PageRenderTime 23ms CodeModel.GetById 16ms app.highlight 5ms RepoModel.GetById 1ms app.codeStats 0ms

Plain Text | 145 lines | 114 code | 31 blank | 0 comment | 0 complexity | 9c35f6fe935ed4175ee1e73ff6cda34c MD5 | raw file
Possible License(s): GPL-2.0, LGPL-2.0, AGPL-1.0
  1			LC-trie implementation notes.
  3Node types
  6	An end node with data. This has a copy of the relevant key, along
  7	with 'hlist' with routing table entries sorted by prefix length.
  8	See struct leaf and struct leaf_info.
 10trie node or tnode
 11	An internal node, holding an array of child (leaf or tnode) pointers,
 12	indexed	through a subset of the key. See Level Compression.
 14A few concepts explained
 16Bits (tnode) 
 17	The number of bits in the key segment used for indexing into the
 18	child array - the "child index". See Level Compression.
 20Pos (tnode)
 21	The position (in the key) of the key segment used for indexing into
 22	the child array. See Path Compression.
 24Path Compression / skipped bits
 25	Any given tnode is linked to from the child array of its parent, using
 26	a segment of the key specified by the parent's "pos" and "bits" 
 27	In certain cases, this tnode's own "pos" will not be immediately
 28	adjacent to the parent (pos+bits), but there will be some bits
 29	in the key skipped over because they represent a single path with no
 30	deviations. These "skipped bits" constitute Path Compression.
 31	Note that the search algorithm will simply skip over these bits when
 32	searching, making it necessary to save the keys in the leaves to
 33	verify that they actually do match the key we are searching for.
 35Level Compression / child arrays
 36	the trie is kept level balanced moving, under certain conditions, the
 37	children of a full child (see "full_children") up one level, so that
 38	instead of a pure binary tree, each internal node ("tnode") may
 39	contain an arbitrarily large array of links to several children.
 40	Conversely, a tnode with a mostly empty	child array (see empty_children)
 41	may be "halved", having some of its children moved downwards one level,
 42	in order to avoid ever-increasing child arrays.
 45	the number of positions in the child array of a given tnode that are
 46	NULL.
 49	the number of children of a given tnode that aren't path compressed.
 50	(in other words, they aren't NULL or leaves and their "pos" is equal
 51	to this	tnode's "pos"+"bits").
 53	(The word "full" here is used more in the sense of "complete" than
 54	as the opposite of "empty", which might be a tad confusing.)
 59We have tried to keep the structure of the code as close to fib_hash as 
 60possible to allow verification and help up reviewing. 
 63	A good start for understanding this code. This function implements a
 64	straightforward trie lookup.
 67	Inserts a new leaf node in the trie. This is bit more complicated than
 68	fib_find_node(). Inserting a new node means we might have to run the
 69	level compression algorithm on part of the trie.
 72	Looks up a key, deletes it and runs the level compression algorithm.
 75	The key function for the dynamic trie after any change in the trie
 76	it is run to optimize and reorganize. Tt will walk the trie upwards 
 77	towards the root from a given tnode, doing a resize() at each step 
 78	to implement level compression.
 81	Analyzes a tnode and optimizes the child array size by either inflating
 82	or shrinking it repeatedly until it fulfills the criteria for optimal
 83	level compression. This part follows the original paper pretty closely
 84	and there may be some room for experimentation here.
 87	Doubles the size of the child array within a tnode. Used by resize().
 90	Halves the size of the child array within a tnode - the inverse of
 91	inflate(). Used by resize();
 93fn_trie_insert(), fn_trie_delete(), fn_trie_select_default()
 94	The route manipulation functions. Should conform pretty closely to the
 95	corresponding functions in fib_hash.
 98	This walks the full trie (using nextleaf()) and searches for empty
 99	leaves which have to be removed.
102	Dumps the routing table ordered by prefix length. This is somewhat
103	slower than the corresponding fib_hash function, as we have to walk the
104	entire trie for each prefix length. In comparison, fib_hash is organized
105	as one "zone"/hash per prefix length.
110fib_lock is used for an RW-lock in the same way that this is done in fib_hash.
111However, the functions are somewhat separated for other possible locking
112scenarios. It might conceivably be possible to run trie_rebalance via RCU
113to avoid read_lock in the fn_trie_lookup() function.
115Main lookup mechanism
117fn_trie_lookup() is the main lookup function.
119The lookup is in its simplest form just like fib_find_node(). We descend the
120trie, key segment by key segment, until we find a leaf. check_leaf() does
121the fib_semantic_match in the leaf's sorted prefix hlist.
123If we find a match, we are done.
125If we don't find a match, we enter prefix matching mode. The prefix length,
126starting out at the same as the key length, is reduced one step at a time,
127and we backtrack upwards through the trie trying to find a longest matching
128prefix. The goal is always to reach a leaf and get a positive result from the
129fib_semantic_match mechanism.
131Inside each tnode, the search for longest matching prefix consists of searching
132through the child array, chopping off (zeroing) the least significant "1" of
133the child index until we find a match or the child index consists of nothing but
136At this point we backtrack (t->stats.backtrack++) up the trie, continuing to
137chop off part of the key in order to find the longest matching prefix.
139At this point we will repeatedly descend subtries to look for a match, and there
140are some optimizations available that can provide us with "shortcuts" to avoid
141descending into dead ends. Look for "HL_OPTIMIZE" sections in the code.
143To alleviate any doubts about the correctness of the route selection process,
144a new netlink operation has been added. Look for NETLINK_FIB_LOOKUP, which
145gives userland access to fib_lookup().