Merge branch 'optimize-bpf_tail_call'

Daniel Borkmann says:

====================
This gets rid of indirect jumps for BPF tail calls whenever possible.
The series adds emission for *direct* jumps for tail call maps in order
to avoid the retpoline overhead from a493a87f38cf ("bpf, x64: implement
retpoline for tail call") for situations that allow for it, meaning,
for known constant keys at verification time which are used as index
into the tail call map. See patch 7/8 for more general details.

Thanks!

v1  -> v2:
  - added more test cases
  - u8 ip_stable -> bool (Andrii)
  - removed bpf_map_poke_{un,}lock and simplified the code (Andrii)
  - added break into prog_array_map_poke_untrack since there's just
    one prog (Andrii)
  - fixed typo: for for in commit msg (Andrii)
  - reworked __bpf_arch_text_poke (Andrii)
  - added subtests, and comment on tests themselves, NULL-NULL
    transistion (Andrii)
  - in constant map key tracking I've moved the map_poke_track callback
    to once we've finished creating the poke tab as otherwise concurrent
    access from tail call map would blow up (since we realloc the table)
rfc -> v1:
  - Applied Alexei's and Andrii's feeback from
    https://lore.kernel.org/bpf/cover.1573779287.git.daniel@iogearbox.net/T/#t
====================

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

1 files changed, 209 insertions, 6 deletions

diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
index 633c8c701ff6..58bdf5fd24cc 100644
--- a/kernel/bpf/arraymap.c
+++ b/kernel/bpf/arraymap.c
@@ -586,10 +586,17 @@ int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
 	if (IS_ERR(new_ptr))
 		return PTR_ERR(new_ptr);
 
-	old_ptr = xchg(array->ptrs + index, new_ptr);
+	if (map->ops->map_poke_run) {
+		mutex_lock(&array->aux->poke_mutex);
+		old_ptr = xchg(array->ptrs + index, new_ptr);
+		map->ops->map_poke_run(map, index, old_ptr, new_ptr);
+		mutex_unlock(&array->aux->poke_mutex);
+	} else {
+		old_ptr = xchg(array->ptrs + index, new_ptr);
+	}
+
 	if (old_ptr)
 		map->ops->map_fd_put_ptr(old_ptr);
-
 	return 0;
 }
 
@@ -602,7 +609,15 @@ static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
 	if (index >= array->map.max_entries)
 		return -E2BIG;
 
-	old_ptr = xchg(array->ptrs + index, NULL);
+	if (map->ops->map_poke_run) {
+		mutex_lock(&array->aux->poke_mutex);
+		old_ptr = xchg(array->ptrs + index, NULL);
+		map->ops->map_poke_run(map, index, old_ptr, NULL);
+		mutex_unlock(&array->aux->poke_mutex);
+	} else {
+		old_ptr = xchg(array->ptrs + index, NULL);
+	}
+
 	if (old_ptr) {
 		map->ops->map_fd_put_ptr(old_ptr);
 		return 0;
@@ -671,17 +686,205 @@ static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
 	rcu_read_unlock();
 }
 
+struct prog_poke_elem {
+	struct list_head list;
+	struct bpf_prog_aux *aux;
+};
+
+static int prog_array_map_poke_track(struct bpf_map *map,
+				     struct bpf_prog_aux *prog_aux)
+{
+	struct prog_poke_elem *elem;
+	struct bpf_array_aux *aux;
+	int ret = 0;
+
+	aux = container_of(map, struct bpf_array, map)->aux;
+	mutex_lock(&aux->poke_mutex);
+	list_for_each_entry(elem, &aux->poke_progs, list) {
+		if (elem->aux == prog_aux)
+			goto out;
+	}
+
+	elem = kmalloc(sizeof(*elem), GFP_KERNEL);
+	if (!elem) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	INIT_LIST_HEAD(&elem->list);
+	/* We must track the program's aux info at this point in time
+	 * since the program pointer itself may not be stable yet, see
+	 * also comment in prog_array_map_poke_run().
+	 */
+	elem->aux = prog_aux;
+
+	list_add_tail(&elem->list, &aux->poke_progs);
+out:
+	mutex_unlock(&aux->poke_mutex);
+	return ret;
+}
+
+static void prog_array_map_poke_untrack(struct bpf_map *map,
+					struct bpf_prog_aux *prog_aux)
+{
+	struct prog_poke_elem *elem, *tmp;
+	struct bpf_array_aux *aux;
+
+	aux = container_of(map, struct bpf_array, map)->aux;
+	mutex_lock(&aux->poke_mutex);
+	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
+		if (elem->aux == prog_aux) {
+			list_del_init(&elem->list);
+			kfree(elem);
+			break;
+		}
+	}
+	mutex_unlock(&aux->poke_mutex);
+}
+
+static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
+				    struct bpf_prog *old,
+				    struct bpf_prog *new)
+{
+	enum bpf_text_poke_type type;
+	struct prog_poke_elem *elem;
+	struct bpf_array_aux *aux;
+
+	if (!old && new)
+		type = BPF_MOD_NOP_TO_JUMP;
+	else if (old && !new)
+		type = BPF_MOD_JUMP_TO_NOP;
+	else if (old && new)
+		type = BPF_MOD_JUMP_TO_JUMP;
+	else
+		return;
+
+	aux = container_of(map, struct bpf_array, map)->aux;
+	WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
+
+	list_for_each_entry(elem, &aux->poke_progs, list) {
+		struct bpf_jit_poke_descriptor *poke;
+		int i, ret;
+
+		for (i = 0; i < elem->aux->size_poke_tab; i++) {
+			poke = &elem->aux->poke_tab[i];
+
+			/* Few things to be aware of:
+			 *
+			 * 1) We can only ever access aux in this context, but
+			 *    not aux->prog since it might not be stable yet and
+			 *    there could be danger of use after free otherwise.
+			 * 2) Initially when we start tracking aux, the program
+			 *    is not JITed yet and also does not have a kallsyms
+			 *    entry. We skip these as poke->ip_stable is not
+			 *    active yet. The JIT will do the final fixup before
+			 *    setting it stable. The various poke->ip_stable are
+			 *    successively activated, so tail call updates can
+			 *    arrive from here while JIT is still finishing its
+			 *    final fixup for non-activated poke entries.
+			 * 3) On program teardown, the program's kallsym entry gets
+			 *    removed out of RCU callback, but we can only untrack
+			 *    from sleepable context, therefore bpf_arch_text_poke()
+			 *    might not see that this is in BPF text section and
+			 *    bails out with -EINVAL. As these are unreachable since
+			 *    RCU grace period already passed, we simply skip them.
+			 * 4) Also programs reaching refcount of zero while patching
+			 *    is in progress is okay since we're protected under
+			 *    poke_mutex and untrack the programs before the JIT
+			 *    buffer is freed. When we're still in the middle of
+			 *    patching and suddenly kallsyms entry of the program
+			 *    gets evicted, we just skip the rest which is fine due
+			 *    to point 3).
+			 * 5) Any other error happening below from bpf_arch_text_poke()
+			 *    is a unexpected bug.
+			 */
+			if (!READ_ONCE(poke->ip_stable))
+				continue;
+			if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
+				continue;
+			if (poke->tail_call.map != map ||
+			    poke->tail_call.key != key)
+				continue;
+
+			ret = bpf_arch_text_poke(poke->ip, type,
+						 old ? (u8 *)old->bpf_func +
+						 poke->adj_off : NULL,
+						 new ? (u8 *)new->bpf_func +
+						 poke->adj_off : NULL);
+			BUG_ON(ret < 0 && ret != -EINVAL);
+		}
+	}
+}
+
+static void prog_array_map_clear_deferred(struct work_struct *work)
+{
+	struct bpf_map *map = container_of(work, struct bpf_array_aux,
+					   work)->map;
+	bpf_fd_array_map_clear(map);
+	bpf_map_put(map);
+}
+
+static void prog_array_map_clear(struct bpf_map *map)
+{
+	struct bpf_array_aux *aux = container_of(map, struct bpf_array,
+						 map)->aux;
+	bpf_map_inc(map);
+	schedule_work(&aux->work);
+}
+
+static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
+{
+	struct bpf_array_aux *aux;
+	struct bpf_map *map;
+
+	aux = kzalloc(sizeof(*aux), GFP_KERNEL);
+	if (!aux)
+		return ERR_PTR(-ENOMEM);
+
+	INIT_WORK(&aux->work, prog_array_map_clear_deferred);
+	INIT_LIST_HEAD(&aux->poke_progs);
+	mutex_init(&aux->poke_mutex);
+
+	map = array_map_alloc(attr);
+	if (IS_ERR(map)) {
+		kfree(aux);
+		return map;
+	}
+
+	container_of(map, struct bpf_array, map)->aux = aux;
+	aux->map = map;
+
+	return map;
+}
+
+static void prog_array_map_free(struct bpf_map *map)
+{
+	struct prog_poke_elem *elem, *tmp;
+	struct bpf_array_aux *aux;
+
+	aux = container_of(map, struct bpf_array, map)->aux;
+	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
+		list_del_init(&elem->list);
+		kfree(elem);
+	}
+	kfree(aux);
+	fd_array_map_free(map);
+}
+
 const struct bpf_map_ops prog_array_map_ops = {
 	.map_alloc_check = fd_array_map_alloc_check,
-	.map_alloc = array_map_alloc,
-	.map_free = fd_array_map_free,
+	.map_alloc = prog_array_map_alloc,
+	.map_free = prog_array_map_free,
+	.map_poke_track = prog_array_map_poke_track,
+	.map_poke_untrack = prog_array_map_poke_untrack,
+	.map_poke_run = prog_array_map_poke_run,
 	.map_get_next_key = array_map_get_next_key,
 	.map_lookup_elem = fd_array_map_lookup_elem,
 	.map_delete_elem = fd_array_map_delete_elem,
 	.map_fd_get_ptr = prog_fd_array_get_ptr,
 	.map_fd_put_ptr = prog_fd_array_put_ptr,
 	.map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
-	.map_release_uref = bpf_fd_array_map_clear,
+	.map_release_uref = prog_array_map_clear,
 	.map_seq_show_elem = prog_array_map_seq_show_elem,
 };