dm cache: add cleaner policy
authorHeinz Mauelshagen <mauelshagen@redhat.com>
Fri, 1 Mar 2013 22:45:52 +0000 (22:45 +0000)
committerAlasdair G Kergon <agk@redhat.com>
Fri, 1 Mar 2013 22:45:52 +0000 (22:45 +0000)
A simple cache policy that writes back all data to the origin.

This is used to decommission a dm cache by emptying it.

Signed-off-by: Heinz Mauelshagen <mauelshagen@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Documentation/device-mapper/cache-policies.txt
drivers/md/Kconfig
drivers/md/Makefile
drivers/md/dm-cache-policy-cleaner.c [new file with mode: 0644]

index 731879f97b8016a0083caf3874ebad8b01f47c80..d7c440b444cc012b8786e4de104577dd47017a41 100644 (file)
@@ -53,6 +53,11 @@ since spindles tend to have good bandwidth. The io_tracker counts
 contiguous I/Os to try to spot when the io is in one of these sequential
 modes.
 
+cleaner
+-------
+
+The cleaner writes back all dirty blocks in a cache to decommission it.
+
 Examples
 ========
 
index 1a96cbc7afda741bbea12e1617ca2ca1cf963282..e30b490055aa8ce3ef3edf2229b6cd2724bf794a 100644 (file)
@@ -291,6 +291,14 @@ config DM_CACHE_MQ
          This is meant to be a general purpose policy.  It prioritises
          reads over writes.
 
+config DM_CACHE_CLEANER
+       tristate "Cleaner Cache Policy (EXPERIMENTAL)"
+       depends on DM_CACHE
+       default y
+       ---help---
+         A simple cache policy that writes back all data to the
+         origin.  Used when decommissioning a dm-cache.
+
 config DM_MIRROR
        tristate "Mirror target"
        depends on BLK_DEV_DM
index adc8710c24083fbb0d83cf36fa1f186214c58b50..7ceeaefc0e95d0fa8b75a6bdc2c2a5846f3144fa 100644 (file)
@@ -13,6 +13,7 @@ dm-log-userspace-y \
 dm-thin-pool-y += dm-thin.o dm-thin-metadata.o
 dm-cache-y     += dm-cache-target.o dm-cache-metadata.o dm-cache-policy.o
 dm-cache-mq-y   += dm-cache-policy-mq.o
+dm-cache-cleaner-y += dm-cache-policy-cleaner.o
 md-mod-y       += md.o bitmap.o
 raid456-y      += raid5.o
 
@@ -48,6 +49,7 @@ obj-$(CONFIG_DM_THIN_PROVISIONING)    += dm-thin-pool.o
 obj-$(CONFIG_DM_VERITY)                += dm-verity.o
 obj-$(CONFIG_DM_CACHE)         += dm-cache.o
 obj-$(CONFIG_DM_CACHE_MQ)      += dm-cache-mq.o
+obj-$(CONFIG_DM_CACHE_CLEANER) += dm-cache-cleaner.o
 
 ifeq ($(CONFIG_DM_UEVENT),y)
 dm-mod-objs                    += dm-uevent.o
diff --git a/drivers/md/dm-cache-policy-cleaner.c b/drivers/md/dm-cache-policy-cleaner.c
new file mode 100644 (file)
index 0000000..cc05d70
--- /dev/null
@@ -0,0 +1,464 @@
+/*
+ * Copyright (C) 2012 Red Hat. All rights reserved.
+ *
+ * writeback cache policy supporting flushing out dirty cache blocks.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-cache-policy.h"
+#include "dm.h"
+
+#include <linux/hash.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+
+/*----------------------------------------------------------------*/
+
+#define DM_MSG_PREFIX "cache cleaner"
+#define CLEANER_VERSION "1.0.0"
+
+/* Cache entry struct. */
+struct wb_cache_entry {
+       struct list_head list;
+       struct hlist_node hlist;
+
+       dm_oblock_t oblock;
+       dm_cblock_t cblock;
+       bool dirty:1;
+       bool pending:1;
+};
+
+struct hash {
+       struct hlist_head *table;
+       dm_block_t hash_bits;
+       unsigned nr_buckets;
+};
+
+struct policy {
+       struct dm_cache_policy policy;
+       spinlock_t lock;
+
+       struct list_head free;
+       struct list_head clean;
+       struct list_head clean_pending;
+       struct list_head dirty;
+
+       /*
+        * We know exactly how many cblocks will be needed,
+        * so we can allocate them up front.
+        */
+       dm_cblock_t cache_size, nr_cblocks_allocated;
+       struct wb_cache_entry *cblocks;
+       struct hash chash;
+};
+
+/*----------------------------------------------------------------------------*/
+
+/*
+ * Low-level functions.
+ */
+static unsigned next_power(unsigned n, unsigned min)
+{
+       return roundup_pow_of_two(max(n, min));
+}
+
+static struct policy *to_policy(struct dm_cache_policy *p)
+{
+       return container_of(p, struct policy, policy);
+}
+
+static struct list_head *list_pop(struct list_head *q)
+{
+       struct list_head *r = q->next;
+
+       list_del(r);
+
+       return r;
+}
+
+/*----------------------------------------------------------------------------*/
+
+/* Allocate/free various resources. */
+static int alloc_hash(struct hash *hash, unsigned elts)
+{
+       hash->nr_buckets = next_power(elts >> 4, 16);
+       hash->hash_bits = ffs(hash->nr_buckets) - 1;
+       hash->table = vzalloc(sizeof(*hash->table) * hash->nr_buckets);
+
+       return hash->table ? 0 : -ENOMEM;
+}
+
+static void free_hash(struct hash *hash)
+{
+       vfree(hash->table);
+}
+
+static int alloc_cache_blocks_with_hash(struct policy *p, dm_cblock_t cache_size)
+{
+       int r = -ENOMEM;
+
+       p->cblocks = vzalloc(sizeof(*p->cblocks) * from_cblock(cache_size));
+       if (p->cblocks) {
+               unsigned u = from_cblock(cache_size);
+
+               while (u--)
+                       list_add(&p->cblocks[u].list, &p->free);
+
+               p->nr_cblocks_allocated = 0;
+
+               /* Cache entries hash. */
+               r = alloc_hash(&p->chash, from_cblock(cache_size));
+               if (r)
+                       vfree(p->cblocks);
+       }
+
+       return r;
+}
+
+static void free_cache_blocks_and_hash(struct policy *p)
+{
+       free_hash(&p->chash);
+       vfree(p->cblocks);
+}
+
+static struct wb_cache_entry *alloc_cache_entry(struct policy *p)
+{
+       struct wb_cache_entry *e;
+
+       BUG_ON(from_cblock(p->nr_cblocks_allocated) >= from_cblock(p->cache_size));
+
+       e = list_entry(list_pop(&p->free), struct wb_cache_entry, list);
+       p->nr_cblocks_allocated = to_cblock(from_cblock(p->nr_cblocks_allocated) + 1);
+
+       return e;
+}
+
+/*----------------------------------------------------------------------------*/
+
+/* Hash functions (lookup, insert, remove). */
+static struct wb_cache_entry *lookup_cache_entry(struct policy *p, dm_oblock_t oblock)
+{
+       struct hash *hash = &p->chash;
+       unsigned h = hash_64(from_oblock(oblock), hash->hash_bits);
+       struct wb_cache_entry *cur;
+       struct hlist_head *bucket = &hash->table[h];
+
+       hlist_for_each_entry(cur, bucket, hlist) {
+               if (cur->oblock == oblock) {
+                       /* Move upfront bucket for faster access. */
+                       hlist_del(&cur->hlist);
+                       hlist_add_head(&cur->hlist, bucket);
+                       return cur;
+               }
+       }
+
+       return NULL;
+}
+
+static void insert_cache_hash_entry(struct policy *p, struct wb_cache_entry *e)
+{
+       unsigned h = hash_64(from_oblock(e->oblock), p->chash.hash_bits);
+
+       hlist_add_head(&e->hlist, &p->chash.table[h]);
+}
+
+static void remove_cache_hash_entry(struct wb_cache_entry *e)
+{
+       hlist_del(&e->hlist);
+}
+
+/* Public interface (see dm-cache-policy.h */
+static int wb_map(struct dm_cache_policy *pe, dm_oblock_t oblock,
+                 bool can_block, bool can_migrate, bool discarded_oblock,
+                 struct bio *bio, struct policy_result *result)
+{
+       struct policy *p = to_policy(pe);
+       struct wb_cache_entry *e;
+       unsigned long flags;
+
+       result->op = POLICY_MISS;
+
+       if (can_block)
+               spin_lock_irqsave(&p->lock, flags);
+
+       else if (!spin_trylock_irqsave(&p->lock, flags))
+               return -EWOULDBLOCK;
+
+       e = lookup_cache_entry(p, oblock);
+       if (e) {
+               result->op = POLICY_HIT;
+               result->cblock = e->cblock;
+
+       }
+
+       spin_unlock_irqrestore(&p->lock, flags);
+
+       return 0;
+}
+
+static int wb_lookup(struct dm_cache_policy *pe, dm_oblock_t oblock, dm_cblock_t *cblock)
+{
+       int r;
+       struct policy *p = to_policy(pe);
+       struct wb_cache_entry *e;
+       unsigned long flags;
+
+       if (!spin_trylock_irqsave(&p->lock, flags))
+               return -EWOULDBLOCK;
+
+       e = lookup_cache_entry(p, oblock);
+       if (e) {
+               *cblock = e->cblock;
+               r = 0;
+
+       } else
+               r = -ENOENT;
+
+       spin_unlock_irqrestore(&p->lock, flags);
+
+       return r;
+}
+
+static void __set_clear_dirty(struct dm_cache_policy *pe, dm_oblock_t oblock, bool set)
+{
+       struct policy *p = to_policy(pe);
+       struct wb_cache_entry *e;
+
+       e = lookup_cache_entry(p, oblock);
+       BUG_ON(!e);
+
+       if (set) {
+               if (!e->dirty) {
+                       e->dirty = true;
+                       list_move(&e->list, &p->dirty);
+               }
+
+       } else {
+               if (e->dirty) {
+                       e->pending = false;
+                       e->dirty = false;
+                       list_move(&e->list, &p->clean);
+               }
+       }
+}
+
+static void wb_set_dirty(struct dm_cache_policy *pe, dm_oblock_t oblock)
+{
+       struct policy *p = to_policy(pe);
+       unsigned long flags;
+
+       spin_lock_irqsave(&p->lock, flags);
+       __set_clear_dirty(pe, oblock, true);
+       spin_unlock_irqrestore(&p->lock, flags);
+}
+
+static void wb_clear_dirty(struct dm_cache_policy *pe, dm_oblock_t oblock)
+{
+       struct policy *p = to_policy(pe);
+       unsigned long flags;
+
+       spin_lock_irqsave(&p->lock, flags);
+       __set_clear_dirty(pe, oblock, false);
+       spin_unlock_irqrestore(&p->lock, flags);
+}
+
+static void add_cache_entry(struct policy *p, struct wb_cache_entry *e)
+{
+       insert_cache_hash_entry(p, e);
+       if (e->dirty)
+               list_add(&e->list, &p->dirty);
+       else
+               list_add(&e->list, &p->clean);
+}
+
+static int wb_load_mapping(struct dm_cache_policy *pe,
+                          dm_oblock_t oblock, dm_cblock_t cblock,
+                          uint32_t hint, bool hint_valid)
+{
+       int r;
+       struct policy *p = to_policy(pe);
+       struct wb_cache_entry *e = alloc_cache_entry(p);
+
+       if (e) {
+               e->cblock = cblock;
+               e->oblock = oblock;
+               e->dirty = false; /* blocks default to clean */
+               add_cache_entry(p, e);
+               r = 0;
+
+       } else
+               r = -ENOMEM;
+
+       return r;
+}
+
+static void wb_destroy(struct dm_cache_policy *pe)
+{
+       struct policy *p = to_policy(pe);
+
+       free_cache_blocks_and_hash(p);
+       kfree(p);
+}
+
+static struct wb_cache_entry *__wb_force_remove_mapping(struct policy *p, dm_oblock_t oblock)
+{
+       struct wb_cache_entry *r = lookup_cache_entry(p, oblock);
+
+       BUG_ON(!r);
+
+       remove_cache_hash_entry(r);
+       list_del(&r->list);
+
+       return r;
+}
+
+static void wb_remove_mapping(struct dm_cache_policy *pe, dm_oblock_t oblock)
+{
+       struct policy *p = to_policy(pe);
+       struct wb_cache_entry *e;
+       unsigned long flags;
+
+       spin_lock_irqsave(&p->lock, flags);
+       e = __wb_force_remove_mapping(p, oblock);
+       list_add_tail(&e->list, &p->free);
+       BUG_ON(!from_cblock(p->nr_cblocks_allocated));
+       p->nr_cblocks_allocated = to_cblock(from_cblock(p->nr_cblocks_allocated) - 1);
+       spin_unlock_irqrestore(&p->lock, flags);
+}
+
+static void wb_force_mapping(struct dm_cache_policy *pe,
+                               dm_oblock_t current_oblock, dm_oblock_t oblock)
+{
+       struct policy *p = to_policy(pe);
+       struct wb_cache_entry *e;
+       unsigned long flags;
+
+       spin_lock_irqsave(&p->lock, flags);
+       e = __wb_force_remove_mapping(p, current_oblock);
+       e->oblock = oblock;
+       add_cache_entry(p, e);
+       spin_unlock_irqrestore(&p->lock, flags);
+}
+
+static struct wb_cache_entry *get_next_dirty_entry(struct policy *p)
+{
+       struct list_head *l;
+       struct wb_cache_entry *r;
+
+       if (list_empty(&p->dirty))
+               return NULL;
+
+       l = list_pop(&p->dirty);
+       r = container_of(l, struct wb_cache_entry, list);
+       list_add(l, &p->clean_pending);
+
+       return r;
+}
+
+static int wb_writeback_work(struct dm_cache_policy *pe,
+                            dm_oblock_t *oblock,
+                            dm_cblock_t *cblock)
+{
+       int r = -ENOENT;
+       struct policy *p = to_policy(pe);
+       struct wb_cache_entry *e;
+       unsigned long flags;
+
+       spin_lock_irqsave(&p->lock, flags);
+
+       e = get_next_dirty_entry(p);
+       if (e) {
+               *oblock = e->oblock;
+               *cblock = e->cblock;
+               r = 0;
+       }
+
+       spin_unlock_irqrestore(&p->lock, flags);
+
+       return r;
+}
+
+static dm_cblock_t wb_residency(struct dm_cache_policy *pe)
+{
+       return to_policy(pe)->nr_cblocks_allocated;
+}
+
+/* Init the policy plugin interface function pointers. */
+static void init_policy_functions(struct policy *p)
+{
+       p->policy.destroy = wb_destroy;
+       p->policy.map = wb_map;
+       p->policy.lookup = wb_lookup;
+       p->policy.set_dirty = wb_set_dirty;
+       p->policy.clear_dirty = wb_clear_dirty;
+       p->policy.load_mapping = wb_load_mapping;
+       p->policy.walk_mappings = NULL;
+       p->policy.remove_mapping = wb_remove_mapping;
+       p->policy.writeback_work = wb_writeback_work;
+       p->policy.force_mapping = wb_force_mapping;
+       p->policy.residency = wb_residency;
+       p->policy.tick = NULL;
+}
+
+static struct dm_cache_policy *wb_create(dm_cblock_t cache_size,
+                                        sector_t origin_size,
+                                        sector_t cache_block_size)
+{
+       int r;
+       struct policy *p = kzalloc(sizeof(*p), GFP_KERNEL);
+
+       if (!p)
+               return NULL;
+
+       init_policy_functions(p);
+       INIT_LIST_HEAD(&p->free);
+       INIT_LIST_HEAD(&p->clean);
+       INIT_LIST_HEAD(&p->clean_pending);
+       INIT_LIST_HEAD(&p->dirty);
+
+       p->cache_size = cache_size;
+       spin_lock_init(&p->lock);
+
+       /* Allocate cache entry structs and add them to free list. */
+       r = alloc_cache_blocks_with_hash(p, cache_size);
+       if (!r)
+               return &p->policy;
+
+       kfree(p);
+
+       return NULL;
+}
+/*----------------------------------------------------------------------------*/
+
+static struct dm_cache_policy_type wb_policy_type = {
+       .name = "cleaner",
+       .hint_size = 0,
+       .owner = THIS_MODULE,
+       .create = wb_create
+};
+
+static int __init wb_init(void)
+{
+       int r = dm_cache_policy_register(&wb_policy_type);
+
+       if (r < 0)
+               DMERR("register failed %d", r);
+       else
+               DMINFO("version " CLEANER_VERSION " loaded");
+
+       return r;
+}
+
+static void __exit wb_exit(void)
+{
+       dm_cache_policy_unregister(&wb_policy_type);
+}
+
+module_init(wb_init);
+module_exit(wb_exit);
+
+MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("cleaner cache policy");