Home | History | Annotate | Line # | Download | only in xen
xbd_xenbus.c revision 1.74
      1 /*      $NetBSD: xbd_xenbus.c,v 1.73 2015/08/27 05:51:50 mlelstv Exp $      */
      2 
      3 /*
      4  * Copyright (c) 2006 Manuel Bouyer.
      5  *
      6  * Redistribution and use in source and binary forms, with or without
      7  * modification, are permitted provided that the following conditions
      8  * are met:
      9  * 1. Redistributions of source code must retain the above copyright
     10  *    notice, this list of conditions and the following disclaimer.
     11  * 2. Redistributions in binary form must reproduce the above copyright
     12  *    notice, this list of conditions and the following disclaimer in the
     13  *    documentation and/or other materials provided with the distribution.
     14  *
     15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     25  *
     26  */
     27 
     28 /*
     29  * The file contains the xbd frontend code required for block-level
     30  * communications (similar to hard disks) between two Xen domains.
     31  *
     32  * We are not supposed to receive solicitations spontaneously from backend. The
     33  * protocol is therefore fairly simple and uses only one ring to communicate
     34  * with backend: frontend posts requests to the ring then wait for their
     35  * replies asynchronously.
     36  *
     37  * xbd follows NetBSD's disk(9) convention. At any time, a LWP can schedule
     38  * an operation request for the device (be it open(), read(), write(), ...).
     39  * Calls are typically processed that way:
     40  * - initiate request: xbdread/write/open/ioctl/..
     41  * - depending on operation, it is handled directly by disk(9) subsystem or
     42  *   goes through physio(9) first.
     43  * - the request is ultimately processed by xbd_diskstart() that prepares the
     44  *   xbd requests, post them in the ring I/O queue, then signal the backend.
     45  *
     46  * When a response is available in the queue, the backend signals the frontend
     47  * via its event channel. This triggers xbd_handler(), which will link back
     48  * the response to its request through the request ID, and mark the I/O as
     49  * completed.
     50  */
     51 
     52 #include <sys/cdefs.h>
     53 __KERNEL_RCSID(0, "$NetBSD: xbd_xenbus.c,v 1.73 2015/08/27 05:51:50 mlelstv Exp $");
     54 
     55 #include "opt_xen.h"
     56 
     57 
     58 #include <sys/param.h>
     59 #include <sys/buf.h>
     60 #include <sys/bufq.h>
     61 #include <sys/device.h>
     62 #include <sys/disk.h>
     63 #include <sys/disklabel.h>
     64 #include <sys/conf.h>
     65 #include <sys/fcntl.h>
     66 #include <sys/kernel.h>
     67 #include <sys/proc.h>
     68 #include <sys/systm.h>
     69 #include <sys/stat.h>
     70 #include <sys/vnode.h>
     71 
     72 #include <dev/dkvar.h>
     73 
     74 #include <uvm/uvm.h>
     75 
     76 #include <xen/hypervisor.h>
     77 #include <xen/evtchn.h>
     78 #include <xen/granttables.h>
     79 #include <xen/xen-public/io/blkif.h>
     80 #include <xen/xen-public/io/protocols.h>
     81 
     82 #include <xen/xenbus.h>
     83 #include "locators.h"
     84 
     85 #undef XBD_DEBUG
     86 #ifdef XBD_DEBUG
     87 #define DPRINTF(x) printf x;
     88 #else
     89 #define DPRINTF(x)
     90 #endif
     91 
     92 #define GRANT_INVALID_REF -1
     93 
     94 #define XBD_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
     95 #define XBD_MAX_XFER (PAGE_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST)
     96 
     97 #define XEN_BSHIFT      9               /* log2(XEN_BSIZE) */
     98 #define XEN_BSIZE       (1 << XEN_BSHIFT)
     99 
    100 struct xbd_req {
    101 	SLIST_ENTRY(xbd_req) req_next;
    102 	uint16_t req_id; /* ID passed to backend */
    103 	union {
    104 	    struct {
    105 		grant_ref_t req_gntref[BLKIF_MAX_SEGMENTS_PER_REQUEST];
    106 		int req_nr_segments; /* number of segments in this request */
    107 		struct buf *req_bp; /* buffer associated with this request */
    108 		void *req_data; /* pointer to the data buffer */
    109 	    } req_rw;
    110 	    struct {
    111 		int s_error;
    112 		volatile int s_done;
    113 	    } req_sync;
    114 	} u;
    115 };
    116 #define req_gntref	u.req_rw.req_gntref
    117 #define req_nr_segments	u.req_rw.req_nr_segments
    118 #define req_bp		u.req_rw.req_bp
    119 #define req_data	u.req_rw.req_data
    120 #define req_sync	u.req_sync
    121 
    122 struct xbd_xenbus_softc {
    123 	struct dk_softc sc_dksc;	/* Must be first in this struct */
    124 	struct xenbus_device *sc_xbusd;
    125 
    126 	blkif_front_ring_t sc_ring;
    127 
    128 	unsigned int sc_evtchn;
    129 
    130 	grant_ref_t sc_ring_gntref;
    131 
    132 	struct xbd_req sc_reqs[XBD_RING_SIZE];
    133 	SLIST_HEAD(,xbd_req) sc_xbdreq_head; /* list of free requests */
    134 	bool sc_xbdreq_wait; /* special waiting on xbd_req */
    135 
    136 	int sc_backend_status; /* our status with backend */
    137 #define BLKIF_STATE_DISCONNECTED 0
    138 #define BLKIF_STATE_CONNECTED    1
    139 #define BLKIF_STATE_SUSPENDED    2
    140 
    141 	int sc_shutdown;
    142 #define BLKIF_SHUTDOWN_RUN    0 /* no shutdown */
    143 #define BLKIF_SHUTDOWN_REMOTE 1 /* backend-initiated shutdown in progress */
    144 #define BLKIF_SHUTDOWN_LOCAL  2 /* locally-initiated shutdown in progress */
    145 
    146 	uint64_t sc_sectors; /* number of sectors for this device */
    147 	u_long sc_secsize; /* sector size */
    148 	uint64_t sc_xbdsize; /* size of disk in DEV_BSIZE */
    149 	u_long sc_info; /* VDISK_* */
    150 	u_long sc_handle; /* from backend */
    151 	int sc_cache_flush; /* backend supports BLKIF_OP_FLUSH_DISKCACHE */
    152 };
    153 
    154 #if 0
    155 /* too big to be on stack */
    156 static multicall_entry_t rq_mcl[XBD_RING_SIZE+1];
    157 static paddr_t rq_pages[XBD_RING_SIZE];
    158 #endif
    159 
    160 static int  xbd_xenbus_match(device_t, cfdata_t, void *);
    161 static void xbd_xenbus_attach(device_t, device_t, void *);
    162 static int  xbd_xenbus_detach(device_t, int);
    163 
    164 static bool xbd_xenbus_suspend(device_t, const pmf_qual_t *);
    165 static bool xbd_xenbus_resume(device_t, const pmf_qual_t *);
    166 
    167 static int  xbd_handler(void *);
    168 static int  xbd_diskstart(device_t, struct buf *);
    169 static void xbd_backend_changed(void *, XenbusState);
    170 static void xbd_connect(struct xbd_xenbus_softc *);
    171 
    172 static int  xbd_map_align(struct xbd_req *);
    173 static void xbd_unmap_align(struct xbd_req *);
    174 
    175 static void xbdminphys(struct buf *);
    176 
    177 CFATTACH_DECL3_NEW(xbd, sizeof(struct xbd_xenbus_softc),
    178     xbd_xenbus_match, xbd_xenbus_attach, xbd_xenbus_detach, NULL, NULL, NULL,
    179     DVF_DETACH_SHUTDOWN);
    180 
    181 dev_type_open(xbdopen);
    182 dev_type_close(xbdclose);
    183 dev_type_read(xbdread);
    184 dev_type_write(xbdwrite);
    185 dev_type_ioctl(xbdioctl);
    186 dev_type_strategy(xbdstrategy);
    187 dev_type_dump(xbddump);
    188 dev_type_size(xbdsize);
    189 
    190 const struct bdevsw xbd_bdevsw = {
    191 	.d_open = xbdopen,
    192 	.d_close = xbdclose,
    193 	.d_strategy = xbdstrategy,
    194 	.d_ioctl = xbdioctl,
    195 	.d_dump = xbddump,
    196 	.d_psize = xbdsize,
    197 	.d_discard = nodiscard,
    198 	.d_flag = D_DISK
    199 };
    200 
    201 const struct cdevsw xbd_cdevsw = {
    202 	.d_open = xbdopen,
    203 	.d_close = xbdclose,
    204 	.d_read = xbdread,
    205 	.d_write = xbdwrite,
    206 	.d_ioctl = xbdioctl,
    207 	.d_stop = nostop,
    208 	.d_tty = notty,
    209 	.d_poll = nopoll,
    210 	.d_mmap = nommap,
    211 	.d_kqfilter = nokqfilter,
    212 	.d_discard = nodiscard,
    213 	.d_flag = D_DISK
    214 };
    215 
    216 extern struct cfdriver xbd_cd;
    217 
    218 static struct dkdriver xbddkdriver = {
    219         .d_strategy = xbdstrategy,
    220 	.d_minphys = xbdminphys,
    221 	.d_open = xbdopen,
    222 	.d_close = xbdclose,
    223 	.d_diskstart = xbd_diskstart,
    224 };
    225 
    226 static int
    227 xbd_xenbus_match(device_t parent, cfdata_t match, void *aux)
    228 {
    229 	struct xenbusdev_attach_args *xa = aux;
    230 
    231 	if (strcmp(xa->xa_type, "vbd") != 0)
    232 		return 0;
    233 
    234 	if (match->cf_loc[XENBUSCF_ID] != XENBUSCF_ID_DEFAULT &&
    235 	    match->cf_loc[XENBUSCF_ID] != xa->xa_id)
    236 		return 0;
    237 
    238 	return 1;
    239 }
    240 
    241 static void
    242 xbd_xenbus_attach(device_t parent, device_t self, void *aux)
    243 {
    244 	struct xbd_xenbus_softc *sc = device_private(self);
    245 	struct xenbusdev_attach_args *xa = aux;
    246 	blkif_sring_t *ring;
    247 	RING_IDX i;
    248 #ifdef XBD_DEBUG
    249 	char **dir, *val;
    250 	int dir_n = 0;
    251 	char id_str[20];
    252 	int err;
    253 #endif
    254 
    255 	config_pending_incr(self);
    256 	aprint_normal(": Xen Virtual Block Device Interface\n");
    257 
    258 	dk_init(&sc->sc_dksc, self, DKTYPE_ESDI);
    259 	disk_init(&sc->sc_dksc.sc_dkdev, device_xname(self), &xbddkdriver);
    260 
    261 #ifdef XBD_DEBUG
    262 	printf("path: %s\n", xa->xa_xbusd->xbusd_path);
    263 	snprintf(id_str, sizeof(id_str), "%d", xa->xa_id);
    264 	err = xenbus_directory(NULL, "device/vbd", id_str, &dir_n, &dir);
    265 	if (err) {
    266 		aprint_error_dev(self, "xenbus_directory err %d\n", err);
    267 	} else {
    268 		printf("%s/\n", xa->xa_xbusd->xbusd_path);
    269 		for (i = 0; i < dir_n; i++) {
    270 			printf("\t/%s", dir[i]);
    271 			err = xenbus_read(NULL, xa->xa_xbusd->xbusd_path,
    272 					  dir[i], NULL, &val);
    273 			if (err) {
    274 				aprint_error_dev(self, "xenbus_read err %d\n",
    275 						 err);
    276 			} else {
    277 				printf(" = %s\n", val);
    278 				free(val, M_DEVBUF);
    279 			}
    280 		}
    281 	}
    282 #endif /* XBD_DEBUG */
    283 	sc->sc_xbusd = xa->xa_xbusd;
    284 	sc->sc_xbusd->xbusd_otherend_changed = xbd_backend_changed;
    285 
    286 	/* initialize free requests list */
    287 	SLIST_INIT(&sc->sc_xbdreq_head);
    288 	for (i = 0; i < XBD_RING_SIZE; i++) {
    289 		sc->sc_reqs[i].req_id = i;
    290 		SLIST_INSERT_HEAD(&sc->sc_xbdreq_head, &sc->sc_reqs[i],
    291 		    req_next);
    292 	}
    293 
    294 	sc->sc_backend_status = BLKIF_STATE_DISCONNECTED;
    295 	sc->sc_shutdown = BLKIF_SHUTDOWN_REMOTE;
    296 
    297 	ring = (void *)uvm_km_alloc(kernel_map, PAGE_SIZE, 0, UVM_KMF_WIRED);
    298 	if (ring == NULL)
    299 		panic("%s: can't alloc ring", device_xname(self));
    300 	sc->sc_ring.sring = ring;
    301 
    302 	/* resume shared structures and tell backend that we are ready */
    303 	if (xbd_xenbus_resume(self, PMF_Q_NONE) == false) {
    304 		uvm_km_free(kernel_map, (vaddr_t)ring, PAGE_SIZE,
    305 		    UVM_KMF_WIRED);
    306 		return;
    307 	}
    308 
    309 	if (!pmf_device_register(self, xbd_xenbus_suspend, xbd_xenbus_resume))
    310 		aprint_error_dev(self, "couldn't establish power handler\n");
    311 
    312 }
    313 
    314 static int
    315 xbd_xenbus_detach(device_t dev, int flags)
    316 {
    317 	struct xbd_xenbus_softc *sc = device_private(dev);
    318 	int bmaj, cmaj, i, mn, rc, s;
    319 
    320 	rc = disk_begindetach(&sc->sc_dksc.sc_dkdev, NULL, dev, flags);
    321 	if (rc != 0)
    322 		return rc;
    323 
    324 	s = splbio();
    325 	DPRINTF(("%s: xbd_detach\n", device_xname(dev)));
    326 	if (sc->sc_shutdown == BLKIF_SHUTDOWN_RUN) {
    327 		sc->sc_shutdown = BLKIF_SHUTDOWN_LOCAL;
    328 		/* wait for requests to complete */
    329 		while (sc->sc_backend_status == BLKIF_STATE_CONNECTED &&
    330 		    sc->sc_dksc.sc_dkdev.dk_stats->io_busy > 0)
    331 			tsleep(xbd_xenbus_detach, PRIBIO, "xbddetach", hz/2);
    332 
    333 		xenbus_switch_state(sc->sc_xbusd, NULL, XenbusStateClosing);
    334 	}
    335 	if ((flags & DETACH_FORCE) == 0) {
    336 		/* xbd_xenbus_detach already in progress */
    337 		wakeup(xbd_xenbus_detach);
    338 		splx(s);
    339 		return EALREADY;
    340 	}
    341 	while (xenbus_read_driver_state(sc->sc_xbusd->xbusd_otherend)
    342 	    != XenbusStateClosed)
    343 		tsleep(xbd_xenbus_detach, PRIBIO, "xbddetach2", hz/2);
    344 	splx(s);
    345 
    346 	/* locate the major number */
    347 	bmaj = bdevsw_lookup_major(&xbd_bdevsw);
    348 	cmaj = cdevsw_lookup_major(&xbd_cdevsw);
    349 
    350 	/* Nuke the vnodes for any open instances. */
    351 	for (i = 0; i < MAXPARTITIONS; i++) {
    352 		mn = DISKMINOR(device_unit(dev), i);
    353 		vdevgone(bmaj, mn, mn, VBLK);
    354 		vdevgone(cmaj, mn, mn, VCHR);
    355 	}
    356 	if (sc->sc_backend_status == BLKIF_STATE_CONNECTED) {
    357 		/* Delete all of our wedges. */
    358 		dkwedge_delall(&sc->sc_dksc.sc_dkdev);
    359 
    360 		/* Kill off any queued buffers. */
    361 		dk_drain(&sc->sc_dksc);
    362 		bufq_free(sc->sc_dksc.sc_bufq);
    363 
    364 		/* detach disk */
    365 		disk_detach(&sc->sc_dksc.sc_dkdev);
    366 		disk_destroy(&sc->sc_dksc.sc_dkdev);
    367 		dk_detach(&sc->sc_dksc);
    368 	}
    369 
    370 	hypervisor_mask_event(sc->sc_evtchn);
    371 	event_remove_handler(sc->sc_evtchn, &xbd_handler, sc);
    372 	while (xengnt_status(sc->sc_ring_gntref)) {
    373 		tsleep(xbd_xenbus_detach, PRIBIO, "xbd_ref", hz/2);
    374 	}
    375 	xengnt_revoke_access(sc->sc_ring_gntref);
    376 	uvm_km_free(kernel_map, (vaddr_t)sc->sc_ring.sring,
    377 	    PAGE_SIZE, UVM_KMF_WIRED);
    378 
    379 	pmf_device_deregister(dev);
    380 
    381 	return 0;
    382 }
    383 
    384 static bool
    385 xbd_xenbus_suspend(device_t dev, const pmf_qual_t *qual) {
    386 
    387 	int s;
    388 	struct xbd_xenbus_softc *sc;
    389 
    390 	sc = device_private(dev);
    391 
    392 	s = splbio();
    393 	/* wait for requests to complete, then suspend device */
    394 	while (sc->sc_backend_status == BLKIF_STATE_CONNECTED &&
    395 	    sc->sc_dksc.sc_dkdev.dk_stats->io_busy > 0)
    396 		tsleep(xbd_xenbus_suspend, PRIBIO, "xbdsuspend", hz/2);
    397 
    398 	hypervisor_mask_event(sc->sc_evtchn);
    399 	sc->sc_backend_status = BLKIF_STATE_SUSPENDED;
    400 	event_remove_handler(sc->sc_evtchn, xbd_handler, sc);
    401 
    402 	splx(s);
    403 
    404 	xenbus_device_suspend(sc->sc_xbusd);
    405 	aprint_verbose_dev(dev, "removed event channel %d\n", sc->sc_evtchn);
    406 
    407 	return true;
    408 }
    409 
    410 static bool
    411 xbd_xenbus_resume(device_t dev, const pmf_qual_t *qual)
    412 {
    413 	struct xbd_xenbus_softc *sc;
    414 	struct xenbus_transaction *xbt;
    415 	int error;
    416 	blkif_sring_t *ring;
    417 	paddr_t ma;
    418 	const char *errmsg;
    419 
    420 	sc = device_private(dev);
    421 
    422 	if (sc->sc_backend_status == BLKIF_STATE_SUSPENDED) {
    423 		/*
    424 		 * Device was suspended, so ensure that access associated to
    425 		 * the block I/O ring is revoked.
    426 		 */
    427 		xengnt_revoke_access(sc->sc_ring_gntref);
    428 	}
    429 	sc->sc_ring_gntref = GRANT_INVALID_REF;
    430 
    431 	/* Initialize ring */
    432 	ring = sc->sc_ring.sring;
    433 	memset(ring, 0, PAGE_SIZE);
    434 	SHARED_RING_INIT(ring);
    435 	FRONT_RING_INIT(&sc->sc_ring, ring, PAGE_SIZE);
    436 
    437 	/*
    438 	 * get MA address of the ring, and use it to set up the grant entry
    439 	 * for the block device
    440 	 */
    441 	(void)pmap_extract_ma(pmap_kernel(), (vaddr_t)ring, &ma);
    442 	error = xenbus_grant_ring(sc->sc_xbusd, ma, &sc->sc_ring_gntref);
    443 	if (error)
    444 		goto abort_resume;
    445 
    446 	error = xenbus_alloc_evtchn(sc->sc_xbusd, &sc->sc_evtchn);
    447 	if (error)
    448 		goto abort_resume;
    449 
    450 	aprint_verbose_dev(dev, "using event channel %d\n",
    451 	    sc->sc_evtchn);
    452 	event_set_handler(sc->sc_evtchn, &xbd_handler, sc,
    453 	    IPL_BIO, device_xname(dev));
    454 
    455 again:
    456 	xbt = xenbus_transaction_start();
    457 	if (xbt == NULL)
    458 		return false;
    459 	error = xenbus_printf(xbt, sc->sc_xbusd->xbusd_path,
    460 	    "ring-ref","%u", sc->sc_ring_gntref);
    461 	if (error) {
    462 		errmsg = "writing ring-ref";
    463 		goto abort_transaction;
    464 	}
    465 	error = xenbus_printf(xbt, sc->sc_xbusd->xbusd_path,
    466 	    "event-channel", "%u", sc->sc_evtchn);
    467 	if (error) {
    468 		errmsg = "writing event channel";
    469 		goto abort_transaction;
    470 	}
    471 	error = xenbus_printf(xbt, sc->sc_xbusd->xbusd_path,
    472 	    "protocol", "%s", XEN_IO_PROTO_ABI_NATIVE);
    473 	if (error) {
    474 		errmsg = "writing protocol";
    475 		goto abort_transaction;
    476 	}
    477 	error = xenbus_transaction_end(xbt, 0);
    478 	if (error == EAGAIN)
    479 		goto again;
    480 	if (error != 0) {
    481 		xenbus_dev_fatal(sc->sc_xbusd, error,
    482 		    "completing transaction");
    483 		return false;
    484 	}
    485 
    486 	xenbus_switch_state(sc->sc_xbusd, NULL, XenbusStateInitialised);
    487 
    488 	if (sc->sc_backend_status == BLKIF_STATE_SUSPENDED) {
    489 		/*
    490 		 * device was suspended, softc structures are
    491 		 * already initialized - we use a shortcut
    492 		 */
    493 		sc->sc_backend_status = BLKIF_STATE_CONNECTED;
    494 		xenbus_device_resume(sc->sc_xbusd);
    495 		hypervisor_enable_event(sc->sc_evtchn);
    496 		xenbus_switch_state(sc->sc_xbusd, NULL, XenbusStateConnected);
    497 	}
    498 
    499 	return true;
    500 
    501 abort_resume:
    502 	xenbus_dev_fatal(sc->sc_xbusd, error, "resuming device");
    503 	return false;
    504 
    505 abort_transaction:
    506 	xenbus_transaction_end(xbt, 1);
    507 	xenbus_dev_fatal(sc->sc_xbusd, error, "%s", errmsg);
    508 	return false;
    509 }
    510 
    511 static void xbd_backend_changed(void *arg, XenbusState new_state)
    512 {
    513 	struct xbd_xenbus_softc *sc = device_private((device_t)arg);
    514 	struct disk_geom *dg;
    515 
    516 	char buf[9];
    517 	int s;
    518 	DPRINTF(("%s: new backend state %d\n",
    519 	    device_xname(sc->sc_dksc.sc_dev), new_state));
    520 
    521 	switch (new_state) {
    522 	case XenbusStateUnknown:
    523 	case XenbusStateInitialising:
    524 	case XenbusStateInitWait:
    525 	case XenbusStateInitialised:
    526 		break;
    527 	case XenbusStateClosing:
    528 		s = splbio();
    529 		if (sc->sc_shutdown == BLKIF_SHUTDOWN_RUN)
    530 			sc->sc_shutdown = BLKIF_SHUTDOWN_REMOTE;
    531 		/* wait for requests to complete */
    532 		while (sc->sc_backend_status == BLKIF_STATE_CONNECTED &&
    533 		    sc->sc_dksc.sc_dkdev.dk_stats->io_busy > 0)
    534 			tsleep(xbd_xenbus_detach, PRIBIO, "xbddetach", hz/2);
    535 		splx(s);
    536 		xenbus_switch_state(sc->sc_xbusd, NULL, XenbusStateClosed);
    537 		break;
    538 	case XenbusStateConnected:
    539 		/*
    540 		 * note that xbd_backend_changed() can only be called by
    541 		 * the xenbus thread.
    542 		 */
    543 
    544 		if (sc->sc_backend_status == BLKIF_STATE_CONNECTED ||
    545 		    sc->sc_backend_status == BLKIF_STATE_SUSPENDED)
    546 			/* already connected */
    547 			return;
    548 
    549 		xbd_connect(sc);
    550 		sc->sc_shutdown = BLKIF_SHUTDOWN_RUN;
    551 		hypervisor_enable_event(sc->sc_evtchn);
    552 
    553 		sc->sc_xbdsize =
    554 		    sc->sc_sectors * (uint64_t)sc->sc_secsize / DEV_BSIZE;
    555 		dg = &sc->sc_dksc.sc_dkdev.dk_geom;
    556 		memset(dg, 0, sizeof(*dg));
    557 
    558 		dg->dg_secperunit = sc->sc_xbdsize;
    559 		dg->dg_secsize = DEV_BSIZE;
    560 		dg->dg_ntracks = 1;
    561 		// XXX: Ok to hard-code DEV_BSIZE?
    562 		dg->dg_nsectors = 1024 * (1024 / dg->dg_secsize);
    563 		dg->dg_ncylinders = dg->dg_secperunit / dg->dg_nsectors;
    564 
    565 		bufq_alloc(&sc->sc_dksc.sc_bufq, "fcfs", 0);
    566 		dk_attach(&sc->sc_dksc);
    567 		disk_attach(&sc->sc_dksc.sc_dkdev);
    568 
    569 		sc->sc_backend_status = BLKIF_STATE_CONNECTED;
    570 
    571 		/* try to read the disklabel */
    572 		dk_getdisklabel(&sc->sc_dksc, 0 /* XXX ? */);
    573 		format_bytes(buf, sizeof(buf), sc->sc_sectors * sc->sc_secsize);
    574 		aprint_verbose_dev(sc->sc_dksc.sc_dev,
    575 				"%s, %d bytes/sect x %" PRIu64 " sectors\n",
    576 				buf, (int)dg->dg_secsize, sc->sc_xbdsize);
    577 		/* Discover wedges on this disk. */
    578 		dkwedge_discover(&sc->sc_dksc.sc_dkdev);
    579 
    580 		disk_set_info(sc->sc_dksc.sc_dev, &sc->sc_dksc.sc_dkdev, NULL);
    581 
    582 		/* the disk should be working now */
    583 		config_pending_decr(sc->sc_dksc.sc_dev);
    584 		break;
    585 	default:
    586 		panic("bad backend state %d", new_state);
    587 	}
    588 }
    589 
    590 static void
    591 xbd_connect(struct xbd_xenbus_softc *sc)
    592 {
    593 	int err;
    594 	unsigned long long sectors;
    595 	u_long cache_flush;
    596 
    597 	err = xenbus_read_ul(NULL,
    598 	    sc->sc_xbusd->xbusd_path, "virtual-device", &sc->sc_handle, 10);
    599 	if (err)
    600 		panic("%s: can't read number from %s/virtual-device\n",
    601 		    device_xname(sc->sc_dksc.sc_dev),
    602 		    sc->sc_xbusd->xbusd_otherend);
    603 	err = xenbus_read_ull(NULL,
    604 	    sc->sc_xbusd->xbusd_otherend, "sectors", &sectors, 10);
    605 	if (err)
    606 		panic("%s: can't read number from %s/sectors\n",
    607 		    device_xname(sc->sc_dksc.sc_dev),
    608 		    sc->sc_xbusd->xbusd_otherend);
    609 	sc->sc_sectors = sectors;
    610 
    611 	err = xenbus_read_ul(NULL,
    612 	    sc->sc_xbusd->xbusd_otherend, "info", &sc->sc_info, 10);
    613 	if (err)
    614 		panic("%s: can't read number from %s/info\n",
    615 		    device_xname(sc->sc_dksc.sc_dev),
    616 		    sc->sc_xbusd->xbusd_otherend);
    617 	err = xenbus_read_ul(NULL,
    618 	    sc->sc_xbusd->xbusd_otherend, "sector-size", &sc->sc_secsize, 10);
    619 	if (err)
    620 		panic("%s: can't read number from %s/sector-size\n",
    621 		    device_xname(sc->sc_dksc.sc_dev),
    622 		    sc->sc_xbusd->xbusd_otherend);
    623 
    624 	err = xenbus_read_ul(NULL, sc->sc_xbusd->xbusd_otherend,
    625 	    "feature-flush-cache", &cache_flush, 10);
    626 	if (err)
    627 		cache_flush = 0;
    628 	if (cache_flush > 0)
    629 		sc->sc_cache_flush = 1;
    630 	else
    631 		sc->sc_cache_flush = 0;
    632 
    633 	xenbus_switch_state(sc->sc_xbusd, NULL, XenbusStateConnected);
    634 }
    635 
    636 static int
    637 xbd_handler(void *arg)
    638 {
    639 	struct xbd_xenbus_softc *sc = arg;
    640 	struct buf *bp;
    641 	RING_IDX resp_prod, i;
    642 	int more_to_do;
    643 	int seg;
    644 
    645 	DPRINTF(("xbd_handler(%s)\n", device_xname(sc->sc_dksc.sc_dev)));
    646 
    647 	if (__predict_false(sc->sc_backend_status != BLKIF_STATE_CONNECTED))
    648 		return 0;
    649 again:
    650 	resp_prod = sc->sc_ring.sring->rsp_prod;
    651 	xen_rmb(); /* ensure we see replies up to resp_prod */
    652 	for (i = sc->sc_ring.rsp_cons; i != resp_prod; i++) {
    653 		blkif_response_t *rep = RING_GET_RESPONSE(&sc->sc_ring, i);
    654 		struct xbd_req *xbdreq = &sc->sc_reqs[rep->id];
    655 		DPRINTF(("xbd_handler(%p): b_bcount = %ld\n",
    656 		    xbdreq->req_bp, (long)bp->b_bcount));
    657 		bp = xbdreq->req_bp;
    658 		if (rep->operation == BLKIF_OP_FLUSH_DISKCACHE) {
    659 			xbdreq->req_sync.s_error = rep->status;
    660 			xbdreq->req_sync.s_done = 1;
    661 			wakeup(xbdreq);
    662 			/* caller will free the req */
    663 			continue;
    664 		}
    665 		for (seg = xbdreq->req_nr_segments - 1; seg >= 0; seg--) {
    666 			if (__predict_false(
    667 			    xengnt_status(xbdreq->req_gntref[seg]))) {
    668 				aprint_verbose_dev(sc->sc_dksc.sc_dev,
    669 					"grant still used by backend\n");
    670 				sc->sc_ring.rsp_cons = i;
    671 				xbdreq->req_nr_segments = seg + 1;
    672 				goto done;
    673 			}
    674 			xengnt_revoke_access(xbdreq->req_gntref[seg]);
    675 			xbdreq->req_nr_segments--;
    676 		}
    677 		if (rep->operation != BLKIF_OP_READ &&
    678 		    rep->operation != BLKIF_OP_WRITE) {
    679 			aprint_error_dev(sc->sc_dksc.sc_dev,
    680 			    "bad operation %d from backend\n", rep->operation);
    681 			bp->b_error = EIO;
    682 			bp->b_resid = bp->b_bcount;
    683 			goto next;
    684 		}
    685 		if (rep->status != BLKIF_RSP_OKAY) {
    686 			bp->b_error = EIO;
    687 			bp->b_resid = bp->b_bcount;
    688 			goto next;
    689 		}
    690 		/* b_resid was set in dk_start */
    691 next:
    692 		if (bp->b_data != xbdreq->req_data)
    693 			xbd_unmap_align(xbdreq);
    694 
    695 		dk_done(&sc->sc_dksc, bp);
    696 
    697 		SLIST_INSERT_HEAD(&sc->sc_xbdreq_head, xbdreq, req_next);
    698 	}
    699 done:
    700 	xen_rmb();
    701 	sc->sc_ring.rsp_cons = i;
    702 
    703 	RING_FINAL_CHECK_FOR_RESPONSES(&sc->sc_ring, more_to_do);
    704 	if (more_to_do)
    705 		goto again;
    706 
    707 	if (sc->sc_xbdreq_wait)
    708 		wakeup(&sc->sc_xbdreq_wait);
    709 	else
    710 		dk_start(&sc->sc_dksc, NULL);
    711 	return 1;
    712 }
    713 
    714 static void
    715 xbdminphys(struct buf *bp)
    716 {
    717 	if (bp->b_bcount > XBD_MAX_XFER) {
    718 		bp->b_bcount = XBD_MAX_XFER;
    719 	}
    720 	minphys(bp);
    721 }
    722 
    723 int
    724 xbdopen(dev_t dev, int flags, int fmt, struct lwp *l)
    725 {
    726 	struct	xbd_xenbus_softc *sc;
    727 
    728 	sc = device_lookup_private(&xbd_cd, DISKUNIT(dev));
    729 	if (sc == NULL)
    730 		return (ENXIO);
    731 	if ((flags & FWRITE) && (sc->sc_info & VDISK_READONLY))
    732 		return EROFS;
    733 
    734 	DPRINTF(("xbdopen(0x%04x, %d)\n", dev, flags));
    735 	return dk_open(&sc->sc_dksc, dev, flags, fmt, l);
    736 }
    737 
    738 int
    739 xbdclose(dev_t dev, int flags, int fmt, struct lwp *l)
    740 {
    741 	struct xbd_xenbus_softc *sc;
    742 
    743 	sc = device_lookup_private(&xbd_cd, DISKUNIT(dev));
    744 
    745 	DPRINTF(("xbdclose(%d, %d)\n", dev, flags));
    746 	return dk_close(&sc->sc_dksc, dev, flags, fmt, l);
    747 }
    748 
    749 void
    750 xbdstrategy(struct buf *bp)
    751 {
    752 	struct xbd_xenbus_softc *sc;
    753 
    754 	sc = device_lookup_private(&xbd_cd, DISKUNIT(bp->b_dev));
    755 
    756 	DPRINTF(("xbdstrategy(%p): b_bcount = %ld\n", bp,
    757 	    (long)bp->b_bcount));
    758 
    759 	if (sc == NULL || sc->sc_shutdown != BLKIF_SHUTDOWN_RUN) {
    760 		bp->b_error = EIO;
    761 		biodone(bp);
    762 		return;
    763 	}
    764 	if (__predict_false((sc->sc_info & VDISK_READONLY) &&
    765 	    (bp->b_flags & B_READ) == 0)) {
    766 		bp->b_error = EROFS;
    767 		biodone(bp);
    768 		return;
    769 	}
    770 
    771 	dk_strategy(&sc->sc_dksc, bp);
    772 	return;
    773 }
    774 
    775 int
    776 xbdsize(dev_t dev)
    777 {
    778 	struct	xbd_xenbus_softc *sc;
    779 
    780 	DPRINTF(("xbdsize(%d)\n", dev));
    781 
    782 	sc = device_lookup_private(&xbd_cd, DISKUNIT(dev));
    783 	if (sc == NULL || sc->sc_shutdown != BLKIF_SHUTDOWN_RUN)
    784 		return -1;
    785 	return dk_size(&sc->sc_dksc, dev);
    786 }
    787 
    788 int
    789 xbdread(dev_t dev, struct uio *uio, int flags)
    790 {
    791 	struct xbd_xenbus_softc *sc =
    792 	    device_lookup_private(&xbd_cd, DISKUNIT(dev));
    793 	struct  dk_softc *dksc = &sc->sc_dksc;
    794 
    795 	if (!DK_ATTACHED(dksc))
    796 		return ENXIO;
    797 	return physio(xbdstrategy, NULL, dev, B_READ, xbdminphys, uio);
    798 }
    799 
    800 int
    801 xbdwrite(dev_t dev, struct uio *uio, int flags)
    802 {
    803 	struct xbd_xenbus_softc *sc =
    804 	    device_lookup_private(&xbd_cd, DISKUNIT(dev));
    805 	struct  dk_softc *dksc = &sc->sc_dksc;
    806 
    807 	if (!DK_ATTACHED(dksc))
    808 		return ENXIO;
    809 	if (__predict_false(sc->sc_info & VDISK_READONLY))
    810 		return EROFS;
    811 	return physio(xbdstrategy, NULL, dev, B_WRITE, xbdminphys, uio);
    812 }
    813 
    814 int
    815 xbdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    816 {
    817 	struct xbd_xenbus_softc *sc =
    818 	    device_lookup_private(&xbd_cd, DISKUNIT(dev));
    819 	struct	dk_softc *dksc;
    820 	int	error;
    821 	int s;
    822 	struct xbd_req *xbdreq;
    823 	blkif_request_t *req;
    824 	int notify;
    825 
    826 	DPRINTF(("xbdioctl(%d, %08lx, %p, %d, %p)\n",
    827 	    dev, cmd, data, flag, l));
    828 	dksc = &sc->sc_dksc;
    829 
    830 	switch (cmd) {
    831 	case DIOCSSTRATEGY:
    832 		error = EOPNOTSUPP;
    833 		break;
    834 	case DIOCCACHESYNC:
    835 		if (sc->sc_cache_flush <= 0) {
    836 			if (sc->sc_cache_flush == 0) {
    837 				aprint_error_dev(sc->sc_dksc.sc_dev,
    838 				    "WARNING: cache flush not supported "
    839 				    "by backend\n");
    840 				sc->sc_cache_flush = -1;
    841 			}
    842 			return EOPNOTSUPP;
    843 		}
    844 
    845 		s = splbio();
    846 
    847 		while (RING_FULL(&sc->sc_ring)) {
    848 			sc->sc_xbdreq_wait = 1;
    849 			tsleep(&sc->sc_xbdreq_wait, PRIBIO, "xbdreq", 0);
    850 		}
    851 		sc->sc_xbdreq_wait = 0;
    852 
    853 		xbdreq = SLIST_FIRST(&sc->sc_xbdreq_head);
    854 		if (__predict_false(xbdreq == NULL)) {
    855 			DPRINTF(("xbdioctl: no req\n"));
    856 			error = ENOMEM;
    857 		} else {
    858 			SLIST_REMOVE_HEAD(&sc->sc_xbdreq_head, req_next);
    859 			req = RING_GET_REQUEST(&sc->sc_ring,
    860 			    sc->sc_ring.req_prod_pvt);
    861 			req->id = xbdreq->req_id;
    862 			req->operation = BLKIF_OP_FLUSH_DISKCACHE;
    863 			req->handle = sc->sc_handle;
    864 			xbdreq->req_sync.s_done = 0;
    865 			sc->sc_ring.req_prod_pvt++;
    866 			RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->sc_ring,
    867 			    notify);
    868 			if (notify)
    869 				hypervisor_notify_via_evtchn(sc->sc_evtchn);
    870 			/* request sent, no wait for completion */
    871 			while (xbdreq->req_sync.s_done == 0) {
    872 				tsleep(xbdreq, PRIBIO, "xbdsync", 0);
    873 			}
    874 			if (xbdreq->req_sync.s_error == BLKIF_RSP_EOPNOTSUPP)
    875 				error = EOPNOTSUPP;
    876 			else if (xbdreq->req_sync.s_error == BLKIF_RSP_OKAY)
    877 				error = 0;
    878 			else
    879 				error = EIO;
    880 			SLIST_INSERT_HEAD(&sc->sc_xbdreq_head, xbdreq,
    881 			    req_next);
    882 		}
    883 		splx(s);
    884 		break;
    885 
    886 	default:
    887 		error = dk_ioctl(dksc, dev, cmd, data, flag, l);
    888 		break;
    889 	}
    890 
    891 	return error;
    892 }
    893 
    894 int
    895 xbddump(dev_t dev, daddr_t blkno, void *va, size_t size)
    896 {
    897 	struct xbd_xenbus_softc *sc;
    898 
    899 	sc  = device_lookup_private(&xbd_cd, DISKUNIT(dev));
    900 	if (sc == NULL)
    901 		return (ENXIO);
    902 
    903 	DPRINTF(("xbddump(%d, %" PRId64 ", %p, %lu)\n", dev, blkno, va,
    904 	    (unsigned long)size));
    905 	return dk_dump(&sc->sc_dksc, dev, blkno, va, size);
    906 }
    907 
    908 static int
    909 xbd_diskstart(device_t self, struct buf *bp)
    910 {
    911 	struct xbd_xenbus_softc *sc = device_private(self);
    912 	struct xbd_req *xbdreq;
    913 	blkif_request_t *req;
    914 	size_t bcount, off;
    915 	paddr_t ma;
    916 	vaddr_t va;
    917 	int nsects, nbytes, seg;
    918 	int notify, error = 0;
    919 
    920 	DPRINTF(("xbd_diskstart(%p): b_bcount = %ld\n",
    921 	    bp, (long)bp->b_bcount));
    922 
    923 	if (sc->sc_shutdown != BLKIF_SHUTDOWN_RUN) {
    924 		error = EIO;
    925 		goto err;
    926 	}
    927 
    928 	if (bp->b_rawblkno < 0 || bp->b_rawblkno > sc->sc_xbdsize) {
    929 		/* invalid block number */
    930 		error = EINVAL;
    931 		goto err;
    932 	}
    933 
    934 	if (__predict_false(
    935 	    sc->sc_backend_status == BLKIF_STATE_SUSPENDED)) {
    936 		/* device is suspended, do not consume buffer */
    937 		DPRINTF(("%s: (xbd_diskstart) device suspended\n",
    938 		    sc->sc_dksc.sc_xname));
    939 		error = EAGAIN;
    940 		goto out;
    941 	}
    942 
    943 	if (RING_FULL(&sc->sc_ring) || sc->sc_xbdreq_wait) {
    944 		DPRINTF(("xbd_diskstart: ring_full\n"));
    945 		error = EAGAIN;
    946 		goto out;
    947 	}
    948 
    949 	xbdreq = SLIST_FIRST(&sc->sc_xbdreq_head);
    950 	if (__predict_false(xbdreq == NULL)) {
    951 		DPRINTF(("xbd_diskstart: no req\n"));
    952 		error = EAGAIN;
    953 		goto out;
    954 	}
    955 
    956 	xbdreq->req_bp = bp;
    957 	xbdreq->req_data = bp->b_data;
    958 	if ((vaddr_t)bp->b_data & (XEN_BSIZE - 1)) {
    959 		if (__predict_false(xbd_map_align(xbdreq) != 0)) {
    960 			DPRINTF(("xbd_diskstart: no align\n"));
    961 			error = EAGAIN;
    962 			goto out;
    963 		}
    964 	}
    965 
    966 	SLIST_REMOVE_HEAD(&sc->sc_xbdreq_head, req_next);
    967 	req = RING_GET_REQUEST(&sc->sc_ring, sc->sc_ring.req_prod_pvt);
    968 	req->id = xbdreq->req_id;
    969 	req->operation =
    970 	    bp->b_flags & B_READ ? BLKIF_OP_READ : BLKIF_OP_WRITE;
    971 	req->sector_number = bp->b_rawblkno;
    972 	req->handle = sc->sc_handle;
    973 
    974 	va = (vaddr_t)xbdreq->req_data & ~PAGE_MASK;
    975 	off = (vaddr_t)xbdreq->req_data & PAGE_MASK;
    976 	bcount = bp->b_bcount;
    977 	bp->b_resid = 0;
    978 	for (seg = 0; bcount > 0;) {
    979 		pmap_extract_ma(pmap_kernel(), va, &ma);
    980 		KASSERT((ma & (XEN_BSIZE - 1)) == 0);
    981 		if (bcount > PAGE_SIZE - off)
    982 			nbytes = PAGE_SIZE - off;
    983 		else
    984 			nbytes = bcount;
    985 		nsects = nbytes >> XEN_BSHIFT;
    986 		req->seg[seg].first_sect = off >> XEN_BSHIFT;
    987 		req->seg[seg].last_sect =
    988 		    (off >> XEN_BSHIFT) + nsects - 1;
    989 		KASSERT(req->seg[seg].first_sect <=
    990 		    req->seg[seg].last_sect);
    991 		KASSERT(req->seg[seg].last_sect < 8);
    992 		if (__predict_false(xengnt_grant_access(
    993 		    sc->sc_xbusd->xbusd_otherend_id, ma,
    994 		    (bp->b_flags & B_READ) == 0,
    995 		    &xbdreq->req_gntref[seg])))
    996 			panic("xbd_diskstart: xengnt_grant_access"); /* XXX XXX !!! */
    997 		req->seg[seg].gref = xbdreq->req_gntref[seg];
    998 		seg++;
    999 		KASSERT(seg <= BLKIF_MAX_SEGMENTS_PER_REQUEST);
   1000 		va += PAGE_SIZE;
   1001 		off = 0;
   1002 		bcount -= nbytes;
   1003 	}
   1004 	xbdreq->req_nr_segments = req->nr_segments = seg;
   1005 	sc->sc_ring.req_prod_pvt++;
   1006 
   1007 out:
   1008 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->sc_ring, notify);
   1009 	if (notify)
   1010 		hypervisor_notify_via_evtchn(sc->sc_evtchn);
   1011 err:
   1012 	return error;
   1013 }
   1014 
   1015 static int
   1016 xbd_map_align(struct xbd_req *req)
   1017 {
   1018 	int s = splvm();
   1019 	int rc;
   1020 
   1021 	rc = uvm_km_kmem_alloc(kmem_va_arena,
   1022 	    req->req_bp->b_bcount, (VM_NOSLEEP | VM_INSTANTFIT),
   1023 	    (vmem_addr_t *)&req->req_data);
   1024 	splx(s);
   1025 	if (__predict_false(rc != 0))
   1026 		return ENOMEM;
   1027 	if ((req->req_bp->b_flags & B_READ) == 0)
   1028 		memcpy(req->req_data, req->req_bp->b_data,
   1029 		    req->req_bp->b_bcount);
   1030 	return 0;
   1031 }
   1032 
   1033 static void
   1034 xbd_unmap_align(struct xbd_req *req)
   1035 {
   1036 	int s;
   1037 	if (req->req_bp->b_flags & B_READ)
   1038 		memcpy(req->req_bp->b_data, req->req_data,
   1039 		    req->req_bp->b_bcount);
   1040 	s = splvm();
   1041 	uvm_km_kmem_free(kmem_va_arena, (vaddr_t)req->req_data, req->req_bp->b_bcount);
   1042 	splx(s);
   1043 }
   1044