FD.io VPP  v18.01-8-g0eacf49
Vector Packet Processing
fib.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 /**
16  * \brief
17  * A IP v4/6 independent FIB.
18  *
19  * The main functions provided by the FIB are as follows;
20  *
21  * - source priorities
22  *
23  * A route can be added to the FIB by more than entity or source. Sources
24  * include, but are not limited to, API, CLI, LISP, MAP, etc (for the full list
25  * see fib_entry.h). Each source provides the forwarding information (FI) that
26  * is has determined as required for that route. Since each source determines the
27  * FI using different best path and loop prevention algorithms, it is not
28  * correct for the FI of multiple sources to be combined. Instead the FIB must
29  * choose to use the FI from only one source. This choose is based on a static
30  * priority assignment. For example;
31  * IF a prefix is added as a result of interface configuration:
32  * set interface address 192.168.1.1/24 GigE0
33  * and then it is also added from the CLI
34  * ip route 192.168.1.1/32 via 2.2.2.2/32
35  * then the 'interface' source will prevail, and the route will remain as
36  * 'local'.
37  * The requirement of the FIB is to always install the FI from the winning
38  * source and thus to maintain the FI added by losing sources so it can be
39  * installed should the winning source be withdrawn.
40  *
41  * - adj-fib maintenance
42  *
43  * When ARP or ND discover a neighbour on a link an adjacency forms for the
44  * address of that neighbour. It is also required to insert a route in the
45  * appropriate FIB table, corresponding to the VRF for the link, an entry for
46  * that neighbour. This entry is often referred to as an adj-fib. Adj-fibs
47  * have a dedicated source; 'ADJ'.
48  * The priority of the ADJ source is lower than most. This is so the following
49  * config;
50  * set interface address 192.168.1.1/32 GigE0
51  * ip arp 192.168.1.2 GigE0 dead.dead.dead
52  * ip route add 192.168.1.2 via 10.10.10.10 GigE1
53  * will forward traffic for 192.168.1.2 via GigE1. That is the route added
54  * by the control plane is favoured over the adjacency discovered by ARP.
55  * The control plane, with its associated authentication, is considered the
56  * authoritative source.
57  * To counter the nefarious addition of adj-fib, through the nefarious injection
58  * of adjacencies, the FIB is also required to ensure that only adj-fibs whose
59  * less specific covering prefix is connected are installed in forwarding. This
60  * requires the use of 'cover tracking', where a route maintains a dependency
61  * relationship with the route that is its less specific cover. When this cover
62  * changes (i.e. there is a new covering route) or the forwarding information
63  * of the cover changes, then the covered route is notified.
64  *
65  * Overlapping sub-nets are not supported, so no adj-fib has multiple paths.
66  * The control plane is expected to remove a prefix configured for an interface
67  * before the interface changes VRF.
68  * So while the following config is accepted:
69  * set interface address 192.168.1.1/32 GigE0
70  * ip arp 192.168.1.2 GigE0 dead.dead.dead
71  * set interface ip table GigE0 2
72  * it does not result in the desired behaviour.
73  *
74  * - attached export.
75  *
76  * Further to adj-fib maintenance above consider the following config:
77  * set interface address 192.168.1.1/24 GigE0
78  * ip route add table 2 192.168.1.0/24 GigE0
79  * Traffic destined for 192.168.1.2 in table 2 will generate an ARP request
80  * on GigE0. However, since GigE0 is in table 0, all adj-fibs will be added in
81  * FIB 0. Hence all hosts in the sub-net are unreachable from table 2. To resolve
82  * this, all adj-fib and local prefixes are exported (i.e. copied) from the
83  * 'export' table 0, to the 'import' table 2. There can be many import tables
84  * for a single export table.
85  *
86  * - recursive route resolution
87  *
88  * A recursive route is of the form:
89  * 1.1.1.1/32 via 10.10.10.10
90  * i.e. a route for which no egress interface is provided. In order to forward
91  * traffic to 1.1.1.1/32 the FIB must therefore first determine how to forward
92  * traffic to 10.10.10.10/32. This is recursive resolution.
93  * Recursive resolution, just like normal resolution, proceeds via a longest
94  * prefix match for the 'via-address' 10.10.10.10. Note it is only possible
95  * to add routes via an address (i.e. a /32 or /128) not via a shorter mask
96  * prefix. There is no use case for the latter.
97  * Since recursive resolution proceeds via a longest prefix match, the entry
98  * in the FIB that will resolve the recursive route, termed the via-entry, may
99  * change as other routes are added to the FIB. Consider the recursive
100  * route shown above, and this non-recursive route:
101  * 10.10.10.0/24 via 192.168.16.1 GigE0
102  * The entry for 10.10.10.0/24 is thus the resolving via-entry. If this entry is
103  * modified, to say;
104  * 10.10.10.0/24 via 192.16.1.3 GigE0
105  * Then packet for 1.1.1.1/32 must also be sent to the new next-hop.
106  * Now consider the addition of;
107  * 10.10.10.0/28 via 192.168.16.2 GigE0
108  * The more specific /28 is a better longest prefix match and thus becomes the
109  * via-entry. Removal of the /28 means the resolution will revert to the /24.
110  * The tracking to the changes in recursive resolution is the requirement of
111  * the FIB. When the forwarding information of the via-entry changes a back-walk
112  * is used to update dependent recursive routes. When new routes are added to
113  * the table the cover tracking feature provides the necessary notifications to
114  * the via-entry routes.
115  * The adjacency constructed for 1.1.1.1/32 will be a recursive adjacency
116  * whose next adjacency will be contributed from the via-entry. Maintaining
117  * the validity of this recursive adjacency is a requirement of the FIB.
118  *
119  * - recursive loop avoidance
120  *
121  * Consider this set of routes:
122  * 1.1.1.1/32 via 2.2.2.2
123  * 2.2.2.2/32 via 3.3.3.3
124  * 3.3.3.3/32 via 1.1.1.1
125  * this is termed a recursion loop - all of the routes in the loop are
126  * unresolved in so far as they do not have a resolving adjacency, but each
127  * is resolved because the via-entry is known. It is important here to note
128  * the distinction between the control-plane objects and the data-plane objects
129  * (more details in the implementation section). The control plane objects must
130  * allow the loop to form (i.e. the graph becomes cyclic), however, the
131  * data-plane absolutely must not allow the loop to form, otherwise the packet
132  * would loop indefinitely and never egress the device - meltdown would follow.
133  * The control plane must allow the loop to form, because when the loop breaks,
134  * all members of the loop need to be updated. Forming the loop allows the
135  * dependencies to be correctly setup to allow this to happen.
136  * There is no limit to the depth of recursion supported by VPP so:
137  * 9.9.9.100/32 via 9.9.9.99
138  * 9.9.9.99/32 via 9.9.9.98
139  * 9.9.9.98/32 via 9.9.9.97
140  * ... turtles, turtles, turtles ...
141  * 9.9.9.1/32 via 10.10.10.10 Gig0
142  * is supported to as many layers of turtles is desired, however, when
143  * back-walking a graph (in this case from 9.9.9.1/32 up toward 9.9.9.100/32)
144  * a FIB needs to differentiate the case where the recursion is deep versus
145  * the case where the recursion is looped. A simple method, employed by VPP FIB,
146  * is to limit the number of steps. VPP FIB limit is 16. Typical BGP scenarios
147  * in the wild do not exceed 3 (BGP Inter-AS option C).
148  *
149  * - Fast Convergence
150  *
151  * After a network topology change, the 'convergence' time, is the time taken
152  * for the router to complete a transition to forward traffic using the new
153  * topology. The convergence time is therefore a summation of the time to;
154  * - detect the failure.
155  * - calculate the new 'best path' information
156  * - download the new best paths to the data-plane.
157  * - install those best best in data-plane forwarding.
158  * The last two points are of relevance to VPP architecture. The download API is
159  * binary and batch, details are not discussed here. There is no HW component to
160  * programme, installation time is bounded by the memory allocation and table
161  * lookup and insert access times.
162  *
163  * 'Fast' convergence refers to a set of technologies that a FIB can employ to
164  * completely or partially restore forwarding whilst the convergence actions
165  * listed above are ongoing. Fast convergence technologies are further
166  * sub-divided into Prefix Independent Convergence (PIC) and Loop Free
167  * Alternate path Fast re-route (LFA-FRR or sometimes called IP-FRR) which
168  * affect recursive and non-recursive routes respectively.
169  *
170  * LFA-FRR
171  *
172  * Consider the network topology below:
173  *
174  * C
175  * / \
176  * X -- A --- B - Y
177  * | |
178  * D F
179  * \ /
180  * E
181  *
182  * all links are equal cost, traffic is passing from X to Y. the best path is
183  * X-A-B-Y. There are two alternative paths, one via C and one via E. An
184  * alternate path is considered to be loop free if no other router on that path
185  * would forward the traffic back to the sender. Consider router C, its best
186  * path to Y is via B, so if A were to send traffic destined to Y to C, then C
187  * would forward that traffic to B - this is a loop-free alternate path. In
188  * contrast consider router D. D's shortest path to Y is via A, so if A were to
189  * send traffic destined to Y via D, then D would send it back to A; this is
190  * not a loop-free alternate path. There are several points of note;
191  * - we are considering the pre-failure routing topology
192  * - any equal-cost multi-path between A and B is also a LFA path.
193  * - in order for A to calculate LFA paths it must be aware of the best-path
194  * to Y from the perspective of D. These calculations are thus limited to
195  * routing protocols that have a full view of the network topology, i.e.
196  * link-state DB protocols like OSPF or an SDN controller. LFA protected
197  * prefixes are thus non-recursive.
198  *
199  * LFA is specified as a 1 to 1 redundancy; a primary path has only one LFA
200  * (a.k.a. backup) path. To my knowledge this limitation is one of complexity
201  * in the calculation of and capacity planning using a 1-n redundancy.
202  *
203  * In the event that the link A-B fails, the alternate path via C can be used.
204  * In order to provide 'fast' failover in the event of a failure, the control
205  * plane will download both the primary and the backup path to the FIB. It is
206  * then a requirement of the FIB to perform the failover (a.k.a cutover) from
207  * the primary to the backup path as quickly as possible, and particularly
208  * without any other control-plane intervention. The expectation is cutover is
209  * less than 50 milli-seconds - a value allegedly from the VOIP QoS. Note that
210  * cutover time still includes the fault detection time, which in a vitalised
211  * environment could be the dominant factor. Failure detection can be either a
212  * link down, which will affect multiple paths on a multi-access interface, or
213  * via a specific path heartbeat (i.e. BFD).
214  * At this time VPP does not support LFA, that is it does not support the
215  * installation of a primary and backup path[s] for a route. However, it does
216  * support ECMP, and VPP FIB is designed to quickly remove failed paths from
217  * the ECMP set, however, it does not insert shared objects specific to the
218  * protected resource into the forwarding object graph, since this would incur
219  * a forwarding/performance cost. Failover time is thus route number dependent.
220  * Details are provided in the implementation section below.
221  *
222  * PIC
223  *
224  * PIC refers to the concept that the converge time should be independent of
225  * the number of prefixes/routes that are affected by the failure. PIC is
226  * therefore most appropriate when considering networks with large number of
227  * prefixes, i.e. BGP networks and thus recursive prefixes. There are several
228  * flavours of PIC covering different locations of protection and failure
229  * scenarios. An outline is given below, see the literature for more details:
230  *
231  * Y/16 - CE1 -- PE1---\
232  * | \ P1---\
233  * | \ PE3 -- CE3 - X/16
234  * | - P2---/
235  * Y/16 - CE2 -- PE2---/
236  *
237  * CE = customer edge, PE = provider edge. external-BGP runs between customer
238  * and provider, internal-BGP runs between provider and provider.
239  *
240  * 1) iBGP PIC-core: consider traffic from CE1 to X/16 via CE3. On PE1 there is
241  * are routes;
242  * X/16 (and hundreds of thousands of others like it)
243  * via PE3
244  * and
245  * PE3/32 (its loopback address)
246  * via 10.0.0.1 Link0 (this is P1)
247  * via 10.1.1.1 Link1 (this is P2)
248  * the failure is the loss of link0 or link1
249  * As in all PIC scenarios, in order to provide prefix independent convergence
250  * it must be that the route for X/16 (and all other routes via PE3) do not
251  * need to be updated in the FIB. The FIB therefore needs to update a single
252  * object that is shared by all routes - once this shared object is updated,
253  * then all routes using it will be instantly updated to use the new forwarding
254  * information. In this case the shared object is the resolving route via PE3.
255  * Once the route via PE3 is updated via IGP (OSPF) convergence, then all
256  * recursive routes that resolve through it are also updated. VPP FIB
257  * implements this scenario via a recursive-adjacency. the X/16 and it sibling
258  * routes share a recursive-adjacency that links to/points at/stacks on the
259  * normal adjacency contributed by the route for PE3. Once this shared
260  * recursive adj is re-linked then all routes are switched to using the new
261  * forwarding information. This is shown below;
262  *
263  * pre-failure;
264  * X/16 --> R-ADJ-1 --> ADJ-1-PE3 (multi-path via P1 and P2)
265  *
266  * post-failure:
267  * X/16 --> R-ADJ-1 --> ADJ-2-PE3 (single path via P1)
268  *
269  * note that R-ADJ-1 (the recursive adj) remains in the forwarding graph,
270  * therefore X/16 (and all its siblings) is not updated.
271  * X/16 and its siblings share the recursive adj since they share the same
272  * path-list. It is the path-list object that contributes the recursive-adj
273  * (see next section for more details)
274  *
275  *
276  * 2) iBGP PIC-edge; Traffic from CE3 to Y/16. On PE3 there is are routes;
277  * Y/16 (and hundreds of thousands of others like it)
278  * via PE1
279  * via PE2
280  * and
281  * PE1/32 (PE1's loopback address)
282  * via 10.0.2.2 Link0 (this is P1)
283  * PE2/32 (PE2's loopback address)
284  * via 10.0.3.3 Link1 (this is P2)
285  *
286  * the failure is the loss of reachability to PE2. this could be either the
287  * loss of the link P2-PE2 or the loss of the node PE2. This is detected either
288  * by the withdrawal of the PE2's loopback route or by some form of failure
289  * detection (i.e. BFD).
290  * VPP FIB again provides PIC via the use of the shared recursive-adj. Y/16 and
291  * its siblings will again share a path-list for the list {PE1,PE2}, this
292  * path-list will contribute a multi-path-recursive-adj, i.e. a multi-path-adj
293  * with each choice therein being another adj;
294  *
295  * Y/16 -> RM-ADJ --> ADJ1 (for PE1)
296  * --> ADJ2 (for PE2)
297  *
298  * when the route for PE1 is withdrawn then the multi-path-recursive-adjacency
299  * is updated to be;
300  *
301  * Y/16 --> RM-ADJ --> ADJ1 (for PE1)
302  * --> ADJ1 (for PE1)
303  *
304  * that is both choices in the ECMP set are the same and thus all traffic is
305  * forwarded to PE1. Eventually the control plane will download a route update
306  * for Y/16 to be via PE1 only. At that time the situation will be:
307  *
308  * Y/16 -> R-ADJ --> ADJ1 (for PE1)
309  *
310  * In the scenario above we assumed that PE1 and PE2 are ECMP for Y/16. eBGP
311  * PIC core is also specified for the case were one PE is primary and the other
312  * backup - VPP FIB does not support that case at this time.
313  *
314  * 3) eBGP PIC Edge; Traffic from CE3 to Y/16. On PE1 there is are routes;
315  * Y/16 (and hundreds of thousands of others like it)
316  * via CE1 (primary)
317  * via PE2 (backup)
318  * and
319  * CE1 (this is an adj-fib)
320  * via 11.0.0.1 Link0 (this is CE1) << this is an adj-fib
321  * PE2 (PE2's loopback address)
322  * via 10.0.5.5 Link1 (this is link PE1-PE2)
323  * the failure is the loss of link0 to CE1. The failure can be detected by FIB
324  * either as a link down event or by the control plane withdrawing the connected
325  * prefix on the link0 (say 10.0.5.4/30). The latter works because the resolving
326  * entry is an adj-fib, so removing the connected will withdraw the adj-fib, and
327  * hence the recursive path becomes unresolved. The former is faster,
328  * particularly in the case of Inter-AS option A where there are many VLAN
329  * sub-interfaces on the PE-CE link, one for each VRF, and so the control plane
330  * must remove the connected prefix for each sub-interface to trigger PIC in
331  * each VRF. Note though that total PIC cutover time will depend on VRF scale
332  * with either trigger.
333  * Primary and backup paths in this eBGP PIC-edge scenario are calculated by
334  * BGP. Each peer is configured to always advertise its best external path to
335  * its iBGP peers. Backup paths therefore send traffic from the PE back into the
336  * core to an alternate PE. A PE may have multiple external paths, i.e. multiple
337  * directly connected CEs, it may also have multiple backup PEs, however there
338  * is no correlation between the two, so unlike LFA-FRR, the redundancy model is
339  * N-M; N primary paths are backed-up by M backup paths - only when all primary
340  * paths fail, then the cutover is performed onto the M backup paths. Note that
341  * PE2 must be suitably configured to forward traffic on its external path that
342  * was received from PE1. VPP FIB does not support external-internal-BGP (eiBGP)
343  * load-balancing.
344  *
345  * As with LFA-FRR the use of primary and backup paths is not currently
346  * supported, however, the use of a recursive-multi-path-adj, and a suitably
347  * constrained hashing algorithm to choose from the primary or backup path sets,
348  * would again provide the necessary shared object and hence the prefix scale
349  * independent cutover.
350  *
351  * Astute readers will recognise that both of the eBGP PIC scenarios refer only
352  * to a BGP free core.
353  *
354  * Fast convergence implementation options come in two flavours:
355  * 1) Insert switches into the data-path. The switch represents the protected
356  * resource. If the switch is 'on' the primary path is taken, otherwise
357  * the backup path is taken. Testing the switch in the data-path comes with
358  * an associated performance cost. A given packet may encounter more than
359  * one protected resource as it is forwarded. This approach minimises
360  * cutover times as packets will be forwarded on the backup path as soon
361  * as the protected resource is detected to be down and the single switch
362  * is tripped. However, it comes at a performance cost, which increases
363  * with each shared resource a packet encounters in the data-path.
364  * This approach is thus best suited to LFA-FRR where the protected routes
365  * are non-recursive (i.e. encounter few shared resources) and the
366  * expectation on cutover times is more stringent (<50msecs).
367  * 2) Update shared objects. Identify objects in the data-path, that are
368  * required to be present whether or not fast convergence is required (i.e.
369  * adjacencies) that can be shared by multiple routes. Create a dependency
370  * between these objects at the protected resource. When the protected
371  * resource fails, each of the shared objects is updated in a way that all
372  * users of it see a consistent change. This approach incurs no performance
373  * penalty as the data-path structure is unchanged, however, the cutover
374  * times are longer as more work is required when the resource fails. This
375  * scheme is thus more appropriate to recursive prefixes (where the packet
376  * will encounter multiple protected resources) and to fast-convergence
377  * technologies where the cutover times are less stringent (i.e. PIC).
378  *
379  * Implementation:
380  * ---------------
381  *
382  * Due to the requirements outlined above, not all routes known to FIB
383  * (e.g. adj-fibs) are installed in forwarding. However, should circumstances
384  * change, those routes will need to be added. This adds the requirement that
385  * a FIB maintains two tables per-VRF, per-AF (where a 'table' is indexed by
386  * prefix); the forwarding and non-forwarding tables.
387  *
388  * For DP speed in VPP we want the lookup in the forwarding table to directly
389  * result in the ADJ. So the two tables; one contains all the routes (a
390  * lookup therein yields a fib_entry_t), the other contains only the forwarding
391  * routes (a lookup therein yields an ip_adjacency_t). The latter is used by the
392  * DP.
393  * This trades memory for forwarding performance. A good trade-off in VPP's
394  * expected operating environments.
395  *
396  * Note these tables are keyed only by the prefix (and since there 2 two
397  * per-VRF, implicitly by the VRF too). The key for an adjacency is the
398  * tuple:{next-hop, address (and it's AF), interface, link/ether-type}.
399  * consider this curious, but allowed, config;
400  *
401  * set int ip addr 10.0.0.1/24 Gig0
402  * set ip arp Gig0 10.0.0.2 dead.dead.dead
403  * # a host in that sub-net is routed via a better next hop (say it avoids a
404  * # big L2 domain)
405  * ip route add 10.0.0.2 Gig1 192.168.1.1
406  * # this recursive should go via Gig1
407  * ip route add 1.1.1.1/32 via 10.0.0.2
408  * # this non-recursive should go via Gig0
409  * ip route add 2.2.2.2/32 via Gig0 10.0.0.2
410  *
411  * for the last route, the lookup for the path (via {Gig0, 10.0.0.2}) in the
412  * prefix table would not yield the correct result. To fix this we need a
413  * separate table for the adjacencies.
414  *
415  * - FIB data structures;
416  *
417  * fib_entry_t:
418  * - a representation of a route.
419  * - has a prefix.
420  * - it maintains an array of path-lists that have been contributed by the
421  * different sources
422  * - install an adjacency in the forwarding table contributed by the best
423  * source's path-list.
424  *
425  * fib_path_list_t:
426  * - a list of paths
427  * - path-lists may be shared between FIB entries. The path-lists are thus
428  * kept in a DB. The key is the combined description of the paths. We share
429  * path-lists when it will aid convergence to do so. Adding path-lists to
430  * this DB that are never shared, or are not shared by prefixes that are
431  * not subject to PIC, will increase the size of the DB unnecessarily and
432  * may lead to increased search times due to hash collisions.
433  * - the path-list contributes the appropriate adj for the entry in the
434  * forwarding table. The adj can be 'normal', multi-path or recursive,
435  * depending on the number of paths and their types.
436  * - since path-lists are shared there is only one instance of the multi-path
437  * adj that they [may] create. As such multi-path adjacencies do not need a
438  * separate DB.
439  * The path-list with recursive paths and the recursive adjacency that it
440  * contributes forms the backbone of the fast convergence architecture (as
441  * described previously).
442  *
443  * fib_path_t:
444  * - a description of how to forward the traffic (i.e. via {Gig1, K}).
445  * - the path describes the intent on how to forward. This differs from how
446  * the path resolves. I.e. it might not be resolved at all (since the
447  * interface is deleted or down).
448  * - paths have different types, most notably recursive or non-recursive.
449  * - a fib_path_t will contribute the appropriate adjacency object. It is from
450  * these contributions that the DP graph/chain for the route is built.
451  * - if the path is recursive and a recursion loop is detected, then the path
452  * will contribute the special DROP adjacency. This way, whilst the control
453  * plane graph is looped, the data-plane graph does not.
454  *
455  * we build a graph of these objects;
456  *
457  * fib_entry_t -> fib_path_list_t -> fib_path_t -> ...
458  *
459  * for recursive paths:
460  *
461  * fib_path_t -> fib_entry_t -> ....
462  *
463  * for non-recursive paths
464  *
465  * fib_path_t -> ip_adjacency_t -> interface
466  *
467  * These objects, which constitute the 'control plane' part of the FIB are used
468  * to represent the resolution of a route. As a whole this is referred to as the
469  * control plane graph. There is a separate DP graph to represent the forwarding
470  * of a packet. In the DP graph each object represents an action that is applied
471  * to a packet as it traverses the graph. For example, a lookup of a IP address
472  * in the forwarding table could result in the following graph:
473  *
474  * recursive-adj --> multi-path-adj --> interface_A
475  * --> interface_B
476  *
477  * A packet traversing this FIB DP graph would thus also traverse a VPP node
478  * graph of:
479  *
480  * ipX_recursive --> ipX_rewrite --> interface_A_tx --> etc
481  *
482  * The taxonomy of objects in a FIB graph is as follows, consider;
483  *
484  * A -->
485  * B --> D
486  * C -->
487  *
488  * Where A,B and C are (for example) routes that resolve through D.
489  * parent; D is the parent of A, B, and C.
490  * children: A, B, and C are children of D.
491  * sibling: A, B and C are siblings of one another.
492  *
493  * All shared objects in the FIB are reference counted. Users of these objects
494  * are thus expected to use the add_lock/unlock semantics (as one would
495  * normally use malloc/free).
496  *
497  * WALKS
498  *
499  * It is necessary to walk/traverse the graph forwards (entry to interface) to
500  * perform a collapse or build a recursive adj and backwards (interface
501  * to entry) to perform updates, i.e. when interface state changes or when
502  * recursive route resolution updates occur.
503  * A forward walk follows simply by navigating an object's parent pointer to
504  * access its parent object. For objects with multiple parents (e.g. a
505  * path-list), each parent is walked in turn.
506  * To support back-walks direct dependencies are maintained between objects,
507  * i.e. in the relationship, {A, B, C} --> D, then object D will maintain a list
508  * of 'pointers' to its children {A, B, C}. Bare C-language pointers are not
509  * allowed, so a pointer is described in terms of an object type (i.e. entry,
510  * path-list, etc) and index - this allows the object to be retrieved from the
511  * appropriate pool. A list is maintained to achieve fast convergence at scale.
512  * When there are millions or recursive prefixes, it is very inefficient to
513  * blindly walk the tables looking for entries that were affected by a given
514  * topology change. The lowest hanging fruit when optimising is to remove
515  * actions that are not required, so all back-walks only traverse objects that
516  * are directly affected by the change.
517  *
518  * PIC Core and fast-reroute rely on FIB reacting quickly to an interface
519  * state change to update the multi-path-adjacencies that use this interface.
520  * An example graph is shown below:
521  *
522  * E_a -->
523  * E_b --> PL_2 --> P_a --> Interface_A
524  * ... --> P_c -\
525  * E_k --> \
526  * Interface_K
527  * /
528  * E_l --> /
529  * E_m --> PL_1 --> P_d -/
530  * ... --> P_f --> Interface_F
531  * E_z -->
532  *
533  * E = fib_entry_t
534  * PL = fib_path_list_t
535  * P = fib_path_t
536  * The subscripts are arbitrary and serve only to distinguish object instances.
537  * This CP graph result in the following DP graph:
538  *
539  * M-ADJ-2 --> Interface_A
540  * \
541  * -> Interface_K
542  * /
543  * M-ADJ-1 --> Interface_F
544  *
545  * M-ADJ = multi-path-adjacency.
546  *
547  * When interface K goes down a back-walk is started over its dependants in the
548  * control plane graph. This back-walk will reach PL_1 and PL_2 and result in
549  * the calculation of new adjacencies that have interface K removed. The walk
550  * will continue to the entry objects and thus the forwarding table is updated
551  * for each prefix with the new adjacency. The DP graph then becomes:
552  *
553  * ADJ-3 --> Interface_A
554  *
555  * ADJ-4 --> Interface_F
556  *
557  * The eBGP PIC scenarios described above relied on the update of a path-list's
558  * recursive-adjacency to provide the shared point of cutover. This is shown
559  * below
560  *
561  * E_a -->
562  * E_b --> PL_2 --> P_a --> E_44 --> PL_a --> P_b --> Interface_A
563  * ... --> P_c -\
564  * E_k --> \
565  * \
566  * E_1 --> PL_k -> P_k --> Interface_K
567  * /
568  * E_l --> /
569  * E_m --> PL_1 --> P_d -/
570  * ... --> P_f --> E_55 --> PL_e --> P_e --> Interface_E
571  * E_z -->
572  *
573  * The failure scenario is the removal of entry E_1 and thus the paths P_c and
574  * P_d become unresolved. To achieve PIC the two shared recursive path-lists,
575  * PL_1 and PL_2 must be updated to remove E_1 from the recursive-multi-path-
576  * adjacencies that they contribute, before any entry E_a to E_z is updated.
577  * This means that as the update propagates backwards (right to left) in the
578  * graph it must do so breadth first not depth first. Note this approach leads
579  * to convergence times that are dependent on the number of path-list and so
580  * the number of combinations of egress PEs - this is desirable as this
581  * scale is considerably lower than the number of prefixes.
582  *
583  * If we consider another section of the graph that is similar to the one
584  * shown above where there is another prefix E_2 in a similar position to E_1
585  * and so also has many dependent children. It is reasonable to expect that a
586  * particular network failure may simultaneously render E_1 and E_2 unreachable.
587  * This means that the update to withdraw E_2 is download immediately after the
588  * update to withdraw E_1. It is a requirement on the FIB to not spend large
589  * amounts of time in a back-walk whilst processing the update for E_1, i.e. the
590  * back-walk must not reach as far as E_a and its siblings. Therefore, after the
591  * back-walk has traversed one generation (breadth first) to update all the
592  * path-lists it should be suspended/back-ground and further updates allowed
593  * to be handled. Once the update queue is empty, the suspended walks can be
594  * resumed. Note that in the case that multiple updates affect the same entry
595  * (say E_1) then this will trigger multiple similar walks, these are merged,
596  * so each child is updated only once.
597  * In the presence of more layers of recursion PIC is still a desirable
598  * feature. Consider an extension to the diagram above, where more recursive
599  * routes (E_100 -> E_200) are added as children of E_a:
600  *
601  * E_100 -->
602  * E_101 --> PL_3 --> P_j-\
603  * ... \
604  * E_199 --> E_a -->
605  * E_b --> PL_2 --> P_a --> E_44 --> ...etc..
606  * ... --> P_c -\
607  * E_k \
608  * E_1 --> ...etc..
609  * /
610  * E_l --> /
611  * E_m --> PL_1 --> P_d -/
612  * ... --> P_e --> E_55 --> ...etc..
613  * E_z -->
614  *
615  * To achieve PIC for the routes E_100->E_199, PL_3 needs to be updated before
616  * E_b -> E_z, a breadth first traversal at each level would not achieve this.
617  * Instead the walk must proceed intelligently. Children on PL_2 are sorted so
618  * those Entry objects that themselves have children appear first in the list,
619  * those without later. When an entry object is walked that has children, a
620  * walk of its children is pushed to the front background queue. The back
621  * ground queue is a priority queue. As the breadth first traversal proceeds
622  * across the dependent entry object E_a to E_k, when the first entry that does
623  * not have children is reached (E_b), the walk is suspended and placed at the
624  * back of the queue. Following this prioritisation method shared path-list
625  * updates are performed before all non-resolving entry objects.
626  * The CPU/core/thread that handles the updates is the same thread that handles
627  * the back-walks. Handling updates has a higher priority than making walk
628  * progress, so a walk is required to be interruptable/suspendable when new
629  * updates are available.
630  * !!! TODO - this section describes how walks should be not how they are !!!
631  *
632  * In the diagram above E_100 is an IP route, however, VPP has no restrictions
633  * on the type of object that can be a dependent of a FIB entry. Children of
634  * a FIB entry can be (and are) GRE & VXLAN tunnels endpoints, L2VPN LSPs etc.
635  * By including all object types into the graph and extending the back-walk, we
636  * can thus deliver fast convergence to technologies that overlay on an IP
637  * network.
638  *
639  * If having read all the above carefully you are still thinking; 'i don't need
640  * all this %&$* i have a route only I know about and I just need to jam it in',
641  * then fib_table_entry_special_add() is your only friend.
642  */
643 
644 #ifndef __FIB_H__
645 #define __FIB_H__
646 
647 #include <vnet/fib/fib_table.h>
648 #include <vnet/fib/fib_entry.h>
649 
650 #endif