1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Xenbus code for blkif backend
3 Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
4 Copyright (C) 2005 XenSource Ltd
9 #define pr_fmt(fmt) "xen-blkback: " fmt
12 #include <linux/module.h>
13 #include <linux/kthread.h>
14 #include <xen/events.h>
15 #include <xen/grant_table.h>
18 /* On the XenBus the max length of 'ring-ref%u'. */
19 #define RINGREF_NAME_LEN (20)
22 struct xenbus_device *dev;
23 struct xen_blkif *blkif;
24 struct xenbus_watch backend_watch;
30 static struct kmem_cache *xen_blkif_cachep;
31 static void connect(struct backend_info *);
32 static int connect_ring(struct backend_info *);
33 static void backend_changed(struct xenbus_watch *, const char *,
35 static void xen_blkif_free(struct xen_blkif *blkif);
36 static void xen_vbd_free(struct xen_vbd *vbd);
38 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
44 * The last request could free the device from softirq context and
45 * xen_blkif_free() can sleep.
47 static void xen_blkif_deferred_free(struct work_struct *work)
49 struct xen_blkif *blkif;
51 blkif = container_of(work, struct xen_blkif, free_work);
52 xen_blkif_free(blkif);
55 static int blkback_name(struct xen_blkif *blkif, char *buf)
57 char *devpath, *devname;
58 struct xenbus_device *dev = blkif->be->dev;
60 devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
62 return PTR_ERR(devpath);
64 devname = strstr(devpath, "/dev/");
66 devname += strlen("/dev/");
70 snprintf(buf, TASK_COMM_LEN, "%d.%s", blkif->domid, devname);
76 static void xen_update_blkif_status(struct xen_blkif *blkif)
79 char name[TASK_COMM_LEN];
80 struct xen_blkif_ring *ring;
83 /* Not ready to connect? */
84 if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
87 /* Already connected? */
88 if (blkif->be->dev->state == XenbusStateConnected)
91 /* Attempt to connect: exit if we fail to. */
93 if (blkif->be->dev->state != XenbusStateConnected)
96 err = blkback_name(blkif, name);
98 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
102 err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
104 xenbus_dev_error(blkif->be->dev, err, "block flush");
107 invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
109 for (i = 0; i < blkif->nr_rings; i++) {
110 ring = &blkif->rings[i];
111 ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
112 if (IS_ERR(ring->xenblkd)) {
113 err = PTR_ERR(ring->xenblkd);
114 ring->xenblkd = NULL;
115 xenbus_dev_fatal(blkif->be->dev, err,
116 "start %s-%d xenblkd", name, i);
124 ring = &blkif->rings[i];
125 kthread_stop(ring->xenblkd);
130 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
134 blkif->rings = kcalloc(blkif->nr_rings, sizeof(struct xen_blkif_ring),
139 for (r = 0; r < blkif->nr_rings; r++) {
140 struct xen_blkif_ring *ring = &blkif->rings[r];
142 spin_lock_init(&ring->blk_ring_lock);
143 init_waitqueue_head(&ring->wq);
144 INIT_LIST_HEAD(&ring->pending_free);
145 INIT_LIST_HEAD(&ring->persistent_purge_list);
146 INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
147 gnttab_page_cache_init(&ring->free_pages);
149 spin_lock_init(&ring->pending_free_lock);
150 init_waitqueue_head(&ring->pending_free_wq);
151 init_waitqueue_head(&ring->shutdown_wq);
153 ring->st_print = jiffies;
160 /* Enable the persistent grants feature. */
161 static bool feature_persistent = true;
162 module_param(feature_persistent, bool, 0644);
163 MODULE_PARM_DESC(feature_persistent, "Enables the persistent grants feature");
165 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
167 struct xen_blkif *blkif;
169 BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
171 blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
173 return ERR_PTR(-ENOMEM);
175 blkif->domid = domid;
176 atomic_set(&blkif->refcnt, 1);
177 init_completion(&blkif->drain_complete);
180 * Because freeing back to the cache may be deferred, it is not
181 * safe to unload the module (and hence destroy the cache) until
182 * this has completed. To prevent premature unloading, take an
183 * extra module reference here and release only when the object
184 * has been freed back to the cache.
186 __module_get(THIS_MODULE);
187 INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
192 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
193 unsigned int nr_grefs, unsigned int evtchn)
196 struct xen_blkif *blkif = ring->blkif;
197 const struct blkif_common_sring *sring_common;
198 RING_IDX rsp_prod, req_prod;
201 /* Already connected through? */
205 err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
210 sring_common = (struct blkif_common_sring *)ring->blk_ring;
211 rsp_prod = READ_ONCE(sring_common->rsp_prod);
212 req_prod = READ_ONCE(sring_common->req_prod);
214 switch (blkif->blk_protocol) {
215 case BLKIF_PROTOCOL_NATIVE:
217 struct blkif_sring *sring_native =
218 (struct blkif_sring *)ring->blk_ring;
220 BACK_RING_ATTACH(&ring->blk_rings.native, sring_native,
221 rsp_prod, XEN_PAGE_SIZE * nr_grefs);
222 size = __RING_SIZE(sring_native, XEN_PAGE_SIZE * nr_grefs);
225 case BLKIF_PROTOCOL_X86_32:
227 struct blkif_x86_32_sring *sring_x86_32 =
228 (struct blkif_x86_32_sring *)ring->blk_ring;
230 BACK_RING_ATTACH(&ring->blk_rings.x86_32, sring_x86_32,
231 rsp_prod, XEN_PAGE_SIZE * nr_grefs);
232 size = __RING_SIZE(sring_x86_32, XEN_PAGE_SIZE * nr_grefs);
235 case BLKIF_PROTOCOL_X86_64:
237 struct blkif_x86_64_sring *sring_x86_64 =
238 (struct blkif_x86_64_sring *)ring->blk_ring;
240 BACK_RING_ATTACH(&ring->blk_rings.x86_64, sring_x86_64,
241 rsp_prod, XEN_PAGE_SIZE * nr_grefs);
242 size = __RING_SIZE(sring_x86_64, XEN_PAGE_SIZE * nr_grefs);
250 if (req_prod - rsp_prod > size)
253 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(blkif->domid,
254 evtchn, xen_blkif_be_int, 0, "blkif-backend", ring);
262 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
263 ring->blk_rings.common.sring = NULL;
267 static int xen_blkif_disconnect(struct xen_blkif *blkif)
269 struct pending_req *req, *n;
273 for (r = 0; r < blkif->nr_rings; r++) {
274 struct xen_blkif_ring *ring = &blkif->rings[r];
281 kthread_stop(ring->xenblkd);
282 ring->xenblkd = NULL;
283 wake_up(&ring->shutdown_wq);
286 /* The above kthread_stop() guarantees that at this point we
287 * don't have any discard_io or other_io requests. So, checking
288 * for inflight IO is enough.
290 if (atomic_read(&ring->inflight) > 0) {
296 unbind_from_irqhandler(ring->irq, ring);
300 if (ring->blk_rings.common.sring) {
301 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
302 ring->blk_rings.common.sring = NULL;
305 /* Remove all persistent grants and the cache of ballooned pages. */
306 xen_blkbk_free_caches(ring);
308 /* Check that there is no request in use */
309 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
310 list_del(&req->free_list);
312 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
313 kfree(req->segments[j]);
315 for (j = 0; j < MAX_INDIRECT_PAGES; j++)
316 kfree(req->indirect_pages[j]);
322 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
323 BUG_ON(!list_empty(&ring->persistent_purge_list));
324 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
325 BUG_ON(ring->free_pages.num_pages != 0);
326 BUG_ON(ring->persistent_gnt_c != 0);
327 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
328 ring->active = false;
333 blkif->nr_ring_pages = 0;
335 * blkif->rings was allocated in connect_ring, so we should free it in
345 static void xen_blkif_free(struct xen_blkif *blkif)
347 WARN_ON(xen_blkif_disconnect(blkif));
348 xen_vbd_free(&blkif->vbd);
349 kfree(blkif->be->mode);
352 /* Make sure everything is drained before shutting down */
353 kmem_cache_free(xen_blkif_cachep, blkif);
354 module_put(THIS_MODULE);
357 int __init xen_blkif_interface_init(void)
359 xen_blkif_cachep = kmem_cache_create("blkif_cache",
360 sizeof(struct xen_blkif),
362 if (!xen_blkif_cachep)
368 void xen_blkif_interface_fini(void)
370 kmem_cache_destroy(xen_blkif_cachep);
371 xen_blkif_cachep = NULL;
375 * sysfs interface for VBD I/O requests
378 #define VBD_SHOW_ALLRING(name, format) \
379 static ssize_t show_##name(struct device *_dev, \
380 struct device_attribute *attr, \
383 struct xenbus_device *dev = to_xenbus_device(_dev); \
384 struct backend_info *be = dev_get_drvdata(&dev->dev); \
385 struct xen_blkif *blkif = be->blkif; \
387 unsigned long long result = 0; \
392 for (i = 0; i < blkif->nr_rings; i++) { \
393 struct xen_blkif_ring *ring = &blkif->rings[i]; \
395 result += ring->st_##name; \
399 return sprintf(buf, format, result); \
401 static DEVICE_ATTR(name, 0444, show_##name, NULL)
403 VBD_SHOW_ALLRING(oo_req, "%llu\n");
404 VBD_SHOW_ALLRING(rd_req, "%llu\n");
405 VBD_SHOW_ALLRING(wr_req, "%llu\n");
406 VBD_SHOW_ALLRING(f_req, "%llu\n");
407 VBD_SHOW_ALLRING(ds_req, "%llu\n");
408 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
409 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
411 static struct attribute *xen_vbdstat_attrs[] = {
412 &dev_attr_oo_req.attr,
413 &dev_attr_rd_req.attr,
414 &dev_attr_wr_req.attr,
415 &dev_attr_f_req.attr,
416 &dev_attr_ds_req.attr,
417 &dev_attr_rd_sect.attr,
418 &dev_attr_wr_sect.attr,
422 static const struct attribute_group xen_vbdstat_group = {
423 .name = "statistics",
424 .attrs = xen_vbdstat_attrs,
427 #define VBD_SHOW(name, format, args...) \
428 static ssize_t show_##name(struct device *_dev, \
429 struct device_attribute *attr, \
432 struct xenbus_device *dev = to_xenbus_device(_dev); \
433 struct backend_info *be = dev_get_drvdata(&dev->dev); \
435 return sprintf(buf, format, ##args); \
437 static DEVICE_ATTR(name, 0444, show_##name, NULL)
439 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
440 VBD_SHOW(mode, "%s\n", be->mode);
442 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
446 error = device_create_file(&dev->dev, &dev_attr_physical_device);
450 error = device_create_file(&dev->dev, &dev_attr_mode);
454 error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
460 fail3: sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
461 fail2: device_remove_file(&dev->dev, &dev_attr_mode);
462 fail1: device_remove_file(&dev->dev, &dev_attr_physical_device);
466 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
468 sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
469 device_remove_file(&dev->dev, &dev_attr_mode);
470 device_remove_file(&dev->dev, &dev_attr_physical_device);
473 static void xen_vbd_free(struct xen_vbd *vbd)
476 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
480 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
481 unsigned major, unsigned minor, int readonly,
485 struct block_device *bdev;
486 struct request_queue *q;
489 vbd->handle = handle;
490 vbd->readonly = readonly;
493 vbd->pdevice = MKDEV(major, minor);
495 bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
496 FMODE_READ : FMODE_WRITE, NULL);
499 pr_warn("xen_vbd_create: device %08x could not be opened\n",
505 if (vbd->bdev->bd_disk == NULL) {
506 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
511 vbd->size = vbd_sz(vbd);
513 if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
514 vbd->type |= VDISK_CDROM;
515 if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
516 vbd->type |= VDISK_REMOVABLE;
518 q = bdev_get_queue(bdev);
519 if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
520 vbd->flush_support = true;
522 if (q && blk_queue_secure_erase(q))
523 vbd->discard_secure = true;
525 pr_debug("Successful creation of handle=%04x (dom=%u)\n",
526 handle, blkif->domid);
530 static int xen_blkbk_remove(struct xenbus_device *dev)
532 struct backend_info *be = dev_get_drvdata(&dev->dev);
534 pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
536 if (be->major || be->minor)
537 xenvbd_sysfs_delif(dev);
539 if (be->backend_watch.node) {
540 unregister_xenbus_watch(&be->backend_watch);
541 kfree(be->backend_watch.node);
542 be->backend_watch.node = NULL;
545 dev_set_drvdata(&dev->dev, NULL);
548 xen_blkif_disconnect(be->blkif);
550 /* Put the reference we set in xen_blkif_alloc(). */
551 xen_blkif_put(be->blkif);
557 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
558 struct backend_info *be, int state)
560 struct xenbus_device *dev = be->dev;
563 err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
566 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
571 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
573 struct xenbus_device *dev = be->dev;
574 struct xen_blkif *blkif = be->blkif;
577 struct block_device *bdev = be->blkif->vbd.bdev;
578 struct request_queue *q = bdev_get_queue(bdev);
580 if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
583 if (blk_queue_discard(q)) {
584 err = xenbus_printf(xbt, dev->nodename,
585 "discard-granularity", "%u",
586 q->limits.discard_granularity);
588 dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
591 err = xenbus_printf(xbt, dev->nodename,
592 "discard-alignment", "%u",
593 q->limits.discard_alignment);
595 dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
600 err = xenbus_printf(xbt, dev->nodename,
601 "discard-secure", "%d",
602 blkif->vbd.discard_secure);
604 dev_warn(&dev->dev, "writing discard-secure (%d)", err);
608 err = xenbus_printf(xbt, dev->nodename, "feature-discard",
611 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
614 int xen_blkbk_barrier(struct xenbus_transaction xbt,
615 struct backend_info *be, int state)
617 struct xenbus_device *dev = be->dev;
620 err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
623 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
629 * Entry point to this code when a new device is created. Allocate the basic
630 * structures, and watch the store waiting for the hotplug scripts to tell us
631 * the device's physical major and minor numbers. Switch to InitWait.
633 static int xen_blkbk_probe(struct xenbus_device *dev,
634 const struct xenbus_device_id *id)
637 struct backend_info *be = kzalloc(sizeof(struct backend_info),
640 /* match the pr_debug in xen_blkbk_remove */
641 pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
644 xenbus_dev_fatal(dev, -ENOMEM,
645 "allocating backend structure");
649 dev_set_drvdata(&dev->dev, be);
651 be->blkif = xen_blkif_alloc(dev->otherend_id);
652 if (IS_ERR(be->blkif)) {
653 err = PTR_ERR(be->blkif);
655 xenbus_dev_fatal(dev, err, "creating block interface");
659 err = xenbus_printf(XBT_NIL, dev->nodename,
660 "feature-max-indirect-segments", "%u",
661 MAX_INDIRECT_SEGMENTS);
664 "writing %s/feature-max-indirect-segments (%d)",
667 /* Multi-queue: advertise how many queues are supported by us.*/
668 err = xenbus_printf(XBT_NIL, dev->nodename,
669 "multi-queue-max-queues", "%u", xenblk_max_queues);
671 pr_warn("Error writing multi-queue-max-queues\n");
673 /* setup back pointer */
676 err = xenbus_watch_pathfmt(dev, &be->backend_watch, NULL,
678 "%s/%s", dev->nodename, "physical-device");
682 err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
683 xen_blkif_max_ring_order);
685 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
687 err = xenbus_switch_state(dev, XenbusStateInitWait);
694 pr_warn("%s failed\n", __func__);
695 xen_blkbk_remove(dev);
700 * Callback received when the hotplug scripts have placed the physical-device
701 * node. Read it and the mode node, and create a vbd. If the frontend is
704 static void backend_changed(struct xenbus_watch *watch,
705 const char *path, const char *token)
710 struct backend_info *be
711 = container_of(watch, struct backend_info, backend_watch);
712 struct xenbus_device *dev = be->dev;
714 unsigned long handle;
717 pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
719 err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
721 if (XENBUS_EXIST_ERR(err)) {
723 * Since this watch will fire once immediately after it is
724 * registered, we expect this. Ignore it, and wait for the
730 xenbus_dev_fatal(dev, err, "reading physical-device");
734 if (be->major | be->minor) {
735 if (be->major != major || be->minor != minor)
736 pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
737 be->major, be->minor, major, minor);
741 be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
742 if (IS_ERR(be->mode)) {
743 err = PTR_ERR(be->mode);
745 xenbus_dev_fatal(dev, err, "reading mode");
749 device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
750 if (!IS_ERR(device_type)) {
751 cdrom = strcmp(device_type, "cdrom") == 0;
755 /* Front end dir is a number, which is used as the handle. */
756 err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
766 err = xen_vbd_create(be->blkif, handle, major, minor,
767 !strchr(be->mode, 'w'), cdrom);
770 xenbus_dev_fatal(dev, err, "creating vbd structure");
772 err = xenvbd_sysfs_addif(dev);
774 xen_vbd_free(&be->blkif->vbd);
775 xenbus_dev_fatal(dev, err, "creating sysfs entries");
785 /* We're potentially connected now */
786 xen_update_blkif_status(be->blkif);
791 * Callback received when the frontend's state changes.
793 static void frontend_changed(struct xenbus_device *dev,
794 enum xenbus_state frontend_state)
796 struct backend_info *be = dev_get_drvdata(&dev->dev);
799 pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
801 switch (frontend_state) {
802 case XenbusStateInitialising:
803 if (dev->state == XenbusStateClosed) {
804 pr_info("%s: prepare for reconnect\n", dev->nodename);
805 xenbus_switch_state(dev, XenbusStateInitWait);
809 case XenbusStateInitialised:
810 case XenbusStateConnected:
812 * Ensure we connect even when two watches fire in
813 * close succession and we miss the intermediate value
816 if (dev->state == XenbusStateConnected)
820 * Enforce precondition before potential leak point.
821 * xen_blkif_disconnect() is idempotent.
823 err = xen_blkif_disconnect(be->blkif);
825 xenbus_dev_fatal(dev, err, "pending I/O");
829 err = connect_ring(be);
832 * Clean up so that memory resources can be used by
833 * other devices. connect_ring reported already error.
835 xen_blkif_disconnect(be->blkif);
838 xen_update_blkif_status(be->blkif);
841 case XenbusStateClosing:
842 xenbus_switch_state(dev, XenbusStateClosing);
845 case XenbusStateClosed:
846 xen_blkif_disconnect(be->blkif);
847 xenbus_switch_state(dev, XenbusStateClosed);
848 if (xenbus_dev_is_online(dev))
852 case XenbusStateUnknown:
853 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
854 device_unregister(&dev->dev);
858 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
864 /* Once a memory pressure is detected, squeeze free page pools for a while. */
865 static unsigned int buffer_squeeze_duration_ms = 10;
866 module_param_named(buffer_squeeze_duration_ms,
867 buffer_squeeze_duration_ms, int, 0644);
868 MODULE_PARM_DESC(buffer_squeeze_duration_ms,
869 "Duration in ms to squeeze pages buffer when a memory pressure is detected");
872 * Callback received when the memory pressure is detected.
874 static void reclaim_memory(struct xenbus_device *dev)
876 struct backend_info *be = dev_get_drvdata(&dev->dev);
880 be->blkif->buffer_squeeze_end = jiffies +
881 msecs_to_jiffies(buffer_squeeze_duration_ms);
884 /* ** Connection ** */
887 * Write the physical details regarding the block device to the store, and
888 * switch to Connected state.
890 static void connect(struct backend_info *be)
892 struct xenbus_transaction xbt;
894 struct xenbus_device *dev = be->dev;
896 pr_debug("%s %s\n", __func__, dev->otherend);
898 /* Supply the information about the device the frontend needs */
900 err = xenbus_transaction_start(&xbt);
902 xenbus_dev_fatal(dev, err, "starting transaction");
906 /* If we can't advertise it is OK. */
907 xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
909 xen_blkbk_discard(xbt, be);
911 xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
913 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
914 be->blkif->vbd.feature_gnt_persistent_parm);
916 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
921 err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
922 (unsigned long long)vbd_sz(&be->blkif->vbd));
924 xenbus_dev_fatal(dev, err, "writing %s/sectors",
929 /* FIXME: use a typename instead */
930 err = xenbus_printf(xbt, dev->nodename, "info", "%u",
931 be->blkif->vbd.type |
932 (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
934 xenbus_dev_fatal(dev, err, "writing %s/info",
938 err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
940 bdev_logical_block_size(be->blkif->vbd.bdev));
942 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
946 err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
947 bdev_physical_block_size(be->blkif->vbd.bdev));
949 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
952 err = xenbus_transaction_end(xbt, 0);
956 xenbus_dev_fatal(dev, err, "ending transaction");
958 err = xenbus_switch_state(dev, XenbusStateConnected);
960 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
965 xenbus_transaction_end(xbt, 1);
969 * Each ring may have multi pages, depends on "ring-page-order".
971 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
973 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
974 struct pending_req *req, *n;
976 struct xen_blkif *blkif = ring->blkif;
977 struct xenbus_device *dev = blkif->be->dev;
978 unsigned int nr_grefs, evtchn;
980 err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
984 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
988 nr_grefs = blkif->nr_ring_pages;
990 if (unlikely(!nr_grefs)) {
995 for (i = 0; i < nr_grefs; i++) {
996 char ring_ref_name[RINGREF_NAME_LEN];
998 if (blkif->multi_ref)
999 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1002 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref");
1005 err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
1006 "%u", &ring_ref[i]);
1010 xenbus_dev_fatal(dev, err, "reading %s/%s",
1011 dir, ring_ref_name);
1017 for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
1018 req = kzalloc(sizeof(*req), GFP_KERNEL);
1021 list_add_tail(&req->free_list, &ring->pending_free);
1022 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1023 req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
1024 if (!req->segments[j])
1027 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1028 req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
1030 if (!req->indirect_pages[j])
1035 /* Map the shared frame, irq etc. */
1036 err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
1038 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
1045 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1046 list_del(&req->free_list);
1047 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1048 if (!req->segments[j])
1050 kfree(req->segments[j]);
1052 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1053 if (!req->indirect_pages[j])
1055 kfree(req->indirect_pages[j]);
1062 static int connect_ring(struct backend_info *be)
1064 struct xenbus_device *dev = be->dev;
1065 struct xen_blkif *blkif = be->blkif;
1066 char protocol[64] = "";
1070 const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1071 unsigned int requested_num_queues = 0;
1072 unsigned int ring_page_order;
1074 pr_debug("%s %s\n", __func__, dev->otherend);
1076 blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1077 err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1080 strcpy(protocol, "unspecified, assuming default");
1081 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1082 blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1083 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1084 blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1085 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1086 blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1088 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1092 blkif->vbd.feature_gnt_persistent_parm = feature_persistent;
1093 blkif->vbd.feature_gnt_persistent =
1094 blkif->vbd.feature_gnt_persistent_parm &&
1095 xenbus_read_unsigned(dev->otherend, "feature-persistent", 0);
1097 blkif->vbd.overflow_max_grants = 0;
1100 * Read the number of hardware queues from frontend.
1102 requested_num_queues = xenbus_read_unsigned(dev->otherend,
1103 "multi-queue-num-queues",
1105 if (requested_num_queues > xenblk_max_queues
1106 || requested_num_queues == 0) {
1107 /* Buggy or malicious guest. */
1108 xenbus_dev_fatal(dev, err,
1109 "guest requested %u queues, exceeding the maximum of %u.",
1110 requested_num_queues, xenblk_max_queues);
1113 blkif->nr_rings = requested_num_queues;
1114 if (xen_blkif_alloc_rings(blkif))
1117 pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1118 blkif->nr_rings, blkif->blk_protocol, protocol,
1119 blkif->vbd.feature_gnt_persistent ? "persistent grants" : "");
1121 err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
1124 blkif->nr_ring_pages = 1;
1125 blkif->multi_ref = false;
1126 } else if (ring_page_order <= xen_blkif_max_ring_order) {
1127 blkif->nr_ring_pages = 1 << ring_page_order;
1128 blkif->multi_ref = true;
1131 xenbus_dev_fatal(dev, err,
1132 "requested ring page order %d exceed max:%d",
1134 xen_blkif_max_ring_order);
1138 if (blkif->nr_rings == 1)
1139 return read_per_ring_refs(&blkif->rings[0], dev->otherend);
1141 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1142 xspath = kmalloc(xspathsize, GFP_KERNEL);
1144 xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1148 for (i = 0; i < blkif->nr_rings; i++) {
1149 memset(xspath, 0, xspathsize);
1150 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1151 err = read_per_ring_refs(&blkif->rings[i], xspath);
1162 static const struct xenbus_device_id xen_blkbk_ids[] = {
1167 static struct xenbus_driver xen_blkbk_driver = {
1168 .ids = xen_blkbk_ids,
1169 .probe = xen_blkbk_probe,
1170 .remove = xen_blkbk_remove,
1171 .otherend_changed = frontend_changed,
1172 .allow_rebind = true,
1173 .reclaim_memory = reclaim_memory,
1176 int xen_blkif_xenbus_init(void)
1178 return xenbus_register_backend(&xen_blkbk_driver);
1181 void xen_blkif_xenbus_fini(void)
1183 xenbus_unregister_driver(&xen_blkbk_driver);