dm cache: add support for discard passdown to the origin device

DM cache now defaults to passing discards down to the origin device.
User may disable this using the "no_discard_passdown" feature when
creating the cache device.

If the cache's underlying origin device doesn't support discards then
passdown is disabled (with warning).  Similarly, if the underlying
origin device's max_discard_sectors is less than a cache block discard
passdown will be disabled (this is required because sizing of the cache
internal discard bitset depends on it).

Signed-off-by: Mike Snitzer <[email protected]>

diff --git a/Documentation/device-mapper/cache.txt b/Documentation/device-mapper/cache.txt
index ff0841711fd524063e2acf0fa4644ec2fbb4b764..8ae1cf8e94daeb6cf142e24624d0e5f4d99d728b 100644 (file)
--- a/Documentation/device-mapper/cache.txt
+++ b/Documentation/device-mapper/cache.txt
@@ -206,6 +206,9 @@ Optional feature arguments are:
                   in a separate btree, which improves speed of shutting
                  down the cache.
 
                   in a separate btree, which improves speed of shutting
                  down the cache.
 

+   no_discard_passdown : disable passing down discards from the cache
+                         to the origin's data device.
+

 A policy called 'default' is always registered.  This is an alias for
 the policy we currently think is giving best all round performance.
 
 A policy called 'default' is always registered.  This is an alias for
 the policy we currently think is giving best all round performance.
 

diff --git a/drivers/md/dm-cache-target.c b/drivers/md/dm-cache-target.c
index adc529f12b6b830bb192503044ef1bf5f2d8d729..d249cf8ac277e4246d8f398c8aa3b2e46fe90724 100644 (file)
--- a/drivers/md/dm-cache-target.c
+++ b/drivers/md/dm-cache-target.c
@@ -353,6 +353,7 @@ struct cache_features {
        enum cache_metadata_mode mode;
        enum cache_io_mode io_mode;
        unsigned metadata_version;
        enum cache_metadata_mode mode;
        enum cache_io_mode io_mode;
        unsigned metadata_version;

+       bool discard_passdown:1;

 };
 
 struct cache_stats {
 };
 
 struct cache_stats {


@@ -1899,7 +1900,11 @@ static bool process_discard_bio(struct cache *cache, struct bio *bio)
                b = to_dblock(from_dblock(b) + 1);
        }
 
                b = to_dblock(from_dblock(b) + 1);
        }
 

-       bio_endio(bio);
+       if (cache->features.discard_passdown) {
+               remap_to_origin(cache, bio);
+               generic_make_request(bio);
+       } else
+               bio_endio(bio);

 
        return false;
 }
 
        return false;
 }


@@ -2233,13 +2238,14 @@ static void init_features(struct cache_features *cf)
        cf->mode = CM_WRITE;
        cf->io_mode = CM_IO_WRITEBACK;
        cf->metadata_version = 1;
        cf->mode = CM_WRITE;
        cf->io_mode = CM_IO_WRITEBACK;
        cf->metadata_version = 1;

+       cf->discard_passdown = true;

 }
 
 static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
                          char **error)
 {
        static const struct dm_arg _args[] = {
 }
 
 static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
                          char **error)
 {
        static const struct dm_arg _args[] = {

-               {0, 2, "Invalid number of cache feature arguments"},
+               {0, 3, "Invalid number of cache feature arguments"},

        };
 
        int r, mode_ctr = 0;
        };
 
        int r, mode_ctr = 0;


@@ -2274,6 +2280,9 @@ static int parse_features(struct cache_args *ca, struct dm_arg_set *as,
                else if (!strcasecmp(arg, "metadata2"))
                        cf->metadata_version = 2;
 
                else if (!strcasecmp(arg, "metadata2"))
                        cf->metadata_version = 2;
 

+               else if (!strcasecmp(arg, "no_discard_passdown"))
+                       cf->discard_passdown = false;
+

                else {
                        *error = "Unrecognised cache feature requested";
                        return -EINVAL;
                else {
                        *error = "Unrecognised cache feature requested";
                        return -EINVAL;


@@ -3119,6 +3128,39 @@ static void cache_resume(struct dm_target *ti)
        do_waker(&cache->waker.work);
 }
 
        do_waker(&cache->waker.work);
 }
 

+static void emit_flags(struct cache *cache, char *result,
+                      unsigned maxlen, ssize_t *sz_ptr)
+{
+       ssize_t sz = *sz_ptr;
+       struct cache_features *cf = &cache->features;
+       unsigned count = (cf->metadata_version == 2) + !cf->discard_passdown + 1;
+
+       DMEMIT("%u ", count);
+
+       if (cf->metadata_version == 2)
+               DMEMIT("metadata2 ");
+
+       if (writethrough_mode(cache))
+               DMEMIT("writethrough ");
+
+       else if (passthrough_mode(cache))
+               DMEMIT("passthrough ");
+
+       else if (writeback_mode(cache))
+               DMEMIT("writeback ");
+
+       else {
+               DMEMIT("unknown ");
+               DMERR("%s: internal error: unknown io mode: %d",
+                     cache_device_name(cache), (int) cf->io_mode);
+       }
+
+       if (!cf->discard_passdown)
+               DMEMIT("no_discard_passdown ");
+
+       *sz_ptr = sz;
+}
+

 /*
  * Status format:
  *
 /*
  * Status format:
  *


@@ -3185,25 +3227,7 @@ static void cache_status(struct dm_target *ti, status_type_t type,
                       (unsigned) atomic_read(&cache->stats.promotion),
                       (unsigned long) atomic_read(&cache->nr_dirty));
 
                       (unsigned) atomic_read(&cache->stats.promotion),
                       (unsigned long) atomic_read(&cache->nr_dirty));
 

-               if (cache->features.metadata_version == 2)
-                       DMEMIT("2 metadata2 ");
-               else
-                       DMEMIT("1 ");
-
-               if (writethrough_mode(cache))
-                       DMEMIT("writethrough ");
-
-               else if (passthrough_mode(cache))
-                       DMEMIT("passthrough ");
-
-               else if (writeback_mode(cache))
-                       DMEMIT("writeback ");
-
-               else {
-                       DMERR("%s: internal error: unknown io mode: %d",
-                             cache_device_name(cache), (int) cache->features.io_mode);
-                       goto err;
-               }
+               emit_flags(cache, result, maxlen, &sz);

 
                DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
 
 
                DMEMIT("2 migration_threshold %llu ", (unsigned long long) cache->migration_threshold);
 


@@ -3432,14 +3456,62 @@ static int cache_iterate_devices(struct dm_target *ti,
        return r;
 }
 
        return r;
 }
 

+static bool origin_dev_supports_discard(struct block_device *origin_bdev)
+{
+       struct request_queue *q = bdev_get_queue(origin_bdev);
+
+       return q && blk_queue_discard(q);
+}
+
+/*
+ * If discard_passdown was enabled verify that the origin device
+ * supports discards.  Disable discard_passdown if not.
+ */
+static void disable_passdown_if_not_supported(struct cache *cache)
+{
+       struct block_device *origin_bdev = cache->origin_dev->bdev;
+       struct queue_limits *origin_limits = &bdev_get_queue(origin_bdev)->limits;
+       const char *reason = NULL;
+       char buf[BDEVNAME_SIZE];
+
+       if (!cache->features.discard_passdown)
+               return;
+
+       if (!origin_dev_supports_discard(origin_bdev))
+               reason = "discard unsupported";
+
+       else if (origin_limits->max_discard_sectors < cache->sectors_per_block)
+               reason = "max discard sectors smaller than a block";
+
+       if (reason) {
+               DMWARN("Origin device (%s) %s: Disabling discard passdown.",
+                      bdevname(origin_bdev, buf), reason);
+               cache->features.discard_passdown = false;
+       }
+}
+

 static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
 {
 static void set_discard_limits(struct cache *cache, struct queue_limits *limits)
 {

+       struct block_device *origin_bdev = cache->origin_dev->bdev;
+       struct queue_limits *origin_limits = &bdev_get_queue(origin_bdev)->limits;
+
+       if (!cache->features.discard_passdown) {
+               /* No passdown is done so setting own virtual limits */
+               limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024,
+                                                   cache->origin_sectors);
+               limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
+               return;
+       }
+

        /*
        /*

-        * FIXME: these limits may be incompatible with the cache device
+        * cache_iterate_devices() is stacking both origin and fast device limits
+        * but discards aren't passed to fast device, so inherit origin's limits.

         */
         */

-       limits->max_discard_sectors = min_t(sector_t, cache->discard_block_size * 1024,
-                                           cache->origin_sectors);
-       limits->discard_granularity = cache->discard_block_size << SECTOR_SHIFT;
+       limits->max_discard_sectors = origin_limits->max_discard_sectors;
+       limits->max_hw_discard_sectors = origin_limits->max_hw_discard_sectors;
+       limits->discard_granularity = origin_limits->discard_granularity;
+       limits->discard_alignment = origin_limits->discard_alignment;
+       limits->discard_misaligned = origin_limits->discard_misaligned;

 }
 
 static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
 }
 
 static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)


@@ -3456,6 +3528,8 @@ static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
                blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT);
                blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
        }
                blk_limits_io_min(limits, cache->sectors_per_block << SECTOR_SHIFT);
                blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
        }

+
+       disable_passdown_if_not_supported(cache);

        set_discard_limits(cache, limits);
 }
 
        set_discard_limits(cache, limits);
 }
 


@@ -3463,7 +3537,7 @@ static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
 
 static struct target_type cache_target = {
        .name = "cache",
 
 static struct target_type cache_target = {
        .name = "cache",

-       .version = {2, 0, 0},
+       .version = {2, 1, 0},

        .module = THIS_MODULE,
        .ctr = cache_ctr,
        .dtr = cache_dtr,
        .module = THIS_MODULE,
        .ctr = cache_ctr,
        .dtr = cache_dtr,