2 * Copyright (c) 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
35 * This module allows disk devices to be created and associated with a
36 * communications pipe or socket. You open the device and issue an
37 * ioctl() to install a new disk along with its communications descriptor.
39 * All further communication occurs via the descriptor using the DMSG
40 * LNK_CONN, LNK_SPAN, and BLOCK protocols. The descriptor can be a
41 * direct connection to a remote machine's disk (in-kernenl), to a remote
42 * cluster controller, to the local cluster controller, etc.
44 * /dev/xdisk is the control device, issue ioctl()s to create the /dev/xa%d
45 * devices. These devices look like raw disks to the system.
48 * Handle circuit disconnects, leave bio's pending
49 * Restart bio's on circuit reconnect.
51 #include <sys/param.h>
52 #include <sys/systm.h>
55 #include <sys/device.h>
56 #include <sys/devicestat.h>
58 #include <sys/kernel.h>
59 #include <sys/malloc.h>
60 #include <sys/sysctl.h>
62 #include <sys/queue.h>
65 #include <sys/kern_syscall.h>
68 #include <sys/xdiskioctl.h>
71 #include <sys/thread2.h>
76 TAILQ_ENTRY(xa_tag) entry;
78 dmsg_blk_error_t status;
80 kdmsg_circuit_t *circ;
82 int running; /* transaction running */
83 int waitseq; /* streaming reply */
84 int done; /* final (transaction closed) */
87 typedef struct xa_tag xa_tag_t;
90 TAILQ_ENTRY(xa_softc) entry;
93 struct xdisk_attach_ioctl xaioc;
94 struct disk_info info;
103 TAILQ_HEAD(, bio) bioq;
104 TAILQ_HEAD(, xa_tag) tag_freeq;
105 TAILQ_HEAD(, xa_tag) tag_pendq;
106 TAILQ_HEAD(, kdmsg_circuit) circq;
107 struct lwkt_token tok;
110 typedef struct xa_softc xa_softc_t;
112 #define MAXTAGS 64 /* no real limit */
114 static int xdisk_attach(struct xdisk_attach_ioctl *xaioc);
115 static int xdisk_detach(struct xdisk_attach_ioctl *xaioc);
116 static void xa_exit(kdmsg_iocom_t *iocom);
117 static void xa_terminate_check(struct xa_softc *xa);
118 static int xa_rcvdmsg(kdmsg_msg_t *msg);
119 static void xa_autodmsg(kdmsg_msg_t *msg);
121 static xa_tag_t *xa_setup_cmd(xa_softc_t *xa, struct bio *bio);
122 static void xa_start(xa_tag_t *tag, kdmsg_msg_t *msg);
123 static uint32_t xa_wait(xa_tag_t *tag, int seq);
124 static void xa_done(xa_tag_t *tag, int wasbio);
125 static int xa_sync_completion(kdmsg_state_t *state, kdmsg_msg_t *msg);
126 static int xa_bio_completion(kdmsg_state_t *state, kdmsg_msg_t *msg);
127 static void xa_restart_deferred(xa_softc_t *xa);
129 MALLOC_DEFINE(M_XDISK, "Networked disk client", "Network Disks");
132 * Control device, issue ioctls to create xa devices.
134 static d_open_t xdisk_open;
135 static d_close_t xdisk_close;
136 static d_ioctl_t xdisk_ioctl;
138 static struct dev_ops xdisk_ops = {
139 { "xdisk", 0, D_MPSAFE | D_TRACKCLOSE },
140 .d_open = xdisk_open,
141 .d_close = xdisk_close,
142 .d_ioctl = xdisk_ioctl
148 static d_open_t xa_open;
149 static d_close_t xa_close;
150 static d_ioctl_t xa_ioctl;
151 static d_strategy_t xa_strategy;
152 static d_psize_t xa_size;
154 static struct dev_ops xa_ops = {
155 { "xa", 0, D_DISK | D_CANFREE | D_MPSAFE | D_TRACKCLOSE },
160 .d_write = physwrite,
161 .d_strategy = xa_strategy,
165 static struct lwkt_token xdisk_token = LWKT_TOKEN_INITIALIZER(xdisk_token);
166 static int xdisk_opencount;
167 static cdev_t xdisk_dev;
168 static TAILQ_HEAD(, xa_softc) xa_queue;
171 * Module initialization
174 xdisk_modevent(module_t mod, int type, void *data)
178 TAILQ_INIT(&xa_queue);
179 xdisk_dev = make_dev(&xdisk_ops, 0,
180 UID_ROOT, GID_WHEEL, 0600, "xdisk");
184 if (xdisk_opencount || TAILQ_FIRST(&xa_queue))
187 destroy_dev(xdisk_dev);
190 dev_ops_remove_all(&xdisk_ops);
191 dev_ops_remove_all(&xa_ops);
199 DEV_MODULE(xdisk, xdisk_modevent, 0);
205 xdisk_open(struct dev_open_args *ap)
207 lwkt_gettoken(&xdisk_token);
209 lwkt_reltoken(&xdisk_token);
214 xdisk_close(struct dev_close_args *ap)
216 lwkt_gettoken(&xdisk_token);
218 lwkt_reltoken(&xdisk_token);
223 xdisk_ioctl(struct dev_ioctl_args *ap)
229 error = xdisk_attach((void *)ap->a_data);
232 error = xdisk_detach((void *)ap->a_data);
241 /************************************************************************
243 ************************************************************************/
246 xdisk_attach(struct xdisk_attach_ioctl *xaioc)
257 * Normalize ioctl params
259 fp = holdfp(curproc->p_fd, xaioc->fd, -1);
262 if (xaioc->cl_label[sizeof(xaioc->cl_label) - 1] != 0)
264 if (xaioc->fs_label[sizeof(xaioc->fs_label) - 1] != 0)
266 if (xaioc->blksize < DEV_BSIZE || xaioc->blksize > MAXBSIZE)
270 * See if the serial number is already present. If we are
271 * racing a termination the disk subsystem may still have
272 * duplicate entries not yet removed so we wait a bit and
275 lwkt_gettoken(&xdisk_token);
277 TAILQ_FOREACH(xa, &xa_queue, entry) {
278 if (strcmp(xa->iocom.auto_lnk_conn.fs_label,
279 xaioc->fs_label) == 0) {
280 if (xa->serializing) {
281 tsleep(xa, 0, "xadelay", hz / 10);
285 kdmsg_iocom_uninit(&xa->iocom);
291 * Create a new xa if not already present
296 TAILQ_FOREACH(xa, &xa_queue, entry) {
297 if (xa->unit == unit)
304 xa = kmalloc(sizeof(*xa), M_XDISK, M_WAITOK|M_ZERO);
307 lwkt_token_init(&xa->tok, "xa");
308 TAILQ_INIT(&xa->circq);
309 TAILQ_INIT(&xa->bioq);
310 TAILQ_INIT(&xa->tag_freeq);
311 TAILQ_INIT(&xa->tag_pendq);
312 for (n = 0; n < MAXTAGS; ++n) {
313 tag = kmalloc(sizeof(*tag), M_XDISK, M_WAITOK|M_ZERO);
315 TAILQ_INSERT_TAIL(&xa->tag_freeq, tag, entry);
317 TAILQ_INSERT_TAIL(&xa_queue, xa, entry);
323 * (xa) is now serializing.
327 lwkt_reltoken(&xdisk_token);
332 if (xa->dev == NULL) {
333 dev = disk_create(unit, &xa->disk, &xa_ops);
338 xa->info.d_media_blksize = xaioc->blksize;
339 xa->info.d_media_blocks = xaioc->bytes / xaioc->blksize;
340 xa->info.d_dsflags = DSO_MBRQUIET | DSO_RAWPSIZE;
341 xa->info.d_secpertrack = 32;
342 xa->info.d_nheads = 64;
343 xa->info.d_secpercyl = xa->info.d_secpertrack * xa->info.d_nheads;
344 xa->info.d_ncylinders = 0;
345 if (xa->xaioc.fs_label[0])
346 xa->info.d_serialno = xa->xaioc.fs_label;
349 * Set up messaging connection
351 ksnprintf(devname, sizeof(devname), "xa%d", unit);
352 kdmsg_iocom_init(&xa->iocom, xa,
353 KDMSG_IOCOMF_AUTOCONN |
354 KDMSG_IOCOMF_AUTOSPAN |
355 KDMSG_IOCOMF_AUTOCIRC |
356 KDMSG_IOCOMF_AUTOFORGE,
357 M_XDISK, xa_rcvdmsg);
358 xa->iocom.exit_func = xa_exit;
360 kdmsg_iocom_reconnect(&xa->iocom, fp, devname);
363 * Setup our LNK_CONN advertisement for autoinitiate.
365 * Our filter is setup to only accept PEER_BLOCK/SERVER
368 xa->iocom.auto_lnk_conn.pfs_type = DMSG_PFSTYPE_CLIENT;
369 xa->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
370 xa->iocom.auto_lnk_conn.peer_type = DMSG_PEER_BLOCK;
371 xa->iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_BLOCK;
372 xa->iocom.auto_lnk_conn.pfs_mask = 1LLU << DMSG_PFSTYPE_SERVER;
373 ksnprintf(xa->iocom.auto_lnk_conn.cl_label,
374 sizeof(xa->iocom.auto_lnk_conn.cl_label),
375 "%s", xaioc->cl_label);
378 * We need a unique pfs_fsid to avoid confusion.
379 * We supply a rendezvous fs_label using the serial number.
381 kern_uuidgen(&xa->pfs_fsid, 1);
382 xa->iocom.auto_lnk_conn.pfs_fsid = xa->pfs_fsid;
383 ksnprintf(xa->iocom.auto_lnk_conn.fs_label,
384 sizeof(xa->iocom.auto_lnk_conn.fs_label),
385 "%s", xaioc->fs_label);
388 * Setup our LNK_SPAN advertisement for autoinitiate
390 xa->iocom.auto_lnk_span.pfs_type = DMSG_PFSTYPE_CLIENT;
391 xa->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
392 xa->iocom.auto_lnk_span.peer_type = DMSG_PEER_BLOCK;
393 ksnprintf(xa->iocom.auto_lnk_span.cl_label,
394 sizeof(xa->iocom.auto_lnk_span.cl_label),
395 "%s", xa->xaioc.cl_label);
397 kdmsg_iocom_autoinitiate(&xa->iocom, xa_autodmsg);
398 disk_setdiskinfo_sync(&xa->disk, &xa->info);
400 lwkt_gettoken(&xdisk_token);
402 xa_terminate_check(xa);
403 lwkt_reltoken(&xdisk_token);
409 xdisk_detach(struct xdisk_attach_ioctl *xaioc)
413 lwkt_gettoken(&xdisk_token);
415 TAILQ_FOREACH(xa, &xa_queue, entry) {
416 if (strcmp(xa->iocom.auto_lnk_conn.fs_label,
417 xaioc->fs_label) == 0) {
421 if (xa == NULL || xa->serializing == 0) {
425 tsleep(xa, 0, "xadet", hz / 10);
428 kdmsg_iocom_uninit(&xa->iocom);
431 lwkt_reltoken(&xdisk_token);
436 * Called from iocom core transmit thread upon disconnect.
440 xa_exit(kdmsg_iocom_t *iocom)
442 struct xa_softc *xa = iocom->handle;
444 lwkt_gettoken(&xa->tok);
445 lwkt_gettoken(&xdisk_token);
448 * We must wait for any I/O's to complete to ensure that all
449 * state structure references are cleaned up before returning.
451 xa->attached = -1; /* force deferral or failure */
452 while (TAILQ_FIRST(&xa->tag_pendq)) {
453 tsleep(xa, 0, "xabiow", hz / 10);
457 * All serializing code checks for de-initialization so only
458 * do it if we aren't already serializing.
460 if (xa->serializing == 0) {
462 kdmsg_iocom_uninit(iocom);
467 * If the drive is not in use and no longer attach it can be
471 xa_terminate_check(xa);
472 lwkt_reltoken(&xdisk_token);
473 lwkt_reltoken(&xa->tok);
477 * Determine if we can destroy the xa_softc.
479 * Called with xdisk_token held.
483 xa_terminate_check(struct xa_softc *xa)
488 if (xa->opencnt || xa->attached || xa->serializing)
491 kdmsg_iocom_uninit(&xa->iocom);
494 * When destroying an xa make sure all pending I/O (typically
495 * from the disk probe) is done.
497 * XXX what about new I/O initiated prior to disk_destroy().
499 while ((tag = TAILQ_FIRST(&xa->tag_pendq)) != NULL) {
500 TAILQ_REMOVE(&xa->tag_pendq, tag, entry);
501 if ((bio = tag->bio) != NULL) {
503 bio->bio_buf->b_error = ENXIO;
504 bio->bio_buf->b_flags |= B_ERROR;
507 TAILQ_INSERT_TAIL(&xa->tag_freeq, tag, entry);
510 disk_destroy(&xa->disk);
511 xa->dev->si_drv1 = NULL;
514 KKASSERT(xa->opencnt == 0 && xa->attached == 0);
515 while ((tag = TAILQ_FIRST(&xa->tag_freeq)) != NULL) {
516 TAILQ_REMOVE(&xa->tag_freeq, tag, entry);
520 KKASSERT(TAILQ_EMPTY(&xa->tag_pendq));
521 TAILQ_REMOVE(&xa_queue, xa, entry); /* XXX */
526 * Shim to catch and record virtual circuit events.
529 xa_autodmsg(kdmsg_msg_t *msg)
531 xa_softc_t *xa = msg->iocom->handle;
533 kdmsg_circuit_t *circ;
534 kdmsg_circuit_t *cscan;
538 * Because this is just a shim we don't have a state callback for
539 * the transactions we are sniffing, so make things easier by
540 * calculating the original command along with the current message's
541 * flags. This is because transactions are made up of numerous
542 * messages and only the first typically specifies the actual command.
545 xcmd = msg->state->icmd |
546 (msg->any.head.cmd & (DMSGF_CREATE |
550 xcmd = msg->any.head.cmd;
554 * Add or remove a circuit, sorted by weight (lower numbers are
558 case DMSG_LNK_CIRC | DMSGF_CREATE | DMSGF_REPLY:
560 * Track established circuits
562 circ = msg->state->any.circ;
563 lwkt_gettoken(&xa->tok);
564 if (circ->recorded == 0) {
565 TAILQ_FOREACH(cscan, &xa->circq, entry) {
566 if (circ->weight < cscan->weight)
570 TAILQ_INSERT_BEFORE(cscan, circ, entry);
572 TAILQ_INSERT_TAIL(&xa->circq, circ, entry);
577 * Restart any deferred I/O.
579 xa_restart_deferred(xa);
580 lwkt_reltoken(&xa->tok);
582 case DMSG_LNK_CIRC | DMSGF_DELETE | DMSGF_REPLY:
584 * Losing virtual circuit. Remove the circ from contention.
586 circ = msg->state->any.circ;
587 lwkt_gettoken(&xa->tok);
588 if (circ->recorded) {
589 TAILQ_REMOVE(&xa->circq, circ, entry);
592 xa_restart_deferred(xa);
593 lwkt_reltoken(&xa->tok);
601 xa_rcvdmsg(kdmsg_msg_t *msg)
603 switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
606 * Execute shell command (not supported atm).
608 * This is a one-way packet but if not (e.g. if part of
609 * a streaming transaction), we will have already closed
612 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
614 case DMSG_DBG_SHELL | DMSGF_REPLY:
616 * Receive one or more replies to a shell command that we
619 * This is a one-way packet but if not (e.g. if part of
620 * a streaming transaction), we will have already closed
624 msg->aux_data[msg->aux_size - 1] = 0;
625 kprintf("xdisk: DEBUGMSG: %s\n", msg->aux_data);
630 * Unsupported LNK message received. We only need to
631 * reply if it's a transaction in order to close our end.
632 * Ignore any one-way messages are any further messages
633 * associated with the transaction.
635 * NOTE: This case also includes DMSG_LNK_ERROR messages
636 * which might be one-way, replying to those would
637 * cause an infinite ping-pong.
639 if (msg->any.head.cmd & DMSGF_CREATE)
640 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
647 /************************************************************************
648 * XA DEVICE INTERFACE *
649 ************************************************************************/
652 xa_open(struct dev_open_args *ap)
654 cdev_t dev = ap->a_head.a_dev;
660 dev->si_bsize_phys = 512;
661 dev->si_bsize_best = 32768;
664 * Interlock open with opencnt, wait for attachment operations
667 lwkt_gettoken(&xdisk_token);
671 lwkt_reltoken(&xdisk_token);
672 return ENXIO; /* raced destruction */
674 if (xa->serializing) {
675 tsleep(xa, 0, "xarace", hz / 10);
678 if (xa->attached == 0) {
679 lwkt_reltoken(&xdisk_token);
680 return ENXIO; /* raced destruction */
684 * Serialize initial open
686 if (xa->opencnt++ > 0) {
687 lwkt_reltoken(&xdisk_token);
691 lwkt_reltoken(&xdisk_token);
693 tag = xa_setup_cmd(xa, NULL);
695 lwkt_gettoken(&xdisk_token);
696 KKASSERT(xa->opencnt > 0);
699 xa_terminate_check(xa);
700 lwkt_reltoken(&xdisk_token);
703 msg = kdmsg_msg_alloc(&xa->iocom, tag->circ,
704 DMSG_BLK_OPEN | DMSGF_CREATE,
705 xa_sync_completion, tag);
706 msg->any.blk_open.modes = DMSG_BLKOPEN_RD | DMSG_BLKOPEN_WR;
708 if (xa_wait(tag, 0) == 0) {
709 xa->keyid = tag->status.keyid;
710 xa->opentag = tag; /* leave tag open */
715 lwkt_gettoken(&xdisk_token);
716 KKASSERT(xa->opencnt > 0);
719 xa_terminate_check(xa);
720 lwkt_reltoken(&xdisk_token);
727 xa_close(struct dev_close_args *ap)
729 cdev_t dev = ap->a_head.a_dev;
735 return ENXIO; /* raced destruction */
737 lwkt_gettoken(&xa->tok);
738 if ((tag = xa->opentag) != NULL) {
740 kdmsg_state_reply(tag->state, 0);
741 while (tag->done == 0)
742 xa_wait(tag, tag->waitseq);
745 lwkt_reltoken(&xa->tok);
747 lwkt_gettoken(&xdisk_token);
748 KKASSERT(xa->opencnt > 0);
750 xa_terminate_check(xa);
751 lwkt_reltoken(&xdisk_token);
757 xa_strategy(struct dev_strategy_args *ap)
759 xa_softc_t *xa = ap->a_head.a_dev->si_drv1;
761 struct bio *bio = ap->a_bio;
764 * Allow potentially temporary link failures to fail the I/Os
765 * only if the device is not open. That is, we allow the disk
766 * probe code prior to mount to fail.
768 if (xa->attached == 0 && xa->opencnt == 0) {
769 bio->bio_buf->b_error = ENXIO;
770 bio->bio_buf->b_flags |= B_ERROR;
775 tag = xa_setup_cmd(xa, bio);
782 xa_ioctl(struct dev_ioctl_args *ap)
788 xa_size(struct dev_psize_args *ap)
792 if ((xa = ap->a_head.a_dev->si_drv1) == NULL)
794 ap->a_result = xa->info.d_media_blocks;
798 /************************************************************************
799 * XA BLOCK PROTOCOL STATE MACHINE *
800 ************************************************************************
802 * Implement tag/msg setup and related functions.
805 xa_setup_cmd(xa_softc_t *xa, struct bio *bio)
807 kdmsg_circuit_t *circ;
811 * Only get a tag if we have a valid virtual circuit to the server.
813 lwkt_gettoken(&xa->tok);
814 TAILQ_FOREACH(circ, &xa->circq, entry) {
818 if (circ == NULL || xa->attached <= 0) {
820 } else if ((tag = TAILQ_FIRST(&xa->tag_freeq)) != NULL) {
821 TAILQ_REMOVE(&xa->tag_freeq, tag, entry);
824 kdmsg_circ_hold(circ);
825 TAILQ_INSERT_TAIL(&xa->tag_pendq, tag, entry);
829 * If we can't dispatch now and this is a bio, queue it for later.
831 if (tag == NULL && bio) {
832 TAILQ_INSERT_TAIL(&xa->bioq, bio, bio_act);
834 lwkt_reltoken(&xa->tok);
840 xa_start(xa_tag_t *tag, kdmsg_msg_t *msg)
842 xa_softc_t *xa = tag->xa;
854 msg = kdmsg_msg_alloc(&xa->iocom, tag->circ,
856 DMSGF_CREATE | DMSGF_DELETE,
857 xa_bio_completion, tag);
858 msg->any.blk_read.keyid = xa->keyid;
859 msg->any.blk_read.offset = bio->bio_offset;
860 msg->any.blk_read.bytes = bp->b_bcount;
863 msg = kdmsg_msg_alloc(&xa->iocom, tag->circ,
865 DMSGF_CREATE | DMSGF_DELETE,
866 xa_bio_completion, tag);
867 msg->any.blk_write.keyid = xa->keyid;
868 msg->any.blk_write.offset = bio->bio_offset;
869 msg->any.blk_write.bytes = bp->b_bcount;
870 msg->aux_data = bp->b_data;
871 msg->aux_size = bp->b_bcount;
874 msg = kdmsg_msg_alloc(&xa->iocom, tag->circ,
876 DMSGF_CREATE | DMSGF_DELETE,
877 xa_bio_completion, tag);
878 msg->any.blk_flush.keyid = xa->keyid;
879 msg->any.blk_flush.offset = bio->bio_offset;
880 msg->any.blk_flush.bytes = bp->b_bcount;
882 case BUF_CMD_FREEBLKS:
883 msg = kdmsg_msg_alloc(&xa->iocom, tag->circ,
885 DMSGF_CREATE | DMSGF_DELETE,
886 xa_bio_completion, tag);
887 msg->any.blk_freeblks.keyid = xa->keyid;
888 msg->any.blk_freeblks.offset = bio->bio_offset;
889 msg->any.blk_freeblks.bytes = bp->b_bcount;
892 bp->b_flags |= B_ERROR;
903 tag->state = msg->state;
904 kdmsg_msg_write(msg);
911 xa_wait(xa_tag_t *tag, int seq)
913 xa_softc_t *xa = tag->xa;
915 lwkt_gettoken(&xa->tok);
916 while (tag->waitseq == seq)
917 tsleep(tag, 0, "xawait", 0);
918 lwkt_reltoken(&xa->tok);
919 return (tag->status.head.error);
923 xa_done(xa_tag_t *tag, int wasbio)
925 xa_softc_t *xa = tag->xa;
928 KKASSERT(tag->bio == NULL);
932 lwkt_gettoken(&xa->tok);
933 if (wasbio && (bio = TAILQ_FIRST(&xa->bioq)) != NULL) {
934 TAILQ_REMOVE(&xa->bioq, bio, bio_act);
936 lwkt_reltoken(&xa->tok);
940 kdmsg_circ_drop(tag->circ);
943 TAILQ_REMOVE(&xa->tag_pendq, tag, entry);
944 TAILQ_INSERT_TAIL(&xa->tag_freeq, tag, entry);
945 lwkt_reltoken(&xa->tok);
950 xa_sync_completion(kdmsg_state_t *state, kdmsg_msg_t *msg)
952 xa_tag_t *tag = state->any.any;
953 xa_softc_t *xa = tag->xa;
955 switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
956 case DMSG_LNK_ERROR | DMSGF_REPLY:
957 bzero(&tag->status, sizeof(tag->status));
958 tag->status.head = msg->any.head;
960 case DMSG_BLK_ERROR | DMSGF_REPLY:
961 tag->status = msg->any.blk_error;
964 lwkt_gettoken(&xa->tok);
965 if (msg->any.head.cmd & DMSGF_DELETE) { /* receive termination */
966 if (xa->opentag == tag) {
967 xa->opentag = NULL; /* XXX */
968 kdmsg_state_reply(tag->state, 0);
970 lwkt_reltoken(&xa->tok);
977 lwkt_reltoken(&xa->tok);
985 xa_bio_completion(kdmsg_state_t *state, kdmsg_msg_t *msg)
987 xa_tag_t *tag = state->any.any;
988 xa_softc_t *xa = tag->xa;
993 * Get the bio from the tag. If no bio is present we just do
996 if ((bio = tag->bio) == NULL)
1001 * Process return status
1003 switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
1004 case DMSG_LNK_ERROR | DMSGF_REPLY:
1005 bzero(&tag->status, sizeof(tag->status));
1006 tag->status.head = msg->any.head;
1007 if (tag->status.head.error)
1008 tag->status.resid = bp->b_bcount;
1010 tag->status.resid = 0;
1012 case DMSG_BLK_ERROR | DMSGF_REPLY:
1013 tag->status = msg->any.blk_error;
1018 * Potentially move the bio back onto the pending queue if the
1019 * device is open and the error is related to losing the virtual
1022 if (tag->status.head.error &&
1023 (msg->any.head.cmd & DMSGF_DELETE) && xa->opencnt) {
1024 if (tag->status.head.error == DMSG_ERR_LOSTLINK ||
1025 tag->status.head.error == DMSG_ERR_CANTCIRC) {
1031 * Process bio completion
1033 * For reads any returned data is zero-extended if necessary, so
1034 * the server can short-cut any all-zeros reads if it desires.
1038 if (msg->aux_data && msg->aux_size) {
1039 if (msg->aux_size < bp->b_bcount) {
1040 bcopy(msg->aux_data, bp->b_data, msg->aux_size);
1041 bzero(bp->b_data + msg->aux_size,
1042 bp->b_bcount - msg->aux_size);
1044 bcopy(msg->aux_data, bp->b_data, bp->b_bcount);
1047 bzero(bp->b_data, bp->b_bcount);
1052 case BUF_CMD_FREEBLKS:
1054 if (tag->status.resid > bp->b_bcount)
1055 tag->status.resid = bp->b_bcount;
1056 bp->b_resid = tag->status.resid;
1057 if ((bp->b_error = tag->status.head.error) != 0) {
1058 bp->b_flags |= B_ERROR;
1068 * Handle completion of the transaction. If the bioq is not empty
1069 * we can initiate another bio on the same tag.
1071 * NOTE: Most of our transactions will be single-message
1072 * CREATE+DELETEs, so we won't have to terminate the
1073 * transaction separately, here. But just in case they
1074 * aren't be sure to terminate the transaction.
1077 if (msg->any.head.cmd & DMSGF_DELETE) {
1079 if ((state->txcmd & DMSGF_DELETE) == 0)
1080 kdmsg_msg_reply(msg, 0);
1085 * Handle the case where the transaction failed due to a
1086 * connectivity issue. The tag is put away with wasbio=0
1087 * and we restart the bio.
1089 * Setting circ->lost causes xa_setup_cmd() to skip the circuit.
1090 * Other circuits might still be live. Once a circuit gets messed
1091 * up it will (eventually) be deleted so we can simply leave (lost)
1092 * set forever after.
1095 lwkt_gettoken(&xa->tok);
1096 kprintf("BIO CIRC FAILURE, REPEND BIO %p\n", bio);
1097 tag->circ->lost = 1;
1100 if ((state->txcmd & DMSGF_DELETE) == 0)
1101 kdmsg_msg_reply(msg, 0);
1104 * Restart or requeue the bio
1106 tag = xa_setup_cmd(xa, bio);
1108 xa_start(tag, NULL);
1109 lwkt_reltoken(&xa->tok);
1114 * Restart as much deferred I/O as we can.
1116 * Called with xa->tok held
1120 xa_restart_deferred(xa_softc_t *xa)
1125 while ((bio = TAILQ_FIRST(&xa->bioq)) != NULL) {
1126 tag = xa_setup_cmd(xa, NULL);
1129 TAILQ_REMOVE(&xa->bioq, bio, bio_act);
1131 xa_start(tag, NULL);