2 * Copyright (c) 2011-2012 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@dragonflybsd.org>
6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * 3. Neither the name of The DragonFly Project nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific, prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 /***************************************************************************
38 ***************************************************************************
40 * The initial public-key exchange is implementing by transmitting a
41 * 512-byte buffer to the other side in a symmetrical fashion. This
42 * buffer contains the following:
44 * (1) A random session key.
46 * (2) A verifier to determine that the decode was successful. It encodes
47 * an XOR of each group of 4 bytes from the session key.
49 * (3) Additional configuration and additional random data.
51 * - The hammer2 message header magic for endian detect
53 * - The hammer2 protocol version. The two sides agree on the
56 * - All unused fields (junk*) are filled with random data.
58 * This structure must be exactly 512 bytes and expects to use 256-byte
61 struct hammer2_handshake {
62 char pad1[8]; /* 000 */
63 uint16_t magic; /* 008 HAMMER2_MSGHDR_MAGIC for endian detect */
64 uint16_t version; /* 00A hammer2 protocol version */
65 uint32_t flags; /* 00C protocol extension flags */
66 uint8_t sess[64]; /* 010 512-bit session key */
67 uint8_t verf[16]; /* 050 verifier = ~sess */
68 char quickmsg[32]; /* 060 reason for connecting */
69 char junk080[128]; /* 080-0FF */
70 char pad2[8]; /* 100-107 */
71 char junk100[256-8]; /* 108-1FF */
74 typedef struct hammer2_handshake hammer2_handshake_t;
76 /***************************************************************************
77 * LOW LEVEL MESSAGING *
78 ***************************************************************************
80 * hammer2_msg - A standalone copy of a message, typically referenced by
81 * or embedded in other structures, or used with I/O queues.
83 * These structures are strictly temporary, so they do not have to be
84 * particularly optimized for size. All possible message headers are
85 * directly embedded (any), and the message may contain a reference
86 * to allocated auxillary data. The structure is recycled quite often
89 * This structure is typically not used for storing persistent message
90 * state (see hammer2_persist for that).
93 struct hammer2_persist;
96 struct hammer2_iocom *iocom;
97 struct hammer2_persist *persist;
98 TAILQ_ENTRY(hammer2_msg) entry; /* queue */
99 char *aux_data; /* aux-data if any */
102 hammer2_any_t any; /* raw extended msg header */
105 typedef struct hammer2_msg hammer2_msg_t;
107 TAILQ_HEAD(hammer2_msg_queue, hammer2_msg);
108 typedef struct hammer2_msg_queue hammer2_msg_queue_t;
110 #define HAMMER2_MSGX_BSWAPPED 0x0001
113 * hammer2_ioq - An embedded component of hammer2_connect, holds state
114 * for the buffering and parsing of incoming and outgoing messages.
117 enum { HAMMER2_MSGQ_STATE_HEADER1,
118 HAMMER2_MSGQ_STATE_HEADER2,
119 HAMMER2_MSGQ_STATE_AUXDATA1,
120 HAMMER2_MSGQ_STATE_AUXDATA2,
121 HAMMER2_MSGQ_STATE_ERROR } state;
122 int fifo_beg; /* buffered data */
124 int hbytes; /* header size */
125 int abytes; /* aux_data size */
127 int seq; /* salt sequencer */
130 hammer2_msg_queue_t msgq;
133 typedef struct hammer2_ioq hammer2_ioq_t;
135 #define HAMMER2_IOQ_ERROR_SYNC 1 /* bad magic / out of sync */
136 #define HAMMER2_IOQ_ERROR_EOF 2 /* unexpected EOF */
137 #define HAMMER2_IOQ_ERROR_SOCK 3 /* read() error on socket */
138 #define HAMMER2_IOQ_ERROR_FIELD 4 /* invalid field */
139 #define HAMMER2_IOQ_ERROR_HCRC 5 /* core header crc bad */
140 #define HAMMER2_IOQ_ERROR_XCRC 6 /* ext header crc bad */
141 #define HAMMER2_IOQ_ERROR_ACRC 7 /* aux data crc bad */
142 #define HAMMER2_IOQ_ERROR_STATE 8 /* bad state */
143 #define HAMMER2_IOQ_ERROR_NOPEER 9 /* bad socket peer */
144 #define HAMMER2_IOQ_ERROR_NORKEY 10 /* no remote keyfile found */
145 #define HAMMER2_IOQ_ERROR_NOLKEY 11 /* no local keyfile found */
146 #define HAMMER2_IOQ_ERROR_KEYXCHGFAIL 12 /* key exchange failed */
147 #define HAMMER2_IOQ_ERROR_KEYFMT 13 /* key file format problem */
148 #define HAMMER2_IOQ_ERROR_BADURANDOM 14 /* /dev/urandom is bad */
150 #define HAMMER2_IOQ_MAXIOVEC 16
153 * hammer2_iocom - governs a messaging stream connection
155 struct hammer2_iocom {
156 hammer2_ioq_t ioq_rx;
157 hammer2_ioq_t ioq_tx;
158 hammer2_msg_queue_t freeq; /* free msgs hdr only */
159 hammer2_msg_queue_t freeq_aux; /* free msgs w/aux_data */
160 void (*recvmsg_callback)(struct hammer2_iocom *);
161 void (*sendmsg_callback)(struct hammer2_iocom *);
162 void (*altmsg_callback)(struct hammer2_iocom *);
163 int sock_fd; /* comm socket or pipe */
164 int alt_fd; /* thread signal, tty, etc */
168 char rxbuf[HAMMER2_MSGBUF_SIZE]; /* for ioq_rx only */
171 typedef struct hammer2_iocom hammer2_iocom_t;
173 #define HAMMER2_IOCOMF_EOF 0x00000001 /* EOF or ERROR on desc */
174 #define HAMMER2_IOCOMF_RREQ 0x00000002 /* request read-data event */
175 #define HAMMER2_IOCOMF_WREQ 0x00000004 /* request write-avail event */
176 #define HAMMER2_IOCOMF_WIDLE 0x00000008 /* request write-avail event */
177 #define HAMMER2_IOCOMF_SIGNAL 0x00000010
179 /***************************************************************************
180 * HIGH LEVEL MESSAGING *
181 ***************************************************************************
183 * Persistent state is stored via the hammer2_persist structure.
185 struct hammer2_persist {
186 uint32_t lcmd; /* recent command direction */
187 uint32_t lrep; /* recent reply direction */
190 typedef struct hammer2_persist hammer2_persist_t;
197 * The global registration structure consolidates information accumulated
198 * via the spanning tree algorithm and tells us which connection (link)
199 * is the best path to get to any given registration.
201 * glob_node - Splay entry for this registration in the global index
202 * of all registrations.
204 * glob_entry - tailq entry when this registration's best_span element
207 * span_list - Head of a simple list of spanning tree entries which
208 * we use to determine the best link.
210 * best_span - Which of the span structure on span_list is the best
213 * source_root - Splay tree root indexing all mesasges sent from this
214 * registration. The messages are indexed by
217 * target_root - Splay tree root indexing all messages being sent to
218 * this registration. The messages are indexed by
219 * {linkid,msgid}. XXX
222 * Whenever spanning tree data causes a registration's best_link field to
223 * change that registration is transmitted as spanning tree data to every
224 * active link. Note that pure clients to the cluster, of which there can
225 * be millions, typically do not transmit spanning tree data to each other.
227 * Each registration is assigned a unique linkid local to the node (another
228 * node might assign a different linkid to the same registration). This
229 * linkid must be persistent as long as messages are active and is used
230 * to identify the message source and target.
232 TAILQ_HEAD(hammer2_span_list, hammer2_span);
233 typedef struct hammer2_span_list hammer2_span_list_t;
236 SPLAY_ENTRY(hammer2_reg) glob_node; /* index of registrations */
237 TAILQ_ENTRY(hammer2_reg) glob_entry; /* when modified */
238 hammer2_span_list_t span_list; /* list of hammer2_span's */
239 hammer2_span_t *best_span; /* best span entry */
240 hammer2_pmsg_splay_head_t source_root; /* msgs sent from reg */
241 hammer2_pmsg_splay_head_t target_root; /* msgs sent to reg */
242 uuid_t pfs_id; /* key field */
243 uuid_t pfs_fsid; /* key field */
249 #define HAMMER2_PROTO_REGF_MODIFIED 0x0001
252 * Each link (connection) collects spanning tree data received via the
253 * link and stores it in these span structures.
255 struct hammer2_span {
256 TAILQ_ENTRY(hammer2_span) span_entry; /* from hammer2_reg */
257 SPLAY_ENTRY(hammer2_span) span_node; /* from hammer2_link */
259 hammer2_link_t *link;
264 * Most hammer2 messages represent transactions and have persistent state
265 * which must be recorded. Some messages, such as cache states and inode
266 * representations are very long-lasting transactions.
268 * Each node in the graph must keep track of the message state in order
269 * to perform the proper action when a connection is lost. To do this
270 * the message is indexed on the source and target (global) registration,
271 * and the actual span element the message was received on and transmitted
272 * to is recorded (allowing us to retrieve the physical links involved).
274 * The {source_reg, target_reg, msgid} uniquely identifies a message. Any
275 * streaming operations using the same msgid use the same rendezvous.
277 * It is important to note that recorded state must use the same physical
278 * link (and thus the same chain of links across the graph) as was 'forged'
279 * by the initial message for that msgid. If the source span a message is
280 * received on does not match the recorded source, or the recorded target
281 * is no longer routeable, the message will be returned or generate an ABORT
282 * with LINKFAIL as appropriate.
284 struct hammer2_pmsg {
285 SPLAY_ENTRY(hammer2_pmsg) source_reg;
286 SPLAY_ENTRY(hammer2_pmsg) target_reg;
287 hammer2_span_t *source;
288 hammer2_span_t *target;