From 9cd3e072b0be17446e37d7414eac8a3499e0601e Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Sun, 29 Nov 2015 20:03:10 -0800 Subject: net: rename SOCK_ASYNC_NOSPACE and SOCK_ASYNC_WAITDATA This patch is a cleanup to make following patch easier to review. Goal is to move SOCK_ASYNC_NOSPACE and SOCK_ASYNC_WAITDATA from (struct socket)->flags to a (struct socket_wq)->flags to benefit from RCU protection in sock_wake_async() To ease backports, we rename both constants. Two new helpers, sk_set_bit(int nr, struct sock *sk) and sk_clear_bit(int net, struct sock *sk) are added so that following patch can change their implementation. Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/socket.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'net/socket.c') diff --git a/net/socket.c b/net/socket.c index dd2c247c99e3..16be908205fc 100644 --- a/net/socket.c +++ b/net/socket.c @@ -1072,11 +1072,11 @@ int sock_wake_async(struct socket *sock, int how, int band) } switch (how) { case SOCK_WAKE_WAITD: - if (test_bit(SOCK_ASYNC_WAITDATA, &sock->flags)) + if (test_bit(SOCKWQ_ASYNC_WAITDATA, &sock->flags)) break; goto call_kill; case SOCK_WAKE_SPACE: - if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags)) + if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sock->flags)) break; /* fall through */ case SOCK_WAKE_IO: -- cgit v1.2.3 From ceb5d58b217098a657f3850b7a2640f995032e62 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Sun, 29 Nov 2015 20:03:11 -0800 Subject: net: fix sock_wake_async() rcu protection Dmitry provided a syzkaller (http://github.com/google/syzkaller) triggering a fault in sock_wake_async() when async IO is requested. Said program stressed af_unix sockets, but the issue is generic and should be addressed in core networking stack. The problem is that by the time sock_wake_async() is called, we should not access the @flags field of 'struct socket', as the inode containing this socket might be freed without further notice, and without RCU grace period. We already maintain an RCU protected structure, "struct socket_wq" so moving SOCKWQ_ASYNC_NOSPACE & SOCKWQ_ASYNC_WAITDATA into it is the safe route. It also reduces number of cache lines needing dirtying, so might provide a performance improvement anyway. In followup patches, we might move remaining flags (SOCK_NOSPACE, SOCK_PASSCRED, SOCK_PASSSEC) to save 8 bytes and let 'struct socket' being mostly read and let it being shared between cpus. Reported-by: Dmitry Vyukov Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/socket.c | 21 +++++++-------------- 1 file changed, 7 insertions(+), 14 deletions(-) (limited to 'net/socket.c') diff --git a/net/socket.c b/net/socket.c index 16be908205fc..456fadb3d819 100644 --- a/net/socket.c +++ b/net/socket.c @@ -1056,27 +1056,20 @@ static int sock_fasync(int fd, struct file *filp, int on) return 0; } -/* This function may be called only under socket lock or callback_lock or rcu_lock */ +/* This function may be called only under rcu_lock */ -int sock_wake_async(struct socket *sock, int how, int band) +int sock_wake_async(struct socket_wq *wq, int how, int band) { - struct socket_wq *wq; - - if (!sock) - return -1; - rcu_read_lock(); - wq = rcu_dereference(sock->wq); - if (!wq || !wq->fasync_list) { - rcu_read_unlock(); + if (!wq || !wq->fasync_list) return -1; - } + switch (how) { case SOCK_WAKE_WAITD: - if (test_bit(SOCKWQ_ASYNC_WAITDATA, &sock->flags)) + if (test_bit(SOCKWQ_ASYNC_WAITDATA, &wq->flags)) break; goto call_kill; case SOCK_WAKE_SPACE: - if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sock->flags)) + if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags)) break; /* fall through */ case SOCK_WAKE_IO: @@ -1086,7 +1079,7 @@ call_kill: case SOCK_WAKE_URG: kill_fasync(&wq->fasync_list, SIGURG, band); } - rcu_read_unlock(); + return 0; } EXPORT_SYMBOL(sock_wake_async); -- cgit v1.2.3 From 130ed5d105dde141e7fe60d5440aa53e0a84f13b Mon Sep 17 00:00:00 2001 From: "tadeusz.struk@intel.com" Date: Tue, 15 Dec 2015 10:46:17 -0800 Subject: net: fix uninitialized variable issue msg_iocb needs to be initialized on the recv/recvfrom path. Otherwise afalg will wrongly interpret it as an async call. Cc: stable@vger.kernel.org Reported-by: Harald Freudenberger Signed-off-by: Tadeusz Struk Signed-off-by: David S. Miller --- net/socket.c | 1 + 1 file changed, 1 insertion(+) (limited to 'net/socket.c') diff --git a/net/socket.c b/net/socket.c index 456fadb3d819..29822d6dd91e 100644 --- a/net/socket.c +++ b/net/socket.c @@ -1695,6 +1695,7 @@ SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size, msg.msg_name = addr ? (struct sockaddr *)&address : NULL; /* We assume all kernel code knows the size of sockaddr_storage */ msg.msg_namelen = 0; + msg.msg_iocb = NULL; if (sock->file->f_flags & O_NONBLOCK) flags |= MSG_DONTWAIT; err = sock_recvmsg(sock, &msg, iov_iter_count(&msg.msg_iter), flags); -- cgit v1.2.3 From 574aab1e02837927e3c94193eedf94128ad10b6d Mon Sep 17 00:00:00 2001 From: Nicolai Stange Date: Tue, 29 Dec 2015 13:29:55 +0100 Subject: net, socket, socket_wq: fix missing initialization of flags Commit ceb5d58b2170 ("net: fix sock_wake_async() rcu protection") from the current 4.4 release cycle introduced a new flags member in struct socket_wq and moved SOCKWQ_ASYNC_NOSPACE and SOCKWQ_ASYNC_WAITDATA from struct socket's flags member into that new place. Unfortunately, the new flags field is never initialized properly, at least not for the struct socket_wq instance created in sock_alloc_inode(). One particular issue I encountered because of this is that my GNU Emacs failed to draw anything on my desktop -- i.e. what I got is a transparent window, including the title bar. Bisection lead to the commit mentioned above and further investigation by means of strace told me that Emacs is indeed speaking to my Xorg through an O_ASYNC AF_UNIX socket. This is reproducible 100% of times and the fact that properly initializing the struct socket_wq ->flags fixes the issue leads me to the conclusion that somehow SOCKWQ_ASYNC_WAITDATA got set in the uninitialized ->flags, preventing my Emacs from receiving any SIGIO's due to data becoming available and it got stuck. Make sock_alloc_inode() set the newly created struct socket_wq's ->flags member to zero. Fixes: ceb5d58b2170 ("net: fix sock_wake_async() rcu protection") Signed-off-by: Nicolai Stange Acked-by: Eric Dumazet Signed-off-by: David S. Miller --- net/socket.c | 1 + 1 file changed, 1 insertion(+) (limited to 'net/socket.c') diff --git a/net/socket.c b/net/socket.c index 29822d6dd91e..d730ef9dfbf0 100644 --- a/net/socket.c +++ b/net/socket.c @@ -257,6 +257,7 @@ static struct inode *sock_alloc_inode(struct super_block *sb) } init_waitqueue_head(&wq->wait); wq->fasync_list = NULL; + wq->flags = 0; RCU_INIT_POINTER(ei->socket.wq, wq); ei->socket.state = SS_UNCONNECTED; -- cgit v1.2.3 From a78cb84c62c427807d917c5aa8797740f00b0bbe Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Fri, 8 Jan 2016 08:37:20 -0800 Subject: net: add scheduling point in recvmmsg/sendmmsg Applications often have to reduce number of datagrams they receive or send per system call to avoid starvation problems. Really the kernel should take care of this by using cond_resched(), so that applications can experiment bigger batch sizes. Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- net/socket.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'net/socket.c') diff --git a/net/socket.c b/net/socket.c index d730ef9dfbf0..91c2de6f5020 100644 --- a/net/socket.c +++ b/net/socket.c @@ -2041,6 +2041,7 @@ int __sys_sendmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen, if (err) break; ++datagrams; + cond_resched(); } fput_light(sock->file, fput_needed); @@ -2236,6 +2237,7 @@ int __sys_recvmmsg(int fd, struct mmsghdr __user *mmsg, unsigned int vlen, /* Out of band data, return right away */ if (msg_sys.msg_flags & MSG_OOB) break; + cond_resched(); } out_put: -- cgit v1.2.3 From 5d097056c9a017a3b720849efb5432f37acabbac Mon Sep 17 00:00:00 2001 From: Vladimir Davydov Date: Thu, 14 Jan 2016 15:18:21 -0800 Subject: kmemcg: account certain kmem allocations to memcg Mark those kmem allocations that are known to be easily triggered from userspace as __GFP_ACCOUNT/SLAB_ACCOUNT, which makes them accounted to memcg. For the list, see below: - threadinfo - task_struct - task_delay_info - pid - cred - mm_struct - vm_area_struct and vm_region (nommu) - anon_vma and anon_vma_chain - signal_struct - sighand_struct - fs_struct - files_struct - fdtable and fdtable->full_fds_bits - dentry and external_name - inode for all filesystems. This is the most tedious part, because most filesystems overwrite the alloc_inode method. The list is far from complete, so feel free to add more objects. Nevertheless, it should be close to "account everything" approach and keep most workloads within bounds. Malevolent users will be able to breach the limit, but this was possible even with the former "account everything" approach (simply because it did not account everything in fact). [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Vladimir Davydov Acked-by: Johannes Weiner Acked-by: Michal Hocko Cc: Tejun Heo Cc: Greg Thelen Cc: Christoph Lameter Cc: Pekka Enberg Cc: David Rientjes Cc: Joonsoo Kim Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- net/socket.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'net/socket.c') diff --git a/net/socket.c b/net/socket.c index 91c2de6f5020..c044d1e8508c 100644 --- a/net/socket.c +++ b/net/socket.c @@ -294,7 +294,7 @@ static int init_inodecache(void) 0, (SLAB_HWCACHE_ALIGN | SLAB_RECLAIM_ACCOUNT | - SLAB_MEM_SPREAD), + SLAB_MEM_SPREAD | SLAB_ACCOUNT), init_once); if (sock_inode_cachep == NULL) return -ENOMEM; -- cgit v1.2.3