dm raid: add region_size parameter
authorJonathan Brassow <jbrassow@redhat.com>
Tue, 2 Aug 2011 11:32:07 +0000 (12:32 +0100)
committerAlasdair G Kergon <agk@redhat.com>
Tue, 2 Aug 2011 11:32:07 +0000 (12:32 +0100)
Allow the user to specify the region_size.

Ensures that the supplied value meets md's constraints, viz. the number of
regions does not exceed 2^21.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Documentation/device-mapper/dm-raid.txt
drivers/md/dm-raid.c

index 4f9dd3cecc114d450724b512e2cee681aedf84a1..be4419a307816b5a21e92454948c1c7978cf5e24 100644 (file)
@@ -52,6 +52,10 @@ The target is named "raid" and it accepts the following parameters:
        [max_recovery_rate <kB/sec/disk>]  Throttle RAID initialization
        [max_write_behind <sectors>]       See '-write-behind=' (man mdadm)
        [stripe_cache <sectors>]           Stripe cache size (higher RAIDs only)
+       [region_size <sectors>]
+               The region_size multiplied by the number of regions is the
+               logical size of the array.  The bitmap records the device
+               synchronisation state for each region.
 
 <#raid_devs>: The number of devices composing the array.
        Each device consists of two entries.  The first is the device
index d4e95b2e39f60b05d64d4d5ccc2685f0cb6dd1ae..a8a1915a450d204a9e1b9400b21d39c8547a56e4 100644 (file)
@@ -52,7 +52,7 @@ struct raid_dev {
 #define DMPF_MAX_RECOVERY_RATE 0x20
 #define DMPF_MAX_WRITE_BEHIND  0x40
 #define DMPF_STRIPE_CACHE      0x80
-
+#define DMPF_REGION_SIZE       0X100
 struct raid_set {
        struct dm_target *ti;
 
@@ -236,6 +236,67 @@ static int dev_parms(struct raid_set *rs, char **argv)
        return 0;
 }
 
+/*
+ * validate_region_size
+ * @rs
+ * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
+ *
+ * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
+ * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
+ *
+ * Returns: 0 on success, -EINVAL on failure.
+ */
+static int validate_region_size(struct raid_set *rs, unsigned long region_size)
+{
+       unsigned long min_region_size = rs->ti->len / (1 << 21);
+
+       if (!region_size) {
+               /*
+                * Choose a reasonable default.  All figures in sectors.
+                */
+               if (min_region_size > (1 << 13)) {
+                       DMINFO("Choosing default region size of %lu sectors",
+                              region_size);
+                       region_size = min_region_size;
+               } else {
+                       DMINFO("Choosing default region size of 4MiB");
+                       region_size = 1 << 13; /* sectors */
+               }
+       } else {
+               /*
+                * Validate user-supplied value.
+                */
+               if (region_size > rs->ti->len) {
+                       rs->ti->error = "Supplied region size is too large";
+                       return -EINVAL;
+               }
+
+               if (region_size < min_region_size) {
+                       DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
+                             region_size, min_region_size);
+                       rs->ti->error = "Supplied region size is too small";
+                       return -EINVAL;
+               }
+
+               if (!is_power_of_2(region_size)) {
+                       rs->ti->error = "Region size is not a power of 2";
+                       return -EINVAL;
+               }
+
+               if (region_size < rs->md.chunk_sectors) {
+                       rs->ti->error = "Region size is smaller than the chunk size";
+                       return -EINVAL;
+               }
+       }
+
+       /*
+        * Convert sectors to bytes.
+        */
+       rs->md.bitmap_info.chunksize = (region_size << 9);
+
+       return 0;
+}
+
 /*
  * Possible arguments are...
  * RAID456:
@@ -249,12 +310,13 @@ static int dev_parms(struct raid_set *rs, char **argv)
  *    [max_recovery_rate <kB/sec/disk>]        Throttle RAID initialization
  *    [max_write_behind <sectors>]     See '-write-behind=' (man mdadm)
  *    [stripe_cache <sectors>]         Stripe cache size for higher RAIDs
+ *    [region_size <sectors>]           Defines granularity of bitmap
  */
 static int parse_raid_params(struct raid_set *rs, char **argv,
                             unsigned num_raid_params)
 {
        unsigned i, rebuild_cnt = 0;
-       unsigned long value;
+       unsigned long value, region_size = 0;
        char *key;
 
        /*
@@ -365,6 +427,9 @@ static int parse_raid_params(struct raid_set *rs, char **argv,
                                return -EINVAL;
                        }
                        rs->md.sync_speed_max = (int)value;
+               } else if (!strcasecmp(key, "region_size")) {
+                       rs->print_flags |= DMPF_REGION_SIZE;
+                       region_size = value;
                } else {
                        DMERR("Unable to parse RAID parameter: %s", key);
                        rs->ti->error = "Unable to parse RAID parameters";
@@ -372,6 +437,14 @@ static int parse_raid_params(struct raid_set *rs, char **argv,
                }
        }
 
+       if (validate_region_size(rs, region_size))
+               return -EINVAL;
+
+       if (rs->md.chunk_sectors)
+               rs->ti->split_io = rs->md.chunk_sectors;
+       else
+               rs->ti->split_io = region_size;
+
        /* Assume there are no metadata devices until the drives are parsed */
        rs->md.persistent = 0;
        rs->md.external = 1;
@@ -469,7 +542,6 @@ static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
                goto bad;
 
        INIT_WORK(&rs->md.event_work, do_table_event);
-       ti->split_io = rs->md.chunk_sectors;
        ti->private = rs;
 
        mutex_lock(&rs->md.reconfig_mutex);
@@ -596,6 +668,10 @@ static int raid_status(struct dm_target *ti, status_type_t type,
                               conf ? conf->max_nr_stripes * 2 : 0);
                }
 
+               if (rs->print_flags & DMPF_REGION_SIZE)
+                       DMEMIT(" region_size %lu",
+                              rs->md.bitmap_info.chunksize >> 9);
+
                DMEMIT(" %d", rs->md.raid_disks);
                for (i = 0; i < rs->md.raid_disks; i++) {
                        DMEMIT(" -"); /* metadata device */