NFSD socket send-path changes
Subsystems (4)
cover NFSD socket send-path changes
Hi-
(cc: linux-nfs only)
This is a set of patches that change the server-side socket send- path to use the kernel socket bvec API. This is a pre-requisite for supporting RPC-on-TLS.
At the same time I've introduced a number of tracepoints that are specific to the server socket transport code, and performed some other clean ups in the area.
---
Chuck Lever (8):
SUNRPC: Remove "#include <trace/events/skb.h>"
SUNRPC: Add more svcsock tracepoints
SUNRPC: Replace dprintk call sites in TCP state change callouts
SUNRPC: Trace server-side rpcbind registration events
SUNRPC: Clean up: Rename svc_sock::sk_reclen
SUNRPC: Restructure svc_tcp_recv_record()
SUNRPC: Refactor svc_recvfrom()
SUNRPC: Restructure svc_udp_recvfrom()
include/linux/sunrpc/svc.h | 1 + include/linux/sunrpc/svcsock.h | 6 +- include/trace/events/sunrpc.h | 270 +++++++++++++++++++++++ net/sunrpc/svc.c | 15 +- net/sunrpc/svcsock.c | 380 ++++++++++++++++----------------- 5 files changed, 456 insertions(+), 216 deletions(-)
--
Chuck Lever
PATCH 1/8 SUNRPC: Remove "#include <trace/events/skb.h>"
Clean up: Commit 850cbaddb52d ("udp: use it's own memory accounting schema") removed the last skb trace event from svcsock.c, so it is no longer necessary to include trace/events/skb.h.
--- net/sunrpc/svcsock.c | 1 - 1 file changed, 1 deletion(-)
net/sunrpc/svcsock.c +0 -1
@@ -45,7 +45,6 @@#include <net/tcp_states.h> #include <linux/uaccess.h> #include <asm/ioctls.h> -#include <trace/events/skb.h> #include <linux/sunrpc/types.h> #include <linux/sunrpc/clnt.h>
PATCH 2/8 SUNRPC: Add more svcsock tracepoints
In reply to Chuck LeverIn addition to tracing recently-updated socket sendto events, this commit adds a trace event class that can be used for additional svcsock-related tracepoints in subsequent patches.
Commit 850cbaddb52d ("udp: use it's own memory accounting schema") removed the last skb trace event from svcsock.c, so it is no longer necessary to include trace/events/skb.h.
--- include/trace/events/sunrpc.h | 97 +++++++++++++++++++++++++++++++++++++++++ net/sunrpc/svcsock.c | 30 ++++--------- 2 files changed, 107 insertions(+), 20 deletions(-)
include/trace/events/sunrpc.h +97 -0
@@ -14,6 +14,39 @@#include <linux/net.h> #include <linux/tracepoint.h> +TRACE_DEFINE_ENUM (SOCK_STREAM ); +TRACE_DEFINE_ENUM (SOCK_DGRAM ); +TRACE_DEFINE_ENUM (SOCK_RAW ); +TRACE_DEFINE_ENUM (SOCK_RDM ); +TRACE_DEFINE_ENUM (SOCK_SEQPACKET ); +TRACE_DEFINE_ENUM (SOCK_DCCP ); +TRACE_DEFINE_ENUM (SOCK_PACKET ); + +#define show_socket_type (type) \ + __print_symbolic(type, \ + { SOCK_STREAM, "STREAM" }, \ + { SOCK_DGRAM, "DGRAM" }, \ + { SOCK_RAW, "RAW" }, \ + { SOCK_RDM, "RDM" }, \ + { SOCK_SEQPACKET, "SEQPACKET" }, \ + { SOCK_DCCP, "DCCP" }, \ + { SOCK_PACKET, "PACKET" }) + +/* This list is known to be incomplete, add new enums as needed. */ +TRACE_DEFINE_ENUM (AF_UNSPEC ); +TRACE_DEFINE_ENUM (AF_UNIX ); +TRACE_DEFINE_ENUM (AF_LOCAL ); +TRACE_DEFINE_ENUM (AF_INET ); +TRACE_DEFINE_ENUM (AF_INET6 ); + +#define rpc_show_address_family (family) \ + __print_symbolic(family, \ + { AF_UNSPEC, "AF_UNSPEC" }, \ + { AF_UNIX, "AF_UNIX" }, \ + { AF_LOCAL, "AF_LOCAL" }, \ + { AF_INET, "AF_INET" }, \ + { AF_INET6, "AF_INET6" }) + DECLARE_EVENT_CLASS(xdr_buf_class, TP_PROTO( const struct xdr_buf *xdr
@@ -1400,6 +1433,70 @@ DEFINE_SVC_DEFERRED_EVENT(drop);DEFINE_SVC_DEFERRED_EVENT (queue);DEFINE_SVC_DEFERRED_EVENT (recv); +TRACE_EVENT (svcsock_new_socket, +TP_PROTO ( + const struct socket *socket + ), + +TP_ARGS (socket), + +TP_STRUCT__entry ( + __field(unsigned long, type) + __field(unsigned long, family) +__field (bool, listener) + ), + +TP_fast_assign ( + __entry->type = socket->type; + __entry->family = socket->sk->sk_family; + __entry->listener = (socket->sk->sk_state == TCP_LISTEN); + ), + +TP_printk ("type=%s family=%s%s" , +show_socket_type (__entry->type), +rpc_show_address_family (__entry->family), + __entry->listener ?" (listener)" :"" + ) +); + +DECLARE_EVENT_CLASS (svcsock_class, +TP_PROTO ( + const struct svc_xprt *xprt, + ssize_t result + ), + +TP_ARGS (xprt, result), + +TP_STRUCT__entry ( + __field(ssize_t, result) + __field(unsigned long, flags) +__string (addr, xprt->xpt_remotebuf) + ), + +TP_fast_assign ( + __entry->result = result; + __entry->flags = xprt->xpt_flags; + __assign_str(addr, xprt->xpt_remotebuf); + ), + +TP_printk ("addr=%s result=%zd flags=%s" ,__get_str (addr), + __entry->result,show_svc_xprt_flags (__entry->flags) + ) +); + +#define DEFINE_SVCSOCK_EVENT (name) \ + DEFINE_EVENT(svcsock_class, svcsock_##name, \ + TP_PROTO( \ + const struct svc_xprt *xprt, \ + ssize_t result \ + ), \ + TP_ARGS(xprt, result)) + +DEFINE_SVCSOCK_EVENT (udp_send); +DEFINE_SVCSOCK_EVENT (tcp_send); +DEFINE_SVCSOCK_EVENT (data_ready); +DEFINE_SVCSOCK_EVENT (write_space); + DECLARE_EVENT_CLASS(cache_event, TP_PROTO( const struct cache_detail *cd,
net/sunrpc/svcsock.c +10 -20
@@ -54,6 +54,8 @@#include <linux/sunrpc/stats.h> #include <linux/sunrpc/xprt.h> +#include <trace/events/sunrpc.h> +#include "socklib.h" #include "sunrpc.h"
@@ -281,13 +283,10 @@ static void svc_data_ready(struct sock *sk)struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;if (svsk) { -dprintk ("svc: socket %p(inet %p), busy=%d\n" , - svsk, sk, -test_bit (XPT_BUSY , &svsk->sk_xprt.xpt_flags)); -/* Refer to svc_setup_socket() for details. */ rmb (); svsk->sk_odata (sk); +trace_svcsock_data_ready (&svsk->sk_xprt,0 );if (!test_and_set_bit (XPT_DATA , &svsk->sk_xprt.xpt_flags))svc_xprt_enqueue (&svsk->sk_xprt); }
@@ -301,11 +300,9 @@ static void svc_write_space(struct sock *sk)struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);if (svsk) { -dprintk ("svc: socket %p(inet %p), write_space busy=%d\n" , - svsk, sk,test_bit (XPT_BUSY , &svsk->sk_xprt.xpt_flags)); -/* Refer to svc_setup_socket() for details. */ rmb (); +trace_svcsock_write_space (&svsk->sk_xprt,0 ); svsk->sk_owspace (sk);svc_xprt_enqueue (&svsk->sk_xprt); }
@@ -545,6 +542,7 @@ static int svc_udp_sendto(struct svc_rqst *rqstp) err =xprt_sock_sendmsg (svsk->sk_sock, &msg, xdr,0 ,0 , &sent);xdr_free_bvec (xdr); } +trace_svcsock_udp_send (xprt, err);mutex_unlock (&xprt->xpt_mutex);if (err <0 )
@@ -616,7 +614,7 @@ static struct svc_xprt_class svc_udp_class = {static void svc_udp_init (struct svc_sock *svsk,struct svc_serv *serv) { -int err, level, optname, one =1 ; +int level, optname, one =1 ;svc_xprt_init (sock_net (svsk->sk_sock->sk), &svc_udp_class, &svsk->sk_xprt, serv);
@@ -647,9 +645,8 @@ static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)default :BUG (); } - err =kernel_setsockopt (svsk->sk_sock, level, optname, - (char *)&one,sizeof (one)); -dprintk ("svc: kernel_setsockopt returned %d\n" , err); +kernel_setsockopt (svsk->sk_sock, level, optname, (char *)&one, +sizeof (one)); }/*
@@ -1100,6 +1097,7 @@ static int svc_tcp_sendto(struct svc_rqst *rqstp) goto out_notconn; err =xprt_sock_sendmsg (svsk->sk_sock, &msg, xdr,0 , marker, &sent);xdr_free_bvec (xdr); +trace_svcsock_tcp_send (xprt, err <0 ? err : sent);if (err <0 || sent != (xdr->len +sizeof (marker))) goto out_close;mutex_unlock (&xprt->xpt_mutex);
@@ -1170,13 +1168,11 @@ static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)set_bit (XPT_CACHE_AUTH , &svsk->sk_xprt.xpt_flags);set_bit (XPT_CONG_CTRL , &svsk->sk_xprt.xpt_flags);if (sk->sk_state ==TCP_LISTEN ) { -dprintk ("setting up TCP socket for listening\n" );strcpy (svsk->sk_xprt.xpt_remotebuf,"listener" );set_bit (XPT_LISTENER , &svsk->sk_xprt.xpt_flags); sk->sk_data_ready = svc_tcp_listen_data_ready;set_bit (XPT_CONN , &svsk->sk_xprt.xpt_flags); }else { -dprintk ("setting up TCP socket for reading\n" ); sk->sk_state_change = svc_tcp_state_change; sk->sk_data_ready = svc_data_ready; sk->sk_write_space = svc_write_space;
@@ -1226,7 +1222,6 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,int pmap_register = !(flags &SVC_SOCK_ANONYMOUS );int err =0 ; -dprintk ("svc: svc_setup_socket %p\n" , sock); svsk =kzalloc (sizeof (*svsk),GFP_KERNEL );if (!svsk)return ERR_PTR (-ENOMEM );
@@ -1263,12 +1258,7 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,else svc_tcp_init (svsk, serv); -dprintk ("svc: svc_setup_socket created %p (inet %p), " -"listen %d close %d\n" , - svsk, svsk->sk_sk, -test_bit (XPT_LISTENER , &svsk->sk_xprt.xpt_flags), -test_bit (XPT_CLOSE , &svsk->sk_xprt.xpt_flags)); - +trace_svcsock_new_socket (sock);return svsk; }
PATCH 3/8 SUNRPC: Replace dprintk call sites in TCP state change callouts
In reply to Chuck LeverReport TCP socket state changes and accept failures via tracepoints, replacing dprintk() call sites.
No tracepoint is added in svc_tcp_listen_data_ready. There's no information available there that isn't also reported by the svcsock_new_socket and the accept failure tracepoints.
--- include/trace/events/sunrpc.h | 67 +++++++++++++++++++++++++++++++++++++++++ net/sunrpc/svcsock.c | 35 ++++----------------- 2 files changed, 73 insertions(+), 29 deletions(-)
include/trace/events/sunrpc.h +67 -0
@@ -1497,6 +1497,73 @@ DEFINE_SVCSOCK_EVENT(tcp_send);DEFINE_SVCSOCK_EVENT (data_ready);DEFINE_SVCSOCK_EVENT (write_space); +TRACE_EVENT (svcsock_tcp_state, +TP_PROTO ( + const struct svc_xprt *xprt, + const struct socket *socket + ), + +TP_ARGS (xprt, socket), + +TP_STRUCT__entry ( + __field(unsigned long, socket_state) + __field(unsigned long, sock_state) + __field(unsigned long, flags) +__string (addr, xprt->xpt_remotebuf) + ), + +TP_fast_assign ( + __entry->socket_state = socket->state; + __entry->sock_state = socket->sk->sk_state; + __entry->flags = xprt->xpt_flags; + __assign_str(addr, xprt->xpt_remotebuf); + ), + +TP_printk ("addr=%s state=%s sk_state=%s flags=%s" ,__get_str (addr), +rpc_show_socket_state (__entry->socket_state), +rpc_show_sock_state (__entry->sock_state), +show_svc_xprt_flags (__entry->flags) + ) +); + +DECLARE_EVENT_CLASS (svcsock_accept_class, +TP_PROTO ( + const struct svc_xprt *xprt, + const char *service, + long status + ), + +TP_ARGS (xprt, service, status), + +TP_STRUCT__entry ( + __field(long, status) + __string(service, service) + __array(unsigned char, addr, sizeof(struct sockaddr_in6)) + ), + +TP_fast_assign ( + __entry->status = status; + __assign_str(service, service); + memcpy(__entry->addr, &xprt->xpt_local, sizeof(__entry->addr)); + ), + +TP_printk ("listener=%pISpc service=%s status=%ld" , + __entry->addr,__get_str (service), __entry->status + ) +); + +#define DEFINE_ACCEPT_EVENT (name) \ + DEFINE_EVENT(svcsock_accept_class, svcsock_##name##_err, \ + TP_PROTO( \ + const struct svc_xprt *xprt, \ + const char *service, \ + long status \ + ), \ + TP_ARGS(xprt, service, status)) + +DEFINE_ACCEPT_EVENT (accept); +DEFINE_ACCEPT_EVENT (getpeername); + DECLARE_EVENT_CLASS(cache_event, TP_PROTO( const struct cache_detail *cd,
net/sunrpc/svcsock.c +6 -29
@@ -657,9 +657,6 @@ static void svc_tcp_listen_data_ready(struct sock *sk) {struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; -dprintk ("svc: socket %p TCP (listen) state change %d\n" , - sk, sk->sk_state); -if (svsk) {/* Refer to svc_setup_socket() for details. */ rmb ();
@@ -680,8 +677,7 @@ static void svc_tcp_listen_data_ready(struct sock *sk)if (svsk) {set_bit (XPT_CONN , &svsk->sk_xprt.xpt_flags);svc_xprt_enqueue (&svsk->sk_xprt); - }else -printk ("svc: socket %p: no user data\n" , sk); + } } }
@@ -692,15 +688,11 @@ static void svc_tcp_state_change(struct sock *sk) {struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; -dprintk ("svc: socket %p TCP (connected) state change %d (svsk %p)\n" , - sk, sk->sk_state, sk->sk_user_data); - -if (!svsk) -printk ("svc: socket %p: no user data\n" , sk); -else { +if (svsk) {/* Refer to svc_setup_socket() for details. */ rmb (); svsk->sk_ostate (sk); +trace_svcsock_tcp_state (&svsk->sk_xprt, svsk->sk_sock);if (sk->sk_state !=TCP_ESTABLISHED ) {set_bit (XPT_CLOSE , &svsk->sk_xprt.xpt_flags);svc_xprt_enqueue (&svsk->sk_xprt);
@@ -721,7 +713,6 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)struct socket *newsock;struct svc_sock *newsvsk;int err, slen; - RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);if (!sock)return NULL ;
@@ -735,30 +726,18 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)else if (err != -EAGAIN )net_warn_ratelimited ("%s: accept failed (err %d)!\n" , serv->sv_name, -err); +trace_svcsock_accept_err (xprt, serv->sv_name, err);return NULL ; }set_bit (XPT_CONN , &svsk->sk_xprt.xpt_flags); err =kernel_getpeername (newsock, sin);if (err <0 ) { -net_warn_ratelimited ("%s: peername failed (err %d)!\n" , - serv->sv_name, -err); +trace_svcsock_getpeername_err (xprt, serv->sv_name, err); goto failed;/* aborted connection or whatever */ } slen = err; -/* Ideally, we would want to reject connections from unauthorized -* hosts here, but when we get encryption, the IP of the host won't -* tell us anything. For now just warn about unpriv connections. -*/ -if (!svc_port_is_privileged (sin)) { -dprintk ("%s: connect from unprivileged port: %s\n" , - serv->sv_name, -__svc_print_addr (sin, buf,sizeof (buf))); - } -dprintk ("%s: connect from %s\n" , serv->sv_name, -__svc_print_addr (sin, buf,sizeof (buf))); -/* Reset the inherited callbacks before calling svc_setup_socket */ newsock->sk->sk_state_change = svsk->sk_ostate; newsock->sk->sk_data_ready = svsk->sk_odata;
@@ -776,10 +755,8 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)svc_xprt_set_remote (&newsvsk->sk_xprt, sin, slen); err =kernel_getsockname (newsock, sin); slen = err; -if (unlikely (err <0 )) { -dprintk ("svc_tcp_accept: kernel_getsockname error %d\n" , -err); +if (unlikely (err <0 )) slen = offsetof(struct sockaddr , sa_data); - }svc_xprt_set_local (&newsvsk->sk_xprt, sin, slen);if (sock_is_loopback (newsock->sk))
PATCH 4/8 SUNRPC: Trace server-side rpcbind registration events
In reply to Chuck Lever--- include/trace/events/sunrpc.h | 80 +++++++++++++++++++++++++++++++++++++++++ net/sunrpc/svc.c | 15 ++------ 2 files changed, 83 insertions(+), 12 deletions(-)
include/trace/events/sunrpc.h +80 -0
@@ -1597,6 +1597,86 @@ DEFINE_CACHE_EVENT(cache_entry_update);DEFINE_CACHE_EVENT (cache_entry_make_negative);DEFINE_CACHE_EVENT (cache_entry_no_listener); +DECLARE_EVENT_CLASS (register_class, +TP_PROTO ( + const char *program, + const u32 version, + const int family, + const unsigned short protocol, + const unsigned short port, + int error + ), + +TP_ARGS (program, version, family, protocol, port, error), + +TP_STRUCT__entry ( + __field(u32, version) + __field(unsigned long, family) + __field(unsigned short, protocol) + __field(unsigned short, port) + __field(int, error) +__string (program, program) + ), + +TP_fast_assign ( + __entry->version = version; + __entry->family = family; + __entry->protocol = protocol; + __entry->port = port; + __entry->error = error; + __assign_str(program, program); + ), + +TP_printk ("program=%sv%u proto=%s port=%u family=%s error=%d" , +__get_str (program), __entry->version, + __entry->protocol ==IPPROTO_UDP ?"udp" :"tcp" , + __entry->port,rpc_show_address_family (__entry->family), + __entry->error + ) +); + +#define DEFINE_REGISTER_EVENT (name) \ + DEFINE_EVENT(register_class, svc_##name, \ + TP_PROTO( \ + const char *program, \ + const u32 version, \ + const int family, \ + const unsigned short protocol, \ + const unsigned short port, \ + int error \ + ), \ + TP_ARGS(program, version, family, protocol, \ + port, error)) + +DEFINE_REGISTER_EVENT (register); +DEFINE_REGISTER_EVENT (noregister); + +TRACE_EVENT (svc_unregister, +TP_PROTO ( + const char *program, + const u32 version, + int error + ), + +TP_ARGS (program, version, error), + +TP_STRUCT__entry ( + __field(u32, version) + __field(int, error) +__string (program, program) + ), + +TP_fast_assign ( + __entry->version = version; + __entry->error = error; + __assign_str(program, program); + ), + +TP_printk ("program=%sv%u error=%d" , +__get_str (program), __entry->version, __entry->error + ) +); +#endif /* _TRACE_SUNRPC_H */ #include <trace/define_trace.h>
net/sunrpc/svc.c +3 -12
@@ -991,6 +991,7 @@ static int __svc_register(struct net *net, const char *progname,#endif } +trace_svc_register (progname, version, protocol, port, family, error);return error; }
@@ -1000,11 +1001,6 @@ int svc_rpcbind_set_version(struct net *net, unsigned short proto, unsigned short port) { - dprintk("svc: svc_register(%sv%d, %s, %u, %u)\n", - progp->pg_name, version, - proto ==IPPROTO_UDP ?"udp" :"tcp" , - port, family); -return __svc_register (net, progp->pg_name, progp->pg_prog, version, family, proto, port);
@@ -1024,11 +1020,7 @@ int svc_generic_rpcbind_set(struct net *net,return 0 ;if (vers->vs_hidden) { -dprintk ("svc: svc_register(%sv%d, %s, %u, %u)" -" (but not telling portmap)\n" , - progp->pg_name, version, - proto ==IPPROTO_UDP ?"udp" :"tcp" , - port, family); +trace_svc_noregister (progp->pg_name, version, proto, port, family,0 );return 0 ; }
@@ -1106,8 +1098,7 @@ static void __svc_unregister(struct net *net, const u32 program, const u32 versiif (error == -EPROTONOSUPPORT ) error =rpcb_register (net, program, version,0 ,0 ); -dprintk ("svc: %s(%sv%u), error %d\n" , - __func__, progname, version, error); +trace_svc_unregister (progname, version, error); }/*
PATCH 5/8 SUNRPC: Clean up: Rename svc_sock::sk_reclen
In reply to Chuck LeverI find the name of the svc_sock::sk_reclen field confusing, so I've changed it to better reflect its function. This field is not read directly to get the record length. Rather, it is a buffer containing a record marker that needs to be decoded.
--- include/linux/sunrpc/svcsock.h | 6 +++--- net/sunrpc/svcsock.c | 6 +++--- 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/sunrpc/svcsock.h +3 -3
@@ -28,7 +28,7 @@ struct svc_sock {/* private TCP part */ /* On-the-wire fragment header: */ -__be32 sk_reclen; +__be32 sk_marker;/* As we receive a record, this includes the length received so * far (including the fragment header): */ u32 sk_tcplen;
@@ -41,12 +41,12 @@ struct svc_sock {static inline u32 svc_sock_reclen (struct svc_sock *svsk) { -return ntohl (svsk->sk_reclen) &RPC_FRAGMENT_SIZE_MASK ; +return be32_to_cpu (svsk->sk_marker) &RPC_FRAGMENT_SIZE_MASK ; }static inline u32 svc_sock_final_rec (struct svc_sock *svsk) { -return ntohl (svsk->sk_reclen) &RPC_LAST_STREAM_FRAGMENT ; +return be32_to_cpu (svsk->sk_marker) &RPC_LAST_STREAM_FRAGMENT ; }/*
net/sunrpc/svcsock.c +3 -3
@@ -841,7 +841,7 @@ static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)struct kvec iov; want =sizeof (rpc_fraghdr) - svsk->sk_tcplen; - iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; + iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen; iov.iov_len = want; len =svc_recvfrom (rqstp, &iov,1 , want,0 );if (len <0 )
@@ -938,7 +938,7 @@ static void svc_tcp_fragment_received(struct svc_sock *svsk)svc_sock_final_rec (svsk) ?"final" :"nonfinal" , svc_sock_reclen(svsk)); svsk->sk_tcplen =0 ; - svsk->sk_reclen =0 ; + svsk->sk_marker = xdr_zero; }/*
@@ -1154,7 +1154,7 @@ static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) sk->sk_data_ready = svc_data_ready; sk->sk_write_space = svc_write_space; - svsk->sk_reclen =0 ; + svsk->sk_marker = xdr_zero; svsk->sk_tcplen =0 ; svsk->sk_datalen =0 ;memset (&svsk->sk_pages[0 ],0 ,sizeof (svsk->sk_pages));
PATCH 6/8 SUNRPC: Restructure svc_tcp_recv_record()
In reply to Chuck LeverRefactor: svc_recvfrom() is going to be converted to read into bio_vecs in a moment. Unhook the only other caller, svc_tcp_recv_record(), which just wants to read the 4-byte stream record marker into a kvec.
While we're in the area, streamline this helper by straight-lining the hot path, replace dprintk call sites with tracepoints, and reduce the number of atomic bit operations in this path.
--- include/trace/events/sunrpc.h | 24 +++++++++++++++++++ net/sunrpc/svcsock.c | 51 ++++++++++++++++++++--------------------- 2 files changed, 49 insertions(+), 26 deletions(-)
include/trace/events/sunrpc.h +24 -0
@@ -1459,6 +1459,30 @@ TRACE_EVENT(svcsock_new_socket, ) ); +TRACE_EVENT (svcsock_marker, +TP_PROTO ( + const struct svc_xprt *xprt, + __be32 marker + ), + +TP_ARGS (xprt, marker), + +TP_STRUCT__entry ( + __field(unsigned int, length) + __field(bool, last) +__string (addr, xprt->xpt_remotebuf) + ), + +TP_fast_assign ( + __entry->length = be32_to_cpu(marker) & RPC_FRAGMENT_SIZE_MASK; + __entry->last = be32_to_cpu(marker) & RPC_LAST_STREAM_FRAGMENT; + __assign_str(addr, xprt->xpt_remotebuf); + ), + +TP_printk ("addr=%s length=%u%s" ,__get_str (addr), + __entry->length, __entry->last ?" (last)" :"" ) +); + DECLARE_EVENT_CLASS(svcsock_class, TP_PROTO( const struct svc_xprt *xprt,
net/sunrpc/svcsock.c +25 -26
84 lines not on this page · raw message
PATCH 7/8 SUNRPC: Refactor svc_recvfrom()
In reply to Chuck LeverThis function is not currently "generic" so remove the documenting comment and rename it appropriately. Its internals are converted to use bio_vecs for reading from the transport socket.
In existing typical sunrpc uses of bio_vecs, the bio_vec array is allocated dynamically. Here, instead, an array of bio_vecs is added to svc_rqst. The lifetime of this array can be greater than one call to xpo_recvfrom():
- Multiple calls to xpo_recvfrom() might be needed to read an RPC message completely.
- At some later point, rq_arg.bvecs will point to this array and it will carry the received message into svc_process().
I also expect that a future optimization will remove either the rq_vec or rq_pages array in favor of rq_bvec, thus conserving the size of struct svc_rqst.
--- include/linux/sunrpc/svc.h | 1 include/trace/events/sunrpc.h | 1 net/sunrpc/svcsock.c | 190 +++++++++++++++++++++-------------------- 3 files changed, 99 insertions(+), 93 deletions(-)
include/linux/sunrpc/svc.h +1 -0
@@ -254,6 +254,7 @@ struct svc_rqst {struct page * *rq_page_end;/* one past the last page */ struct kvec rq_vec[RPCSVC_MAXPAGES ];/* generally useful.. */ +struct bio_vec rq_bvec[RPCSVC_MAXPAGES ];__be32 rq_xid;/* transmission id */ u32 rq_prog;/* program number */
include/trace/events/sunrpc.h +1 -0
@@ -1518,6 +1518,7 @@ DECLARE_EVENT_CLASS(svcsock_class,DEFINE_SVCSOCK_EVENT (udp_send);DEFINE_SVCSOCK_EVENT (tcp_send); +DEFINE_SVCSOCK_EVENT (tcp_recv);DEFINE_SVCSOCK_EVENT (data_ready);DEFINE_SVCSOCK_EVENT (write_space);
net/sunrpc/svcsock.c +97 -93
303 lines not on this page · raw message
PATCH 8/8 SUNRPC: Restructure svc_udp_recvfrom()
In reply to Chuck LeverAt this point, we are not ready yet to support bio_vecs in the UDP transport. However, we can clean up svc_udp_recvfrom to match the tracing and straight-lining changes made in svc_tcp_recvfrom.
--- include/trace/events/sunrpc.h | 1 + net/sunrpc/svcsock.c | 68 +++++++++++++++++++++++------------------ 2 files changed, 39 insertions(+), 30 deletions(-)
include/trace/events/sunrpc.h +1 -0
@@ -1517,6 +1517,7 @@ DECLARE_EVENT_CLASS(svcsock_class, TP_ARGS(xprt, result))DEFINE_SVCSOCK_EVENT (udp_send); +DEFINE_SVCSOCK_EVENT (udp_recv);DEFINE_SVCSOCK_EVENT (tcp_send);DEFINE_SVCSOCK_EVENT (tcp_recv);DEFINE_SVCSOCK_EVENT (data_ready);
net/sunrpc/svcsock.c +38 -30
126 lines not on this page · raw message