[PATCH 0/5] libceph: crush tunables5

[PATCH 0/5] libceph: crush tunables5

[Ilya Dryomov]

Hello,

This series adds support for the new chooseleaf_stable CRUSH tunable
and enables TUNABLES5 feature bit.  It's likely that TUNABLES5 bit will
be shared and cover request_redirect_t encoding in MOSDOpReply and
possibly new file layouts but those aren't in master yet.

Thanks,

                Ilya


Ilya Dryomov (5):
  crush: ensure bucket id is valid before indexing buckets array
  crush: ensure take bucket value is valid
  crush: add chooseleaf_stable tunable
  crush: decode and initialize chooseleaf_stable
  libceph: advertise support for TUNABLES5

 include/linux/ceph/ceph_features.h | 13 ++++++++++++-
 include/linux/crush/crush.h        |  8 +++++++-
 net/ceph/crush/mapper.c            | 33 ++++++++++++++++++++++++++-------
 net/ceph/osdmap.c                  | 19 ++++++++++++++-----
 4 files changed, 59 insertions(+), 14 deletions(-)

-- 
2.4.3

[PATCH 1/5] crush: ensure bucket id is valid before indexing buckets array

[Ilya Dryomov]

We were indexing the buckets array without verifying the index was
within the [0,max_buckets) range.  This could happen because
a multistep rule does not have enough buckets and has CRUSH_ITEM_NONE
for an intermediate result, which would feed in CRUSH_ITEM_NONE and
make us crash.

Reflects ceph.git commit 976a24a326da8931e689ee22fce35feab5b67b76.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
---
 net/ceph/crush/mapper.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/net/ceph/crush/mapper.c b/net/ceph/crush/mapper.c
index 393bfb22d5bb..97ecf6f262aa 100644
--- a/net/ceph/crush/mapper.c
+++ b/net/ceph/crush/mapper.c
@@ -888,6 +888,7 @@ int crush_do_rule(const struct crush_map *map,
 			osize = 0;
 
 			for (i = 0; i < wsize; i++) {
+				int bno;
 				/*
 				 * see CRUSH_N, CRUSH_N_MINUS macros.
 				 * basically, numrep <= 0 means relative to
@@ -900,6 +901,13 @@ int crush_do_rule(const struct crush_map *map,
 						continue;
 				}
 				j = 0;
+				/* make sure bucket id is valid */
+				bno = -1 - w[i];
+				if (bno < 0 || bno >= map->max_buckets) {
+					/* w[i] is probably CRUSH_ITEM_NONE */
+					dprintk("  bad w[i] %d\n", w[i]);
+					continue;
+				}
 				if (firstn) {
 					int recurse_tries;
 					if (choose_leaf_tries)
@@ -911,7 +919,7 @@ int crush_do_rule(const struct crush_map *map,
 						recurse_tries = choose_tries;
 					osize += crush_choose_firstn(
 						map,
-						map->buckets[-1-w[i]],
+						map->buckets[bno],
 						weight, weight_max,
 						x, numrep,
 						curstep->arg2,
@@ -930,7 +938,7 @@ int crush_do_rule(const struct crush_map *map,
 						    numrep : (result_max-osize));
 					crush_choose_indep(
 						map,
-						map->buckets[-1-w[i]],
+						map->buckets[bno],
 						weight, weight_max,
 						x, out_size, numrep,
 						curstep->arg2,
-- 
2.4.3

[PATCH 2/5] crush: ensure take bucket value is valid

[Ilya Dryomov]

Ensure that the take argument is a valid bucket ID before indexing the
buckets array.

Reflects ceph.git commit 93ec538e8a667699876b72459b8ad78966d89c61.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
---
 net/ceph/crush/mapper.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/ceph/crush/mapper.c b/net/ceph/crush/mapper.c
index 97ecf6f262aa..abb700621e4a 100644
--- a/net/ceph/crush/mapper.c
+++ b/net/ceph/crush/mapper.c
@@ -835,7 +835,8 @@ int crush_do_rule(const struct crush_map *map,
 		case CRUSH_RULE_TAKE:
 			if ((curstep->arg1 >= 0 &&
 			     curstep->arg1 < map->max_devices) ||
-			    (-1-curstep->arg1 < map->max_buckets &&
+			    (-1-curstep->arg1 >= 0 &&
+			     -1-curstep->arg1 < map->max_buckets &&
 			     map->buckets[-1-curstep->arg1])) {
 				w[0] = curstep->arg1;
 				wsize = 1;
-- 
2.4.3

[PATCH 3/5] crush: add chooseleaf_stable tunable

[Ilya Dryomov]

Add a tunable to fix the bug that chooseleaf may cause unnecessary pg
migrations when some device fails.

Reflects ceph.git commit fdb3f664448e80d984470f32f04e2e6f03ab52ec.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
---
 include/linux/crush/crush.h |  8 +++++++-
 net/ceph/crush/mapper.c     | 18 ++++++++++++++----
 2 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/include/linux/crush/crush.h b/include/linux/crush/crush.h
index 48b49305716b..be8f12b8f195 100644
--- a/include/linux/crush/crush.h
+++ b/include/linux/crush/crush.h
@@ -59,7 +59,8 @@ enum {
 	CRUSH_RULE_SET_CHOOSELEAF_TRIES = 9, /* override chooseleaf_descend_once */
 	CRUSH_RULE_SET_CHOOSE_LOCAL_TRIES = 10,
 	CRUSH_RULE_SET_CHOOSE_LOCAL_FALLBACK_TRIES = 11,
-	CRUSH_RULE_SET_CHOOSELEAF_VARY_R = 12
+	CRUSH_RULE_SET_CHOOSELEAF_VARY_R = 12,
+	CRUSH_RULE_SET_CHOOSELEAF_STABLE = 13
 };
 
 /*
@@ -205,6 +206,11 @@ struct crush_map {
 	 * mappings line up a bit better with previous mappings. */
 	__u8 chooseleaf_vary_r;
 
+	/* if true, it makes chooseleaf firstn to return stable results (if
+	 * no local retry) so that data migrations would be optimal when some
+	 * device fails. */
+	__u8 chooseleaf_stable;
+
 #ifndef __KERNEL__
 	/*
 	 * version 0 (original) of straw_calc has various flaws.  version 1
diff --git a/net/ceph/crush/mapper.c b/net/ceph/crush/mapper.c
index abb700621e4a..5fcfb98f309e 100644
--- a/net/ceph/crush/mapper.c
+++ b/net/ceph/crush/mapper.c
@@ -403,6 +403,7 @@ static int is_out(const struct crush_map *map,
  * @local_retries: localized retries
  * @local_fallback_retries: localized fallback retries
  * @recurse_to_leaf: true if we want one device under each item of given type (chooseleaf instead of choose)
+ * @stable: stable mode starts rep=0 in the recursive call for all replicas
  * @vary_r: pass r to recursive calls
  * @out2: second output vector for leaf items (if @recurse_to_leaf)
  * @parent_r: r value passed from the parent
@@ -419,6 +420,7 @@ static int crush_choose_firstn(const struct crush_map *map,
 			       unsigned int local_fallback_retries,
 			       int recurse_to_leaf,
 			       unsigned int vary_r,
+			       unsigned int stable,
 			       int *out2,
 			       int parent_r)
 {
@@ -433,13 +435,13 @@ static int crush_choose_firstn(const struct crush_map *map,
 	int collide, reject;
 	int count = out_size;
 
-	dprintk("CHOOSE%s bucket %d x %d outpos %d numrep %d tries %d recurse_tries %d local_retries %d local_fallback_retries %d parent_r %d\n",
+	dprintk("CHOOSE%s bucket %d x %d outpos %d numrep %d tries %d recurse_tries %d local_retries %d local_fallback_retries %d parent_r %d stable %d\n",
 		recurse_to_leaf ? "_LEAF" : "",
 		bucket->id, x, outpos, numrep,
 		tries, recurse_tries, local_retries, local_fallback_retries,
-		parent_r);
+		parent_r, stable);
 
-	for (rep = outpos; rep < numrep && count > 0 ; rep++) {
+	for (rep = stable ? 0 : outpos; rep < numrep && count > 0 ; rep++) {
 		/* keep trying until we get a non-out, non-colliding item */
 		ftotal = 0;
 		skip_rep = 0;
@@ -512,13 +514,14 @@ static int crush_choose_firstn(const struct crush_map *map,
 						if (crush_choose_firstn(map,
 							 map->buckets[-1-item],
 							 weight, weight_max,
-							 x, outpos+1, 0,
+							 x, stable ? 1 : outpos+1, 0,
 							 out2, outpos, count,
 							 recurse_tries, 0,
 							 local_retries,
 							 local_fallback_retries,
 							 0,
 							 vary_r,
+							 stable,
 							 NULL,
 							 sub_r) <= outpos)
 							/* didn't get leaf */
@@ -816,6 +819,7 @@ int crush_do_rule(const struct crush_map *map,
 	int choose_local_fallback_retries = map->choose_local_fallback_tries;
 
 	int vary_r = map->chooseleaf_vary_r;
+	int stable = map->chooseleaf_stable;
 
 	if ((__u32)ruleno >= map->max_rules) {
 		dprintk(" bad ruleno %d\n", ruleno);
@@ -870,6 +874,11 @@ int crush_do_rule(const struct crush_map *map,
 				vary_r = curstep->arg1;
 			break;
 
+		case CRUSH_RULE_SET_CHOOSELEAF_STABLE:
+			if (curstep->arg1 >= 0)
+				stable = curstep->arg1;
+			break;
+
 		case CRUSH_RULE_CHOOSELEAF_FIRSTN:
 		case CRUSH_RULE_CHOOSE_FIRSTN:
 			firstn = 1;
@@ -932,6 +941,7 @@ int crush_do_rule(const struct crush_map *map,
 						choose_local_fallback_retries,
 						recurse_to_leaf,
 						vary_r,
+						stable,
 						c+osize,
 						0);
 				} else {
-- 
2.4.3

[PATCH 4/5] crush: decode and initialize chooseleaf_stable

[Ilya Dryomov]

Also add missing \n while at it.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
---
 net/ceph/osdmap.c | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/net/ceph/osdmap.c b/net/ceph/osdmap.c
index 7d8f581d9f1f..243574c8cf33 100644
--- a/net/ceph/osdmap.c
+++ b/net/ceph/osdmap.c
@@ -342,23 +342,32 @@ static struct crush_map *crush_decode(void *pbyval, void *end)
         c->choose_local_tries = ceph_decode_32(p);
         c->choose_local_fallback_tries =  ceph_decode_32(p);
         c->choose_total_tries = ceph_decode_32(p);
-        dout("crush decode tunable choose_local_tries = %d",
+        dout("crush decode tunable choose_local_tries = %d\n",
              c->choose_local_tries);
-        dout("crush decode tunable choose_local_fallback_tries = %d",
+        dout("crush decode tunable choose_local_fallback_tries = %d\n",
              c->choose_local_fallback_tries);
-        dout("crush decode tunable choose_total_tries = %d",
+        dout("crush decode tunable choose_total_tries = %d\n",
              c->choose_total_tries);
 
 	ceph_decode_need(p, end, sizeof(u32), done);
 	c->chooseleaf_descend_once = ceph_decode_32(p);
-	dout("crush decode tunable chooseleaf_descend_once = %d",
+	dout("crush decode tunable chooseleaf_descend_once = %d\n",
 	     c->chooseleaf_descend_once);
 
 	ceph_decode_need(p, end, sizeof(u8), done);
 	c->chooseleaf_vary_r = ceph_decode_8(p);
-	dout("crush decode tunable chooseleaf_vary_r = %d",
+	dout("crush decode tunable chooseleaf_vary_r = %d\n",
 	     c->chooseleaf_vary_r);
 
+	/* skip straw_calc_version, allowed_bucket_algs */
+	ceph_decode_need(p, end, sizeof(u8) + sizeof(u32), done);
+	*p += sizeof(u8) + sizeof(u32);
+
+	ceph_decode_need(p, end, sizeof(u8), done);
+	c->chooseleaf_stable = ceph_decode_8(p);
+	dout("crush decode tunable chooseleaf_stable = %d\n",
+	     c->chooseleaf_stable);
+
 done:
 	dout("crush_decode success\n");
 	return c;
-- 
2.4.3

[PATCH 5/5] libceph: advertise support for TUNABLES5

[Ilya Dryomov]

Add TUNABLES5 feature (chooseleaf_stable tunable) to a set of features
supported by default.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
---
 include/linux/ceph/ceph_features.h | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/include/linux/ceph/ceph_features.h b/include/linux/ceph/ceph_features.h
index fa3f82aae3be..6b4b059c158f 100644
--- a/include/linux/ceph/ceph_features.h
+++ b/include/linux/ceph/ceph_features.h
@@ -63,6 +63,16 @@
 #define CEPH_FEATURE_OSD_MIN_SIZE_RECOVERY (1ULL<<49)
 // duplicated since it was introduced at the same time as MIN_SIZE_RECOVERY
 #define CEPH_FEATURE_OSD_PROXY_FEATURES (1ULL<<49)  /* overlap w/ above */
+#define CEPH_FEATURE_MON_METADATA (1ULL<<50)
+#define CEPH_FEATURE_OSD_BITWISE_HOBJ_SORT (1ULL<<51) /* can sort objs bitwise */
+#define CEPH_FEATURE_OSD_PROXY_WRITE_FEATURES (1ULL<<52)
+#define CEPH_FEATURE_ERASURE_CODE_PLUGINS_V3 (1ULL<<53)
+#define CEPH_FEATURE_OSD_HITSET_GMT (1ULL<<54)
+#define CEPH_FEATURE_HAMMER_0_94_4 (1ULL<<55)
+#define CEPH_FEATURE_NEW_OSDOP_ENCODING   (1ULL<<56) /* New, v7 encoding */
+#define CEPH_FEATURE_MON_STATEFUL_SUB (1ULL<<57) /* stateful mon subscription */
+#define CEPH_FEATURE_MON_ROUTE_OSDMAP (1ULL<<57) /* peon sends osdmaps */
+#define CEPH_FEATURE_CRUSH_TUNABLES5	(1ULL<<58) /* chooseleaf stable mode */
 
 /*
  * The introduction of CEPH_FEATURE_OSD_SNAPMAPPER caused the feature
@@ -109,7 +119,8 @@ static inline u64 ceph_sanitize_features(u64 features)
 	 CEPH_FEATURE_CRUSH_TUNABLES3 |		\
 	 CEPH_FEATURE_OSD_PRIMARY_AFFINITY |	\
 	 CEPH_FEATURE_MSGR_KEEPALIVE2 |		\
-	 CEPH_FEATURE_CRUSH_V4)
+	 CEPH_FEATURE_CRUSH_V4 |		\
+	 CEPH_FEATURE_CRUSH_TUNABLES5)
 
 #define CEPH_FEATURES_REQUIRED_DEFAULT   \
 	(CEPH_FEATURE_NOSRCADDR |	 \
-- 
2.4.3

Re: [PATCH 0/5] libceph: crush tunables5

[Sage Weil]

On Wed, 3 Feb 2016, Ilya Dryomov wrote:
> Hello,
> 
> This series adds support for the new chooseleaf_stable CRUSH tunable
> and enables TUNABLES5 feature bit.  It's likely that TUNABLES5 bit will
> be shared and cover request_redirect_t encoding in MOSDOpReply and
> possibly new file layouts but those aren't in master yet.
> 
> Thanks,
> 
>                 Ilya
> 
> 
> Ilya Dryomov (5):
>   crush: ensure bucket id is valid before indexing buckets array
>   crush: ensure take bucket value is valid
>   crush: add chooseleaf_stable tunable
>   crush: decode and initialize chooseleaf_stable
>   libceph: advertise support for TUNABLES5
> 
>  include/linux/ceph/ceph_features.h | 13 ++++++++++++-
>  include/linux/crush/crush.h        |  8 +++++++-
>  net/ceph/crush/mapper.c            | 33 ++++++++++++++++++++++++++-------
>  net/ceph/osdmap.c                  | 19 ++++++++++++++-----
>  4 files changed, 59 insertions(+), 14 deletions(-)

Whole series 

Reviewed-by: Sage Weil <sage@redhat.com>

Thanks!
sage

Re: [PATCH 0/5] libceph: crush tunables5

[Yan, Zheng]

On Wed, Feb 3, 2016 at 10:23 PM, Ilya Dryomov <idryomov@gmail.com> wrote:
> Hello,
>
> This series adds support for the new chooseleaf_stable CRUSH tunable
> and enables TUNABLES5 feature bit.  It's likely that TUNABLES5 bit will
> be shared and cover request_redirect_t encoding in MOSDOpReply and
> possibly new file layouts but those aren't in master yet.
>

file layout RP hasn't been merged. I will push the kernel patch once
it get merged

Regards
Yan, Zheng

> Thanks,
>
>                 Ilya
>
>
> Ilya Dryomov (5):
>   crush: ensure bucket id is valid before indexing buckets array
>   crush: ensure take bucket value is valid
>   crush: add chooseleaf_stable tunable
>   crush: decode and initialize chooseleaf_stable
>   libceph: advertise support for TUNABLES5
>
>  include/linux/ceph/ceph_features.h | 13 ++++++++++++-
>  include/linux/crush/crush.h        |  8 +++++++-
>  net/ceph/crush/mapper.c            | 33 ++++++++++++++++++++++++++-------
>  net/ceph/osdmap.c                  | 19 ++++++++++++++-----
>  4 files changed, 59 insertions(+), 14 deletions(-)
>
> --
> 2.4.3
>
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html