/*
- * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <config.h>
#include "classifier.h"
+#include "classifier-private.h"
#include <errno.h>
#include <netinet/in.h>
#include "byte-order.h"
#include "dynamic-string.h"
-#include "flow.h"
-#include "hash.h"
-#include "hindex.h"
-#include "hmap.h"
-#include "list.h"
#include "odp-util.h"
#include "ofp-util.h"
-#include "ovs-thread.h"
#include "packets.h"
-#include "tag.h"
#include "util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(classifier);
-struct trie_node;
struct trie_ctx;
/* Ports trie depends on both ports sharing the same ovs_be32. */
#define TP_PORTS_OFS32 (offsetof(struct flow, tp_src) / 4)
BUILD_ASSERT_DECL(TP_PORTS_OFS32 == offsetof(struct flow, tp_dst) / 4);
-/* Prefix trie for a 'field' */
-struct cls_trie {
- const struct mf_field *field; /* Trie field, or NULL. */
- struct trie_node *root; /* NULL if none. */
-};
-
-struct cls_subtable_entry {
- struct cls_subtable *subtable;
- tag_type tag;
- unsigned int max_priority;
-};
-
-struct cls_subtables {
- size_t count; /* One past last valid array element. */
- size_t alloc_size; /* Number of allocated elements. */
- struct cls_subtable_entry *array;
-};
-
-enum {
- CLS_MAX_INDICES = 3 /* Maximum number of lookup indices per subtable. */
-};
-
-struct cls_classifier {
- int n_rules; /* Total number of rules. */
- uint8_t n_flow_segments;
- uint8_t flow_segments[CLS_MAX_INDICES]; /* Flow segment boundaries to use
- * for staged lookup. */
- struct hmap subtables_map; /* Contains "struct cls_subtable"s. */
- struct cls_subtables subtables;
- struct hmap partitions; /* Contains "struct cls_partition"s. */
- struct cls_trie tries[CLS_MAX_TRIES]; /* Prefix tries. */
- unsigned int n_tries;
-};
-
-/* A set of rules that all have the same fields wildcarded. */
-struct cls_subtable {
- struct hmap_node hmap_node; /* Within struct cls_classifier 'subtables_map'
- * hmap. */
- struct hmap rules; /* Contains "struct cls_rule"s. */
- int n_rules; /* Number of rules, including duplicates. */
- unsigned int max_priority; /* Max priority of any rule in the subtable. */
- unsigned int max_count; /* Count of max_priority rules. */
- tag_type tag; /* Tag generated from mask for partitioning. */
- uint8_t n_indices; /* How many indices to use. */
- uint8_t index_ofs[CLS_MAX_INDICES]; /* u32 flow segment boundaries. */
- struct hindex indices[CLS_MAX_INDICES]; /* Staged lookup indices. */
- unsigned int trie_plen[CLS_MAX_TRIES]; /* Trie prefix length in 'mask'. */
- int ports_mask_len;
- struct trie_node *ports_trie; /* NULL if none. */
- struct minimask mask; /* Wildcards for fields. */
- /* 'mask' must be the last field. */
-};
-
-/* Associates a metadata value (that is, a value of the OpenFlow 1.1+ metadata
- * field) with tags for the "cls_subtable"s that contain rules that match that
- * metadata value. */
-struct cls_partition {
- struct hmap_node hmap_node; /* In struct cls_classifier's 'partitions'
- * hmap. */
- ovs_be64 metadata; /* metadata value for this partition. */
- tag_type tags; /* OR of each flow's cls_subtable tag. */
- struct tag_tracker tracker; /* Tracks the bits in 'tags'. */
-};
-
-/* Internal representation of a rule in a "struct cls_subtable". */
-struct cls_match {
- struct cls_rule *cls_rule;
- struct hindex_node index_nodes[CLS_MAX_INDICES]; /* Within subtable's
- * 'indices'. */
- struct hmap_node hmap_node; /* Within struct cls_subtable 'rules'. */
- unsigned int priority; /* Larger numbers are higher priorities. */
- struct cls_partition *partition;
- struct list list; /* List of identical, lower-priority rules. */
- struct miniflow flow; /* Matching rule. Mask is in the subtable. */
- /* 'flow' must be the last field. */
-};
-
static struct cls_match *
cls_match_alloc(struct cls_rule *rule)
{
= xmalloc(sizeof *cls_match - sizeof cls_match->flow.inline_values
+ MINIFLOW_VALUES_SIZE(count));
- cls_match->cls_rule = rule;
- miniflow_clone_inline(&cls_match->flow, &rule->match.flow, count);
- cls_match->priority = rule->priority;
+ *CONST_CAST(const struct cls_rule **, &cls_match->cls_rule) = rule;
+ *CONST_CAST(int *, &cls_match->priority) = rule->priority;
+ miniflow_clone_inline(CONST_CAST(struct miniflow *, &cls_match->flow),
+ &rule->match.flow, count);
rule->cls_match = cls_match;
return cls_match;
}
-static struct cls_subtable *find_subtable(const struct cls_classifier *,
+static struct cls_subtable *find_subtable(const struct classifier *cls,
const struct minimask *);
-static struct cls_subtable *insert_subtable(struct cls_classifier *,
- const struct minimask *);
-
-static void destroy_subtable(struct cls_classifier *, struct cls_subtable *);
-
-static void update_subtables_after_insertion(struct cls_classifier *,
- struct cls_subtable *,
- unsigned int new_priority);
-static void update_subtables_after_removal(struct cls_classifier *,
- struct cls_subtable *,
- unsigned int del_priority);
-
-static struct cls_match *find_match_wc(const struct cls_subtable *,
- const struct flow *, struct trie_ctx *,
- unsigned int n_tries,
- struct flow_wildcards *);
-static struct cls_match *find_equal(struct cls_subtable *,
+static struct cls_subtable *insert_subtable(struct classifier *cls,
+ const struct minimask *)
+ OVS_REQUIRES(cls->mutex);
+static void destroy_subtable(struct classifier *cls, struct cls_subtable *)
+ OVS_REQUIRES(cls->mutex);
+static struct cls_match *insert_rule(struct classifier *cls,
+ struct cls_subtable *, struct cls_rule *)
+ OVS_REQUIRES(cls->mutex);
+
+static const struct cls_match *find_match_wc(const struct cls_subtable *,
+ const struct flow *,
+ struct trie_ctx *,
+ unsigned int n_tries,
+ struct flow_wildcards *);
+static struct cls_match *find_equal(const struct cls_subtable *,
const struct miniflow *, uint32_t hash);
-static struct cls_match *insert_rule(struct cls_classifier *,
- struct cls_subtable *, struct cls_rule *);
-
-/* Iterates RULE over HEAD and all of the cls_rules on HEAD->list. */
-#define FOR_EACH_RULE_IN_LIST(RULE, HEAD) \
- for ((RULE) = (HEAD); (RULE) != NULL; (RULE) = next_rule_in_list(RULE))
-#define FOR_EACH_RULE_IN_LIST_SAFE(RULE, NEXT, HEAD) \
- for ((RULE) = (HEAD); \
- (RULE) != NULL && ((NEXT) = next_rule_in_list(RULE), true); \
- (RULE) = (NEXT))
-
-static struct cls_match *next_rule_in_list__(struct cls_match *);
-static struct cls_match *next_rule_in_list(struct cls_match *);
-
-static unsigned int minimask_get_prefix_len(const struct minimask *,
- const struct mf_field *);
-static void trie_init(struct cls_classifier *, int trie_idx,
- const struct mf_field *);
-static unsigned int trie_lookup(const struct cls_trie *, const struct flow *,
- unsigned int *checkbits);
-static unsigned int trie_lookup_value(const struct trie_node *,
- const ovs_be32 value[],
- unsigned int value_bits,
- unsigned int *checkbits);
-static void trie_destroy(struct trie_node *);
-static void trie_insert(struct cls_trie *, const struct cls_rule *, int mlen);
-static void trie_insert_prefix(struct trie_node **, const ovs_be32 *prefix,
- int mlen);
-static void trie_remove(struct cls_trie *, const struct cls_rule *, int mlen);
-static void trie_remove_prefix(struct trie_node **, const ovs_be32 *prefix,
- int mlen);
-static void mask_set_prefix_bits(struct flow_wildcards *, uint8_t be32ofs,
- unsigned int nbits);
-static bool mask_prefix_bits_set(const struct flow_wildcards *,
- uint8_t be32ofs, unsigned int nbits);
-
-static void
-cls_subtables_init(struct cls_subtables *subtables)
-{
- memset(subtables, 0, sizeof *subtables);
-}
-
-static void
-cls_subtables_destroy(struct cls_subtables *subtables)
-{
- free(subtables->array);
- memset(subtables, 0, sizeof *subtables);
-}
-
-/* Subtables insertion. */
-static void
-cls_subtables_push_back(struct cls_subtables *subtables,
- struct cls_subtable_entry a)
-{
- if (subtables->count == subtables->alloc_size) {
- subtables->array = x2nrealloc(subtables->array, &subtables->alloc_size,
- sizeof a);
- }
-
- subtables->array[subtables->count++] = a;
-}
-
-/* Move subtable entry at 'from' to 'to', shifting the elements in between
- * (including the one at 'to') accordingly. */
-static inline void
-cls_subtables_move(struct cls_subtable_entry *to,
- struct cls_subtable_entry *from)
-{
- if (to != from) {
- struct cls_subtable_entry temp = *from;
-
- if (to > from) {
- /* Shift entries (from,to] backwards to make space at 'to'. */
- memmove(from, from + 1, (to - from) * sizeof *to);
- } else {
- /* Shift entries [to,from) forward to make space at 'to'. */
- memmove(to + 1, to, (from - to) * sizeof *to);
- }
-
- *to = temp;
- }
-}
-
-/* Subtables removal. */
-static inline void
-cls_subtables_remove(struct cls_subtables *subtables,
- struct cls_subtable_entry *elem)
-{
- ssize_t size = (&subtables->array[subtables->count]
- - (elem + 1)) * sizeof *elem;
- if (size > 0) {
- memmove(elem, elem + 1, size);
- }
- subtables->count--;
-}
-
-#define CLS_SUBTABLES_FOR_EACH(SUBTABLE, ITER, SUBTABLES) \
- for ((ITER) = (SUBTABLES)->array; \
- (ITER) < &(SUBTABLES)->array[(SUBTABLES)->count] \
- && OVS_LIKELY((SUBTABLE) = (ITER)->subtable); \
- ++(ITER))
-#define CLS_SUBTABLES_FOR_EACH_CONTINUE(SUBTABLE, ITER, SUBTABLES) \
- for (++(ITER); \
- (ITER) < &(SUBTABLES)->array[(SUBTABLES)->count] \
- && OVS_LIKELY((SUBTABLE) = (ITER)->subtable); \
- ++(ITER))
-#define CLS_SUBTABLES_FOR_EACH_REVERSE(SUBTABLE, ITER, SUBTABLES) \
- for ((ITER) = &(SUBTABLES)->array[(SUBTABLES)->count]; \
- (ITER) > (SUBTABLES)->array \
- && OVS_LIKELY((SUBTABLE) = (--(ITER))->subtable);)
-
-static void
-cls_subtables_verify(struct cls_subtables *subtables)
-{
- struct cls_subtable *table;
- struct cls_subtable_entry *iter;
- unsigned int priority = 0;
-
- CLS_SUBTABLES_FOR_EACH_REVERSE (table, iter, subtables) {
- if (iter->max_priority != table->max_priority) {
- VLOG_WARN("Subtable %p has mismatching priority in cache (%u != %u)",
- table, iter->max_priority, table->max_priority);
- }
- if (iter->max_priority < priority) {
- VLOG_WARN("Subtable cache is out of order (%u < %u)",
- iter->max_priority, priority);
- }
- priority = iter->max_priority;
- }
-}
-
-static void
-cls_subtables_reset(struct cls_classifier *cls)
-{
- struct cls_subtables old = cls->subtables;
- struct cls_subtable *subtable;
-
- VLOG_WARN("Resetting subtable cache.");
-
- cls_subtables_verify(&cls->subtables);
-
- cls_subtables_init(&cls->subtables);
-
- HMAP_FOR_EACH (subtable, hmap_node, &cls->subtables_map) {
- struct cls_match *head;
- struct cls_subtable_entry elem;
- struct cls_subtable *table;
- struct cls_subtable_entry *iter, *from = NULL;
- unsigned int new_max = 0;
- unsigned int max_count = 0;
- bool found;
-
- /* Verify max_priority. */
- HMAP_FOR_EACH (head, hmap_node, &subtable->rules) {
- if (head->priority > new_max) {
- new_max = head->priority;
- max_count = 1;
- } else if (head->priority == new_max) {
- max_count++;
- }
- }
- if (new_max != subtable->max_priority ||
- max_count != subtable->max_count) {
- VLOG_WARN("subtable %p (%u rules) has mismatching max_priority "
- "(%u) or max_count (%u). Highest priority found was %u, "
- "count: %u",
- subtable, subtable->n_rules, subtable->max_priority,
- subtable->max_count, new_max, max_count);
- subtable->max_priority = new_max;
- subtable->max_count = max_count;
- }
-
- /* Locate the subtable from the old cache. */
- found = false;
- CLS_SUBTABLES_FOR_EACH (table, iter, &old) {
- if (table == subtable) {
- if (iter->max_priority != new_max) {
- VLOG_WARN("Subtable %p has wrong max priority (%u != %u) "
- "in the old cache.",
- subtable, iter->max_priority, new_max);
- }
- if (found) {
- VLOG_WARN("Subtable %p duplicated in the old cache.",
- subtable);
- }
- found = true;
- }
- }
- if (!found) {
- VLOG_WARN("Subtable %p not found from the old cache.", subtable);
- }
-
- elem.subtable = subtable;
- elem.tag = subtable->tag;
- elem.max_priority = subtable->max_priority;
- cls_subtables_push_back(&cls->subtables, elem);
-
- /* Possibly move 'subtable' earlier in the priority array. If
- * we break out of the loop, then the subtable (at 'from')
- * should be moved to the position right after the current
- * element. If the loop terminates normally, then 'iter' will
- * be at the first array element and we'll move the subtable
- * to the front of the array. */
- CLS_SUBTABLES_FOR_EACH_REVERSE (table, iter, &cls->subtables) {
- if (table == subtable) {
- from = iter; /* Locate the subtable as we go. */
- } else if (table->max_priority >= new_max) {
- ovs_assert(from != NULL);
- iter++; /* After this. */
- break;
- }
- }
-
- /* Move subtable at 'from' to 'iter'. */
- cls_subtables_move(iter, from);
- }
-
- /* Verify that the old and the new have the same size. */
- if (old.count != cls->subtables.count) {
- VLOG_WARN("subtables cache sizes differ: old (%"PRIuSIZE
- ") != new (%"PRIuSIZE").",
- old.count, cls->subtables.count);
- }
-
- cls_subtables_destroy(&old);
-
- cls_subtables_verify(&cls->subtables);
-}
-
-\f
-/* flow/miniflow/minimask/minimatch utilities.
- * These are only used by the classifier, so place them here to allow
- * for better optimization. */
-
-static inline uint64_t
-miniflow_get_map_in_range(const struct miniflow *miniflow,
- uint8_t start, uint8_t end, unsigned int *offset)
-{
- uint64_t map = miniflow->map;
- *offset = 0;
-
- if (start > 0) {
- uint64_t msk = (UINT64_C(1) << start) - 1; /* 'start' LSBs set */
- *offset = count_1bits(map & msk);
- map &= ~msk;
- }
- if (end < FLOW_U32S) {
- uint64_t msk = (UINT64_C(1) << end) - 1; /* 'end' LSBs set */
- map &= msk;
- }
- return map;
-}
-
-/* Returns a hash value for the bits of 'flow' where there are 1-bits in
- * 'mask', given 'basis'.
- *
- * The hash values returned by this function are the same as those returned by
- * miniflow_hash_in_minimask(), only the form of the arguments differ. */
-static inline uint32_t
-flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask,
- uint32_t basis)
-{
- const uint32_t *mask_values = miniflow_get_u32_values(&mask->masks);
- const uint32_t *flow_u32 = (const uint32_t *)flow;
- const uint32_t *p = mask_values;
- uint32_t hash;
- uint64_t map;
-
- hash = basis;
- for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) {
- hash = mhash_add(hash, flow_u32[raw_ctz(map)] & *p++);
- }
-
- return mhash_finish(hash, (p - mask_values) * 4);
-}
-
-/* Returns a hash value for the bits of 'flow' where there are 1-bits in
- * 'mask', given 'basis'.
- *
- * The hash values returned by this function are the same as those returned by
- * flow_hash_in_minimask(), only the form of the arguments differ. */
-static inline uint32_t
-miniflow_hash_in_minimask(const struct miniflow *flow,
- const struct minimask *mask, uint32_t basis)
-{
- const uint32_t *mask_values = miniflow_get_u32_values(&mask->masks);
- const uint32_t *p = mask_values;
- uint32_t hash = basis;
- uint32_t flow_u32;
-
- MINIFLOW_FOR_EACH_IN_MAP(flow_u32, flow, mask->masks.map) {
- hash = mhash_add(hash, flow_u32 & *p++);
- }
- return mhash_finish(hash, (p - mask_values) * 4);
-}
-
-/* Returns a hash value for the bits of range [start, end) in 'flow',
- * where there are 1-bits in 'mask', given 'hash'.
- *
- * The hash values returned by this function are the same as those returned by
- * minimatch_hash_range(), only the form of the arguments differ. */
-static inline uint32_t
-flow_hash_in_minimask_range(const struct flow *flow,
- const struct minimask *mask,
- uint8_t start, uint8_t end, uint32_t *basis)
+static inline const struct cls_match *
+next_rule_in_list__(const struct cls_match *rule)
{
- const uint32_t *mask_values = miniflow_get_u32_values(&mask->masks);
- const uint32_t *flow_u32 = (const uint32_t *)flow;
- unsigned int offset;
- uint64_t map = miniflow_get_map_in_range(&mask->masks, start, end,
- &offset);
- const uint32_t *p = mask_values + offset;
- uint32_t hash = *basis;
-
- for (; map; map = zero_rightmost_1bit(map)) {
- hash = mhash_add(hash, flow_u32[raw_ctz(map)] & *p++);
- }
-
- *basis = hash; /* Allow continuation from the unfinished value. */
- return mhash_finish(hash, (p - mask_values) * 4);
-}
-
-/* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */
-static inline void
-flow_wildcards_fold_minimask(struct flow_wildcards *wc,
- const struct minimask *mask)
-{
- flow_union_with_miniflow(&wc->masks, &mask->masks);
-}
-
-/* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask
- * in range [start, end). */
-static inline void
-flow_wildcards_fold_minimask_range(struct flow_wildcards *wc,
- const struct minimask *mask,
- uint8_t start, uint8_t end)
-{
- uint32_t *dst_u32 = (uint32_t *)&wc->masks;
- unsigned int offset;
- uint64_t map = miniflow_get_map_in_range(&mask->masks, start, end,
- &offset);
- const uint32_t *p = miniflow_get_u32_values(&mask->masks) + offset;
-
- for (; map; map = zero_rightmost_1bit(map)) {
- dst_u32[raw_ctz(map)] |= *p++;
- }
+ const struct cls_match *next = NULL;
+ next = OBJECT_CONTAINING(rculist_next(&rule->list), next, list);
+ return next;
}
-/* Returns a hash value for 'flow', given 'basis'. */
-static inline uint32_t
-miniflow_hash(const struct miniflow *flow, uint32_t basis)
+static inline const struct cls_match *
+next_rule_in_list(const struct cls_match *rule)
{
- const uint32_t *values = miniflow_get_u32_values(flow);
- const uint32_t *p = values;
- uint32_t hash = basis;
- uint64_t hash_map = 0;
- uint64_t map;
-
- for (map = flow->map; map; map = zero_rightmost_1bit(map)) {
- if (*p) {
- hash = mhash_add(hash, *p);
- hash_map |= rightmost_1bit(map);
- }
- p++;
- }
- hash = mhash_add(hash, hash_map);
- hash = mhash_add(hash, hash_map >> 32);
-
- return mhash_finish(hash, p - values);
+ const struct cls_match *next = next_rule_in_list__(rule);
+ return next->priority < rule->priority ? next : NULL;
}
-/* Returns a hash value for 'mask', given 'basis'. */
-static inline uint32_t
-minimask_hash(const struct minimask *mask, uint32_t basis)
+static inline struct cls_match *
+next_rule_in_list_protected__(struct cls_match *rule)
{
- return miniflow_hash(&mask->masks, basis);
+ struct cls_match *next = NULL;
+ next = OBJECT_CONTAINING(rculist_next_protected(&rule->list), next, list);
+ return next;
}
-/* Returns a hash value for 'match', given 'basis'. */
-static inline uint32_t
-minimatch_hash(const struct minimatch *match, uint32_t basis)
+static inline struct cls_match *
+next_rule_in_list_protected(struct cls_match *rule)
{
- return miniflow_hash(&match->flow, minimask_hash(&match->mask, basis));
+ struct cls_match *next = next_rule_in_list_protected__(rule);
+ return next->priority < rule->priority ? next : NULL;
}
-/* Returns a hash value for the bits of range [start, end) in 'minimatch',
- * given 'basis'.
- *
- * The hash values returned by this function are the same as those returned by
- * flow_hash_in_minimask_range(), only the form of the arguments differ. */
-static inline uint32_t
-minimatch_hash_range(const struct minimatch *match, uint8_t start, uint8_t end,
- uint32_t *basis)
-{
- unsigned int offset;
- const uint32_t *p, *q;
- uint32_t hash = *basis;
- int n, i;
-
- n = count_1bits(miniflow_get_map_in_range(&match->mask.masks, start, end,
- &offset));
- q = miniflow_get_u32_values(&match->mask.masks) + offset;
- p = miniflow_get_u32_values(&match->flow) + offset;
-
- for (i = 0; i < n; i++) {
- hash = mhash_add(hash, p[i] & q[i]);
- }
- *basis = hash; /* Allow continuation from the unfinished value. */
- return mhash_finish(hash, (offset + n) * 4);
-}
+/* Iterates RULE over HEAD and all of the cls_rules on HEAD->list.
+ * Classifier's mutex must be held while iterating, as the list is
+ * protoceted by it. */
+#define FOR_EACH_RULE_IN_LIST(RULE, HEAD) \
+ for ((RULE) = (HEAD); (RULE) != NULL; (RULE) = next_rule_in_list(RULE))
+#define FOR_EACH_RULE_IN_LIST_PROTECTED(RULE, HEAD) \
+ for ((RULE) = (HEAD); (RULE) != NULL; \
+ (RULE) = next_rule_in_list_protected(RULE))
+static unsigned int minimask_get_prefix_len(const struct minimask *,
+ const struct mf_field *);
+static void trie_init(struct classifier *cls, int trie_idx,
+ const struct mf_field *)
+ OVS_REQUIRES(cls->mutex);
+static unsigned int trie_lookup(const struct cls_trie *, const struct flow *,
+ union mf_value *plens);
+static unsigned int trie_lookup_value(const rcu_trie_ptr *,
+ const ovs_be32 value[], ovs_be32 plens[],
+ unsigned int value_bits);
+static void trie_destroy(rcu_trie_ptr *);
+static void trie_insert(struct cls_trie *, const struct cls_rule *, int mlen);
+static void trie_insert_prefix(rcu_trie_ptr *, const ovs_be32 *prefix,
+ int mlen);
+static void trie_remove(struct cls_trie *, const struct cls_rule *, int mlen);
+static void trie_remove_prefix(rcu_trie_ptr *, const ovs_be32 *prefix,
+ int mlen);
+static void mask_set_prefix_bits(struct flow_wildcards *, uint8_t be32ofs,
+ unsigned int n_bits);
+static bool mask_prefix_bits_set(const struct flow_wildcards *,
+ uint8_t be32ofs, unsigned int n_bits);
\f
/* cls_rule. */
*
* The caller must eventually destroy 'rule' with cls_rule_destroy().
*
- * (OpenFlow uses priorities between 0 and UINT16_MAX, inclusive, but
- * internally Open vSwitch supports a wider range.) */
+ * Clients should not use priority INT_MIN. (OpenFlow uses priorities between
+ * 0 and UINT16_MAX, inclusive.) */
void
-cls_rule_init(struct cls_rule *rule,
- const struct match *match, unsigned int priority)
+cls_rule_init(struct cls_rule *rule, const struct match *match, int priority)
{
minimatch_init(&rule->match, match);
rule->priority = priority;
/* Same as cls_rule_init() for initialization from a "struct minimatch". */
void
cls_rule_init_from_minimatch(struct cls_rule *rule,
- const struct minimatch *match,
- unsigned int priority)
+ const struct minimatch *match, int priority)
{
minimatch_clone(&rule->match, match);
rule->priority = priority;
/* Initializes 'cls' as a classifier that initially contains no classification
* rules. */
void
-classifier_init(struct classifier *cls_, const uint8_t *flow_segments)
+classifier_init(struct classifier *cls, const uint8_t *flow_segments)
+ OVS_EXCLUDED(cls->mutex)
{
- struct cls_classifier *cls = xmalloc(sizeof *cls);
-
- fat_rwlock_init(&cls_->rwlock);
-
- cls_->cls = cls;
-
+ ovs_mutex_init(&cls->mutex);
+ ovs_mutex_lock(&cls->mutex);
cls->n_rules = 0;
- hmap_init(&cls->subtables_map);
- cls_subtables_init(&cls->subtables);
- hmap_init(&cls->partitions);
+ cmap_init(&cls->subtables_map);
+ pvector_init(&cls->subtables);
+ cmap_init(&cls->partitions);
cls->n_flow_segments = 0;
if (flow_segments) {
while (cls->n_flow_segments < CLS_MAX_INDICES
}
}
cls->n_tries = 0;
+ for (int i = 0; i < CLS_MAX_TRIES; i++) {
+ trie_init(cls, i, NULL);
+ }
+ ovs_mutex_unlock(&cls->mutex);
}
/* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the
- * caller's responsibility. */
+ * caller's responsibility.
+ * May only be called after all the readers have been terminated. */
void
-classifier_destroy(struct classifier *cls_)
+classifier_destroy(struct classifier *cls)
+ OVS_EXCLUDED(cls->mutex)
{
- if (cls_) {
- struct cls_classifier *cls = cls_->cls;
- struct cls_partition *partition, *next_partition;
- struct cls_subtable *subtable, *next_subtable;
+ if (cls) {
+ struct cls_partition *partition;
+ struct cls_subtable *subtable;
int i;
- fat_rwlock_destroy(&cls_->rwlock);
- if (!cls) {
- return;
- }
-
+ ovs_mutex_lock(&cls->mutex);
for (i = 0; i < cls->n_tries; i++) {
- trie_destroy(cls->tries[i].root);
+ trie_destroy(&cls->tries[i].root);
}
- HMAP_FOR_EACH_SAFE (subtable, next_subtable, hmap_node,
- &cls->subtables_map) {
+ CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) {
destroy_subtable(cls, subtable);
}
- hmap_destroy(&cls->subtables_map);
+ cmap_destroy(&cls->subtables_map);
- HMAP_FOR_EACH_SAFE (partition, next_partition, hmap_node,
- &cls->partitions) {
- hmap_remove(&cls->partitions, &partition->hmap_node);
- free(partition);
+ CMAP_FOR_EACH (partition, cmap_node, &cls->partitions) {
+ ovsrcu_postpone(free, partition);
}
- hmap_destroy(&cls->partitions);
+ cmap_destroy(&cls->partitions);
- cls_subtables_destroy(&cls->subtables);
- free(cls);
+ pvector_destroy(&cls->subtables);
+ ovs_mutex_unlock(&cls->mutex);
+ ovs_mutex_destroy(&cls->mutex);
}
}
-/* We use uint64_t as a set for the fields below. */
-BUILD_ASSERT_DECL(MFF_N_IDS <= 64);
-
/* Set the fields for which prefix lookup should be performed. */
-void
-classifier_set_prefix_fields(struct classifier *cls_,
+bool
+classifier_set_prefix_fields(struct classifier *cls,
const enum mf_field_id *trie_fields,
unsigned int n_fields)
+ OVS_EXCLUDED(cls->mutex)
{
- struct cls_classifier *cls = cls_->cls;
- uint64_t fields = 0;
- int i, trie;
+ const struct mf_field * new_fields[CLS_MAX_TRIES];
+ struct mf_bitmap fields = MF_BITMAP_INITIALIZER;
+ int i, n_tries = 0;
+ bool changed = false;
- for (i = 0, trie = 0; i < n_fields && trie < CLS_MAX_TRIES; i++) {
+ ovs_mutex_lock(&cls->mutex);
+ for (i = 0; i < n_fields && n_tries < CLS_MAX_TRIES; i++) {
const struct mf_field *field = mf_from_id(trie_fields[i]);
if (field->flow_be32ofs < 0 || field->n_bits % 32) {
/* Incompatible field. This is the only place where we
continue;
}
- if (fields & (UINT64_C(1) << trie_fields[i])) {
+ if (bitmap_is_set(fields.bm, trie_fields[i])) {
/* Duplicate field, there is no need to build more than
* one index for any one field. */
continue;
}
- fields |= UINT64_C(1) << trie_fields[i];
+ bitmap_set1(fields.bm, trie_fields[i]);
- if (trie >= cls->n_tries || field != cls->tries[trie].field) {
- trie_init(cls, trie, field);
+ new_fields[n_tries] = NULL;
+ if (n_tries >= cls->n_tries || field != cls->tries[n_tries].field) {
+ new_fields[n_tries] = field;
+ changed = true;
+ }
+ n_tries++;
+ }
+
+ if (changed || n_tries < cls->n_tries) {
+ struct cls_subtable *subtable;
+
+ /* Trie configuration needs to change. Disable trie lookups
+ * for the tries that are changing and wait all the current readers
+ * with the old configuration to be done. */
+ changed = false;
+ CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) {
+ for (i = 0; i < cls->n_tries; i++) {
+ if ((i < n_tries && new_fields[i]) || i >= n_tries) {
+ if (subtable->trie_plen[i]) {
+ subtable->trie_plen[i] = 0;
+ changed = true;
+ }
+ }
+ }
+ }
+ /* Synchronize if any readers were using tries. The readers may
+ * temporarily function without the trie lookup based optimizations. */
+ if (changed) {
+ /* ovsrcu_synchronize() functions as a memory barrier, so it does
+ * not matter that subtable->trie_plen is not atomic. */
+ ovsrcu_synchronize();
}
- trie++;
- }
- /* Destroy the rest. */
- for (i = trie; i < cls->n_tries; i++) {
- trie_init(cls, i, NULL);
+ /* Now set up the tries. */
+ for (i = 0; i < n_tries; i++) {
+ if (new_fields[i]) {
+ trie_init(cls, i, new_fields[i]);
+ }
+ }
+ /* Destroy the rest, if any. */
+ for (; i < cls->n_tries; i++) {
+ trie_init(cls, i, NULL);
+ }
+
+ cls->n_tries = n_tries;
+ ovs_mutex_unlock(&cls->mutex);
+ return true;
}
- cls->n_tries = trie;
+
+ ovs_mutex_unlock(&cls->mutex);
+ return false; /* No change. */
}
static void
-trie_init(struct cls_classifier *cls, int trie_idx,
- const struct mf_field *field)
+trie_init(struct classifier *cls, int trie_idx, const struct mf_field *field)
+ OVS_REQUIRES(cls->mutex)
{
struct cls_trie *trie = &cls->tries[trie_idx];
struct cls_subtable *subtable;
- struct cls_subtable_entry *iter;
if (trie_idx < cls->n_tries) {
- trie_destroy(trie->root);
+ trie_destroy(&trie->root);
+ } else {
+ ovsrcu_set_hidden(&trie->root, NULL);
}
- trie->root = NULL;
trie->field = field;
- /* Add existing rules to the trie. */
- CLS_SUBTABLES_FOR_EACH (subtable, iter, &cls->subtables) {
+ /* Add existing rules to the new trie. */
+ CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) {
unsigned int plen;
plen = field ? minimask_get_prefix_len(&subtable->mask, field) : 0;
- /* Initialize subtable's prefix length on this field. */
- subtable->trie_plen[trie_idx] = plen;
-
if (plen) {
struct cls_match *head;
- HMAP_FOR_EACH (head, hmap_node, &subtable->rules) {
+ CMAP_FOR_EACH (head, cmap_node, &subtable->rules) {
struct cls_match *match;
- FOR_EACH_RULE_IN_LIST (match, head) {
+ FOR_EACH_RULE_IN_LIST_PROTECTED (match, head) {
trie_insert(trie, match->cls_rule, plen);
}
}
}
+ /* Initialize subtable's prefix length on this field. This will
+ * allow readers to use the trie. */
+ atomic_thread_fence(memory_order_release);
+ subtable->trie_plen[trie_idx] = plen;
}
}
-/* Returns true if 'cls' contains no classification rules, false otherwise. */
+/* Returns true if 'cls' contains no classification rules, false otherwise.
+ * Checking the cmap requires no locking. */
bool
classifier_is_empty(const struct classifier *cls)
{
- return cls->cls->n_rules == 0;
+ return cmap_is_empty(&cls->subtables_map);
}
/* Returns the number of rules in 'cls'. */
int
classifier_count(const struct classifier *cls)
+ OVS_NO_THREAD_SAFETY_ANALYSIS
{
- return cls->cls->n_rules;
+ /* n_rules is an int, so in the presence of concurrent writers this will
+ * return either the old or a new value. */
+ return cls->n_rules;
}
static uint32_t
}
static struct cls_partition *
-find_partition(const struct cls_classifier *cls, ovs_be64 metadata,
- uint32_t hash)
+find_partition(const struct classifier *cls, ovs_be64 metadata, uint32_t hash)
{
struct cls_partition *partition;
- HMAP_FOR_EACH_IN_BUCKET (partition, hmap_node, hash, &cls->partitions) {
+ CMAP_FOR_EACH_WITH_HASH (partition, cmap_node, hash, &cls->partitions) {
if (partition->metadata == metadata) {
return partition;
}
}
static struct cls_partition *
-create_partition(struct cls_classifier *cls, struct cls_subtable *subtable,
+create_partition(struct classifier *cls, struct cls_subtable *subtable,
ovs_be64 metadata)
+ OVS_REQUIRES(cls->mutex)
{
uint32_t hash = hash_metadata(metadata);
struct cls_partition *partition = find_partition(cls, metadata, hash);
partition->metadata = metadata;
partition->tags = 0;
tag_tracker_init(&partition->tracker);
- hmap_insert(&cls->partitions, &partition->hmap_node, hash);
+ cmap_insert(&cls->partitions, &partition->cmap_node, hash);
}
tag_tracker_add(&partition->tracker, &partition->tags, subtable->tag);
return partition;
* inserting the new rule. In this case, no rules are displaced by the new
* rule, even rules that cannot have any effect because the new rule matches a
* superset of their flows and has higher priority. */
-struct cls_rule *
-classifier_replace(struct classifier *cls_, struct cls_rule *rule)
+const struct cls_rule *
+classifier_replace(struct classifier *cls, struct cls_rule *rule)
+ OVS_EXCLUDED(cls->mutex)
{
- struct cls_classifier *cls = cls_->cls;
struct cls_match *old_rule;
struct cls_subtable *subtable;
+ const struct cls_rule *old_cls_rule = NULL;
+ ovs_mutex_lock(&cls->mutex);
subtable = find_subtable(cls, &rule->match.mask);
if (!subtable) {
subtable = insert_subtable(cls, &rule->match.mask);
old_rule = insert_rule(cls, subtable, rule);
if (!old_rule) {
- int i;
+ old_cls_rule = NULL;
rule->cls_match->partition = NULL;
if (minimask_get_metadata_mask(&rule->match.mask) == OVS_BE64_MAX) {
metadata);
}
- subtable->n_rules++;
cls->n_rules++;
- for (i = 0; i < cls->n_tries; i++) {
+ for (int i = 0; i < cls->n_tries; i++) {
if (subtable->trie_plen[i]) {
trie_insert(&cls->tries[i], rule, subtable->trie_plen[i]);
}
trie_insert_prefix(&subtable->ports_trie, &masked_ports,
subtable->ports_mask_len);
}
-
- return NULL;
} else {
- struct cls_rule *old_cls_rule = old_rule->cls_rule;
-
+ old_cls_rule = old_rule->cls_rule;
rule->cls_match->partition = old_rule->partition;
- old_cls_rule->cls_match = NULL;
- free(old_rule);
- return old_cls_rule;
+ CONST_CAST(struct cls_rule *, old_cls_rule)->cls_match = NULL;
+
+ /* 'old_rule' contains a cmap_node, which may not be freed
+ * immediately. */
+ ovsrcu_postpone(free, old_rule);
}
+ ovs_mutex_unlock(&cls->mutex);
+ return old_cls_rule;
}
/* Inserts 'rule' into 'cls'. Until 'rule' is removed from 'cls', the caller
void
classifier_insert(struct classifier *cls, struct cls_rule *rule)
{
- struct cls_rule *displaced_rule = classifier_replace(cls, rule);
+ const struct cls_rule *displaced_rule = classifier_replace(cls, rule);
ovs_assert(!displaced_rule);
}
/* Removes 'rule' from 'cls'. It is the caller's responsibility to destroy
* 'rule' with cls_rule_destroy(), freeing the memory block in which 'rule'
- * resides, etc., as necessary. */
-void
-classifier_remove(struct classifier *cls_, struct cls_rule *rule)
+ * resides, etc., as necessary.
+ *
+ * Does nothing if 'rule' has been already removed, or was never inserted.
+ *
+ * Returns the removed rule, or NULL, if it was already removed.
+ */
+const struct cls_rule *
+classifier_remove(struct classifier *cls, const struct cls_rule *rule)
+ OVS_EXCLUDED(cls->mutex)
{
- struct cls_classifier *cls = cls_->cls;
struct cls_partition *partition;
- struct cls_match *cls_match = rule->cls_match;
+ struct cls_match *cls_match;
struct cls_match *head;
struct cls_subtable *subtable;
int i;
+ uint32_t basis = 0, hash, ihash[CLS_MAX_INDICES];
+ uint8_t prev_be32ofs = 0;
- ovs_assert(cls_match);
+ ovs_mutex_lock(&cls->mutex);
+ cls_match = rule->cls_match;
+ if (!cls_match) {
+ rule = NULL;
+ goto unlock; /* Already removed. */
+ }
subtable = find_subtable(cls, &rule->match.mask);
ovs_assert(subtable);
/* Remove rule node from indices. */
for (i = 0; i < subtable->n_indices; i++) {
- hindex_remove(&subtable->indices[i], &cls_match->index_nodes[i]);
+ ihash[i] = minimatch_hash_range(&rule->match, prev_be32ofs,
+ subtable->index_ofs[i], &basis);
+ cmap_remove(&subtable->indices[i], &cls_match->index_nodes[i],
+ ihash[i]);
+ prev_be32ofs = subtable->index_ofs[i];
}
+ hash = minimatch_hash_range(&rule->match, prev_be32ofs, FLOW_U32S, &basis);
- head = find_equal(subtable, &rule->match.flow, cls_match->hmap_node.hash);
+ head = find_equal(subtable, &rule->match.flow, hash);
if (head != cls_match) {
- list_remove(&cls_match->list);
- } else if (list_is_empty(&cls_match->list)) {
- hmap_remove(&subtable->rules, &cls_match->hmap_node);
+ rculist_remove(&cls_match->list);
+ } else if (rculist_is_empty(&cls_match->list)) {
+ cmap_remove(&subtable->rules, &cls_match->cmap_node, hash);
} else {
- struct cls_match *next = CONTAINER_OF(cls_match->list.next,
- struct cls_match, list);
+ struct cls_match *next = next_rule_in_list_protected(cls_match);
- list_remove(&cls_match->list);
- hmap_replace(&subtable->rules, &cls_match->hmap_node,
- &next->hmap_node);
+ rculist_remove(&cls_match->list);
+ cmap_replace(&subtable->rules, &cls_match->cmap_node,
+ &next->cmap_node, hash);
}
partition = cls_match->partition;
tag_tracker_subtract(&partition->tracker, &partition->tags,
subtable->tag);
if (!partition->tags) {
- hmap_remove(&cls->partitions, &partition->hmap_node);
- free(partition);
+ cmap_remove(&cls->partitions, &partition->cmap_node,
+ hash_metadata(partition->metadata));
+ ovsrcu_postpone(free, partition);
}
}
if (--subtable->n_rules == 0) {
destroy_subtable(cls, subtable);
- } else {
- update_subtables_after_removal(cls, subtable, cls_match->priority);
+ } else if (subtable->max_priority == cls_match->priority
+ && --subtable->max_count == 0) {
+ /* Find the new 'max_priority' and 'max_count'. */
+ struct cls_match *head;
+ int max_priority = INT_MIN;
+
+ CMAP_FOR_EACH (head, cmap_node, &subtable->rules) {
+ if (head->priority > max_priority) {
+ max_priority = head->priority;
+ subtable->max_count = 1;
+ } else if (head->priority == max_priority) {
+ ++subtable->max_count;
+ }
+ }
+ subtable->max_priority = max_priority;
+ pvector_change_priority(&cls->subtables, subtable, max_priority);
}
cls->n_rules--;
- rule->cls_match = NULL;
- free(cls_match);
+ ovsrcu_postpone(free, cls_match);
+ CONST_CAST(struct cls_rule *, rule)->cls_match = NULL;
+unlock:
+ ovs_mutex_unlock(&cls->mutex);
+
+ return rule;
}
/* Prefix tree context. Valid when 'lookup_done' is true. Can skip all
- * subtables which have more than 'match_plen' bits in their corresponding
- * field at offset 'be32ofs'. If skipped, 'maskbits' prefix bits should be
- * unwildcarded to quarantee datapath flow matches only packets it should. */
+ * subtables which have a prefix match on the trie field, but whose prefix
+ * length is not indicated in 'match_plens'. For example, a subtable that
+ * has a 8-bit trie field prefix match can be skipped if
+ * !be_get_bit_at(&match_plens, 8 - 1). If skipped, 'maskbits' prefix bits
+ * must be unwildcarded to make datapath flow only match packets it should. */
struct trie_ctx {
const struct cls_trie *trie;
bool lookup_done; /* Status of the lookup. */
uint8_t be32ofs; /* U32 offset of the field in question. */
- unsigned int match_plen; /* Longest prefix than could possibly match. */
unsigned int maskbits; /* Prefix length needed to avoid false matches. */
+ union mf_value match_plens; /* Bitmask of prefix lengths with possible
+ * matches. */
};
static void
ctx->lookup_done = false;
}
-static inline void
-lookahead_subtable(const struct cls_subtable_entry *subtables)
-{
- ovs_prefetch_range(subtables->subtable, sizeof *subtables->subtable);
-}
-
/* Finds and returns the highest-priority rule in 'cls' that matches 'flow'.
* Returns a null pointer if no rules in 'cls' match 'flow'. If multiple rules
* of equal priority match 'flow', returns one arbitrarily.
* set of bits that were significant in the lookup. At some point
* earlier, 'wc' should have been initialized (e.g., by
* flow_wildcards_init_catchall()). */
-struct cls_rule *
-classifier_lookup(const struct classifier *cls_, const struct flow *flow,
+const struct cls_rule *
+classifier_lookup(const struct classifier *cls, const struct flow *flow,
struct flow_wildcards *wc)
{
- struct cls_classifier *cls = cls_->cls;
const struct cls_partition *partition;
tag_type tags;
- struct cls_match *best;
+ int best_priority = INT_MIN;
+ const struct cls_match *best;
struct trie_ctx trie_ctx[CLS_MAX_TRIES];
- int i;
- struct cls_subtable_entry *subtables = cls->subtables.array;
- int n_subtables = cls->subtables.count;
- int64_t best_priority = -1;
+ struct cls_subtable *subtable;
- /* Prefetch the subtables array. */
- ovs_prefetch_range(subtables, n_subtables * sizeof *subtables);
+ /* Synchronize for cls->n_tries and subtable->trie_plen. They can change
+ * when table configuration changes, which happens typically only on
+ * startup. */
+ atomic_thread_fence(memory_order_acquire);
/* Determine 'tags' such that, if 'subtable->tag' doesn't intersect them,
* then 'flow' cannot possibly match in 'subtable':
* that 'tags' always intersects such a cls_subtable's 'tags', so we don't
* need a special case.
*/
- partition = (hmap_is_empty(&cls->partitions)
+ partition = (cmap_is_empty(&cls->partitions)
? NULL
: find_partition(cls, flow->metadata,
hash_metadata(flow->metadata)));
tags = partition ? partition->tags : TAG_ARBITRARY;
- /* Initialize trie contexts for match_find_wc(). */
- for (i = 0; i < cls->n_tries; i++) {
+ /* Initialize trie contexts for find_match_wc(). */
+ for (int i = 0; i < cls->n_tries; i++) {
trie_ctx_init(&trie_ctx[i], &cls->tries[i]);
}
- /* Prefetch the first subtables. */
- if (n_subtables > 1) {
- lookahead_subtable(subtables);
- lookahead_subtable(subtables + 1);
- }
-
best = NULL;
- for (i = 0; OVS_LIKELY(i < n_subtables); i++) {
- struct cls_match *rule;
+ PVECTOR_FOR_EACH_PRIORITY(subtable, best_priority, 2,
+ sizeof(struct cls_subtable), &cls->subtables) {
+ const struct cls_match *rule;
- if ((int64_t)subtables[i].max_priority <= best_priority) {
- /* Subtables are in descending priority order,
- * can not find anything better. */
- break;
- }
-
- /* Prefetch a forthcoming subtable. */
- if (i + 2 < n_subtables) {
- lookahead_subtable(&subtables[i + 2]);
- }
-
- if (!tag_intersects(tags, subtables[i].tag)) {
+ if (!tag_intersects(tags, subtable->tag)) {
continue;
}
- rule = find_match_wc(subtables[i].subtable, flow, trie_ctx,
- cls->n_tries, wc);
- if (rule && (int64_t)rule->priority > best_priority) {
- best_priority = (int64_t)rule->priority;
+ rule = find_match_wc(subtable, flow, trie_ctx, cls->n_tries, wc);
+ if (rule && rule->priority > best_priority) {
+ best_priority = rule->priority;
best = rule;
}
}
return best ? best->cls_rule : NULL;
}
-/* Returns true if 'target' satisifies 'match', that is, if each bit for which
- * 'match' specifies a particular value has the correct value in 'target'.
- *
- * 'flow' and 'mask' have the same mask! */
-static bool
-miniflow_and_mask_matches_miniflow(const struct miniflow *flow,
- const struct minimask *mask,
- const struct miniflow *target)
-{
- const uint32_t *flowp = miniflow_get_u32_values(flow);
- const uint32_t *maskp = miniflow_get_u32_values(&mask->masks);
- uint32_t target_u32;
-
- MINIFLOW_FOR_EACH_IN_MAP(target_u32, target, mask->masks.map) {
- if ((*flowp++ ^ target_u32) & *maskp++) {
- return false;
- }
- }
-
- return true;
-}
-
-static inline struct cls_match *
-find_match_miniflow(const struct cls_subtable *subtable,
- const struct miniflow *flow,
- uint32_t hash)
-{
- struct cls_match *rule;
-
- HMAP_FOR_EACH_WITH_HASH (rule, hmap_node, hash, &subtable->rules) {
- if (miniflow_and_mask_matches_miniflow(&rule->flow, &subtable->mask,
- flow)) {
- return rule;
- }
- }
-
- return NULL;
-}
-
-/* Finds and returns the highest-priority rule in 'cls' that matches
- * 'miniflow'. Returns a null pointer if no rules in 'cls' match 'flow'.
- * If multiple rules of equal priority match 'flow', returns one arbitrarily.
- *
- * This function is optimized for the userspace datapath, which only ever has
- * one priority value for it's flows!
- */
-struct cls_rule *classifier_lookup_miniflow_first(const struct classifier *cls_,
- const struct miniflow *flow)
-{
- struct cls_classifier *cls = cls_->cls;
- struct cls_subtable *subtable;
- struct cls_subtable_entry *iter;
-
- CLS_SUBTABLES_FOR_EACH (subtable, iter, &cls->subtables) {
- struct cls_match *rule;
-
- rule = find_match_miniflow(subtable, flow,
- miniflow_hash_in_minimask(flow,
- &subtable->mask,
- 0));
- if (rule) {
- return rule->cls_rule;
- }
- }
-
- return NULL;
-}
-
/* Finds and returns a rule in 'cls' with exactly the same priority and
* matching criteria as 'target'. Returns a null pointer if 'cls' doesn't
* contain an exact match. */
-struct cls_rule *
-classifier_find_rule_exactly(const struct classifier *cls_,
+const struct cls_rule *
+classifier_find_rule_exactly(const struct classifier *cls,
const struct cls_rule *target)
{
- struct cls_classifier *cls = cls_->cls;
- struct cls_match *head, *rule;
- struct cls_subtable *subtable;
+ const struct cls_match *head, *rule;
+ const struct cls_subtable *subtable;
subtable = find_subtable(cls, &target->match.mask);
if (!subtable) {
return NULL;
}
- /* Skip if there is no hope. */
- if (target->priority > subtable->max_priority) {
- return NULL;
- }
-
head = find_equal(subtable, &target->match.flow,
miniflow_hash_in_minimask(&target->match.flow,
&target->match.mask, 0));
+ if (!head) {
+ return NULL;
+ }
FOR_EACH_RULE_IN_LIST (rule, head) {
if (target->priority >= rule->priority) {
return target->priority == rule->priority ? rule->cls_rule : NULL;
/* Finds and returns a rule in 'cls' with priority 'priority' and exactly the
* same matching criteria as 'target'. Returns a null pointer if 'cls' doesn't
* contain an exact match. */
-struct cls_rule *
+const struct cls_rule *
classifier_find_match_exactly(const struct classifier *cls,
- const struct match *target,
- unsigned int priority)
+ const struct match *target, int priority)
{
- struct cls_rule *retval;
+ const struct cls_rule *retval;
struct cls_rule cr;
cls_rule_init(&cr, target, priority);
* considered to overlap if both rules have the same priority and a packet
* could match both. */
bool
-classifier_rule_overlaps(const struct classifier *cls_,
+classifier_rule_overlaps(const struct classifier *cls,
const struct cls_rule *target)
+ OVS_EXCLUDED(cls->mutex)
{
- struct cls_classifier *cls = cls_->cls;
struct cls_subtable *subtable;
- struct cls_subtable_entry *iter;
+ ovs_mutex_lock(&cls->mutex);
/* Iterate subtables in the descending max priority order. */
- CLS_SUBTABLES_FOR_EACH (subtable, iter, &cls->subtables) {
+ PVECTOR_FOR_EACH_PRIORITY (subtable, target->priority - 1, 2,
+ sizeof(struct cls_subtable), &cls->subtables) {
uint32_t storage[FLOW_U32S];
struct minimask mask;
struct cls_match *head;
- if (target->priority > iter->max_priority) {
- break; /* Can skip this and the rest of the subtables. */
- }
-
minimask_combine(&mask, &target->match.mask, &subtable->mask, storage);
- HMAP_FOR_EACH (head, hmap_node, &subtable->rules) {
+ CMAP_FOR_EACH (head, cmap_node, &subtable->rules) {
struct cls_match *rule;
- FOR_EACH_RULE_IN_LIST (rule, head) {
+ FOR_EACH_RULE_IN_LIST_PROTECTED (rule, head) {
if (rule->priority < target->priority) {
break; /* Rules in descending priority order. */
}
if (rule->priority == target->priority
&& miniflow_equal_in_minimask(&target->match.flow,
&rule->flow, &mask)) {
+ ovs_mutex_unlock(&cls->mutex);
return true;
}
}
}
}
+ ovs_mutex_unlock(&cls->mutex);
return false;
}
&target->match.mask));
}
-static struct cls_match *
+static const struct cls_match *
search_subtable(const struct cls_subtable *subtable,
- const struct cls_rule *target)
+ struct cls_cursor *cursor)
{
- if (!target || !minimask_has_extra(&subtable->mask, &target->match.mask)) {
- struct cls_match *rule;
+ if (!cursor->target
+ || !minimask_has_extra(&subtable->mask, &cursor->target->match.mask)) {
+ const struct cls_match *rule;
- HMAP_FOR_EACH (rule, hmap_node, &subtable->rules) {
- if (rule_matches(rule, target)) {
+ CMAP_CURSOR_FOR_EACH (rule, cmap_node, &cursor->rules,
+ &subtable->rules) {
+ if (rule_matches(rule, cursor->target)) {
return rule;
}
}
return NULL;
}
-/* Initializes 'cursor' for iterating through rules in 'cls':
+/* Initializes 'cursor' for iterating through rules in 'cls', and returns the
+ * first matching cls_rule via '*pnode', or NULL if there are no matches.
*
* - If 'target' is null, the cursor will visit every rule in 'cls'.
*
* such that cls_rule_is_loose_match(rule, target) returns true.
*
* Ignores target->priority. */
-void
-cls_cursor_init(struct cls_cursor *cursor, const struct classifier *cls,
- const struct cls_rule *target)
-{
- cursor->cls = cls->cls;
- cursor->target = target && !cls_rule_is_catchall(target) ? target : NULL;
-}
-
-/* Returns the first matching cls_rule in 'cursor''s iteration, or a null
- * pointer if there are no matches. */
-struct cls_rule *
-cls_cursor_first(struct cls_cursor *cursor)
+struct cls_cursor cls_cursor_start(const struct classifier *cls,
+ const struct cls_rule *target,
+ bool safe)
+ OVS_NO_THREAD_SAFETY_ANALYSIS
{
+ struct cls_cursor cursor;
struct cls_subtable *subtable;
- HMAP_FOR_EACH (subtable, hmap_node, &cursor->cls->subtables_map) {
- struct cls_match *rule = search_subtable(subtable, cursor->target);
+ cursor.safe = safe;
+ cursor.cls = cls;
+ cursor.target = target && !cls_rule_is_catchall(target) ? target : NULL;
+ cursor.rule = NULL;
+
+ /* Find first rule. */
+ ovs_mutex_lock(&cursor.cls->mutex);
+ CMAP_CURSOR_FOR_EACH (subtable, cmap_node, &cursor.subtables,
+ &cursor.cls->subtables_map) {
+ const struct cls_match *rule = search_subtable(subtable, &cursor);
+
if (rule) {
- cursor->subtable = subtable;
- return rule->cls_rule;
+ cursor.subtable = subtable;
+ cursor.rule = rule->cls_rule;
+ break;
}
}
- return NULL;
+ /* Leave locked if requested and have a rule. */
+ if (safe || !cursor.rule) {
+ ovs_mutex_unlock(&cursor.cls->mutex);
+ }
+ return cursor;
}
-/* Returns the next matching cls_rule in 'cursor''s iteration, or a null
- * pointer if there are no more matches. */
-struct cls_rule *
-cls_cursor_next(struct cls_cursor *cursor, const struct cls_rule *rule_)
+static const struct cls_rule *
+cls_cursor_next(struct cls_cursor *cursor)
+ OVS_NO_THREAD_SAFETY_ANALYSIS
{
- struct cls_match *rule = CONST_CAST(struct cls_match *, rule_->cls_match);
+ const struct cls_match *rule = cursor->rule->cls_match;
const struct cls_subtable *subtable;
- struct cls_match *next;
+ const struct cls_match *next;
next = next_rule_in_list__(rule);
if (next->priority < rule->priority) {
}
/* 'next' is the head of the list, that is, the rule that is included in
- * the subtable's hmap. (This is important when the classifier contains
+ * the subtable's map. (This is important when the classifier contains
* rules that differ only in priority.) */
rule = next;
- HMAP_FOR_EACH_CONTINUE (rule, hmap_node, &cursor->subtable->rules) {
+ CMAP_CURSOR_FOR_EACH_CONTINUE (rule, cmap_node, &cursor->rules) {
if (rule_matches(rule, cursor->target)) {
return rule->cls_rule;
}
}
subtable = cursor->subtable;
- HMAP_FOR_EACH_CONTINUE (subtable, hmap_node, &cursor->cls->subtables_map) {
- rule = search_subtable(subtable, cursor->target);
+ CMAP_CURSOR_FOR_EACH_CONTINUE (subtable, cmap_node, &cursor->subtables) {
+ rule = search_subtable(subtable, cursor);
if (rule) {
cursor->subtable = subtable;
return rule->cls_rule;
return NULL;
}
+
+/* Sets 'cursor->rule' to the next matching cls_rule in 'cursor''s iteration,
+ * or to null if all matching rules have been visited. */
+void
+cls_cursor_advance(struct cls_cursor *cursor)
+ OVS_NO_THREAD_SAFETY_ANALYSIS
+{
+ if (cursor->safe) {
+ ovs_mutex_lock(&cursor->cls->mutex);
+ }
+ cursor->rule = cls_cursor_next(cursor);
+ if (cursor->safe || !cursor->rule) {
+ ovs_mutex_unlock(&cursor->cls->mutex);
+ }
+}
\f
static struct cls_subtable *
-find_subtable(const struct cls_classifier *cls, const struct minimask *mask)
+find_subtable(const struct classifier *cls, const struct minimask *mask)
{
struct cls_subtable *subtable;
- HMAP_FOR_EACH_IN_BUCKET (subtable, hmap_node, minimask_hash(mask, 0),
+ CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, minimask_hash(mask, 0),
&cls->subtables_map) {
if (minimask_equal(mask, &subtable->mask)) {
return subtable;
return NULL;
}
+/* The new subtable will be visible to the readers only after this. */
static struct cls_subtable *
-insert_subtable(struct cls_classifier *cls, const struct minimask *mask)
+insert_subtable(struct classifier *cls, const struct minimask *mask)
+ OVS_REQUIRES(cls->mutex)
{
uint32_t hash = minimask_hash(mask, 0);
struct cls_subtable *subtable;
int i, index = 0;
struct flow_wildcards old, new;
uint8_t prev;
- struct cls_subtable_entry elem;
int count = count_1bits(mask->masks.map);
subtable = xzalloc(sizeof *subtable - sizeof mask->masks.inline_values
+ MINIFLOW_VALUES_SIZE(count));
- hmap_init(&subtable->rules);
- miniflow_clone_inline(&subtable->mask.masks, &mask->masks, count);
+ cmap_init(&subtable->rules);
+ miniflow_clone_inline(CONST_CAST(struct miniflow *, &subtable->mask.masks),
+ &mask->masks, count);
/* Init indices for segmented lookup, if any. */
flow_wildcards_init_catchall(&new);
cls->flow_segments[i]);
/* Add an index if it adds mask bits. */
if (!flow_wildcards_equal(&new, &old)) {
- hindex_init(&subtable->indices[index]);
- subtable->index_ofs[index] = cls->flow_segments[i];
+ cmap_init(&subtable->indices[index]);
+ *CONST_CAST(uint8_t *, &subtable->index_ofs[index])
+ = cls->flow_segments[i];
index++;
old = new;
}
flow_wildcards_fold_minimask_range(&new, mask, prev, FLOW_U32S);
if (flow_wildcards_equal(&new, &old)) {
--index;
- subtable->index_ofs[index] = 0;
- hindex_destroy(&subtable->indices[index]);
+ *CONST_CAST(uint8_t *, &subtable->index_ofs[index]) = 0;
+ cmap_destroy(&subtable->indices[index]);
}
}
- subtable->n_indices = index;
+ *CONST_CAST(uint8_t *, &subtable->n_indices) = index;
- subtable->tag = (minimask_get_metadata_mask(mask) == OVS_BE64_MAX
- ? tag_create_deterministic(hash)
- : TAG_ALL);
+ *CONST_CAST(tag_type *, &subtable->tag) =
+ (minimask_get_metadata_mask(mask) == OVS_BE64_MAX
+ ? tag_create_deterministic(hash)
+ : TAG_ALL);
for (i = 0; i < cls->n_tries; i++) {
subtable->trie_plen[i] = minimask_get_prefix_len(mask,
}
/* Ports trie. */
- subtable->ports_trie = NULL;
- subtable->ports_mask_len
+ ovsrcu_set_hidden(&subtable->ports_trie, NULL);
+ *CONST_CAST(int *, &subtable->ports_mask_len)
= 32 - ctz32(ntohl(MINIFLOW_GET_BE32(&mask->masks, tp_src)));
- hmap_insert(&cls->subtables_map, &subtable->hmap_node, hash);
- elem.subtable = subtable;
- elem.tag = subtable->tag;
- elem.max_priority = subtable->max_priority;
- cls_subtables_push_back(&cls->subtables, elem);
+ cmap_insert(&cls->subtables_map, &subtable->cmap_node, hash);
return subtable;
}
+/* RCU readers may still access the subtable before it is actually freed. */
static void
-destroy_subtable(struct cls_classifier *cls, struct cls_subtable *subtable)
+destroy_subtable(struct classifier *cls, struct cls_subtable *subtable)
+ OVS_REQUIRES(cls->mutex)
{
int i;
- struct cls_subtable *table = NULL;
- struct cls_subtable_entry *iter;
- CLS_SUBTABLES_FOR_EACH (table, iter, &cls->subtables) {
- if (table == subtable) {
- cls_subtables_remove(&cls->subtables, iter);
- break;
- }
- }
+ pvector_remove(&cls->subtables, subtable);
+ cmap_remove(&cls->subtables_map, &subtable->cmap_node,
+ minimask_hash(&subtable->mask, 0));
- trie_destroy(subtable->ports_trie);
+ ovs_assert(ovsrcu_get_protected(struct trie_node *, &subtable->ports_trie)
+ == NULL);
+ ovs_assert(cmap_is_empty(&subtable->rules));
for (i = 0; i < subtable->n_indices; i++) {
- hindex_destroy(&subtable->indices[i]);
- }
- minimask_destroy(&subtable->mask);
- hmap_remove(&cls->subtables_map, &subtable->hmap_node);
- hmap_destroy(&subtable->rules);
- free(subtable);
-}
-
-/* This function performs the following updates for 'subtable' in 'cls'
- * following the addition of a new rule with priority 'new_priority' to
- * 'subtable':
- *
- * - Update 'subtable->max_priority' and 'subtable->max_count' if necessary.
- *
- * - Update 'subtable''s position in 'cls->subtables' if necessary.
- *
- * This function should only be called after adding a new rule, not after
- * replacing a rule by an identical one or modifying a rule in-place. */
-static void
-update_subtables_after_insertion(struct cls_classifier *cls,
- struct cls_subtable *subtable,
- unsigned int new_priority)
-{
- if (new_priority == subtable->max_priority) {
- ++subtable->max_count;
- } else if (new_priority > subtable->max_priority) {
- struct cls_subtable *table;
- struct cls_subtable_entry *iter, *from = NULL;
-
- subtable->max_priority = new_priority;
- subtable->max_count = 1;
-
- /* Possibly move 'subtable' earlier in the priority array. If
- * we break out of the loop, then the subtable (at 'from')
- * should be moved to the position right after the current
- * element. If the loop terminates normally, then 'iter' will
- * be at the first array element and we'll move the subtable
- * to the front of the array. */
- CLS_SUBTABLES_FOR_EACH_REVERSE (table, iter, &cls->subtables) {
- if (table == subtable) {
- from = iter; /* Locate the subtable as we go. */
- iter->max_priority = new_priority;
- } else if (table->max_priority >= new_priority) {
- if (from == NULL) {
- /* Corrupted cache? */
- cls_subtables_reset(cls);
- VLOG_ABORT("update_subtables_after_insertion(): Subtable priority list corrupted.");
- OVS_NOT_REACHED();
- }
- iter++; /* After this. */
- break;
- }
- }
-
- /* Move subtable at 'from' to 'iter'. */
- cls_subtables_move(iter, from);
- }
-}
-
-/* This function performs the following updates for 'subtable' in 'cls'
- * following the deletion of a rule with priority 'del_priority' from
- * 'subtable':
- *
- * - Update 'subtable->max_priority' and 'subtable->max_count' if necessary.
- *
- * - Update 'subtable''s position in 'cls->subtables' if necessary.
- *
- * This function should only be called after removing a rule, not after
- * replacing a rule by an identical one or modifying a rule in-place. */
-static void
-update_subtables_after_removal(struct cls_classifier *cls,
- struct cls_subtable *subtable,
- unsigned int del_priority)
-{
- if (del_priority == subtable->max_priority && --subtable->max_count == 0) {
- struct cls_match *head;
- struct cls_subtable *table;
- struct cls_subtable_entry *iter, *from = NULL;
-
- subtable->max_priority = 0;
- HMAP_FOR_EACH (head, hmap_node, &subtable->rules) {
- if (head->priority > subtable->max_priority) {
- subtable->max_priority = head->priority;
- subtable->max_count = 1;
- } else if (head->priority == subtable->max_priority) {
- ++subtable->max_count;
- }
- }
-
- /* Possibly move 'subtable' later in the priority array.
- * After the loop the 'iter' will point right after the position
- * at which the subtable should be moved (either at a subtable
- * with an equal or lower priority, or just past the array),
- * so it is decremented once. */
- CLS_SUBTABLES_FOR_EACH (table, iter, &cls->subtables) {
- if (table == subtable) {
- from = iter; /* Locate the subtable as we go. */
- iter->max_priority = subtable->max_priority;
- } else if (table->max_priority <= subtable->max_priority) {
- if (from == NULL) {
- /* Corrupted cache? */
- cls_subtables_reset(cls);
- VLOG_ABORT("update_subtables_after_removal(): Subtable priority list corrupted.");
- OVS_NOT_REACHED();
- }
- break;
- }
- }
- /* Now at one past the destination. */
- iter--;
-
- /* Move subtable at 'from' to 'iter'. */
- cls_subtables_move(iter, from);
+ cmap_destroy(&subtable->indices[i]);
}
+ cmap_destroy(&subtable->rules);
+ ovsrcu_postpone(free, subtable);
}
struct range {
uint8_t end;
};
+static unsigned int be_get_bit_at(const ovs_be32 value[], unsigned int ofs);
+
/* Return 'true' if can skip rest of the subtable based on the prefix trie
* lookup results. */
static inline bool
if (be32ofs >= ofs.start && be32ofs < ofs.end) {
/* On-demand trie lookup. */
if (!ctx->lookup_done) {
- ctx->match_plen = trie_lookup(ctx->trie, flow,
- &ctx->maskbits);
+ memset(&ctx->match_plens, 0, sizeof ctx->match_plens);
+ ctx->maskbits = trie_lookup(ctx->trie, flow,
+ &ctx->match_plens);
ctx->lookup_done = true;
}
/* Possible to skip the rest of the subtable if subtable's
- * prefix on the field is longer than what is known to match
- * based on the trie lookup. */
- if (field_plen[j] > ctx->match_plen) {
- /* RFC: We want the trie lookup to never result in
- * unwildcarding any bits that would not be unwildcarded
- * otherwise. Since the trie is shared by the whole
- * classifier, it is possible that the 'maskbits' contain
- * bits that are irrelevant for the partition of the
- * classifier relevant for the current flow. */
+ * prefix on the field is not included in the lookup result. */
+ if (!be_get_bit_at(&ctx->match_plens.be32, field_plen[j] - 1)) {
+ /* We want the trie lookup to never result in unwildcarding
+ * any bits that would not be unwildcarded otherwise.
+ * Since the trie is shared by the whole classifier, it is
+ * possible that the 'maskbits' contain bits that are
+ * irrelevant for the partition relevant for the current
+ * packet. Hence the checks below. */
- /* Can skip if the field is already unwildcarded. */
- if (mask_prefix_bits_set(wc, be32ofs, ctx->maskbits)) {
- return true;
- }
/* Check that the trie result will not unwildcard more bits
- * than this stage will. */
+ * than this subtable would otherwise. */
if (ctx->maskbits <= field_plen[j]) {
/* Unwildcard the bits and skip the rest. */
mask_set_prefix_bits(wc, be32ofs, ctx->maskbits);
/* Note: Prerequisite already unwildcarded, as the only
* prerequisite of the supported trie lookup fields is
- * the ethertype, which is currently always
- * unwildcarded.
- */
+ * the ethertype, which is always unwildcarded. */
+ return true;
+ }
+ /* Can skip if the field is already unwildcarded. */
+ if (mask_prefix_bits_set(wc, be32ofs, ctx->maskbits)) {
return true;
}
}
* value has the correct value in 'target'.
*
* This function is equivalent to miniflow_equal_flow_in_minimask(flow,
- * target, mask) but it is faster because of the invariant that
- * flow->map and mask->masks.map are the same. */
+ * target, mask) but this is faster because of the invariant that
+ * flow->map and mask->masks.map are the same, and that this version
+ * takes the 'wc'. */
static inline bool
miniflow_and_mask_matches_flow(const struct miniflow *flow,
const struct minimask *mask,
{
const uint32_t *flowp = miniflow_get_u32_values(flow);
const uint32_t *maskp = miniflow_get_u32_values(&mask->masks);
- uint32_t target_u32;
+ uint32_t idx;
+
+ MAP_FOR_EACH_INDEX(idx, mask->masks.map) {
+ uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & *maskp++;
- FLOW_FOR_EACH_IN_MAP(target_u32, target, mask->masks.map) {
- if ((*flowp++ ^ target_u32) & *maskp++) {
+ if (diff) {
return false;
}
}
return true;
}
-static inline struct cls_match *
+static inline const struct cls_match *
find_match(const struct cls_subtable *subtable, const struct flow *flow,
uint32_t hash)
{
- struct cls_match *rule;
+ const struct cls_match *rule;
- HMAP_FOR_EACH_WITH_HASH (rule, hmap_node, hash, &subtable->rules) {
+ CMAP_FOR_EACH_WITH_HASH (rule, cmap_node, hash, &subtable->rules) {
if (miniflow_and_mask_matches_flow(&rule->flow, &subtable->mask,
flow)) {
return rule;
return NULL;
}
-static struct cls_match *
+/* Returns true if 'target' satisifies 'flow'/'mask', that is, if each bit
+ * for which 'flow', for which 'mask' has a bit set, specifies a particular
+ * value has the correct value in 'target'.
+ *
+ * This function is equivalent to miniflow_and_mask_matches_flow() but this
+ * version fills in the mask bits in 'wc'. */
+static inline bool
+miniflow_and_mask_matches_flow_wc(const struct miniflow *flow,
+ const struct minimask *mask,
+ const struct flow *target,
+ struct flow_wildcards *wc)
+{
+ const uint32_t *flowp = miniflow_get_u32_values(flow);
+ const uint32_t *maskp = miniflow_get_u32_values(&mask->masks);
+ uint32_t idx;
+
+ MAP_FOR_EACH_INDEX(idx, mask->masks.map) {
+ uint32_t mask = *maskp++;
+ uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & mask;
+
+ if (diff) {
+ /* Only unwildcard if none of the differing bits is already
+ * exact-matched. */
+ if (!(flow_u32_value(&wc->masks, idx) & diff)) {
+ /* Keep one bit of the difference. */
+ *flow_u32_lvalue(&wc->masks, idx) |= rightmost_1bit(diff);
+ }
+ return false;
+ }
+ /* Fill in the bits that were looked at. */
+ *flow_u32_lvalue(&wc->masks, idx) |= mask;
+ }
+
+ return true;
+}
+
+/* Unwildcard the fields looked up so far, if any. */
+static void
+fill_range_wc(const struct cls_subtable *subtable, struct flow_wildcards *wc,
+ uint8_t to)
+{
+ if (to) {
+ flow_wildcards_fold_minimask_range(wc, &subtable->mask, 0, to);
+ }
+}
+
+static const struct cls_match *
find_match_wc(const struct cls_subtable *subtable, const struct flow *flow,
struct trie_ctx trie_ctx[CLS_MAX_TRIES], unsigned int n_tries,
struct flow_wildcards *wc)
{
uint32_t basis = 0, hash;
- struct cls_match *rule = NULL;
+ const struct cls_match *rule = NULL;
int i;
struct range ofs;
ofs.start = 0;
/* Try to finish early by checking fields in segments. */
for (i = 0; i < subtable->n_indices; i++) {
- struct hindex_node *inode;
+ const struct cmap_node *inode;
+
ofs.end = subtable->index_ofs[i];
if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow,
wc)) {
- goto range_out;
+ /* 'wc' bits for the trie field set, now unwildcard the preceding
+ * bits used so far. */
+ fill_range_wc(subtable, wc, ofs.start);
+ return NULL;
}
hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start,
ofs.end, &basis);
- ofs.start = ofs.end;
- inode = hindex_node_with_hash(&subtable->indices[i], hash);
+ inode = cmap_find(&subtable->indices[i], hash);
if (!inode) {
- /* No match, can stop immediately, but must fold in the mask
- * covered so far. */
- goto range_out;
+ /* No match, can stop immediately, but must fold in the bits
+ * used in lookup so far. */
+ fill_range_wc(subtable, wc, ofs.end);
+ return NULL;
}
/* If we have narrowed down to a single rule already, check whether
- * that rule matches. If it does match, then we're done. If it does
- * not match, then we know that we will never get a match, but we do
- * not yet know how many wildcards we need to fold into 'wc' so we
- * continue iterating through indices to find that out. (We won't
- * waste time calling miniflow_and_mask_matches_flow() again because
- * we've set 'rule' nonnull.)
- *
- * This check shows a measurable benefit with non-trivial flow tables.
+ * that rule matches. Either way, we're done.
*
* (Rare) hash collisions may cause us to miss the opportunity for this
* optimization. */
- if (!inode->s && !rule) {
+ if (!cmap_node_next(inode)) {
ASSIGN_CONTAINER(rule, inode - i, index_nodes);
- if (miniflow_and_mask_matches_flow(&rule->flow, &subtable->mask,
- flow)) {
- goto out;
+ if (miniflow_and_mask_matches_flow_wc(&rule->flow, &subtable->mask,
+ flow, wc)) {
+ return rule;
}
+ return NULL;
}
+ ofs.start = ofs.end;
}
ofs.end = FLOW_U32S;
/* Trie check for the final range. */
if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow, wc)) {
- goto range_out;
- }
- if (!rule) {
- /* Multiple potential matches exist, look for one. */
- hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start,
- ofs.end, &basis);
- rule = find_match(subtable, flow, hash);
- } else {
- /* We already narrowed the matching candidates down to just 'rule',
- * but it didn't match. */
- rule = NULL;
+ fill_range_wc(subtable, wc, ofs.start);
+ return NULL;
}
+ hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start,
+ ofs.end, &basis);
+ rule = find_match(subtable, flow, hash);
if (!rule && subtable->ports_mask_len) {
/* Ports are always part of the final range, if any.
* No match was found for the ports. Use the ports trie to figure out
* which ports bits to unwildcard. */
unsigned int mbits;
- ovs_be32 value, mask;
+ ovs_be32 value, plens, mask;
mask = MINIFLOW_GET_BE32(&subtable->mask.masks, tp_src);
value = ((OVS_FORCE ovs_be32 *)flow)[TP_PORTS_OFS32] & mask;
- trie_lookup_value(subtable->ports_trie, &value, 32, &mbits);
+ mbits = trie_lookup_value(&subtable->ports_trie, &value, &plens, 32);
((OVS_FORCE ovs_be32 *)&wc->masks)[TP_PORTS_OFS32] |=
- mask & htonl(~0 << (32 - mbits));
+ mask & be32_prefix_mask(mbits);
- ofs.start = TP_PORTS_OFS32;
- goto range_out;
+ /* Unwildcard all bits in the mask upto the ports, as they were used
+ * to determine there is no match. */
+ fill_range_wc(subtable, wc, TP_PORTS_OFS32);
+ return NULL;
}
- out:
+
/* Must unwildcard all the fields, as they were looked at. */
flow_wildcards_fold_minimask(wc, &subtable->mask);
return rule;
-
- range_out:
- /* Must unwildcard the fields looked up so far, if any. */
- if (ofs.start) {
- flow_wildcards_fold_minimask_range(wc, &subtable->mask, 0, ofs.start);
- }
- return NULL;
}
static struct cls_match *
-find_equal(struct cls_subtable *subtable, const struct miniflow *flow,
+find_equal(const struct cls_subtable *subtable, const struct miniflow *flow,
uint32_t hash)
{
struct cls_match *head;
- HMAP_FOR_EACH_WITH_HASH (head, hmap_node, hash, &subtable->rules) {
+ CMAP_FOR_EACH_WITH_HASH (head, cmap_node, hash, &subtable->rules) {
if (miniflow_equal(&head->flow, flow)) {
return head;
}
return NULL;
}
+/*
+ * As the readers are operating concurrently with the modifications, a
+ * concurrent reader may or may not see the new rule, depending on how
+ * the concurrent events overlap with each other. This is no
+ * different from the former locked behavior, but there the visibility
+ * of the new rule only depended on the timing of the locking
+ * functions.
+ *
+ * The new rule is first added to the segment indices, so the readers
+ * may find the rule in the indices before the rule is visible in the
+ * subtables 'rules' map. This may result in us losing the
+ * opportunity to quit lookups earlier, resulting in sub-optimal
+ * wildcarding. This will be fixed by forthcoming revalidation always
+ * scheduled after flow table changes.
+ *
+ * Similar behavior may happen due to us removing the overlapping rule
+ * (if any) from the indices only after the new rule has been added.
+ *
+ * The subtable's max priority is updated only after the rule is
+ * inserted, so the concurrent readers may not see the rule, as the
+ * updated priority ordered subtable list will only be visible after
+ * the subtable's max priority is updated.
+ *
+ * Similarly, the classifier's partitions for new rules are updated by
+ * the caller after this function, so the readers may keep skipping
+ * the subtable until they see the updated partitions.
+ */
static struct cls_match *
-insert_rule(struct cls_classifier *cls, struct cls_subtable *subtable,
- struct cls_rule *new)
+insert_rule(struct classifier *cls, struct cls_subtable *subtable,
+ struct cls_rule *new_rule)
+ OVS_REQUIRES(cls->mutex)
{
- struct cls_match *cls_match = cls_match_alloc(new);
- struct cls_match *head;
struct cls_match *old = NULL;
+ struct cls_match *new = cls_match_alloc(new_rule);
+ struct cls_match *head;
int i;
- uint32_t basis = 0, hash;
+ uint32_t basis = 0, hash, ihash[CLS_MAX_INDICES];
uint8_t prev_be32ofs = 0;
/* Add new node to segment indices. */
for (i = 0; i < subtable->n_indices; i++) {
- hash = minimatch_hash_range(&new->match, prev_be32ofs,
- subtable->index_ofs[i], &basis);
- hindex_insert(&subtable->indices[i], &cls_match->index_nodes[i], hash);
+ ihash[i] = minimatch_hash_range(&new_rule->match, prev_be32ofs,
+ subtable->index_ofs[i], &basis);
+ cmap_insert(&subtable->indices[i], &new->index_nodes[i], ihash[i]);
prev_be32ofs = subtable->index_ofs[i];
}
- hash = minimatch_hash_range(&new->match, prev_be32ofs, FLOW_U32S, &basis);
- head = find_equal(subtable, &new->match.flow, hash);
+ hash = minimatch_hash_range(&new_rule->match, prev_be32ofs, FLOW_U32S,
+ &basis);
+ head = find_equal(subtable, &new_rule->match.flow, hash);
if (!head) {
- hmap_insert(&subtable->rules, &cls_match->hmap_node, hash);
- list_init(&cls_match->list);
+ cmap_insert(&subtable->rules, &new->cmap_node, hash);
+ rculist_init(&new->list);
goto out;
} else {
/* Scan the list for the insertion point that will keep the list in
* order of decreasing priority. */
struct cls_match *rule;
- cls_match->hmap_node.hash = hash; /* Otherwise done by hmap_insert. */
-
- FOR_EACH_RULE_IN_LIST (rule, head) {
- if (cls_match->priority >= rule->priority) {
+ FOR_EACH_RULE_IN_LIST_PROTECTED (rule, head) {
+ if (new->priority >= rule->priority) {
if (rule == head) {
/* 'new' is the new highest-priority flow in the list. */
- hmap_replace(&subtable->rules,
- &rule->hmap_node, &cls_match->hmap_node);
+ cmap_replace(&subtable->rules, &rule->cmap_node,
+ &new->cmap_node, hash);
}
- if (cls_match->priority == rule->priority) {
- list_replace(&cls_match->list, &rule->list);
+ if (new->priority == rule->priority) {
+ rculist_replace(&new->list, &rule->list);
old = rule;
- goto out;
} else {
- list_insert(&rule->list, &cls_match->list);
- goto out;
+ rculist_insert(&rule->list, &new->list);
}
+ goto out;
}
}
/* Insert 'new' at the end of the list. */
- list_push_back(&head->list, &cls_match->list);
+ rculist_push_back(&head->list, &new->list);
}
out:
if (!old) {
- update_subtables_after_insertion(cls, subtable, cls_match->priority);
+ subtable->n_rules++;
+
+ /* Rule was added, not replaced. Update 'subtable's 'max_priority'
+ * and 'max_count', if necessary. */
+ if (subtable->n_rules == 1) {
+ subtable->max_priority = new->priority;
+ subtable->max_count = 1;
+ pvector_insert(&cls->subtables, subtable, new->priority);
+ } else if (subtable->max_priority == new->priority) {
+ ++subtable->max_count;
+ } else if (new->priority > subtable->max_priority) {
+ subtable->max_priority = new->priority;
+ subtable->max_count = 1;
+ pvector_change_priority(&cls->subtables, subtable, new->priority);
+ }
} else {
/* Remove old node from indices. */
for (i = 0; i < subtable->n_indices; i++) {
- hindex_remove(&subtable->indices[i], &old->index_nodes[i]);
+ cmap_remove(&subtable->indices[i], &old->index_nodes[i], ihash[i]);
}
}
return old;
}
-
-static struct cls_match *
-next_rule_in_list__(struct cls_match *rule)
-{
- struct cls_match *next = OBJECT_CONTAINING(rule->list.next, next, list);
- return next;
-}
-
-static struct cls_match *
-next_rule_in_list(struct cls_match *rule)
-{
- struct cls_match *next = next_rule_in_list__(rule);
- return next->priority < rule->priority ? next : NULL;
-}
\f
/* A longest-prefix match tree. */
-struct trie_node {
- uint32_t prefix; /* Prefix bits for this node, MSB first. */
- uint8_t nbits; /* Never zero, except for the root node. */
- unsigned int n_rules; /* Number of rules that have this prefix. */
- struct trie_node *edges[2]; /* Both NULL if leaf. */
-};
-
-/* Max bits per node. Must fit in struct trie_node's 'prefix'.
- * Also tested with 16, 8, and 5 to stress the implementation. */
-#define TRIE_PREFIX_BITS 32
/* Return at least 'plen' bits of the 'prefix', starting at bit offset 'ofs'.
* Prefixes are in the network byte order, and the offset 0 corresponds to
return raw_get_prefix(pr, ofs, plen) & ~0u << (32 - plen);
}
-/* Return the number of equal bits in 'nbits' of 'prefix's MSBs and a 'value'
+/* Return the number of equal bits in 'n_bits' of 'prefix's MSBs and a 'value'
* starting at "MSB 0"-based offset 'ofs'. */
static unsigned int
-prefix_equal_bits(uint32_t prefix, unsigned int nbits, const ovs_be32 value[],
+prefix_equal_bits(uint32_t prefix, unsigned int n_bits, const ovs_be32 value[],
unsigned int ofs)
{
- uint64_t diff = prefix ^ raw_get_prefix(value, ofs, nbits);
+ uint64_t diff = prefix ^ raw_get_prefix(value, ofs, n_bits);
/* Set the bit after the relevant bits to limit the result. */
- return raw_clz64(diff << 32 | UINT64_C(1) << (63 - nbits));
+ return raw_clz64(diff << 32 | UINT64_C(1) << (63 - n_bits));
}
/* Return the number of equal bits in 'node' prefix and a 'prefix' of length
trie_prefix_equal_bits(const struct trie_node *node, const ovs_be32 prefix[],
unsigned int ofs, unsigned int plen)
{
- return prefix_equal_bits(node->prefix, MIN(node->nbits, plen - ofs),
+ return prefix_equal_bits(node->prefix, MIN(node->n_bits, plen - ofs),
prefix, ofs);
}
node->prefix = trie_get_prefix(prefix, ofs, plen);
if (plen <= TRIE_PREFIX_BITS) {
- node->nbits = plen;
- node->edges[0] = NULL;
- node->edges[1] = NULL;
+ node->n_bits = plen;
+ ovsrcu_set_hidden(&node->edges[0], NULL);
+ ovsrcu_set_hidden(&node->edges[1], NULL);
node->n_rules = n_rules;
} else { /* Need intermediate nodes. */
struct trie_node *subnode = trie_branch_create(prefix,
plen - TRIE_PREFIX_BITS,
n_rules);
int bit = get_bit_at(subnode->prefix, 0);
- node->nbits = TRIE_PREFIX_BITS;
- node->edges[bit] = subnode;
- node->edges[!bit] = NULL;
+ node->n_bits = TRIE_PREFIX_BITS;
+ ovsrcu_set_hidden(&node->edges[bit], subnode);
+ ovsrcu_set_hidden(&node->edges[!bit], NULL);
node->n_rules = 0;
}
return node;
}
static void
-trie_node_destroy(struct trie_node *node)
+trie_node_destroy(const struct trie_node *node)
{
- free(node);
+ ovsrcu_postpone(free, CONST_CAST(struct trie_node *, node));
}
+/* Copy a trie node for modification and postpone delete the old one. */
+static struct trie_node *
+trie_node_rcu_realloc(const struct trie_node *node)
+{
+ struct trie_node *new_node = xmalloc(sizeof *node);
+
+ *new_node = *node;
+ trie_node_destroy(node);
+
+ return new_node;
+}
+
+/* May only be called while holding the classifier mutex. */
static void
-trie_destroy(struct trie_node *node)
+trie_destroy(rcu_trie_ptr *trie)
{
+ struct trie_node *node = ovsrcu_get_protected(struct trie_node *, trie);
+
if (node) {
- trie_destroy(node->edges[0]);
- trie_destroy(node->edges[1]);
- free(node);
+ ovsrcu_set_hidden(trie, NULL);
+ trie_destroy(&node->edges[0]);
+ trie_destroy(&node->edges[1]);
+ trie_node_destroy(node);
}
}
static bool
trie_is_leaf(const struct trie_node *trie)
{
- return !trie->edges[0] && !trie->edges[1]; /* No children. */
+ /* No children? */
+ return !ovsrcu_get(struct trie_node *, &trie->edges[0])
+ && !ovsrcu_get(struct trie_node *, &trie->edges[1]);
}
static void
mask_set_prefix_bits(struct flow_wildcards *wc, uint8_t be32ofs,
- unsigned int nbits)
+ unsigned int n_bits)
{
ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs];
unsigned int i;
- for (i = 0; i < nbits / 32; i++) {
+ for (i = 0; i < n_bits / 32; i++) {
mask[i] = OVS_BE32_MAX;
}
- if (nbits % 32) {
- mask[i] |= htonl(~0u << (32 - nbits % 32));
+ if (n_bits % 32) {
+ mask[i] |= htonl(~0u << (32 - n_bits % 32));
}
}
static bool
mask_prefix_bits_set(const struct flow_wildcards *wc, uint8_t be32ofs,
- unsigned int nbits)
+ unsigned int n_bits)
{
ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs];
unsigned int i;
ovs_be32 zeroes = 0;
- for (i = 0; i < nbits / 32; i++) {
+ for (i = 0; i < n_bits / 32; i++) {
zeroes |= ~mask[i];
}
- if (nbits % 32) {
- zeroes |= ~mask[i] & htonl(~0u << (32 - nbits % 32));
+ if (n_bits % 32) {
+ zeroes |= ~mask[i] & htonl(~0u << (32 - n_bits % 32));
}
- return !zeroes; /* All 'nbits' bits set. */
+ return !zeroes; /* All 'n_bits' bits set. */
}
-static struct trie_node **
+static rcu_trie_ptr *
trie_next_edge(struct trie_node *node, const ovs_be32 value[],
unsigned int ofs)
{
trie_next_node(const struct trie_node *node, const ovs_be32 value[],
unsigned int ofs)
{
- return node->edges[be_get_bit_at(value, ofs)];
+ return ovsrcu_get(struct trie_node *,
+ &node->edges[be_get_bit_at(value, ofs)]);
}
-/* Return the prefix mask length necessary to find the longest-prefix match for
- * the '*value' in the prefix tree 'node'.
- * '*checkbits' is set to the number of bits in the prefix mask necessary to
- * determine a mismatch, in case there are longer prefixes in the tree below
- * the one that matched.
+/* Set the bit at ("MSB 0"-based) offset 'ofs'. 'ofs' can be greater than 31.
+ */
+static void
+be_set_bit_at(ovs_be32 value[], unsigned int ofs)
+{
+ ((uint8_t *)value)[ofs / 8] |= 1u << (7 - ofs % 8);
+}
+
+/* Returns the number of bits in the prefix mask necessary to determine a
+ * mismatch, in case there are longer prefixes in the tree below the one that
+ * matched.
+ * '*plens' will have a bit set for each prefix length that may have matching
+ * rules. The caller is responsible for clearing the '*plens' prior to
+ * calling this.
*/
static unsigned int
-trie_lookup_value(const struct trie_node *node, const ovs_be32 value[],
- unsigned int n_bits, unsigned int *checkbits)
+trie_lookup_value(const rcu_trie_ptr *trie, const ovs_be32 value[],
+ ovs_be32 plens[], unsigned int n_bits)
{
- unsigned int ofs = 0, match_len = 0;
const struct trie_node *prev = NULL;
+ const struct trie_node *node = ovsrcu_get(struct trie_node *, trie);
+ unsigned int match_len = 0; /* Number of matching bits. */
- for (; node; prev = node, node = trie_next_node(node, value, ofs)) {
+ for (; node; prev = node, node = trie_next_node(node, value, match_len)) {
unsigned int eqbits;
/* Check if this edge can be followed. */
- eqbits = prefix_equal_bits(node->prefix, node->nbits, value, ofs);
- ofs += eqbits;
- if (eqbits < node->nbits) { /* Mismatch, nothing more to be found. */
- /* Bit at offset 'ofs' differed. */
- *checkbits = ofs + 1; /* Includes the first mismatching bit. */
- return match_len;
+ eqbits = prefix_equal_bits(node->prefix, node->n_bits, value,
+ match_len);
+ match_len += eqbits;
+ if (eqbits < node->n_bits) { /* Mismatch, nothing more to be found. */
+ /* Bit at offset 'match_len' differed. */
+ return match_len + 1; /* Includes the first mismatching bit. */
}
/* Full match, check if rules exist at this prefix length. */
if (node->n_rules > 0) {
- match_len = ofs;
+ be_set_bit_at(plens, match_len - 1);
}
- if (ofs >= n_bits) {
- *checkbits = n_bits; /* Full prefix. */
- return match_len;
+ if (match_len >= n_bits) {
+ return n_bits; /* Full prefix. */
}
}
- /* node == NULL. Full match so far, but we came to a dead end.
- * need to exclude the other branch if it exists. */
- *checkbits = !prev || trie_is_leaf(prev) ? ofs : ofs + 1;
- return match_len;
+ /* node == NULL. Full match so far, but we tried to follow an
+ * non-existing branch. Need to exclude the other branch if it exists
+ * (it does not if we were called on an empty trie or 'prev' is a leaf
+ * node). */
+ return !prev || trie_is_leaf(prev) ? match_len : match_len + 1;
}
static unsigned int
trie_lookup(const struct cls_trie *trie, const struct flow *flow,
- unsigned int *checkbits)
+ union mf_value *plens)
{
const struct mf_field *mf = trie->field;
* field. Some match fields are used for multiple purposes, so we
* must check that the trie is relevant for this flow. */
if (mf_are_prereqs_ok(mf, flow)) {
- return trie_lookup_value(trie->root,
+ return trie_lookup_value(&trie->root,
&((ovs_be32 *)flow)[mf->flow_be32ofs],
- mf->n_bits, checkbits);
+ &plens->be32, mf->n_bits);
}
- *checkbits = 0; /* Value not used in this case. */
- return UINT_MAX;
+ memset(plens, 0xff, sizeof *plens); /* All prefixes, no skipping. */
+ return 0; /* Value not used in this case. */
}
/* Returns the length of a prefix match mask for the field 'mf' in 'minimask'.
minimask_get_prefix_len(const struct minimask *minimask,
const struct mf_field *mf)
{
- unsigned int nbits = 0, mask_tz = 0; /* Non-zero when end of mask seen. */
+ unsigned int n_bits = 0, mask_tz = 0; /* Non-zero when end of mask seen. */
uint8_t u32_ofs = mf->flow_be32ofs;
uint8_t u32_end = u32_ofs + mf->n_bytes / 4;
return 0; /* Mask not contiguous. */
}
mask_tz = ctz32(mask);
- nbits += 32 - mask_tz;
+ n_bits += 32 - mask_tz;
}
}
- return nbits;
+ return n_bits;
}
/*
}
static void
-trie_insert_prefix(struct trie_node **edge, const ovs_be32 *prefix, int mlen)
+trie_insert_prefix(rcu_trie_ptr *edge, const ovs_be32 *prefix, int mlen)
{
struct trie_node *node;
int ofs = 0;
/* Walk the tree. */
- for (; (node = *edge) != NULL;
+ for (; (node = ovsrcu_get_protected(struct trie_node *, edge));
edge = trie_next_edge(node, prefix, ofs)) {
unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen);
ofs += eqbits;
- if (eqbits < node->nbits) {
+ if (eqbits < node->n_bits) {
/* Mismatch, new node needs to be inserted above. */
int old_branch = get_bit_at(node->prefix, eqbits);
+ struct trie_node *new_parent;
- /* New parent node. */
- *edge = trie_branch_create(prefix, ofs - eqbits, eqbits,
- ofs == mlen ? 1 : 0);
-
- /* Adjust old node for its new position in the tree. */
+ new_parent = trie_branch_create(prefix, ofs - eqbits, eqbits,
+ ofs == mlen ? 1 : 0);
+ /* Copy the node to modify it. */
+ node = trie_node_rcu_realloc(node);
+ /* Adjust the new node for its new position in the tree. */
node->prefix <<= eqbits;
- node->nbits -= eqbits;
- (*edge)->edges[old_branch] = node;
+ node->n_bits -= eqbits;
+ ovsrcu_set_hidden(&new_parent->edges[old_branch], node);
/* Check if need a new branch for the new rule. */
if (ofs < mlen) {
- (*edge)->edges[!old_branch]
- = trie_branch_create(prefix, ofs, mlen - ofs, 1);
+ ovsrcu_set_hidden(&new_parent->edges[!old_branch],
+ trie_branch_create(prefix, ofs, mlen - ofs,
+ 1));
}
+ ovsrcu_set(edge, new_parent); /* Publish changes. */
return;
}
/* Full match so far. */
}
}
/* Must insert a new tree branch for the new rule. */
- *edge = trie_branch_create(prefix, ofs, mlen - ofs, 1);
+ ovsrcu_set(edge, trie_branch_create(prefix, ofs, mlen - ofs, 1));
}
/* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask
/* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask
* in 'rule'. */
static void
-trie_remove_prefix(struct trie_node **root, const ovs_be32 *prefix, int mlen)
+trie_remove_prefix(rcu_trie_ptr *root, const ovs_be32 *prefix, int mlen)
{
struct trie_node *node;
- struct trie_node **edges[sizeof(union mf_value) * 8];
+ rcu_trie_ptr *edges[sizeof(union mf_value) * 8];
int depth = 0, ofs = 0;
/* Walk the tree. */
for (edges[0] = root;
- (node = *edges[depth]) != NULL;
+ (node = ovsrcu_get_protected(struct trie_node *, edges[depth]));
edges[++depth] = trie_next_edge(node, prefix, ofs)) {
unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen);
- if (eqbits < node->nbits) {
+ if (eqbits < node->n_bits) {
/* Mismatch, nothing to be removed. This should never happen, as
* only rules in the classifier are ever removed. */
break; /* Log a warning. */
node->n_rules--;
/* Check if can prune the tree. */
- while (!node->n_rules && !(node->edges[0] && node->edges[1])) {
- /* No rules and at most one child node, remove this node. */
- struct trie_node *next;
- next = node->edges[0] ? node->edges[0] : node->edges[1];
+ while (!node->n_rules) {
+ struct trie_node *next,
+ *edge0 = ovsrcu_get_protected(struct trie_node *,
+ &node->edges[0]),
+ *edge1 = ovsrcu_get_protected(struct trie_node *,
+ &node->edges[1]);
+
+ if (edge0 && edge1) {
+ break; /* A branching point, cannot prune. */
+ }
+
+ /* Else have at most one child node, remove this node. */
+ next = edge0 ? edge0 : edge1;
if (next) {
- if (node->nbits + next->nbits > TRIE_PREFIX_BITS) {
+ if (node->n_bits + next->n_bits > TRIE_PREFIX_BITS) {
break; /* Cannot combine. */
}
+ next = trie_node_rcu_realloc(next); /* Modify. */
+
/* Combine node with next. */
- next->prefix = node->prefix | next->prefix >> node->nbits;
- next->nbits += node->nbits;
+ next->prefix = node->prefix | next->prefix >> node->n_bits;
+ next->n_bits += node->n_bits;
}
- trie_node_destroy(node);
/* Update the parent's edge. */
- *edges[depth] = next;
+ ovsrcu_set(edges[depth], next); /* Publish changes. */
+ trie_node_destroy(node);
+
if (next || !depth) {
/* Branch not pruned or at root, nothing more to do. */
break;
}
- node = *edges[--depth];
+ node = ovsrcu_get_protected(struct trie_node *,
+ edges[--depth]);
}
return;
}
projects / cascardo / ovs.git / blobdiff