All of lore.kernel.org
 help / color / mirror / Atom feed
From: Mike Snitzer <snitzer@redhat.com>
To: Alasdair G Kergon <agk@redhat.com>
Cc: device-mapper development <dm-devel@redhat.com>,
	Joe Thornber <ejt@redhat.com>
Subject: [PATCH] dm thin: fix pool target flags that control discard
Date: Fri, 23 Mar 2012 17:55:02 -0400	[thread overview]
Message-ID: <20120323215502.GB11054@redhat.com> (raw)
In-Reply-To: <20120323123747.GB16214@agk-dp.fab.redhat.com>

On Fri, Mar 23 2012 at  8:37am -0400,
Alasdair G Kergon <agk@redhat.com> wrote:

> On Fri, Mar 16, 2012 at 03:22:34PM +0000, Joe Thornber wrote:
> > +      'ignore_discard': disable discard support
> > +      'no_discard_passdown': don't pass discards down to the underlying data device
> 
> If someone reloads the pool target changing either of these options, how do connected
> thin targets pick up the change?  If they don't pick it up automatically, how do you
> determine whether they did or didn't pick it up - is that internal state not
> exposed to userspace yet?

Here are various fixes for dm_thin-add-pool-target-flags-to-control-discard.patch

o wire up 'discard_passdown' so that it does control whether or not the
  discard is passed to the pool's underlying data device
o disallow disabling discards if a pool was already configured with
  discards enabled (the default)
  - justification is inlined in the code
  - required pool_ctr knowing whether pool was created or not; so added
    'created' flag to __pool_find()
o if 'discard_passdown' is enabled (the default) verify that the pool's
  data device supports discards; if not warn and disable discard_passdown
o the pool device's discard support should _not_ be limited by whether
  'discard_passdown' is enabled or not.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
---
 drivers/md/dm-thin.c |   41 ++++++++++++++++++++++++++++++++++-------
 1 file changed, 34 insertions(+), 7 deletions(-)

Index: linux/drivers/md/dm-thin.c
===================================================================
--- linux.orig/drivers/md/dm-thin.c
+++ linux/drivers/md/dm-thin.c
@@ -1209,7 +1209,7 @@ static void process_discard(struct thin_
 			 */
 			m = get_next_mapping(pool);
 			m->tc = tc;
-			m->pass_discard = !lookup_result.shared;
+			m->pass_discard = (!lookup_result.shared) & pool->pf.discard_passdown;
 			m->virt_block = block;
 			m->data_block = lookup_result.block;
 			m->cell = cell;
@@ -1790,7 +1790,8 @@ static void __pool_dec(struct pool *pool
 
 static struct pool *__pool_find(struct mapped_device *pool_md,
 				struct block_device *metadata_dev,
-				unsigned long block_size, char **error)
+				unsigned long block_size, char **error,
+				int *created)
 {
 	struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
 
@@ -1806,8 +1807,10 @@ static struct pool *__pool_find(struct m
 				return ERR_PTR(-EINVAL);
 			__pool_inc(pool);
 
-		} else
+		} else {
 			pool = pool_create(pool_md, metadata_dev, block_size, error);
+			*created = 1;
+		}
 	}
 
 	return pool;
@@ -1887,7 +1890,7 @@ static int parse_pool_features(struct dm
  */
 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
 {
-	int r;
+	int r, pool_created = 0;
 	struct pool_c *pt;
 	struct pool *pool;
 	struct pool_features pf;
@@ -1961,12 +1964,36 @@ static int pool_ctr(struct dm_target *ti
 	}
 
 	pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
-			   block_size, &ti->error);
+			   block_size, &ti->error, &pool_created);
 	if (IS_ERR(pool)) {
 		r = PTR_ERR(pool);
 		goto out_free_pt;
 	}
 
+	/*
+	 * 'pool_created' reflects whether this is the first table load.
+	 * Discard support is not allowed to be disabled after initial load.
+	 * Disabling it would require a pool reload to trigger thin device
+	 * changes (e.g. ti->discards_supported and QUEUE_FLAG_DISCARD).
+	 */
+	if (!pool_created && !pf.discard_enabled && pool->pf.discard_enabled) {
+		ti->error = "Discard support cannot be disabled once enabled";
+		r = -EINVAL;
+		goto out_free_pt;
+	}
+
+	/*
+	 * If discard_passdown was enabled verify that the data device
+	 * supports discards.  Disable discard_passdown if not.
+	 */
+	if (pf.discard_passdown) {
+		struct request_queue *q = bdev_get_queue(data_dev->bdev);
+		if (!q || !blk_queue_discard(q)) {
+			DMWARN("Discard unsupported by data device, disabling discard passdown");
+			pf.discard_passdown = 0;
+		}
+	}
+
 	pt->pool = pool;
 	pt->ti = ti;
 	pt->metadata_dev = metadata_dev;
@@ -1974,9 +2001,9 @@ static int pool_ctr(struct dm_target *ti
 	pt->low_water_blocks = low_water_blocks;
 	pt->pf = pf;
 	ti->num_flush_requests = 1;
-	if (pf.discard_enabled && pf.discard_passdown) {
-		ti->discards_supported = 1;
+	if (pf.discard_enabled) {
 		ti->num_discard_requests = 1;
+		ti->discards_supported = 1;
 	}
 	ti->private = pt;
 

  reply	other threads:[~2012-03-23 21:55 UTC|newest]

Thread overview: 26+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-03-16 15:21 dm-thin patches for 3.4 Joe Thornber
2012-03-16 15:22 ` [PATCH 01/14] dm-thin: don't use the bi_next field for the holder of a cell Joe Thornber
2012-03-16 15:22   ` [PATCH 02/14] dm-thin: remove some documentation for the unimplemented 'trim' target message Joe Thornber
2012-03-16 15:22   ` [PATCH 03/14] dm_thin: dm_sm_root_size() was being called for the wrong space-map Joe Thornber
2012-03-16 15:22   ` [PATCH 04/14] dm_thin: tweak a comment Joe Thornber
2012-03-16 15:22   ` [PATCH 05/14] dm_btree: remove redundant arg from value_ptr() Joe Thornber
2012-03-16 15:22   ` [PATCH 06/14] dm_btree: fix rebalancing of 3 nodes after remove Joe Thornber
2012-03-16 15:22   ` [PATCH 07/14] dm_space_map: remove entries from the ref_count tree if they're no longer needed Joe Thornber
2012-03-16 15:22   ` [PATCH 08/14] dm_thin: add support for read-only external snapshot origins Joe Thornber
2012-03-16 15:22   ` [PATCH 09/14] dm_thin: foundation for discard support Joe Thornber
2012-03-16 15:22   ` [PATCH 10/14] dm_thin: add support for REQ_DISCARD Joe Thornber
2012-03-23 12:45     ` Alasdair G Kergon
2012-03-16 15:22   ` [PATCH 11/14] dm_thin: add pool target flags to control discard Joe Thornber
2012-03-23 12:37     ` Alasdair G Kergon
2012-03-23 21:55       ` Mike Snitzer [this message]
2012-03-26 14:15         ` [PATCH] dm thin: fix pool target flags that " Joe Thornber
2012-03-26 15:33           ` Mike Snitzer
2012-03-26 19:56             ` Mike Snitzer
2012-03-26 15:34           ` [PATCH] " Joe Thornber
2012-03-26 15:46           ` Joe Thornber
2012-03-16 15:22   ` [PATCH 12/14] dm_thin: commit metadata every second Joe Thornber
2012-03-16 15:22   ` [PATCH 13/14] dm_thin: commit just before processing a pool target info request Joe Thornber
2012-03-19 14:00     ` Alasdair G Kergon
2012-03-20 10:12       ` Joe Thornber
2012-03-16 15:22   ` [PATCH 14/14] dm_thin: bump the target versions Joe Thornber
2012-03-20 18:24   ` [PATCH 01/14] dm-thin: don't use the bi_next field for the holder of a cell Alasdair G Kergon

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20120323215502.GB11054@redhat.com \
    --to=snitzer@redhat.com \
    --cc=agk@redhat.com \
    --cc=dm-devel@redhat.com \
    --cc=ejt@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.