FAILED: patch "[PATCH] bpf: support deferring bpf_link dealloc to after RCU grace" failed to apply to 6.6-stable tree
Subsystems (6)
[email protected]
maintainerThe patch below does not apply to the 6.6-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.
To reproduce the conflict and resubmit, you may use the following commands:
git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.6.y
git checkout FETCH_HEAD
git cherry-pick -x 1a80dbcb2dbaf6e4c216e62e30fa7d3daa8001ce
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2024040548-lid-mahogany-fd86@gregkh' --subject-prefix 'PATCH 6.6.y' HEAD^..
Possible dependencies:
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
From 1a80dbcb2dbaf6e4c216e62e30fa7d3daa8001ce Mon Sep 17 00:00:00 2001
From: Andrii Nakryiko <[email protected]>
Date: Wed, 27 Mar 2024 22:24:26 -0700
Subject: [PATCH] bpf: support deferring bpf_link dealloc to after RCU grace
period
BPF link for some program types is passed as a "context" which can be used by those BPF programs to look up additional information. E.g., for multi-kprobes and multi-uprobes, link is used to fetch BPF cookie values.
Because of this runtime dependency, when bpf_link refcnt drops to zero there could still be active BPF programs running accessing link data.
This patch adds generic support to defer bpf_link dealloc callback to after RCU GP, if requested. This is done by exposing two different deallocation callbacks, one synchronous and one deferred. If deferred one is provided, bpf_link_free() will schedule dealloc_deferred() callback to happen after RCU GP.
BPF is using two flavors of RCU: "classic" non-sleepable one and RCU tasks trace one. The latter is used when sleepable BPF programs are used. bpf_link_free() accommodates that by checking underlying BPF program's sleepable flag, and goes either through normal RCU GP only for non-sleepable, or through RCU tasks trace GP *and* then normal RCU GP (taking into account rcu_trace_implies_rcu_gp() optimization), if BPF program is sleepable.
We use this for multi-kprobe and multi-uprobe links, which dereference link during program run. We also preventively switch raw_tp link to use deferred dealloc callback, as upcoming changes in bpf-next tree expose raw_tp link data (specifically, cookie value) to BPF program at runtime as well.
include/linux/bpf.h +15 -1
@@ -1574,12 +1574,26 @@ struct bpf_link {enum bpf_link_type type;const struct bpf_link_ops *ops;struct bpf_prog *prog; -struct work_struct work; +/* rcu is used before freeing, work can be used to schedule that +* RCU-based freeing before that, so they never overlap +*/ +union { +struct rcu_head rcu; +struct work_struct work; + }; };struct bpf_link_ops {void (*release)(struct bpf_link *link); +/* deallocate link resources callback, called without RCU grace period +* waiting +*/ void (*dealloc)(struct bpf_link *link); +/* deallocate link resources callback, called after RCU grace period; +* if underlying BPF program is sleepable we go through tasks trace +* RCU GP and then "classic" RCU GP +*/ +void (*dealloc_deferred)(struct bpf_link *link);int (*detach)(struct bpf_link *link);int (*update_prog)(struct bpf_link *link,struct bpf_prog *new_prog,struct bpf_prog *old_prog);
kernel/bpf/syscall.c +32 -3
@@ -3024,17 +3024,46 @@ void bpf_link_inc(struct bpf_link *link)atomic64_inc (&link->refcnt); } +static void bpf_link_defer_dealloc_rcu_gp (struct rcu_head *rcu) +{ + struct bpf_link *link = container_of(rcu, struct bpf_link, rcu); + +/* free bpf_link and its containing memory */ + link->ops->dealloc_deferred (link); +} + +static void bpf_link_defer_dealloc_mult_rcu_gp (struct rcu_head *rcu) +{ +if (rcu_trace_implies_rcu_gp ()) +bpf_link_defer_dealloc_rcu_gp (rcu); +else +call_rcu (rcu, bpf_link_defer_dealloc_rcu_gp); +} +/* bpf_link_free is guaranteed to be called from process context */ static void bpf_link_free (struct bpf_link *link) { +bool sleepable = false; +bpf_link_free_id (link->id);if (link->prog) { + sleepable = link->prog->sleepable;/* detach BPF program, clean up used resources */ link->ops->release (link);bpf_prog_put (link->prog); } -/* free bpf_link and its containing memory */ - link->ops->dealloc (link); +if (link->ops->dealloc_deferred) { +/* schedule BPF link deallocation; if underlying BPF program +* is sleepable, we need to first wait for RCU tasks trace +* sync, then go through "classic" RCU grace period +*/ +if (sleepable) +call_rcu_tasks_trace (&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp); +else +call_rcu (&link->rcu, bpf_link_defer_dealloc_rcu_gp); + } +if (link->ops->dealloc) + link->ops->dealloc (link); }static void bpf_link_put_deferred (struct work_struct *work)
@@ -3544,7 +3573,7 @@ static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,static const struct bpf_link_ops bpf_raw_tp_link_lops = { .release = bpf_raw_tp_link_release, - .dealloc = bpf_raw_tp_link_dealloc, + .dealloc_deferred = bpf_raw_tp_link_dealloc, .show_fdinfo = bpf_raw_tp_link_show_fdinfo, .fill_link_info = bpf_raw_tp_link_fill_link_info, };
kernel/trace/bpf_trace.c +2 -2
@@ -2728,7 +2728,7 @@ static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,static const struct bpf_link_ops bpf_kprobe_multi_link_lops = { .release = bpf_kprobe_multi_link_release, - .dealloc = bpf_kprobe_multi_link_dealloc, + .dealloc_deferred = bpf_kprobe_multi_link_dealloc, .fill_link_info = bpf_kprobe_multi_link_fill_link_info, };
@@ -3242,7 +3242,7 @@ static int bpf_uprobe_multi_link_fill_link_info(const struct bpf_link *link,static const struct bpf_link_ops bpf_uprobe_multi_link_lops = { .release = bpf_uprobe_multi_link_release, - .dealloc = bpf_uprobe_multi_link_dealloc, + .dealloc_deferred = bpf_uprobe_multi_link_dealloc, .fill_link_info = bpf_uprobe_multi_link_fill_link_info, };
Andrii Nakryiko
maintainerPATCH 1/2 bpf: put uprobe link's path and task in release callback
There is no need to delay putting either path or task to deallocation step. It can be done right after bpf_uprobe_unregister. Between release and dealloc, there could be still some running BPF programs, but they don't access either task or path, only data in link->uprobes, so it is safe to do.
On the other hand, doing path_put() in dealloc callback makes this dealloc sleepable because path_put() itself might sleep. Which is problematic due to the need to call uprobe's dealloc through call_rcu(), which is what is done in the next bug fix patch. So solve the problem by releasing these resources early.
--- kernel/trace/bpf_trace.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
kernel/trace/bpf_trace.c +3 -3
@@ -3065,6 +3065,9 @@ static void bpf_uprobe_multi_link_release(struct bpf_link *link) umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);bpf_uprobe_unregister (&umulti_link->path, umulti_link->uprobes, umulti_link->cnt); +if (umulti_link->task) +put_task_struct (umulti_link->task); +path_put (&umulti_link->path); }static void bpf_uprobe_multi_link_dealloc (struct bpf_link *link)
@@ -3072,9 +3075,6 @@ static void bpf_uprobe_multi_link_dealloc(struct bpf_link *link)struct bpf_uprobe_multi_link *umulti_link; umulti_link = container_of(link, struct bpf_uprobe_multi_link, link); -if (umulti_link->task) -put_task_struct (umulti_link->task); -path_put (&umulti_link->path);kvfree (umulti_link->uprobes);kfree (umulti_link); }
Signature · 1 line
2.43.0
Andrii Nakryiko
maintainerPATCH 2/2 bpf: support deferring bpf_link dealloc to after RCU grace period
BPF link for some program types is passed as a "context" which can be used by those BPF programs to look up additional information. E.g., for multi-kprobes and multi-uprobes, link is used to fetch BPF cookie values.
Because of this runtime dependency, when bpf_link refcnt drops to zero there could still be active BPF programs running accessing link data.
This patch adds generic support to defer bpf_link dealloc callback to after RCU GP, if requested. This is done by exposing two different deallocation callbacks, one synchronous and one deferred. If deferred one is provided, bpf_link_free() will schedule dealloc_deferred() callback to happen after RCU GP.
BPF is using two flavors of RCU: "classic" non-sleepable one and RCU tasks trace one. The latter is used when sleepable BPF programs are used. bpf_link_free() accommodates that by checking underlying BPF program's sleepable flag, and goes either through normal RCU GP only for non-sleepable, or through RCU tasks trace GP *and* then normal RCU GP (taking into account rcu_trace_implies_rcu_gp() optimization), if BPF program is sleepable.
We use this for multi-kprobe and multi-uprobe links, which dereference link during program run. We also preventively switch raw_tp link to use deferred dealloc callback, as upcoming changes in bpf-next tree expose raw_tp link data (specifically, cookie value) to BPF program at runtime as well.
--- include/linux/bpf.h | 16 +++++++++++++++- kernel/bpf/syscall.c | 35 ++++++++++++++++++++++++++++++++--- kernel/trace/bpf_trace.c | 4 ++-- 3 files changed, 49 insertions(+), 6 deletions(-)
include/linux/bpf.h +15 -1
@@ -1524,12 +1524,26 @@ struct bpf_link {enum bpf_link_type type;const struct bpf_link_ops *ops;struct bpf_prog *prog; -struct work_struct work; +/* rcu is used before freeing, work can be used to schedule that +* RCU-based freeing before that, so they never overlap +*/ +union { +struct rcu_head rcu; +struct work_struct work; + }; };struct bpf_link_ops {void (*release)(struct bpf_link *link); +/* deallocate link resources callback, called without RCU grace period +* waiting +*/ void (*dealloc)(struct bpf_link *link); +/* deallocate link resources callback, called after RCU grace period; +* if underlying BPF program is sleepable we go through tasks trace +* RCU GP and then "classic" RCU GP +*/ +void (*dealloc_deferred)(struct bpf_link *link);int (*detach)(struct bpf_link *link);int (*update_prog)(struct bpf_link *link,struct bpf_prog *new_prog,struct bpf_prog *old_prog);
kernel/bpf/syscall.c +32 -3
@@ -2866,17 +2866,46 @@ void bpf_link_inc(struct bpf_link *link)atomic64_inc (&link->refcnt); } +static void bpf_link_defer_dealloc_rcu_gp (struct rcu_head *rcu) +{ + struct bpf_link *link = container_of(rcu, struct bpf_link, rcu); + +/* free bpf_link and its containing memory */ + link->ops->dealloc_deferred (link); +} + +static void bpf_link_defer_dealloc_mult_rcu_gp (struct rcu_head *rcu) +{ +if (rcu_trace_implies_rcu_gp ()) +bpf_link_defer_dealloc_rcu_gp (rcu); +else +call_rcu (rcu, bpf_link_defer_dealloc_rcu_gp); +} +/* bpf_link_free is guaranteed to be called from process context */ static void bpf_link_free (struct bpf_link *link) { +bool sleepable = false; +bpf_link_free_id (link->id);if (link->prog) { + sleepable = link->prog->aux->sleepable;/* detach BPF program, clean up used resources */ link->ops->release (link);bpf_prog_put (link->prog); } -/* free bpf_link and its containing memory */ - link->ops->dealloc (link); +if (link->ops->dealloc_deferred) { +/* schedule BPF link deallocation; if underlying BPF program +* is sleepable, we need to first wait for RCU tasks trace +* sync, then go through "classic" RCU grace period +*/ +if (sleepable) +call_rcu_tasks_trace (&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp); +else +call_rcu (&link->rcu, bpf_link_defer_dealloc_rcu_gp); + } +if (link->ops->dealloc) + link->ops->dealloc (link); }static void bpf_link_put_deferred (struct work_struct *work)
@@ -3381,7 +3410,7 @@ static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,static const struct bpf_link_ops bpf_raw_tp_link_lops = { .release = bpf_raw_tp_link_release, - .dealloc = bpf_raw_tp_link_dealloc, + .dealloc_deferred = bpf_raw_tp_link_dealloc, .show_fdinfo = bpf_raw_tp_link_show_fdinfo, .fill_link_info = bpf_raw_tp_link_fill_link_info, };
kernel/trace/bpf_trace.c +2 -2
@@ -2639,7 +2639,7 @@ static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,static const struct bpf_link_ops bpf_kprobe_multi_link_lops = { .release = bpf_kprobe_multi_link_release, - .dealloc = bpf_kprobe_multi_link_dealloc, + .dealloc_deferred = bpf_kprobe_multi_link_dealloc, .fill_link_info = bpf_kprobe_multi_link_fill_link_info, };
@@ -3081,7 +3081,7 @@ static void bpf_uprobe_multi_link_dealloc(struct bpf_link *link)static const struct bpf_link_ops bpf_uprobe_multi_link_lops = { .release = bpf_uprobe_multi_link_release, - .dealloc = bpf_uprobe_multi_link_dealloc, + .dealloc_deferred = bpf_uprobe_multi_link_dealloc, }; static int uprobe_prog_run(struct bpf_uprobe *uprobe,
Signature · 1 line
2.43.0
Greg KH
maintainerRe: [PATCH 6.6.y 1/2] bpf: put uprobe link's path and task in release callback
In reply to Andrii NakryikoQuoted 12 lines · Andrii Nakryiko
There is no need to delay putting either path or task to deallocation step. It can be done right after
bpf_uprobe_unregister. Between release and dealloc, there could be still some running BPF programs, but they don't access either task or path, only data inlink->uprobes, so it is safe to do.On the other hand, doing
path_put()in dealloc callback makes this dealloc sleepable becausepath_put()itself might sleep. Which is problematic due to the need to call uprobe's dealloc throughcall_rcu(), which is what is done in the next bug fix patch. So solve the problem by releasing these resources early.
Quoted 7 lines
Signed-off-by Andrii Nakryiko <[email protected]>Signed-off-by Alexei Starovoitov <[email protected]>(cherry picked from commit e9c856cabefb71d47b2eeb197f72c9c88e9b45b0)--- kernel/trace/bpf_trace.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
All now queued up, thanks.
greg k-h