diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c index 742b2908ff68..b3dbf6c76e98 100644 --- a/drivers/block/drbd/drbd_actlog.c +++ b/drivers/block/drbd/drbd_actlog.c @@ -483,38 +483,20 @@ void drbd_al_begin_io(struct drbd_device *device, struct drbd_interval *i) int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval *i) { - struct lru_cache *al = device->act_log; /* for bios crossing activity log extent boundaries, * we may need to activate two extents in one go */ unsigned first = i->sector >> (AL_EXTENT_SHIFT-9); unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9); - unsigned nr_al_extents; - unsigned available_update_slots; unsigned enr; - D_ASSERT(device, first <= last); - - nr_al_extents = 1 + last - first; /* worst case: all touched extends are cold. */ - available_update_slots = min(al->nr_elements - al->used, - al->max_pending_changes - al->pending_changes); - - /* We want all necessary updates for a given request within the same transaction - * We could first check how many updates are *actually* needed, - * and use that instead of the worst-case nr_al_extents */ - if (available_update_slots < nr_al_extents) { - /* Too many activity log extents are currently "hot". - * - * If we have accumulated pending changes already, - * we made progress. - * - * If we cannot get even a single pending change through, - * stop the fast path until we made some progress, - * or requests to "cold" extents could be starved. */ - if (!al->pending_changes) - __set_bit(__LC_STARVING, &device->act_log->flags); - return -ENOBUFS; + if (i->partially_in_al_next_enr) { + D_ASSERT(device, first < i->partially_in_al_next_enr); + D_ASSERT(device, last >= i->partially_in_al_next_enr); + first = i->partially_in_al_next_enr; } + D_ASSERT(device, first <= last); + /* Is resync active in this area? */ for (enr = first; enr <= last; enr++) { struct lc_element *tmp; @@ -529,14 +511,21 @@ int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval * } } - /* Checkout the refcounts. - * Given that we checked for available elements and update slots above, - * this has to be successful. */ + /* Try to checkout the refcounts. */ for (enr = first; enr <= last; enr++) { struct lc_element *al_ext; al_ext = lc_get_cumulative(device->act_log, enr); - if (!al_ext) - drbd_info(device, "LOGIC BUG for enr=%u\n", enr); + + if (!al_ext) { + /* Did not work. We may have exhausted the possible + * changes per transaction. Or raced with someone + * "locking" it against changes. + * Remember where to continue from. + */ + if (enr > first) + i->partially_in_al_next_enr = enr; + return -ENOBUFS; + } } return 0; } @@ -556,7 +545,11 @@ void drbd_al_complete_io(struct drbd_device *device, struct drbd_interval *i) for (enr = first; enr <= last; enr++) { extent = lc_find(device->act_log, enr); - if (!extent) { + /* Yes, this masks a bug elsewhere. However, during normal + * operation this is harmless, so no need to crash the kernel + * by the BUG_ON(refcount == 0) in lc_put(). + */ + if (!extent || extent->refcnt == 0) { drbd_err(device, "al_complete_io() called on inactive extent %u\n", enr); continue; } diff --git a/drivers/block/drbd/drbd_interval.h b/drivers/block/drbd/drbd_interval.h index 366489b72fe9..5d3213b81eed 100644 --- a/drivers/block/drbd/drbd_interval.h +++ b/drivers/block/drbd/drbd_interval.h @@ -8,12 +8,15 @@ struct drbd_interval { struct rb_node rb; sector_t sector; /* start sector of the interval */ - unsigned int size; /* size in bytes */ sector_t end; /* highest interval end in subtree */ + unsigned int size; /* size in bytes */ unsigned int local:1 /* local or remote request? */; unsigned int waiting:1; /* someone is waiting for completion */ unsigned int completed:1; /* this has been completed already; * ignore for conflict detection */ + + /* to resume a partially successful drbd_al_begin_io_nonblock(); */ + unsigned int partially_in_al_next_enr; }; static inline void drbd_clear_interval(struct drbd_interval *i)