Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1733856

[PATCH 4.13 01/52] Revert "net: use lib/percpu_counter API for fragmentation mem accounting"

From Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Newsgroups linux.kernel
Subject [PATCH 4.13 01/52] Revert "net: use lib/percpu_counter API for fragmentation mem accounting"
Date 2017-09-18 11:20 +0200
Message-ID <ur0nK-4ck-161@gated-at.bofh.it> (permalink)
References <ur0nE-4ck-9@gated-at.bofh.it>
Organization linux.* mail to news gateway

Show all headers | View raw


4.13-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Jesper Dangaard Brouer <brouer@redhat.com>


[ Upstream commit fb452a1aa3fd4034d7999e309c5466ff2d7005aa ]

This reverts commit 6d7b857d541ecd1d9bd997c97242d4ef94b19de2.

There is a bug in fragmentation codes use of the percpu_counter API,
that can cause issues on systems with many CPUs.

The frag_mem_limit() just reads the global counter (fbc->count),
without considering other CPUs can have upto batch size (130K) that
haven't been subtracted yet.  Due to the 3MBytes lower thresh limit,
this become dangerous at >=24 CPUs (3*1024*1024/130000=24).

The correct API usage would be to use __percpu_counter_compare() which
does the right thing, and takes into account the number of (online)
CPUs and batch size, to account for this and call __percpu_counter_sum()
when needed.

We choose to revert the use of the lib/percpu_counter API for frag
memory accounting for several reasons:

1) On systems with CPUs > 24, the heavier fully locked
   __percpu_counter_sum() is always invoked, which will be more
   expensive than the atomic_t that is reverted to.

Given systems with more than 24 CPUs are becoming common this doesn't
seem like a good option.  To mitigate this, the batch size could be
decreased and thresh be increased.

2) The add_frag_mem_limit+sub_frag_mem_limit pairs happen on the RX
   CPU, before SKBs are pushed into sockets on remote CPUs.  Given
   NICs can only hash on L2 part of the IP-header, the NIC-RXq's will
   likely be limited.  Thus, a fair chance that atomic add+dec happen
   on the same CPU.

Revert note that commit 1d6119baf061 ("net: fix percpu memory leaks")
removed init_frag_mem_limit() and instead use inet_frags_init_net().
After this revert, inet_frags_uninit_net() becomes empty.

Fixes: 6d7b857d541e ("net: use lib/percpu_counter API for fragmentation mem accounting")
Fixes: 1d6119baf061 ("net: fix percpu memory leaks")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
 include/net/inet_frag.h  |   30 +++++++++---------------------
 net/ipv4/inet_fragment.c |    4 +---
 2 files changed, 10 insertions(+), 24 deletions(-)

--- a/include/net/inet_frag.h
+++ b/include/net/inet_frag.h
@@ -1,14 +1,9 @@
 #ifndef __NET_FRAG_H__
 #define __NET_FRAG_H__
 
-#include <linux/percpu_counter.h>
-
 struct netns_frags {
-	/* The percpu_counter "mem" need to be cacheline aligned.
-	 *  mem.count must not share cacheline with other writers
-	 */
-	struct percpu_counter   mem ____cacheline_aligned_in_smp;
-
+	/* Keep atomic mem on separate cachelines in structs that include it */
+	atomic_t		mem ____cacheline_aligned_in_smp;
 	/* sysctls */
 	int			timeout;
 	int			high_thresh;
@@ -110,11 +105,11 @@ void inet_frags_fini(struct inet_frags *
 
 static inline int inet_frags_init_net(struct netns_frags *nf)
 {
-	return percpu_counter_init(&nf->mem, 0, GFP_KERNEL);
+	atomic_set(&nf->mem, 0);
+	return 0;
 }
 static inline void inet_frags_uninit_net(struct netns_frags *nf)
 {
-	percpu_counter_destroy(&nf->mem);
 }
 
 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
@@ -140,31 +135,24 @@ static inline bool inet_frag_evicting(st
 
 /* Memory Tracking Functions. */
 
-/* The default percpu_counter batch size is not big enough to scale to
- * fragmentation mem acct sizes.
- * The mem size of a 64K fragment is approx:
- *  (44 fragments * 2944 truesize) + frag_queue struct(200) = 129736 bytes
- */
-static unsigned int frag_percpu_counter_batch = 130000;
-
 static inline int frag_mem_limit(struct netns_frags *nf)
 {
-	return percpu_counter_read(&nf->mem);
+	return atomic_read(&nf->mem);
 }
 
 static inline void sub_frag_mem_limit(struct netns_frags *nf, int i)
 {
-	percpu_counter_add_batch(&nf->mem, -i, frag_percpu_counter_batch);
+	atomic_sub(i, &nf->mem);
 }
 
 static inline void add_frag_mem_limit(struct netns_frags *nf, int i)
 {
-	percpu_counter_add_batch(&nf->mem, i, frag_percpu_counter_batch);
+	atomic_add(i, &nf->mem);
 }
 
-static inline unsigned int sum_frag_mem_limit(struct netns_frags *nf)
+static inline int sum_frag_mem_limit(struct netns_frags *nf)
 {
-	return percpu_counter_sum_positive(&nf->mem);
+	return atomic_read(&nf->mem);
 }
 
 /* RFC 3168 support :
--- a/net/ipv4/inet_fragment.c
+++ b/net/ipv4/inet_fragment.c
@@ -234,10 +234,8 @@ evict_again:
 	cond_resched();
 
 	if (read_seqretry(&f->rnd_seqlock, seq) ||
-	    percpu_counter_sum(&nf->mem))
+	    sum_frag_mem_limit(nf))
 		goto evict_again;
-
-	percpu_counter_destroy(&nf->mem);
 }
 EXPORT_SYMBOL(inet_frags_exit_net);
 

Back to linux.kernel | Previous | NextPrevious in thread | Next in thread | Find similar | Unroll thread


Thread

[PATCH 4.13 00/52] 4.13.3-stable review Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 14/52] thunderbolt: Remove superfluous check Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 16/52] thunderbolt: Allow clearing the key Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 30/52] xfs: fix recovery failure when log record header wraps log end Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 04/52] vhost_net: correctly check tx avail during rx busy polling Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 20/52] x86/mm, mm/hwpoison: Clear PRESENT bit for kernel 1:1 mappings of poison pages Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 35/52] xfs: evict all inodes involved with log redo item Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 21/52] ovl: fix false positive ESTALE on lookup Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 37/52] xfs: open-code xfs_buf_item_dirty() Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 11/52] ip_tunnel: fix setting ttl and tos value in collect_md mode Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 01/52] Revert "net: use lib/percpu_counter API for fragmentation mem accounting" Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 12/52] f2fs: let fill_super handle roll-forward errors Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 15/52] thunderbolt: Make key root-only accessible Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 47/52] xfs: fix incorrect log_flushed on fsync Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 40/52] xfs: refactor buffer logging into buffer dirtying helper Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 52/52] md/raid5: release/flush io in raid5_do_work() Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 11:20 +0200
  [PATCH 4.13 08/52] ipv6: fix typo in fib6_net_exit() Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 50/52] xfs: use kmem_free to free return value of kmem_zalloc Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 28/52] xfs: Add infrastructure needed for error propagation during buffer IO failure Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 24/52] libnvdimm, btt: check memory allocation failure Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 38/52] xfs: remove unnecessary dirty bli format check for ordered bufs Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 46/52] xfs: disable per-inode DAX flag Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 23/52] idr: remove WARN_ON_ONCE() when trying to replace negative ID Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 31/52] xfs: always verify the log tail during recovery Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 41/52] xfs: dont log dirty ranges for ordered buffers Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 43/52] xfs: move bmbt owner change to last step of extent swap Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 48/52] xfs: dont set v3 xflags for v2 inodes Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 13/52] f2fs: check hot_data for roll-forward recovery Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 33/52] xfs: handle -EFSCORRUPTED during head/tail verification Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 39/52] xfs: ordered buffer log items are never formatted Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 10/52] tcp: fix a request socket leak Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 34/52] xfs: stop searching for free slots in an inode chunk when there are none Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 09/52] sctp: fix missing wake ups in some situations Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 42/52] xfs: skip bmbt block ino validation during owner change Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 51/52] md/raid1/10: reset bio allocated from mempool Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 36/52] xfs: check for race with xfs_reclaim_inode() in xfs_ifree_cluster() Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 25/52] libnvdimm: fix integer overflow static analysis warning Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 32/52] xfs: fix log recovery corruption error due to tail overwrite Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 29/52] xfs: Properly retry failed inode items in case of error during buffer writeback Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:00 +0200
  [PATCH 4.13 03/52] gianfar: Fix Tx flow control deactivation Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:10 +0200
  [PATCH 4.13 02/52] Revert "net: fix percpu memory leaks" Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:10 +0200
  [PATCH 4.13 17/52] x86/fsgsbase/64: Fully initialize FS and GS state in start_thread_common Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:10 +0200
  [PATCH 4.13 06/52] udp: drop head states only when all skb references are gone Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-18 12:10 +0200
  Re: [PATCH 4.13 00/52] 4.13.3-stable review Guenter Roeck <linux@roeck-us.net> - 2017-09-18 21:30 +0200
  Re: [PATCH 4.13 00/52] 4.13.3-stable review Shuah Khan <shuahkh@osg.samsung.com> - 2017-09-18 22:20 +0200
    Re: [PATCH 4.13 00/52] 4.13.3-stable review Greg Kroah-Hartman <gregkh@linuxfoundation.org> - 2017-09-19 08:40 +0200

csiph-web