Merge branch 'sleepable uprobe support'

Delyan Kratunov says:

====================

This series implements support for sleepable uprobe programs.
Key work is in patches 2 and 3, the rest is plumbing and tests.

The main observation is that the only obstacle in the way of sleepable uprobe
programs is not the uprobe infrastructure, which already runs in a user-like
context, but the rcu usage around bpf_prog_array.

Details are in patch 2 but the tl;dr is that we chain trace_tasks and normal rcu
grace periods when releasing to array to accommodate users of either rcu type.
This introduces latency for non-sleepable users (kprobe, tp) but that's deemed
acceptable, given recent benchmarks by Andrii [1]. We're a couple of orders of
magnitude under the rate of bpf_prog_array churn that would raise flags (~1MM/s per Paul).

  [1]: https://lore.kernel.org/bpf/CAEf4BzbpjN6ca7D9KOTiFPOoBYkciYvTz0UJNp5c-_3ptm=Mrg@mail.gmail.com/

v3 -> v4:
 * Fix kdoc and inline issues
 * Rebase

v2 -> v3:
 * Inline uprobe_call_bpf into trace_uprobe.c, it's just a bpf_prog_run_array_sleepable call now.
 * Do not disable preemption for uprobe non-sleepable programs.
 * Add acks.

v1 -> v2:
 * Fix lockdep annotations in bpf_prog_run_array_sleepable
 * Chain rcu grace periods only for perf_event-attached programs. This limits
   the additional latency on the free path to use cases where we know it won't
   be a problem.
 * Add tests calling helpers only available in sleepable programs.
 * Remove kprobe.s support from libbpf.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

2 files changed, 88 insertions, 34 deletions

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 8e6092d0ea95..f3e88afdaffe 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -5,6 +5,7 @@
 #define _LINUX_BPF_H 1
 
 #include <uapi/linux/bpf.h>
+#include <uapi/linux/filter.h>
 
 #include <linux/workqueue.h>
 #include <linux/file.h>
@@ -22,8 +23,10 @@
 #include <linux/sched/mm.h>
 #include <linux/slab.h>
 #include <linux/percpu-refcount.h>
+#include <linux/stddef.h>
 #include <linux/bpfptr.h>
 #include <linux/btf.h>
+#include <linux/rcupdate_trace.h>
 
 struct bpf_verifier_env;
 struct bpf_verifier_log;
@@ -1084,6 +1087,40 @@ struct bpf_prog_aux {
 	};
 };
 
+struct bpf_prog {
+	u16			pages;		/* Number of allocated pages */
+	u16			jited:1,	/* Is our filter JIT'ed? */
+				jit_requested:1,/* archs need to JIT the prog */
+				gpl_compatible:1, /* Is filter GPL compatible? */
+				cb_access:1,	/* Is control block accessed? */
+				dst_needed:1,	/* Do we need dst entry? */
+				blinding_requested:1, /* needs constant blinding */
+				blinded:1,	/* Was blinded */
+				is_func:1,	/* program is a bpf function */
+				kprobe_override:1, /* Do we override a kprobe? */
+				has_callchain_buf:1, /* callchain buffer allocated? */
+				enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
+				call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */
+				call_get_func_ip:1, /* Do we call get_func_ip() */
+				tstamp_type_access:1; /* Accessed __sk_buff->tstamp_type */
+	enum bpf_prog_type	type;		/* Type of BPF program */
+	enum bpf_attach_type	expected_attach_type; /* For some prog types */
+	u32			len;		/* Number of filter blocks */
+	u32			jited_len;	/* Size of jited insns in bytes */
+	u8			tag[BPF_TAG_SIZE];
+	struct bpf_prog_stats __percpu *stats;
+	int __percpu		*active;
+	unsigned int		(*bpf_func)(const void *ctx,
+					    const struct bpf_insn *insn);
+	struct bpf_prog_aux	*aux;		/* Auxiliary fields */
+	struct sock_fprog_kern	*orig_prog;	/* Original BPF program */
+	/* Instructions for interpreter */
+	union {
+		DECLARE_FLEX_ARRAY(struct sock_filter, insns);
+		DECLARE_FLEX_ARRAY(struct bpf_insn, insnsi);
+	};
+};
+
 struct bpf_array_aux {
 	/* Programs with direct jumps into programs part of this array. */
 	struct list_head poke_progs;
@@ -1336,6 +1373,8 @@ extern struct bpf_empty_prog_array bpf_empty_prog_array;
 
 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags);
 void bpf_prog_array_free(struct bpf_prog_array *progs);
+/* Use when traversal over the bpf_prog_array uses tasks_trace rcu */
+void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs);
 int bpf_prog_array_length(struct bpf_prog_array *progs);
 bool bpf_prog_array_is_empty(struct bpf_prog_array *array);
 int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs,
@@ -1427,6 +1466,55 @@ bpf_prog_run_array(const struct bpf_prog_array *array,
 	return ret;
 }
 
+/* Notes on RCU design for bpf_prog_arrays containing sleepable programs:
+ *
+ * We use the tasks_trace rcu flavor read section to protect the bpf_prog_array
+ * overall. As a result, we must use the bpf_prog_array_free_sleepable
+ * in order to use the tasks_trace rcu grace period.
+ *
+ * When a non-sleepable program is inside the array, we take the rcu read
+ * section and disable preemption for that program alone, so it can access
+ * rcu-protected dynamically sized maps.
+ */
+static __always_inline u32
+bpf_prog_run_array_sleepable(const struct bpf_prog_array __rcu *array_rcu,
+			     const void *ctx, bpf_prog_run_fn run_prog)
+{
+	const struct bpf_prog_array_item *item;
+	const struct bpf_prog *prog;
+	const struct bpf_prog_array *array;
+	struct bpf_run_ctx *old_run_ctx;
+	struct bpf_trace_run_ctx run_ctx;
+	u32 ret = 1;
+
+	might_fault();
+
+	rcu_read_lock_trace();
+	migrate_disable();
+
+	array = rcu_dereference_check(array_rcu, rcu_read_lock_trace_held());
+	if (unlikely(!array))
+		goto out;
+	old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
+	item = &array->items[0];
+	while ((prog = READ_ONCE(item->prog))) {
+		if (!prog->aux->sleepable)
+			rcu_read_lock();
+
+		run_ctx.bpf_cookie = item->bpf_cookie;
+		ret &= run_prog(prog, ctx);
+		item++;
+
+		if (!prog->aux->sleepable)
+			rcu_read_unlock();
+	}
+	bpf_reset_run_ctx(old_run_ctx);
+out:
+	migrate_enable();
+	rcu_read_unlock_trace();
+	return ret;
+}
+
 #ifdef CONFIG_BPF_SYSCALL
 DECLARE_PER_CPU(int, bpf_prog_active);
 extern struct mutex bpf_stats_enabled_mutex;
diff --git a/include/linux/filter.h b/include/linux/filter.h
index ed0c0ff42ad5..d0cbb31b1b4d 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -559,40 +559,6 @@ struct bpf_prog_stats {
 	struct u64_stats_sync syncp;
 } __aligned(2 * sizeof(u64));
 
-struct bpf_prog {
-	u16			pages;		/* Number of allocated pages */
-	u16			jited:1,	/* Is our filter JIT'ed? */
-				jit_requested:1,/* archs need to JIT the prog */
-				gpl_compatible:1, /* Is filter GPL compatible? */
-				cb_access:1,	/* Is control block accessed? */
-				dst_needed:1,	/* Do we need dst entry? */
-				blinding_requested:1, /* needs constant blinding */
-				blinded:1,	/* Was blinded */
-				is_func:1,	/* program is a bpf function */
-				kprobe_override:1, /* Do we override a kprobe? */
-				has_callchain_buf:1, /* callchain buffer allocated? */
-				enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
-				call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */
-				call_get_func_ip:1, /* Do we call get_func_ip() */
-				tstamp_type_access:1; /* Accessed __sk_buff->tstamp_type */
-	enum bpf_prog_type	type;		/* Type of BPF program */
-	enum bpf_attach_type	expected_attach_type; /* For some prog types */
-	u32			len;		/* Number of filter blocks */
-	u32			jited_len;	/* Size of jited insns in bytes */
-	u8			tag[BPF_TAG_SIZE];
-	struct bpf_prog_stats __percpu *stats;
-	int __percpu		*active;
-	unsigned int		(*bpf_func)(const void *ctx,
-					    const struct bpf_insn *insn);
-	struct bpf_prog_aux	*aux;		/* Auxiliary fields */
-	struct sock_fprog_kern	*orig_prog;	/* Original BPF program */
-	/* Instructions for interpreter */
-	union {
-		DECLARE_FLEX_ARRAY(struct sock_filter, insns);
-		DECLARE_FLEX_ARRAY(struct bpf_insn, insnsi);
-	};
-};
-
 struct sk_filter {
 	refcount_t	refcnt;
 	struct rcu_head	rcu;