summaryrefslogtreecommitdiff
path: root/drivers/block
diff options
context:
space:
mode:
authorLars Ellenberg <lars.ellenberg@linbit.com>2012-04-25 11:46:14 +0200
committerPhilipp Reisner <philipp.reisner@linbit.com>2012-05-09 15:16:58 +0200
commitba280c092e6eca8a70c502e4510061535fdce382 (patch)
tree70bf9476e9667ec5afec09f5a2c8f4b9cc91e75c /drivers/block
parent5de738272e38f7051c7a44c42631b71a0e2a1e80 (diff)
downloadlinux-stable-ba280c092e6eca8a70c502e4510061535fdce382.tar.gz
linux-stable-ba280c092e6eca8a70c502e4510061535fdce382.tar.bz2
linux-stable-ba280c092e6eca8a70c502e4510061535fdce382.zip
drbd: fix resend/resubmit of frozen IO
DRBD can freeze IO, due to fencing policy (fencing resource-and-stonith), or because we lost access to data (on-no-data-accessible suspend-io). Resuming from there (re-connect, or re-attach, or explicit admin intervention) should "just work". Unfortunately, if the re-attach/re-connect did not happen within the timeout, since the commit drbd: Implemented real timeout checking for request processing time if so configured, the request_timer_fn() would timeout and detach/disconnect virtually immediately. This change tracks the most recent attach and connect, and does not timeout within <configured timeout interval> after attach/connect. Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Diffstat (limited to 'drivers/block')
-rw-r--r--drivers/block/drbd/drbd_int.h2
-rw-r--r--drivers/block/drbd/drbd_main.c9
-rw-r--r--drivers/block/drbd/drbd_req.c52
3 files changed, 48 insertions, 15 deletions
diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index f215ad430bb8..302a6e786f76 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -1049,6 +1049,8 @@ struct drbd_conf {
struct crypto_hash *csums_tfm;
struct crypto_hash *verify_tfm;
+ unsigned long last_reattach_jif;
+ unsigned long last_reconnect_jif;
struct drbd_thread receiver;
struct drbd_thread worker;
struct drbd_thread asender;
diff --git a/drivers/block/drbd/drbd_main.c b/drivers/block/drbd/drbd_main.c
index deccff3af774..ab501b23b50e 100644
--- a/drivers/block/drbd/drbd_main.c
+++ b/drivers/block/drbd/drbd_main.c
@@ -1326,6 +1326,15 @@ __drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
if (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED)
drbd_resume_al(mdev);
+ /* remember last connect and attach times so request_timer_fn() won't
+ * kill newly established sessions while we are still trying to thaw
+ * previously frozen IO */
+ if (os.conn != C_WF_REPORT_PARAMS && ns.conn == C_WF_REPORT_PARAMS)
+ mdev->last_reconnect_jif = jiffies;
+ if ((os.disk == D_ATTACHING || os.disk == D_NEGOTIATING) &&
+ ns.disk > D_NEGOTIATING)
+ mdev->last_reattach_jif = jiffies;
+
ascw = kmalloc(sizeof(*ascw), GFP_ATOMIC);
if (ascw) {
ascw->os = os;
diff --git a/drivers/block/drbd/drbd_req.c b/drivers/block/drbd/drbd_req.c
index 340d57b98565..4a642ce62bae 100644
--- a/drivers/block/drbd/drbd_req.c
+++ b/drivers/block/drbd/drbd_req.c
@@ -1200,43 +1200,65 @@ void request_timer_fn(unsigned long data)
struct drbd_request *req; /* oldest request */
struct list_head *le;
unsigned long ent = 0, dt = 0, et, nt; /* effective timeout = ko_count * timeout */
+ unsigned long now;
if (get_net_conf(mdev)) {
- ent = mdev->net_conf->timeout*HZ/10 * mdev->net_conf->ko_count;
+ if (mdev->state.conn >= C_WF_REPORT_PARAMS)
+ ent = mdev->net_conf->timeout*HZ/10
+ * mdev->net_conf->ko_count;
put_net_conf(mdev);
}
- if (get_ldev(mdev)) {
+ if (get_ldev(mdev)) { /* implicit state.disk >= D_INCONSISTENT */
dt = mdev->ldev->dc.disk_timeout * HZ / 10;
put_ldev(mdev);
}
et = min_not_zero(dt, ent);
- if (!et || (mdev->state.conn < C_WF_REPORT_PARAMS && mdev->state.disk <= D_FAILED))
+ if (!et)
return; /* Recurring timer stopped */
+ now = jiffies;
+
spin_lock_irq(&mdev->req_lock);
le = &mdev->oldest_tle->requests;
if (list_empty(le)) {
spin_unlock_irq(&mdev->req_lock);
- mod_timer(&mdev->request_timer, jiffies + et);
+ mod_timer(&mdev->request_timer, now + et);
return;
}
le = le->prev;
req = list_entry(le, struct drbd_request, tl_requests);
- if (ent && req->rq_state & RQ_NET_PENDING) {
- if (time_is_before_eq_jiffies(req->start_time + ent)) {
- dev_warn(DEV, "Remote failed to finish a request within ko-count * timeout\n");
- _drbd_set_state(_NS(mdev, conn, C_TIMEOUT), CS_VERBOSE | CS_HARD, NULL);
- }
+
+ /* The request is considered timed out, if
+ * - we have some effective timeout from the configuration,
+ * with above state restrictions applied,
+ * - the oldest request is waiting for a response from the network
+ * resp. the local disk,
+ * - the oldest request is in fact older than the effective timeout,
+ * - the connection was established (resp. disk was attached)
+ * for longer than the timeout already.
+ * Note that for 32bit jiffies and very stable connections/disks,
+ * we may have a wrap around, which is catched by
+ * !time_in_range(now, last_..._jif, last_..._jif + timeout).
+ *
+ * Side effect: once per 32bit wrap-around interval, which means every
+ * ~198 days with 250 HZ, we have a window where the timeout would need
+ * to expire twice (worst case) to become effective. Good enough.
+ */
+ if (ent && req->rq_state & RQ_NET_PENDING &&
+ time_after(now, req->start_time + ent) &&
+ !time_in_range(now, mdev->last_reconnect_jif, mdev->last_reconnect_jif + ent)) {
+ dev_warn(DEV, "Remote failed to finish a request within ko-count * timeout\n");
+ _drbd_set_state(_NS(mdev, conn, C_TIMEOUT), CS_VERBOSE | CS_HARD, NULL);
}
- if (dt && req->rq_state & RQ_LOCAL_PENDING) {
- if (time_is_before_eq_jiffies(req->start_time + dt)) {
- dev_warn(DEV, "Local backing device failed to meet the disk-timeout\n");
- __drbd_chk_io_error(mdev, 1);
- }
+ if (dt && req->rq_state & RQ_LOCAL_PENDING &&
+ time_after(now, req->start_time + dt) &&
+ !time_in_range(now, mdev->last_reattach_jif, mdev->last_reattach_jif + dt)) {
+ dev_warn(DEV, "Local backing device failed to meet the disk-timeout\n");
+ __drbd_chk_io_error(mdev, 1);
}
- nt = (time_is_before_eq_jiffies(req->start_time + et) ? jiffies : req->start_time) + et;
+ nt = (time_after(now, req->start_time + et) ? now : req->start_time) + et;
spin_unlock_irq(&mdev->req_lock);
mod_timer(&mdev->request_timer, nt);
}