2 * Copyright (c) 2012-2014 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.
47 #include <sys/param.h>
48 #include <sys/systm.h>
51 #include <sys/device.h>
52 #include <sys/devicestat.h>
54 #include <sys/kernel.h>
55 #include <sys/malloc.h>
56 #include <sys/sysctl.h>
58 #include <sys/queue.h>
62 #include <sys/kern_syscall.h>
65 #include <sys/xdiskioctl.h>
68 #include <sys/thread2.h>
72 RB_HEAD(xa_softc_tree, xa_softc);
73 RB_PROTOTYPE(xa_softc_tree, xa_softc, rbnode, xa_softc_cmp);
76 SYSCTL_INT(_debug, OID_AUTO, xa_active, CTLFLAG_RW, &xa_active, 0,
77 "Number of active xdisk IOs");
78 static uint64_t xa_last;
79 SYSCTL_ULONG(_debug, OID_AUTO, xa_last, CTLFLAG_RW, &xa_last, 0,
80 "Offset of last xdisk IO");
86 TAILQ_ENTRY(xa_tag) entry;
88 dmsg_blk_error_t status;
96 typedef struct xa_tag xa_tag_t;
102 struct kdmsg_state_list spanq;
103 RB_ENTRY(xa_softc) rbnode;
105 struct devstat stats;
106 struct disk_info info;
115 char cl_label[64]; /* from LNK_SPAN cl_label (host/dev) */
116 char fs_label[64]; /* from LNK_SPAN fs_label (serno str) */
118 TAILQ_HEAD(, bio) bioq; /* pending BIOs */
119 TAILQ_HEAD(, xa_tag) tag_freeq; /* available I/O tags */
120 TAILQ_HEAD(, xa_tag) tag_pendq; /* running I/O tags */
121 struct lwkt_token tok;
124 typedef struct xa_softc xa_softc_t;
127 TAILQ_ENTRY(xa_iocom) entry;
132 typedef struct xa_iocom xa_iocom_t;
134 static int xa_softc_cmp(xa_softc_t *sc1, xa_softc_t *sc2);
135 RB_GENERATE(xa_softc_tree, xa_softc, rbnode, xa_softc_cmp);
136 static struct xa_softc_tree xa_device_tree;
138 #define MAXTAGS 64 /* no real limit */
140 static int xdisk_attach(struct xdisk_attach_ioctl *xaioc);
141 static int xdisk_detach(struct xdisk_attach_ioctl *xaioc);
142 static void xaio_exit(kdmsg_iocom_t *iocom);
143 static int xaio_rcvdmsg(kdmsg_msg_t *msg);
145 static void xa_terminate_check(struct xa_softc *sc);
147 static xa_tag_t *xa_setup_cmd(xa_softc_t *sc, struct bio *bio);
148 static void xa_start(xa_tag_t *tag, kdmsg_msg_t *msg, int async);
149 static void xa_done(xa_tag_t *tag, int wasbio);
150 static void xa_release(xa_tag_t *tag, int wasbio);
151 static uint32_t xa_wait(xa_tag_t *tag);
152 static int xa_sync_completion(kdmsg_state_t *state, kdmsg_msg_t *msg);
153 static int xa_bio_completion(kdmsg_state_t *state, kdmsg_msg_t *msg);
154 static void xa_restart_deferred(xa_softc_t *sc);
156 MALLOC_DEFINE(M_XDISK, "Networked disk client", "Network Disks");
159 * Control device, issue ioctls to create xa devices.
161 static d_open_t xdisk_open;
162 static d_close_t xdisk_close;
163 static d_ioctl_t xdisk_ioctl;
165 static struct dev_ops xdisk_ops = {
166 { "xdisk", 0, D_MPSAFE | D_TRACKCLOSE },
167 .d_open = xdisk_open,
168 .d_close = xdisk_close,
169 .d_ioctl = xdisk_ioctl
175 static d_open_t xa_open;
176 static d_close_t xa_close;
177 static d_ioctl_t xa_ioctl;
178 static d_strategy_t xa_strategy;
179 static d_psize_t xa_size;
181 static struct dev_ops xa_ops = {
182 { "xa", 0, D_DISK | D_CANFREE | D_MPSAFE | D_TRACKCLOSE },
187 .d_write = physwrite,
188 .d_strategy = xa_strategy,
192 static struct lwkt_token xdisk_token = LWKT_TOKEN_INITIALIZER(xdisk_token);
193 static int xdisk_opencount;
194 static cdev_t xdisk_dev;
195 static TAILQ_HEAD(, xa_iocom) xaiocomq;
198 * Module initialization
201 xdisk_modevent(module_t mod, int type, void *data)
205 TAILQ_INIT(&xaiocomq);
206 RB_INIT(&xa_device_tree);
207 xdisk_dev = make_dev(&xdisk_ops, 0,
208 UID_ROOT, GID_WHEEL, 0600, "xdisk");
212 if (xdisk_opencount || TAILQ_FIRST(&xaiocomq))
215 destroy_dev(xdisk_dev);
218 dev_ops_remove_all(&xdisk_ops);
219 dev_ops_remove_all(&xa_ops);
227 DEV_MODULE(xdisk, xdisk_modevent, 0);
230 xa_softc_cmp(xa_softc_t *sc1, xa_softc_t *sc2)
232 return(strcmp(sc1->fs_label, sc2->fs_label));
239 xdisk_open(struct dev_open_args *ap)
241 lwkt_gettoken(&xdisk_token);
243 lwkt_reltoken(&xdisk_token);
248 xdisk_close(struct dev_close_args *ap)
250 lwkt_gettoken(&xdisk_token);
252 lwkt_reltoken(&xdisk_token);
257 xdisk_ioctl(struct dev_ioctl_args *ap)
263 error = xdisk_attach((void *)ap->a_data);
266 error = xdisk_detach((void *)ap->a_data);
275 /************************************************************************
277 ************************************************************************/
280 xdisk_attach(struct xdisk_attach_ioctl *xaioc)
286 * Normalize ioctl params
288 kprintf("xdisk_attach1\n");
289 fp = holdfp(curproc->p_fd, xaioc->fd, -1);
292 kprintf("xdisk_attach2\n");
295 * See if the serial number is already present. If we are
296 * racing a termination the disk subsystem may still have
297 * duplicate entries not yet removed so we wait a bit and
300 lwkt_gettoken(&xdisk_token);
302 xaio = kmalloc(sizeof(*xaio), M_XDISK, M_WAITOK | M_ZERO);
303 kprintf("xdisk_attach3\n");
304 kdmsg_iocom_init(&xaio->iocom, xaio,
305 KDMSG_IOCOMF_AUTOCONN,
306 M_XDISK, xaio_rcvdmsg);
307 xaio->iocom.exit_func = xaio_exit;
309 kdmsg_iocom_reconnect(&xaio->iocom, fp, "xdisk");
312 * Setup our LNK_CONN advertisement for autoinitiate.
314 * Our filter is setup to only accept PEER_BLOCK/SERVER
317 * We need a unique pfs_fsid to avoid confusion.
319 xaio->iocom.auto_lnk_conn.pfs_type = DMSG_PFSTYPE_CLIENT;
320 xaio->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
321 xaio->iocom.auto_lnk_conn.peer_type = DMSG_PEER_BLOCK;
322 xaio->iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_BLOCK;
323 xaio->iocom.auto_lnk_conn.pfs_mask = 1LLU << DMSG_PFSTYPE_SERVER;
324 ksnprintf(xaio->iocom.auto_lnk_conn.fs_label,
325 sizeof(xaio->iocom.auto_lnk_conn.fs_label),
327 kern_uuidgen(&xaio->iocom.auto_lnk_conn.pfs_fsid, 1);
330 * Setup our LNK_SPAN advertisement for autoinitiate
332 TAILQ_INSERT_TAIL(&xaiocomq, xaio, entry);
333 kdmsg_iocom_autoinitiate(&xaio->iocom, NULL);
334 lwkt_reltoken(&xdisk_token);
340 xdisk_detach(struct xdisk_attach_ioctl *xaioc)
346 * Called from iocom core transmit thread upon disconnect.
350 xaio_exit(kdmsg_iocom_t *iocom)
352 xa_iocom_t *xaio = iocom->handle;
354 kprintf("xdisk_detach -xaio_exit\n");
355 lwkt_gettoken(&xdisk_token);
356 TAILQ_REMOVE(&xaiocomq, xaio, entry);
357 lwkt_reltoken(&xdisk_token);
359 kdmsg_iocom_uninit(&xaio->iocom);
361 kfree(xaio, M_XDISK);
365 * Called from iocom core to handle messages that the iocom core does not
366 * handle itself and for which a state function callback has not yet been
369 * We primarily care about LNK_SPAN transactions here.
372 xaio_rcvdmsg(kdmsg_msg_t *msg)
374 kdmsg_state_t *state = msg->state;
375 xa_iocom_t *xaio = state->iocom->handle;
378 kprintf("xdisk_rcvdmsg %08x (state cmd %08x)\n",
379 msg->any.head.cmd, msg->tcmd);
380 lwkt_gettoken(&xdisk_token);
383 case DMSG_LNK_SPAN | DMSGF_CREATE | DMSGF_DELETE:
385 * A LNK_SPAN transaction which is opened and closed
386 * degenerately is not useful to us, just ignore it.
388 kdmsg_msg_reply(msg, 0);
390 case DMSG_LNK_SPAN | DMSGF_CREATE:
392 * Manage the tracking node for the remote LNK_SPAN.
394 * Return a streaming result, leaving the transaction open
395 * in both directions to allow sub-transactions.
397 bcopy(msg->any.lnk_span.cl_label, xaio->dummysc.cl_label,
398 sizeof(xaio->dummysc.cl_label));
399 xaio->dummysc.cl_label[sizeof(xaio->dummysc.cl_label) - 1] = 0;
401 bcopy(msg->any.lnk_span.fs_label, xaio->dummysc.fs_label,
402 sizeof(xaio->dummysc.fs_label));
403 xaio->dummysc.fs_label[sizeof(xaio->dummysc.fs_label) - 1] = 0;
405 kprintf("xdisk: %s LNK_SPAN create state=%p ",
406 msg->any.lnk_span.fs_label,
409 sc = RB_FIND(xa_softc_tree, &xa_device_tree, &xaio->dummysc);
417 sc = kmalloc(sizeof(*sc), M_XDISK, M_WAITOK | M_ZERO);
418 kprintf("(not found - create %p)\n", sc);
419 bcopy(msg->any.lnk_span.cl_label, sc->cl_label,
420 sizeof(sc->cl_label));
421 sc->cl_label[sizeof(sc->cl_label) - 1] = 0;
422 bcopy(msg->any.lnk_span.fs_label, sc->fs_label,
423 sizeof(sc->fs_label));
424 sc->fs_label[sizeof(sc->fs_label) - 1] = 0;
426 /* XXX FIXME O(N^2) */
430 RB_FOREACH(sctmp, xa_softc_tree,
432 if (sctmp->unit == unit)
440 lwkt_token_init(&sc->tok, "xa");
441 TAILQ_INIT(&sc->spanq);
442 TAILQ_INIT(&sc->bioq);
443 TAILQ_INIT(&sc->tag_freeq);
444 TAILQ_INIT(&sc->tag_pendq);
445 RB_INSERT(xa_softc_tree, &xa_device_tree, sc);
446 TAILQ_INSERT_TAIL(&sc->spanq, msg->state, user_entry);
447 msg->state->any.xa_sc = sc;
452 for (n = 0; n < MAXTAGS; ++n) {
453 tag = kmalloc(sizeof(*tag),
454 M_XDISK, M_WAITOK|M_ZERO);
456 TAILQ_INSERT_TAIL(&sc->tag_freeq, tag, entry);
459 if (sc->dev == NULL) {
460 dev = disk_create(unit, &sc->disk, &xa_ops);
463 devstat_add_entry(&sc->stats, "xa", unit,
465 DEVSTAT_NO_ORDERED_TAGS,
466 DEVSTAT_TYPE_DIRECT |
467 DEVSTAT_TYPE_IF_OTHER,
468 DEVSTAT_PRIORITY_OTHER);
471 sc->info.d_media_blksize =
472 msg->any.lnk_span.media.block.blksize;
473 if (sc->info.d_media_blksize <= 0)
474 sc->info.d_media_blksize = 1;
475 sc->info.d_media_blocks =
476 msg->any.lnk_span.media.block.bytes /
477 sc->info.d_media_blksize;
478 sc->info.d_dsflags = DSO_MBRQUIET | DSO_RAWPSIZE;
479 sc->info.d_secpertrack = 32;
480 sc->info.d_nheads = 64;
481 sc->info.d_secpercyl = sc->info.d_secpertrack *
483 sc->info.d_ncylinders = 0;
485 sc->info.d_serialno = sc->fs_label;
487 * WARNING! disk_setdiskinfo() must be asynchronous
488 * because we are in the rxmsg thread. If
489 * it is synchronous and issues more disk
490 * I/Os, we will deadlock.
492 disk_setdiskinfo(&sc->disk, &sc->info);
493 xa_restart_deferred(sc); /* eats serializing */
495 kprintf("(found spancnt %d sc=%p)\n", sc->spancnt, sc);
497 TAILQ_INSERT_TAIL(&sc->spanq, msg->state, user_entry);
498 msg->state->any.xa_sc = sc;
499 if (sc->serializing == 0 && sc->open_tag == NULL) {
501 xa_restart_deferred(sc); /* eats serializing */
504 kdmsg_msg_result(msg, 0);
506 case DMSG_LNK_SPAN | DMSGF_DELETE:
508 * Manage the tracking node for the remote LNK_SPAN.
510 * Return a final result, closing our end of the transaction.
512 sc = msg->state->any.xa_sc;
513 kprintf("xdisk: %s LNK_SPAN terminate state=%p\n",
514 sc->fs_label, msg->state);
515 msg->state->any.xa_sc = NULL;
516 TAILQ_REMOVE(&sc->spanq, msg->state, user_entry);
518 xa_terminate_check(sc);
519 kdmsg_msg_reply(msg, 0);
521 case DMSG_LNK_SPAN | DMSGF_DELETE | DMSGF_REPLY:
522 case DMSG_LNK_SPAN | DMSGF_REPLY:
524 * Ignore unimplemented streaming replies on our LNK_SPAN
530 * Execute shell command (not supported atm).
532 * This is a one-way packet but if not (e.g. if part of
533 * a streaming transaction), we will have already closed
536 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
538 case DMSG_DBG_SHELL | DMSGF_REPLY:
540 * Receive one or more replies to a shell command
541 * that we sent. Just dump it to the console.
543 * This is a one-way packet but if not (e.g. if
544 * part of a streaming transaction), we will have
545 * already closed our end.
548 msg->aux_data[msg->aux_size - 1] = 0;
549 kprintf("xdisk: DEBUGMSG: %s\n",
555 * Unsupported one-way message, streaming message, or
558 * Terminate any unsupported transactions with an error
559 * and ignore any unsupported streaming messages.
561 * NOTE: This case also includes DMSG_LNK_ERROR messages
562 * which might be one-way, replying to those would
563 * cause an infinite ping-pong.
565 if (msg->any.head.cmd & DMSGF_CREATE)
566 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
569 lwkt_reltoken(&xdisk_token);
575 * Determine if we can destroy the xa_softc.
577 * Called with xdisk_token held.
581 xa_terminate_check(struct xa_softc *sc)
586 * Determine if we can destroy the softc.
588 kprintf("xdisk: terminate check xa%d (%d,%d,%d) sc=%p ",
590 sc->opencnt, sc->serializing, sc->spancnt,
593 if (sc->opencnt || sc->serializing || sc->spancnt) {
594 kprintf("(leave intact)\n");
597 kprintf("(remove from tree)\n");
599 KKASSERT(TAILQ_EMPTY(&sc->tag_pendq));
601 RB_REMOVE(xa_softc_tree, &xa_device_tree, sc);
604 disk_destroy(&sc->disk);
605 devstat_remove_entry(&sc->stats);
606 sc->dev->si_drv1 = NULL;
609 KKASSERT(sc->opencnt == 0);
610 KKASSERT(TAILQ_EMPTY(&sc->tag_pendq));
612 while ((tag = TAILQ_FIRST(&sc->tag_freeq)) != NULL) {
613 TAILQ_REMOVE(&sc->tag_freeq, tag, entry);
621 /************************************************************************
622 * XA DEVICE INTERFACE *
623 ************************************************************************/
626 xa_open(struct dev_open_args *ap)
628 cdev_t dev = ap->a_head.a_dev;
632 dev->si_bsize_phys = 512;
633 dev->si_bsize_best = 32768;
636 * Interlock open with opencnt, wait for attachment operations
639 lwkt_gettoken(&xdisk_token);
643 lwkt_reltoken(&xdisk_token);
644 return ENXIO; /* raced destruction */
646 if (sc->serializing) {
647 tsleep(sc, 0, "xarace", hz / 10);
653 * Serialize initial open
655 if (sc->opencnt++ > 0) {
658 lwkt_reltoken(&xdisk_token);
661 lwkt_reltoken(&xdisk_token);
664 * Issue BLK_OPEN if necessary. ENXIO is returned if we have trouble.
666 if (sc->open_tag == NULL) {
667 xa_restart_deferred(sc); /* eats serializing */
674 * Wait for completion of the BLK_OPEN
676 lwkt_gettoken(&xdisk_token);
677 while (sc->serializing)
678 tsleep(sc, 0, "xaopen", hz);
680 error = sc->last_error;
682 KKASSERT(sc->opencnt > 0);
684 xa_terminate_check(sc);
685 sc = NULL; /* sc may be invalid now */
687 lwkt_reltoken(&xdisk_token);
693 xa_close(struct dev_close_args *ap)
695 cdev_t dev = ap->a_head.a_dev;
701 return ENXIO; /* raced destruction */
702 lwkt_gettoken(&xdisk_token);
703 lwkt_gettoken(&sc->tok);
706 * NOTE: Clearing open_tag allows a concurrent open to re-open
707 * the device and prevents autonomous completion of the tag.
709 if (sc->opencnt == 1 && sc->open_tag) {
712 kdmsg_state_reply(tag->state, 0); /* close our side */
713 xa_wait(tag); /* wait on remote */
715 lwkt_reltoken(&sc->tok);
716 KKASSERT(sc->opencnt > 0);
718 xa_terminate_check(sc);
719 lwkt_reltoken(&xdisk_token);
725 xa_strategy(struct dev_strategy_args *ap)
727 xa_softc_t *sc = ap->a_head.a_dev->si_drv1;
729 struct bio *bio = ap->a_bio;
732 * Only BUF_CMD_READ is allowed (for probes) if opencnt is zero.
733 * Otherwise a BLK_OPEN transaction is required.
735 if (sc->opencnt == 0 && bio->bio_buf->b_cmd != BUF_CMD_READ) {
736 bio->bio_buf->b_error = ENXIO;
737 bio->bio_buf->b_flags |= B_ERROR;
741 devstat_start_transaction(&sc->stats);
742 atomic_add_int(&xa_active, 1);
743 xa_last = bio->bio_offset;
745 lwkt_gettoken(&sc->tok);
746 tag = xa_setup_cmd(sc, bio);
748 xa_start(tag, NULL, 1);
749 lwkt_reltoken(&sc->tok);
754 xa_ioctl(struct dev_ioctl_args *ap)
760 xa_size(struct dev_psize_args *ap)
764 if ((sc = ap->a_head.a_dev->si_drv1) == NULL)
766 ap->a_result = sc->info.d_media_blocks;
770 /************************************************************************
771 * XA BLOCK PROTOCOL STATE MACHINE *
772 ************************************************************************
774 * Implement tag/msg setup and related functions.
777 xa_setup_cmd(xa_softc_t *sc, struct bio *bio)
782 * Only get a tag if we have a valid virtual circuit to the server.
784 lwkt_gettoken(&sc->tok);
785 if ((tag = TAILQ_FIRST(&sc->tag_freeq)) != NULL) {
786 TAILQ_REMOVE(&sc->tag_freeq, tag, entry);
788 TAILQ_INSERT_TAIL(&sc->tag_pendq, tag, entry);
792 * If we can't dispatch now and this is a bio, queue it for later.
794 if (tag == NULL && bio) {
795 TAILQ_INSERT_TAIL(&sc->bioq, bio, bio_act);
797 lwkt_reltoken(&sc->tok);
803 xa_start(xa_tag_t *tag, kdmsg_msg_t *msg, int async)
805 xa_softc_t *sc = tag->sc;
820 if (sc->opencnt == 0 || sc->open_tag == NULL) {
823 TAILQ_FOREACH(span, &sc->spanq, user_entry) {
824 if ((span->rxcmd & DMSGF_DELETE) == 0)
829 msg = kdmsg_msg_alloc(span,
833 xa_bio_completion, tag);
835 msg = kdmsg_msg_alloc(sc->open_tag->state,
839 xa_bio_completion, tag);
841 msg->any.blk_read.keyid = sc->keyid;
842 msg->any.blk_read.offset = bio->bio_offset;
843 msg->any.blk_read.bytes = bp->b_bcount;
846 msg = kdmsg_msg_alloc(sc->open_tag->state,
848 DMSGF_CREATE | DMSGF_DELETE,
849 xa_bio_completion, tag);
850 msg->any.blk_write.keyid = sc->keyid;
851 msg->any.blk_write.offset = bio->bio_offset;
852 msg->any.blk_write.bytes = bp->b_bcount;
853 msg->aux_data = bp->b_data;
854 msg->aux_size = bp->b_bcount;
857 msg = kdmsg_msg_alloc(sc->open_tag->state,
859 DMSGF_CREATE | DMSGF_DELETE,
860 xa_bio_completion, tag);
861 msg->any.blk_flush.keyid = sc->keyid;
862 msg->any.blk_flush.offset = bio->bio_offset;
863 msg->any.blk_flush.bytes = bp->b_bcount;
865 case BUF_CMD_FREEBLKS:
866 msg = kdmsg_msg_alloc(sc->open_tag->state,
868 DMSGF_CREATE | DMSGF_DELETE,
869 xa_bio_completion, tag);
870 msg->any.blk_freeblks.keyid = sc->keyid;
871 msg->any.blk_freeblks.offset = bio->bio_offset;
872 msg->any.blk_freeblks.bytes = bp->b_bcount;
875 bp->b_flags |= B_ERROR;
877 devstat_end_transaction_buf(&sc->stats, bp);
878 atomic_add_int(&xa_active, -1);
886 tag->state = msg->state;
887 kdmsg_msg_write(msg);
889 tag->status.head.error = DMSG_ERR_IO;
895 xa_wait(xa_tag_t *tag)
897 xa_softc_t *sc = tag->sc;
900 kprintf("xdisk: xa_wait %p\n", tag);
902 lwkt_gettoken(&sc->tok);
904 while (tag->done == 0)
905 tsleep(tag, 0, "xawait", 0);
906 lwkt_reltoken(&sc->tok);
907 error = tag->status.head.error;
915 xa_done(xa_tag_t *tag, int wasbio)
917 KKASSERT(tag->bio == NULL);
924 xa_release(tag, wasbio);
928 * Release a tag. If everything looks ok and there are pending BIOs
929 * (due to all tags in-use), we can use the tag to start the next BIO.
930 * Do not try to restart if the connection is currently failed.
934 xa_release(xa_tag_t *tag, int wasbio)
936 xa_softc_t *sc = tag->sc;
939 lwkt_gettoken(&sc->tok);
940 if (wasbio && sc->open_tag &&
941 (bio = TAILQ_FIRST(&sc->bioq)) != NULL) {
942 TAILQ_REMOVE(&sc->bioq, bio, bio_act);
944 xa_start(tag, NULL, 1);
945 lwkt_reltoken(&sc->tok);
947 TAILQ_REMOVE(&sc->tag_pendq, tag, entry);
948 TAILQ_INSERT_TAIL(&sc->tag_freeq, tag, entry);
949 lwkt_reltoken(&sc->tok);
954 * Handle messages under the BLKOPEN transaction.
957 xa_sync_completion(kdmsg_state_t *state, kdmsg_msg_t *msg)
959 xa_tag_t *tag = state->any.any;
964 * If the tag has been cleaned out we already closed our side
965 * of the transaction and we are waiting for the other side to
969 if (msg->any.head.cmd & DMSGF_CREATE)
970 kdmsg_state_reply(state, DMSG_ERR_LOSTLINK);
978 lwkt_gettoken(&sc->tok);
981 * Handle initial response to our open and restart any deferred
984 * NOTE: DELETE may also be set.
986 if (msg->any.head.cmd & DMSGF_CREATE) {
987 switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
988 case DMSG_LNK_ERROR | DMSGF_REPLY:
989 bzero(&tag->status, sizeof(tag->status));
990 tag->status.head = msg->any.head;
992 case DMSG_BLK_ERROR | DMSGF_REPLY:
993 tag->status = msg->any.blk_error;
996 sc->last_error = tag->status.head.error;
997 kprintf("xdisk: blk_open completion status %d\n",
999 if (sc->last_error == 0) {
1000 while ((bio = TAILQ_FIRST(&sc->bioq)) != NULL) {
1001 tag = xa_setup_cmd(sc, NULL);
1004 TAILQ_REMOVE(&sc->bioq, bio, bio_act);
1006 xa_start(tag, NULL, 1);
1009 sc->serializing = 0;
1014 * Handle unexpected termination (or lost comm channel) from other
1015 * side. Autonomous completion only if open_tag matches,
1016 * otherwise another thread is probably waiting on the tag.
1018 * (see xa_close() for other interactions)
1020 if (msg->any.head.cmd & DMSGF_DELETE) {
1021 kdmsg_state_reply(tag->state, 0);
1022 if (sc->open_tag == tag) {
1023 sc->open_tag = NULL;
1030 lwkt_reltoken(&sc->tok);
1035 xa_bio_completion(kdmsg_state_t *state, kdmsg_msg_t *msg)
1037 xa_tag_t *tag = state->any.any;
1038 xa_softc_t *sc = tag->sc;
1043 * Get the bio from the tag. If no bio is present we just do
1046 if ((bio = tag->bio) == NULL)
1051 * Process return status
1053 switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
1054 case DMSG_LNK_ERROR | DMSGF_REPLY:
1055 bzero(&tag->status, sizeof(tag->status));
1056 tag->status.head = msg->any.head;
1057 if (tag->status.head.error)
1058 tag->status.resid = bp->b_bcount;
1060 tag->status.resid = 0;
1062 case DMSG_BLK_ERROR | DMSGF_REPLY:
1063 tag->status = msg->any.blk_error;
1068 * If the device is open stall the bio on DMSG errors. If an
1069 * actual I/O error occured on the remote device, DMSG_ERR_IO
1072 if (tag->status.head.error &&
1073 (msg->any.head.cmd & DMSGF_DELETE) && sc->opencnt) {
1074 if (tag->status.head.error != DMSG_ERR_IO)
1079 * Process bio completion
1081 * For reads any returned data is zero-extended if necessary, so
1082 * the server can short-cut any all-zeros reads if it desires.
1086 if (msg->aux_data && msg->aux_size) {
1087 if (msg->aux_size < bp->b_bcount) {
1088 bcopy(msg->aux_data, bp->b_data, msg->aux_size);
1089 bzero(bp->b_data + msg->aux_size,
1090 bp->b_bcount - msg->aux_size);
1092 bcopy(msg->aux_data, bp->b_data, bp->b_bcount);
1095 bzero(bp->b_data, bp->b_bcount);
1100 case BUF_CMD_FREEBLKS:
1102 if (tag->status.resid > bp->b_bcount)
1103 tag->status.resid = bp->b_bcount;
1104 bp->b_resid = tag->status.resid;
1105 if (tag->status.head.error != 0) {
1107 bp->b_flags |= B_ERROR;
1111 devstat_end_transaction_buf(&sc->stats, bp);
1112 atomic_add_int(&xa_active, -1);
1119 * Handle completion of the transaction. If the bioq is not empty
1120 * we can initiate another bio on the same tag.
1122 * NOTE: Most of our transactions will be single-message
1123 * CREATE+DELETEs, so we won't have to terminate the
1124 * transaction separately, here. But just in case they
1125 * aren't be sure to terminate the transaction.
1128 if (msg->any.head.cmd & DMSGF_DELETE) {
1130 if ((state->txcmd & DMSGF_DELETE) == 0)
1131 kdmsg_msg_reply(msg, 0);
1136 * Handle the case where the transaction failed due to a
1137 * connectivity issue. The tag is put away with wasbio=0
1138 * and we put the BIO back onto the bioq for a later restart.
1141 lwkt_gettoken(&sc->tok);
1142 kprintf("BIO CIRC FAILURE, REPEND BIO %p\n", bio);
1145 if ((state->txcmd & DMSGF_DELETE) == 0)
1146 kdmsg_msg_reply(msg, 0);
1151 TAILQ_INSERT_TAIL(&sc->bioq, bio, bio_act);
1153 lwkt_reltoken(&sc->tok);
1158 * Restart as much deferred I/O as we can. The serializer is set and we
1159 * eat it (clear it) when done.
1161 * Called with sc->tok held
1165 xa_restart_deferred(xa_softc_t *sc)
1167 kdmsg_state_t *span;
1172 KKASSERT(sc->serializing);
1175 * Determine if a restart is needed.
1177 if (sc->opencnt == 0) {
1179 * Device is not open, nothing to do, eat serializing.
1181 sc->serializing = 0;
1183 } else if (sc->open_tag == NULL) {
1185 * BLK_OPEN required before we can restart any BIOs.
1186 * Select the best LNK_SPAN to issue the BLK_OPEN under.
1188 * serializing interlocks waiting open()s.
1191 TAILQ_FOREACH(span, &sc->spanq, user_entry) {
1192 if ((span->rxcmd & DMSGF_DELETE) == 0)
1199 tag = xa_setup_cmd(sc, NULL);
1204 kprintf("xdisk: BLK_OPEN\n");
1206 msg = kdmsg_msg_alloc(span,
1209 xa_sync_completion, tag);
1210 msg->any.blk_open.modes = DMSG_BLKOPEN_RD;
1211 xa_start(tag, msg, 0);
1214 sc->serializing = 0;
1217 /* else leave serializing set until BLK_OPEN response */
1220 sc->serializing = 0;