Invalidation is removing an entry from the cache without writing it
back. Cache blocks can be invalidated via the invalidate_cblocks
message, which takes an arbitrary number of cblock ranges. Each cblock
-must be expressed as a decimal value, in the future a variant message
-that takes cblock ranges expressed in hexidecimal may be needed to
-better support efficient invalidation of larger caches. The cache must
-be in passthrough mode when invalidate_cblocks is used.
+range's end value is "one past the end", meaning 5-10 expresses a range
+of values from 5 to 9. Each cblock must be expressed as a decimal
+value, in the future a variant message that takes cblock ranges
+expressed in hexidecimal may be needed to better support efficient
+invalidation of larger caches. The cache must be in passthrough mode
+when invalidate_cblocks is used.
invalidate_cblocks [<cblock>|<cblock begin>-<cblock end>]*
{
__u64 mem;
+ dm_bufio_allocated_kmem_cache = 0;
+ dm_bufio_allocated_get_free_pages = 0;
+ dm_bufio_allocated_vmalloc = 0;
+ dm_bufio_current_allocated = 0;
+
memset(&dm_bufio_caches, 0, sizeof dm_bufio_caches);
memset(&dm_bufio_cache_names, 0, sizeof dm_bufio_cache_names);
int r = 0;
bool updated = updated_this_tick(mq, e);
- requeue_and_update_tick(mq, e);
-
if ((!discarded_oblock && updated) ||
- !should_promote(mq, e, discarded_oblock, data_dir))
+ !should_promote(mq, e, discarded_oblock, data_dir)) {
+ requeue_and_update_tick(mq, e);
result->op = POLICY_MISS;
- else if (!can_migrate)
+
+ } else if (!can_migrate)
r = -EWOULDBLOCK;
- else
+
+ else {
+ requeue_and_update_tick(mq, e);
r = pre_cache_to_cache(mq, e, result);
+ }
return r;
}
{
int r;
- r = dm_cache_resize(cache->cmd, cache->cache_size);
+ r = dm_cache_resize(cache->cmd, new_size);
if (r) {
DMERR("could not resize cache metadata");
return r;
struct delay_c {
struct timer_list delay_timer;
struct mutex timer_lock;
+ struct workqueue_struct *kdelayd_wq;
struct work_struct flush_expired_bios;
struct list_head delayed_bios;
atomic_t may_delay;
static DEFINE_MUTEX(delayed_bios_lock);
-static struct workqueue_struct *kdelayd_wq;
static struct kmem_cache *delayed_cache;
static void handle_delayed_timer(unsigned long data)
{
struct delay_c *dc = (struct delay_c *)data;
- queue_work(kdelayd_wq, &dc->flush_expired_bios);
+ queue_work(dc->kdelayd_wq, &dc->flush_expired_bios);
}
static void queue_timeout(struct delay_c *dc, unsigned long expires)
goto bad_dev_write;
}
+ dc->kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
+ if (!dc->kdelayd_wq) {
+ DMERR("Couldn't start kdelayd");
+ goto bad_queue;
+ }
+
setup_timer(&dc->delay_timer, handle_delayed_timer, (unsigned long)dc);
INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
ti->private = dc;
return 0;
+bad_queue:
+ mempool_destroy(dc->delayed_pool);
bad_dev_write:
if (dc->dev_write)
dm_put_device(ti, dc->dev_write);
{
struct delay_c *dc = ti->private;
- flush_workqueue(kdelayd_wq);
+ destroy_workqueue(dc->kdelayd_wq);
dm_put_device(ti, dc->dev_read);
{
int r = -ENOMEM;
- kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
- if (!kdelayd_wq) {
- DMERR("Couldn't start kdelayd");
- goto bad_queue;
- }
-
delayed_cache = KMEM_CACHE(dm_delay_info, 0);
if (!delayed_cache) {
DMERR("Couldn't create delayed bio cache.");
bad_register:
kmem_cache_destroy(delayed_cache);
bad_memcache:
- destroy_workqueue(kdelayd_wq);
-bad_queue:
return r;
}
{
dm_unregister_target(&delay_target);
kmem_cache_destroy(delayed_cache);
- destroy_workqueue(kdelayd_wq);
}
/* Module hooks */
atomic_t pending_exceptions_count;
+ /* Protected by "lock" */
+ sector_t exception_start_sequence;
+
+ /* Protected by kcopyd single-threaded callback */
+ sector_t exception_complete_sequence;
+
+ /*
+ * A list of pending exceptions that completed out of order.
+ * Protected by kcopyd single-threaded callback.
+ */
+ struct list_head out_of_order_list;
+
mempool_t *pending_pool;
struct dm_exception_table pending;
*/
int started;
+ /* There was copying error. */
+ int copy_error;
+
+ /* A sequence number, it is used for in-order completion. */
+ sector_t exception_sequence;
+
+ struct list_head out_of_order_entry;
+
/*
* For writing a complete chunk, bypassing the copy.
*/
s->valid = 1;
s->active = 0;
atomic_set(&s->pending_exceptions_count, 0);
+ s->exception_start_sequence = 0;
+ s->exception_complete_sequence = 0;
+ INIT_LIST_HEAD(&s->out_of_order_list);
init_rwsem(&s->lock);
INIT_LIST_HEAD(&s->list);
spin_lock_init(&s->pe_lock);
pending_complete(pe, success);
}
+static void complete_exception(struct dm_snap_pending_exception *pe)
+{
+ struct dm_snapshot *s = pe->snap;
+
+ if (unlikely(pe->copy_error))
+ pending_complete(pe, 0);
+
+ else
+ /* Update the metadata if we are persistent */
+ s->store->type->commit_exception(s->store, &pe->e,
+ commit_callback, pe);
+}
+
/*
* Called when the copy I/O has finished. kcopyd actually runs
* this code so don't block.
struct dm_snap_pending_exception *pe = context;
struct dm_snapshot *s = pe->snap;
- if (read_err || write_err)
- pending_complete(pe, 0);
+ pe->copy_error = read_err || write_err;
- else
- /* Update the metadata if we are persistent */
- s->store->type->commit_exception(s->store, &pe->e,
- commit_callback, pe);
+ if (pe->exception_sequence == s->exception_complete_sequence) {
+ s->exception_complete_sequence++;
+ complete_exception(pe);
+
+ while (!list_empty(&s->out_of_order_list)) {
+ pe = list_entry(s->out_of_order_list.next,
+ struct dm_snap_pending_exception, out_of_order_entry);
+ if (pe->exception_sequence != s->exception_complete_sequence)
+ break;
+ s->exception_complete_sequence++;
+ list_del(&pe->out_of_order_entry);
+ complete_exception(pe);
+ }
+ } else {
+ struct list_head *lh;
+ struct dm_snap_pending_exception *pe2;
+
+ list_for_each_prev(lh, &s->out_of_order_list) {
+ pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry);
+ if (pe2->exception_sequence < pe->exception_sequence)
+ break;
+ }
+ list_add(&pe->out_of_order_entry, lh);
+ }
}
/*
return NULL;
}
+ pe->exception_sequence = s->exception_start_sequence++;
+
dm_insert_exception(&s->pending, &pe->e);
return pe;
static struct target_type snapshot_target = {
.name = "snapshot",
- .version = {1, 11, 1},
+ .version = {1, 12, 0},
.module = THIS_MODULE,
.ctr = snapshot_ctr,
.dtr = snapshot_dtr,
int __init dm_statistics_init(void)
{
+ shared_memory_amount = 0;
dm_stat_need_rcu_barrier = 0;
return 0;
}
num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
+ if (!num_targets) {
+ kfree(t);
+ return -ENOMEM;
+ }
+
if (alloc_targets(t, num_targets)) {
kfree(t);
return -ENOMEM;
up_write(&pmd->root_lock);
}
+void dm_pool_metadata_read_write(struct dm_pool_metadata *pmd)
+{
+ down_write(&pmd->root_lock);
+ pmd->read_only = false;
+ dm_bm_set_read_write(pmd->bm);
+ up_write(&pmd->root_lock);
+}
+
int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
dm_block_t threshold,
dm_sm_threshold_fn fn,
* that nothing is changing.
*/
void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd);
+void dm_pool_metadata_read_write(struct dm_pool_metadata *pmd);
int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
dm_block_t threshold,
*/
r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
if (r) {
- DMERR_LIMIT("dm_thin_insert_block() failed");
+ DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d",
+ dm_device_name(pool->pool_md), r);
+ set_pool_mode(pool, PM_READ_ONLY);
cell_error(pool, m->cell);
goto out;
}
}
}
-static int commit(struct pool *pool)
-{
- int r;
-
- r = dm_pool_commit_metadata(pool->pmd);
- if (r)
- DMERR_LIMIT("%s: commit failed: error = %d",
- dm_device_name(pool->pool_md), r);
-
- return r;
-}
-
/*
* A non-zero return indicates read_only or fail_io mode.
* Many callers don't care about the return value.
*/
-static int commit_or_fallback(struct pool *pool)
+static int commit(struct pool *pool)
{
int r;
if (get_pool_mode(pool) != PM_WRITE)
return -EINVAL;
- r = commit(pool);
- if (r)
+ r = dm_pool_commit_metadata(pool->pmd);
+ if (r) {
+ DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d",
+ dm_device_name(pool->pool_md), r);
set_pool_mode(pool, PM_READ_ONLY);
+ }
return r;
}
* Try to commit to see if that will free up some
* more space.
*/
- (void) commit_or_fallback(pool);
+ r = commit(pool);
+ if (r)
+ return r;
r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
if (r)
* table reload).
*/
if (!free_blocks) {
- DMWARN("%s: no free space available.",
+ DMWARN("%s: no free data space available.",
dm_device_name(pool->pool_md));
spin_lock_irqsave(&pool->lock, flags);
pool->no_free_space = 1;
}
r = dm_pool_alloc_data_block(pool->pmd, result);
- if (r)
+ if (r) {
+ if (r == -ENOSPC &&
+ !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) &&
+ !free_blocks) {
+ DMWARN("%s: no free metadata space available.",
+ dm_device_name(pool->pool_md));
+ set_pool_mode(pool, PM_READ_ONLY);
+ }
return r;
+ }
return 0;
}
if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
return;
- if (commit_or_fallback(pool)) {
+ if (commit(pool)) {
while ((bio = bio_list_pop(&bios)))
bio_io_error(bio);
return;
case PM_FAIL:
DMERR("%s: switching pool to failure mode",
dm_device_name(pool->pool_md));
+ dm_pool_metadata_read_only(pool->pmd);
pool->process_bio = process_bio_fail;
pool->process_discard = process_bio_fail;
pool->process_prepared_mapping = process_prepared_mapping_fail;
break;
case PM_WRITE:
+ dm_pool_metadata_read_write(pool->pmd);
pool->process_bio = process_bio;
pool->process_discard = process_discard;
pool->process_prepared_mapping = process_prepared_mapping;
struct pool_c *pt = ti->private;
/*
- * We want to make sure that degraded pools are never upgraded.
+ * We want to make sure that a pool in PM_FAIL mode is never upgraded.
*/
enum pool_mode old_mode = pool->pf.mode;
enum pool_mode new_mode = pt->adjusted_pf.mode;
- if (old_mode > new_mode)
+ /*
+ * If we were in PM_FAIL mode, rollback of metadata failed. We're
+ * not going to recover without a thin_repair. So we never let the
+ * pool move out of the old mode. On the other hand a PM_READ_ONLY
+ * may have been due to a lack of metadata or data space, and may
+ * now work (ie. if the underlying devices have been resized).
+ */
+ if (old_mode == PM_FAIL)
new_mode = old_mode;
pool->ti = ti;
return r;
if (need_commit1 || need_commit2)
- (void) commit_or_fallback(pool);
+ (void) commit(pool);
return 0;
}
cancel_delayed_work(&pool->waker);
flush_workqueue(pool->wq);
- (void) commit_or_fallback(pool);
+ (void) commit(pool);
}
static int check_arg_count(unsigned argc, unsigned args_required)
if (r)
return r;
- (void) commit_or_fallback(pool);
+ (void) commit(pool);
r = dm_pool_reserve_metadata_snap(pool->pmd);
if (r)
DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
if (!r)
- (void) commit_or_fallback(pool);
+ (void) commit(pool);
return r;
}
/* Commit to ensure statistics aren't out-of-date */
if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
- (void) commit_or_fallback(pool);
+ (void) commit(pool);
r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
if (r) {
* The shadow op will often be a noop. Only insert if it really
* copied data.
*/
- if (dm_block_location(*block) != b)
+ if (dm_block_location(*block) != b) {
+ /*
+ * dm_tm_shadow_block will have already decremented the old
+ * block, but it is still referenced by the btree. We
+ * increment to stop the insert decrementing it below zero
+ * when overwriting the old value.
+ */
+ dm_tm_inc(info->btree_info.tm, b);
r = insert_ablock(info, index, *block, root);
+ }
return r;
}
}
EXPORT_SYMBOL_GPL(dm_bm_set_read_only);
+void dm_bm_set_read_write(struct dm_block_manager *bm)
+{
+ bm->read_only = false;
+}
+EXPORT_SYMBOL_GPL(dm_bm_set_read_write);
+
u32 dm_bm_checksum(const void *data, size_t len, u32 init_xor)
{
return crc32c(~(u32) 0, data, len) ^ init_xor;
int dm_bm_flush_and_unlock(struct dm_block_manager *bm,
struct dm_block *superblock);
- /*
- * Request data be prefetched into the cache.
- */
+/*
+ * Request data is prefetched into the cache.
+ */
void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b);
/*
* be returned if you do.
*/
void dm_bm_set_read_only(struct dm_block_manager *bm);
+void dm_bm_set_read_write(struct dm_block_manager *bm);
u32 dm_bm_checksum(const void *data, size_t len, u32 init_xor);
}
static int sm_ll_mutate(struct ll_disk *ll, dm_block_t b,
- uint32_t (*mutator)(void *context, uint32_t old),
+ int (*mutator)(void *context, uint32_t old, uint32_t *new),
void *context, enum allocation_event *ev)
{
int r;
if (old > 2) {
r = sm_ll_lookup_big_ref_count(ll, b, &old);
- if (r < 0)
+ if (r < 0) {
+ dm_tm_unlock(ll->tm, nb);
return r;
+ }
}
- ref_count = mutator(context, old);
+ r = mutator(context, old, &ref_count);
+ if (r) {
+ dm_tm_unlock(ll->tm, nb);
+ return r;
+ }
if (ref_count <= 2) {
sm_set_bitmap(bm_le, bit, ref_count);
return ll->save_ie(ll, index, &ie_disk);
}
-static uint32_t set_ref_count(void *context, uint32_t old)
+static int set_ref_count(void *context, uint32_t old, uint32_t *new)
{
- return *((uint32_t *) context);
+ *new = *((uint32_t *) context);
+ return 0;
}
int sm_ll_insert(struct ll_disk *ll, dm_block_t b,
return sm_ll_mutate(ll, b, set_ref_count, &ref_count, ev);
}
-static uint32_t inc_ref_count(void *context, uint32_t old)
+static int inc_ref_count(void *context, uint32_t old, uint32_t *new)
{
- return old + 1;
+ *new = old + 1;
+ return 0;
}
int sm_ll_inc(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
return sm_ll_mutate(ll, b, inc_ref_count, NULL, ev);
}
-static uint32_t dec_ref_count(void *context, uint32_t old)
+static int dec_ref_count(void *context, uint32_t old, uint32_t *new)
{
- return old - 1;
+ if (!old) {
+ DMERR_LIMIT("unable to decrement a reference count below 0");
+ return -EINVAL;
+ }
+
+ *new = old - 1;
+ return 0;
}
int sm_ll_dec(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
int r = sm_metadata_new_block_(sm, b);
- if (r)
+ if (r) {
DMERR("unable to allocate new metadata block");
+ return r;
+ }
r = sm_metadata_get_nr_free(sm, &count);
- if (r)
+ if (r) {
DMERR("couldn't get free block count");
+ return r;
+ }
check_threshold(&smm->threshold, count);