block: fixes for always associate blkg

PATCH In review Dennis Zhou 2 participants 4 messages Oct 20, 2018

Subsystems (2)

Block layer cgroup blkio

lore

Share & tools

lore

Dennis Zhou

author

2018-10-20 18:56 UTC #0 lore raw

cover block: fixes for always associate blkg

Hi everyone,

It was reported in [1] that blk_get_rl() cleanup patch was causing a null pointer dereference. After some back and forth debugging with Valdis, it turns out I wasn't properly handling association with recursive calls to make_request().

Another issue was identified with the blk_get_rl() update as it is possible under certain circumstances that a blkg cannot be allocated when called in blk_get_rl(). This could result in the blkcg_root being returned. However, the blkcg_root is a special case where all blkgs share the request_queue's request_list.

The original series can be found at [2].

[1] https://lore.kernel.org/lkml/[email protected]/
[2] https://lore.kernel.org/lkml/[email protected]/

This patchset contains the following 2 patches:
0001-blkcg-fix-edge-case-for-blk_get_rl-under-memory-pres.patch 0002-blkcg-reassociate-bios-when-make_request-is-called-r.patch

0001 addresses an edge case where a blkg cannot be created and can possibly return a blkg associated with the blkcg_root. 0002 fixes the stale association when make_request() is called recursively.

This patchset is on top of axboe#for-4.20/block bbc152825afc.

diffstats below:

Dennis Zhou (2):
blkcg: fix edge case for blk_get_rl() under memory pressure
blkcg: reassociate bios when make_request() is called recursively

 block/bio.c                | 20 ++++++++++++++++++++
 block/blk-core.c           |  1 +
 include/linux/bio.h        |  3 +++
 include/linux/blk-cgroup.h |  2 +-
 4 files changed, 25 insertions(+), 1 deletion(-)

Thanks,
Dennis

Dennis Zhou

author

2018-10-20 18:56 UTC #1 lore raw

PATCH 1/2 blkcg: fix edge case for blk_get_rl() under memory pressure

It is possible for blkg creation to fail when in blk_get_rl(). In this situation, the fallback logic returns the nearest created blkg. There is however special handling for the request_list for the root blkcg. This fixes the missing edge case from the earlier series changing blk_get_rl().

Fixes e2b0989954ae ("blkcg: cleanup and make blk_get_rl use blkg_lookup_create")

Signed-off-by Dennis Zhou <[email protected]>

---
 include/linux/blk-cgroup.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

include/linux/blk-cgroup.h +1 -1

index b7fd08013de2..1e76ceebeb5d 100644
@@ -597,7 +597,7 @@ static inline struct request_list *blk_get_rl(struct request_queue *q,
 	if (unlikely(!blkg))
 		blkg = __blkg_lookup_create(blkcg, q);
 
-	if (!blkg_tryget(blkg))
+	if (blkg->blkcg == &blkcg_root || !blkg_tryget(blkg))
 		goto rl_use_root;
 
 	rcu_read_unlock();

Signature · 1 line

2.17.1

Dennis Zhou

author

2018-10-20 18:56 UTC #2 lore raw

PATCH 2/2 blkcg: reassociate bios when make_request() is called recursively

In reply to Dennis Zhou

When submitting a bio, multiple recursive calls to make_request() may occur. This causes the initial associate done in blkcg_bio_issue_check() to be incorrect and reference the prior request_queue. This introduces a helper to do reassociation when make_request() is recursively called.

Fixes a7b39b4e961c ("blkcg: always associate a bio with a blkg")

Reported-by Valdis Kletnieks <[email protected]>

Signed-off-by Dennis Zhou <[email protected]>

Tested-by Valdis Kletnieks <[email protected]>

---
 block/bio.c         | 20 ++++++++++++++++++++
 block/blk-core.c    |  1 +
 include/linux/bio.h |  3 +++
 3 files changed, 24 insertions(+)

block/bio.c +20 -0

index 17a8b0aa7050..bbfeb4ee2892 100644
@@ -2083,6 +2083,26 @@ int bio_associate_create_blkg(struct request_queue *q, struct bio *bio)
 	return ret;
 }
 
+/**
+ * bio_reassociate_blkg - reassociate a bio with a blkg from q
+ * @q: request_queue where bio is going
+ * @bio: target bio
+ *
+ * When submitting a bio, multiple recursive calls to make_request() may occur.
+ * This causes the initial associate done in blkcg_bio_issue_check() to be
+ * incorrect and reference the prior request_queue.  This performs reassociation
+ * when this situation happens.
+ */
+int bio_reassociate_blkg(struct request_queue *q, struct bio *bio)
+{
+	if (bio->bi_blkg) {
+		blkg_put(bio->bi_blkg);
+		bio->bi_blkg = NULL;
+	}
+
+	return bio_associate_create_blkg(q, bio);
+}
+
 /**
  * bio_disassociate_task - undo bio_associate_current()
  * @bio: target bio

block/blk-core.c +1 -0

index cdfabc5646da..3ed60723e242 100644
@@ -2433,6 +2433,7 @@ blk_qc_t generic_make_request(struct bio *bio)
 			if (q)
 				blk_queue_exit(q);
 			q = bio->bi_disk->queue;
+			bio_reassociate_blkg(q, bio);
 			flags = 0;
 			if (bio->bi_opf & REQ_NOWAIT)
 				flags = BLK_MQ_REQ_NOWAIT;

include/linux/bio.h +3 -0

index f447b0ebb288..b47c7f716731 100644
@@ -514,6 +514,7 @@ int bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg);
 int bio_associate_blkg_from_css(struct bio *bio,
 				struct cgroup_subsys_state *css);
 int bio_associate_create_blkg(struct request_queue *q, struct bio *bio);
+int bio_reassociate_blkg(struct request_queue *q, struct bio *bio);
 void bio_disassociate_task(struct bio *bio);
 void bio_clone_blkg_association(struct bio *dst, struct bio *src);
 #else	/* CONFIG_BLK_CGROUP */

@@ -522,6 +523,8 @@ static inline int bio_associate_blkg_from_css(struct bio *bio,
 { return 0; }
 static inline int bio_associate_create_blkg(struct request_queue *q,
 					    struct bio *bio) { return 0; }
+static inline int bio_reassociate_blkg(struct request_queue *q, struct bio *bio)
+{ return 0; }
 static inline void bio_disassociate_task(struct bio *bio) { }
 static inline void bio_clone_blkg_association(struct bio *dst,
 					      struct bio *src) { }

Signature · 1 line

2.17.1

Jens Axboe

maintainer

2018-10-20 21:40 UTC #3 lore raw

In reply to Dennis Zhou

Quoted 18 lines · Dennis Zhou

Hi everyone,
It was reported in [1] that blk_get_rl() cleanup patch was causing a null pointer dereference. After some back and forth debugging with Valdis, it turns out I wasn't properly handling association with recursive calls to make_request().
Another issue was identified with the blk_get_rl() update as it is possible under certain circumstances that a blkg cannot be allocated when called in blk_get_rl(). This could result in the blkcg_root being returned. However, the blkcg_root is a special case where all blkgs share the request_queue's request_list.
The original series can be found at [2].
[1] https://lore.kernel.org/lkml/[email protected]/
[2] https://lore.kernel.org/lkml/[email protected]/

Quoted 7 lines

This patchset contains the following 2 patches:
0001-blkcg-fix-edge-case-for-blk_get_rl-under-memory-pres.patch 0002-blkcg-reassociate-bios-when-make_request-is-called-r.patch
0001 addresses an edge case where a blkg cannot be created and can possibly return a blkg associated with the blkcg_root. 0002 fixes the stale association when make_request() is called recursively.

Thanks Dennis, applied.

Signature · 1 line

Jens Axboe