net/9p: handle correctly interrupted 9P requests
authorLatchesar Ionkov <lucho@ionkov.net>
Sun, 5 Apr 2009 21:28:59 +0000 (16:28 -0500)
committerEric Van Hensbergen <ericvh@vTrogdor.(none)>
Sun, 5 Apr 2009 21:54:53 +0000 (16:54 -0500)
Currently the 9p code crashes when a operation is interrupted, i.e. for
example when the user presses ^C while reading from a file.

This patch fixes the code that is responsible for interruption and flushing
of 9P operations.

Signed-off-by: Latchesar Ionkov <lucho@ionkov.net>
include/net/9p/client.h
net/9p/client.c
net/9p/trans_fd.c
net/9p/trans_rdma.c
net/9p/trans_virtio.c

index 4012e07..e268122 100644 (file)
@@ -98,7 +98,6 @@ enum p9_req_status_t {
 struct p9_req_t {
        int status;
        int t_err;
-       u16 flush_tag;
        wait_queue_head_t *wq;
        struct p9_fcall *tc;
        struct p9_fcall *rc;
index 781d89a..dd43a82 100644 (file)
@@ -203,7 +203,6 @@ static struct p9_req_t *p9_tag_alloc(struct p9_client *c, u16 tag)
        p9pdu_reset(req->tc);
        p9pdu_reset(req->rc);
 
-       req->flush_tag = 0;
        req->tc->tag = tag-1;
        req->status = REQ_STATUS_ALLOC;
 
@@ -324,35 +323,9 @@ static void p9_free_req(struct p9_client *c, struct p9_req_t *r)
  */
 void p9_client_cb(struct p9_client *c, struct p9_req_t *req)
 {
-       struct p9_req_t *other_req;
-       unsigned long flags;
-
        P9_DPRINTK(P9_DEBUG_MUX, " tag %d\n", req->tc->tag);
-
-       if (req->status == REQ_STATUS_ERROR)
-               wake_up(req->wq);
-
-       if (req->flush_tag) {                   /* flush receive path */
-               P9_DPRINTK(P9_DEBUG_9P, "<<< RFLUSH %d\n", req->tc->tag);
-               spin_lock_irqsave(&c->lock, flags);
-               other_req = p9_tag_lookup(c, req->flush_tag);
-               if (other_req->status != REQ_STATUS_FLSH) /* stale flush */
-                       spin_unlock_irqrestore(&c->lock, flags);
-               else {
-                       other_req->status = REQ_STATUS_FLSHD;
-                       spin_unlock_irqrestore(&c->lock, flags);
-                       wake_up(other_req->wq);
-               }
-               p9_free_req(c, req);
-       } else {                                /* normal receive path */
-               P9_DPRINTK(P9_DEBUG_MUX, "normal: tag %d\n", req->tc->tag);
-               spin_lock_irqsave(&c->lock, flags);
-               if (req->status != REQ_STATUS_FLSHD)
-                       req->status = REQ_STATUS_RCVD;
-               spin_unlock_irqrestore(&c->lock, flags);
-               wake_up(req->wq);
-               P9_DPRINTK(P9_DEBUG_MUX, "wakeup: %d\n", req->tc->tag);
-       }
+       wake_up(req->wq);
+       P9_DPRINTK(P9_DEBUG_MUX, "wakeup: %d\n", req->tc->tag);
 }
 EXPORT_SYMBOL(p9_client_cb);
 
@@ -486,9 +459,15 @@ static int p9_client_flush(struct p9_client *c, struct p9_req_t *oldreq)
        if (IS_ERR(req))
                return PTR_ERR(req);
 
-       req->flush_tag = oldtag;
 
-       /* we don't free anything here because RPC isn't complete */
+       /* if we haven't received a response for oldreq,
+          remove it from the list. */
+       spin_lock(&c->lock);
+       if (oldreq->status == REQ_STATUS_FLSH)
+               list_del(&oldreq->req_list);
+       spin_unlock(&c->lock);
+
+       p9_free_req(c, req);
        return 0;
 }
 
@@ -509,7 +488,6 @@ p9_client_rpc(struct p9_client *c, int8_t type, const char *fmt, ...)
        struct p9_req_t *req;
        unsigned long flags;
        int sigpending;
-       int flushed = 0;
 
        P9_DPRINTK(P9_DEBUG_MUX, "client %p op %d\n", c, type);
 
@@ -546,42 +524,28 @@ p9_client_rpc(struct p9_client *c, int8_t type, const char *fmt, ...)
                goto reterr;
        }
 
-       /* if it was a flush we just transmitted, return our tag */
-       if (type == P9_TFLUSH)
-               return req;
-again:
        P9_DPRINTK(P9_DEBUG_MUX, "wait %p tag: %d\n", req->wq, tag);
        err = wait_event_interruptible(*req->wq,
                                                req->status >= REQ_STATUS_RCVD);
-       P9_DPRINTK(P9_DEBUG_MUX, "wait %p tag: %d returned %d (flushed=%d)\n",
-                                               req->wq, tag, err, flushed);
+       P9_DPRINTK(P9_DEBUG_MUX, "wait %p tag: %d returned %d\n",
+                                               req->wq, tag, err);
 
        if (req->status == REQ_STATUS_ERROR) {
                P9_DPRINTK(P9_DEBUG_ERROR, "req_status error %d\n", req->t_err);
                err = req->t_err;
-       } else if (err == -ERESTARTSYS && flushed) {
-               P9_DPRINTK(P9_DEBUG_MUX, "flushed - going again\n");
-               goto again;
-       } else if (req->status == REQ_STATUS_FLSHD) {
-               P9_DPRINTK(P9_DEBUG_MUX, "flushed - erestartsys\n");
-               err = -ERESTARTSYS;
        }
 
-       if ((err == -ERESTARTSYS) && (c->status == Connected) && (!flushed)) {
+       if ((err == -ERESTARTSYS) && (c->status == Connected)) {
                P9_DPRINTK(P9_DEBUG_MUX, "flushing\n");
-               spin_lock_irqsave(&c->lock, flags);
-               if (req->status == REQ_STATUS_SENT)
-                       req->status = REQ_STATUS_FLSH;
-               spin_unlock_irqrestore(&c->lock, flags);
                sigpending = 1;
-               flushed = 1;
                clear_thread_flag(TIF_SIGPENDING);
 
-               if (c->trans_mod->cancel(c, req)) {
-                       err = p9_client_flush(c, req);
-                       if (err == 0)
-                               goto again;
-               }
+               if (c->trans_mod->cancel(c, req))
+                       p9_client_flush(c, req);
+
+               /* if we received the response anyway, don't signal error */
+               if (req->status == REQ_STATUS_RCVD)
+                       err = 0;
        }
 
        if (sigpending) {
index c613ed0..a2a1814 100644 (file)
@@ -213,8 +213,8 @@ static void p9_conn_cancel(struct p9_conn *m, int err)
        spin_unlock_irqrestore(&m->client->lock, flags);
 
        list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
-               list_del(&req->req_list);
                P9_DPRINTK(P9_DEBUG_ERROR, "call back req %p\n", req);
+               list_del(&req->req_list);
                p9_client_cb(m->client, req);
        }
 }
@@ -336,7 +336,8 @@ static void p9_read_work(struct work_struct *work)
                        "mux %p pkt: size: %d bytes tag: %d\n", m, n, tag);
 
                m->req = p9_tag_lookup(m->client, tag);
-               if (!m->req) {
+               if (!m->req || (m->req->status != REQ_STATUS_SENT &&
+                                       m->req->status != REQ_STATUS_FLSH)) {
                        P9_DPRINTK(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
                                                                 tag);
                        err = -EIO;
@@ -361,10 +362,11 @@ static void p9_read_work(struct work_struct *work)
        if ((m->req) && (m->rpos == m->rsize)) { /* packet is read in */
                P9_DPRINTK(P9_DEBUG_TRANS, "got new packet\n");
                spin_lock(&m->client->lock);
+               if (m->req->status != REQ_STATUS_ERROR)
+                       m->req->status = REQ_STATUS_RCVD;
                list_del(&m->req->req_list);
                spin_unlock(&m->client->lock);
                p9_client_cb(m->client, m->req);
-
                m->rbuf = NULL;
                m->rpos = 0;
                m->rsize = 0;
@@ -454,6 +456,7 @@ static void p9_write_work(struct work_struct *work)
                req = list_entry(m->unsent_req_list.next, struct p9_req_t,
                               req_list);
                req->status = REQ_STATUS_SENT;
+               P9_DPRINTK(P9_DEBUG_TRANS, "move req %p\n", req);
                list_move_tail(&req->req_list, &m->req_list);
 
                m->wbuf = req->tc->sdata;
@@ -683,12 +686,13 @@ static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
        P9_DPRINTK(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
 
        spin_lock(&client->lock);
-       list_del(&req->req_list);
 
        if (req->status == REQ_STATUS_UNSENT) {
+               list_del(&req->req_list);
                req->status = REQ_STATUS_FLSHD;
                ret = 0;
-       }
+       } else if (req->status == REQ_STATUS_SENT)
+               req->status = REQ_STATUS_FLSH;
 
        spin_unlock(&client->lock);
 
index 7fa0eb2..ac49900 100644 (file)
@@ -295,6 +295,7 @@ handle_recv(struct p9_client *client, struct p9_trans_rdma *rdma,
                goto err_out;
 
        req->rc = c->rc;
+       req->status = REQ_STATUS_RCVD;
        p9_client_cb(client, req);
 
        return;
index 2d7781e..bb8579a 100644 (file)
@@ -134,6 +134,7 @@ static void req_done(struct virtqueue *vq)
                P9_DPRINTK(P9_DEBUG_TRANS, ": rc %p\n", rc);
                P9_DPRINTK(P9_DEBUG_TRANS, ": lookup tag %d\n", rc->tag);
                req = p9_tag_lookup(chan->client, rc->tag);
+               req->status = REQ_STATUS_RCVD;
                p9_client_cb(chan->client, req);
        }
 }