Skip to contentSkip to conversation

Introduce bpf_wq

PATCH Applied Benjamin Tissoires 4 participants 28 messages Apr 20, 2024 – Apr 25, 2024

Subsystems (6)
Networking Kselftest BPF Arm STM32 +2 more subsystems
lore
Share & tools
lore
#0 lore raw

v2 cover Introduce bpf_wq

This is a followup of sleepable bpf_timer[0].

When discussing sleepable bpf_timer, it was thought that we should give a try to bpf_wq, as the 2 APIs are similar but distinct enough to justify a new one.

So here it is.

I tried to keep as much as possible common code in kernel/bpf/helpers.c but I couldn't get away with code duplication in kernel/bpf/verifier.c.

This series introduces a basic bpf_wq support:
- creation is supported
- assignment is supported
- running a simple bpf_wq is also supported.

We will probably need to extend the API further with:
- a full delayed_work API (can be piggy backed on top with a correct flag)
- bpf_wq_cancel() <- apparently not, this is shooting ourself in the foot
- bpf_wq_cancel_sync() (for sleepable programs)
- documentation

---

For reference, the use cases I have in mind:

---

Basically, I need to be able to defer a HID-BPF program for the following reasons (from the aforementioned patch):
1. defer an event:
Sometimes we receive an out of proximity event, but the device can not be trusted enough, and we need to ensure that we won't receive another one in the following n milliseconds. So we need to wait those n milliseconds, and eventually re-inject that event in the stack.

2. inject new events in reaction to one given event:
We might want to transform one given event into several. This is the case for macro keys where a single key press is supposed to send a sequence of key presses. But this could also be used to patch a faulty behavior, if a device forgets to send a release event.

3. communicate with the device in reaction to one event:
We might want to communicate back to the device after a given event. For example a device might send us an event saying that it came back from sleeping state and needs to be re-initialized.

Currently we can achieve that by keeping a userspace program around, raise a bpf event, and let that userspace program inject the events and commands.
However, we are just keeping that program alive as a daemon for just scheduling commands. There is no logic in it, so it doesn't really justify an actual userspace wakeup. So a kernel workqueue seems simpler to handle.

bpf_timers are currently running in a soft IRQ context, this patch series implements a sleppable context for them.

Cheers,
Benjamin

To: Alexei Starovoitov <[email protected]>
To: Daniel Borkmann <[email protected]>
To: Andrii Nakryiko <[email protected]>
To: Martin KaFai Lau <[email protected]>
To: Eduard Zingerman <[email protected]>
To: Song Liu <[email protected]>
To: Yonghong Song <[email protected]>
To: John Fastabend <[email protected]>
To: KP Singh <[email protected]>
To: Stanislav Fomichev <[email protected]>
To: Hao Luo <[email protected]>
To: Jiri Olsa <[email protected]>
To: Mykola Lysenko <[email protected]>
To: Shuah Khan <[email protected]>

Cc:  <[email protected]>
Cc:  <[email protected]>
Cc:  <[email protected]>

[0] https://lore.kernel.org/all/[email protected]/

---
Changes in v2:
- took previous review into account
- mainly dropped BPF_F_WQ_SLEEPABLE
- Link to v1: https://lore.kernel.org/r/[email protected]

---
Benjamin Tissoires (16):
bpf: make timer data struct more generic
bpf: replace bpf_timer_init with a generic helper
bpf: replace bpf_timer_set_callback with a generic helper
bpf: replace bpf_timer_cancel_and_free with a generic helper
bpf: add support for bpf_wq user type
tools: sync include/uapi/linux/bpf.h
bpf: verifier: bail out if the argument is not a map
bpf: add support for KF_ARG_PTR_TO_WORKQUEUE
bpf: allow struct bpf_wq to be embedded in arraymaps and hashmaps selftests/bpf: add bpf_wq tests
bpf: wq: add bpf_wq_init
selftests/bpf: wq: add bpf_wq_init() checks
bpf: wq: add bpf_wq_set_callback_impl
selftests/bpf: add checks for bpf_wq_set_callback()
bpf: add bpf_wq_start
selftests/bpf: wq: add bpf_wq_start() checks

 include/linux/bpf.h                                |  13 +-
 include/linux/bpf_verifier.h                       |   1 +
 include/uapi/linux/bpf.h                           |   4 +
 kernel/bpf/arraymap.c                              |  18 +-
 kernel/bpf/btf.c                                   |  17 +
 kernel/bpf/hashtab.c                               |  55 +++-
 kernel/bpf/helpers.c                               | 348 ++++++++++++++++-----
 kernel/bpf/syscall.c                               |  15 +-
 kernel/bpf/verifier.c                              | 154 ++++++++-
 tools/include/uapi/linux/bpf.h                     |   4 +
 tools/testing/selftests/bpf/bpf_experimental.h     |   7 +
 .../selftests/bpf/bpf_testmod/bpf_testmod.c        |   5 +
 .../selftests/bpf/bpf_testmod/bpf_testmod_kfunc.h  |   1 +
 tools/testing/selftests/bpf/prog_tests/wq.c        |  41 +++
 tools/testing/selftests/bpf/progs/wq.c             | 186 +++++++++++
 tools/testing/selftests/bpf/progs/wq_failures.c    | 144 +++++++++
 16 files changed, 910 insertions(+), 103 deletions(-)

---
base-commit: ffa6b26b4d8a0520b78636ca9373ab842cb3b1a8
change-id: 20240411-bpf_wq-fe24e8d24f5e

Best regards,

Signature · 1 line
Benjamin Tissoires <[email protected]>
#1 lore raw

PATCH v2 1/16 bpf: make timer data struct more generic

To be able to add workqueues and reuse most of the timer code, we need to make bpf_hrtimer more generic.

There is no code change except that the new struct gets a new u64 flags attribute. We are still below 2 cache lines, so this shouldn't impact the current running codes.

The ordering is also changed. Everything related to async callback is now on top of bpf_hrtimer.

---

no changes in v2

---
 kernel/bpf/helpers.c | 71 ++++++++++++++++++++++++++++------------------------
 1 file changed, 38 insertions(+), 33 deletions(-)
kernel/bpf/helpers.c +38 -33
index 8cde717137bd..5a069c70b5e6 100644
@@ -1079,11 +1079,20 @@ const struct bpf_func_proto bpf_snprintf_proto = {
 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
 };
 
+struct bpf_async_cb {
+	struct bpf_map *map;
+	struct bpf_prog *prog;
+	void __rcu *callback_fn;
+	void *value;
+	struct rcu_head rcu;
+	u64 flags;
+};
+
 /* BPF map elements can contain 'struct bpf_timer'.
  * Such map owns all of its BPF timers.
  * 'struct bpf_timer' is allocated as part of map element allocation
  * and it's zero initialized.
- * That space is used to keep 'struct bpf_timer_kern'.
+ * That space is used to keep 'struct bpf_async_kern'.
  * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
  * remembers 'struct bpf_map *' pointer it's part of.
  * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
@@ -1096,16 +1105,12 @@ const struct bpf_func_proto bpf_snprintf_proto = {
  * freeing the timers when inner map is replaced or deleted by user space.
  */
 struct bpf_hrtimer {
+	struct bpf_async_cb cb;
 	struct hrtimer timer;
-	struct bpf_map *map;
-	struct bpf_prog *prog;
-	void __rcu *callback_fn;
-	void *value;
-	struct rcu_head rcu;
 };
 
 /* the actual struct hidden inside uapi struct bpf_timer */
-struct bpf_timer_kern {
+struct bpf_async_kern {
 	struct bpf_hrtimer *timer;
 	/* bpf_spin_lock is used here instead of spinlock_t to make
 	 * sure that it always fits into space reserved by struct bpf_timer
@@ -1119,14 +1124,14 @@ static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
 {
 	struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
-	struct bpf_map *map = t->map;
-	void *value = t->value;
+	struct bpf_map *map = t->cb.map;
+	void *value = t->cb.value;
 	bpf_callback_t callback_fn;
 	void *key;
 	u32 idx;
 
 	BTF_TYPE_EMIT(struct bpf_timer);
-	callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
+	callback_fn = rcu_dereference_check(t->cb.callback_fn, rcu_read_lock_bh_held());
 	if (!callback_fn)
 		goto out;
 
@@ -1155,7 +1160,7 @@ static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
 	return HRTIMER_NORESTART;
 }
 
-BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
+BPF_CALL_3(bpf_timer_init, struct bpf_async_kern *, timer, struct bpf_map *, map,
 	   u64, flags)
 {
 	clockid_t clockid = flags & (MAX_CLOCKS - 1);
@@ -1163,8 +1168,8 @@ BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map
 	int ret = 0;
 
 	BUILD_BUG_ON(MAX_CLOCKS != 16);
-	BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
-	BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
+	BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_timer));
+	BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_timer));
 
 	if (in_nmi())
 		return -EOPNOTSUPP;
@@ -1187,10 +1192,10 @@ BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map
 		ret = -ENOMEM;
 		goto out;
 	}
-	t->value = (void *)timer - map->record->timer_off;
-	t->map = map;
-	t->prog = NULL;
-	rcu_assign_pointer(t->callback_fn, NULL);
+	t->cb.value = (void *)timer - map->record->timer_off;
+	t->cb.map = map;
+	t->cb.prog = NULL;
+	rcu_assign_pointer(t->cb.callback_fn, NULL);
 	hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
 	t->timer.function = bpf_timer_cb;
 	WRITE_ONCE(timer->timer, t);
@@ -1222,7 +1227,7 @@ static const struct bpf_func_proto bpf_timer_init_proto = {
 	.arg3_type	= ARG_ANYTHING,
 };
 
-BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
+BPF_CALL_3(bpf_timer_set_callback, struct bpf_async_kern *, timer, void *, callback_fn,
 	   struct bpf_prog_aux *, aux)
 {
 	struct bpf_prog *prev, *prog = aux->prog;
@@ -1237,7 +1242,7 @@ BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callb
 		ret = -EINVAL;
 		goto out;
 	}
-	if (!atomic64_read(&t->map->usercnt)) {
+	if (!atomic64_read(&t->cb.map->usercnt)) {
 		/* maps with timers must be either held by user space
 		 * or pinned in bpffs. Otherwise timer might still be
 		 * running even when bpf prog is detached and user space
@@ -1246,7 +1251,7 @@ BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callb
 		ret = -EPERM;
 		goto out;
 	}
-	prev = t->prog;
+	prev = t->cb.prog;
 	if (prev != prog) {
 		/* Bump prog refcnt once. Every bpf_timer_set_callback()
 		 * can pick different callback_fn-s within the same prog.
@@ -1259,9 +1264,9 @@ BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callb
 		if (prev)
 			/* Drop prev prog refcnt when swapping with new prog */
 			bpf_prog_put(prev);
-		t->prog = prog;
+		t->cb.prog = prog;
 	}
-	rcu_assign_pointer(t->callback_fn, callback_fn);
+	rcu_assign_pointer(t->cb.callback_fn, callback_fn);
 out:
 	__bpf_spin_unlock_irqrestore(&timer->lock);
 	return ret;
@@ -1275,7 +1280,7 @@ static const struct bpf_func_proto bpf_timer_set_callback_proto = {
 	.arg2_type	= ARG_PTR_TO_FUNC,
 };
 
-BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
+BPF_CALL_3(bpf_timer_start, struct bpf_async_kern *, timer, u64, nsecs, u64, flags)
 {
 	struct bpf_hrtimer *t;
 	int ret = 0;
@@ -1287,7 +1292,7 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
 		return -EINVAL;
 	__bpf_spin_lock_irqsave(&timer->lock);
 	t = timer->timer;
-	if (!t || !t->prog) {
+	if (!t || !t->cb.prog) {
 		ret = -EINVAL;
 		goto out;
 	}
@@ -1315,18 +1320,18 @@ static const struct bpf_func_proto bpf_timer_start_proto = {
 	.arg3_type	= ARG_ANYTHING,
 };
 
-static void drop_prog_refcnt(struct bpf_hrtimer *t)
+static void drop_prog_refcnt(struct bpf_async_cb *async)
 {
-	struct bpf_prog *prog = t->prog;
+	struct bpf_prog *prog = async->prog;
 
 	if (prog) {
 		bpf_prog_put(prog);
-		t->prog = NULL;
-		rcu_assign_pointer(t->callback_fn, NULL);
+		async->prog = NULL;
+		rcu_assign_pointer(async->callback_fn, NULL);
 	}
 }
 
-BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
+BPF_CALL_1(bpf_timer_cancel, struct bpf_async_kern *, timer)
 {
 	struct bpf_hrtimer *t;
 	int ret = 0;
@@ -1348,7 +1353,7 @@ BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
 		ret = -EDEADLK;
 		goto out;
 	}
-	drop_prog_refcnt(t);
+	drop_prog_refcnt(&t->cb);
 out:
 	__bpf_spin_unlock_irqrestore(&timer->lock);
 	/* Cancel the timer and wait for associated callback to finish
@@ -1371,7 +1376,7 @@ static const struct bpf_func_proto bpf_timer_cancel_proto = {
  */
 void bpf_timer_cancel_and_free(void *val)
 {
-	struct bpf_timer_kern *timer = val;
+	struct bpf_async_kern *timer = val;
 	struct bpf_hrtimer *t;
 
 	/* Performance optimization: read timer->timer without lock first. */
@@ -1383,7 +1388,7 @@ void bpf_timer_cancel_and_free(void *val)
 	t = timer->timer;
 	if (!t)
 		goto out;
-	drop_prog_refcnt(t);
+	drop_prog_refcnt(&t->cb);
 	/* The subsequent bpf_timer_start/cancel() helpers won't be able to use
 	 * this timer, since it won't be initialized.
 	 */
@@ -1410,7 +1415,7 @@ void bpf_timer_cancel_and_free(void *val)
 	 */
 	if (this_cpu_read(hrtimer_running) != t)
 		hrtimer_cancel(&t->timer);
-	kfree_rcu(t, rcu);
+	kfree_rcu(t, cb.rcu);
 }
 
 BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
Signature · 1 line
2.44.0
#2 lore raw

PATCH v2 2/16 bpf: replace bpf_timer_init with a generic helper

In reply to Benjamin Tissoires

No code change except for the new flags argument being stored in the local data struct.

---

no changes in v2

---
 kernel/bpf/helpers.c | 91 ++++++++++++++++++++++++++++++++++++----------------
 1 file changed, 63 insertions(+), 28 deletions(-)
kernel/bpf/helpers.c +63 -28
index 5a069c70b5e6..1e6d1011303b 100644
@@ -1111,7 +1111,10 @@ struct bpf_hrtimer {
 
 /* the actual struct hidden inside uapi struct bpf_timer */
 struct bpf_async_kern {
-	struct bpf_hrtimer *timer;
+	union {
+		struct bpf_async_cb *cb;
+		struct bpf_hrtimer *timer;
+	};
 	/* bpf_spin_lock is used here instead of spinlock_t to make
 	 * sure that it always fits into space reserved by struct bpf_timer
 	 * regardless of LOCKDEP and spinlock debug flags.
@@ -1119,6 +1122,10 @@ struct bpf_async_kern {
 	struct bpf_spin_lock lock;
 } __attribute__((aligned(8)));
 
+enum bpf_async_type {
+	BPF_ASYNC_TYPE_TIMER = 0,
+};
+
 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
 
 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
@@ -1160,46 +1167,55 @@ static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
 	return HRTIMER_NORESTART;
 }
 
-BPF_CALL_3(bpf_timer_init, struct bpf_async_kern *, timer, struct bpf_map *, map,
-	   u64, flags)
+static int __bpf_async_init(struct bpf_async_kern *async, struct bpf_map *map, u64 flags,
+			    enum bpf_async_type type)
 {
-	clockid_t clockid = flags & (MAX_CLOCKS - 1);
+	struct bpf_async_cb *cb;
 	struct bpf_hrtimer *t;
+	clockid_t clockid;
+	size_t size;
 	int ret = 0;
 
-	BUILD_BUG_ON(MAX_CLOCKS != 16);
-	BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_timer));
-	BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_timer));
-
 	if (in_nmi())
 		return -EOPNOTSUPP;
 
-	if (flags >= MAX_CLOCKS ||
-	    /* similar to timerfd except _ALARM variants are not supported */
-	    (clockid != CLOCK_MONOTONIC &&
-	     clockid != CLOCK_REALTIME &&
-	     clockid != CLOCK_BOOTTIME))
+	switch (type) {
+	case BPF_ASYNC_TYPE_TIMER:
+		size = sizeof(struct bpf_hrtimer);
+		break;
+	default:
 		return -EINVAL;
-	__bpf_spin_lock_irqsave(&timer->lock);
-	t = timer->timer;
+	}
+
+	__bpf_spin_lock_irqsave(&async->lock);
+	t = async->timer;
 	if (t) {
 		ret = -EBUSY;
 		goto out;
 	}
+
 	/* allocate hrtimer via map_kmalloc to use memcg accounting */
-	t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node);
-	if (!t) {
+	cb = bpf_map_kmalloc_node(map, size, GFP_ATOMIC, map->numa_node);
+	if (!cb) {
 		ret = -ENOMEM;
 		goto out;
 	}
-	t->cb.value = (void *)timer - map->record->timer_off;
-	t->cb.map = map;
-	t->cb.prog = NULL;
-	rcu_assign_pointer(t->cb.callback_fn, NULL);
-	hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
-	t->timer.function = bpf_timer_cb;
-	WRITE_ONCE(timer->timer, t);
-	/* Guarantee the order between timer->timer and map->usercnt. So
+
+	if (type == BPF_ASYNC_TYPE_TIMER) {
+		clockid = flags & (MAX_CLOCKS - 1);
+		t = (struct bpf_hrtimer *)cb;
+
+		hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
+		t->timer.function = bpf_timer_cb;
+		cb->value = (void *)async - map->record->timer_off;
+	}
+	cb->map = map;
+	cb->prog = NULL;
+	cb->flags = flags;
+	rcu_assign_pointer(cb->callback_fn, NULL);
+
+	WRITE_ONCE(async->cb, cb);
+	/* Guarantee the order between async->cb and map->usercnt. So
 	 * when there are concurrent uref release and bpf timer init, either
 	 * bpf_timer_cancel_and_free() called by uref release reads a no-NULL
 	 * timer or atomic64_read() below returns a zero usercnt.
@@ -1209,15 +1225,34 @@ BPF_CALL_3(bpf_timer_init, struct bpf_async_kern *, timer, struct bpf_map *, map
 		/* maps with timers must be either held by user space
 		 * or pinned in bpffs.
 		 */
-		WRITE_ONCE(timer->timer, NULL);
-		kfree(t);
+		WRITE_ONCE(async->cb, NULL);
+		kfree(cb);
 		ret = -EPERM;
 	}
 out:
-	__bpf_spin_unlock_irqrestore(&timer->lock);
+	__bpf_spin_unlock_irqrestore(&async->lock);
 	return ret;
 }
 
+BPF_CALL_3(bpf_timer_init, struct bpf_async_kern *, timer, struct bpf_map *, map,
+	   u64, flags)
+{
+	clock_t clockid = flags & (MAX_CLOCKS - 1);
+
+	BUILD_BUG_ON(MAX_CLOCKS != 16);
+	BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_timer));
+	BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_timer));
+
+	if (flags >= MAX_CLOCKS ||
+	    /* similar to timerfd except _ALARM variants are not supported */
+	    (clockid != CLOCK_MONOTONIC &&
+	     clockid != CLOCK_REALTIME &&
+	     clockid != CLOCK_BOOTTIME))
+		return -EINVAL;
+
+	return __bpf_async_init(timer, map, flags, BPF_ASYNC_TYPE_TIMER);
+}
+
 static const struct bpf_func_proto bpf_timer_init_proto = {
 	.func		= bpf_timer_init,
 	.gpl_only	= true,
Signature · 1 line
2.44.0
#3 lore raw

PATCH v2 3/16 bpf: replace bpf_timer_set_callback with a generic helper

In reply to Benjamin Tissoires

In the same way we have a generic __bpf_async_init(), we also need to share code between timer and workqueue for the set_callback call.

We just add an unused flags parameter, as it will be used for workqueues.

---

no changes in v2

---
 kernel/bpf/helpers.c | 29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)
kernel/bpf/helpers.c +18 -11
index 1e6d1011303b..d0a645b09d3d 100644
@@ -1262,22 +1262,23 @@ static const struct bpf_func_proto bpf_timer_init_proto = {
 	.arg3_type	= ARG_ANYTHING,
 };
 
-BPF_CALL_3(bpf_timer_set_callback, struct bpf_async_kern *, timer, void *, callback_fn,
-	   struct bpf_prog_aux *, aux)
+static int __bpf_async_set_callback(struct bpf_async_kern *async, void *callback_fn,
+				    struct bpf_prog_aux *aux, unsigned int flags,
+				    enum bpf_async_type type)
 {
 	struct bpf_prog *prev, *prog = aux->prog;
-	struct bpf_hrtimer *t;
+	struct bpf_async_cb *cb;
 	int ret = 0;
 
 	if (in_nmi())
 		return -EOPNOTSUPP;
-	__bpf_spin_lock_irqsave(&timer->lock);
-	t = timer->timer;
-	if (!t) {
+	__bpf_spin_lock_irqsave(&async->lock);
+	cb = async->cb;
+	if (!cb) {
 		ret = -EINVAL;
 		goto out;
 	}
-	if (!atomic64_read(&t->cb.map->usercnt)) {
+	if (!atomic64_read(&cb->map->usercnt)) {
 		/* maps with timers must be either held by user space
 		 * or pinned in bpffs. Otherwise timer might still be
 		 * running even when bpf prog is detached and user space
@@ -1286,7 +1287,7 @@ BPF_CALL_3(bpf_timer_set_callback, struct bpf_async_kern *, timer, void *, callb
 		ret = -EPERM;
 		goto out;
 	}
-	prev = t->cb.prog;
+	prev = cb->prog;
 	if (prev != prog) {
 		/* Bump prog refcnt once. Every bpf_timer_set_callback()
 		 * can pick different callback_fn-s within the same prog.
@@ -1299,14 +1300,20 @@ BPF_CALL_3(bpf_timer_set_callback, struct bpf_async_kern *, timer, void *, callb
 		if (prev)
 			/* Drop prev prog refcnt when swapping with new prog */
 			bpf_prog_put(prev);
-		t->cb.prog = prog;
+		cb->prog = prog;
 	}
-	rcu_assign_pointer(t->cb.callback_fn, callback_fn);
+	rcu_assign_pointer(cb->callback_fn, callback_fn);
 out:
-	__bpf_spin_unlock_irqrestore(&timer->lock);
+	__bpf_spin_unlock_irqrestore(&async->lock);
 	return ret;
 }
 
+BPF_CALL_3(bpf_timer_set_callback, struct bpf_async_kern *, timer, void *, callback_fn,
+	   struct bpf_prog_aux *, aux)
+{
+	return __bpf_async_set_callback(timer, callback_fn, aux, 0, BPF_ASYNC_TYPE_TIMER);
+}
+
 static const struct bpf_func_proto bpf_timer_set_callback_proto = {
 	.func		= bpf_timer_set_callback,
 	.gpl_only	= true,
Signature · 1 line
2.44.0
#4 lore raw

PATCH v2 4/16 bpf: replace bpf_timer_cancel_and_free with a generic helper

In reply to Benjamin Tissoires

Same reason than most bpf_timer* functions, we need almost the same for workqueues.
So extract the generic part out of it so bpf_wq_cancel_and_free can reuse it.

---

no changes in v2

---
 kernel/bpf/helpers.c | 42 +++++++++++++++++++++++++-----------------
 1 file changed, 25 insertions(+), 17 deletions(-)
kernel/bpf/helpers.c +25 -17
index d0a645b09d3d..78847f444f79 100644
@@ -1413,36 +1413,44 @@ static const struct bpf_func_proto bpf_timer_cancel_proto = {
 	.arg1_type	= ARG_PTR_TO_TIMER,
 };
 
-/* This function is called by map_delete/update_elem for individual element and
- * by ops->map_release_uref when the user space reference to a map reaches zero.
- */
-void bpf_timer_cancel_and_free(void *val)
+static struct bpf_async_cb *__bpf_async_cancel_and_free(struct bpf_async_kern *async)
 {
-	struct bpf_async_kern *timer = val;
-	struct bpf_hrtimer *t;
+	struct bpf_async_cb *cb;
 
-	/* Performance optimization: read timer->timer without lock first. */
-	if (!READ_ONCE(timer->timer))
-		return;
+	/* Performance optimization: read async->cb without lock first. */
+	if (!READ_ONCE(async->cb))
+		return NULL;
 
-	__bpf_spin_lock_irqsave(&timer->lock);
+	__bpf_spin_lock_irqsave(&async->lock);
 	/* re-read it under lock */
-	t = timer->timer;
-	if (!t)
+	cb = async->cb;
+	if (!cb)
 		goto out;
-	drop_prog_refcnt(&t->cb);
+	drop_prog_refcnt(cb);
 	/* The subsequent bpf_timer_start/cancel() helpers won't be able to use
 	 * this timer, since it won't be initialized.
 	 */
-	WRITE_ONCE(timer->timer, NULL);
+	WRITE_ONCE(async->cb, NULL);
 out:
-	__bpf_spin_unlock_irqrestore(&timer->lock);
+	__bpf_spin_unlock_irqrestore(&async->lock);
+	return cb;
+}
+
+/* This function is called by map_delete/update_elem for individual element and
+ * by ops->map_release_uref when the user space reference to a map reaches zero.
+ */
+void bpf_timer_cancel_and_free(void *val)
+{
+	struct bpf_hrtimer *t;
+
+	t = (struct bpf_hrtimer *)__bpf_async_cancel_and_free(val);
+
 	if (!t)
 		return;
 	/* Cancel the timer and wait for callback to complete if it was running.
 	 * If hrtimer_cancel() can be safely called it's safe to call kfree(t)
 	 * right after for both preallocated and non-preallocated maps.
-	 * The timer->timer = NULL was already done and no code path can
+	 * The async->cb = NULL was already done and no code path can
 	 * see address 't' anymore.
 	 *
 	 * Check that bpf_map_delete/update_elem() wasn't called from timer
@@ -1451,7 +1459,7 @@ void bpf_timer_cancel_and_free(void *val)
 	 * return -1). Though callback_fn is still running on this cpu it's
 	 * safe to do kfree(t) because bpf_timer_cb() read everything it needed
 	 * from 't'. The bpf subprog callback_fn won't be able to access 't',
-	 * since timer->timer = NULL was already done. The timer will be
+	 * since async->cb = NULL was already done. The timer will be
 	 * effectively cancelled because bpf_timer_cb() will return
 	 * HRTIMER_NORESTART.
 	 */
Signature · 1 line
2.44.0
#5 lore raw

PATCH v2 5/16 bpf: add support for bpf_wq user type

In reply to Benjamin Tissoires

Mostly a copy/paste from the bpf_timer API, without the initialization and free, as they will be done in a separate patch.

---

no changes in v2

---
 include/linux/bpf.h      | 11 ++++++++++-
 include/uapi/linux/bpf.h |  4 ++++
 kernel/bpf/btf.c         | 17 +++++++++++++++++
 kernel/bpf/syscall.c     |  6 +++++-
 kernel/bpf/verifier.c    |  9 +++++++++
 5 files changed, 45 insertions(+), 2 deletions(-)
include/linux/bpf.h +10 -1
index 5034c1b4ded7..c7dcfd395555 100644
@@ -185,7 +185,7 @@ struct bpf_map_ops {
 
 enum {
 	/* Support at most 10 fields in a BTF type */
-	BTF_FIELDS_MAX	   = 10,
+	BTF_FIELDS_MAX	   = 11,
 };
 
 enum btf_field_type {
@@ -202,6 +202,7 @@ enum btf_field_type {
 	BPF_GRAPH_NODE = BPF_RB_NODE | BPF_LIST_NODE,
 	BPF_GRAPH_ROOT = BPF_RB_ROOT | BPF_LIST_HEAD,
 	BPF_REFCOUNT   = (1 << 9),
+	BPF_WORKQUEUE  = (1 << 10),
 };
 
 typedef void (*btf_dtor_kfunc_t)(void *);
@@ -238,6 +239,7 @@ struct btf_record {
 	u32 field_mask;
 	int spin_lock_off;
 	int timer_off;
+	int wq_off;
 	int refcount_off;
 	struct btf_field fields[];
 };
@@ -312,6 +314,8 @@ static inline const char *btf_field_type_name(enum btf_field_type type)
 		return "bpf_spin_lock";
 	case BPF_TIMER:
 		return "bpf_timer";
+	case BPF_WORKQUEUE:
+		return "bpf_wq";
 	case BPF_KPTR_UNREF:
 	case BPF_KPTR_REF:
 		return "kptr";
@@ -340,6 +344,8 @@ static inline u32 btf_field_type_size(enum btf_field_type type)
 		return sizeof(struct bpf_spin_lock);
 	case BPF_TIMER:
 		return sizeof(struct bpf_timer);
+	case BPF_WORKQUEUE:
+		return sizeof(struct bpf_wq);
 	case BPF_KPTR_UNREF:
 	case BPF_KPTR_REF:
 	case BPF_KPTR_PERCPU:
@@ -367,6 +373,8 @@ static inline u32 btf_field_type_align(enum btf_field_type type)
 		return __alignof__(struct bpf_spin_lock);
 	case BPF_TIMER:
 		return __alignof__(struct bpf_timer);
+	case BPF_WORKQUEUE:
+		return __alignof__(struct bpf_wq);
 	case BPF_KPTR_UNREF:
 	case BPF_KPTR_REF:
 	case BPF_KPTR_PERCPU:
@@ -406,6 +414,7 @@ static inline void bpf_obj_init_field(const struct btf_field *field, void *addr)
 		/* RB_ROOT_CACHED 0-inits, no need to do anything after memset */
 	case BPF_SPIN_LOCK:
 	case BPF_TIMER:
+	case BPF_WORKQUEUE:
 	case BPF_KPTR_UNREF:
 	case BPF_KPTR_REF:
 	case BPF_KPTR_PERCPU:
include/uapi/linux/bpf.h +4 -0
index cee0a7915c08..e4ae83550fb3 100644
@@ -7306,6 +7306,10 @@ struct bpf_timer {
 	__u64 __opaque[2];
 } __attribute__((aligned(8)));
 
+struct bpf_wq {
+	__u64 __opaque[2];
+} __attribute__((aligned(8)));
+
 struct bpf_dynptr {
 	__u64 __opaque[2];
 } __attribute__((aligned(8)));
kernel/bpf/btf.c +17 -0
index 90c4a32d89ff..a9b5b28a2630 100644
@@ -3464,6 +3464,15 @@ static int btf_get_field_type(const char *name, u32 field_mask, u32 *seen_mask,
 			goto end;
 		}
 	}
+	if (field_mask & BPF_WORKQUEUE) {
+		if (!strcmp(name, "bpf_wq")) {
+			if (*seen_mask & BPF_WORKQUEUE)
+				return -E2BIG;
+			*seen_mask |= BPF_WORKQUEUE;
+			type = BPF_WORKQUEUE;
+			goto end;
+		}
+	}
 	field_mask_test_name(BPF_LIST_HEAD, "bpf_list_head");
 	field_mask_test_name(BPF_LIST_NODE, "bpf_list_node");
 	field_mask_test_name(BPF_RB_ROOT,   "bpf_rb_root");
@@ -3515,6 +3524,7 @@ static int btf_find_struct_field(const struct btf *btf,
 		switch (field_type) {
 		case BPF_SPIN_LOCK:
 		case BPF_TIMER:
+		case BPF_WORKQUEUE:
 		case BPF_LIST_NODE:
 		case BPF_RB_NODE:
 		case BPF_REFCOUNT:
@@ -3582,6 +3592,7 @@ static int btf_find_datasec_var(const struct btf *btf, const struct btf_type *t,
 		switch (field_type) {
 		case BPF_SPIN_LOCK:
 		case BPF_TIMER:
+		case BPF_WORKQUEUE:
 		case BPF_LIST_NODE:
 		case BPF_RB_NODE:
 		case BPF_REFCOUNT:
@@ -3816,6 +3827,7 @@ struct btf_record *btf_parse_fields(const struct btf *btf, const struct btf_type
 
 	rec->spin_lock_off = -EINVAL;
 	rec->timer_off = -EINVAL;
+	rec->wq_off = -EINVAL;
 	rec->refcount_off = -EINVAL;
 	for (i = 0; i < cnt; i++) {
 		field_type_size = btf_field_type_size(info_arr[i].type);
@@ -3846,6 +3858,11 @@ struct btf_record *btf_parse_fields(const struct btf *btf, const struct btf_type
 			/* Cache offset for faster lookup at runtime */
 			rec->timer_off = rec->fields[i].offset;
 			break;
+		case BPF_WORKQUEUE:
+			WARN_ON_ONCE(rec->wq_off >= 0);
+			/* Cache offset for faster lookup at runtime */
+			rec->wq_off = rec->fields[i].offset;
+			break;
 		case BPF_REFCOUNT:
 			WARN_ON_ONCE(rec->refcount_off >= 0);
 			/* Cache offset for faster lookup at runtime */
kernel/bpf/syscall.c +5 -1
index 7d392ec83655..0848e4141b00 100644
@@ -559,6 +559,7 @@ void btf_record_free(struct btf_record *rec)
 		case BPF_SPIN_LOCK:
 		case BPF_TIMER:
 		case BPF_REFCOUNT:
+		case BPF_WORKQUEUE:
 			/* Nothing to release */
 			break;
 		default:
@@ -608,6 +609,7 @@ struct btf_record *btf_record_dup(const struct btf_record *rec)
 		case BPF_SPIN_LOCK:
 		case BPF_TIMER:
 		case BPF_REFCOUNT:
+		case BPF_WORKQUEUE:
 			/* Nothing to acquire */
 			break;
 		default:
@@ -679,6 +681,8 @@ void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 		case BPF_TIMER:
 			bpf_timer_cancel_and_free(field_ptr);
 			break;
+		case BPF_WORKQUEUE:
+			break;
 		case BPF_KPTR_UNREF:
 			WRITE_ONCE(*(u64 *)field_ptr, 0);
 			break;
@@ -1085,7 +1089,7 @@ static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
 
 	map->record = btf_parse_fields(btf, value_type,
 				       BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
-				       BPF_RB_ROOT | BPF_REFCOUNT,
+				       BPF_RB_ROOT | BPF_REFCOUNT | BPF_WORKQUEUE,
 				       map->value_size);
 	if (!IS_ERR_OR_NULL(map->record)) {
 		int i;
kernel/bpf/verifier.c +9 -0
index 2aad6d90550f..deaf2e1ab690 100644
@@ -1838,6 +1838,8 @@ static void mark_ptr_not_null_reg(struct bpf_reg_state *reg)
 			 */
 			if (btf_record_has_field(map->inner_map_meta->record, BPF_TIMER))
 				reg->map_uid = reg->id;
+			if (btf_record_has_field(map->inner_map_meta->record, BPF_WORKQUEUE))
+				reg->map_uid = reg->id;
 		} else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
 			reg->type = PTR_TO_XDP_SOCK;
 		} else if (map->map_type == BPF_MAP_TYPE_SOCKMAP ||
@@ -18155,6 +18157,13 @@ static int check_map_prog_compatibility(struct bpf_verifier_env *env,
 		}
 	}
 
+	if (btf_record_has_field(map->record, BPF_WORKQUEUE)) {
+		if (is_tracing_prog_type(prog_type)) {
+			verbose(env, "tracing progs cannot use bpf_wq yet\n");
+			return -EINVAL;
+		}
+	}
+
 	if ((bpf_prog_is_offloaded(prog->aux) || bpf_map_is_offloaded(map)) &&
 	    !bpf_offload_prog_map_match(prog, map)) {
 		verbose(env, "offload device mismatch between prog and map\n");
Signature · 1 line
2.44.0
#6 lore raw

PATCH v2 6/16 tools: sync include/uapi/linux/bpf.h

In reply to Benjamin Tissoires

cp include/uapi/linux/bpf.h tools/include/uapi/linux/bpf.h

---

no changes in v2

---
 tools/include/uapi/linux/bpf.h | 4 ++++
 1 file changed, 4 insertions(+)
tools/include/uapi/linux/bpf.h +4 -0
index cee0a7915c08..e4ae83550fb3 100644
@@ -7306,6 +7306,10 @@ struct bpf_timer {
 	__u64 __opaque[2];
 } __attribute__((aligned(8)));
 
+struct bpf_wq {
+	__u64 __opaque[2];
+} __attribute__((aligned(8)));
+
 struct bpf_dynptr {
 	__u64 __opaque[2];
 } __attribute__((aligned(8)));
Signature · 1 line
2.44.0
#7 lore raw

PATCH v2 7/16 bpf: verifier: bail out if the argument is not a map

In reply to Benjamin Tissoires

When a kfunc is declared with a KF_ARG_PTR_TO_MAP, we should have reg->map_ptr set to a non NULL value, otherwise, that means that the underlying type is not a map.

---

new in v2

---
 kernel/bpf/verifier.c | 5 +++++
 1 file changed, 5 insertions(+)
kernel/bpf/verifier.c +5 -0
index deaf2e1ab690..b95fd63864f1 100644
@@ -11720,6 +11720,11 @@ static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_
 		case KF_ARG_PTR_TO_NULL:
 			continue;
 		case KF_ARG_PTR_TO_MAP:
+			if (!reg->map_ptr) {
+				verbose(env, "pointer in R%d isn't map pointer\n", regno);
+				return -EINVAL;
+			}
+			fallthrough;
 		case KF_ARG_PTR_TO_ALLOC_BTF_ID:
 		case KF_ARG_PTR_TO_BTF_ID:
 			if (!is_kfunc_trusted_args(meta) && !is_kfunc_rcu(meta))
Signature · 1 line
2.44.0
#8 lore raw

PATCH v2 8/16 bpf: add support for KF_ARG_PTR_TO_WORKQUEUE

In reply to Benjamin Tissoires

Introduce support for KF_ARG_PTR_TO_WORKQUEUE. The kfuncs will use bpf_wq as argument and that will be recognized as workqueue argument by verifier. bpf_wq_kern casting can happen inside kfunc, but using bpf_wq in argument makes life easier for users who work with non-kern type in BPF progs.

Duplicate process_timer_func into process_wq_func.
meta argument is only needed to ensure bpf_wq_init's workqueue and map arguments are coming from the same map (map_uid logic is necessary for correct inner-map handling), so also amend check_kfunc_args() to match what helpers functions check is doing.

---

changes in v2:
- clean up already done checks in process_wq_func()
- adjust comment
- gate check_kfunc_args check with map->record->wq_off.

---
 kernel/bpf/verifier.c | 65 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 65 insertions(+)
kernel/bpf/verifier.c +65 -0
index b95fd63864f1..5ff7b52d9032 100644
@@ -332,6 +332,10 @@ struct bpf_kfunc_call_arg_meta {
 		u8 spi;
 		u8 frameno;
 	} iter;
+	struct {
+		struct bpf_map *ptr;
+		int uid;
+	} map;
 	u64 mem_size;
 };
 
@@ -7598,6 +7602,23 @@ static int process_timer_func(struct bpf_verifier_env *env, int regno,
 	return 0;
 }
 
+static int process_wq_func(struct bpf_verifier_env *env, int regno,
+			   struct bpf_kfunc_call_arg_meta *meta)
+{
+	struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
+	struct bpf_map *map = reg->map_ptr;
+	u64 val = reg->var_off.value;
+
+	if (map->record->wq_off != val + reg->off) {
+		verbose(env, "off %lld doesn't point to 'struct bpf_wq' that is at %d\n",
+			val + reg->off, map->record->wq_off);
+		return -EINVAL;
+	}
+	meta->map.uid = reg->map_uid;
+	meta->map.ptr = map;
+	return 0;
+}
+
 static int process_kptr_func(struct bpf_verifier_env *env, int regno,
 			     struct bpf_call_arg_meta *meta)
 {
@@ -10843,6 +10864,7 @@ enum {
 	KF_ARG_LIST_NODE_ID,
 	KF_ARG_RB_ROOT_ID,
 	KF_ARG_RB_NODE_ID,
+	KF_ARG_WORKQUEUE_ID,
 };
 
 BTF_ID_LIST(kf_arg_btf_ids)
@@ -10851,6 +10873,7 @@ BTF_ID(struct, bpf_list_head)
 BTF_ID(struct, bpf_list_node)
 BTF_ID(struct, bpf_rb_root)
 BTF_ID(struct, bpf_rb_node)
+BTF_ID(struct, bpf_wq)
 
 static bool __is_kfunc_ptr_arg_type(const struct btf *btf,
 				    const struct btf_param *arg, int type)
@@ -10894,6 +10917,11 @@ static bool is_kfunc_arg_rbtree_node(const struct btf *btf, const struct btf_par
 	return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_RB_NODE_ID);
 }
 
+static bool is_kfunc_arg_wq(const struct btf *btf, const struct btf_param *arg)
+{
+	return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_WORKQUEUE_ID);
+}
+
 static bool is_kfunc_arg_callback(struct bpf_verifier_env *env, const struct btf *btf,
 				  const struct btf_param *arg)
 {
@@ -10963,6 +10991,7 @@ enum kfunc_ptr_arg_type {
 	KF_ARG_PTR_TO_NULL,
 	KF_ARG_PTR_TO_CONST_STR,
 	KF_ARG_PTR_TO_MAP,
+	KF_ARG_PTR_TO_WORKQUEUE,
 };
 
 enum special_kfunc_type {
@@ -11119,6 +11148,9 @@ get_kfunc_ptr_arg_type(struct bpf_verifier_env *env,
 	if (is_kfunc_arg_map(meta->btf, &args[argno]))
 		return KF_ARG_PTR_TO_MAP;
 
+	if (is_kfunc_arg_wq(meta->btf, &args[argno]))
+		return KF_ARG_PTR_TO_WORKQUEUE;
+
 	if ((base_type(reg->type) == PTR_TO_BTF_ID || reg2btf_ids[base_type(reg->type)])) {
 		if (!btf_type_is_struct(ref_t)) {
 			verbose(env, "kernel function %s args#%d pointer type %s %s is not supported\n",
@@ -11724,6 +11756,29 @@ static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_
 				verbose(env, "pointer in R%d isn't map pointer\n", regno);
 				return -EINVAL;
 			}
+			if (meta->map.ptr && reg->map_ptr->record->wq_off >= 0) {
+				/* Use map_uid (which is unique id of inner map) to reject:
+				 * inner_map1 = bpf_map_lookup_elem(outer_map, key1)
+				 * inner_map2 = bpf_map_lookup_elem(outer_map, key2)
+				 * if (inner_map1 && inner_map2) {
+				 *     wq = bpf_map_lookup_elem(inner_map1);
+				 *     if (wq)
+				 *         // mismatch would have been allowed
+				 *         bpf_wq_init(wq, inner_map2);
+				 * }
+				 *
+				 * Comparing map_ptr is enough to distinguish normal and outer maps.
+				 */
+				if (meta->map.ptr != reg->map_ptr ||
+				    meta->map.uid != reg->map_uid) {
+					verbose(env,
+						"workqueue pointer in R1 map_uid=%d doesn't match map pointer in R2 map_uid=%d\n",
+						meta->map.uid, reg->map_uid);
+					return -EINVAL;
+				}
+			}
+			meta->map.ptr = reg->map_ptr;
+			meta->map.uid = reg->map_uid;
 			fallthrough;
 		case KF_ARG_PTR_TO_ALLOC_BTF_ID:
 		case KF_ARG_PTR_TO_BTF_ID:
@@ -11757,6 +11812,7 @@ static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_
 		case KF_ARG_PTR_TO_CALLBACK:
 		case KF_ARG_PTR_TO_REFCOUNTED_KPTR:
 		case KF_ARG_PTR_TO_CONST_STR:
+		case KF_ARG_PTR_TO_WORKQUEUE:
 			/* Trusted by default */
 			break;
 		default:
@@ -12043,6 +12099,15 @@ static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_
 			if (ret)
 				return ret;
 			break;
+		case KF_ARG_PTR_TO_WORKQUEUE:
+			if (reg->type != PTR_TO_MAP_VALUE) {
+				verbose(env, "arg#%d doesn't point to a map value\n", i);
+				return -EINVAL;
+			}
+			ret = process_wq_func(env, regno, meta);
+			if (ret < 0)
+				return ret;
+			break;
 		}
 	}
 
Signature · 1 line
2.44.0
#9 lore raw

PATCH v2 9/16 bpf: allow struct bpf_wq to be embedded in arraymaps and hashmaps

In reply to Benjamin Tissoires

Currently bpf_wq_cancel_and_free() is just a placeholder as there is no memory allocation for bpf_wq just yet.

Again, duplication of the bpf_timer approach

---

changes in v2:
- removed extra fallthrough

---
 include/linux/bpf.h   |  2 ++
 kernel/bpf/arraymap.c | 18 ++++++++++-------
 kernel/bpf/hashtab.c  | 55 ++++++++++++++++++++++++++++++++++++++++-----------
 kernel/bpf/helpers.c  |  8 ++++++++
 kernel/bpf/syscall.c  |  9 +++++++++
 5 files changed, 73 insertions(+), 19 deletions(-)
include/linux/bpf.h +2 -0
index c7dcfd395555..64bf0cf3ee95 100644
@@ -534,6 +534,7 @@ static inline void zero_map_value(struct bpf_map *map, void *dst)
 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
 			   bool lock_src);
 void bpf_timer_cancel_and_free(void *timer);
+void bpf_wq_cancel_and_free(void *timer);
 void bpf_list_head_free(const struct btf_field *field, void *list_head,
 			struct bpf_spin_lock *spin_lock);
 void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
@@ -2204,6 +2205,7 @@ void bpf_map_free_record(struct bpf_map *map);
 struct btf_record *btf_record_dup(const struct btf_record *rec);
 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b);
 void bpf_obj_free_timer(const struct btf_record *rec, void *obj);
+void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj);
 void bpf_obj_free_fields(const struct btf_record *rec, void *obj);
 void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu);
 
kernel/bpf/arraymap.c +11 -7
index 8c1e6d7654bb..580d07b15471 100644
@@ -428,17 +428,21 @@ static void *array_map_vmalloc_addr(struct bpf_array *array)
 	return (void *)round_down((unsigned long)array, PAGE_SIZE);
 }
 
-static void array_map_free_timers(struct bpf_map *map)
+static void array_map_free_timers_wq(struct bpf_map *map)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	int i;
 
-	/* We don't reset or free fields other than timer on uref dropping to zero. */
-	if (!btf_record_has_field(map->record, BPF_TIMER))
-		return;
+	/* We don't reset or free fields other than timer and workqueue
+	 * on uref dropping to zero.
+	 */
+	if (btf_record_has_field(map->record, BPF_TIMER))
+		for (i = 0; i < array->map.max_entries; i++)
+			bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
 
-	for (i = 0; i < array->map.max_entries; i++)
-		bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
+	if (btf_record_has_field(map->record, BPF_WORKQUEUE))
+		for (i = 0; i < array->map.max_entries; i++)
+			bpf_obj_free_workqueue(map->record, array_map_elem_ptr(array, i));
 }
 
 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
@@ -782,7 +786,7 @@ const struct bpf_map_ops array_map_ops = {
 	.map_alloc = array_map_alloc,
 	.map_free = array_map_free,
 	.map_get_next_key = array_map_get_next_key,
-	.map_release_uref = array_map_free_timers,
+	.map_release_uref = array_map_free_timers_wq,
 	.map_lookup_elem = array_map_lookup_elem,
 	.map_update_elem = array_map_update_elem,
 	.map_delete_elem = array_map_delete_elem,
kernel/bpf/hashtab.c +43 -12

94 lines not on this page · raw message

kernel/bpf/helpers.c +8 -0
index 78847f444f79..9fd12d480b8b 100644
@@ -1468,6 +1468,14 @@ void bpf_timer_cancel_and_free(void *val)
 	kfree_rcu(t, cb.rcu);
 }
 
+/* This function is called by map_delete/update_elem for individual element and
+ * by ops->map_release_uref when the user space reference to a map reaches zero.
+ */
+void bpf_wq_cancel_and_free(void *val)
+{
+	BTF_TYPE_EMIT(struct bpf_wq);
+}
+
 BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
 {
 	unsigned long *kptr = map_value;
kernel/bpf/syscall.c +9 -0
index 0848e4141b00..63e368337483 100644
@@ -661,6 +661,13 @@ void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 	bpf_timer_cancel_and_free(obj + rec->timer_off);
 }
 
+void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj)
+{
+	if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_WORKQUEUE)))
+		return;
+	bpf_wq_cancel_and_free(obj + rec->wq_off);
+}
+
 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 {
 	const struct btf_field *fields;
@@ -682,6 +689,7 @@ void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 			bpf_timer_cancel_and_free(field_ptr);
 			break;
 		case BPF_WORKQUEUE:
+			bpf_wq_cancel_and_free(field_ptr);
 			break;
 		case BPF_KPTR_UNREF:
 			WRITE_ONCE(*(u64 *)field_ptr, 0);
@@ -1119,6 +1127,7 @@ static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
 				}
 				break;
 			case BPF_TIMER:
+			case BPF_WORKQUEUE:
 				if (map->map_type != BPF_MAP_TYPE_HASH &&
 				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
 				    map->map_type != BPF_MAP_TYPE_ARRAY) {
Signature · 1 line
2.44.0
#10 lore raw

Re: [PATCH bpf-next v2 09/16] bpf: allow struct bpf_wq to be embedded in arraymaps and hashmaps

Quoted 17 lines · Benjamin Tissoires
kernel/bpf/hashtab.c +4 -8
-static void htab_map_free_timers(struct bpf_map *map)
+static void htab_map_free_timers_and_wq(struct bpf_map *map)
 {
        struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 
-       /* We only free timer on uref dropping to zero */
-       if (!btf_record_has_field(htab->map.record, BPF_TIMER))
-               return;
-       if (!htab_is_prealloc(htab))
-               htab_free_malloced_timers(htab);
-       else
-               htab_free_prealloced_timers(htab);
+       /* We only free timer and workqueue on uref dropping to zero */
+       if (btf_record_has_field(htab->map.record, BPF_TIMER)) {
+               if (!htab_is_prealloc(htab))

kernel/bpf/hashtab.c static void htab_free_malloced_timers(struct bpf_htab *htab) · line 1558

Quoted 10 lines
kernel/bpf/hashtab.c +10 -0
+                       htab_free_malloced_timers_or_wq(htab, true);
+               else
+                       htab_free_prealloced_timers(htab);
+       }
+       if (btf_record_has_field(htab->map.record, BPF_WORKQUEUE)) {
+               if (!htab_is_prealloc(htab))
+                       htab_free_malloced_timers_or_wq(htab, false);
+               else
+                       htab_free_prealloced_wq(htab);
+       }

I missed this earlier, but pls follow up to improve this part. Double walk of all elements once for timer and 2nd for wq
is inefficient. Better to handle it in one go in prealloc and malloced cases. Until we have delayed_wq it's quite possible that somebody will have both timers and wq in map elements.

#11 lore raw

PATCH v2 10/16 selftests/bpf: add bpf_wq tests

In reply to Benjamin Tissoires

We simply try in all supported map types if we can store/load a bpf_wq.

---

no changes in v2

---
 tools/testing/selftests/bpf/prog_tests/wq.c |  11 +++
 tools/testing/selftests/bpf/progs/wq.c      | 135 ++++++++++++++++++++++++++++
 2 files changed, 146 insertions(+)
tools/testing/selftests/bpf/prog_tests/wq.c new file +11 -0
new file mode 100644
index 000000000000..9a07b8bc2c52
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2024 Benjamin Tissoires */
+#include <test_progs.h>
+#include "wq.skel.h"
+
+void serial_test_wq(void)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, topts);
+
+	RUN_TESTS(wq);
+}
tools/testing/selftests/bpf/progs/wq.c new file +135 -0

135 lines not on this page · raw message

Signature · 1 line
2.44.0
#12 lore raw

Re: [PATCH bpf-next v2 10/16] selftests/bpf: add bpf_wq tests

Quoted 6 lines · Benjamin Tissoires
tools/testing/selftests/bpf/progs/wq.c new file +5 -0
+
+SEC("tc")
+/* test that workqueues can be used from an array
+ */
+__retval(0)

tools/testing/selftests/bpf/progs/wq.c line 94

Quoted 10 lines
tools/testing/selftests/bpf/progs/wq.c new file +10 -0
+long test_call_array_sleepable(void *ctx)
+{
+       int key = 0;
+
+       return test_elem_callback(&array, &key);
+}
+
+SEC("syscall")
+/* Same test than above but from a sleepable context.
+ */

I adjusted all these comments to be normal single line comments.

The main thing to follow up is to figure out how to wait for wq to finish.
Since we have this message:
./test_progs -t wq
...
#521 wq:OK
Summary: 2/9 PASSED, 0 SKIPPED, 0 FAILED
Failed to unload bpf_testmod.ko from kernel: -11

and doing manual "rmmod bpf_testmod"
after the test run is quite annoying.

Overall great stuff. Thank you for pushing it through. Much appreciated!

#13 lore raw

Re: [PATCH bpf-next v2 10/16] selftests/bpf: add bpf_wq tests

Quoted 30 lines · Alexei Starovoitov
Quoted 6 lines · Benjamin Tissoires
tools/testing/selftests/bpf/progs/wq.c new file +5 -0
+
+SEC("tc")
+/* test that workqueues can be used from an array
+ */
+__retval(0)

tools/testing/selftests/bpf/progs/wq.c line 94

Quoted 10 lines
tools/testing/selftests/bpf/progs/wq.c new file +10 -0
+long test_call_array_sleepable(void *ctx)
+{
+       int key = 0;
+
+       return test_elem_callback(&array, &key);
+}
+
+SEC("syscall")
+/* Same test than above but from a sleepable context.
+ */

I adjusted all these comments to be normal single line comments.

The main thing to follow up is to figure out how to wait for wq to finish.
Since we have this message:
./test_progs -t wq
...
#521 wq:OK
Summary: 2/9 PASSED, 0 SKIPPED, 0 FAILED
Failed to unload bpf_testmod.ko from kernel: -11

Quoted 5 lines

and doing manual "rmmod bpf_testmod"
after the test run is quite annoying.

Overall great stuff. Thank you for pushing it through. Much appreciated!

Hi Benjamin,

I've ended up fixing it as well,
because it was causing CI issues.

See
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/commit/?id=82e38a505c9868e784ec31e743fd8a9fa5ca1084

#14 lore raw

PATCH v2 11/16 bpf: wq: add bpf_wq_init

In reply to Benjamin Tissoires

We need to teach the verifier about the second argument which is declared as void * but which is of type KF_ARG_PTR_TO_MAP. We could have dropped this extra case if we declared the second argument as struct bpf_map *, but that means users will have to do extra casting to have their program compile.

We also need to duplicate the timer code for the checking if the map argument is matching the provided workqueue.

---

FWIW, I still have one concern with this implementation:
- bpf_wq_work() access ->prog without protection, but I think this might be racing with bpf_wq_set_callback(): if we have the following:

  CPU 0                                     CPU 1
  bpf_wq_set_callback()
  bpf_start()
                                            bpf_wq_work():
                                              prog = cb->prog;

  bpf_wq_set_callback()
    cb->prog = prog;
    bpf_prog_put(prev)
    rcu_assign_ptr(cb->callback_fn,
                   callback_fn);
                                           callback = READ_ONCE(w->cb.callback_fn);

As I understand callback_fn is fine, prog might be, but we clearly have an inconstency between "prog" and "callback_fn" as they can come from 2 different bpf_wq_set_callback() calls.

IMO we should protect this by the async->lock, but I'm not sure if it's OK or not.

---

changes in v2:
- dropped BPF_F_WQ_SLEEPABLE
- use __map suffix to drop special case in verifier
- in bpf_wq_work: use new local variable cb

---
 kernel/bpf/helpers.c | 104 ++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 102 insertions(+), 2 deletions(-)
kernel/bpf/helpers.c +102 -2
index 9fd12d480b8b..b12486adc47a 100644
@@ -1109,11 +1109,18 @@ struct bpf_hrtimer {
 	struct hrtimer timer;
 };
 
-/* the actual struct hidden inside uapi struct bpf_timer */
+struct bpf_work {
+	struct bpf_async_cb cb;
+	struct work_struct work;
+	struct work_struct delete_work;
+};
+
+/* the actual struct hidden inside uapi struct bpf_timer and bpf_wq */
 struct bpf_async_kern {
 	union {
 		struct bpf_async_cb *cb;
 		struct bpf_hrtimer *timer;
+		struct bpf_work *work;
 	};
 	/* bpf_spin_lock is used here instead of spinlock_t to make
 	 * sure that it always fits into space reserved by struct bpf_timer
@@ -1124,6 +1131,7 @@ struct bpf_async_kern {
 
 enum bpf_async_type {
 	BPF_ASYNC_TYPE_TIMER = 0,
+	BPF_ASYNC_TYPE_WQ,
 };
 
 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
@@ -1167,11 +1175,64 @@ static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
 	return HRTIMER_NORESTART;
 }
 
+static void bpf_wq_work(struct work_struct *work)
+{
+	struct bpf_work *w = container_of(work, struct bpf_work, work);
+	struct bpf_tramp_run_ctx __maybe_unused run_ctx;
+	struct bpf_async_cb *cb = &w->cb;
+	struct bpf_prog *prog = cb->prog;
+	struct bpf_map *map = cb->map;
+	bpf_callback_t callback_fn;
+	void *value = cb->value;
+	void *key;
+	u32 idx;
+
+	BTF_TYPE_EMIT(struct bpf_wq);
+
+	callback_fn = READ_ONCE(cb->callback_fn);
+	if (!callback_fn || !prog)
+		return;
+
+	if (map->map_type == BPF_MAP_TYPE_ARRAY) {
+		struct bpf_array *array = container_of(map, struct bpf_array, map);
+
+		/* compute the key */
+		idx = ((char *)value - array->value) / array->elem_size;
+		key = &idx;
+	} else { /* hash or lru */
+		key = value - round_up(map->key_size, 8);
+	}
+
+	run_ctx.bpf_cookie = 0;
+
+	if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
+		/* recursion detected */
+		__bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
+		return;
+	}
+
+	callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
+	/* The verifier checked that return value is zero. */
+
+	__bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
+					&run_ctx);
+}
+
+static void bpf_wq_delete_work(struct work_struct *work)
+{
+	struct bpf_work *w = container_of(work, struct bpf_work, delete_work);
+
+	cancel_work_sync(&w->work);
+
+	kfree_rcu(w, cb.rcu);
+}
+
 static int __bpf_async_init(struct bpf_async_kern *async, struct bpf_map *map, u64 flags,
 			    enum bpf_async_type type)
 {
 	struct bpf_async_cb *cb;
 	struct bpf_hrtimer *t;
+	struct bpf_work *w;
 	clockid_t clockid;
 	size_t size;
 	int ret = 0;
@@ -1183,6 +1244,9 @@ static int __bpf_async_init(struct bpf_async_kern *async, struct bpf_map *map, u
 	case BPF_ASYNC_TYPE_TIMER:
 		size = sizeof(struct bpf_hrtimer);
 		break;
+	case BPF_ASYNC_TYPE_WQ:
+		size = sizeof(struct bpf_work);
+		break;
 	default:
 		return -EINVAL;
 	}
@@ -1201,13 +1265,22 @@ static int __bpf_async_init(struct bpf_async_kern *async, struct bpf_map *map, u
 		goto out;
 	}
 
-	if (type == BPF_ASYNC_TYPE_TIMER) {
+	switch (type) {
+	case BPF_ASYNC_TYPE_TIMER:
 		clockid = flags & (MAX_CLOCKS - 1);
 		t = (struct bpf_hrtimer *)cb;
 
 		hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
 		t->timer.function = bpf_timer_cb;
 		cb->value = (void *)async - map->record->timer_off;
+		break;
+	case BPF_ASYNC_TYPE_WQ:
+		w = (struct bpf_work *)cb;
+
+		INIT_WORK(&w->work, bpf_wq_work);
+		INIT_WORK(&w->delete_work, bpf_wq_delete_work);
+		cb->value = (void *)async - map->record->wq_off;
+		break;
 	}
 	cb->map = map;
 	cb->prog = NULL;
@@ -1473,7 +1546,19 @@ void bpf_timer_cancel_and_free(void *val)
  */
 void bpf_wq_cancel_and_free(void *val)
 {
+	struct bpf_work *work;
+
 	BTF_TYPE_EMIT(struct bpf_wq);
+
+	work = (struct bpf_work *)__bpf_async_cancel_and_free(val);
+	if (!work)
+		return;
+	/* Trigger cancel of the sleepable work, but *do not* wait for
+	 * it to finish if it was running as we might not be in a
+	 * sleepable context.
+	 * kfree will be called once the work has finished.
+	 */
+	schedule_work(&work->delete_work);
 }
 
 BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
@@ -2612,6 +2697,20 @@ __bpf_kfunc void bpf_throw(u64 cookie)
 	WARN(1, "A call to BPF exception callback should never return\n");
 }
 
+__bpf_kfunc int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags)
+{
+	struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
+	struct bpf_map *map = p__map;
+
+	BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_wq));
+	BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_wq));
+
+	if (flags)
+		return -EINVAL;
+
+	return __bpf_async_init(async, map, flags, BPF_ASYNC_TYPE_WQ);
+}
+
 __bpf_kfunc_end_defs();
 
 BTF_KFUNCS_START(generic_btf_ids)
@@ -2689,6 +2788,7 @@ BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
 BTF_ID_FLAGS(func, bpf_dynptr_size)
 BTF_ID_FLAGS(func, bpf_dynptr_clone)
 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
+BTF_ID_FLAGS(func, bpf_wq_init)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
Signature · 1 line
2.44.0
#15 lore raw

Re: [PATCH bpf-next v2 11/16] bpf: wq: add bpf_wq_init

Quoted 31 lines · Benjamin Tissoires

We need to teach the verifier about the second argument which is declared as void * but which is of type KF_ARG_PTR_TO_MAP. We could have dropped this extra case if we declared the second argument as struct bpf_map *, but that means users will have to do extra casting to have their program compile.

We also need to duplicate the timer code for the checking if the map argument is matching the provided workqueue.

---

FWIW, I still have one concern with this implementation:
- bpf_wq_work() access ->prog without protection, but I think this might be racing with bpf_wq_set_callback(): if we have the following:

  CPU 0                                     CPU 1
  bpf_wq_set_callback()
  bpf_start()
                                            bpf_wq_work():
                                              prog = cb->prog;

  bpf_wq_set_callback()
    cb->prog = prog;
    bpf_prog_put(prev)
    rcu_assign_ptr(cb->callback_fn,
                   callback_fn);
                                           callback = READ_ONCE(w->cb.callback_fn);
Quoted 6 lines

As I understand callback_fn is fine, prog might be, but we clearly have an inconstency between "prog" and "callback_fn" as they can come from 2 different bpf_wq_set_callback() calls.

IMO we should protect this by the async->lock, but I'm not sure if it's OK or not.

I see the concern, but I think it's overkill.
Here 'prog' is used to pass it into __bpf_prog_enter_sleepable_recur() to keep the standard pattern of calling into sleepable prog. But it won't recurse.
We can open code migrate_disable,etc from there except this_cpu_inc_return, but it's an overkill.
The passed 'prog' is irrelevant.
If somebody tries really hard by having two progs sharing the same map with bpf_wq and racing to set_callback... I can see how prog won't match callback, but it won't make a difference.
prog is not going trigger recursion check (unless somebody tries is obsessed) and not going to UAF.
I imagine it's possible to attach somewhere in core wq callback invocation path with fentry, set_callback to the same prog,
and technically it's kinda sorta recursion, but different subprogs, so not a safety issue.
The code as-is is fine. imo.

#16 lore raw

Re: [PATCH bpf-next v2 11/16] bpf: wq: add bpf_wq_init

Quoted 49 lines · Alexei Starovoitov
Quoted 31 lines · Benjamin Tissoires

We need to teach the verifier about the second argument which is declared as void * but which is of type KF_ARG_PTR_TO_MAP. We could have dropped this extra case if we declared the second argument as struct bpf_map *, but that means users will have to do extra casting to have their program compile.

We also need to duplicate the timer code for the checking if the map argument is matching the provided workqueue.

---

FWIW, I still have one concern with this implementation:
- bpf_wq_work() access ->prog without protection, but I think this might be racing with bpf_wq_set_callback(): if we have the following:

  CPU 0                                     CPU 1
  bpf_wq_set_callback()
  bpf_start()
                                            bpf_wq_work():
                                              prog = cb->prog;

  bpf_wq_set_callback()
    cb->prog = prog;
    bpf_prog_put(prev)
    rcu_assign_ptr(cb->callback_fn,
                   callback_fn);
                                           callback = READ_ONCE(w->cb.callback_fn);
Quoted 6 lines

As I understand callback_fn is fine, prog might be, but we clearly have an inconstency between "prog" and "callback_fn" as they can come from 2 different bpf_wq_set_callback() calls.

IMO we should protect this by the async->lock, but I'm not sure if it's OK or not.

I see the concern, but I think it's overkill.
Here 'prog' is used to pass it into __bpf_prog_enter_sleepable_recur() to keep the standard pattern of calling into sleepable prog. But it won't recurse.
We can open code migrate_disable,etc from there except this_cpu_inc_return, but it's an overkill.
The passed 'prog' is irrelevant.
If somebody tries really hard by having two progs sharing the same map with bpf_wq and racing to set_callback... I can see how

Quoted 8 lines

prog won't match callback, but it won't make a difference. prog is not going trigger recursion check (unless somebody tries is obsessed) and not going to UAF.
I imagine it's possible to attach somewhere in core wq callback invocation path with fentry, set_callback to the same prog, and technically it's kinda sorta recursion, but different subprogs, so not a safety issue.
The code as-is is fine. imo.

After sleeping on it, I realized that the use of __bpf_prog_enter_sleepable_recur() here is very much incorrect :( The tests are passing only because we don't inc prog->active when we run the prog via prog_run cmd.
Adding the following:

net/bpf/test_run.c +2 -0
index f6aad4ed2ab2..0732dfe22204 100644
@@ -1514,7 +1514,9 @@ int bpf_prog_test_run_syscall(struct bpf_prog *prog,
        }
 
        rcu_read_lock_trace();
+       this_cpu_inc_return(*(prog->active));
        retval = bpf_prog_run_pin_on_cpu(prog, ctx);
+       this_cpu_dec(*(prog->active));
        rcu_read_unlock_trace();
 

makes the test fail sporadically.
Or 100% fail when the kernel is booted with 1 cpu.

Could you send a quick follow up to
replace __bpf_prog_enter_sleepable_recur() with

        rcu_read_lock_trace();
        migrate_disable();

?

Or I'll do it in an hour or so.

#17 lore raw

Re: [PATCH bpf-next v2 11/16] bpf: wq: add bpf_wq_init

Quoted 82 lines · Alexei Starovoitov
Quoted 49 lines · Alexei Starovoitov
Quoted 31 lines · Benjamin Tissoires

We need to teach the verifier about the second argument which is declared as void * but which is of type KF_ARG_PTR_TO_MAP. We could have dropped this extra case if we declared the second argument as struct bpf_map *, but that means users will have to do extra casting to have their program compile.

We also need to duplicate the timer code for the checking if the map argument is matching the provided workqueue.

---

FWIW, I still have one concern with this implementation:
- bpf_wq_work() access ->prog without protection, but I think this might be racing with bpf_wq_set_callback(): if we have the following:

  CPU 0                                     CPU 1
  bpf_wq_set_callback()
  bpf_start()
                                            bpf_wq_work():
                                              prog = cb->prog;

  bpf_wq_set_callback()
    cb->prog = prog;
    bpf_prog_put(prev)
    rcu_assign_ptr(cb->callback_fn,
                   callback_fn);
                                           callback = READ_ONCE(w->cb.callback_fn);
Quoted 6 lines

As I understand callback_fn is fine, prog might be, but we clearly have an inconstency between "prog" and "callback_fn" as they can come from 2 different bpf_wq_set_callback() calls.

IMO we should protect this by the async->lock, but I'm not sure if it's OK or not.

I see the concern, but I think it's overkill.
Here 'prog' is used to pass it into __bpf_prog_enter_sleepable_recur() to keep the standard pattern of calling into sleepable prog. But it won't recurse.
We can open code migrate_disable,etc from there except this_cpu_inc_return, but it's an overkill.
The passed 'prog' is irrelevant.
If somebody tries really hard by having two progs sharing the same map with bpf_wq and racing to set_callback... I can see how

Quoted 8 lines

prog won't match callback, but it won't make a difference. prog is not going trigger recursion check (unless somebody tries is obsessed) and not going to UAF.
I imagine it's possible to attach somewhere in core wq callback invocation path with fentry, set_callback to the same prog, and technically it's kinda sorta recursion, but different subprogs, so not a safety issue.
The code as-is is fine. imo.

After sleeping on it, I realized that the use of __bpf_prog_enter_sleepable_recur() here is very much incorrect :( The tests are passing only because we don't inc prog->active when we run the prog via prog_run cmd.
Adding the following:

net/bpf/test_run.c +2 -0
index f6aad4ed2ab2..0732dfe22204 100644
@@ -1514,7 +1514,9 @@ int bpf_prog_test_run_syscall(struct bpf_prog *prog,
        }
 
        rcu_read_lock_trace();
+       this_cpu_inc_return(*(prog->active));
        retval = bpf_prog_run_pin_on_cpu(prog, ctx);
+       this_cpu_dec(*(prog->active));
        rcu_read_unlock_trace();
 

makes the test fail sporadically.
Or 100% fail when the kernel is booted with 1 cpu.

Quoted 7 lines

Could you send a quick follow up to
replace __bpf_prog_enter_sleepable_recur() with

        rcu_read_lock_trace();
        migrate_disable();

?

Or I'll do it in an hour or so.

Considering two broken-build reports already
I've applied the following fix:
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/commit/?id=dc92febf7b93da5049fe177804e6b1961fcc6bd7

that addresses the build issue on !JIT and fixes this recursion problem.

#18 lore raw

Re: [PATCH bpf-next v2 11/16] bpf: wq: add bpf_wq_init

Quoted 90 lines · Alexei Starovoitov
Quoted 88 lines · Alexei Starovoitov
Quoted 49 lines · Alexei Starovoitov
Quoted 31 lines · Benjamin Tissoires

We need to teach the verifier about the second argument which is declared as void * but which is of type KF_ARG_PTR_TO_MAP. We could have dropped this extra case if we declared the second argument as struct bpf_map *, but that means users will have to do extra casting to have their program compile.

We also need to duplicate the timer code for the checking if the map argument is matching the provided workqueue.

---

FWIW, I still have one concern with this implementation:
- bpf_wq_work() access ->prog without protection, but I think this might be racing with bpf_wq_set_callback(): if we have the following:

  CPU 0                                     CPU 1
  bpf_wq_set_callback()
  bpf_start()
                                            bpf_wq_work():
                                              prog = cb->prog;

  bpf_wq_set_callback()
    cb->prog = prog;
    bpf_prog_put(prev)
    rcu_assign_ptr(cb->callback_fn,
                   callback_fn);
                                           callback = READ_ONCE(w->cb.callback_fn);
Quoted 6 lines

As I understand callback_fn is fine, prog might be, but we clearly have an inconstency between "prog" and "callback_fn" as they can come from 2 different bpf_wq_set_callback() calls.

IMO we should protect this by the async->lock, but I'm not sure if it's OK or not.

I see the concern, but I think it's overkill.
Here 'prog' is used to pass it into __bpf_prog_enter_sleepable_recur() to keep the standard pattern of calling into sleepable prog. But it won't recurse.
We can open code migrate_disable,etc from there except this_cpu_inc_return, but it's an overkill.
The passed 'prog' is irrelevant.
If somebody tries really hard by having two progs sharing the same map with bpf_wq and racing to set_callback... I can see how

Quoted 8 lines

prog won't match callback, but it won't make a difference. prog is not going trigger recursion check (unless somebody tries is obsessed) and not going to UAF.
I imagine it's possible to attach somewhere in core wq callback invocation path with fentry, set_callback to the same prog, and technically it's kinda sorta recursion, but different subprogs, so not a safety issue.
The code as-is is fine. imo.

After sleeping on it, I realized that the use of __bpf_prog_enter_sleepable_recur() here is very much incorrect :( The tests are passing only because we don't inc prog->active when we run the prog via prog_run cmd.
Adding the following:

net/bpf/test_run.c +2 -0
index f6aad4ed2ab2..0732dfe22204 100644
@@ -1514,7 +1514,9 @@ int bpf_prog_test_run_syscall(struct bpf_prog *prog,
        }
 
        rcu_read_lock_trace();
+       this_cpu_inc_return(*(prog->active));
        retval = bpf_prog_run_pin_on_cpu(prog, ctx);
+       this_cpu_dec(*(prog->active));
        rcu_read_unlock_trace();
 

makes the test fail sporadically.
Or 100% fail when the kernel is booted with 1 cpu.

Could you send a quick follow up to
replace __bpf_prog_enter_sleepable_recur() with

        rcu_read_lock_trace();
        migrate_disable();

?

Quoted 7 lines
Quoted 1 line

Or I'll do it in an hour or so.

Considering two broken-build reports already
I've applied the following fix:
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/commit/?id=dc92febf7b93da5049fe177804e6b1961fcc6bd7

that addresses the build issue on !JIT and fixes this recursion problem.

Thanks a lot for fixing this on my behalf. I was slightly puzzled by the broken-build report but really I wasn't in position to think straight yesterday (got a pretty big muscular pain in the back and had to go to the doctor to get painkillers)

I haven't forgoten about the double list walk in 9/16, I'll hopefully send the fix today.

Cheers,
Benjamin

#19 lore raw

PATCH v2 12/16 selftests/bpf: wq: add bpf_wq_init() checks

In reply to Benjamin Tissoires

Allows to test if allocation/free works

---

changes in v2:
- dropped mark_precise checks

---
 tools/testing/selftests/bpf/bpf_experimental.h  |  1 +
 tools/testing/selftests/bpf/prog_tests/wq.c     |  8 +++
 tools/testing/selftests/bpf/progs/wq.c          | 10 ++++
 tools/testing/selftests/bpf/progs/wq_failures.c | 78 +++++++++++++++++++++++++
 4 files changed, 97 insertions(+)
tools/testing/selftests/bpf/bpf_experimental.h +1 -0
index 3329ea080865..785b91b629be 100644
@@ -470,4 +470,5 @@ extern int bpf_iter_css_new(struct bpf_iter_css *it,
 extern struct cgroup_subsys_state *bpf_iter_css_next(struct bpf_iter_css *it) __weak __ksym;
 extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym;
 
+extern int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags) __weak __ksym;
 #endif
tools/testing/selftests/bpf/prog_tests/wq.c +8 -0
index 9a07b8bc2c52..26ab69796103 100644
@@ -2,6 +2,7 @@
 /* Copyright (c) 2024 Benjamin Tissoires */
 #include <test_progs.h>
 #include "wq.skel.h"
+#include "wq_failures.skel.h"
 
 void serial_test_wq(void)
 {
@@ -9,3 +10,10 @@ void serial_test_wq(void)
 
 	RUN_TESTS(wq);
 }
+
+void serial_test_failures_wq(void)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, topts);
+
+	RUN_TESTS(wq_failures);
+}
tools/testing/selftests/bpf/progs/wq.c +10 -0
index 8c668ad04059..f92466eb8fb1 100644
@@ -52,6 +52,7 @@ struct {
 static int test_elem_callback(void *map, int *key)
 {
 	struct elem init = {}, *val;
+	struct bpf_wq *wq;
 
 	if (map == &lru &&
 	    bpf_map_update_elem(map, key, &init, 0))
@@ -61,12 +62,17 @@ static int test_elem_callback(void *map, int *key)
 	if (!val)
 		return -2;
 
+	wq = &val->w;
+	if (bpf_wq_init(wq, map, 0) != 0)
+		return -3;
+
 	return 0;
 }
 
 static int test_hmap_elem_callback(void *map, int *key)
 {
 	struct hmap_elem init = {}, *val;
+	struct bpf_wq *wq;
 
 	if (bpf_map_update_elem(map, key, &init, 0))
 		return -1;
@@ -75,6 +81,10 @@ static int test_hmap_elem_callback(void *map, int *key)
 	if (!val)
 		return -2;
 
+	wq = &val->work;
+	if (bpf_wq_init(wq, map, 0) != 0)
+		return -3;
+
 	return 0;
 }
 
tools/testing/selftests/bpf/progs/wq_failures.c new file +78 -0

78 lines not on this page · raw message

Signature · 1 line
2.44.0
#20 lore raw

Re: [PATCH bpf-next v2 12/16] selftests/bpf: wq: add bpf_wq_init() checks

Quoted 39 lines · Benjamin Tissoires

Allows to test if allocation/free works

---

changes in v2:
- dropped mark_precise checks

---
 tools/testing/selftests/bpf/bpf_experimental.h  |  1 +
 tools/testing/selftests/bpf/prog_tests/wq.c     |  8 +++
 tools/testing/selftests/bpf/progs/wq.c          | 10 ++++
 tools/testing/selftests/bpf/progs/wq_failures.c | 78 +++++++++++++++++++++++++
 4 files changed, 97 insertions(+)
tools/testing/selftests/bpf/bpf_experimental.h +1 -0
index 3329ea080865..785b91b629be 100644
@@ -470,4 +470,5 @@ extern int bpf_iter_css_new(struct bpf_iter_css *it,
 extern struct cgroup_subsys_state *bpf_iter_css_next(struct bpf_iter_css *it) __weak __ksym;
 extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym;
 
+extern int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags) __weak __ksym;
 #endif
tools/testing/selftests/bpf/prog_tests/wq.c +1 -0
index 9a07b8bc2c52..26ab69796103 100644
@@ -2,6 +2,7 @@
 /* Copyright (c) 2024 Benjamin Tissoires */
 #include <test_progs.h>
 #include "wq.skel.h"
+#include "wq_failures.skel.h"
 
 void serial_test_wq(void)
 {

tools/testing/selftests/bpf/prog_tests/wq.c void serial_test_wq(void) · line 17

Quoted 8 lines
tools/testing/selftests/bpf/prog_tests/wq.c +5 -0
@@ -9,3 +10,10 @@ void serial_test_wq(void)
 
        RUN_TESTS(wq);
 }
+
+void serial_test_failures_wq(void)
+{
+       LIBBPF_OPTS(bpf_test_run_opts, topts);
+

unused leftover?

Quoted 2 lines
tools/testing/selftests/bpf/prog_tests/wq.c +2 -0
+       RUN_TESTS(wq_failures);
+}

[...]

#21 lore raw

Re: [PATCH bpf-next v2 12/16] selftests/bpf: wq: add bpf_wq_init() checks

Quoted 41 lines · Andrii Nakryiko
Quoted 40 lines · Benjamin Tissoires

Allows to test if allocation/free works

---

changes in v2:
- dropped mark_precise checks

---
 tools/testing/selftests/bpf/bpf_experimental.h  |  1 +
 tools/testing/selftests/bpf/prog_tests/wq.c     |  8 +++
 tools/testing/selftests/bpf/progs/wq.c          | 10 ++++
 tools/testing/selftests/bpf/progs/wq_failures.c | 78 +++++++++++++++++++++++++
 4 files changed, 97 insertions(+)
tools/testing/selftests/bpf/bpf_experimental.h +1 -0
index 3329ea080865..785b91b629be 100644
@@ -470,4 +470,5 @@ extern int bpf_iter_css_new(struct bpf_iter_css *it,
 extern struct cgroup_subsys_state *bpf_iter_css_next(struct bpf_iter_css *it) __weak __ksym;
 extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym;
 
+extern int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags) __weak __ksym;
 #endif
tools/testing/selftests/bpf/prog_tests/wq.c +1 -0
index 9a07b8bc2c52..26ab69796103 100644
@@ -2,6 +2,7 @@
 /* Copyright (c) 2024 Benjamin Tissoires */
 #include <test_progs.h>
 #include "wq.skel.h"
+#include "wq_failures.skel.h"
 
 void serial_test_wq(void)
 {
@@ -9,3 +10,10 @@ void serial_test_wq(void)
 
Quoted 9 lines

tools/testing/selftests/bpf/prog_tests/wq.c void serial_test_wq(void) · line 17

Quoted 7 lines
tools/testing/selftests/bpf/prog_tests/wq.c +5 -0
        RUN_TESTS(wq);
 }
+
+void serial_test_failures_wq(void)
+{
+       LIBBPF_OPTS(bpf_test_run_opts, topts);
+

unused leftover?

Oops, yeah. Looks like it.

I'll send a removal of it while fixing 9/16

Cheers,
Benjamin

Quoted 5 lines
Quoted 3 lines
tools/testing/selftests/bpf/prog_tests/wq.c +2 -0
+       RUN_TESTS(wq_failures);
+}

[...]

#22 lore raw

PATCH v2 13/16 bpf: wq: add bpf_wq_set_callback_impl

In reply to Benjamin Tissoires

To support sleepable async callbacks, we need to tell push_async_cb() whether the cb is sleepable or not.

The verifier now detects that we are in bpf_wq_set_callback_impl and can allow a sleepable callback to happen.

---

changes in v2:
- dropped flags check for sleepable
- squashed with v1 14/18 as there is no point in having it separated now

---
 include/linux/bpf_verifier.h |  1 +
 kernel/bpf/helpers.c         | 15 +++++++++
 kernel/bpf/verifier.c        | 75 ++++++++++++++++++++++++++++++++++++++++----
 3 files changed, 85 insertions(+), 6 deletions(-)
include/linux/bpf_verifier.h +1 -0
index 36d19cd32eb5..9db35530c878 100644
@@ -426,6 +426,7 @@ struct bpf_verifier_state {
 	 * while they are still in use.
 	 */
 	bool used_as_loop_entry;
+	bool in_sleepable;
 
 	/* first and last insn idx of this verifier state */
 	u32 first_insn_idx;
kernel/bpf/helpers.c +15 -0
index b12486adc47a..85d3f483f27f 100644
@@ -2711,6 +2711,20 @@ __bpf_kfunc int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags)
 	return __bpf_async_init(async, map, flags, BPF_ASYNC_TYPE_WQ);
 }
 
+__bpf_kfunc int bpf_wq_set_callback_impl(struct bpf_wq *wq,
+					 int (callback_fn)(void *map, int *key, struct bpf_wq *wq),
+					 unsigned int flags,
+					 void *aux__ign)
+{
+	struct bpf_prog_aux *aux = (struct bpf_prog_aux *)aux__ign;
+	struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
+
+	if (flags)
+		return -EINVAL;
+
+	return __bpf_async_set_callback(async, callback_fn, aux, flags, BPF_ASYNC_TYPE_WQ);
+}
+
 __bpf_kfunc_end_defs();
 
 BTF_KFUNCS_START(generic_btf_ids)
@@ -2789,6 +2803,7 @@ BTF_ID_FLAGS(func, bpf_dynptr_size)
 BTF_ID_FLAGS(func, bpf_dynptr_clone)
 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
 BTF_ID_FLAGS(func, bpf_wq_init)
+BTF_ID_FLAGS(func, bpf_wq_set_callback_impl)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
kernel/bpf/verifier.c +69 -6

175 lines not on this page · raw message

Signature · 1 line
2.44.0
#23 lore raw

Re: [PATCH bpf-next v2 13/16] bpf: wq: add bpf_wq_set_callback_impl

Quoted 17 lines · Benjamin Tissoires
kernel/bpf/verifier.c +2 -0
@@ -11018,6 +11027,7 @@ enum special_kfunc_type {
        KF_bpf_percpu_obj_drop_impl,
        KF_bpf_throw,
        KF_bpf_iter_css_task_new,
+       KF_bpf_wq_set_callback_impl,
 };
 
 BTF_SET_START(special_kfunc_set)
@@ -11044,6 +11054,7 @@ BTF_ID(func, bpf_throw)
 #ifdef CONFIG_CGROUPS
 BTF_ID(func, bpf_iter_css_task_new)
 #endif
+BTF_ID(func, bpf_wq_set_callback_impl)
 BTF_SET_END(special_kfunc_set)
 
 BTF_ID_LIST(special_kfunc_list)

kernel/bpf/verifier.c BTF_ID(func, bpf_iter_css_task_new) · line 11088

Quoted 4 lines
kernel/bpf/verifier.c +1 -0
@@ -11074,6 +11085,7 @@ BTF_ID(func, bpf_iter_css_task_new)
 #else
 BTF_ID_UNUSED
 #endif
+BTF_ID(func, bpf_wq_set_callback_impl)

This is broken on !CONFIG_CGROUPS. KF_bpf_wq_set_callback_impl in enum special_kfunc_type won't match special_kfunc_list.
I moved this line up while applying.

Quoted 67 lines
kernel/bpf/verifier.c +33 -0
 static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta)
 {
@@ -11402,12 +11414,28 @@ static bool is_sync_callback_calling_kfunc(u32 btf_id)
        return btf_id == special_kfunc_list[KF_bpf_rbtree_add_impl];
 }
 
+static bool is_async_callback_calling_kfunc(u32 btf_id)
+{
+       return btf_id == special_kfunc_list[KF_bpf_wq_set_callback_impl];
+}
+
 static bool is_bpf_throw_kfunc(struct bpf_insn *insn)
 {
        return bpf_pseudo_kfunc_call(insn) && insn->off == 0 &&
               insn->imm == special_kfunc_list[KF_bpf_throw];
 }
 
+static bool is_bpf_wq_set_callback_impl_kfunc(u32 btf_id)
+{
+       return btf_id == special_kfunc_list[KF_bpf_wq_set_callback_impl];
+}
+
+static bool is_callback_calling_kfunc(u32 btf_id)
+{
+       return is_sync_callback_calling_kfunc(btf_id) ||
+              is_async_callback_calling_kfunc(btf_id);
+}
+
 static bool is_rbtree_lock_required_kfunc(u32 btf_id)
 {
        return is_bpf_rbtree_api_kfunc(btf_id);
@@ -12219,6 +12247,16 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
                }
        }
 
+       if (is_bpf_wq_set_callback_impl_kfunc(meta.func_id)) {
+               err = push_callback_call(env, insn, insn_idx, meta.subprogno,
+                                        set_timer_callback_state);
+               if (err) {
+                       verbose(env, "kfunc %s#%d failed callback verification\n",
+                               func_name, meta.func_id);
+                       return err;
+               }
+       }
+
        rcu_lock = is_kfunc_bpf_rcu_read_lock(&meta);
        rcu_unlock = is_kfunc_bpf_rcu_read_unlock(&meta);
 
@@ -16982,6 +17020,9 @@ static bool states_equal(struct bpf_verifier_env *env,
        if (old->active_rcu_lock != cur->active_rcu_lock)
                return false;
 
+       if (old->in_sleepable != cur->in_sleepable)
+               return false;
+
        /* for states to be equal callsites have to be the same
         * and all frame states need to be equivalent
         */
@@ -19653,6 +19694,28 @@ static int fixup_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
                   desc->func_id == special_kfunc_list[KF_bpf_rdonly_cast]) {
                insn_buf[0] = BPF_MOV64_REG(BPF_REG_0, BPF_REG_1);
                *cnt = 1;
+       } else if (is_bpf_wq_set_callback_impl_kfunc(desc->func_id)) {
+               /* The verifier will process callback_fn as many times as necessary
+                * with different maps and the register states prepared by
+                * set_timer_callback_state will be accurate.

kernel/bpf/verifier.c static int fixup_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, · line 19710

Quoted 10 lines
kernel/bpf/verifier.c +10 -0
+                *
+                * The following use case is valid:
+                *   map1 is shared by prog1, prog2, prog3.
+                *   prog1 calls bpf_timer_init for some map1 elements
+                *   prog2 calls bpf_timer_set_callback for some map1 elements.
+                *     Those that were not bpf_timer_init-ed will return -EINVAL.
+                *   prog3 calls bpf_timer_start for some map1 elements.
+                *     Those that were not both bpf_timer_init-ed and
+                *     bpf_timer_set_callback-ed will return -EINVAL.
+                */

Also removed this comment. It's not correct here.

Quoted 15 lines
kernel/bpf/verifier.c +8 -1
+               struct bpf_insn ld_addrs[2] = {
+                       BPF_LD_IMM64(BPF_REG_4, (long)env->prog->aux),
+               };
+
+               insn_buf[0] = ld_addrs[0];
+               insn_buf[1] = ld_addrs[1];
+               insn_buf[2] = *insn;
+               *cnt = 3;
        }
        return 0;
 }
 
--

2.44.0

#24 lore raw

PATCH v2 14/16 selftests/bpf: add checks for bpf_wq_set_callback()

In reply to Benjamin Tissoires

We assign the callback and set everything up.
The actual tests of these callbacks will be done when bpf_wq_start() is available.

---

changes in v2:
- removed mark_precise checks
- adapted without BPF_F_WQ_SLEEPABLE

---
 tools/testing/selftests/bpf/bpf_experimental.h     |  5 ++
 .../selftests/bpf/bpf_testmod/bpf_testmod.c        |  5 ++
 .../selftests/bpf/bpf_testmod/bpf_testmod_kfunc.h  |  1 +
 tools/testing/selftests/bpf/progs/wq.c             | 41 +++++++++++---
 tools/testing/selftests/bpf/progs/wq_failures.c    | 66 ++++++++++++++++++++++
 5 files changed, 111 insertions(+), 7 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +5 -0
index 785b91b629be..b80b39f76034 100644
@@ -471,4 +471,9 @@ extern struct cgroup_subsys_state *bpf_iter_css_next(struct bpf_iter_css *it) __
 extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym;
 
 extern int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags) __weak __ksym;
+extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
+		int (callback_fn)(void *map, int *key, struct bpf_wq *wq),
+		unsigned int flags__k, void *aux__ign) __ksym;
+#define bpf_wq_set_callback(timer, cb, flags) \
+	bpf_wq_set_callback_impl(timer, cb, flags, NULL)
 #endif
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +5 -0
index 39ad96a18123..eb2b78552ca2 100644
@@ -494,6 +494,10 @@ __bpf_kfunc static u32 bpf_kfunc_call_test_static_unused_arg(u32 arg, u32 unused
 	return arg;
 }
 
+__bpf_kfunc void bpf_kfunc_call_test_sleepable(void)
+{
+}
+
 BTF_KFUNCS_START(bpf_testmod_check_kfunc_ids)
 BTF_ID_FLAGS(func, bpf_testmod_test_mod_kfunc)
 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
@@ -520,6 +524,7 @@ BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS | KF_RCU)
 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE)
 BTF_ID_FLAGS(func, bpf_kfunc_call_test_static_unused_arg)
 BTF_ID_FLAGS(func, bpf_kfunc_call_test_offset)
+BTF_ID_FLAGS(func, bpf_kfunc_call_test_sleepable, KF_SLEEPABLE)
 BTF_KFUNCS_END(bpf_testmod_check_kfunc_ids)
 
 static int bpf_testmod_ops_init(struct btf *btf)
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod_kfunc.h +1 -0
index 7c664dd61059..ce5cd763561c 100644
@@ -96,6 +96,7 @@ void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p) __ksym;
 void bpf_kfunc_call_test_mem_len_fail2(__u64 *mem, int len) __ksym;
 
 void bpf_kfunc_call_test_destructive(void) __ksym;
+void bpf_kfunc_call_test_sleepable(void) __ksym;
 
 void bpf_kfunc_call_test_offset(struct prog_test_ref_kfunc *p);
 struct prog_test_member *bpf_kfunc_call_memb_acquire(void);
tools/testing/selftests/bpf/progs/wq.c +34 -7

90 lines not on this page · raw message

tools/testing/selftests/bpf/progs/wq_failures.c +66 -0

75 lines not on this page · raw message

Signature · 1 line
2.44.0
#25 lore raw

PATCH v2 15/16 bpf: add bpf_wq_start

In reply to Benjamin Tissoires

again, copy/paste from bpf_timer_start().

---

changes in v2:
- dropped spinlock and use READ_ONCE() instead

---
 kernel/bpf/helpers.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)
kernel/bpf/helpers.c +18 -0
index 85d3f483f27f..07550a657d7f 100644
@@ -2711,6 +2711,23 @@ __bpf_kfunc int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags)
 	return __bpf_async_init(async, map, flags, BPF_ASYNC_TYPE_WQ);
 }
 
+__bpf_kfunc int bpf_wq_start(struct bpf_wq *wq, unsigned int flags)
+{
+	struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
+	struct bpf_work *w;
+
+	if (in_nmi())
+		return -EOPNOTSUPP;
+	if (flags)
+		return -EINVAL;
+	w = READ_ONCE(async->work);
+	if (!w || !READ_ONCE(w->cb.prog))
+		return -EINVAL;
+
+	schedule_work(&w->work);
+	return 0;
+}
+
 __bpf_kfunc int bpf_wq_set_callback_impl(struct bpf_wq *wq,
 					 int (callback_fn)(void *map, int *key, struct bpf_wq *wq),
 					 unsigned int flags,
@@ -2804,6 +2821,7 @@ BTF_ID_FLAGS(func, bpf_dynptr_clone)
 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
 BTF_ID_FLAGS(func, bpf_wq_init)
 BTF_ID_FLAGS(func, bpf_wq_set_callback_impl)
+BTF_ID_FLAGS(func, bpf_wq_start)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
Signature · 1 line
2.44.0
#26 lore raw

PATCH v2 16/16 selftests/bpf: wq: add bpf_wq_start() checks

In reply to Benjamin Tissoires

Allows to test if allocation/free works

---

changes in v2:
- remove #define CLOCK_MONOTONIC 1 leftover

---
 tools/testing/selftests/bpf/bpf_experimental.h |  1 +
 tools/testing/selftests/bpf/prog_tests/wq.c    | 22 ++++++++++++++++++++++
 tools/testing/selftests/bpf/progs/wq.c         | 20 +++++++++++++++++---
 3 files changed, 40 insertions(+), 3 deletions(-)
tools/testing/selftests/bpf/bpf_experimental.h +1 -0
index b80b39f76034..93c5a6c446b3 100644
@@ -471,6 +471,7 @@ extern struct cgroup_subsys_state *bpf_iter_css_next(struct bpf_iter_css *it) __
 extern void bpf_iter_css_destroy(struct bpf_iter_css *it) __weak __ksym;
 
 extern int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags) __weak __ksym;
+extern int bpf_wq_start(struct bpf_wq *wq, unsigned int flags) __weak __ksym;
 extern int bpf_wq_set_callback_impl(struct bpf_wq *wq,
 		int (callback_fn)(void *map, int *key, struct bpf_wq *wq),
 		unsigned int flags__k, void *aux__ign) __ksym;
tools/testing/selftests/bpf/prog_tests/wq.c +22 -0
index 26ab69796103..8a4a91d944cc 100644
@@ -6,9 +6,31 @@
 
 void serial_test_wq(void)
 {
+	struct wq *wq_skel = NULL;
+	int err, prog_fd;
+
 	LIBBPF_OPTS(bpf_test_run_opts, topts);
 
 	RUN_TESTS(wq);
+
+	/* re-run the success test to check if the timer was actually executed */
+
+	wq_skel = wq__open_and_load();
+	if (!ASSERT_OK_PTR(wq_skel, "wq_skel_load"))
+		return;
+
+	err = wq__attach(wq_skel);
+	if (!ASSERT_OK(err, "wq_attach"))
+		return;
+
+	prog_fd = bpf_program__fd(wq_skel->progs.test_syscall_array_sleepable);
+	err = bpf_prog_test_run_opts(prog_fd, &topts);
+	ASSERT_OK(err, "test_run");
+	ASSERT_EQ(topts.retval, 0, "test_run");
+
+	usleep(50); /* 10 usecs should be enough, but give it extra */
+
+	ASSERT_EQ(wq_skel->bss->ok_sleepable, (1 << 1), "ok_sleepable");
 }
 
 void serial_test_failures_wq(void)
tools/testing/selftests/bpf/progs/wq.c +17 -3
index f746296d32b1..8011a27a2fbd 100644
@@ -49,12 +49,19 @@ struct {
 	__type(value, struct elem);
 } lru SEC(".maps");
 
+__u32 ok;
+__u32 ok_sleepable;
+
 static int test_elem_callback(void *map, int *key,
 		int (callback_fn)(void *map, int *key, struct bpf_wq *wq))
 {
 	struct elem init = {}, *val;
 	struct bpf_wq *wq;
 
+	if ((ok & (1 << *key) ||
+	    (ok_sleepable & (1 << *key))))
+		return -22;
+
 	if (map == &lru &&
 	    bpf_map_update_elem(map, key, &init, 0))
 		return -1;
@@ -70,6 +77,9 @@ static int test_elem_callback(void *map, int *key,
 	if (bpf_wq_set_callback(wq, callback_fn, 0))
 		return -4;
 
+	if (bpf_wq_start(wq, 0))
+		return -5;
+
 	return 0;
 }
 
@@ -79,6 +89,10 @@ static int test_hmap_elem_callback(void *map, int *key,
 	struct hmap_elem init = {}, *val;
 	struct bpf_wq *wq;
 
+	if ((ok & (1 << *key) ||
+	    (ok_sleepable & (1 << *key))))
+		return -22;
+
 	if (bpf_map_update_elem(map, key, &init, 0))
 		return -1;
 
@@ -93,12 +107,12 @@ static int test_hmap_elem_callback(void *map, int *key,
 	if (bpf_wq_set_callback(wq, callback_fn, 0))
 		return -4;
 
+	if (bpf_wq_start(wq, 0))
+		return -5;
+
 	return 0;
 }
 
-__u32 ok;
-__u32 ok_sleepable;
-
 /* callback for non sleepable workqueue */
 static int wq_callback(void *map, int *key, struct bpf_wq *work)
 {
Signature · 1 line
2.44.0
#27 lore raw
In reply to Benjamin Tissoires
This series was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

This series was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:

Quoted 9 lines · On Sat, 20 Apr 2024 11:09:00 +0200 you wrote:

This is a followup of sleepable bpf_timer[0].

When discussing sleepable bpf_timer, it was thought that we should give a try to bpf_wq, as the 2 APIs are similar but distinct enough to justify a new one.

So here it is.

[...]

Here is the summary with links:
- [bpf-next,v2,01/16] bpf: make timer data struct more generic

    https://git.kernel.org/bpf/bpf-next/c/be2749beff62
  - [bpf-next,v2,02/16] bpf: replace bpf_timer_init with a generic helper
    https://git.kernel.org/bpf/bpf-next/c/56b4a177ae63
  - [bpf-next,v2,03/16] bpf: replace bpf_timer_set_callback with a generic helper
    https://git.kernel.org/bpf/bpf-next/c/073f11b02643
  - [bpf-next,v2,04/16] bpf: replace bpf_timer_cancel_and_free with a generic helper
    https://git.kernel.org/bpf/bpf-next/c/fc22d9495f0b
  - [bpf-next,v2,05/16] bpf: add support for bpf_wq user type
    https://git.kernel.org/bpf/bpf-next/c/d56b63cf0c0f
  - [bpf-next,v2,06/16] tools: sync include/uapi/linux/bpf.h
    https://git.kernel.org/bpf/bpf-next/c/f1d0a2fbb008
  - [bpf-next,v2,07/16] bpf: verifier: bail out if the argument is not a map
    https://git.kernel.org/bpf/bpf-next/c/ad2c03e691be
  - [bpf-next,v2,08/16] bpf: add support for KF_ARG_PTR_TO_WORKQUEUE
    https://git.kernel.org/bpf/bpf-next/c/d940c9b94d7e
  - [bpf-next,v2,09/16] bpf: allow struct bpf_wq to be embedded in arraymaps and hashmaps
    https://git.kernel.org/bpf/bpf-next/c/246331e3f1ea
  - [bpf-next,v2,10/16] selftests/bpf: add bpf_wq tests
    https://git.kernel.org/bpf/bpf-next/c/b4abee7c1ae3
  - [bpf-next,v2,11/16] bpf: wq: add bpf_wq_init
    https://git.kernel.org/bpf/bpf-next/c/eb48f6cd41a0
  - [bpf-next,v2,12/16] selftests/bpf: wq: add bpf_wq_init() checks
    https://git.kernel.org/bpf/bpf-next/c/e3d9eac99afd
  - [bpf-next,v2,13/16] bpf: wq: add bpf_wq_set_callback_impl
    https://git.kernel.org/bpf/bpf-next/c/81f1d7a583fa
  - [bpf-next,v2,14/16] selftests/bpf: add checks for bpf_wq_set_callback()
    https://git.kernel.org/bpf/bpf-next/c/01b7b1c5f3cc
  - [bpf-next,v2,15/16] bpf: add bpf_wq_start
    https://git.kernel.org/bpf/bpf-next/c/8e83da9732d9
  - [bpf-next,v2,16/16] selftests/bpf: wq: add bpf_wq_start() checks
    https://git.kernel.org/bpf/bpf-next/c/8290dba51910

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html