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>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 #ifndef _SYS_MALLOC_H_
39 #include <sys/malloc.h>
44 #ifndef _SYS_THREAD_H_
45 #include <sys/thread.h>
52 * Mesh network protocol structures.
56 * The mesh is constructed from point-to-point streaming links with varying
57 * levels of interconnectedness, forming a graph. Terminii in the graph
58 * are entities such as a HAMMER2 PFS or a network mount or other types
61 * Upon connecting and after authentication, a LNK_CONN transaction is opened
62 * on circuit 0 by both ends. This configures and enables the SPAN protocol.
63 * The LNK_CONN transaction remains open for the life of the connection.
67 * Once enabled, termini transmits a representitive LNK_SPAN out all
68 * available connections advertising what it is. Nodes maintaing multiple
69 * connections will relay received LNK_SPANs out available connections
70 * with some filtering based on the CONN configuration. A distance metric
71 * and per-node random value (rnss) is aggregated.
73 * Since LNK_SPANs can rapidly multiply in a complex graph, not all incoming
74 * LNK_SPANs will be relayed. Only the top N over all collect LNK_SPANs for
75 * any given advertisement are relayed.
77 * It is possible to code the SPANning tree algorithm to guarantee that
78 * symmetrical spans will be generated after stabilization. The RNSS field
79 * is used to help distinguish and reduce paths in complex graphs when
80 * symmetric spans are desired. We always generate RNSS but we currently do
81 * not implement symmetrical SPAN guarantees.
85 * We aren't done yet. Before transactions can be relayed, symmetric paths
86 * must be formed via the LNK_CIRC protocol. The LNK_CIRC protocol
87 * establishes a virtual circuit from any node to any other node, creating
88 * a circuit id which is stored in dmsg_hdr.circuit. Messages received on
89 * one side or forwarded to the other. Forwarded messages bypass normal
92 * A virtual circuit is forged by working the propogated SPANs backwards.
93 * Each node in the graph helps propagate the virtual circuit by attach the
94 * LNK_CIRC transaction it receives to a LNK_CIRC transaction it initiates
95 * out the other interface.
97 * Since SPANs are link-state transactions any change in related span(s)
98 * will also force-terminate VC's using those spans.
100 * MESSAGE TRANSACTIONAL STATES
102 * Message state is handled by the CREATE, DELETE, REPLY, and ABORT
103 * flags. Message state is typically recorded at the end points and
104 * at each hop until a DELETE is received from both sides.
106 * One-way messages such as those used by spanning tree commands are not
107 * recorded. These are sent without the CREATE, DELETE, or ABORT flags set.
108 * ABORT is not supported for one-off messages. The REPLY bit can be used
109 * to distinguish between command and status if desired.
111 * Persistent-state messages are messages which require a reply to be
112 * returned. These messages can also consist of multiple message elements
113 * for the command or reply or both (or neither). The command message
114 * sequence sets CREATE on the first message and DELETE on the last message.
115 * A single message command sets both (CREATE|DELETE). The reply message
116 * sequence works the same way but of course also sets the REPLY bit.
118 * Persistent-state messages can be aborted by sending a message element
119 * with the ABORT flag set. This flag can be combined with either or both
120 * the CREATE and DELETE flags. When combined with the CREATE flag the
121 * command is treated as non-blocking but still executes. Whem combined
122 * with the DELETE flag no additional message elements are required.
124 * ABORT SPECIAL CASE - Mid-stream aborts. A mid-stream abort can be sent
125 * when supported by the sender by sending an ABORT message with neither
126 * CREATE or DELETE set. This effectively turns the message into a
127 * non-blocking message (but depending on what is being represented can also
128 * cut short prior data elements in the stream).
130 * ABORT SPECIAL CASE - Abort-after-DELETE. Persistent messages have to be
131 * abortable if the stream/pipe/whatever is lost. In this situation any
132 * forwarding relay needs to unconditionally abort commands and replies that
133 * are still active. This is done by sending an ABORT|DELETE even in
134 * situations where a DELETE has already been sent in that direction. This
135 * is done, for example, when links are in a half-closed state. In this
136 * situation it is possible for the abort request to race a transition to the
137 * fully closed state. ABORT|DELETE messages which race the fully closed
138 * state are expected to be discarded by the other end.
142 * All base and extended message headers are 64-byte aligned, and all
143 * transports must support extended message headers up to DMSG_HDR_MAX.
144 * Currently we allow extended message headers up to 2048 bytes. Note
145 * that the extended header size is encoded in the 'cmd' field of the header.
147 * Any in-band data is padded to a 64-byte alignment and placed directly
148 * after the extended header (after the higher-level cmd/rep structure).
149 * The actual unaligned size of the in-band data is encoded in the aux_bytes
150 * field in this case. Maximum data sizes are negotiated during registration.
152 * Auxillary data can be in-band or out-of-band. In-band data sets aux_descr
153 * equal to 0. Any out-of-band data must be negotiated by the SPAN protocol.
155 * Auxillary data, whether in-band or out-of-band, must be at-least 64-byte
156 * aligned. The aux_bytes field contains the actual byte-granular length
157 * and not the aligned length.
159 * hdr_crc is calculated over the entire, ALIGNED extended header. For
160 * the purposes of calculating the crc, the hdr_crc field is 0. That is,
161 * if calculating the crc in HW a 32-bit '0' must be inserted in place of
162 * the hdr_crc field when reading the entire header and compared at the
163 * end (but the actual hdr_crc must be left intact in memory). A simple
164 * counter to replace the field going into the CRC generator does the job
165 * in HW. The CRC endian is based on the magic number field and may have
166 * to be byte-swapped, too (which is also easy to do in HW).
168 * aux_crc is calculated over the entire, ALIGNED auxillary data.
170 * SHARED MEMORY IMPLEMENTATIONS
172 * Shared-memory implementations typically use a pipe to transmit the extended
173 * message header and shared memory to store any auxilary data. Auxillary
174 * data in one-way (non-transactional) messages is typically required to be
175 * inline. CRCs are still recommended and required at the beginning, but
176 * may be negotiated away later.
179 uint16_t magic; /* 00 sanity, synchro, endian */
180 uint16_t reserved02; /* 02 */
181 uint32_t salt; /* 04 random salt helps w/crypto */
183 uint64_t msgid; /* 08 message transaction id */
184 uint64_t circuit; /* 10 circuit id or 0 */
185 uint64_t reserved18; /* 18 */
187 uint32_t cmd; /* 20 flags | cmd | hdr_size / ALIGN */
188 uint32_t aux_crc; /* 24 auxillary data crc */
189 uint32_t aux_bytes; /* 28 auxillary data length (bytes) */
190 uint32_t error; /* 2C error code or 0 */
191 uint64_t aux_descr; /* 30 negotiated OOB data descr */
192 uint32_t reserved38; /* 38 */
193 uint32_t hdr_crc; /* 3C (aligned) extended header crc */
196 typedef struct dmsg_hdr dmsg_hdr_t;
198 #define DMSG_HDR_MAGIC 0x4832
199 #define DMSG_HDR_MAGIC_REV 0x3248
200 #define DMSG_HDR_CRCOFF offsetof(dmsg_hdr_t, salt)
201 #define DMSG_HDR_CRCBYTES (sizeof(dmsg_hdr_t) - DMSG_HDR_CRCOFF)
204 * Administrative protocol limits.
206 #define DMSG_HDR_MAX 2048 /* <= 65535 */
207 #define DMSG_AUX_MAX 65536 /* <= 1MB */
208 #define DMSG_BUF_SIZE (DMSG_HDR_MAX * 4)
209 #define DMSG_BUF_MASK (DMSG_BUF_SIZE - 1)
212 * The message (cmd) field also encodes various flags and the total size
213 * of the message header. This allows the protocol processors to validate
214 * persistency and structural settings for every command simply by
215 * switch()ing on the (cmd) field.
217 #define DMSGF_CREATE 0x80000000U /* msg start */
218 #define DMSGF_DELETE 0x40000000U /* msg end */
219 #define DMSGF_REPLY 0x20000000U /* reply path */
220 #define DMSGF_ABORT 0x10000000U /* abort req */
221 #define DMSGF_AUXOOB 0x08000000U /* aux-data is OOB */
222 #define DMSGF_FLAG2 0x04000000U
223 #define DMSGF_FLAG1 0x02000000U
224 #define DMSGF_FLAG0 0x01000000U
226 #define DMSGF_FLAGS 0xFF000000U /* all flags */
227 #define DMSGF_PROTOS 0x00F00000U /* all protos */
228 #define DMSGF_CMDS 0x000FFF00U /* all cmds */
229 #define DMSGF_SIZE 0x000000FFU /* N*32 */
231 #define DMSGF_CMDSWMASK (DMSGF_CMDS | \
236 #define DMSGF_BASECMDMASK (DMSGF_CMDS | \
240 #define DMSGF_TRANSMASK (DMSGF_CMDS | \
247 #define DMSG_PROTO_LNK 0x00000000U
248 #define DMSG_PROTO_DBG 0x00100000U
249 #define DMSG_PROTO_DOM 0x00200000U
250 #define DMSG_PROTO_CAC 0x00300000U
251 #define DMSG_PROTO_QRM 0x00400000U
252 #define DMSG_PROTO_BLK 0x00500000U
253 #define DMSG_PROTO_VOP 0x00600000U
256 * Message command constructors, sans flags
258 #define DMSG_ALIGN 64
259 #define DMSG_ALIGNMASK (DMSG_ALIGN - 1)
260 #define DMSG_DOALIGN(bytes) (((bytes) + DMSG_ALIGNMASK) & \
263 #define DMSG_HDR_ENCODE(elm) (((uint32_t)sizeof(struct elm) + \
267 #define DMSG_LNK(cmd, elm) (DMSG_PROTO_LNK | \
269 DMSG_HDR_ENCODE(elm))
271 #define DMSG_DBG(cmd, elm) (DMSG_PROTO_DBG | \
273 DMSG_HDR_ENCODE(elm))
275 #define DMSG_DOM(cmd, elm) (DMSG_PROTO_DOM | \
277 DMSG_HDR_ENCODE(elm))
279 #define DMSG_CAC(cmd, elm) (DMSG_PROTO_CAC | \
281 DMSG_HDR_ENCODE(elm))
283 #define DMSG_QRM(cmd, elm) (DMSG_PROTO_QRM | \
285 DMSG_HDR_ENCODE(elm))
287 #define DMSG_BLK(cmd, elm) (DMSG_PROTO_BLK | \
289 DMSG_HDR_ENCODE(elm))
291 #define DMSG_VOP(cmd, elm) (DMSG_PROTO_VOP | \
293 DMSG_HDR_ENCODE(elm))
296 * Link layer ops basically talk to just the other side of a direct
299 * LNK_PAD - One-way message on circuit 0, ignored by target. Used to
300 * pad message buffers on shared-memory transports. Not
301 * typically used with TCP.
303 * LNK_PING - One-way message on circuit-0, keep-alive, run by both sides
304 * typically 1/sec on idle link, link is lost after 10 seconds
307 * LNK_AUTH - Authenticate the connection, negotiate administrative
308 * rights & encryption, protocol class, etc. Only PAD and
309 * AUTH messages (not even PING) are accepted until
310 * authentication is complete. This message also identifies
313 * LNK_CONN - Enable the SPAN protocol on circuit-0, possibly also
314 * installing a PFS filter (by cluster id, unique id, and/or
317 * LNK_SPAN - A SPAN transaction on circuit-0 enables messages to be
318 * relayed to/from a particular cluster node. SPANs are
319 * received, sorted, aggregated, filtered, and retransmitted
320 * back out across all applicable connections.
322 * The leaf protocol also uses this to make a PFS available
323 * to the cluster (e.g. on-mount).
325 * LNK_CIRC - a CIRC transaction establishes a circuit from source to
326 * target by creating pairs of open transactions across each
329 * LNK_VOLCONF - Volume header configuration change. All hammer2
330 * connections (hammer2 connect ...) stored in the volume
331 * header are spammed on circuit 0 to the hammer2
332 * service daemon, and any live configuration change
335 #define DMSG_LNK_PAD DMSG_LNK(0x000, dmsg_hdr)
336 #define DMSG_LNK_PING DMSG_LNK(0x001, dmsg_hdr)
337 #define DMSG_LNK_AUTH DMSG_LNK(0x010, dmsg_lnk_auth)
338 #define DMSG_LNK_CONN DMSG_LNK(0x011, dmsg_lnk_conn)
339 #define DMSG_LNK_SPAN DMSG_LNK(0x012, dmsg_lnk_span)
340 #define DMSG_LNK_CIRC DMSG_LNK(0x013, dmsg_lnk_circ)
341 #define DMSG_LNK_VOLCONF DMSG_LNK(0x020, dmsg_lnk_volconf)
342 #define DMSG_LNK_ERROR DMSG_LNK(0xFFF, dmsg_hdr)
345 * LNK_AUTH - Authentication (often omitted)
347 struct dmsg_lnk_auth {
353 * LNK_CONN - Register connection info for SPAN protocol
354 * (transaction, left open, circuit 0 only).
356 * LNK_CONN identifies a streaming connection into the cluster and serves
357 * to identify, enable, and specify filters for the SPAN protocol.
359 * peer_mask serves to filter the SPANs we receive by peer_type. A cluster
360 * controller typically sets this to (uint64_t)-1, indicating that it wants
361 * everything. A block devfs interface might set it to 1 << DMSG_PEER_DISK,
362 * and a hammer2 mount might set it to 1 << DMSG_PEER_HAMMER2.
364 * mediaid allows multiple (e.g. HAMMER2) connections belonging to the same
365 * media to transmit duplicative LNK_VOLCONF updates without causing
366 * confusion in the cluster controller.
368 * pfs_clid, pfs_fsid, pfs_type, and label are peer-specific and must be
369 * left empty (zero-fill) if not supported by a particular peer.
371 * DMSG_PEER_CLUSTER filter: none
372 * DMSG_PEER_BLOCK filter: label
373 * DMSG_PEER_HAMMER2 filter: pfs_clid if not empty, and label
375 struct dmsg_lnk_conn {
377 uuid_t mediaid; /* media configuration id */
378 uuid_t pfs_clid; /* rendezvous pfs uuid */
379 uuid_t pfs_fsid; /* unique pfs uuid */
380 uint64_t peer_mask; /* PEER mask for SPAN filtering */
381 uint8_t peer_type; /* see DMSG_PEER_xxx */
382 uint8_t pfs_type; /* pfs type */
383 uint16_t proto_version; /* high level protocol support */
384 uint32_t status; /* status flags */
385 uint32_t rnss; /* node's generated rnss */
386 uint8_t reserved02[8];
387 uint32_t reserved03[12];
388 uint64_t pfs_mask; /* PFS mask for SPAN filtering */
389 char cl_label[128]; /* cluster label (for PEER_BLOCK) */
390 char fs_label[128]; /* PFS label (for PEER_HAMMER2) */
393 typedef struct dmsg_lnk_conn dmsg_lnk_conn_t;
395 #define DMSG_PFSTYPE_NONE 0
396 #define DMSG_PFSTYPE_ADMIN 1
397 #define DMSG_PFSTYPE_CLIENT 2
398 #define DMSG_PFSTYPE_CACHE 3
399 #define DMSG_PFSTYPE_COPY 4
400 #define DMSG_PFSTYPE_SLAVE 5
401 #define DMSG_PFSTYPE_SOFT_SLAVE 6
402 #define DMSG_PFSTYPE_SOFT_MASTER 7
403 #define DMSG_PFSTYPE_MASTER 8
404 #define DMSG_PFSTYPE_SERVER 9
405 #define DMSG_PFSTYPE_MAX 10 /* 0-9 */
407 #define DMSG_PEER_NONE 0
408 #define DMSG_PEER_CLUSTER 1 /* a cluster controller */
409 #define DMSG_PEER_BLOCK 2 /* block devices */
410 #define DMSG_PEER_HAMMER2 3 /* hammer2-mounted volumes */
413 * Structures embedded in LNK_SPAN
415 struct dmsg_media_block {
416 uint64_t bytes; /* media size in bytes */
417 uint32_t blksize; /* media block size */
420 typedef struct dmsg_media_block dmsg_media_block_t;
423 * LNK_SPAN - Initiate or relay a SPAN
424 * (transaction, left open, circuit 0 only)
426 * This message registers an end-point with the other end of the connection,
427 * telling the other end who we are and what we can provide or intend to
428 * consume. Multiple registrations can be maintained as open transactions
429 * with each one specifying a unique end-point.
431 * Registrations are sent from {source}=S {1...n} to {target}=0 and maintained
432 * as open transactions. Registrations are also received and maintains as
433 * open transactions, creating a matrix of linkid's.
435 * While these transactions are open additional transactions can be executed
436 * between any two linkid's {source}=S (registrations we sent) to {target}=T
437 * (registrations we received).
439 * Closure of any registration transaction will automatically abort any open
440 * transactions using the related linkids. Closure can be initiated
441 * voluntarily from either side with either end issuing a DELETE, or they
444 * Status updates are performed via the open transaction.
448 * A registration identifies a node and its various PFS parameters including
449 * the PFS_TYPE. For example, a diskless HAMMER2 client typically identifies
450 * itself as PFSTYPE_CLIENT.
452 * Any node may serve as a cluster controller, aggregating and passing
453 * on received registrations, but end-points do not have to implement this
454 * ability. Most end-points typically implement a single client-style or
455 * server-style PFS_TYPE and rendezvous at a cluster controller.
457 * The cluster controller does not aggregate/pass-on all received
458 * registrations. It typically filters what gets passed on based on what it
459 * receives, passing on only the best candidates.
461 * If a symmetric spanning tree is desired additional candidates whos
462 * {dist, rnss} fields match the last best candidate must also be propagated.
463 * This feature is not currently enabled.
465 * STATUS UPDATES: Status updates use the same structure but typically
466 * only contain incremental changes to e.g. pfs_type, with
467 * a text description sent as out-of-band data.
469 struct dmsg_lnk_span {
471 uuid_t pfs_clid; /* rendezvous pfs uuid */
472 uuid_t pfs_fsid; /* unique pfs id (differentiate node) */
473 uint8_t pfs_type; /* PFS type */
474 uint8_t peer_type; /* PEER type */
475 uint16_t proto_version; /* high level protocol support */
476 uint32_t status; /* status flags */
477 uint8_t reserved02[8];
478 uint32_t dist; /* span distance */
479 uint32_t rnss; /* random number sub-sort */
481 uint32_t reserved03[14];
482 dmsg_media_block_t block;
486 * NOTE: for PEER_HAMMER2 cl_label is typically empty and fs_label
487 * is the superroot directory name.
489 * for PEER_BLOCK cl_label is typically host/device and
490 * fs_label is typically the serial number string.
492 char cl_label[128]; /* cluster label */
493 char fs_label[128]; /* PFS label */
496 typedef struct dmsg_lnk_span dmsg_lnk_span_t;
498 #define DMSG_SPAN_PROTO_1 1
501 * LNK_CIRC - Establish a circuit
502 * (transaction, left open, circuit 0 only)
504 * Establish a circuit to the specified target. The msgid for the open
505 * transaction is used to transit messages in both directions.
507 * For circuit establishment the receiving entity looks up the outgoing
508 * relayed SPAN on the incoming iocom based on the target field and then
509 * creates peer circuit on the interface the SPAN originally came in on.
510 * Messages received on one side or forwarded to the other side and vise-versa.
511 * Any link state loss causes all related circuits to be lost.
513 struct dmsg_lnk_circ {
519 typedef struct dmsg_lnk_circ dmsg_lnk_circ_t;
524 * All HAMMER2 directories directly under the super-root on your local
525 * media can be mounted separately, even if they share the same physical
528 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2
529 * cluster via local media. The local media does not have to participate
530 * in the cluster, other than to provide the dmsg_vol_data[] array and
531 * root inode for the mount.
533 * This is important: The mount device path you specify serves to bootstrap
534 * your entry into the cluster, but your mount will make active connections
535 * to ALL copy elements in the dmsg_vol_data[] array which match the
536 * PFSID of the directory in the super-root that you specified. The local
537 * media path does not have to be mentioned in this array but becomes part
538 * of the cluster based on its type and access rights. ALL ELEMENTS ARE
539 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM.
541 * The actual cluster may be far larger than the elements you list in the
542 * dmsg_vol_data[] array. You list only the elements you wish to
543 * directly connect to and you are able to access the rest of the cluster
544 * indirectly through those connections.
546 * This structure must be exactly 128 bytes long.
548 * WARNING! dmsg_vol_data is embedded in the hammer2 media volume header
550 struct dmsg_vol_data {
551 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */
552 uint8_t inprog; /* 01 operation in progress, or 0 */
553 uint8_t chain_to; /* 02 operation chaining to, or 0 */
554 uint8_t chain_from; /* 03 operation chaining from, or 0 */
555 uint16_t flags; /* 04-05 flags field */
556 uint8_t error; /* 06 last operational error */
557 uint8_t priority; /* 07 priority and round-robin flag */
558 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */
559 uint8_t reserved08[23]; /* 09-1F */
560 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */
561 uint8_t label[16]; /* 30-3F import/export label */
562 uint8_t path[64]; /* 40-7F target specification string or key */
565 typedef struct dmsg_vol_data dmsg_vol_data_t;
567 #define DMSG_VOLF_ENABLED 0x0001
568 #define DMSG_VOLF_INPROG 0x0002
569 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */
570 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */
571 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */
573 #define DMSG_COPYID_COUNT 256 /* WARNING! embedded in hammer2 vol */
575 struct dmsg_lnk_volconf {
577 dmsg_vol_data_t copy; /* copy spec */
581 int64_t reserved02[32];
584 typedef struct dmsg_lnk_volconf dmsg_lnk_volconf_t;
587 * Debug layer ops operate on any link
589 * SHELL - Persist stream, access the debug shell on the target
590 * registration. Multiple shells can be operational.
592 #define DMSG_DBG_SHELL DMSG_DBG(0x001, dmsg_dbg_shell)
594 struct dmsg_dbg_shell {
597 typedef struct dmsg_dbg_shell dmsg_dbg_shell_t;
600 * Domain layer ops operate on any link, link-0 may be used when the
601 * directory connected target is the desired registration.
607 * Cache layer ops operate on any link, link-0 may be used when the
608 * directly connected target is the desired registration.
610 * LOCK - Persist state, blockable, abortable.
612 * Obtain cache state (MODIFIED, EXCLUSIVE, SHARED, or INVAL)
613 * in any of three domains (TREE, INUM, ATTR, DIRENT) for a
614 * particular key relative to cache state already owned.
616 * TREE - Effects entire sub-tree at the specified element
617 * and will cause existing cache state owned by
618 * other nodes to be adjusted such that the request
621 * INUM - Only effects inode creation/deletion of an existing
622 * element or a new element, by inumber and/or name.
623 * typically can be held for very long periods of time
624 * (think the vnode cache), directly relates to
625 * hammer2_chain structures representing inodes.
627 * ATTR - Only effects an inode's attributes, such as
628 * ownership, modes, etc. Used for lookups, chdir,
629 * open, etc. mtime has no affect.
631 * DIRENT - Only affects an inode's attributes plus the
632 * attributes or names related to any directory entry
633 * directly under this inode (non-recursively). Can
634 * be retained for medium periods of time when doing
637 * This function may block and can be aborted. You may be
638 * granted cache state that is more broad than the state you
639 * requested (e.g. a different set of domains and/or an element
640 * at a higher layer in the tree). When quorum operations
641 * are used you may have to reconcile these grants to the
642 * lowest common denominator.
644 * In order to grant your request either you or the target
645 * (or both) may have to obtain a quorum agreement. Deadlock
646 * resolution may be required. When doing it yourself you
647 * will typically maintain an active message to each master
648 * node in the system. You can only grant the cache state
649 * when a quorum of nodes agree.
651 * The cache state includes transaction id information which
652 * can be used to resolve data requests.
654 #define DMSG_CAC_LOCK DMSG_CAC(0x001, dmsg_cac_lock)
657 * Quorum layer ops operate on any link, link-0 may be used when the
658 * directly connected target is the desired registration.
660 * COMMIT - Persist state, blockable, abortable
662 * Issue a COMMIT in two phases. A quorum must acknowledge
663 * the operation to proceed to phase-2. Message-update to
664 * proceed to phase-2.
666 #define DMSG_QRM_COMMIT DMSG_QRM(0x001, dmsg_qrm_commit)
669 * DMSG_PROTO_BLK Protocol
671 * BLK_OPEN - Open device. This transaction must be left open for the
672 * duration and the returned keyid passed in all associated
675 * BLK_READ - Strategy read
677 * BLK_WRITE - Strategy write
679 * BLK_FLUSH - Strategy flush
681 #define DMSG_BLK_OPEN DMSG_BLK(0x001, dmsg_blk_open)
682 #define DMSG_BLK_READ DMSG_BLK(0x002, dmsg_blk_read)
683 #define DMSG_BLK_WRITE DMSG_BLK(0x003, dmsg_blk_write)
684 #define DMSG_BLK_FLUSH DMSG_BLK(0x004, dmsg_blk_flush)
685 #define DMSG_BLK_FREEBLKS DMSG_BLK(0x005, dmsg_blk_freeblks)
686 #define DMSG_BLK_ERROR DMSG_BLK(0xFFF, dmsg_blk_error)
688 struct dmsg_blk_open {
694 #define DMSG_BLKOPEN_RD 0x0001
695 #define DMSG_BLKOPEN_WR 0x0002
698 * DMSG_LNK_ERROR is returned for simple results,
699 * DMSG_BLK_ERROR is returned for extended results.
701 struct dmsg_blk_error {
709 struct dmsg_blk_read {
719 struct dmsg_blk_write {
729 struct dmsg_blk_flush {
739 struct dmsg_blk_freeblks {
749 typedef struct dmsg_blk_open dmsg_blk_open_t;
750 typedef struct dmsg_blk_read dmsg_blk_read_t;
751 typedef struct dmsg_blk_write dmsg_blk_write_t;
752 typedef struct dmsg_blk_flush dmsg_blk_flush_t;
753 typedef struct dmsg_blk_freeblks dmsg_blk_freeblks_t;
754 typedef struct dmsg_blk_error dmsg_blk_error_t;
757 * NOTE!!!! ALL EXTENDED HEADER STRUCTURES MUST BE 64-BYTE ALIGNED!!!
759 * General message errors
761 * 0x00 - 0x1F Local iocomm errors
762 * 0x20 - 0x2F Global errors
764 #define DMSG_ERR_NOSUPP 0x20
765 #define DMSG_ERR_LOSTLINK 0x21
768 char buf[DMSG_HDR_MAX];
771 dmsg_lnk_conn_t lnk_conn;
772 dmsg_lnk_span_t lnk_span;
773 dmsg_lnk_circ_t lnk_circ;
774 dmsg_lnk_volconf_t lnk_volconf;
776 dmsg_blk_open_t blk_open;
777 dmsg_blk_error_t blk_error;
778 dmsg_blk_read_t blk_read;
779 dmsg_blk_write_t blk_write;
780 dmsg_blk_flush_t blk_flush;
781 dmsg_blk_freeblks_t blk_freeblks;
784 typedef union dmsg_any dmsg_any_t;
787 * Kernel iocom structures and prototypes for kern/kern_dmsg.c
789 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
791 struct hammer2_pfsmount;
797 * msg_ctl flags (atomic)
799 #define KDMSG_CLUSTERCTL_KILL 0x00000001
800 #define KDMSG_CLUSTERCTL_KILLRX 0x00000002 /* staged helper exit */
801 #define KDMSG_CLUSTERCTL_KILLTX 0x00000004 /* staged helper exit */
802 #define KDMSG_CLUSTERCTL_SLEEPING 0x00000008 /* interlocked w/msglk */
805 * When the KDMSG_IOCOMF_AUTOCIRC flag is set the kdmsg code in
806 * the kernel automatically tries to forge a virtual circuit for
807 * any active SPAN state received.
809 * This is only done when the received SPANs are significantly filtered
810 * by the transmitted LNK_CONN. That is, it is done only by clients who
811 * connect to specific services over the cluster.
813 struct kdmsg_circuit {
814 TAILQ_ENTRY(kdmsg_circuit) entry;
815 struct kdmsg_iocom *iocom;
816 struct kdmsg_state *span_state;
817 struct kdmsg_state *circ_state;
818 int recorded; /* used by shim */
822 typedef struct kdmsg_circuit kdmsg_circuit_t;
825 * Transactional state structure, representing an open transaction. The
826 * transaction might represent a cache state (and thus have a chain
827 * association), or a VOP op, LNK_SPAN, or other things.
830 RB_ENTRY(kdmsg_state) rbnode; /* indexed by msgid */
831 struct kdmsg_iocom *iocom;
833 uint32_t icmd; /* record cmd creating state */
834 uint32_t txcmd; /* mostly for CMDF flags */
835 uint32_t rxcmd; /* mostly for CMDF flags */
836 uint64_t msgid; /* {circuit,msgid} uniq */
839 void *chain; /* (caller's state) */
840 struct kdmsg_msg *msg;
841 int (*func)(struct kdmsg_state *, struct kdmsg_msg *);
844 struct hammer2_pfsmount *pmp;
845 struct kdmsg_circuit *circ;
849 #define KDMSG_STATE_INSERTED 0x0001
850 #define KDMSG_STATE_DYNAMIC 0x0002
851 #define KDMSG_STATE_DELPEND 0x0004 /* transmit delete pending */
854 TAILQ_ENTRY(kdmsg_msg) qentry; /* serialized queue */
855 struct kdmsg_iocom *iocom;
856 struct kdmsg_state *state;
864 #define KDMSG_FLAG_AUXALLOC 0x0001
866 typedef struct kdmsg_link kdmsg_link_t;
867 typedef struct kdmsg_state kdmsg_state_t;
868 typedef struct kdmsg_msg kdmsg_msg_t;
870 struct kdmsg_state_tree;
871 RB_HEAD(kdmsg_state_tree, kdmsg_state);
872 int kdmsg_state_cmp(kdmsg_state_t *state1, kdmsg_state_t *state2);
873 RB_PROTOTYPE(kdmsg_state_tree, kdmsg_state, rbnode, kdmsg_state_cmp);
876 * Structure embedded in e.g. mount, master control structure for
877 * DMSG stream handling.
880 struct malloc_type *mmsg;
881 struct file *msg_fp; /* cluster pipe->userland */
882 thread_t msgrd_td; /* cluster thread */
883 thread_t msgwr_td; /* cluster thread */
884 int msg_ctl; /* wakeup flags */
885 int msg_seq; /* cluster msg sequence id */
887 struct lock msglk; /* lockmgr lock */
888 TAILQ_HEAD(, kdmsg_msg) msgq; /* transmit queue */
890 void (*auto_callback)(kdmsg_msg_t *);
891 int (*rcvmsg)(kdmsg_msg_t *);
892 void (*exit_func)(struct kdmsg_iocom *);
893 struct kdmsg_state *conn_state; /* active LNK_CONN state */
894 struct kdmsg_state *freerd_state; /* allocation cache */
895 struct kdmsg_state *freewr_state; /* allocation cache */
896 struct kdmsg_state_tree staterd_tree; /* active messages */
897 struct kdmsg_state_tree statewr_tree; /* active messages */
898 dmsg_lnk_conn_t auto_lnk_conn;
899 dmsg_lnk_span_t auto_lnk_span;
902 typedef struct kdmsg_iocom kdmsg_iocom_t;
904 #define KDMSG_IOCOMF_AUTOCONN 0x0001 /* handle received LNK_CONN */
905 #define KDMSG_IOCOMF_AUTOSPAN 0x0002 /* handle received LNK_SPAN */
906 #define KDMSG_IOCOMF_AUTOCIRC 0x0004 /* handle received LNK_CIRC */
907 #define KDMSG_IOCOMF_AUTOFORGE 0x0008 /* auto initiate LNK_CIRC */
909 #define KDMSG_IOCOMF_AUTOANY (KDMSG_IOCOMF_AUTOCONN | \
910 KDMSG_IOCOMF_AUTOSPAN | \
911 KDMSG_IOCOMF_AUTOCIRC)
913 uint32_t kdmsg_icrc32(const void *buf, size_t size);
914 uint32_t kdmsg_icrc32c(const void *buf, size_t size, uint32_t crc);
919 void kdmsg_iocom_init(kdmsg_iocom_t *iocom, void *handle, u_int32_t flags,
920 struct malloc_type *mmsg,
921 int (*rcvmsg)(kdmsg_msg_t *msg));
922 void kdmsg_iocom_reconnect(kdmsg_iocom_t *iocom, struct file *fp,
923 const char *subsysname);
924 void kdmsg_iocom_autoinitiate(kdmsg_iocom_t *iocom,
925 void (*conn_callback)(kdmsg_msg_t *msg));
926 void kdmsg_iocom_uninit(kdmsg_iocom_t *iocom);
927 void kdmsg_drain_msgq(kdmsg_iocom_t *iocom);
929 int kdmsg_state_msgrx(kdmsg_msg_t *msg);
930 int kdmsg_state_msgtx(kdmsg_msg_t *msg);
931 void kdmsg_state_cleanuprx(kdmsg_msg_t *msg);
932 void kdmsg_state_cleanuptx(kdmsg_msg_t *msg);
933 int kdmsg_msg_execute(kdmsg_msg_t *msg);
934 void kdmsg_state_free(kdmsg_state_t *state);
935 void kdmsg_msg_free(kdmsg_msg_t *msg);
936 kdmsg_msg_t *kdmsg_msg_alloc(kdmsg_iocom_t *iocom, uint64_t circuit,
938 int (*func)(kdmsg_state_t *, kdmsg_msg_t *),
940 void kdmsg_msg_write(kdmsg_msg_t *msg);
941 void kdmsg_msg_reply(kdmsg_msg_t *msg, uint32_t error);
942 void kdmsg_msg_result(kdmsg_msg_t *msg, uint32_t error);
943 void kdmsg_state_reply(kdmsg_state_t *state, uint32_t error);
944 void kdmsg_state_result(kdmsg_state_t *state, uint32_t error);