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");
81 static int xa_debug = 1;
82 SYSCTL_INT(_debug, OID_AUTO, xa_debug, CTLFLAG_RW, &xa_debug, 0,
89 TAILQ_ENTRY(xa_tag) entry;
91 dmsg_blk_error_t status;
99 typedef struct xa_tag xa_tag_t;
105 struct kdmsg_state_list spanq;
106 RB_ENTRY(xa_softc) rbnode;
108 struct devstat stats;
109 struct disk_info info;
119 char peer_label[64]; /* from LNK_SPAN host/dev */
120 char pfs_label[64]; /* from LNK_SPAN serno */
122 TAILQ_HEAD(, bio) bioq; /* pending BIOs */
123 TAILQ_HEAD(, xa_tag) tag_freeq; /* available I/O tags */
124 TAILQ_HEAD(, xa_tag) tag_pendq; /* running I/O tags */
128 typedef struct xa_softc xa_softc_t;
131 TAILQ_ENTRY(xa_iocom) entry;
136 typedef struct xa_iocom xa_iocom_t;
138 static int xa_softc_cmp(xa_softc_t *sc1, xa_softc_t *sc2);
139 RB_GENERATE(xa_softc_tree, xa_softc, rbnode, xa_softc_cmp);
140 static struct xa_softc_tree xa_device_tree;
142 #define MAXTAGS 64 /* no real limit */
144 static int xdisk_attach(struct xdisk_attach_ioctl *xaioc);
145 static int xdisk_detach(struct xdisk_attach_ioctl *xaioc);
146 static void xaio_exit(kdmsg_iocom_t *iocom);
147 static int xaio_rcvdmsg(kdmsg_msg_t *msg);
149 static void xa_terminate_check(struct xa_softc *sc);
151 static xa_tag_t *xa_setup_cmd(xa_softc_t *sc, struct bio *bio);
152 static void xa_start(xa_tag_t *tag, kdmsg_msg_t *msg, int async);
153 static void xa_done(xa_tag_t *tag, int wasbio);
154 static void xa_release(xa_tag_t *tag, int wasbio);
155 static uint32_t xa_wait(xa_tag_t *tag);
156 static int xa_sync_completion(kdmsg_state_t *state, kdmsg_msg_t *msg);
157 static int xa_bio_completion(kdmsg_state_t *state, kdmsg_msg_t *msg);
158 static void xa_restart_deferred(xa_softc_t *sc);
160 #define xa_printf(level, ctl, ...) \
161 if (xa_debug >= (level)) kprintf("xdisk: " ctl, __VA_ARGS__)
163 MALLOC_DEFINE(M_XDISK, "Networked disk client", "Network Disks");
166 * Control device, issue ioctls to create xa devices.
168 static d_open_t xdisk_open;
169 static d_close_t xdisk_close;
170 static d_ioctl_t xdisk_ioctl;
172 static struct dev_ops xdisk_ops = {
173 { "xdisk", 0, D_MPSAFE | D_TRACKCLOSE },
174 .d_open = xdisk_open,
175 .d_close = xdisk_close,
176 .d_ioctl = xdisk_ioctl
182 static d_open_t xa_open;
183 static d_close_t xa_close;
184 static d_ioctl_t xa_ioctl;
185 static d_strategy_t xa_strategy;
186 static d_psize_t xa_size;
188 static struct dev_ops xa_ops = {
189 { "xa", 0, D_DISK | D_CANFREE | D_MPSAFE | D_TRACKCLOSE },
194 .d_write = physwrite,
195 .d_strategy = xa_strategy,
199 static int xdisk_opencount;
200 static cdev_t xdisk_dev;
201 struct lock xdisk_lk;
202 static TAILQ_HEAD(, xa_iocom) xaiocomq;
205 * Module initialization
208 xdisk_modevent(module_t mod, int type, void *data)
212 TAILQ_INIT(&xaiocomq);
213 RB_INIT(&xa_device_tree);
214 lockinit(&xdisk_lk, "xdisk", 0, 0);
215 xdisk_dev = make_dev(&xdisk_ops, 0,
216 UID_ROOT, GID_WHEEL, 0600, "xdisk");
220 if (!RB_EMPTY(&xa_device_tree))
222 if (xdisk_opencount || TAILQ_FIRST(&xaiocomq))
225 destroy_dev(xdisk_dev);
228 dev_ops_remove_all(&xdisk_ops);
229 dev_ops_remove_all(&xa_ops);
237 DEV_MODULE(xdisk, xdisk_modevent, 0);
240 xa_softc_cmp(xa_softc_t *sc1, xa_softc_t *sc2)
242 return(strcmp(sc1->pfs_label, sc2->pfs_label));
249 xdisk_open(struct dev_open_args *ap)
251 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
253 lockmgr(&xdisk_lk, LK_RELEASE);
258 xdisk_close(struct dev_close_args *ap)
260 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
262 lockmgr(&xdisk_lk, LK_RELEASE);
267 xdisk_ioctl(struct dev_ioctl_args *ap)
273 error = xdisk_attach((void *)ap->a_data);
276 error = xdisk_detach((void *)ap->a_data);
285 /************************************************************************
287 ************************************************************************/
290 xdisk_attach(struct xdisk_attach_ioctl *xaioc)
296 * Normalize ioctl params
298 fp = holdfp(curproc->p_fd, xaioc->fd, -1);
301 xa_printf(1, "xdisk_attach fp=%p\n", fp);
304 * See if the serial number is already present. If we are
305 * racing a termination the disk subsystem may still have
306 * duplicate entries not yet removed so we wait a bit and
309 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
311 xaio = kmalloc(sizeof(*xaio), M_XDISK, M_WAITOK | M_ZERO);
312 kdmsg_iocom_init(&xaio->iocom, xaio,
313 KDMSG_IOCOMF_AUTOCONN,
314 M_XDISK, xaio_rcvdmsg);
315 xaio->iocom.exit_func = xaio_exit;
317 kdmsg_iocom_reconnect(&xaio->iocom, fp, "xdisk");
320 * Setup our LNK_CONN advertisement for autoinitiate.
322 * Our filter is setup to only accept PEER_BLOCK advertisements.
323 * XXX no peer_id filter.
325 * We need a unique pfs_fsid to avoid confusion.
327 xaio->iocom.auto_lnk_conn.peer_type = DMSG_PEER_CLIENT;
328 xaio->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
329 xaio->iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_BLOCK;
330 ksnprintf(xaio->iocom.auto_lnk_conn.peer_label,
331 sizeof(xaio->iocom.auto_lnk_conn.peer_label),
334 /* kern_uuidgen(&xaio->iocom.auto_lnk_conn.pfs_fsid, 1); */
337 * Setup our LNK_SPAN advertisement for autoinitiate
339 TAILQ_INSERT_TAIL(&xaiocomq, xaio, entry);
340 kdmsg_iocom_autoinitiate(&xaio->iocom, NULL);
342 lockmgr(&xdisk_lk, LK_RELEASE);
348 xdisk_detach(struct xdisk_attach_ioctl *xaioc)
354 * Called from iocom core transmit thread upon disconnect.
358 xaio_exit(kdmsg_iocom_t *iocom)
360 xa_iocom_t *xaio = iocom->handle;
362 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
363 xa_printf(1, "%s", "xdisk_detach [xaio_exit()]\n");
364 TAILQ_REMOVE(&xaiocomq, xaio, entry);
365 lockmgr(&xdisk_lk, LK_RELEASE);
367 kdmsg_iocom_uninit(&xaio->iocom);
369 kfree(xaio, M_XDISK);
373 * Called from iocom core to handle messages that the iocom core does not
374 * handle itself and for which a state function callback has not yet been
377 * We primarily care about LNK_SPAN transactions here.
380 xaio_rcvdmsg(kdmsg_msg_t *msg)
382 kdmsg_state_t *state = msg->state;
383 xa_iocom_t *xaio = state->iocom->handle;
388 "xdisk - rcvmsg state=%p rx=%08x tx=%08x msgcmd=%08x\n",
389 state, state->rxcmd, state->txcmd,
392 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
395 case DMSG_LNK_SPAN | DMSGF_CREATE | DMSGF_DELETE:
397 * A LNK_SPAN transaction which is opened and closed
398 * degenerately is not useful to us, just ignore it.
400 kdmsg_msg_reply(msg, 0);
402 case DMSG_LNK_SPAN | DMSGF_CREATE:
404 * Manage the tracking node for the remote LNK_SPAN.
406 * Return a streaming result, leaving the transaction open
407 * in both directions to allow sub-transactions.
409 bcopy(msg->any.lnk_span.peer_label, xaio->dummysc.peer_label,
410 sizeof(xaio->dummysc.peer_label));
411 xaio->dummysc.peer_label[
412 sizeof(xaio->dummysc.peer_label) - 1] = 0;
414 bcopy(msg->any.lnk_span.pfs_label, xaio->dummysc.pfs_label,
415 sizeof(xaio->dummysc.pfs_label));
416 xaio->dummysc.pfs_label[
417 sizeof(xaio->dummysc.pfs_label) - 1] = 0;
419 xa_printf(3, "LINK_SPAN state %p create for %s\n",
420 msg->state, msg->any.lnk_span.pfs_label);
422 sc = RB_FIND(xa_softc_tree, &xa_device_tree, &xaio->dummysc);
430 sc = kmalloc(sizeof(*sc), M_XDISK, M_WAITOK | M_ZERO);
431 bcopy(msg->any.lnk_span.peer_label, sc->peer_label,
432 sizeof(sc->peer_label));
433 sc->peer_label[sizeof(sc->peer_label) - 1] = 0;
434 bcopy(msg->any.lnk_span.pfs_label, sc->pfs_label,
435 sizeof(sc->pfs_label));
436 sc->pfs_label[sizeof(sc->pfs_label) - 1] = 0;
438 /* XXX FIXME O(N^2) */
442 RB_FOREACH(sctmp, xa_softc_tree,
444 if (sctmp->unit == unit)
452 lockinit(&sc->lk, "xalk", 0, 0);
453 TAILQ_INIT(&sc->spanq);
454 TAILQ_INIT(&sc->bioq);
455 TAILQ_INIT(&sc->tag_freeq);
456 TAILQ_INIT(&sc->tag_pendq);
458 lockmgr(&sc->lk, LK_EXCLUSIVE);
459 RB_INSERT(xa_softc_tree, &xa_device_tree, sc);
460 TAILQ_INSERT_TAIL(&sc->spanq, msg->state, user_entry);
461 msg->state->any.xa_sc = sc;
466 for (n = 0; n < MAXTAGS; ++n) {
467 tag = kmalloc(sizeof(*tag),
468 M_XDISK, M_WAITOK|M_ZERO);
470 TAILQ_INSERT_TAIL(&sc->tag_freeq, tag, entry);
473 if (sc->dev == NULL) {
474 dev = disk_create(unit, &sc->disk, &xa_ops);
477 devstat_add_entry(&sc->stats, "xa", unit,
479 DEVSTAT_NO_ORDERED_TAGS,
480 DEVSTAT_TYPE_DIRECT |
481 DEVSTAT_TYPE_IF_OTHER,
482 DEVSTAT_PRIORITY_OTHER);
485 sc->info.d_media_blksize =
486 msg->any.lnk_span.media.block.blksize;
487 if (sc->info.d_media_blksize <= 0)
488 sc->info.d_media_blksize = 1;
489 sc->info.d_media_blocks =
490 msg->any.lnk_span.media.block.bytes /
491 sc->info.d_media_blksize;
492 sc->info.d_dsflags = DSO_MBRQUIET | DSO_RAWPSIZE;
493 sc->info.d_secpertrack = 32;
494 sc->info.d_nheads = 64;
495 sc->info.d_secpercyl = sc->info.d_secpertrack *
497 sc->info.d_ncylinders = 0;
498 if (sc->pfs_label[0])
499 sc->info.d_serialno = sc->pfs_label;
501 * WARNING! disk_setdiskinfo() must be asynchronous
502 * because we are in the rxmsg thread. If
503 * it is synchronous and issues more disk
504 * I/Os, we will deadlock.
506 disk_setdiskinfo(&sc->disk, &sc->info);
507 xa_restart_deferred(sc); /* eats serializing */
508 lockmgr(&sc->lk, LK_RELEASE);
510 lockmgr(&sc->lk, LK_EXCLUSIVE);
512 TAILQ_INSERT_TAIL(&sc->spanq, msg->state, user_entry);
513 msg->state->any.xa_sc = sc;
514 if (sc->serializing == 0 && sc->open_tag == NULL) {
516 xa_restart_deferred(sc); /* eats serializing */
518 lockmgr(&sc->lk, LK_RELEASE);
519 if (sc->dev && sc->dev->si_disk) {
520 xa_printf(1, "reprobe disk: %s\n",
522 disk_msg_send(DISK_DISK_REPROBE,
527 xa_printf(2, "sc %p spancnt %d\n", sc, sc->spancnt);
528 kdmsg_msg_result(msg, 0);
530 case DMSG_LNK_SPAN | DMSGF_DELETE:
532 * Manage the tracking node for the remote LNK_SPAN.
534 * Return a final result, closing our end of the transaction.
536 sc = msg->state->any.xa_sc;
537 xa_printf(3, "LINK_SPAN state %p delete for %s (sc=%p)\n",
538 msg->state, (sc ? sc->pfs_label : "(null)"), sc);
539 lockmgr(&sc->lk, LK_EXCLUSIVE);
540 msg->state->any.xa_sc = NULL;
541 TAILQ_REMOVE(&sc->spanq, msg->state, user_entry);
544 xa_printf(2, "sc %p spancnt %d\n", sc, sc->spancnt);
547 * Spans can come and go as the graph stabilizes, so if
548 * we lose a span along with sc->open_tag we may be able
549 * to restart the I/Os on a different span.
552 sc->serializing == 0 && sc->open_tag == NULL) {
554 xa_restart_deferred(sc);
556 lockmgr(&sc->lk, LK_RELEASE);
557 kdmsg_msg_reply(msg, 0);
563 if (sc->spancnt == 0)
564 xa_terminate_check(sc);
567 case DMSG_LNK_SPAN | DMSGF_DELETE | DMSGF_REPLY:
569 * Ignore unimplemented streaming replies on our LNK_SPAN
572 xa_printf(3, "LINK_SPAN state %p delete+reply\n",
575 case DMSG_LNK_SPAN | DMSGF_REPLY:
577 * Ignore unimplemented streaming replies on our LNK_SPAN
580 xa_printf(3, "LINK_SPAN state %p reply\n",
585 * Execute shell command (not supported atm).
587 * This is a one-way packet but if not (e.g. if part of
588 * a streaming transaction), we will have already closed
591 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
593 case DMSG_DBG_SHELL | DMSGF_REPLY:
595 * Receive one or more replies to a shell command
596 * that we sent. Just dump it to the console.
598 * This is a one-way packet but if not (e.g. if
599 * part of a streaming transaction), we will have
600 * already closed our end.
603 msg->aux_data[msg->aux_size - 1] = 0;
604 xa_printf(0, "DEBUGMSG: %s\n", msg->aux_data);
609 * Unsupported one-way message, streaming message, or
612 * Terminate any unsupported transactions with an error
613 * and ignore any unsupported streaming messages.
615 * NOTE: This case also includes DMSG_LNK_ERROR messages
616 * which might be one-way, replying to those would
617 * cause an infinite ping-pong.
619 if (msg->any.head.cmd & DMSGF_CREATE)
620 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
623 lockmgr(&xdisk_lk, LK_RELEASE);
629 * Determine if we can destroy the xa_softc.
631 * Called with xdisk_lk held.
635 xa_terminate_check(struct xa_softc *sc)
640 * Determine if we can destroy the softc.
642 xa_printf(1, "Terminate check xa%d (%d,%d,%d) sc=%p ",
644 sc->opencnt, sc->serializing, sc->spancnt,
647 if (sc->opencnt || sc->serializing || sc->spancnt ||
648 TAILQ_FIRST(&sc->bioq) || TAILQ_FIRST(&sc->tag_pendq)) {
649 xa_printf(1, "%s", "(leave intact)\n");
654 * Remove from device tree, a race with a new incoming span
655 * will create a new softc and disk.
657 RB_REMOVE(xa_softc_tree, &xa_device_tree, sc);
661 * Device has to go first to prevent device ops races.
664 disk_destroy(&sc->disk);
665 devstat_remove_entry(&sc->stats);
666 sc->dev->si_drv1 = NULL;
670 xa_printf(1, "%s", "(remove from tree)\n");
672 KKASSERT(sc->opencnt == 0);
673 KKASSERT(TAILQ_EMPTY(&sc->tag_pendq));
675 while ((tag = TAILQ_FIRST(&sc->tag_freeq)) != NULL) {
676 TAILQ_REMOVE(&sc->tag_freeq, tag, entry);
684 /************************************************************************
685 * XA DEVICE INTERFACE *
686 ************************************************************************/
689 xa_open(struct dev_open_args *ap)
691 cdev_t dev = ap->a_head.a_dev;
695 dev->si_bsize_phys = 512;
696 dev->si_bsize_best = 32768;
699 * Interlock open with opencnt, wait for attachment operations
702 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
706 lockmgr(&xdisk_lk, LK_RELEASE);
707 return ENXIO; /* raced destruction */
709 if (sc->serializing) {
710 tsleep(sc, 0, "xarace", hz / 10);
713 if (sc->terminating) {
714 lockmgr(&xdisk_lk, LK_RELEASE);
715 return ENXIO; /* raced destruction */
720 * Serialize initial open
722 if (sc->opencnt++ > 0) {
725 lockmgr(&xdisk_lk, LK_RELEASE);
730 * Issue BLK_OPEN if necessary. ENXIO is returned if we have trouble.
732 if (sc->open_tag == NULL) {
733 lockmgr(&sc->lk, LK_EXCLUSIVE);
734 xa_restart_deferred(sc); /* eats serializing */
735 lockmgr(&sc->lk, LK_RELEASE);
740 lockmgr(&xdisk_lk, LK_RELEASE);
743 * Wait for completion of the BLK_OPEN
745 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
746 while (sc->serializing)
747 lksleep(sc, &xdisk_lk, 0, "xaopen", hz);
749 error = sc->last_error;
751 KKASSERT(sc->opencnt > 0);
753 xa_terminate_check(sc);
754 sc = NULL; /* sc may be invalid now */
756 lockmgr(&xdisk_lk, LK_RELEASE);
762 xa_close(struct dev_close_args *ap)
764 cdev_t dev = ap->a_head.a_dev;
768 lockmgr(&xdisk_lk, LK_EXCLUSIVE);
771 lockmgr(&sc->lk, LK_RELEASE);
772 return ENXIO; /* raced destruction */
774 if (sc->terminating) {
775 lockmgr(&sc->lk, LK_RELEASE);
776 return ENXIO; /* raced destruction */
778 lockmgr(&sc->lk, LK_EXCLUSIVE);
781 * NOTE: Clearing open_tag allows a concurrent open to re-open
782 * the device and prevents autonomous completion of the tag.
784 if (sc->opencnt == 1 && sc->open_tag) {
787 lockmgr(&sc->lk, LK_RELEASE);
788 kdmsg_state_reply(tag->state, 0); /* close our side */
789 xa_wait(tag); /* wait on remote */
791 lockmgr(&sc->lk, LK_RELEASE);
793 KKASSERT(sc->opencnt > 0);
795 xa_terminate_check(sc);
796 lockmgr(&xdisk_lk, LK_RELEASE);
802 xa_strategy(struct dev_strategy_args *ap)
804 xa_softc_t *sc = ap->a_head.a_dev->si_drv1;
806 struct bio *bio = ap->a_bio;
808 devstat_start_transaction(&sc->stats);
809 atomic_add_int(&xa_active, 1);
810 xa_last = bio->bio_offset;
813 * If no tags are available NULL is returned and the bio is
814 * placed on sc->bioq.
816 lockmgr(&sc->lk, LK_EXCLUSIVE);
817 tag = xa_setup_cmd(sc, bio);
819 xa_start(tag, NULL, 1);
820 lockmgr(&sc->lk, LK_RELEASE);
826 xa_ioctl(struct dev_ioctl_args *ap)
832 xa_size(struct dev_psize_args *ap)
836 if ((sc = ap->a_head.a_dev->si_drv1) == NULL)
838 ap->a_result = sc->info.d_media_blocks;
842 /************************************************************************
843 * XA BLOCK PROTOCOL STATE MACHINE *
844 ************************************************************************
846 * Implement tag/msg setup and related functions.
847 * Called with sc->lk held.
850 xa_setup_cmd(xa_softc_t *sc, struct bio *bio)
855 * Only get a tag if we have a valid virtual circuit to the server.
857 if ((tag = TAILQ_FIRST(&sc->tag_freeq)) != NULL) {
858 TAILQ_REMOVE(&sc->tag_freeq, tag, entry);
860 TAILQ_INSERT_TAIL(&sc->tag_pendq, tag, entry);
864 * If we can't dispatch now and this is a bio, queue it for later.
866 if (tag == NULL && bio) {
867 TAILQ_INSERT_TAIL(&sc->bioq, bio, bio_act);
874 * Called with sc->lk held
877 xa_start(xa_tag_t *tag, kdmsg_msg_t *msg, int async)
879 xa_softc_t *sc = tag->sc;
883 tag->status.head.error = DMSG_ERR_IO; /* fallback error */
888 kdmsg_state_t *trans;
890 if (sc->opencnt == 0 || sc->open_tag == NULL) {
891 TAILQ_FOREACH(trans, &sc->spanq, user_entry) {
892 if ((trans->rxcmd & DMSGF_DELETE) == 0)
896 trans = sc->open_tag->state;
907 msg = kdmsg_msg_alloc(trans,
911 xa_bio_completion, tag);
912 msg->any.blk_read.keyid = sc->keyid;
913 msg->any.blk_read.offset = bio->bio_offset;
914 msg->any.blk_read.bytes = bp->b_bcount;
917 msg = kdmsg_msg_alloc(trans,
919 DMSGF_CREATE | DMSGF_DELETE,
920 xa_bio_completion, tag);
921 msg->any.blk_write.keyid = sc->keyid;
922 msg->any.blk_write.offset = bio->bio_offset;
923 msg->any.blk_write.bytes = bp->b_bcount;
924 msg->aux_data = bp->b_data;
925 msg->aux_size = bp->b_bcount;
928 msg = kdmsg_msg_alloc(trans,
930 DMSGF_CREATE | DMSGF_DELETE,
931 xa_bio_completion, tag);
932 msg->any.blk_flush.keyid = sc->keyid;
933 msg->any.blk_flush.offset = bio->bio_offset;
934 msg->any.blk_flush.bytes = bp->b_bcount;
936 case BUF_CMD_FREEBLKS:
937 msg = kdmsg_msg_alloc(trans,
939 DMSGF_CREATE | DMSGF_DELETE,
940 xa_bio_completion, tag);
941 msg->any.blk_freeblks.keyid = sc->keyid;
942 msg->any.blk_freeblks.offset = bio->bio_offset;
943 msg->any.blk_freeblks.bytes = bp->b_bcount;
946 bp->b_flags |= B_ERROR;
948 devstat_end_transaction_buf(&sc->stats, bp);
949 atomic_add_int(&xa_active, -1);
957 * If no msg was allocated we likely could not find a good span.
962 * Message was passed in or constructed.
964 tag->state = msg->state;
965 lockmgr(&sc->lk, LK_RELEASE);
966 kdmsg_msg_write(msg);
967 lockmgr(&sc->lk, LK_EXCLUSIVE);
968 } else if (tag->bio &&
969 (tag->bio->bio_buf->b_flags & B_FAILONDIS) == 0) {
971 * No spans available but BIO is not allowed to fail
972 * on connectivity problems. Requeue the BIO.
974 TAILQ_INSERT_TAIL(&sc->bioq, tag->bio, bio_act);
976 lockmgr(&sc->lk, LK_RELEASE);
978 lockmgr(&sc->lk, LK_EXCLUSIVE);
981 * No spans available, bio is allowed to fail.
983 lockmgr(&sc->lk, LK_RELEASE);
984 tag->status.head.error = DMSG_ERR_IO;
986 lockmgr(&sc->lk, LK_EXCLUSIVE);
991 xa_wait(xa_tag_t *tag)
993 xa_softc_t *sc = tag->sc;
996 lockmgr(&sc->lk, LK_EXCLUSIVE);
998 while (tag->done == 0)
999 lksleep(tag, &sc->lk, 0, "xawait", 0);
1000 lockmgr(&sc->lk, LK_RELEASE);
1002 error = tag->status.head.error;
1010 xa_done(xa_tag_t *tag, int wasbio)
1012 KKASSERT(tag->bio == NULL);
1019 xa_release(tag, wasbio);
1023 * Release a tag. If everything looks ok and there are pending BIOs
1024 * (due to all tags in-use), we can use the tag to start the next BIO.
1025 * Do not try to restart if the connection is currently failed.
1029 xa_release(xa_tag_t *tag, int wasbio)
1031 xa_softc_t *sc = tag->sc;
1034 if ((bio = tag->bio) != NULL) {
1035 struct buf *bp = bio->bio_buf;
1038 bp->b_flags |= B_ERROR;
1039 devstat_end_transaction_buf(&sc->stats, bp);
1040 atomic_add_int(&xa_active, -1);
1045 lockmgr(&sc->lk, LK_EXCLUSIVE);
1047 if (wasbio && sc->open_tag &&
1048 (bio = TAILQ_FIRST(&sc->bioq)) != NULL) {
1049 TAILQ_REMOVE(&sc->bioq, bio, bio_act);
1051 xa_start(tag, NULL, 1);
1053 TAILQ_REMOVE(&sc->tag_pendq, tag, entry);
1054 TAILQ_INSERT_TAIL(&sc->tag_freeq, tag, entry);
1056 lockmgr(&sc->lk, LK_RELEASE);
1060 * Handle messages under the BLKOPEN transaction.
1063 xa_sync_completion(kdmsg_state_t *state, kdmsg_msg_t *msg)
1065 xa_tag_t *tag = state->any.any;
1070 * If the tag has been cleaned out we already closed our side
1071 * of the transaction and we are waiting for the other side to
1074 xa_printf(1, "xa_sync_completion: tag %p msg %08x state %p\n",
1075 tag, msg->any.head.cmd, msg->state);
1078 if (msg->any.head.cmd & DMSGF_CREATE)
1079 kdmsg_state_reply(state, DMSG_ERR_LOSTLINK);
1087 lockmgr(&sc->lk, LK_EXCLUSIVE);
1090 * Handle initial response to our open and restart any deferred
1093 * NOTE: DELETE may also be set.
1095 if (msg->any.head.cmd & DMSGF_CREATE) {
1096 switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
1097 case DMSG_LNK_ERROR | DMSGF_REPLY:
1098 bzero(&tag->status, sizeof(tag->status));
1099 tag->status.head = msg->any.head;
1101 case DMSG_BLK_ERROR | DMSGF_REPLY:
1102 tag->status = msg->any.blk_error;
1105 sc->last_error = tag->status.head.error;
1106 xa_printf(1, "blk_open completion status %d\n",
1108 if (sc->last_error == 0) {
1109 while ((bio = TAILQ_FIRST(&sc->bioq)) != NULL) {
1110 tag = xa_setup_cmd(sc, NULL);
1113 TAILQ_REMOVE(&sc->bioq, bio, bio_act);
1115 xa_start(tag, NULL, 1);
1118 sc->serializing = 0;
1123 * Handle unexpected termination (or lost comm channel) from other
1124 * side. Autonomous completion only if open_tag matches,
1125 * otherwise another thread is probably waiting on the tag.
1127 * (see xa_close() for other interactions)
1129 if (msg->any.head.cmd & DMSGF_DELETE) {
1130 kdmsg_state_reply(tag->state, 0);
1131 if (sc->open_tag == tag) {
1132 sc->open_tag = NULL;
1139 lockmgr(&sc->lk, LK_RELEASE);
1145 xa_bio_completion(kdmsg_state_t *state, kdmsg_msg_t *msg)
1147 xa_tag_t *tag = state->any.any;
1148 xa_softc_t *sc = tag->sc;
1153 * Get the bio from the tag. If no bio is present we just do
1156 if ((bio = tag->bio) == NULL)
1161 * Process return status
1163 switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
1164 case DMSG_LNK_ERROR | DMSGF_REPLY:
1165 bzero(&tag->status, sizeof(tag->status));
1166 tag->status.head = msg->any.head;
1167 if (tag->status.head.error)
1168 tag->status.resid = bp->b_bcount;
1170 tag->status.resid = 0;
1172 case DMSG_BLK_ERROR | DMSGF_REPLY:
1173 tag->status = msg->any.blk_error;
1178 * If the device is open stall the bio on DMSG errors. If an
1179 * actual I/O error occured on the remote device, DMSG_ERR_IO
1182 if (tag->status.head.error &&
1183 (msg->any.head.cmd & DMSGF_DELETE) && sc->opencnt) {
1184 if (tag->status.head.error != DMSG_ERR_IO)
1189 * Process bio completion
1191 * For reads any returned data is zero-extended if necessary, so
1192 * the server can short-cut any all-zeros reads if it desires.
1196 if (msg->aux_data && msg->aux_size) {
1197 if (msg->aux_size < bp->b_bcount) {
1198 bcopy(msg->aux_data, bp->b_data, msg->aux_size);
1199 bzero(bp->b_data + msg->aux_size,
1200 bp->b_bcount - msg->aux_size);
1202 bcopy(msg->aux_data, bp->b_data, bp->b_bcount);
1205 bzero(bp->b_data, bp->b_bcount);
1210 case BUF_CMD_FREEBLKS:
1212 if (tag->status.resid > bp->b_bcount)
1213 tag->status.resid = bp->b_bcount;
1214 bp->b_resid = tag->status.resid;
1215 if (tag->status.head.error != 0) {
1217 bp->b_flags |= B_ERROR;
1221 devstat_end_transaction_buf(&sc->stats, bp);
1222 atomic_add_int(&xa_active, -1);
1229 * Handle completion of the transaction. If the bioq is not empty
1230 * we can initiate another bio on the same tag.
1232 * NOTE: Most of our transactions will be single-message
1233 * CREATE+DELETEs, so we won't have to terminate the
1234 * transaction separately, here. But just in case they
1235 * aren't be sure to terminate the transaction.
1238 if (msg->any.head.cmd & DMSGF_DELETE) {
1240 if ((state->txcmd & DMSGF_DELETE) == 0)
1241 kdmsg_msg_reply(msg, 0);
1246 * Handle the case where the transaction failed due to a
1247 * connectivity issue. The tag is put away with wasbio=0
1248 * and we put the BIO back onto the bioq for a later restart.
1250 * probe I/Os (where the device is not open) will be failed
1251 * instead of requeued.
1255 if (bio->bio_buf->b_flags & B_FAILONDIS) {
1256 xa_printf(1, "xa_strategy: lost link, fail probe bp %p\n",
1258 bio->bio_buf->b_error = ENXIO;
1259 bio->bio_buf->b_flags |= B_ERROR;
1263 xa_printf(1, "xa_strategy: lost link, requeue bp %p\n",
1267 if ((state->txcmd & DMSGF_DELETE) == 0)
1268 kdmsg_msg_reply(msg, 0);
1274 lockmgr(&sc->lk, LK_EXCLUSIVE);
1275 TAILQ_INSERT_TAIL(&sc->bioq, bio, bio_act);
1276 lockmgr(&sc->lk, LK_RELEASE);
1282 * Restart as much deferred I/O as we can. The serializer is set and we
1283 * eat it (clear it) when done.
1285 * Called with sc->lk held
1289 xa_restart_deferred(xa_softc_t *sc)
1291 kdmsg_state_t *span;
1296 KKASSERT(sc->serializing);
1299 * Determine if a restart is needed.
1301 if (sc->opencnt == 0) {
1303 * Device is not open, nothing to do, eat serializing.
1305 sc->serializing = 0;
1307 } else if (sc->open_tag == NULL) {
1309 * BLK_OPEN required before we can restart any BIOs.
1310 * Select the best LNK_SPAN to issue the BLK_OPEN under.
1312 * serializing interlocks waiting open()s.
1315 TAILQ_FOREACH(span, &sc->spanq, user_entry) {
1316 if ((span->rxcmd & DMSGF_DELETE) == 0)
1323 tag = xa_setup_cmd(sc, NULL);
1329 msg = kdmsg_msg_alloc(span,
1332 xa_sync_completion, tag);
1333 msg->any.blk_open.modes = DMSG_BLKOPEN_RD;
1335 "BLK_OPEN tag %p state %p "
1337 tag, msg->state, span);
1338 xa_start(tag, msg, 0);
1341 sc->serializing = 0;
1344 /* else leave serializing set until BLK_OPEN response */
1347 sc->serializing = 0;