table *mrt, int flags) }if (flags &MRT6_FLUSH_MFC ) { -if (atomic_read (&mrt->cache_resolve_queue_len) !=0 ) { +if (!list_empty (&mrt->mfc_unres_queue)) {spin_lock_bh (&mfc_unres_lock);list_for_each_entry_safe (c, tmp, &mrt->mfc_unres_queue, list) {list_del (&c->list);
Signature · 1 line
2.19.2
Quoted 21 lines · Hangbin Liu
This is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source
addresses(cache_resolve_queue_len)to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.To resolve this issue, we need either add a sysctl entry to make the
cache_resolve_queue_lenconfigurable, or just removecache_resolve_queue_lendirectly, as we already have the socket receive queue limits of mrouted socket, pointed by David.From my side, I'd perfer to remove the
cache_resolve_queue_leninstead of creating two more(IPv4 and IPv6 version) sysctl entry.(quoted diff) +2
-0+static int queue_count(struct mr_table *mrt) +{
Review
Quoted 10 lines
(quoted diff) +10
-0+ struct list_head *pos; + int count = 0; + + spin_lock_bh(&mfc_unres_lock); + list_for_each(pos, &mrt->mfc_unres_queue) + count++; + spin_unlock_bh(&mfc_unres_lock); + + return count; +}
I guess that even if we remove a limit on the number of items, we probably should keep the atomic counter (no code churn, patch much easier to review...)
Your patch could be a one liner really [1]
Eventually replacing this linear list with an RB-tree, so that we can be on the safe side.
[1]
net/ipv4/ipmr.c +2 -2
@@ -1134,8 +1134,8 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,if (!found) {/* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ipmr_cache_alloc_unres ()) ==NULL ) { + c =ipmr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
Quoted 26 lines · Eric Dumazet
Review
Quoted 12 lines
(quoted diff) +12
-0+static int queue_count(struct mr_table *mrt) +{ + struct list_head *pos; + int count = 0; + + spin_lock_bh(&mfc_unres_lock); + list_for_each(pos, &mrt->mfc_unres_queue) + count++; + spin_unlock_bh(&mfc_unres_lock); + + return count; +}I guess that even if we remove a limit on the number of items, we probably should keep the atomic counter (no code churn, patch much easier to review...)
Your patch could be a one liner really [1]
Eventually replacing this linear list with an RB-tree, so that we can be on the safe side.
[1]
net/ipv4/ipmr.c +0
-0
net/ipv4/ipmr.c static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, · line 1141
Quoted 10 lines
net/ipv4/ipmr.c +2
-2@@ -1134,8 +1134,8 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,if (!found) {/* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ipmr_cache_alloc_unres ()) ==NULL ) { + c =ipmr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
hmm, that looks more clear and easy to review..
Hi David, Alexey,
What do you think? If you also agree, I could post a new version patch.
Thanks
Hangbin
PATCH v3 1/1 ipmr: remove hard code cache_resolve_queue_len limit
In reply to Hangbin LiuThis is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source addresses(cache_resolve_queue_len) to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.
To resolve this issue, we need either add a sysctl entry to make the cache_resolve_queue_len configurable, or just remove cache_resolve_queue_len limit directly, as we already have the socket receive queue limits of mrouted socket, pointed by David.
From my side, I'd perfer to remove the cache_resolve_queue_len limit instead of creating two more(IPv4 and IPv6 version) sysctl entry.
[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343
v3: instead of remove cache_resolve_queue_len totally, let's only remove the hard code limit when allocate the unresolved cache, as Eric Dumazet suggested, so we don't need to re-count it in other places.
v2: hold the mfc_unres_lock while walking the unresolved list in queue_count(), as Nikolay Aleksandrov remind.
--- net/ipv4/ipmr.c | 4 ++-- net/ipv6/ip6mr.c | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-)
net/ipv4/ipmr.c +2 -2
@@ -1134,8 +1134,8 @@ static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,if (!found) {/* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ipmr_cache_alloc_unres ()) ==NULL ) { + c =ipmr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
net/ipv6/ip6mr.c +2 -2
@@ -1148,8 +1148,8 @@ static int ip6mr_cache_unresolved(struct mr_table *mrt, mifi_t mifi,* Create a new entry if allowable */ -if (atomic_read (&mrt->cache_resolve_queue_len) >=10 || - (c =ip6mr_cache_alloc_unres ()) ==NULL ) { + c =ip6mr_cache_alloc_unres (); +if (!c) {spin_unlock_bh (&mfc_unres_lock);kfree_skb (skb);
Signature · 1 line
2.19.2
Re: [PATCHv3 net-next] ipmr: remove hard code cache_resolve_queue_len limit
Quoted 29 lines · Hangbin Liu
This is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source
addresses(cache_resolve_queue_len)to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.To resolve this issue, we need either add a sysctl entry to make the
cache_resolve_queue_lenconfigurable, or just removecache_resolve_queue_lenlimit directly, as we already have the socket receive queue limits of mrouted socket, pointed by David.From my side, I'd perfer to remove the
cache_resolve_queue_lenlimit instead of creating two more(IPv4 and IPv6 version) sysctl entry.[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343v3: instead of remove
cache_resolve_queue_lentotally, let's only remove the hard code limit when allocate the unresolved cache, as Eric Dumazet suggested, so we don't need to re-count it in other places.v2: hold the
mfc_unres_lockwhile walking the unresolved list inqueue_count(), as Nikolay Aleksandrov remind.
Quoted 7 lines
Reported-by Phil Karn <[email protected]>Signed-off-by Hangbin Liu <[email protected]>--- net/ipv4/ipmr.c | 4 ++-- net/ipv6/ip6mr.c | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-)
Please CC all interested parties who have reviewed/commented on previous versions.
I'd also add a Suggested-by tag such as:
Looks good to me, thanks!
Re: [PATCHv3 net-next] ipmr: remove hard code cache_resolve_queue_len limit
In reply to Hangbin LiuFrom: Hangbin Liu <[email protected]>
Date: Fri, 6 Sep 2019 15:36:01 +0800
Quoted 26 lines
This is a re-post of previous patch wrote by David Miller[1].
Phil Karn reported[2] that on busy networks with lots of unresolved multicast routing entries, the creation of new multicast group routes can be extremely slow and unreliable.
The reason is we hard-coded multicast route entries with unresolved source
addresses(cache_resolve_queue_len)to 10. If some multicast route never resolves and the unresolved source addresses increased, there will be no ability to create new multicast route cache.To resolve this issue, we need either add a sysctl entry to make the
cache_resolve_queue_lenconfigurable, or just removecache_resolve_queue_lenlimit directly, as we already have the socket receive queue limits of mrouted socket, pointed by David.From my side, I'd perfer to remove the
cache_resolve_queue_lenlimit instead of creating two more(IPv4 and IPv6 version) sysctl entry.[1] https://lkml.org/lkml/2018/7/22/11
[2] https://lkml.org/lkml/2018/7/21/343v3: instead of remove
cache_resolve_queue_lentotally, let's only remove the hard code limit when allocate the unresolved cache, as Eric Dumazet suggested, so we don't need to re-count it in other places.
Quoted 5 lines
v2: hold the
mfc_unres_lockwhile walking the unresolved list inqueue_count(), as Nikolay Aleksandrov remind.Reported-by Phil Karn <[email protected]>Signed-off-by Hangbin Liu <[email protected]>
Applied to net-next.