1 /* Xenbus code for blkif backend
2 Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
3 Copyright (C) 2005 XenSource Ltd
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
18 #include <linux/module.h>
19 #include <linux/kthread.h>
20 #include <xen/events.h>
21 #include <xen/grant_table.h>
25 #define DPRINTK(fmt, args...) \
26 pr_debug("blkback/xenbus (%s:%d) " fmt ".\n", \
27 __func__, __LINE__, ##args)
30 struct xenbus_device
*dev
;
31 struct blkif_st
*blkif
;
32 struct xenbus_watch backend_watch
;
38 static struct kmem_cache
*blkif_cachep
;
39 static void connect(struct backend_info
*);
40 static int connect_ring(struct backend_info
*);
41 static void backend_changed(struct xenbus_watch
*, const char **,
44 struct xenbus_device
*blkback_xenbus(struct backend_info
*be
)
49 static int blkback_name(struct blkif_st
*blkif
, char *buf
)
51 char *devpath
, *devname
;
52 struct xenbus_device
*dev
= blkif
->be
->dev
;
54 devpath
= xenbus_read(XBT_NIL
, dev
->nodename
, "dev", NULL
);
56 return PTR_ERR(devpath
);
58 devname
= strstr(devpath
, "/dev/");
60 devname
+= strlen("/dev/");
64 snprintf(buf
, TASK_COMM_LEN
, "blkback.%d.%s", blkif
->domid
, devname
);
70 static void update_blkif_status(struct blkif_st
*blkif
)
73 char name
[TASK_COMM_LEN
];
75 /* Not ready to connect? */
76 if (!blkif
->irq
|| !blkif
->vbd
.bdev
)
79 /* Already connected? */
80 if (blkif
->be
->dev
->state
== XenbusStateConnected
)
83 /* Attempt to connect: exit if we fail to. */
85 if (blkif
->be
->dev
->state
!= XenbusStateConnected
)
88 err
= blkback_name(blkif
, name
);
90 xenbus_dev_error(blkif
->be
->dev
, err
, "get blkback dev name");
94 err
= filemap_write_and_wait(blkif
->vbd
.bdev
->bd_inode
->i_mapping
);
96 xenbus_dev_error(blkif
->be
->dev
, err
, "block flush");
99 invalidate_inode_pages2(blkif
->vbd
.bdev
->bd_inode
->i_mapping
);
101 blkif
->xenblkd
= kthread_run(blkif_schedule
, blkif
, name
);
102 if (IS_ERR(blkif
->xenblkd
)) {
103 err
= PTR_ERR(blkif
->xenblkd
);
104 blkif
->xenblkd
= NULL
;
105 xenbus_dev_error(blkif
->be
->dev
, err
, "start xenblkd");
109 struct blkif_st
*blkif_alloc(domid_t domid
)
111 struct blkif_st
*blkif
;
113 blkif
= kmem_cache_alloc(blkif_cachep
, GFP_KERNEL
);
115 return ERR_PTR(-ENOMEM
);
117 memset(blkif
, 0, sizeof(*blkif
));
118 blkif
->domid
= domid
;
119 spin_lock_init(&blkif
->blk_ring_lock
);
120 atomic_set(&blkif
->refcnt
, 1);
121 init_waitqueue_head(&blkif
->wq
);
122 blkif
->st_print
= jiffies
;
123 init_waitqueue_head(&blkif
->waiting_to_free
);
128 static int map_frontend_page(struct blkif_st
*blkif
, unsigned long shared_page
)
130 struct gnttab_map_grant_ref op
;
132 gnttab_set_map_op(&op
, (unsigned long)blkif
->blk_ring_area
->addr
,
133 GNTMAP_host_map
, shared_page
, blkif
->domid
);
135 if (HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref
, &op
, 1))
139 DPRINTK(" Grant table operation failure !\n");
143 blkif
->shmem_ref
= shared_page
;
144 blkif
->shmem_handle
= op
.handle
;
149 static void unmap_frontend_page(struct blkif_st
*blkif
)
151 struct gnttab_unmap_grant_ref op
;
153 gnttab_set_unmap_op(&op
, (unsigned long)blkif
->blk_ring_area
->addr
,
154 GNTMAP_host_map
, blkif
->shmem_handle
);
156 if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref
, &op
, 1))
160 int blkif_map(struct blkif_st
*blkif
, unsigned long shared_page
,
165 /* Already connected through? */
169 blkif
->blk_ring_area
= alloc_vm_area(PAGE_SIZE
);
170 if (!blkif
->blk_ring_area
)
173 err
= map_frontend_page(blkif
, shared_page
);
175 free_vm_area(blkif
->blk_ring_area
);
179 switch (blkif
->blk_protocol
) {
180 case BLKIF_PROTOCOL_NATIVE
:
182 struct blkif_sring
*sring
;
183 sring
= (struct blkif_sring
*)blkif
->blk_ring_area
->addr
;
184 BACK_RING_INIT(&blkif
->blk_rings
.native
, sring
, PAGE_SIZE
);
187 case BLKIF_PROTOCOL_X86_32
:
189 struct blkif_x86_32_sring
*sring_x86_32
;
190 sring_x86_32
= (struct blkif_x86_32_sring
*)blkif
->blk_ring_area
->addr
;
191 BACK_RING_INIT(&blkif
->blk_rings
.x86_32
, sring_x86_32
, PAGE_SIZE
);
194 case BLKIF_PROTOCOL_X86_64
:
196 struct blkif_x86_64_sring
*sring_x86_64
;
197 sring_x86_64
= (struct blkif_x86_64_sring
*)blkif
->blk_ring_area
->addr
;
198 BACK_RING_INIT(&blkif
->blk_rings
.x86_64
, sring_x86_64
, PAGE_SIZE
);
205 err
= bind_interdomain_evtchn_to_irqhandler(
206 blkif
->domid
, evtchn
, blkif_be_int
, 0, "blkif-backend", blkif
);
208 unmap_frontend_page(blkif
);
209 free_vm_area(blkif
->blk_ring_area
);
210 blkif
->blk_rings
.common
.sring
= NULL
;
218 void blkif_disconnect(struct blkif_st
*blkif
)
220 if (blkif
->xenblkd
) {
221 kthread_stop(blkif
->xenblkd
);
222 blkif
->xenblkd
= NULL
;
225 atomic_dec(&blkif
->refcnt
);
226 wait_event(blkif
->waiting_to_free
, atomic_read(&blkif
->refcnt
) == 0);
227 atomic_inc(&blkif
->refcnt
);
230 unbind_from_irqhandler(blkif
->irq
, blkif
);
234 if (blkif
->blk_rings
.common
.sring
) {
235 unmap_frontend_page(blkif
);
236 free_vm_area(blkif
->blk_ring_area
);
237 blkif
->blk_rings
.common
.sring
= NULL
;
241 void blkif_free(struct blkif_st
*blkif
)
243 if (!atomic_dec_and_test(&blkif
->refcnt
))
245 kmem_cache_free(blkif_cachep
, blkif
);
248 int __init
blkif_interface_init(void)
250 blkif_cachep
= kmem_cache_create("blkif_cache", sizeof(struct blkif_st
),
259 * sysfs interface for VBD I/O requests
262 #define VBD_SHOW(name, format, args...) \
263 static ssize_t show_##name(struct device *_dev, \
264 struct device_attribute *attr, \
267 struct xenbus_device *dev = to_xenbus_device(_dev); \
268 struct backend_info *be = dev_get_drvdata(&dev->dev); \
270 return sprintf(buf, format, ##args); \
272 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
274 VBD_SHOW(oo_req
, "%d\n", be
->blkif
->st_oo_req
);
275 VBD_SHOW(rd_req
, "%d\n", be
->blkif
->st_rd_req
);
276 VBD_SHOW(wr_req
, "%d\n", be
->blkif
->st_wr_req
);
277 VBD_SHOW(br_req
, "%d\n", be
->blkif
->st_br_req
);
278 VBD_SHOW(rd_sect
, "%d\n", be
->blkif
->st_rd_sect
);
279 VBD_SHOW(wr_sect
, "%d\n", be
->blkif
->st_wr_sect
);
281 static struct attribute
*vbdstat_attrs
[] = {
282 &dev_attr_oo_req
.attr
,
283 &dev_attr_rd_req
.attr
,
284 &dev_attr_wr_req
.attr
,
285 &dev_attr_br_req
.attr
,
286 &dev_attr_rd_sect
.attr
,
287 &dev_attr_wr_sect
.attr
,
291 static struct attribute_group vbdstat_group
= {
292 .name
= "statistics",
293 .attrs
= vbdstat_attrs
,
296 VBD_SHOW(physical_device
, "%x:%x\n", be
->major
, be
->minor
);
297 VBD_SHOW(mode
, "%s\n", be
->mode
);
299 int xenvbd_sysfs_addif(struct xenbus_device
*dev
)
303 error
= device_create_file(&dev
->dev
, &dev_attr_physical_device
);
307 error
= device_create_file(&dev
->dev
, &dev_attr_mode
);
311 error
= sysfs_create_group(&dev
->dev
.kobj
, &vbdstat_group
);
317 fail3
: sysfs_remove_group(&dev
->dev
.kobj
, &vbdstat_group
);
318 fail2
: device_remove_file(&dev
->dev
, &dev_attr_mode
);
319 fail1
: device_remove_file(&dev
->dev
, &dev_attr_physical_device
);
323 void xenvbd_sysfs_delif(struct xenbus_device
*dev
)
325 sysfs_remove_group(&dev
->dev
.kobj
, &vbdstat_group
);
326 device_remove_file(&dev
->dev
, &dev_attr_mode
);
327 device_remove_file(&dev
->dev
, &dev_attr_physical_device
);
331 static void vbd_free(struct vbd
*vbd
)
334 blkdev_put(vbd
->bdev
, vbd
->readonly
? FMODE_READ
: FMODE_WRITE
);
338 static int vbd_create(struct blkif_st
*blkif
, blkif_vdev_t handle
,
339 unsigned major
, unsigned minor
, int readonly
,
343 struct block_device
*bdev
;
346 vbd
->handle
= handle
;
347 vbd
->readonly
= readonly
;
350 vbd
->pdevice
= MKDEV(major
, minor
);
352 bdev
= blkdev_get_by_dev(vbd
->pdevice
, vbd
->readonly
?
353 FMODE_READ
: FMODE_WRITE
, NULL
);
356 DPRINTK("vbd_creat: device %08x could not be opened.\n",
362 vbd
->size
= vbd_sz(vbd
);
364 if (vbd
->bdev
->bd_disk
== NULL
) {
365 DPRINTK("vbd_creat: device %08x doesn't exist.\n",
371 if (vbd
->bdev
->bd_disk
->flags
& GENHD_FL_CD
|| cdrom
)
372 vbd
->type
|= VDISK_CDROM
;
373 if (vbd
->bdev
->bd_disk
->flags
& GENHD_FL_REMOVABLE
)
374 vbd
->type
|= VDISK_REMOVABLE
;
376 DPRINTK("Successful creation of handle=%04x (dom=%u)\n",
377 handle
, blkif
->domid
);
380 static int blkback_remove(struct xenbus_device
*dev
)
382 struct backend_info
*be
= dev_get_drvdata(&dev
->dev
);
386 if (be
->major
|| be
->minor
)
387 xenvbd_sysfs_delif(dev
);
389 if (be
->backend_watch
.node
) {
390 unregister_xenbus_watch(&be
->backend_watch
);
391 kfree(be
->backend_watch
.node
);
392 be
->backend_watch
.node
= NULL
;
396 blkif_disconnect(be
->blkif
);
397 vbd_free(&be
->blkif
->vbd
);
398 blkif_free(be
->blkif
);
403 dev_set_drvdata(&dev
->dev
, NULL
);
407 int blkback_barrier(struct xenbus_transaction xbt
,
408 struct backend_info
*be
, int state
)
410 struct xenbus_device
*dev
= be
->dev
;
413 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-barrier",
416 xenbus_dev_fatal(dev
, err
, "writing feature-barrier");
422 * Entry point to this code when a new device is created. Allocate the basic
423 * structures, and watch the store waiting for the hotplug scripts to tell us
424 * the device's physical major and minor numbers. Switch to InitWait.
426 static int blkback_probe(struct xenbus_device
*dev
,
427 const struct xenbus_device_id
*id
)
430 struct backend_info
*be
= kzalloc(sizeof(struct backend_info
),
433 xenbus_dev_fatal(dev
, -ENOMEM
,
434 "allocating backend structure");
438 dev_set_drvdata(&dev
->dev
, be
);
440 be
->blkif
= blkif_alloc(dev
->otherend_id
);
441 if (IS_ERR(be
->blkif
)) {
442 err
= PTR_ERR(be
->blkif
);
444 xenbus_dev_fatal(dev
, err
, "creating block interface");
448 /* setup back pointer */
451 err
= xenbus_watch_pathfmt(dev
, &be
->backend_watch
, backend_changed
,
452 "%s/%s", dev
->nodename
, "physical-device");
456 err
= xenbus_switch_state(dev
, XenbusStateInitWait
);
470 * Callback received when the hotplug scripts have placed the physical-device
471 * node. Read it and the mode node, and create a vbd. If the frontend is
474 static void backend_changed(struct xenbus_watch
*watch
,
475 const char **vec
, unsigned int len
)
480 struct backend_info
*be
481 = container_of(watch
, struct backend_info
, backend_watch
);
482 struct xenbus_device
*dev
= be
->dev
;
488 err
= xenbus_scanf(XBT_NIL
, dev
->nodename
, "physical-device", "%x:%x",
490 if (XENBUS_EXIST_ERR(err
)) {
491 /* Since this watch will fire once immediately after it is
492 registered, we expect this. Ignore it, and wait for the
497 xenbus_dev_fatal(dev
, err
, "reading physical-device");
501 if ((be
->major
|| be
->minor
) &&
502 ((be
->major
!= major
) || (be
->minor
!= minor
))) {
504 "blkback: changing physical device (from %x:%x to "
505 "%x:%x) not supported.\n", be
->major
, be
->minor
,
510 be
->mode
= xenbus_read(XBT_NIL
, dev
->nodename
, "mode", NULL
);
511 if (IS_ERR(be
->mode
)) {
512 err
= PTR_ERR(be
->mode
);
514 xenbus_dev_fatal(dev
, err
, "reading mode");
518 device_type
= xenbus_read(XBT_NIL
, dev
->otherend
, "device-type", NULL
);
519 if (!IS_ERR(device_type
)) {
520 cdrom
= strcmp(device_type
, "cdrom") == 0;
524 if (be
->major
== 0 && be
->minor
== 0) {
525 /* Front end dir is a number, which is used as the handle. */
527 char *p
= strrchr(dev
->otherend
, '/') + 1;
529 err
= strict_strtoul(p
, 0, &handle
);
536 err
= vbd_create(be
->blkif
, handle
, major
, minor
,
537 (NULL
== strchr(be
->mode
, 'w')), cdrom
);
539 be
->major
= be
->minor
= 0;
540 xenbus_dev_fatal(dev
, err
, "creating vbd structure");
544 err
= xenvbd_sysfs_addif(dev
);
546 vbd_free(&be
->blkif
->vbd
);
547 be
->major
= be
->minor
= 0;
548 xenbus_dev_fatal(dev
, err
, "creating sysfs entries");
552 /* We're potentially connected now */
553 update_blkif_status(be
->blkif
);
559 * Callback received when the frontend's state changes.
561 static void frontend_changed(struct xenbus_device
*dev
,
562 enum xenbus_state frontend_state
)
564 struct backend_info
*be
= dev_get_drvdata(&dev
->dev
);
567 DPRINTK("%s", xenbus_strstate(frontend_state
));
569 switch (frontend_state
) {
570 case XenbusStateInitialising
:
571 if (dev
->state
== XenbusStateClosed
) {
572 printk(KERN_INFO
"%s: %s: prepare for reconnect\n",
573 __func__
, dev
->nodename
);
574 xenbus_switch_state(dev
, XenbusStateInitWait
);
578 case XenbusStateInitialised
:
579 case XenbusStateConnected
:
580 /* Ensure we connect even when two watches fire in
581 close successsion and we miss the intermediate value
582 of frontend_state. */
583 if (dev
->state
== XenbusStateConnected
)
586 /* Enforce precondition before potential leak point.
587 * blkif_disconnect() is idempotent.
589 blkif_disconnect(be
->blkif
);
591 err
= connect_ring(be
);
594 update_blkif_status(be
->blkif
);
597 case XenbusStateClosing
:
598 blkif_disconnect(be
->blkif
);
599 xenbus_switch_state(dev
, XenbusStateClosing
);
602 case XenbusStateClosed
:
603 xenbus_switch_state(dev
, XenbusStateClosed
);
604 if (xenbus_dev_is_online(dev
))
606 /* fall through if not online */
607 case XenbusStateUnknown
:
608 /* implies blkif_disconnect() via blkback_remove() */
609 device_unregister(&dev
->dev
);
613 xenbus_dev_fatal(dev
, -EINVAL
, "saw state %d at frontend",
620 /* ** Connection ** */
624 * Write the physical details regarding the block device to the store, and
625 * switch to Connected state.
627 static void connect(struct backend_info
*be
)
629 struct xenbus_transaction xbt
;
631 struct xenbus_device
*dev
= be
->dev
;
633 DPRINTK("%s", dev
->otherend
);
635 /* Supply the information about the device the frontend needs */
637 err
= xenbus_transaction_start(&xbt
);
639 xenbus_dev_fatal(dev
, err
, "starting transaction");
643 err
= blkback_barrier(xbt
, be
, 1);
647 err
= xenbus_printf(xbt
, dev
->nodename
, "sectors", "%llu",
648 (unsigned long long)vbd_sz(&be
->blkif
->vbd
));
650 xenbus_dev_fatal(dev
, err
, "writing %s/sectors",
655 /* FIXME: use a typename instead */
656 err
= xenbus_printf(xbt
, dev
->nodename
, "info", "%u",
657 be
->blkif
->vbd
.type
|
658 (be
->blkif
->vbd
.readonly
? VDISK_READONLY
: 0));
660 xenbus_dev_fatal(dev
, err
, "writing %s/info",
664 err
= xenbus_printf(xbt
, dev
->nodename
, "sector-size", "%lu",
666 bdev_logical_block_size(be
->blkif
->vbd
.bdev
));
668 xenbus_dev_fatal(dev
, err
, "writing %s/sector-size",
673 err
= xenbus_transaction_end(xbt
, 0);
677 xenbus_dev_fatal(dev
, err
, "ending transaction");
679 err
= xenbus_switch_state(dev
, XenbusStateConnected
);
681 xenbus_dev_fatal(dev
, err
, "switching to Connected state",
686 xenbus_transaction_end(xbt
, 1);
690 static int connect_ring(struct backend_info
*be
)
692 struct xenbus_device
*dev
= be
->dev
;
693 unsigned long ring_ref
;
695 char protocol
[64] = "";
698 DPRINTK("%s", dev
->otherend
);
700 err
= xenbus_gather(XBT_NIL
, dev
->otherend
, "ring-ref", "%lu",
701 &ring_ref
, "event-channel", "%u", &evtchn
, NULL
);
703 xenbus_dev_fatal(dev
, err
,
704 "reading %s/ring-ref and event-channel",
709 be
->blkif
->blk_protocol
= BLKIF_PROTOCOL_NATIVE
;
710 err
= xenbus_gather(XBT_NIL
, dev
->otherend
, "protocol",
711 "%63s", protocol
, NULL
);
713 strcpy(protocol
, "unspecified, assuming native");
714 else if (0 == strcmp(protocol
, XEN_IO_PROTO_ABI_NATIVE
))
715 be
->blkif
->blk_protocol
= BLKIF_PROTOCOL_NATIVE
;
716 else if (0 == strcmp(protocol
, XEN_IO_PROTO_ABI_X86_32
))
717 be
->blkif
->blk_protocol
= BLKIF_PROTOCOL_X86_32
;
718 else if (0 == strcmp(protocol
, XEN_IO_PROTO_ABI_X86_64
))
719 be
->blkif
->blk_protocol
= BLKIF_PROTOCOL_X86_64
;
721 xenbus_dev_fatal(dev
, err
, "unknown fe protocol %s", protocol
);
725 "blkback: ring-ref %ld, event-channel %d, protocol %d (%s)\n",
726 ring_ref
, evtchn
, be
->blkif
->blk_protocol
, protocol
);
728 /* Map the shared frame, irq etc. */
729 err
= blkif_map(be
->blkif
, ring_ref
, evtchn
);
731 xenbus_dev_fatal(dev
, err
, "mapping ring-ref %lu port %u",
740 /* ** Driver Registration ** */
743 static const struct xenbus_device_id blkback_ids
[] = {
749 static struct xenbus_driver blkback
= {
751 .owner
= THIS_MODULE
,
753 .probe
= blkback_probe
,
754 .remove
= blkback_remove
,
755 .otherend_changed
= frontend_changed
759 int blkif_xenbus_init(void)
761 return xenbus_register_backend(&blkback
);