Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / md / bcache / journal.c
diff --git a/kernel/drivers/md/bcache/journal.c b/kernel/drivers/md/bcache/journal.c
new file mode 100644 (file)
index 0000000..fe080ad
--- /dev/null
@@ -0,0 +1,821 @@
+/*
+ * bcache journalling code, for btree insertions
+ *
+ * Copyright 2012 Google, Inc.
+ */
+
+#include "bcache.h"
+#include "btree.h"
+#include "debug.h"
+#include "extents.h"
+
+#include <trace/events/bcache.h>
+
+/*
+ * Journal replay/recovery:
+ *
+ * This code is all driven from run_cache_set(); we first read the journal
+ * entries, do some other stuff, then we mark all the keys in the journal
+ * entries (same as garbage collection would), then we replay them - reinserting
+ * them into the cache in precisely the same order as they appear in the
+ * journal.
+ *
+ * We only journal keys that go in leaf nodes, which simplifies things quite a
+ * bit.
+ */
+
+static void journal_read_endio(struct bio *bio, int error)
+{
+       struct closure *cl = bio->bi_private;
+       closure_put(cl);
+}
+
+static int journal_read_bucket(struct cache *ca, struct list_head *list,
+                              unsigned bucket_index)
+{
+       struct journal_device *ja = &ca->journal;
+       struct bio *bio = &ja->bio;
+
+       struct journal_replay *i;
+       struct jset *j, *data = ca->set->journal.w[0].data;
+       struct closure cl;
+       unsigned len, left, offset = 0;
+       int ret = 0;
+       sector_t bucket = bucket_to_sector(ca->set, ca->sb.d[bucket_index]);
+
+       closure_init_stack(&cl);
+
+       pr_debug("reading %u", bucket_index);
+
+       while (offset < ca->sb.bucket_size) {
+reread:                left = ca->sb.bucket_size - offset;
+               len = min_t(unsigned, left, PAGE_SECTORS << JSET_BITS);
+
+               bio_reset(bio);
+               bio->bi_iter.bi_sector  = bucket + offset;
+               bio->bi_bdev    = ca->bdev;
+               bio->bi_rw      = READ;
+               bio->bi_iter.bi_size    = len << 9;
+
+               bio->bi_end_io  = journal_read_endio;
+               bio->bi_private = &cl;
+               bch_bio_map(bio, data);
+
+               closure_bio_submit(bio, &cl, ca);
+               closure_sync(&cl);
+
+               /* This function could be simpler now since we no longer write
+                * journal entries that overlap bucket boundaries; this means
+                * the start of a bucket will always have a valid journal entry
+                * if it has any journal entries at all.
+                */
+
+               j = data;
+               while (len) {
+                       struct list_head *where;
+                       size_t blocks, bytes = set_bytes(j);
+
+                       if (j->magic != jset_magic(&ca->sb)) {
+                               pr_debug("%u: bad magic", bucket_index);
+                               return ret;
+                       }
+
+                       if (bytes > left << 9 ||
+                           bytes > PAGE_SIZE << JSET_BITS) {
+                               pr_info("%u: too big, %zu bytes, offset %u",
+                                       bucket_index, bytes, offset);
+                               return ret;
+                       }
+
+                       if (bytes > len << 9)
+                               goto reread;
+
+                       if (j->csum != csum_set(j)) {
+                               pr_info("%u: bad csum, %zu bytes, offset %u",
+                                       bucket_index, bytes, offset);
+                               return ret;
+                       }
+
+                       blocks = set_blocks(j, block_bytes(ca->set));
+
+                       while (!list_empty(list)) {
+                               i = list_first_entry(list,
+                                       struct journal_replay, list);
+                               if (i->j.seq >= j->last_seq)
+                                       break;
+                               list_del(&i->list);
+                               kfree(i);
+                       }
+
+                       list_for_each_entry_reverse(i, list, list) {
+                               if (j->seq == i->j.seq)
+                                       goto next_set;
+
+                               if (j->seq < i->j.last_seq)
+                                       goto next_set;
+
+                               if (j->seq > i->j.seq) {
+                                       where = &i->list;
+                                       goto add;
+                               }
+                       }
+
+                       where = list;
+add:
+                       i = kmalloc(offsetof(struct journal_replay, j) +
+                                   bytes, GFP_KERNEL);
+                       if (!i)
+                               return -ENOMEM;
+                       memcpy(&i->j, j, bytes);
+                       list_add(&i->list, where);
+                       ret = 1;
+
+                       ja->seq[bucket_index] = j->seq;
+next_set:
+                       offset  += blocks * ca->sb.block_size;
+                       len     -= blocks * ca->sb.block_size;
+                       j = ((void *) j) + blocks * block_bytes(ca);
+               }
+       }
+
+       return ret;
+}
+
+int bch_journal_read(struct cache_set *c, struct list_head *list)
+{
+#define read_bucket(b)                                                 \
+       ({                                                              \
+               int ret = journal_read_bucket(ca, list, b);             \
+               __set_bit(b, bitmap);                                   \
+               if (ret < 0)                                            \
+                       return ret;                                     \
+               ret;                                                    \
+       })
+
+       struct cache *ca;
+       unsigned iter;
+
+       for_each_cache(ca, c, iter) {
+               struct journal_device *ja = &ca->journal;
+               unsigned long bitmap[SB_JOURNAL_BUCKETS / BITS_PER_LONG];
+               unsigned i, l, r, m;
+               uint64_t seq;
+
+               bitmap_zero(bitmap, SB_JOURNAL_BUCKETS);
+               pr_debug("%u journal buckets", ca->sb.njournal_buckets);
+
+               /*
+                * Read journal buckets ordered by golden ratio hash to quickly
+                * find a sequence of buckets with valid journal entries
+                */
+               for (i = 0; i < ca->sb.njournal_buckets; i++) {
+                       l = (i * 2654435769U) % ca->sb.njournal_buckets;
+
+                       if (test_bit(l, bitmap))
+                               break;
+
+                       if (read_bucket(l))
+                               goto bsearch;
+               }
+
+               /*
+                * If that fails, check all the buckets we haven't checked
+                * already
+                */
+               pr_debug("falling back to linear search");
+
+               for (l = find_first_zero_bit(bitmap, ca->sb.njournal_buckets);
+                    l < ca->sb.njournal_buckets;
+                    l = find_next_zero_bit(bitmap, ca->sb.njournal_buckets, l + 1))
+                       if (read_bucket(l))
+                               goto bsearch;
+
+               /* no journal entries on this device? */
+               if (l == ca->sb.njournal_buckets)
+                       continue;
+bsearch:
+               BUG_ON(list_empty(list));
+
+               /* Binary search */
+               m = l;
+               r = find_next_bit(bitmap, ca->sb.njournal_buckets, l + 1);
+               pr_debug("starting binary search, l %u r %u", l, r);
+
+               while (l + 1 < r) {
+                       seq = list_entry(list->prev, struct journal_replay,
+                                        list)->j.seq;
+
+                       m = (l + r) >> 1;
+                       read_bucket(m);
+
+                       if (seq != list_entry(list->prev, struct journal_replay,
+                                             list)->j.seq)
+                               l = m;
+                       else
+                               r = m;
+               }
+
+               /*
+                * Read buckets in reverse order until we stop finding more
+                * journal entries
+                */
+               pr_debug("finishing up: m %u njournal_buckets %u",
+                        m, ca->sb.njournal_buckets);
+               l = m;
+
+               while (1) {
+                       if (!l--)
+                               l = ca->sb.njournal_buckets - 1;
+
+                       if (l == m)
+                               break;
+
+                       if (test_bit(l, bitmap))
+                               continue;
+
+                       if (!read_bucket(l))
+                               break;
+               }
+
+               seq = 0;
+
+               for (i = 0; i < ca->sb.njournal_buckets; i++)
+                       if (ja->seq[i] > seq) {
+                               seq = ja->seq[i];
+                               /*
+                                * When journal_reclaim() goes to allocate for
+                                * the first time, it'll use the bucket after
+                                * ja->cur_idx
+                                */
+                               ja->cur_idx = i;
+                               ja->last_idx = ja->discard_idx = (i + 1) %
+                                       ca->sb.njournal_buckets;
+
+                       }
+       }
+
+       if (!list_empty(list))
+               c->journal.seq = list_entry(list->prev,
+                                           struct journal_replay,
+                                           list)->j.seq;
+
+       return 0;
+#undef read_bucket
+}
+
+void bch_journal_mark(struct cache_set *c, struct list_head *list)
+{
+       atomic_t p = { 0 };
+       struct bkey *k;
+       struct journal_replay *i;
+       struct journal *j = &c->journal;
+       uint64_t last = j->seq;
+
+       /*
+        * journal.pin should never fill up - we never write a journal
+        * entry when it would fill up. But if for some reason it does, we
+        * iterate over the list in reverse order so that we can just skip that
+        * refcount instead of bugging.
+        */
+
+       list_for_each_entry_reverse(i, list, list) {
+               BUG_ON(last < i->j.seq);
+               i->pin = NULL;
+
+               while (last-- != i->j.seq)
+                       if (fifo_free(&j->pin) > 1) {
+                               fifo_push_front(&j->pin, p);
+                               atomic_set(&fifo_front(&j->pin), 0);
+                       }
+
+               if (fifo_free(&j->pin) > 1) {
+                       fifo_push_front(&j->pin, p);
+                       i->pin = &fifo_front(&j->pin);
+                       atomic_set(i->pin, 1);
+               }
+
+               for (k = i->j.start;
+                    k < bset_bkey_last(&i->j);
+                    k = bkey_next(k))
+                       if (!__bch_extent_invalid(c, k)) {
+                               unsigned j;
+
+                               for (j = 0; j < KEY_PTRS(k); j++)
+                                       if (ptr_available(c, k, j))
+                                               atomic_inc(&PTR_BUCKET(c, k, j)->pin);
+
+                               bch_initial_mark_key(c, 0, k);
+                       }
+       }
+}
+
+int bch_journal_replay(struct cache_set *s, struct list_head *list)
+{
+       int ret = 0, keys = 0, entries = 0;
+       struct bkey *k;
+       struct journal_replay *i =
+               list_entry(list->prev, struct journal_replay, list);
+
+       uint64_t start = i->j.last_seq, end = i->j.seq, n = start;
+       struct keylist keylist;
+
+       list_for_each_entry(i, list, list) {
+               BUG_ON(i->pin && atomic_read(i->pin) != 1);
+
+               cache_set_err_on(n != i->j.seq, s,
+"bcache: journal entries %llu-%llu missing! (replaying %llu-%llu)",
+                                n, i->j.seq - 1, start, end);
+
+               for (k = i->j.start;
+                    k < bset_bkey_last(&i->j);
+                    k = bkey_next(k)) {
+                       trace_bcache_journal_replay_key(k);
+
+                       bch_keylist_init_single(&keylist, k);
+
+                       ret = bch_btree_insert(s, &keylist, i->pin, NULL);
+                       if (ret)
+                               goto err;
+
+                       BUG_ON(!bch_keylist_empty(&keylist));
+                       keys++;
+
+                       cond_resched();
+               }
+
+               if (i->pin)
+                       atomic_dec(i->pin);
+               n = i->j.seq + 1;
+               entries++;
+       }
+
+       pr_info("journal replay done, %i keys in %i entries, seq %llu",
+               keys, entries, end);
+err:
+       while (!list_empty(list)) {
+               i = list_first_entry(list, struct journal_replay, list);
+               list_del(&i->list);
+               kfree(i);
+       }
+
+       return ret;
+}
+
+/* Journalling */
+
+static void btree_flush_write(struct cache_set *c)
+{
+       /*
+        * Try to find the btree node with that references the oldest journal
+        * entry, best is our current candidate and is locked if non NULL:
+        */
+       struct btree *b, *best;
+       unsigned i;
+retry:
+       best = NULL;
+
+       for_each_cached_btree(b, c, i)
+               if (btree_current_write(b)->journal) {
+                       if (!best)
+                               best = b;
+                       else if (journal_pin_cmp(c,
+                                       btree_current_write(best)->journal,
+                                       btree_current_write(b)->journal)) {
+                               best = b;
+                       }
+               }
+
+       b = best;
+       if (b) {
+               mutex_lock(&b->write_lock);
+               if (!btree_current_write(b)->journal) {
+                       mutex_unlock(&b->write_lock);
+                       /* We raced */
+                       goto retry;
+               }
+
+               __bch_btree_node_write(b, NULL);
+               mutex_unlock(&b->write_lock);
+       }
+}
+
+#define last_seq(j)    ((j)->seq - fifo_used(&(j)->pin) + 1)
+
+static void journal_discard_endio(struct bio *bio, int error)
+{
+       struct journal_device *ja =
+               container_of(bio, struct journal_device, discard_bio);
+       struct cache *ca = container_of(ja, struct cache, journal);
+
+       atomic_set(&ja->discard_in_flight, DISCARD_DONE);
+
+       closure_wake_up(&ca->set->journal.wait);
+       closure_put(&ca->set->cl);
+}
+
+static void journal_discard_work(struct work_struct *work)
+{
+       struct journal_device *ja =
+               container_of(work, struct journal_device, discard_work);
+
+       submit_bio(0, &ja->discard_bio);
+}
+
+static void do_journal_discard(struct cache *ca)
+{
+       struct journal_device *ja = &ca->journal;
+       struct bio *bio = &ja->discard_bio;
+
+       if (!ca->discard) {
+               ja->discard_idx = ja->last_idx;
+               return;
+       }
+
+       switch (atomic_read(&ja->discard_in_flight)) {
+       case DISCARD_IN_FLIGHT:
+               return;
+
+       case DISCARD_DONE:
+               ja->discard_idx = (ja->discard_idx + 1) %
+                       ca->sb.njournal_buckets;
+
+               atomic_set(&ja->discard_in_flight, DISCARD_READY);
+               /* fallthrough */
+
+       case DISCARD_READY:
+               if (ja->discard_idx == ja->last_idx)
+                       return;
+
+               atomic_set(&ja->discard_in_flight, DISCARD_IN_FLIGHT);
+
+               bio_init(bio);
+               bio->bi_iter.bi_sector  = bucket_to_sector(ca->set,
+                                               ca->sb.d[ja->discard_idx]);
+               bio->bi_bdev            = ca->bdev;
+               bio->bi_rw              = REQ_WRITE|REQ_DISCARD;
+               bio->bi_max_vecs        = 1;
+               bio->bi_io_vec          = bio->bi_inline_vecs;
+               bio->bi_iter.bi_size    = bucket_bytes(ca);
+               bio->bi_end_io          = journal_discard_endio;
+
+               closure_get(&ca->set->cl);
+               INIT_WORK(&ja->discard_work, journal_discard_work);
+               schedule_work(&ja->discard_work);
+       }
+}
+
+static void journal_reclaim(struct cache_set *c)
+{
+       struct bkey *k = &c->journal.key;
+       struct cache *ca;
+       uint64_t last_seq;
+       unsigned iter, n = 0;
+       atomic_t p;
+
+       while (!atomic_read(&fifo_front(&c->journal.pin)))
+               fifo_pop(&c->journal.pin, p);
+
+       last_seq = last_seq(&c->journal);
+
+       /* Update last_idx */
+
+       for_each_cache(ca, c, iter) {
+               struct journal_device *ja = &ca->journal;
+
+               while (ja->last_idx != ja->cur_idx &&
+                      ja->seq[ja->last_idx] < last_seq)
+                       ja->last_idx = (ja->last_idx + 1) %
+                               ca->sb.njournal_buckets;
+       }
+
+       for_each_cache(ca, c, iter)
+               do_journal_discard(ca);
+
+       if (c->journal.blocks_free)
+               goto out;
+
+       /*
+        * Allocate:
+        * XXX: Sort by free journal space
+        */
+
+       for_each_cache(ca, c, iter) {
+               struct journal_device *ja = &ca->journal;
+               unsigned next = (ja->cur_idx + 1) % ca->sb.njournal_buckets;
+
+               /* No space available on this device */
+               if (next == ja->discard_idx)
+                       continue;
+
+               ja->cur_idx = next;
+               k->ptr[n++] = PTR(0,
+                                 bucket_to_sector(c, ca->sb.d[ja->cur_idx]),
+                                 ca->sb.nr_this_dev);
+       }
+
+       bkey_init(k);
+       SET_KEY_PTRS(k, n);
+
+       if (n)
+               c->journal.blocks_free = c->sb.bucket_size >> c->block_bits;
+out:
+       if (!journal_full(&c->journal))
+               __closure_wake_up(&c->journal.wait);
+}
+
+void bch_journal_next(struct journal *j)
+{
+       atomic_t p = { 1 };
+
+       j->cur = (j->cur == j->w)
+               ? &j->w[1]
+               : &j->w[0];
+
+       /*
+        * The fifo_push() needs to happen at the same time as j->seq is
+        * incremented for last_seq() to be calculated correctly
+        */
+       BUG_ON(!fifo_push(&j->pin, p));
+       atomic_set(&fifo_back(&j->pin), 1);
+
+       j->cur->data->seq       = ++j->seq;
+       j->cur->dirty           = false;
+       j->cur->need_write      = false;
+       j->cur->data->keys      = 0;
+
+       if (fifo_full(&j->pin))
+               pr_debug("journal_pin full (%zu)", fifo_used(&j->pin));
+}
+
+static void journal_write_endio(struct bio *bio, int error)
+{
+       struct journal_write *w = bio->bi_private;
+
+       cache_set_err_on(error, w->c, "journal io error");
+       closure_put(&w->c->journal.io);
+}
+
+static void journal_write(struct closure *);
+
+static void journal_write_done(struct closure *cl)
+{
+       struct journal *j = container_of(cl, struct journal, io);
+       struct journal_write *w = (j->cur == j->w)
+               ? &j->w[1]
+               : &j->w[0];
+
+       __closure_wake_up(&w->wait);
+       continue_at_nobarrier(cl, journal_write, system_wq);
+}
+
+static void journal_write_unlock(struct closure *cl)
+{
+       struct cache_set *c = container_of(cl, struct cache_set, journal.io);
+
+       c->journal.io_in_flight = 0;
+       spin_unlock(&c->journal.lock);
+}
+
+static void journal_write_unlocked(struct closure *cl)
+       __releases(c->journal.lock)
+{
+       struct cache_set *c = container_of(cl, struct cache_set, journal.io);
+       struct cache *ca;
+       struct journal_write *w = c->journal.cur;
+       struct bkey *k = &c->journal.key;
+       unsigned i, sectors = set_blocks(w->data, block_bytes(c)) *
+               c->sb.block_size;
+
+       struct bio *bio;
+       struct bio_list list;
+       bio_list_init(&list);
+
+       if (!w->need_write) {
+               closure_return_with_destructor(cl, journal_write_unlock);
+       } else if (journal_full(&c->journal)) {
+               journal_reclaim(c);
+               spin_unlock(&c->journal.lock);
+
+               btree_flush_write(c);
+               continue_at(cl, journal_write, system_wq);
+       }
+
+       c->journal.blocks_free -= set_blocks(w->data, block_bytes(c));
+
+       w->data->btree_level = c->root->level;
+
+       bkey_copy(&w->data->btree_root, &c->root->key);
+       bkey_copy(&w->data->uuid_bucket, &c->uuid_bucket);
+
+       for_each_cache(ca, c, i)
+               w->data->prio_bucket[ca->sb.nr_this_dev] = ca->prio_buckets[0];
+
+       w->data->magic          = jset_magic(&c->sb);
+       w->data->version        = BCACHE_JSET_VERSION;
+       w->data->last_seq       = last_seq(&c->journal);
+       w->data->csum           = csum_set(w->data);
+
+       for (i = 0; i < KEY_PTRS(k); i++) {
+               ca = PTR_CACHE(c, k, i);
+               bio = &ca->journal.bio;
+
+               atomic_long_add(sectors, &ca->meta_sectors_written);
+
+               bio_reset(bio);
+               bio->bi_iter.bi_sector  = PTR_OFFSET(k, i);
+               bio->bi_bdev    = ca->bdev;
+               bio->bi_rw      = REQ_WRITE|REQ_SYNC|REQ_META|REQ_FLUSH|REQ_FUA;
+               bio->bi_iter.bi_size = sectors << 9;
+
+               bio->bi_end_io  = journal_write_endio;
+               bio->bi_private = w;
+               bch_bio_map(bio, w->data);
+
+               trace_bcache_journal_write(bio);
+               bio_list_add(&list, bio);
+
+               SET_PTR_OFFSET(k, i, PTR_OFFSET(k, i) + sectors);
+
+               ca->journal.seq[ca->journal.cur_idx] = w->data->seq;
+       }
+
+       atomic_dec_bug(&fifo_back(&c->journal.pin));
+       bch_journal_next(&c->journal);
+       journal_reclaim(c);
+
+       spin_unlock(&c->journal.lock);
+
+       while ((bio = bio_list_pop(&list)))
+               closure_bio_submit(bio, cl, c->cache[0]);
+
+       continue_at(cl, journal_write_done, NULL);
+}
+
+static void journal_write(struct closure *cl)
+{
+       struct cache_set *c = container_of(cl, struct cache_set, journal.io);
+
+       spin_lock(&c->journal.lock);
+       journal_write_unlocked(cl);
+}
+
+static void journal_try_write(struct cache_set *c)
+       __releases(c->journal.lock)
+{
+       struct closure *cl = &c->journal.io;
+       struct journal_write *w = c->journal.cur;
+
+       w->need_write = true;
+
+       if (!c->journal.io_in_flight) {
+               c->journal.io_in_flight = 1;
+               closure_call(cl, journal_write_unlocked, NULL, &c->cl);
+       } else {
+               spin_unlock(&c->journal.lock);
+       }
+}
+
+static struct journal_write *journal_wait_for_write(struct cache_set *c,
+                                                   unsigned nkeys)
+{
+       size_t sectors;
+       struct closure cl;
+       bool wait = false;
+
+       closure_init_stack(&cl);
+
+       spin_lock(&c->journal.lock);
+
+       while (1) {
+               struct journal_write *w = c->journal.cur;
+
+               sectors = __set_blocks(w->data, w->data->keys + nkeys,
+                                      block_bytes(c)) * c->sb.block_size;
+
+               if (sectors <= min_t(size_t,
+                                    c->journal.blocks_free * c->sb.block_size,
+                                    PAGE_SECTORS << JSET_BITS))
+                       return w;
+
+               if (wait)
+                       closure_wait(&c->journal.wait, &cl);
+
+               if (!journal_full(&c->journal)) {
+                       if (wait)
+                               trace_bcache_journal_entry_full(c);
+
+                       /*
+                        * XXX: If we were inserting so many keys that they
+                        * won't fit in an _empty_ journal write, we'll
+                        * deadlock. For now, handle this in
+                        * bch_keylist_realloc() - but something to think about.
+                        */
+                       BUG_ON(!w->data->keys);
+
+                       journal_try_write(c); /* unlocks */
+               } else {
+                       if (wait)
+                               trace_bcache_journal_full(c);
+
+                       journal_reclaim(c);
+                       spin_unlock(&c->journal.lock);
+
+                       btree_flush_write(c);
+               }
+
+               closure_sync(&cl);
+               spin_lock(&c->journal.lock);
+               wait = true;
+       }
+}
+
+static void journal_write_work(struct work_struct *work)
+{
+       struct cache_set *c = container_of(to_delayed_work(work),
+                                          struct cache_set,
+                                          journal.work);
+       spin_lock(&c->journal.lock);
+       if (c->journal.cur->dirty)
+               journal_try_write(c);
+       else
+               spin_unlock(&c->journal.lock);
+}
+
+/*
+ * Entry point to the journalling code - bio_insert() and btree_invalidate()
+ * pass bch_journal() a list of keys to be journalled, and then
+ * bch_journal() hands those same keys off to btree_insert_async()
+ */
+
+atomic_t *bch_journal(struct cache_set *c,
+                     struct keylist *keys,
+                     struct closure *parent)
+{
+       struct journal_write *w;
+       atomic_t *ret;
+
+       if (!CACHE_SYNC(&c->sb))
+               return NULL;
+
+       w = journal_wait_for_write(c, bch_keylist_nkeys(keys));
+
+       memcpy(bset_bkey_last(w->data), keys->keys, bch_keylist_bytes(keys));
+       w->data->keys += bch_keylist_nkeys(keys);
+
+       ret = &fifo_back(&c->journal.pin);
+       atomic_inc(ret);
+
+       if (parent) {
+               closure_wait(&w->wait, parent);
+               journal_try_write(c);
+       } else if (!w->dirty) {
+               w->dirty = true;
+               schedule_delayed_work(&c->journal.work,
+                                     msecs_to_jiffies(c->journal_delay_ms));
+               spin_unlock(&c->journal.lock);
+       } else {
+               spin_unlock(&c->journal.lock);
+       }
+
+
+       return ret;
+}
+
+void bch_journal_meta(struct cache_set *c, struct closure *cl)
+{
+       struct keylist keys;
+       atomic_t *ref;
+
+       bch_keylist_init(&keys);
+
+       ref = bch_journal(c, &keys, cl);
+       if (ref)
+               atomic_dec_bug(ref);
+}
+
+void bch_journal_free(struct cache_set *c)
+{
+       free_pages((unsigned long) c->journal.w[1].data, JSET_BITS);
+       free_pages((unsigned long) c->journal.w[0].data, JSET_BITS);
+       free_fifo(&c->journal.pin);
+}
+
+int bch_journal_alloc(struct cache_set *c)
+{
+       struct journal *j = &c->journal;
+
+       spin_lock_init(&j->lock);
+       INIT_DELAYED_WORK(&j->work, journal_write_work);
+
+       c->journal_delay_ms = 100;
+
+       j->w[0].c = c;
+       j->w[1].c = c;
+
+       if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)) ||
+           !(j->w[0].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS)) ||
+           !(j->w[1].data = (void *) __get_free_pages(GFP_KERNEL, JSET_BITS)))
+               return -ENOMEM;
+
+       return 0;
+}