nfsd41: sunrpc: Added rpc server-side backchannel handling
authorRahul Iyer <iyer@netapp.com>
Thu, 10 Sep 2009 14:32:28 +0000 (17:32 +0300)
committerJ. Bruce Fields <bfields@citi.umich.edu>
Fri, 11 Sep 2009 19:04:16 +0000 (15:04 -0400)
When the call direction is a reply, copy the xid and call direction into the
req->rq_private_buf.head[0].iov_base otherwise rpc_verify_header returns
rpc_garbage.

Signed-off-by: Rahul Iyer <iyer@netapp.com>
Signed-off-by: Mike Sager <sager@netapp.com>
Signed-off-by: Marc Eshel <eshel@almaden.ibm.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[get rid of CONFIG_NFSD_V4_1]
[sunrpc: refactoring of svc_tcp_recvfrom]
[nfsd41: sunrpc: create common send routine for the fore and the back channels]
[nfsd41: sunrpc: Use free_page() to free server backchannel pages]
[nfsd41: sunrpc: Document server backchannel locking]
[nfsd41: sunrpc: remove bc_connect_worker()]
[nfsd41: sunrpc: Define xprt_server_backchannel()[
[nfsd41: sunrpc: remove bc_close and bc_init_auto_disconnect dummy functions]
[nfsd41: sunrpc: eliminate unneeded switch statement in xs_setup_tcp()]
[nfsd41: sunrpc: Don't auto close the server backchannel connection]
[nfsd41: sunrpc: Remove unused functions]
Signed-off-by: Alexandros Batsakis <batsakis@netapp.com>
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[nfsd41: change bc_sock to bc_xprt]
[nfsd41: sunrpc: move struct rpc_buffer def into a common header file]
[nfsd41: sunrpc: use rpc_sleep in bc_send_request so not to block on mutex]
[removed cosmetic changes]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[sunrpc: add new xprt class for nfsv4.1 backchannel]
[sunrpc: v2.1 change handling of auto_close and init_auto_disconnect operations for the nfsv4.1 backchannel]
Signed-off-by: Alexandros Batsakis <batsakis@netapp.com>
[reverted more cosmetic leftovers]
[got rid of xprt_server_backchannel]
[separated "nfsd41: sunrpc: add new xprt class for nfsv4.1 backchannel"]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Cc: Trond Myklebust <trond.myklebust@netapp.com>
[sunrpc: change idle timeout value for the backchannel]
Signed-off-by: Alexandros Batsakis <batsakis@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Acked-by: Trond Myklebust <trond.myklebust@netapp.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
include/linux/sunrpc/svc_xprt.h
include/linux/sunrpc/svcsock.h
include/linux/sunrpc/xprt.h
net/sunrpc/sunrpc.h
net/sunrpc/svc_xprt.c
net/sunrpc/svcsock.c
net/sunrpc/xprt.c
net/sunrpc/xprtsock.c

index 2223ae0..5f4e18b 100644 (file)
@@ -65,6 +65,7 @@ struct svc_xprt {
        size_t                  xpt_locallen;   /* length of address */
        struct sockaddr_storage xpt_remote;     /* remote peer's address */
        size_t                  xpt_remotelen;  /* length of address */
+       struct rpc_wait_queue   xpt_bc_pending; /* backchannel wait queue */
 };
 
 int    svc_reg_xprt_class(struct svc_xprt_class *);
index 04dba23..1b353a7 100644 (file)
@@ -28,6 +28,7 @@ struct svc_sock {
        /* private TCP part */
        u32                     sk_reclen;      /* length of record */
        u32                     sk_tcplen;      /* current read length */
+       struct rpc_xprt         *sk_bc_xprt;    /* NFSv4.1 backchannel xprt */
 };
 
 /*
index c090df4..228d694 100644 (file)
@@ -179,6 +179,7 @@ struct rpc_xprt {
        spinlock_t              reserve_lock;   /* lock slot table */
        u32                     xid;            /* Next XID value to use */
        struct rpc_task *       snd_task;       /* Task blocked in send */
+       struct svc_xprt         *bc_xprt;       /* NFSv4.1 backchannel */
 #if defined(CONFIG_NFS_V4_1)
        struct svc_serv         *bc_serv;       /* The RPC service which will */
                                                /* process the callback */
index 13171e6..90c292e 100644 (file)
@@ -43,5 +43,9 @@ static inline int rpc_reply_expected(struct rpc_task *task)
                (task->tk_msg.rpc_proc->p_decode != NULL);
 }
 
+int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
+                   struct page *headpage, unsigned long headoffset,
+                   struct page *tailpage, unsigned long tailoffset);
+
 #endif /* _NET_SUNRPC_SUNRPC_H */
 
index 912dea5..df124f7 100644 (file)
@@ -160,6 +160,7 @@ void svc_xprt_init(struct svc_xprt_class *xcl, struct svc_xprt *xprt,
        mutex_init(&xprt->xpt_mutex);
        spin_lock_init(&xprt->xpt_lock);
        set_bit(XPT_BUSY, &xprt->xpt_flags);
+       rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending");
 }
 EXPORT_SYMBOL_GPL(svc_xprt_init);
 
@@ -810,6 +811,7 @@ int svc_send(struct svc_rqst *rqstp)
        else
                len = xprt->xpt_ops->xpo_sendto(rqstp);
        mutex_unlock(&xprt->xpt_mutex);
+       rpc_wake_up(&xprt->xpt_bc_pending);
        svc_xprt_release(rqstp);
 
        if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
index 76a380d..ccc5e83 100644 (file)
@@ -49,6 +49,7 @@
 #include <linux/sunrpc/msg_prot.h>
 #include <linux/sunrpc/svcsock.h>
 #include <linux/sunrpc/stats.h>
+#include <linux/sunrpc/xprt.h>
 
 #define RPCDBG_FACILITY        RPCDBG_SVCXPRT
 
@@ -153,49 +154,27 @@ static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
 }
 
 /*
- * Generic sendto routine
+ * send routine intended to be shared by the fore- and back-channel
  */
-static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
+int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
+                   struct page *headpage, unsigned long headoffset,
+                   struct page *tailpage, unsigned long tailoffset)
 {
-       struct svc_sock *svsk =
-               container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
-       struct socket   *sock = svsk->sk_sock;
-       int             slen;
-       union {
-               struct cmsghdr  hdr;
-               long            all[SVC_PKTINFO_SPACE / sizeof(long)];
-       } buffer;
-       struct cmsghdr *cmh = &buffer.hdr;
-       int             len = 0;
        int             result;
        int             size;
        struct page     **ppage = xdr->pages;
        size_t          base = xdr->page_base;
        unsigned int    pglen = xdr->page_len;
        unsigned int    flags = MSG_MORE;
-       RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
+       int             slen;
+       int             len = 0;
 
        slen = xdr->len;
 
-       if (rqstp->rq_prot == IPPROTO_UDP) {
-               struct msghdr msg = {
-                       .msg_name       = &rqstp->rq_addr,
-                       .msg_namelen    = rqstp->rq_addrlen,
-                       .msg_control    = cmh,
-                       .msg_controllen = sizeof(buffer),
-                       .msg_flags      = MSG_MORE,
-               };
-
-               svc_set_cmsg_data(rqstp, cmh);
-
-               if (sock_sendmsg(sock, &msg, 0) < 0)
-                       goto out;
-       }
-
        /* send head */
        if (slen == xdr->head[0].iov_len)
                flags = 0;
-       len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
+       len = kernel_sendpage(sock, headpage, headoffset,
                                  xdr->head[0].iov_len, flags);
        if (len != xdr->head[0].iov_len)
                goto out;
@@ -219,16 +198,58 @@ static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
                base = 0;
                ppage++;
        }
+
        /* send tail */
        if (xdr->tail[0].iov_len) {
-               result = kernel_sendpage(sock, rqstp->rq_respages[0],
-                                            ((unsigned long)xdr->tail[0].iov_base)
-                                               & (PAGE_SIZE-1),
-                                            xdr->tail[0].iov_len, 0);
-
+               result = kernel_sendpage(sock, tailpage, tailoffset,
+                                  xdr->tail[0].iov_len, 0);
                if (result > 0)
                        len += result;
        }
+
+out:
+       return len;
+}
+
+
+/*
+ * Generic sendto routine
+ */
+static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
+{
+       struct svc_sock *svsk =
+               container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
+       struct socket   *sock = svsk->sk_sock;
+       union {
+               struct cmsghdr  hdr;
+               long            all[SVC_PKTINFO_SPACE / sizeof(long)];
+       } buffer;
+       struct cmsghdr *cmh = &buffer.hdr;
+       int             len = 0;
+       unsigned long tailoff;
+       unsigned long headoff;
+       RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
+
+       if (rqstp->rq_prot == IPPROTO_UDP) {
+               struct msghdr msg = {
+                       .msg_name       = &rqstp->rq_addr,
+                       .msg_namelen    = rqstp->rq_addrlen,
+                       .msg_control    = cmh,
+                       .msg_controllen = sizeof(buffer),
+                       .msg_flags      = MSG_MORE,
+               };
+
+               svc_set_cmsg_data(rqstp, cmh);
+
+               if (sock_sendmsg(sock, &msg, 0) < 0)
+                       goto out;
+       }
+
+       tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
+       headoff = 0;
+       len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
+                              rqstp->rq_respages[0], tailoff);
+
 out:
        dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
                svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
@@ -951,6 +972,57 @@ static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
        return -EAGAIN;
 }
 
+static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
+                              struct rpc_rqst **reqpp, struct kvec *vec)
+{
+       struct rpc_rqst *req = NULL;
+       u32 *p;
+       u32 xid;
+       u32 calldir;
+       int len;
+
+       len = svc_recvfrom(rqstp, vec, 1, 8);
+       if (len < 0)
+               goto error;
+
+       p = (u32 *)rqstp->rq_arg.head[0].iov_base;
+       xid = *p++;
+       calldir = *p;
+
+       if (calldir == 0) {
+               /* REQUEST is the most common case */
+               vec[0] = rqstp->rq_arg.head[0];
+       } else {
+               /* REPLY */
+               if (svsk->sk_bc_xprt)
+                       req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid);
+
+               if (!req) {
+                       printk(KERN_NOTICE
+                               "%s: Got unrecognized reply: "
+                               "calldir 0x%x sk_bc_xprt %p xid %08x\n",
+                               __func__, ntohl(calldir),
+                               svsk->sk_bc_xprt, xid);
+                       vec[0] = rqstp->rq_arg.head[0];
+                       goto out;
+               }
+
+               memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
+                      sizeof(struct xdr_buf));
+               /* copy the xid and call direction */
+               memcpy(req->rq_private_buf.head[0].iov_base,
+                      rqstp->rq_arg.head[0].iov_base, 8);
+               vec[0] = req->rq_private_buf.head[0];
+       }
+ out:
+       vec[0].iov_base += 8;
+       vec[0].iov_len -= 8;
+       len = svsk->sk_reclen - 8;
+ error:
+       *reqpp = req;
+       return len;
+}
+
 /*
  * Receive data from a TCP socket.
  */
@@ -962,6 +1034,7 @@ static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
        int             len;
        struct kvec *vec;
        int pnum, vlen;
+       struct rpc_rqst *req = NULL;
 
        dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
                svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
@@ -975,9 +1048,27 @@ static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
        vec = rqstp->rq_vec;
        vec[0] = rqstp->rq_arg.head[0];
        vlen = PAGE_SIZE;
+
+       /*
+        * We have enough data for the whole tcp record. Let's try and read the
+        * first 8 bytes to get the xid and the call direction. We can use this
+        * to figure out if this is a call or a reply to a callback. If
+        * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
+        * In that case, don't bother with the calldir and just read the data.
+        * It will be rejected in svc_process.
+        */
+       if (len >= 8) {
+               len = svc_process_calldir(svsk, rqstp, &req, vec);
+               if (len < 0)
+                       goto err_again;
+               vlen -= 8;
+       }
+
        pnum = 1;
        while (vlen < len) {
-               vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
+               vec[pnum].iov_base = (req) ?
+                       page_address(req->rq_private_buf.pages[pnum - 1]) :
+                       page_address(rqstp->rq_pages[pnum]);
                vec[pnum].iov_len = PAGE_SIZE;
                pnum++;
                vlen += PAGE_SIZE;
@@ -989,6 +1080,16 @@ static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
        if (len < 0)
                goto err_again;
 
+       /*
+        * Account for the 8 bytes we read earlier
+        */
+       len += 8;
+
+       if (req) {
+               xprt_complete_rqst(req->rq_task, len);
+               len = 0;
+               goto out;
+       }
        dprintk("svc: TCP complete record (%d bytes)\n", len);
        rqstp->rq_arg.len = len;
        rqstp->rq_arg.page_base = 0;
@@ -1002,6 +1103,7 @@ static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
        rqstp->rq_xprt_ctxt   = NULL;
        rqstp->rq_prot        = IPPROTO_TCP;
 
+out:
        /* Reset TCP read info */
        svsk->sk_reclen = 0;
        svsk->sk_tcplen = 0;
index f412a85..fd46d42 100644 (file)
@@ -832,6 +832,11 @@ static void xprt_timer(struct rpc_task *task)
        spin_unlock_bh(&xprt->transport_lock);
 }
 
+static inline int xprt_has_timer(struct rpc_xprt *xprt)
+{
+       return xprt->idle_timeout != 0;
+}
+
 /**
  * xprt_prepare_transmit - reserve the transport before sending a request
  * @task: RPC task about to send a request
@@ -1013,7 +1018,7 @@ void xprt_release(struct rpc_task *task)
        if (!list_empty(&req->rq_list))
                list_del(&req->rq_list);
        xprt->last_used = jiffies;
-       if (list_empty(&xprt->recv))
+       if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
                mod_timer(&xprt->timer,
                                xprt->last_used + xprt->idle_timeout);
        spin_unlock_bh(&xprt->transport_lock);
@@ -1082,8 +1087,11 @@ found:
 #endif /* CONFIG_NFS_V4_1 */
 
        INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
-       setup_timer(&xprt->timer, xprt_init_autodisconnect,
-                       (unsigned long)xprt);
+       if (xprt_has_timer(xprt))
+               setup_timer(&xprt->timer, xprt_init_autodisconnect,
+                           (unsigned long)xprt);
+       else
+               init_timer(&xprt->timer);
        xprt->last_used = jiffies;
        xprt->cwnd = RPC_INITCWND;
        xprt->bind_index = 0;
@@ -1102,7 +1110,6 @@ found:
 
        dprintk("RPC:       created transport %p with %u slots\n", xprt,
                        xprt->max_reqs);
-
        return xprt;
 }
 
index 62438f3..d9a2b81 100644 (file)
@@ -32,6 +32,7 @@
 #include <linux/tcp.h>
 #include <linux/sunrpc/clnt.h>
 #include <linux/sunrpc/sched.h>
+#include <linux/sunrpc/svcsock.h>
 #include <linux/sunrpc/xprtsock.h>
 #include <linux/file.h>
 #ifdef CONFIG_NFS_V4_1
@@ -43,6 +44,7 @@
 #include <net/udp.h>
 #include <net/tcp.h>
 
+#include "sunrpc.h"
 /*
  * xprtsock tunables
  */
@@ -2098,6 +2100,134 @@ static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
                        xprt->stat.bklog_u);
 }
 
+/*
+ * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
+ * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
+ * to use the server side send routines.
+ */
+void *bc_malloc(struct rpc_task *task, size_t size)
+{
+       struct page *page;
+       struct rpc_buffer *buf;
+
+       BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
+       page = alloc_page(GFP_KERNEL);
+
+       if (!page)
+               return NULL;
+
+       buf = page_address(page);
+       buf->len = PAGE_SIZE;
+
+       return buf->data;
+}
+
+/*
+ * Free the space allocated in the bc_alloc routine
+ */
+void bc_free(void *buffer)
+{
+       struct rpc_buffer *buf;
+
+       if (!buffer)
+               return;
+
+       buf = container_of(buffer, struct rpc_buffer, data);
+       free_page((unsigned long)buf);
+}
+
+/*
+ * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
+ * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
+ */
+static int bc_sendto(struct rpc_rqst *req)
+{
+       int len;
+       struct xdr_buf *xbufp = &req->rq_snd_buf;
+       struct rpc_xprt *xprt = req->rq_xprt;
+       struct sock_xprt *transport =
+                               container_of(xprt, struct sock_xprt, xprt);
+       struct socket *sock = transport->sock;
+       unsigned long headoff;
+       unsigned long tailoff;
+
+       /*
+        * Set up the rpc header and record marker stuff
+        */
+       xs_encode_tcp_record_marker(xbufp);
+
+       tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
+       headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
+       len = svc_send_common(sock, xbufp,
+                             virt_to_page(xbufp->head[0].iov_base), headoff,
+                             xbufp->tail[0].iov_base, tailoff);
+
+       if (len != xbufp->len) {
+               printk(KERN_NOTICE "Error sending entire callback!\n");
+               len = -EAGAIN;
+       }
+
+       return len;
+}
+
+/*
+ * The send routine. Borrows from svc_send
+ */
+static int bc_send_request(struct rpc_task *task)
+{
+       struct rpc_rqst *req = task->tk_rqstp;
+       struct svc_xprt *xprt;
+       struct svc_sock         *svsk;
+       u32                     len;
+
+       dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
+       /*
+        * Get the server socket associated with this callback xprt
+        */
+       xprt = req->rq_xprt->bc_xprt;
+       svsk = container_of(xprt, struct svc_sock, sk_xprt);
+
+       /*
+        * Grab the mutex to serialize data as the connection is shared
+        * with the fore channel
+        */
+       if (!mutex_trylock(&xprt->xpt_mutex)) {
+               rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
+               if (!mutex_trylock(&xprt->xpt_mutex))
+                       return -EAGAIN;
+               rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
+       }
+       if (test_bit(XPT_DEAD, &xprt->xpt_flags))
+               len = -ENOTCONN;
+       else
+               len = bc_sendto(req);
+       mutex_unlock(&xprt->xpt_mutex);
+
+       if (len > 0)
+               len = 0;
+
+       return len;
+}
+
+/*
+ * The close routine. Since this is client initiated, we do nothing
+ */
+
+static void bc_close(struct rpc_xprt *xprt)
+{
+       return;
+}
+
+/*
+ * The xprt destroy routine. Again, because this connection is client
+ * initiated, we do nothing
+ */
+
+static void bc_destroy(struct rpc_xprt *xprt)
+{
+       return;
+}
+
 static struct rpc_xprt_ops xs_udp_ops = {
        .set_buffer_size        = xs_udp_set_buffer_size,
        .reserve_xprt           = xprt_reserve_xprt_cong,
@@ -2134,6 +2264,22 @@ static struct rpc_xprt_ops xs_tcp_ops = {
        .print_stats            = xs_tcp_print_stats,
 };
 
+/*
+ * The rpc_xprt_ops for the server backchannel
+ */
+
+static struct rpc_xprt_ops bc_tcp_ops = {
+       .reserve_xprt           = xprt_reserve_xprt,
+       .release_xprt           = xprt_release_xprt,
+       .buf_alloc              = bc_malloc,
+       .buf_free               = bc_free,
+       .send_request           = bc_send_request,
+       .set_retrans_timeout    = xprt_set_retrans_timeout_def,
+       .close                  = bc_close,
+       .destroy                = bc_destroy,
+       .print_stats            = xs_tcp_print_stats,
+};
+
 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
                                      unsigned int slot_table_size)
 {