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 struct xenbus_device *dev;
26 struct xen_blkif *blkif;
27 struct xenbus_watch backend_watch;
33 static struct kmem_cache *xen_blkif_cachep;
34 static void connect(struct backend_info *);
35 static int connect_ring(struct backend_info *);
36 static void backend_changed(struct xenbus_watch *, const char **,
39 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
44 static int blkback_name(struct xen_blkif *blkif, char *buf)
46 char *devpath, *devname;
47 struct xenbus_device *dev = blkif->be->dev;
49 devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
51 return PTR_ERR(devpath);
53 devname = strstr(devpath, "/dev/");
55 devname += strlen("/dev/");
59 snprintf(buf, TASK_COMM_LEN, "blkback.%d.%s", blkif->domid, devname);
65 static void xen_update_blkif_status(struct xen_blkif *blkif)
68 char name[TASK_COMM_LEN];
70 /* Not ready to connect? */
71 if (!blkif->irq || !blkif->vbd.bdev)
74 /* Already connected? */
75 if (blkif->be->dev->state == XenbusStateConnected)
78 /* Attempt to connect: exit if we fail to. */
80 if (blkif->be->dev->state != XenbusStateConnected)
83 err = blkback_name(blkif, name);
85 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
89 err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
91 xenbus_dev_error(blkif->be->dev, err, "block flush");
94 invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
96 blkif->xenblkd = kthread_run(xen_blkif_schedule, blkif, name);
97 if (IS_ERR(blkif->xenblkd)) {
98 err = PTR_ERR(blkif->xenblkd);
99 blkif->xenblkd = NULL;
100 xenbus_dev_error(blkif->be->dev, err, "start xenblkd");
104 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
106 struct xen_blkif *blkif;
108 blkif = kmem_cache_alloc(xen_blkif_cachep, GFP_KERNEL);
110 return ERR_PTR(-ENOMEM);
112 memset(blkif, 0, sizeof(*blkif));
113 blkif->domid = domid;
114 spin_lock_init(&blkif->blk_ring_lock);
115 atomic_set(&blkif->refcnt, 1);
116 init_waitqueue_head(&blkif->wq);
117 init_completion(&blkif->drain_complete);
118 atomic_set(&blkif->drain, 0);
119 blkif->st_print = jiffies;
120 init_waitqueue_head(&blkif->waiting_to_free);
125 static int map_frontend_page(struct xen_blkif *blkif, unsigned long shared_page)
127 struct gnttab_map_grant_ref op;
129 gnttab_set_map_op(&op, (unsigned long)blkif->blk_ring_area->addr,
130 GNTMAP_host_map, shared_page, blkif->domid);
132 if (HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, &op, 1))
136 DPRINTK("Grant table operation failure !\n");
140 blkif->shmem_ref = shared_page;
141 blkif->shmem_handle = op.handle;
146 static void unmap_frontend_page(struct xen_blkif *blkif)
148 struct gnttab_unmap_grant_ref op;
150 gnttab_set_unmap_op(&op, (unsigned long)blkif->blk_ring_area->addr,
151 GNTMAP_host_map, blkif->shmem_handle);
153 if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, &op, 1))
157 static int xen_blkif_map(struct xen_blkif *blkif, unsigned long shared_page,
162 /* Already connected through? */
166 blkif->blk_ring_area = alloc_vm_area(PAGE_SIZE);
167 if (!blkif->blk_ring_area)
170 err = map_frontend_page(blkif, shared_page);
172 free_vm_area(blkif->blk_ring_area);
176 switch (blkif->blk_protocol) {
177 case BLKIF_PROTOCOL_NATIVE:
179 struct blkif_sring *sring;
180 sring = (struct blkif_sring *)blkif->blk_ring_area->addr;
181 BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE);
184 case BLKIF_PROTOCOL_X86_32:
186 struct blkif_x86_32_sring *sring_x86_32;
187 sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring_area->addr;
188 BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE);
191 case BLKIF_PROTOCOL_X86_64:
193 struct blkif_x86_64_sring *sring_x86_64;
194 sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring_area->addr;
195 BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
202 err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
204 "blkif-backend", blkif);
206 unmap_frontend_page(blkif);
207 free_vm_area(blkif->blk_ring_area);
208 blkif->blk_rings.common.sring = NULL;
216 static void xen_blkif_disconnect(struct xen_blkif *blkif)
218 if (blkif->xenblkd) {
219 kthread_stop(blkif->xenblkd);
220 blkif->xenblkd = NULL;
223 atomic_dec(&blkif->refcnt);
224 wait_event(blkif->waiting_to_free, atomic_read(&blkif->refcnt) == 0);
225 atomic_inc(&blkif->refcnt);
228 unbind_from_irqhandler(blkif->irq, blkif);
232 if (blkif->blk_rings.common.sring) {
233 unmap_frontend_page(blkif);
234 free_vm_area(blkif->blk_ring_area);
235 blkif->blk_rings.common.sring = NULL;
239 void xen_blkif_free(struct xen_blkif *blkif)
241 if (!atomic_dec_and_test(&blkif->refcnt))
243 kmem_cache_free(xen_blkif_cachep, blkif);
246 int __init xen_blkif_interface_init(void)
248 xen_blkif_cachep = kmem_cache_create("blkif_cache",
249 sizeof(struct xen_blkif),
251 if (!xen_blkif_cachep)
258 * sysfs interface for VBD I/O requests
261 #define VBD_SHOW(name, format, args...) \
262 static ssize_t show_##name(struct device *_dev, \
263 struct device_attribute *attr, \
266 struct xenbus_device *dev = to_xenbus_device(_dev); \
267 struct backend_info *be = dev_get_drvdata(&dev->dev); \
269 return sprintf(buf, format, ##args); \
271 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
273 VBD_SHOW(oo_req, "%d\n", be->blkif->st_oo_req);
274 VBD_SHOW(rd_req, "%d\n", be->blkif->st_rd_req);
275 VBD_SHOW(wr_req, "%d\n", be->blkif->st_wr_req);
276 VBD_SHOW(f_req, "%d\n", be->blkif->st_f_req);
277 VBD_SHOW(ds_req, "%d\n", be->blkif->st_ds_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 *xen_vbdstat_attrs[] = {
282 &dev_attr_oo_req.attr,
283 &dev_attr_rd_req.attr,
284 &dev_attr_wr_req.attr,
285 &dev_attr_f_req.attr,
286 &dev_attr_ds_req.attr,
287 &dev_attr_rd_sect.attr,
288 &dev_attr_wr_sect.attr,
292 static struct attribute_group xen_vbdstat_group = {
293 .name = "statistics",
294 .attrs = xen_vbdstat_attrs,
297 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
298 VBD_SHOW(mode, "%s\n", be->mode);
300 int xenvbd_sysfs_addif(struct xenbus_device *dev)
304 error = device_create_file(&dev->dev, &dev_attr_physical_device);
308 error = device_create_file(&dev->dev, &dev_attr_mode);
312 error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
318 fail3: sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
319 fail2: device_remove_file(&dev->dev, &dev_attr_mode);
320 fail1: device_remove_file(&dev->dev, &dev_attr_physical_device);
324 void xenvbd_sysfs_delif(struct xenbus_device *dev)
326 sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
327 device_remove_file(&dev->dev, &dev_attr_mode);
328 device_remove_file(&dev->dev, &dev_attr_physical_device);
332 static void xen_vbd_free(struct xen_vbd *vbd)
335 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
339 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
340 unsigned major, unsigned minor, int readonly,
344 struct block_device *bdev;
345 struct request_queue *q;
348 vbd->handle = handle;
349 vbd->readonly = readonly;
352 vbd->pdevice = MKDEV(major, minor);
354 bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
355 FMODE_READ : FMODE_WRITE, NULL);
358 DPRINTK("xen_vbd_create: device %08x could not be opened.\n",
364 if (vbd->bdev->bd_disk == NULL) {
365 DPRINTK("xen_vbd_create: device %08x doesn't exist.\n",
370 vbd->size = vbd_sz(vbd);
372 if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
373 vbd->type |= VDISK_CDROM;
374 if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
375 vbd->type |= VDISK_REMOVABLE;
377 q = bdev_get_queue(bdev);
378 if (q && q->flush_flags)
379 vbd->flush_support = true;
381 DPRINTK("Successful creation of handle=%04x (dom=%u)\n",
382 handle, blkif->domid);
385 static int xen_blkbk_remove(struct xenbus_device *dev)
387 struct backend_info *be = dev_get_drvdata(&dev->dev);
391 if (be->major || be->minor)
392 xenvbd_sysfs_delif(dev);
394 if (be->backend_watch.node) {
395 unregister_xenbus_watch(&be->backend_watch);
396 kfree(be->backend_watch.node);
397 be->backend_watch.node = NULL;
401 xen_blkif_disconnect(be->blkif);
402 xen_vbd_free(&be->blkif->vbd);
403 xen_blkif_free(be->blkif);
408 dev_set_drvdata(&dev->dev, NULL);
412 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
413 struct backend_info *be, int state)
415 struct xenbus_device *dev = be->dev;
418 err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
421 xenbus_dev_fatal(dev, err, "writing feature-flush-cache");
426 int xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
428 struct xenbus_device *dev = be->dev;
429 struct xen_blkif *blkif = be->blkif;
434 type = xenbus_read(XBT_NIL, dev->nodename, "type", NULL);
436 if (strncmp(type, "file", 4) == 0) {
438 blkif->blk_backend_type = BLKIF_BACKEND_FILE;
440 if (strncmp(type, "phy", 3) == 0) {
441 struct block_device *bdev = be->blkif->vbd.bdev;
442 struct request_queue *q = bdev_get_queue(bdev);
443 if (blk_queue_discard(q)) {
444 err = xenbus_printf(xbt, dev->nodename,
445 "discard-granularity", "%u",
446 q->limits.discard_granularity);
448 xenbus_dev_fatal(dev, err,
449 "writing discard-granularity");
452 err = xenbus_printf(xbt, dev->nodename,
453 "discard-alignment", "%u",
454 q->limits.discard_alignment);
456 xenbus_dev_fatal(dev, err,
457 "writing discard-alignment");
461 blkif->blk_backend_type = BLKIF_BACKEND_PHY;
466 xenbus_dev_fatal(dev, err, "reading type");
470 err = xenbus_printf(xbt, dev->nodename, "feature-discard",
473 xenbus_dev_fatal(dev, err, "writing feature-discard");
479 int xen_blkbk_barrier(struct xenbus_transaction xbt,
480 struct backend_info *be, int state)
482 struct xenbus_device *dev = be->dev;
485 err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
488 xenbus_dev_fatal(dev, err, "writing feature-barrier");
494 * Entry point to this code when a new device is created. Allocate the basic
495 * structures, and watch the store waiting for the hotplug scripts to tell us
496 * the device's physical major and minor numbers. Switch to InitWait.
498 static int xen_blkbk_probe(struct xenbus_device *dev,
499 const struct xenbus_device_id *id)
502 struct backend_info *be = kzalloc(sizeof(struct backend_info),
505 xenbus_dev_fatal(dev, -ENOMEM,
506 "allocating backend structure");
510 dev_set_drvdata(&dev->dev, be);
512 be->blkif = xen_blkif_alloc(dev->otherend_id);
513 if (IS_ERR(be->blkif)) {
514 err = PTR_ERR(be->blkif);
516 xenbus_dev_fatal(dev, err, "creating block interface");
520 /* setup back pointer */
523 err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
524 "%s/%s", dev->nodename, "physical-device");
528 err = xenbus_switch_state(dev, XenbusStateInitWait);
536 xen_blkbk_remove(dev);
542 * Callback received when the hotplug scripts have placed the physical-device
543 * node. Read it and the mode node, and create a vbd. If the frontend is
546 static void backend_changed(struct xenbus_watch *watch,
547 const char **vec, unsigned int len)
552 struct backend_info *be
553 = container_of(watch, struct backend_info, backend_watch);
554 struct xenbus_device *dev = be->dev;
560 err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
562 if (XENBUS_EXIST_ERR(err)) {
564 * Since this watch will fire once immediately after it is
565 * registered, we expect this. Ignore it, and wait for the
571 xenbus_dev_fatal(dev, err, "reading physical-device");
575 if ((be->major || be->minor) &&
576 ((be->major != major) || (be->minor != minor))) {
577 pr_warn(DRV_PFX "changing physical device (from %x:%x to %x:%x) not supported.\n",
578 be->major, be->minor, major, minor);
582 be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
583 if (IS_ERR(be->mode)) {
584 err = PTR_ERR(be->mode);
586 xenbus_dev_fatal(dev, err, "reading mode");
590 device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
591 if (!IS_ERR(device_type)) {
592 cdrom = strcmp(device_type, "cdrom") == 0;
596 if (be->major == 0 && be->minor == 0) {
597 /* Front end dir is a number, which is used as the handle. */
599 char *p = strrchr(dev->otherend, '/') + 1;
601 err = strict_strtoul(p, 0, &handle);
608 err = xen_vbd_create(be->blkif, handle, major, minor,
609 (NULL == strchr(be->mode, 'w')), cdrom);
613 xenbus_dev_fatal(dev, err, "creating vbd structure");
617 err = xenvbd_sysfs_addif(dev);
619 xen_vbd_free(&be->blkif->vbd);
622 xenbus_dev_fatal(dev, err, "creating sysfs entries");
626 /* We're potentially connected now */
627 xen_update_blkif_status(be->blkif);
633 * Callback received when the frontend's state changes.
635 static void frontend_changed(struct xenbus_device *dev,
636 enum xenbus_state frontend_state)
638 struct backend_info *be = dev_get_drvdata(&dev->dev);
641 DPRINTK("%s", xenbus_strstate(frontend_state));
643 switch (frontend_state) {
644 case XenbusStateInitialising:
645 if (dev->state == XenbusStateClosed) {
646 pr_info(DRV_PFX "%s: prepare for reconnect\n",
648 xenbus_switch_state(dev, XenbusStateInitWait);
652 case XenbusStateInitialised:
653 case XenbusStateConnected:
655 * Ensure we connect even when two watches fire in
656 * close successsion and we miss the intermediate value
659 if (dev->state == XenbusStateConnected)
663 * Enforce precondition before potential leak point.
664 * xen_blkif_disconnect() is idempotent.
666 xen_blkif_disconnect(be->blkif);
668 err = connect_ring(be);
671 xen_update_blkif_status(be->blkif);
674 case XenbusStateClosing:
675 xenbus_switch_state(dev, XenbusStateClosing);
678 case XenbusStateClosed:
679 xen_blkif_disconnect(be->blkif);
680 xenbus_switch_state(dev, XenbusStateClosed);
681 if (xenbus_dev_is_online(dev))
683 /* fall through if not online */
684 case XenbusStateUnknown:
685 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
686 device_unregister(&dev->dev);
690 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
697 /* ** Connection ** */
701 * Write the physical details regarding the block device to the store, and
702 * switch to Connected state.
704 static void connect(struct backend_info *be)
706 struct xenbus_transaction xbt;
708 struct xenbus_device *dev = be->dev;
710 DPRINTK("%s", dev->otherend);
712 /* Supply the information about the device the frontend needs */
714 err = xenbus_transaction_start(&xbt);
716 xenbus_dev_fatal(dev, err, "starting transaction");
720 err = xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
724 err = xen_blkbk_discard(xbt, be);
726 /* If we can't advertise it is OK. */
727 err = xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
729 err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
730 (unsigned long long)vbd_sz(&be->blkif->vbd));
732 xenbus_dev_fatal(dev, err, "writing %s/sectors",
737 /* FIXME: use a typename instead */
738 err = xenbus_printf(xbt, dev->nodename, "info", "%u",
739 be->blkif->vbd.type |
740 (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
742 xenbus_dev_fatal(dev, err, "writing %s/info",
746 err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
748 bdev_logical_block_size(be->blkif->vbd.bdev));
750 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
755 err = xenbus_transaction_end(xbt, 0);
759 xenbus_dev_fatal(dev, err, "ending transaction");
761 err = xenbus_switch_state(dev, XenbusStateConnected);
763 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
768 xenbus_transaction_end(xbt, 1);
772 static int connect_ring(struct backend_info *be)
774 struct xenbus_device *dev = be->dev;
775 unsigned long ring_ref;
777 char protocol[64] = "";
780 DPRINTK("%s", dev->otherend);
782 err = xenbus_gather(XBT_NIL, dev->otherend, "ring-ref", "%lu",
783 &ring_ref, "event-channel", "%u", &evtchn, NULL);
785 xenbus_dev_fatal(dev, err,
786 "reading %s/ring-ref and event-channel",
791 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
792 err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
793 "%63s", protocol, NULL);
795 strcpy(protocol, "unspecified, assuming native");
796 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
797 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
798 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
799 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
800 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
801 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
803 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
806 pr_info(DRV_PFX "ring-ref %ld, event-channel %d, protocol %d (%s)\n",
807 ring_ref, evtchn, be->blkif->blk_protocol, protocol);
809 /* Map the shared frame, irq etc. */
810 err = xen_blkif_map(be->blkif, ring_ref, evtchn);
812 xenbus_dev_fatal(dev, err, "mapping ring-ref %lu port %u",
821 /* ** Driver Registration ** */
824 static const struct xenbus_device_id xen_blkbk_ids[] = {
830 static struct xenbus_driver xen_blkbk = {
832 .owner = THIS_MODULE,
833 .ids = xen_blkbk_ids,
834 .probe = xen_blkbk_probe,
835 .remove = xen_blkbk_remove,
836 .otherend_changed = frontend_changed
840 int xen_blkif_xenbus_init(void)
842 return xenbus_register_backend(&xen_blkbk);