- patches.arch/x86_mce_intel_decode_physical_address.patch:
[linux-flexiantxendom0-3.2.10.git] / drivers / block / drbd / drbd_req.c
index de81ab7..654f1ef 100644 (file)
@@ -102,32 +102,7 @@ static void _req_is_done(struct drbd_conf *mdev, struct drbd_request *req, const
                }
        }
 
-       /* if it was a local io error, we want to notify our
-        * peer about that, and see if we need to
-        * detach the disk and stuff.
-        * to avoid allocating some special work
-        * struct, reuse the request. */
-
-       /* THINK
-        * why do we do this not when we detect the error,
-        * but delay it until it is "done", i.e. possibly
-        * until the next barrier ack? */
-
-       if (rw == WRITE &&
-           ((s & RQ_LOCAL_MASK) && !(s & RQ_LOCAL_OK))) {
-               if (!(req->w.list.next == LIST_POISON1 ||
-                     list_empty(&req->w.list))) {
-                       /* DEBUG ASSERT only; if this triggers, we
-                        * probably corrupt the worker list here */
-                       dev_err(DEV, "req->w.list.next = %p\n", req->w.list.next);
-                       dev_err(DEV, "req->w.list.prev = %p\n", req->w.list.prev);
-               }
-               req->w.cb = w_io_error;
-               drbd_queue_work(&mdev->data.work, &req->w);
-               /* drbd_req_free() is done in w_io_error */
-       } else {
-               drbd_req_free(req);
-       }
+       drbd_req_free(req);
 }
 
 static void queue_barrier(struct drbd_conf *mdev)
@@ -453,9 +428,6 @@ void __req_mod(struct drbd_request *req, enum drbd_req_event what,
                req->rq_state |= RQ_LOCAL_COMPLETED;
                req->rq_state &= ~RQ_LOCAL_PENDING;
 
-               dev_alert(DEV, "Local WRITE failed sec=%llus size=%u\n",
-                     (unsigned long long)req->sector, req->size);
-               /* and now: check how to handle local io error. */
                __drbd_chk_io_error(mdev, FALSE);
                _req_may_be_done(req, m);
                put_ldev(mdev);
@@ -475,22 +447,21 @@ void __req_mod(struct drbd_request *req, enum drbd_req_event what,
                req->rq_state |= RQ_LOCAL_COMPLETED;
                req->rq_state &= ~RQ_LOCAL_PENDING;
 
-               dev_alert(DEV, "Local READ failed sec=%llus size=%u\n",
-                     (unsigned long long)req->sector, req->size);
-               /* _req_mod(req,to_be_send); oops, recursion... */
                D_ASSERT(!(req->rq_state & RQ_NET_MASK));
-               req->rq_state |= RQ_NET_PENDING;
-               inc_ap_pending(mdev);
 
                __drbd_chk_io_error(mdev, FALSE);
                put_ldev(mdev);
-               /* NOTE: if we have no connection,
-                * or know the peer has no good data either,
-                * then we don't actually need to "queue_for_net_read",
-                * but we do so anyways, since the drbd_io_error()
-                * and the potential state change to "Diskless"
-                * needs to be done from process context */
 
+               /* no point in retrying if there is no good remote data,
+                * or we have no connection. */
+               if (mdev->state.pdsk != D_UP_TO_DATE) {
+                       _req_may_be_done(req, m);
+                       break;
+               }
+
+               /* _req_mod(req,to_be_send); oops, recursion... */
+               req->rq_state |= RQ_NET_PENDING;
+               inc_ap_pending(mdev);
                /* fall through: _req_mod(req,queue_for_net_read); */
 
        case queue_for_net_read:
@@ -600,6 +571,9 @@ void __req_mod(struct drbd_request *req, enum drbd_req_event what,
                _req_may_be_done(req, m);
                break;
 
+       case read_retry_remote_canceled:
+               req->rq_state &= ~RQ_NET_QUEUED;
+               /* fall through, in case we raced with drbd_disconnect */
        case connection_lost_while_pending:
                /* transfer log cleanup after connection loss */
                /* assert something? */
@@ -722,6 +696,7 @@ static int drbd_make_request_common(struct drbd_conf *mdev, struct bio *bio)
        struct drbd_request *req;
        int local, remote;
        int err = -EIO;
+       int ret = 0;
 
        /* allocate outside of all locks; */
        req = drbd_req_new(mdev, bio);
@@ -784,7 +759,7 @@ static int drbd_make_request_common(struct drbd_conf *mdev, struct bio *bio)
                            (mdev->state.pdsk == D_INCONSISTENT &&
                             mdev->state.conn >= C_CONNECTED));
 
-       if (!(local || remote)) {
+       if (!(local || remote) && !mdev->state.susp) {
                dev_err(DEV, "IO ERROR: neither local nor remote disk\n");
                goto fail_free_complete;
        }
@@ -810,6 +785,16 @@ allocate_barrier:
        /* GOOD, everything prepared, grab the spin_lock */
        spin_lock_irq(&mdev->req_lock);
 
+       if (mdev->state.susp) {
+               /* If we got suspended, use the retry mechanism of
+                  generic_make_request() to restart processing of this
+                  bio. In the next call to drbd_make_request_26
+                  we sleep in inc_ap_bio() */
+               ret = 1;
+               spin_unlock_irq(&mdev->req_lock);
+               goto fail_free_complete;
+       }
+
        if (remote) {
                remote = (mdev->state.pdsk == D_UP_TO_DATE ||
                            (mdev->state.pdsk == D_INCONSISTENT &&
@@ -947,12 +932,14 @@ fail_and_free_req:
                req->private_bio = NULL;
                put_ldev(mdev);
        }
-       bio_endio(bio, err);
+       if (!ret)
+               bio_endio(bio, err);
+
        drbd_req_free(req);
        dec_ap_bio(mdev);
        kfree(b);
 
-       return 0;
+       return ret;
 }
 
 /* helper function for drbd_make_request
@@ -962,11 +949,6 @@ fail_and_free_req:
  */
 static int drbd_fail_request_early(struct drbd_conf *mdev, int is_write)
 {
-       /* Unconfigured */
-       if (mdev->state.conn == C_DISCONNECTING &&
-           mdev->state.disk == D_DISKLESS)
-               return 1;
-
        if (mdev->state.role != R_PRIMARY &&
                (!allow_oos || is_write)) {
                if (__ratelimit(&drbd_ratelimit_state)) {
@@ -1070,15 +1052,21 @@ int drbd_make_request_26(struct request_queue *q, struct bio *bio)
 
                /* we need to get a "reference count" (ap_bio_cnt)
                 * to avoid races with the disconnect/reconnect/suspend code.
-                * In case we need to split the bio here, we need to get two references
+                * In case we need to split the bio here, we need to get three references
                 * atomically, otherwise we might deadlock when trying to submit the
                 * second one! */
-               inc_ap_bio(mdev, 2);
+               inc_ap_bio(mdev, 3);
 
                D_ASSERT(e_enr == s_enr + 1);
 
-               drbd_make_request_common(mdev, &bp->bio1);
-               drbd_make_request_common(mdev, &bp->bio2);
+               while (drbd_make_request_common(mdev, &bp->bio1))
+                       inc_ap_bio(mdev, 1);
+
+               while (drbd_make_request_common(mdev, &bp->bio2))
+                       inc_ap_bio(mdev, 1);
+
+               dec_ap_bio(mdev);
+
                bio_pair_release(bp);
        }
        return 0;
@@ -1115,7 +1103,7 @@ int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct
        } else if (limit && get_ldev(mdev)) {
                struct request_queue * const b =
                        mdev->ldev->backing_bdev->bd_disk->queue;
-               if (b->merge_bvec_fn && mdev->ldev->dc.use_bmbv) {
+               if (b->merge_bvec_fn) {
                        backing_limit = b->merge_bvec_fn(b, bvm, bvec);
                        limit = min(limit, backing_limit);
                }