Skip to contentSkip to conversation

cgroup: Replace cgroup->ancestor_ids[] with ->ancestors[]

PATCH Superseded Tejun Heo 7 participants 28 messages Jul 29, 2022 – Oct 14, 2022

Subsystems (5)
Networking netfilter BPF perf events +1 more subsystems
lore
Share & tools
lore

Tejun Heo

author maintainer
#0 lore raw

PATCH 1/1 cgroup: Replace cgroup->ancestor_ids[] with ->ancestors[]

Every cgroup knows all its ancestors through its ->ancestor_ids[]. While all the existing users only need the IDs, there's no advantage to remembering the IDs instead of the pointers directly. Let's replace cgroup->ancestor_ids[] with ->ancestors[] so that it's easy to access non-ID ancestor fields.

This patch shouldn't cause any behavior differences.

---
Namhyung, I think the change in bperf_cgroup is correct but couldn't figure out how to test it (normal perf build wouldn't compile it). Can you please see whether it's correct?

Thanks.

 include/linux/cgroup-defs.h                 |   10 +++++-----
 include/linux/cgroup.h                      |    2 +-
 kernel/cgroup/cgroup.c                      |    7 +++----
 net/netfilter/nft_socket.c                  |    5 +++--
 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c |    2 +-
 5 files changed, 13 insertions(+), 13 deletions(-)
include/linux/cgroup-defs.h +5 -5
index 63bf43c7ca3b..51fa744c2e9d 100644
@@ -379,7 +379,7 @@ struct cgroup {
 	/*
 	 * The depth this cgroup is at.  The root is at depth zero and each
 	 * step down the hierarchy increments the level.  This along with
-	 * ancestor_ids[] can determine whether a given cgroup is a
+	 * ancestors[] can determine whether a given cgroup is a
 	 * descendant of another without traversing the hierarchy.
 	 */
 	int level;
@@ -499,8 +499,8 @@ struct cgroup {
 	/* Used to store internal freezer state */
 	struct cgroup_freezer_state freezer;
 
-	/* ids of the ancestors at each level including self */
-	u64 ancestor_ids[];
+	/* All ancestors including self */
+	struct cgroup *ancestors[];
 };
 
 /*
@@ -520,8 +520,8 @@ struct cgroup_root {
 	/* The root cgroup.  Root is destroyed on its release. */
 	struct cgroup cgrp;
 
-	/* for cgrp->ancestor_ids[0] */
-	u64 cgrp_ancestor_id_storage;
+	/* for cgrp->ancestors[0] */
+	u64 cgrp_ancestor_storage;
 
 	/* Number of cgroups in the hierarchy, used only for /proc/cgroups */
 	atomic_t nr_cgrps;
include/linux/cgroup.h +1 -1
index ed53bfe7c46c..5334b6134399 100644
@@ -574,7 +574,7 @@ static inline bool cgroup_is_descendant(struct cgroup *cgrp,
 {
 	if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
 		return false;
-	return cgrp->ancestor_ids[ancestor->level] == cgroup_id(ancestor);
+	return cgrp->ancestors[ancestor->level] == ancestor;
 }
 
 /**
kernel/cgroup/cgroup.c +3 -4
index 85fa4c8587a8..ce587fe43dab 100644
@@ -2047,7 +2047,7 @@ int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
 	}
 	root_cgrp->kn = kernfs_root_to_node(root->kf_root);
 	WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
-	root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
+	root_cgrp->ancestors[0] = root_cgrp;
 
 	ret = css_populate_dir(&root_cgrp->self);
 	if (ret)
@@ -5391,8 +5391,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 	int ret;
 
 	/* allocate the cgroup and its ID, 0 is reserved for the root */
-	cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
-		       GFP_KERNEL);
+	cgrp = kzalloc(struct_size(cgrp, ancestors, (level + 1)), GFP_KERNEL);
 	if (!cgrp)
 		return ERR_PTR(-ENOMEM);
 
@@ -5444,7 +5443,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 
 	spin_lock_irq(&css_set_lock);
 	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
-		cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
+		cgrp->ancestors[tcgrp->level] = tcgrp;
 
 		if (tcgrp != cgrp) {
 			tcgrp->nr_descendants++;
net/netfilter/nft_socket.c +3 -2
index 05ae5a338b6f..d64784d4bd02 100644
@@ -40,6 +40,7 @@ static noinline bool
 nft_sock_get_eval_cgroupv2(u32 *dest, struct sock *sk, const struct nft_pktinfo *pkt, u32 level)
 {
 	struct cgroup *cgrp;
+	u64 cgid;
 
 	if (!sk_fullsock(sk))
 		return false;
@@ -48,8 +49,8 @@ nft_sock_get_eval_cgroupv2(u32 *dest, struct sock *sk, const struct nft_pktinfo
 	if (level > cgrp->level)
 		return false;
 
-	memcpy(dest, &cgrp->ancestor_ids[level], sizeof(u64));
-
+	cgid = cgroup_id(cgrp->ancestors[level]);
+	memcpy(dest, &cgid, sizeof(u64));
 	return true;
 }
 #endif
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +1 -1
index 292c430768b5..bd6a420acc8f 100644
@@ -68,7 +68,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 			break;
 
 		// convert cgroup-id to a map index
-		cgrp_id = BPF_CORE_READ(cgrp, ancestor_ids[i]);
+		cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
 		elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
 		if (!elem)
 			continue;
maintainer Acked-by on v1 1/1
#1 lore raw

Hi Tejun,

Quoted 7 lines · Tejun Heo

Every cgroup knows all its ancestors through its ->ancestor_ids[]. While all the existing users only need the IDs, there's no advantage to remembering the IDs instead of the pointers directly. Let's replace cgroup->ancestor_ids[] with ->ancestors[] so that it's easy to access non-ID ancestor fields.

Quoted 7 lines

This patch shouldn't cause any behavior differences.

---
Namhyung, I think the change in bperf_cgroup is correct but couldn't figure out how to test it (normal perf build wouldn't compile it). Can you please see whether it's correct?

Looks ok to me. For the perf part,

Note that you need to pass BUILD_BPF_SKEL=1 when you build perf to enable BPF stuff. IIRC Arnaldo will make it default at some point.

Thanks,
Namhyung

Quoted 126 lines

Thanks.

 include/linux/cgroup-defs.h                 |   10 +++++-----
 include/linux/cgroup.h                      |    2 +-
 kernel/cgroup/cgroup.c                      |    7 +++----
 net/netfilter/nft_socket.c                  |    5 +++--
 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c |    2 +-
 5 files changed, 13 insertions(+), 13 deletions(-)
include/linux/cgroup-defs.h +5 -5
index 63bf43c7ca3b..51fa744c2e9d 100644
@@ -379,7 +379,7 @@ struct cgroup {
        /*
         * The depth this cgroup is at.  The root is at depth zero and each
         * step down the hierarchy increments the level.  This along with
-        * ancestor_ids[] can determine whether a given cgroup is a
+        * ancestors[] can determine whether a given cgroup is a
         * descendant of another without traversing the hierarchy.
         */
        int level;
@@ -499,8 +499,8 @@ struct cgroup {
        /* Used to store internal freezer state */
        struct cgroup_freezer_state freezer;
 
-       /* ids of the ancestors at each level including self */
-       u64 ancestor_ids[];
+       /* All ancestors including self */
+       struct cgroup *ancestors[];
 };
 
 /*
@@ -520,8 +520,8 @@ struct cgroup_root {
        /* The root cgroup.  Root is destroyed on its release. */
        struct cgroup cgrp;
 
-       /* for cgrp->ancestor_ids[0] */
-       u64 cgrp_ancestor_id_storage;
+       /* for cgrp->ancestors[0] */
+       u64 cgrp_ancestor_storage;
 
        /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
        atomic_t nr_cgrps;
include/linux/cgroup.h +1 -1
index ed53bfe7c46c..5334b6134399 100644
@@ -574,7 +574,7 @@ static inline bool cgroup_is_descendant(struct cgroup *cgrp,
 {
        if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
                return false;
-       return cgrp->ancestor_ids[ancestor->level] == cgroup_id(ancestor);
+       return cgrp->ancestors[ancestor->level] == ancestor;
 }
 
 /**
kernel/cgroup/cgroup.c +3 -4
index 85fa4c8587a8..ce587fe43dab 100644
@@ -2047,7 +2047,7 @@ int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
        }
        root_cgrp->kn = kernfs_root_to_node(root->kf_root);
        WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
-       root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
+       root_cgrp->ancestors[0] = root_cgrp;
 
        ret = css_populate_dir(&root_cgrp->self);
        if (ret)
@@ -5391,8 +5391,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
        int ret;
 
        /* allocate the cgroup and its ID, 0 is reserved for the root */
-       cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
-                      GFP_KERNEL);
+       cgrp = kzalloc(struct_size(cgrp, ancestors, (level + 1)), GFP_KERNEL);
        if (!cgrp)
                return ERR_PTR(-ENOMEM);
 
@@ -5444,7 +5443,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 
        spin_lock_irq(&css_set_lock);
        for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
-               cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
+               cgrp->ancestors[tcgrp->level] = tcgrp;
 
                if (tcgrp != cgrp) {
                        tcgrp->nr_descendants++;
net/netfilter/nft_socket.c +3 -2
index 05ae5a338b6f..d64784d4bd02 100644
@@ -40,6 +40,7 @@ static noinline bool
 nft_sock_get_eval_cgroupv2(u32 *dest, struct sock *sk, const struct nft_pktinfo *pkt, u32 level)
 {
        struct cgroup *cgrp;
+       u64 cgid;
 
        if (!sk_fullsock(sk))
                return false;
@@ -48,8 +49,8 @@ nft_sock_get_eval_cgroupv2(u32 *dest, struct sock *sk, const struct nft_pktinfo
        if (level > cgrp->level)
                return false;
 
-       memcpy(dest, &cgrp->ancestor_ids[level], sizeof(u64));
-
+       cgid = cgroup_id(cgrp->ancestors[level]);
+       memcpy(dest, &cgid, sizeof(u64));
        return true;
 }
 #endif
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +1 -1
index 292c430768b5..bd6a420acc8f 100644
@@ -68,7 +68,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
                        break;
 
                // convert cgroup-id to a map index
-               cgrp_id = BPF_CORE_READ(cgrp, ancestor_ids[i]);
+               cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
                elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
                if (!elem)
                        continue;
author maintainer
#2 lore raw

PATCH v2 1/1 cgroup: Replace cgroup->ancestor_ids[] with ->ancestors[]

In reply to Tejun Heo

Every cgroup knows all its ancestors through its ->ancestor_ids[]. There's no advantage to remembering the IDs instead of the pointers directly and this makes the array useless for finding an actual ancestor cgroup forcing cgroup_ancestor() to iteratively walk up the hierarchy instead. Let's replace cgroup->ancestor_ids[] with ->ancestors[] and remove the walking-up from cgroup_ancestor().

This patch shouldn't cause user-visible behavior differences.

v2: Update cgroup_ancestor() to use ->ancestors[].

---
 include/linux/cgroup-defs.h                 |   10 +++++-----
 include/linux/cgroup.h                      |    8 +++-----
 kernel/cgroup/cgroup.c                      |    7 +++----
 net/netfilter/nft_socket.c                  |    9 +++++----
 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c |    2 +-
 5 files changed, 17 insertions(+), 19 deletions(-)
include/linux/cgroup-defs.h +5 -5
index 63bf43c7ca3b..51fa744c2e9d 100644
@@ -379,7 +379,7 @@ struct cgroup {
 	/*
 	 * The depth this cgroup is at.  The root is at depth zero and each
 	 * step down the hierarchy increments the level.  This along with
-	 * ancestor_ids[] can determine whether a given cgroup is a
+	 * ancestors[] can determine whether a given cgroup is a
 	 * descendant of another without traversing the hierarchy.
 	 */
 	int level;
@@ -499,8 +499,8 @@ struct cgroup {
 	/* Used to store internal freezer state */
 	struct cgroup_freezer_state freezer;
 
-	/* ids of the ancestors at each level including self */
-	u64 ancestor_ids[];
+	/* All ancestors including self */
+	struct cgroup *ancestors[];
 };
 
 /*
@@ -520,8 +520,8 @@ struct cgroup_root {
 	/* The root cgroup.  Root is destroyed on its release. */
 	struct cgroup cgrp;
 
-	/* for cgrp->ancestor_ids[0] */
-	u64 cgrp_ancestor_id_storage;
+	/* for cgrp->ancestors[0] */
+	u64 cgrp_ancestor_storage;
 
 	/* Number of cgroups in the hierarchy, used only for /proc/cgroups */
 	atomic_t nr_cgrps;
include/linux/cgroup.h +3 -5
index ed53bfe7c46c..4d143729b246 100644
@@ -574,7 +574,7 @@ static inline bool cgroup_is_descendant(struct cgroup *cgrp,
 {
 	if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
 		return false;
-	return cgrp->ancestor_ids[ancestor->level] == cgroup_id(ancestor);
+	return cgrp->ancestors[ancestor->level] == ancestor;
 }
 
 /**
@@ -591,11 +591,9 @@ static inline bool cgroup_is_descendant(struct cgroup *cgrp,
 static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp,
 					     int ancestor_level)
 {
-	if (cgrp->level < ancestor_level)
+	if (ancestor_level < 0 || ancestor_level > cgrp->level)
 		return NULL;
-	while (cgrp && cgrp->level > ancestor_level)
-		cgrp = cgroup_parent(cgrp);
-	return cgrp;
+	return cgrp->ancestors[ancestor_level];
 }
 
 /**
kernel/cgroup/cgroup.c +3 -4
index 85fa4c8587a8..ce587fe43dab 100644
@@ -2047,7 +2047,7 @@ int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
 	}
 	root_cgrp->kn = kernfs_root_to_node(root->kf_root);
 	WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
-	root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
+	root_cgrp->ancestors[0] = root_cgrp;
 
 	ret = css_populate_dir(&root_cgrp->self);
 	if (ret)
@@ -5391,8 +5391,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 	int ret;
 
 	/* allocate the cgroup and its ID, 0 is reserved for the root */
-	cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
-		       GFP_KERNEL);
+	cgrp = kzalloc(struct_size(cgrp, ancestors, (level + 1)), GFP_KERNEL);
 	if (!cgrp)
 		return ERR_PTR(-ENOMEM);
 
@@ -5444,7 +5443,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 
 	spin_lock_irq(&css_set_lock);
 	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
-		cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
+		cgrp->ancestors[tcgrp->level] = tcgrp;
 
 		if (tcgrp != cgrp) {
 			tcgrp->nr_descendants++;
net/netfilter/nft_socket.c +5 -4
index 05ae5a338b6f..d982a7c22a77 100644
@@ -40,16 +40,17 @@ static noinline bool
 nft_sock_get_eval_cgroupv2(u32 *dest, struct sock *sk, const struct nft_pktinfo *pkt, u32 level)
 {
 	struct cgroup *cgrp;
+	u64 cgid;
 
 	if (!sk_fullsock(sk))
 		return false;
 
-	cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
-	if (level > cgrp->level)
+	cgrp = cgroup_ancestor(sock_cgroup_ptr(&sk->sk_cgrp_data), level);
+	if (!cgrp)
 		return false;
 
-	memcpy(dest, &cgrp->ancestor_ids[level], sizeof(u64));
-
+	cgid = cgroup_id(cgrp);
+	memcpy(dest, &cgid, sizeof(u64));
 	return true;
 }
 #endif
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +1 -1
index 292c430768b5..bd6a420acc8f 100644
@@ -68,7 +68,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 			break;
 
 		// convert cgroup-id to a map index
-		cgrp_id = BPF_CORE_READ(cgrp, ancestor_ids[i]);
+		cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
 		elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
 		if (!elem)
 			continue;
maintainer
#3 lore raw

include/linux/cgroup-defs.h struct cgroup_root { · line 524

Quoted 8 lines · Tejun Heo
include/linux/cgroup-defs.h +2 -2
@@ -520,8 +520,8 @@ struct cgroup_root {
 	/* The root cgroup.  Root is destroyed on its release. */
 	struct cgroup cgrp;
 
-	/* for cgrp->ancestor_ids[0] */
-	u64 cgrp_ancestor_id_storage;
+	/* for cgrp->ancestors[0] */
+	u64 cgrp_ancestor_storage;

Just noticed, this member is (and was AFAICS) superfluous.

Michal

author maintainer
#4 lore raw
Quoted 11 lines · Michal Koutný

include/linux/cgroup-defs.h struct cgroup_root { · line 524

Quoted 8 lines · Tejun Heo
include/linux/cgroup-defs.h +2 -2
@@ -520,8 +520,8 @@ struct cgroup_root {
 	/* The root cgroup.  Root is destroyed on its release. */
 	struct cgroup cgrp;
 
-	/* for cgrp->ancestor_ids[0] */
-	u64 cgrp_ancestor_id_storage;
+	/* for cgrp->ancestors[0] */
+	u64 cgrp_ancestor_storage;

Just noticed, this member is (and was AFAICS) superfluous.

I should have changed the type to struct cgroup * but that's the space into which cgroup_root->cgrp->ancestors[] stretch into.

Thanks.

Signature · 1 line
tejun
author maintainer
#5 lore raw

PATCH v3 1/1 cgroup: Replace cgroup->ancestor_ids[] with ->ancestors[]

In reply to Tejun Heo

Every cgroup knows all its ancestors through its ->ancestor_ids[]. There's no advantage to remembering the IDs instead of the pointers directly and this makes the array useless for finding an actual ancestor cgroup forcing cgroup_ancestor() to iteratively walk up the hierarchy instead. Let's replace cgroup->ancestor_ids[] with ->ancestors[] and remove the walking-up from cgroup_ancestor().

While at it, improve comments around cgroup_root->cgrp_ancestor_storage.

This patch shouldn't cause user-visible behavior differences.

v2: Update cgroup_ancestor() to use ->ancestors[].

v3: cgroup_root->cgrp_ancestor_storage's type is updated to match

    cgroup->ancestors[]. Better comments.
---
 include/linux/cgroup-defs.h                 |   16 ++++++++++------
 include/linux/cgroup.h                      |    8 +++-----
 kernel/cgroup/cgroup.c                      |    7 +++----
 net/netfilter/nft_socket.c                  |    9 +++++----
 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c |    2 +-
 5 files changed, 22 insertions(+), 20 deletions(-)
include/linux/cgroup-defs.h +10 -6
index 63bf43c7ca3b..52a3c47c89bc 100644
@@ -379,7 +379,7 @@ struct cgroup {
 	/*
 	 * The depth this cgroup is at.  The root is at depth zero and each
 	 * step down the hierarchy increments the level.  This along with
-	 * ancestor_ids[] can determine whether a given cgroup is a
+	 * ancestors[] can determine whether a given cgroup is a
 	 * descendant of another without traversing the hierarchy.
 	 */
 	int level;
@@ -499,8 +499,8 @@ struct cgroup {
 	/* Used to store internal freezer state */
 	struct cgroup_freezer_state freezer;
 
-	/* ids of the ancestors at each level including self */
-	u64 ancestor_ids[];
+	/* All ancestors including self */
+	struct cgroup *ancestors[];
 };
 
 /*
@@ -517,11 +517,15 @@ struct cgroup_root {
 	/* Unique id for this hierarchy. */
 	int hierarchy_id;
 
-	/* The root cgroup.  Root is destroyed on its release. */
+	/*
+	 * The root cgroup. The containing cgroup_root will be destroyed on its
+	 * release. cgrp->ancestors[0] will be used overflowing into the
+	 * following field. cgrp_ancestor_storage must immediately follow.
+	 */
 	struct cgroup cgrp;
 
-	/* for cgrp->ancestor_ids[0] */
-	u64 cgrp_ancestor_id_storage;
+	/* must follow cgrp for cgrp->ancestors[0], see above */
+	struct cgroup *cgrp_ancestor_storage;
 
 	/* Number of cgroups in the hierarchy, used only for /proc/cgroups */
 	atomic_t nr_cgrps;
include/linux/cgroup.h +3 -5
index ed53bfe7c46c..4d143729b246 100644
@@ -574,7 +574,7 @@ static inline bool cgroup_is_descendant(struct cgroup *cgrp,
 {
 	if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
 		return false;
-	return cgrp->ancestor_ids[ancestor->level] == cgroup_id(ancestor);
+	return cgrp->ancestors[ancestor->level] == ancestor;
 }
 
 /**
@@ -591,11 +591,9 @@ static inline bool cgroup_is_descendant(struct cgroup *cgrp,
 static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp,
 					     int ancestor_level)
 {
-	if (cgrp->level < ancestor_level)
+	if (ancestor_level < 0 || ancestor_level > cgrp->level)
 		return NULL;
-	while (cgrp && cgrp->level > ancestor_level)
-		cgrp = cgroup_parent(cgrp);
-	return cgrp;
+	return cgrp->ancestors[ancestor_level];
 }
 
 /**
kernel/cgroup/cgroup.c +3 -4
index 85fa4c8587a8..ce587fe43dab 100644
@@ -2047,7 +2047,7 @@ int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
 	}
 	root_cgrp->kn = kernfs_root_to_node(root->kf_root);
 	WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
-	root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
+	root_cgrp->ancestors[0] = root_cgrp;
 
 	ret = css_populate_dir(&root_cgrp->self);
 	if (ret)
@@ -5391,8 +5391,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 	int ret;
 
 	/* allocate the cgroup and its ID, 0 is reserved for the root */
-	cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
-		       GFP_KERNEL);
+	cgrp = kzalloc(struct_size(cgrp, ancestors, (level + 1)), GFP_KERNEL);
 	if (!cgrp)
 		return ERR_PTR(-ENOMEM);
 
@@ -5444,7 +5443,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
 
 	spin_lock_irq(&css_set_lock);
 	for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
-		cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
+		cgrp->ancestors[tcgrp->level] = tcgrp;
 
 		if (tcgrp != cgrp) {
 			tcgrp->nr_descendants++;
net/netfilter/nft_socket.c +5 -4
index 05ae5a338b6f..d982a7c22a77 100644
@@ -40,16 +40,17 @@ static noinline bool
 nft_sock_get_eval_cgroupv2(u32 *dest, struct sock *sk, const struct nft_pktinfo *pkt, u32 level)
 {
 	struct cgroup *cgrp;
+	u64 cgid;
 
 	if (!sk_fullsock(sk))
 		return false;
 
-	cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
-	if (level > cgrp->level)
+	cgrp = cgroup_ancestor(sock_cgroup_ptr(&sk->sk_cgrp_data), level);
+	if (!cgrp)
 		return false;
 
-	memcpy(dest, &cgrp->ancestor_ids[level], sizeof(u64));
-
+	cgid = cgroup_id(cgrp);
+	memcpy(dest, &cgid, sizeof(u64));
 	return true;
 }
 #endif
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +1 -1
index 292c430768b5..bd6a420acc8f 100644
@@ -68,7 +68,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 			break;
 
 		// convert cgroup-id to a map index
-		cgrp_id = BPF_CORE_READ(cgrp, ancestor_ids[i]);
+		cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
 		elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
 		if (!elem)
 			continue;
author maintainer Applied
#6 lore raw
Quoted 11 lines · Tejun Heo

Every cgroup knows all its ancestors through its ->ancestor_ids[]. There's no advantage to remembering the IDs instead of the pointers directly and this makes the array useless for finding an actual ancestor cgroup forcing cgroup_ancestor() to iteratively walk up the hierarchy instead. Let's replace cgroup->ancestor_ids[] with ->ancestors[] and remove the walking-up from cgroup_ancestor().

While at it, improve comments around cgroup_root->cgrp_ancestor_storage.

This patch shouldn't cause user-visible behavior differences.

Quoted 7 lines

v2: Update cgroup_ancestor() to use ->ancestors[].

v3: cgroup_root->cgrp_ancestor_storage's type is updated to match

    cgroup->ancestors[]. Better comments.

Applied to cgroup/for-6.1.

Signature · 1 line
tejun
maintainer
#7 lore raw

Hi Tejun,

Quoted 15 lines · Tejun Heo
Quoted 13 lines · Tejun Heo

Every cgroup knows all its ancestors through its ->ancestor_ids[]. There's no advantage to remembering the IDs instead of the pointers directly and this makes the array useless for finding an actual ancestor cgroup forcing cgroup_ancestor() to iteratively walk up the hierarchy instead. Let's replace cgroup->ancestor_ids[] with ->ancestors[] and remove the walking-up from cgroup_ancestor().

While at it, improve comments around cgroup_root->cgrp_ancestor_storage.

This patch shouldn't cause user-visible behavior differences.

v2: Update cgroup_ancestor() to use ->ancestors[].

Quoted 7 lines
Quoted 5 lines

v3: cgroup_root->cgrp_ancestor_storage's type is updated to match

    cgroup->ancestors[]. Better comments.

Applied to cgroup/for-6.1.

I've realized that perf stat change needs backward compatibility. Will send a fix soon.

Thanks,
Namhyung

maintainer
#8 lore raw

PATCH 1/1 perf stat: Support old kernels for bperf cgroup counting

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 29 ++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +28 -1
index 488bd398f01d..4fe61043de04 100644
@@ -43,12 +43,39 @@ struct {
 	__uint(value_size, sizeof(struct bpf_perf_event_value));
 } cgrp_readings SEC(".maps");
 
+/* new kernel cgroup definition */
+struct cgroup___new {
+	int level;
+	struct cgroup *ancestors[];
+} __attribute__((preserve_access_index));
+
+/* old kernel cgroup definition */
+struct cgroup___old {
+	int level;
+	u64 ancestor_ids[];
+} __attribute__((preserve_access_index));
+
 const volatile __u32 num_events = 1;
 const volatile __u32 num_cpus = 1;
 
 int enabled = 0;
 int use_cgroup_v2 = 0;
 
+static inline __u64 get_cgroup_v1_ancestor_id(struct cgroup *cgrp, int level)
+{
+	/* recast pointer to capture new type for compiler */
+	struct cgroup___new *cgrp_new = (void *)cgrp;
+
+	if (bpf_core_field_exists(cgrp_new->ancestors)) {
+		return BPF_CORE_READ(cgrp_new, ancestors[level], kn, id);
+	} else {
+		/* recast pointer to capture old type for compiler */
+		struct cgroup___old *cgrp_old = (void *)cgrp;
+
+		return BPF_CORE_READ(cgrp_old, ancestor_ids[level]);
+	}
+}
+
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
 	struct task_struct *p = (void *)bpf_get_current_task();
@@ -70,7 +97,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 			break;
 
 		// convert cgroup-id to a map index
-		cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
+		cgrp_id = get_cgroup_v1_ancestor_id(cgrp, i);
 		elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
 		if (!elem)
 			continue;
Signature · 1 line
2.37.3.968.ga6b4b080e4-goog
author maintainer
#9 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Quoted 4 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Quoted 8 lines

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

FWIW, looks fine to me and I'd be happy to route this through the cgroup tree once it gets acked.

Thanks.

Signature · 1 line
tejun
maintainer
#10 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Quoted 9 lines · Tejun Heo
Quoted 7 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Quoted 8 lines
Quoted 5 lines

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

FWIW, looks fine to me and I'd be happy to route this through the cgroup tree once it gets acked.

Thanks Tejun!

Can any perf + bpf folks take a look?

#11 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

In reply to Namhyung Kim
Quoted 4 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Quoted 8 lines

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

could you please paste the cgroup tree link?

thanks,
jirka

Quoted 60 lines
 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 29 ++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +28 -1
index 488bd398f01d..4fe61043de04 100644
@@ -43,12 +43,39 @@ struct {
 	__uint(value_size, sizeof(struct bpf_perf_event_value));
 } cgrp_readings SEC(".maps");
 
+/* new kernel cgroup definition */
+struct cgroup___new {
+	int level;
+	struct cgroup *ancestors[];
+} __attribute__((preserve_access_index));
+
+/* old kernel cgroup definition */
+struct cgroup___old {
+	int level;
+	u64 ancestor_ids[];
+} __attribute__((preserve_access_index));
+
 const volatile __u32 num_events = 1;
 const volatile __u32 num_cpus = 1;
 
 int enabled = 0;
 int use_cgroup_v2 = 0;
 
+static inline __u64 get_cgroup_v1_ancestor_id(struct cgroup *cgrp, int level)
+{
+	/* recast pointer to capture new type for compiler */
+	struct cgroup___new *cgrp_new = (void *)cgrp;
+
+	if (bpf_core_field_exists(cgrp_new->ancestors)) {
+		return BPF_CORE_READ(cgrp_new, ancestors[level], kn, id);
+	} else {
+		/* recast pointer to capture old type for compiler */
+		struct cgroup___old *cgrp_old = (void *)cgrp;
+
+		return BPF_CORE_READ(cgrp_old, ancestor_ids[level]);
+	}
+}
+
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
 	struct task_struct *p = (void *)bpf_get_current_task();
@@ -70,7 +97,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 			break;
 
 		// convert cgroup-id to a map index
-		cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
+		cgrp_id = get_cgroup_v1_ancestor_id(cgrp, i);
 		elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
 		if (!elem)
 			continue;
Signature · 1 line
2.37.3.968.ga6b4b080e4-goog
maintainer
#12 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Hi Jiri,

Quoted 9 lines · Jiri Olsa
Quoted 7 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Quoted 7 lines
Quoted 5 lines

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

could you please paste the cgroup tree link?

Do you mean this?

https://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git

Thanks,.
Namhyung

#13 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Quoted 18 lines · Namhyung Kim

Hi Jiri,

Quoted 15 lines · Jiri Olsa
Quoted 12 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

Quoted 6 lines
Quoted 1 line

could you please paste the cgroup tree link?

Do you mean this?

https://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git

Which branch and cset in that tree does you perf skel depends on?

- Arnaldo

Quoted 2 lines

Thanks,.
Namhyung

maintainer
#14 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Quoted 24 lines · Arnaldo Carvalho de Melo
Quoted 20 lines · Namhyung Kim

Hi Jiri,

Quoted 16 lines · Jiri Olsa
Quoted 4 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Quoted 8 lines

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

could you please paste the cgroup tree link?

Quoted 7 lines
Quoted 4 lines

Do you mean this?

https://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git

Which branch and cset in that tree does you perf skel depends on?

I believe it's for-6.1 and the cset is in

https://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git/commit/?h=for-6.1&id=7f203bc89eb66d6afde7eae91347fc0352090cc3

Thanks,
Namhyung

#15 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

In reply to Namhyung Kim
Quoted 18 lines · Namhyung Kim

Hi Jiri,

Quoted 15 lines · Jiri Olsa
Quoted 12 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

Quoted 5 lines
Quoted 1 line

could you please paste the cgroup tree link?

Do you mean this?

https://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git

yea, wanted to try that.. I cherry-picked the 7f203bc89eb6 ;-)

thanks,
jirka

Quoted 3 lines

Thanks,.
Namhyung

Acked-by on v1 1/1
#16 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

In reply to Namhyung Kim
Quoted 40 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 29 ++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +12 -0
index 488bd398f01d..4fe61043de04 100644
@@ -43,12 +43,39 @@ struct {
        __uint(value_size, sizeof(struct bpf_perf_event_value));
 } cgrp_readings SEC(".maps");
 
+/* new kernel cgroup definition */
+struct cgroup___new {
+       int level;
+       struct cgroup *ancestors[];
+} __attribute__((preserve_access_index));
+
+/* old kernel cgroup definition */
+struct cgroup___old {
+       int level;
+       u64 ancestor_ids[];
+} __attribute__((preserve_access_index));
+
 const volatile __u32 num_events = 1;
 const volatile __u32 num_cpus = 1;
 

tools/perf/util/bpf_skel/bperf_cgroup.bpf.c struct { · line 70

Quoted 10 lines
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +7 -0
 int enabled = 0;
 int use_cgroup_v2 = 0;
 
+static inline __u64 get_cgroup_v1_ancestor_id(struct cgroup *cgrp, int level)
+{
+       /* recast pointer to capture new type for compiler */
+       struct cgroup___new *cgrp_new = (void *)cgrp;
+
+       if (bpf_core_field_exists(cgrp_new->ancestors)) {
+               return BPF_CORE_READ(cgrp_new, ancestors[level], kn, id);

have you checked generated BPF code for this ancestors[level] access? I'd expect CO-RE relocation for finding ancestors offset and then just normal + level * 8 arithmetic, but would be nice to confirm. Apart from this, looks good to me:

Quoted 23 lines
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +9 -2
+       } else {
+               /* recast pointer to capture old type for compiler */
+               struct cgroup___old *cgrp_old = (void *)cgrp;
+
+               return BPF_CORE_READ(cgrp_old, ancestor_ids[level]);
+       }
+}
+
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
        struct task_struct *p = (void *)bpf_get_current_task();
@@ -70,7 +97,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
                        break;
 
                // convert cgroup-id to a map index
-               cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
+               cgrp_id = get_cgroup_v1_ancestor_id(cgrp, i);
                elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
                if (!elem)
                        continue;
--

2.37.3.968.ga6b4b080e4-goog

maintainer
#17 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Hello,

Quoted 58 lines · Andrii Nakryiko
Quoted 40 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 29 ++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +12 -0
index 488bd398f01d..4fe61043de04 100644
@@ -43,12 +43,39 @@ struct {
        __uint(value_size, sizeof(struct bpf_perf_event_value));
 } cgrp_readings SEC(".maps");
 
+/* new kernel cgroup definition */
+struct cgroup___new {
+       int level;
+       struct cgroup *ancestors[];
+} __attribute__((preserve_access_index));
+
+/* old kernel cgroup definition */
+struct cgroup___old {
+       int level;
+       u64 ancestor_ids[];
+} __attribute__((preserve_access_index));
+
 const volatile __u32 num_events = 1;
 const volatile __u32 num_cpus = 1;
 

tools/perf/util/bpf_skel/bperf_cgroup.bpf.c struct { · line 70

Quoted 10 lines
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +7 -0
 int enabled = 0;
 int use_cgroup_v2 = 0;
 
+static inline __u64 get_cgroup_v1_ancestor_id(struct cgroup *cgrp, int level)
+{
+       /* recast pointer to capture new type for compiler */
+       struct cgroup___new *cgrp_new = (void *)cgrp;
+
+       if (bpf_core_field_exists(cgrp_new->ancestors)) {
+               return BPF_CORE_READ(cgrp_new, ancestors[level], kn, id);

have you checked generated BPF code for this ancestors[level] access? I'd expect CO-RE relocation for finding ancestors offset and then just normal + level * 8 arithmetic, but would be nice to confirm. Apart from this, looks good to me:

Quoted 1 line

Thanks for your review!

How can I check the generated code? Do you have something works with skeletons or do I have to save the BPF object somehow during the build?

Thanks,
Namhyung

#18 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Quoted 63 lines · Namhyung Kim

Hello,

Quoted 58 lines · Andrii Nakryiko
Quoted 40 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 29 ++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +12 -0
index 488bd398f01d..4fe61043de04 100644
@@ -43,12 +43,39 @@ struct {
        __uint(value_size, sizeof(struct bpf_perf_event_value));
 } cgrp_readings SEC(".maps");
 
+/* new kernel cgroup definition */
+struct cgroup___new {
+       int level;
+       struct cgroup *ancestors[];
+} __attribute__((preserve_access_index));
+
+/* old kernel cgroup definition */
+struct cgroup___old {
+       int level;
+       u64 ancestor_ids[];
+} __attribute__((preserve_access_index));
+
 const volatile __u32 num_events = 1;
 const volatile __u32 num_cpus = 1;
 

tools/perf/util/bpf_skel/bperf_cgroup.bpf.c struct { · line 70

Quoted 10 lines
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +7 -0
 int enabled = 0;
 int use_cgroup_v2 = 0;
 
+static inline __u64 get_cgroup_v1_ancestor_id(struct cgroup *cgrp, int level)
+{
+       /* recast pointer to capture new type for compiler */
+       struct cgroup___new *cgrp_new = (void *)cgrp;
+
+       if (bpf_core_field_exists(cgrp_new->ancestors)) {
+               return BPF_CORE_READ(cgrp_new, ancestors[level], kn, id);

have you checked generated BPF code for this ancestors[level] access? I'd expect CO-RE relocation for finding ancestors offset and then just normal + level * 8 arithmetic, but would be nice to confirm. Apart from this, looks good to me:

Quoted 7 lines
Quoted 1 line

Thanks for your review!

How can I check the generated code? Do you have something works with skeletons or do I have to save the BPF object somehow during the build?

skeleton is generated from ELF BPF object file. You can do llvm-objdump -d <obj.bpf.o> to see instructions. Unfortunately you can't see BPF CO-RE relocations this way, you'd have to use something like my custom tool ([0]).

But anyways, I checked locally similar code pattern and I think it's all good from BPF CO-RE perspective. I see appropriate relocations in all the necessary places. So this should work.

[0] https://github.com/anakryiko/btfdump

Quoted 2 lines

Thanks,
Namhyung

Acked-by on v1 1/1
#19 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

In reply to Namhyung Kim
Quoted 8 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

lgtm

jirka

Quoted 64 lines

---
Arnaldo, I think this should go through the cgroup tree since it depends on the earlier change there. I don't think it'd conflict with other perf changes but please let me know if you see any trouble, thanks!

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 29 ++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +28 -1
index 488bd398f01d..4fe61043de04 100644
@@ -43,12 +43,39 @@ struct {
 	__uint(value_size, sizeof(struct bpf_perf_event_value));
 } cgrp_readings SEC(".maps");
 
+/* new kernel cgroup definition */
+struct cgroup___new {
+	int level;
+	struct cgroup *ancestors[];
+} __attribute__((preserve_access_index));
+
+/* old kernel cgroup definition */
+struct cgroup___old {
+	int level;
+	u64 ancestor_ids[];
+} __attribute__((preserve_access_index));
+
 const volatile __u32 num_events = 1;
 const volatile __u32 num_cpus = 1;
 
 int enabled = 0;
 int use_cgroup_v2 = 0;
 
+static inline __u64 get_cgroup_v1_ancestor_id(struct cgroup *cgrp, int level)
+{
+	/* recast pointer to capture new type for compiler */
+	struct cgroup___new *cgrp_new = (void *)cgrp;
+
+	if (bpf_core_field_exists(cgrp_new->ancestors)) {
+		return BPF_CORE_READ(cgrp_new, ancestors[level], kn, id);
+	} else {
+		/* recast pointer to capture old type for compiler */
+		struct cgroup___old *cgrp_old = (void *)cgrp;
+
+		return BPF_CORE_READ(cgrp_old, ancestor_ids[level]);
+	}
+}
+
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
 	struct task_struct *p = (void *)bpf_get_current_task();
@@ -70,7 +97,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 			break;
 
 		// convert cgroup-id to a map index
-		cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
+		cgrp_id = get_cgroup_v1_ancestor_id(cgrp, i);
 		elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
 		if (!elem)
 			continue;
Signature · 1 line
2.37.3.968.ga6b4b080e4-goog
author maintainer
#20 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

In reply to Namhyung Kim
Quoted 8 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Looks like it's acked enough but the patch doesn't apply anymore. Namhyung, can you please refresh the patch? I'll route this through cgroup/for-6.1-fixes unless somebody objects.

Thanks.

Signature · 1 line
tejun
maintainer
#21 lore raw

Re: [PATCH] perf stat: Support old kernels for bperf cgroup counting

Hi Tejun,

Quoted 6 lines · Tejun Heo
Quoted 4 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Quoted 8 lines
Quoted 4 lines

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Looks like it's acked enough but the patch doesn't apply anymore. Namhyung, can you please refresh the patch? I'll route this through cgroup/for-6.1-fixes unless somebody objects.

Sorry about the conflict. There was another change to get the perf cgroup subsys id properly. Will send v2 soon.

Thanks,
Namhyung

maintainer
#22 lore raw

PATCH v2 1/1 perf stat: Support old kernels for bperf cgroup counting

In reply to Tejun Heo

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

---
 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 29 ++++++++++++++++++++-
 1 file changed, 28 insertions(+), 1 deletion(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +28 -1
index 435a87556688..6a438e0102c5 100644
@@ -43,6 +43,18 @@ struct {
 	__uint(value_size, sizeof(struct bpf_perf_event_value));
 } cgrp_readings SEC(".maps");
 
+/* new kernel cgroup definition */
+struct cgroup___new {
+	int level;
+	struct cgroup *ancestors[];
+} __attribute__((preserve_access_index));
+
+/* old kernel cgroup definition */
+struct cgroup___old {
+	int level;
+	u64 ancestor_ids[];
+} __attribute__((preserve_access_index));
+
 const volatile __u32 num_events = 1;
 const volatile __u32 num_cpus = 1;
 
@@ -50,6 +62,21 @@ int enabled = 0;
 int use_cgroup_v2 = 0;
 int perf_subsys_id = -1;
 
+static inline __u64 get_cgroup_v1_ancestor_id(struct cgroup *cgrp, int level)
+{
+	/* recast pointer to capture new type for compiler */
+	struct cgroup___new *cgrp_new = (void *)cgrp;
+
+	if (bpf_core_field_exists(cgrp_new->ancestors)) {
+		return BPF_CORE_READ(cgrp_new, ancestors[level], kn, id);
+	} else {
+		/* recast pointer to capture old type for compiler */
+		struct cgroup___old *cgrp_old = (void *)cgrp;
+
+		return BPF_CORE_READ(cgrp_old, ancestor_ids[level]);
+	}
+}
+
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
 	struct task_struct *p = (void *)bpf_get_current_task();
@@ -77,7 +104,7 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 			break;
 
 		// convert cgroup-id to a map index
-		cgrp_id = BPF_CORE_READ(cgrp, ancestors[i], kn, id);
+		cgrp_id = get_cgroup_v1_ancestor_id(cgrp, i);
 		elem = bpf_map_lookup_elem(&cgrp_idx, &cgrp_id);
 		if (!elem)
 			continue;
Signature · 1 line
2.38.0.rc1.362.ged0d419d3c-goog
author maintainer Applied
#23 lore raw

Re: [PATCH v2] perf stat: Support old kernels for bperf cgroup counting

Quoted 10 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Applied to cgroup/for-6.1-fixes.

Thanks.

Signature · 1 line
tejun
#24 lore raw

Re: [PATCH v2] perf stat: Support old kernels for bperf cgroup counting

Em Tue, Oct 11, 2022 at 06:53:43AM -1000, Tejun Heo escreveu:

Quoted 5 lines · Namhyung Kim
Quoted 3 lines

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Quoted 8 lines
Quoted 6 lines

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Applied to cgroup/for-6.1-fixes.

Hey, I noticed that the perf build is broken for the tools/perf/util/bpf_skel/bperf_cgroup.bpf.c skell, so I tried using b4 on this Namhyung patch, it ended up getting a newer version, by Tejun, that mixes up kernel code and tooling, which, when I tried to apply upstream didn't work.

Please try not to mix up kernel and tools/ changes in the same patch to avoid these issues.

Also when changing tools/perf, please CC me.

I'm now back trying to apply v2 of this patch to see if it fixes my build.

Thanks,

- Arnaldo

#25 lore raw

Re: [PATCH v2] perf stat: Support old kernels for bperf cgroup counting

Em Fri, Oct 14, 2022 at 10:27:40AM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 21 lines

Em Tue, Oct 11, 2022 at 06:53:43AM -1000, Tejun Heo escreveu:

Quoted 5 lines · Namhyung Kim
Quoted 3 lines

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Quoted 8 lines
Quoted 6 lines

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Applied to cgroup/for-6.1-fixes.

Hey, I noticed that the perf build is broken for the tools/perf/util/bpf_skel/bperf_cgroup.bpf.c skell, so I tried using b4 on this Namhyung patch, it ended up getting a newer version, by Tejun, that mixes up kernel code and tooling, which, when I tried to apply upstream didn't work.

Quoted 7 lines

Please try not to mix up kernel and tools/ changes in the same patch to avoid these issues.

Also when changing tools/perf, please CC me.

I'm now back trying to apply v2 of this patch to see if it fixes my build.

Yeah, applying just Namhyung's v2 patch gets perf back building, I'll keep it there while processing the other patches so that I can test them all together.

- Arnaldo

author maintainer
#26 lore raw

Re: [PATCH v2] perf stat: Support old kernels for bperf cgroup counting

In reply to Arnaldo Carvalho de Melo
Quoted 8 lines · Arnaldo Carvalho de Melo

Hey, I noticed that the perf build is broken for the tools/perf/util/bpf_skel/bperf_cgroup.bpf.c skell, so I tried using b4 on this Namhyung patch, it ended up getting a newer version, by Tejun, that mixes up kernel code and tooling, which, when I tried to apply upstream didn't work.

Please try not to mix up kernel and tools/ changes in the same patch to avoid these issues.

I didn't write a newer version of this patch. What are you talking about?

Signature · 1 line
tejun
maintainer Applied
#27 lore raw

Re: [PATCH v2] perf stat: Support old kernels for bperf cgroup counting

Em Fri, Oct 14, 2022 at 06:40:56AM -1000, Tejun Heo escreveu:

Quoted 11 lines · Arnaldo Carvalho de Melo
Quoted 8 lines

Hey, I noticed that the perf build is broken for the tools/perf/util/bpf_skel/bperf_cgroup.bpf.c skell, so I tried using b4 on this Namhyung patch, it ended up getting a newer version, by Tejun, that mixes up kernel code and tooling, which, when I tried to apply upstream didn't work.

Please try not to mix up kernel and tools/ changes in the same patch to avoid these issues.

I didn't write a newer version of this patch. What are you talking about?

So, I saw this message from you in reply to Namhyung's v2 patch:

--------------------------

Date: Tue, 11 Oct 2022 06:53:43 -1000
From: Tejun Heo <[email protected]>
Subject: Re: [PATCH v2] perf stat: Support old kernels for bperf cgroup counting
To: Namhyung Kim <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>, Zefan Li <[email protected]>, Johannes Weiner <[email protected]>, [email protected], Jiri Olsa <[email protected]>, LKML <[email protected]>, [email protected], Song Liu
<[email protected]>, [email protected], Andrii Nakryiko <[email protected]>
Sender: Tejun Heo <[email protected]>
Message-ID: <Y0Wfl88objrECjSo-NiLfg/[email protected]>

Quoted 10 lines · Namhyung Kim

The recent change in the cgroup will break the backward compatiblity in the BPF program. It should support both old and new kernels using BPF CO-RE technique.

Like the task_struct->__state handling in the offcpu analysis, we can check the field name in the cgroup struct.

Applied to cgroup/for-6.1-fixes.

Thanks.

--
tejun
--------------------------

So, I picked the message id, Y0Wfl88objrECjSo-NiLfg/[email protected], and asked b4 to pick the patch:

⬢[acme@toolbox perf]$ b4 am --help | grep -A1 -- -c,
-c, --check-newer-revisions
Check if newer patch revisions exist
⬢[acme@toolbox perf]$

⬢[acme@toolbox perf]$ b4 am -ctsl --cc-trailers Y0Wfl88objrECjSo-NiLfg/[email protected] Grabbing thread from lore.kernel.org/all/Y0Wfl88objrECjSo%40slm.duckdns.org/t.mbox.gz Checking for newer revisions on https://lore.kernel.org/all/
Analyzing 27 messages in the thread
('Acked-by', 'Andrii Nakryiko <[email protected]>', None) Will use the latest revision: v3
You can pick other revisions using the -vN flag
Checking attestation on all messages, may take a moment...
---
✓ [PATCH v3] cgroup: Replace cgroup->ancestor_ids[] with ->ancestors[]
---
✓ Signed: DKIM/gmail.com (From: [email protected])
---
Total patches: 1
---

Base: not specified

       git am ./v3_20220729_tj_cgroup_replace_cgroup_ancestor_ids_with_ancestors.mbx

⬢[acme@toolbox perf]$

Which got me this:

⬢[acme@toolbox perf]$ diffstat ./v3_20220729_tj_cgroup_replace_cgroup_ancestor_ids_with_ancestors.mbx

 include/linux/cgroup-defs.h                 |   16 ++++++++++------
 include/linux/cgroup.h                      |    8 +++-----
 kernel/cgroup/cgroup.c                      |    7 +++----
 net/netfilter/nft_socket.c                  |    9 +++++----
 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c |    2 +-
 5 files changed, 22 insertions(+), 20 deletions(-)

⬢[acme@toolbox perf]$

⬢[acme@toolbox perf]$ grep From: ./v3_20220729_tj_cgroup_replace_cgroup_ancestor_ids_with_ancestors.mbx
From: Tejun Heo <[email protected]>
⬢[acme@toolbox perf]$

That mixes kernel and tools bits and touches tools/perf/util/bpf_skel/bperf_cgroup.bpf.c, hence my request to add me to the CC list for patches touching tools/perf/.

My assumption that it was a new patch was because b4 somehow got to v3_20220729_tj_cgroup_replace_cgroup_ancestor_ids_with_ancestors, which has v3 and touches the tools cgroup bpf skel.

So it seems b4 is confused somehow.

Hope this clarifies.

- Arnaldo