xref: /linux/drivers/block/xen-blkback/xenbus.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*  Xenbus code for blkif backend
2     Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
3     Copyright (C) 2005 XenSource Ltd
4 
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.
9 
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.
14 
15 */
16 
17 #define pr_fmt(fmt) "xen-blkback: " fmt
18 
19 #include <stdarg.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 #include <xen/events.h>
23 #include <xen/grant_table.h>
24 #include "common.h"
25 
26 /* Enlarge the array size in order to fully show blkback name. */
27 #define BLKBACK_NAME_LEN (20)
28 #define RINGREF_NAME_LEN (20)
29 
30 struct backend_info {
31 	struct xenbus_device	*dev;
32 	struct xen_blkif	*blkif;
33 	struct xenbus_watch	backend_watch;
34 	unsigned		major;
35 	unsigned		minor;
36 	char			*mode;
37 };
38 
39 static struct kmem_cache *xen_blkif_cachep;
40 static void connect(struct backend_info *);
41 static int connect_ring(struct backend_info *);
42 static void backend_changed(struct xenbus_watch *, const char **,
43 			    unsigned int);
44 static void xen_blkif_free(struct xen_blkif *blkif);
45 static void xen_vbd_free(struct xen_vbd *vbd);
46 
47 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
48 {
49 	return be->dev;
50 }
51 
52 /*
53  * The last request could free the device from softirq context and
54  * xen_blkif_free() can sleep.
55  */
56 static void xen_blkif_deferred_free(struct work_struct *work)
57 {
58 	struct xen_blkif *blkif;
59 
60 	blkif = container_of(work, struct xen_blkif, free_work);
61 	xen_blkif_free(blkif);
62 }
63 
64 static int blkback_name(struct xen_blkif *blkif, char *buf)
65 {
66 	char *devpath, *devname;
67 	struct xenbus_device *dev = blkif->be->dev;
68 
69 	devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
70 	if (IS_ERR(devpath))
71 		return PTR_ERR(devpath);
72 
73 	devname = strstr(devpath, "/dev/");
74 	if (devname != NULL)
75 		devname += strlen("/dev/");
76 	else
77 		devname  = devpath;
78 
79 	snprintf(buf, BLKBACK_NAME_LEN, "blkback.%d.%s", blkif->domid, devname);
80 	kfree(devpath);
81 
82 	return 0;
83 }
84 
85 static void xen_update_blkif_status(struct xen_blkif *blkif)
86 {
87 	int err;
88 	char name[BLKBACK_NAME_LEN];
89 
90 	/* Not ready to connect? */
91 	if (!blkif->irq || !blkif->vbd.bdev)
92 		return;
93 
94 	/* Already connected? */
95 	if (blkif->be->dev->state == XenbusStateConnected)
96 		return;
97 
98 	/* Attempt to connect: exit if we fail to. */
99 	connect(blkif->be);
100 	if (blkif->be->dev->state != XenbusStateConnected)
101 		return;
102 
103 	err = blkback_name(blkif, name);
104 	if (err) {
105 		xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
106 		return;
107 	}
108 
109 	err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
110 	if (err) {
111 		xenbus_dev_error(blkif->be->dev, err, "block flush");
112 		return;
113 	}
114 	invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
115 
116 	blkif->xenblkd = kthread_run(xen_blkif_schedule, blkif, "%s", name);
117 	if (IS_ERR(blkif->xenblkd)) {
118 		err = PTR_ERR(blkif->xenblkd);
119 		blkif->xenblkd = NULL;
120 		xenbus_dev_error(blkif->be->dev, err, "start xenblkd");
121 		return;
122 	}
123 }
124 
125 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
126 {
127 	struct xen_blkif *blkif;
128 
129 	BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
130 
131 	blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
132 	if (!blkif)
133 		return ERR_PTR(-ENOMEM);
134 
135 	blkif->domid = domid;
136 	spin_lock_init(&blkif->blk_ring_lock);
137 	atomic_set(&blkif->refcnt, 1);
138 	init_waitqueue_head(&blkif->wq);
139 	init_completion(&blkif->drain_complete);
140 	atomic_set(&blkif->drain, 0);
141 	blkif->st_print = jiffies;
142 	blkif->persistent_gnts.rb_node = NULL;
143 	spin_lock_init(&blkif->free_pages_lock);
144 	INIT_LIST_HEAD(&blkif->free_pages);
145 	INIT_LIST_HEAD(&blkif->persistent_purge_list);
146 	blkif->free_pages_num = 0;
147 	atomic_set(&blkif->persistent_gnt_in_use, 0);
148 	atomic_set(&blkif->inflight, 0);
149 	INIT_WORK(&blkif->persistent_purge_work, xen_blkbk_unmap_purged_grants);
150 
151 	INIT_LIST_HEAD(&blkif->pending_free);
152 	INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
153 	spin_lock_init(&blkif->pending_free_lock);
154 	init_waitqueue_head(&blkif->pending_free_wq);
155 	init_waitqueue_head(&blkif->shutdown_wq);
156 
157 	return blkif;
158 }
159 
160 static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t *gref,
161 			 unsigned int nr_grefs, unsigned int evtchn)
162 {
163 	int err;
164 
165 	/* Already connected through? */
166 	if (blkif->irq)
167 		return 0;
168 
169 	err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
170 				     &blkif->blk_ring);
171 	if (err < 0)
172 		return err;
173 
174 	switch (blkif->blk_protocol) {
175 	case BLKIF_PROTOCOL_NATIVE:
176 	{
177 		struct blkif_sring *sring;
178 		sring = (struct blkif_sring *)blkif->blk_ring;
179 		BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE * nr_grefs);
180 		break;
181 	}
182 	case BLKIF_PROTOCOL_X86_32:
183 	{
184 		struct blkif_x86_32_sring *sring_x86_32;
185 		sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring;
186 		BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE * nr_grefs);
187 		break;
188 	}
189 	case BLKIF_PROTOCOL_X86_64:
190 	{
191 		struct blkif_x86_64_sring *sring_x86_64;
192 		sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring;
193 		BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE * nr_grefs);
194 		break;
195 	}
196 	default:
197 		BUG();
198 	}
199 
200 	err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
201 						    xen_blkif_be_int, 0,
202 						    "blkif-backend", blkif);
203 	if (err < 0) {
204 		xenbus_unmap_ring_vfree(blkif->be->dev, blkif->blk_ring);
205 		blkif->blk_rings.common.sring = NULL;
206 		return err;
207 	}
208 	blkif->irq = err;
209 
210 	return 0;
211 }
212 
213 static int xen_blkif_disconnect(struct xen_blkif *blkif)
214 {
215 	struct pending_req *req, *n;
216 	int i = 0, j;
217 
218 	if (blkif->xenblkd) {
219 		kthread_stop(blkif->xenblkd);
220 		wake_up(&blkif->shutdown_wq);
221 		blkif->xenblkd = NULL;
222 	}
223 
224 	/* The above kthread_stop() guarantees that at this point we
225 	 * don't have any discard_io or other_io requests. So, checking
226 	 * for inflight IO is enough.
227 	 */
228 	if (atomic_read(&blkif->inflight) > 0)
229 		return -EBUSY;
230 
231 	if (blkif->irq) {
232 		unbind_from_irqhandler(blkif->irq, blkif);
233 		blkif->irq = 0;
234 	}
235 
236 	if (blkif->blk_rings.common.sring) {
237 		xenbus_unmap_ring_vfree(blkif->be->dev, blkif->blk_ring);
238 		blkif->blk_rings.common.sring = NULL;
239 	}
240 
241 	/* Remove all persistent grants and the cache of ballooned pages. */
242 	xen_blkbk_free_caches(blkif);
243 
244 	/* Check that there is no request in use */
245 	list_for_each_entry_safe(req, n, &blkif->pending_free, free_list) {
246 		list_del(&req->free_list);
247 
248 		for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
249 			kfree(req->segments[j]);
250 
251 		for (j = 0; j < MAX_INDIRECT_PAGES; j++)
252 			kfree(req->indirect_pages[j]);
253 
254 		kfree(req);
255 		i++;
256 	}
257 
258 	WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
259 	blkif->nr_ring_pages = 0;
260 
261 	return 0;
262 }
263 
264 static void xen_blkif_free(struct xen_blkif *blkif)
265 {
266 
267 	xen_blkif_disconnect(blkif);
268 	xen_vbd_free(&blkif->vbd);
269 
270 	/* Make sure everything is drained before shutting down */
271 	BUG_ON(blkif->persistent_gnt_c != 0);
272 	BUG_ON(atomic_read(&blkif->persistent_gnt_in_use) != 0);
273 	BUG_ON(blkif->free_pages_num != 0);
274 	BUG_ON(!list_empty(&blkif->persistent_purge_list));
275 	BUG_ON(!list_empty(&blkif->free_pages));
276 	BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts));
277 
278 	kmem_cache_free(xen_blkif_cachep, blkif);
279 }
280 
281 int __init xen_blkif_interface_init(void)
282 {
283 	xen_blkif_cachep = kmem_cache_create("blkif_cache",
284 					     sizeof(struct xen_blkif),
285 					     0, 0, NULL);
286 	if (!xen_blkif_cachep)
287 		return -ENOMEM;
288 
289 	return 0;
290 }
291 
292 /*
293  *  sysfs interface for VBD I/O requests
294  */
295 
296 #define VBD_SHOW(name, format, args...)					\
297 	static ssize_t show_##name(struct device *_dev,			\
298 				   struct device_attribute *attr,	\
299 				   char *buf)				\
300 	{								\
301 		struct xenbus_device *dev = to_xenbus_device(_dev);	\
302 		struct backend_info *be = dev_get_drvdata(&dev->dev);	\
303 									\
304 		return sprintf(buf, format, ##args);			\
305 	}								\
306 	static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
307 
308 VBD_SHOW(oo_req,  "%llu\n", be->blkif->st_oo_req);
309 VBD_SHOW(rd_req,  "%llu\n", be->blkif->st_rd_req);
310 VBD_SHOW(wr_req,  "%llu\n", be->blkif->st_wr_req);
311 VBD_SHOW(f_req,  "%llu\n", be->blkif->st_f_req);
312 VBD_SHOW(ds_req,  "%llu\n", be->blkif->st_ds_req);
313 VBD_SHOW(rd_sect, "%llu\n", be->blkif->st_rd_sect);
314 VBD_SHOW(wr_sect, "%llu\n", be->blkif->st_wr_sect);
315 
316 static struct attribute *xen_vbdstat_attrs[] = {
317 	&dev_attr_oo_req.attr,
318 	&dev_attr_rd_req.attr,
319 	&dev_attr_wr_req.attr,
320 	&dev_attr_f_req.attr,
321 	&dev_attr_ds_req.attr,
322 	&dev_attr_rd_sect.attr,
323 	&dev_attr_wr_sect.attr,
324 	NULL
325 };
326 
327 static struct attribute_group xen_vbdstat_group = {
328 	.name = "statistics",
329 	.attrs = xen_vbdstat_attrs,
330 };
331 
332 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
333 VBD_SHOW(mode, "%s\n", be->mode);
334 
335 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
336 {
337 	int error;
338 
339 	error = device_create_file(&dev->dev, &dev_attr_physical_device);
340 	if (error)
341 		goto fail1;
342 
343 	error = device_create_file(&dev->dev, &dev_attr_mode);
344 	if (error)
345 		goto fail2;
346 
347 	error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
348 	if (error)
349 		goto fail3;
350 
351 	return 0;
352 
353 fail3:	sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
354 fail2:	device_remove_file(&dev->dev, &dev_attr_mode);
355 fail1:	device_remove_file(&dev->dev, &dev_attr_physical_device);
356 	return error;
357 }
358 
359 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
360 {
361 	sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
362 	device_remove_file(&dev->dev, &dev_attr_mode);
363 	device_remove_file(&dev->dev, &dev_attr_physical_device);
364 }
365 
366 
367 static void xen_vbd_free(struct xen_vbd *vbd)
368 {
369 	if (vbd->bdev)
370 		blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
371 	vbd->bdev = NULL;
372 }
373 
374 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
375 			  unsigned major, unsigned minor, int readonly,
376 			  int cdrom)
377 {
378 	struct xen_vbd *vbd;
379 	struct block_device *bdev;
380 	struct request_queue *q;
381 
382 	vbd = &blkif->vbd;
383 	vbd->handle   = handle;
384 	vbd->readonly = readonly;
385 	vbd->type     = 0;
386 
387 	vbd->pdevice  = MKDEV(major, minor);
388 
389 	bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
390 				 FMODE_READ : FMODE_WRITE, NULL);
391 
392 	if (IS_ERR(bdev)) {
393 		pr_warn("xen_vbd_create: device %08x could not be opened\n",
394 			vbd->pdevice);
395 		return -ENOENT;
396 	}
397 
398 	vbd->bdev = bdev;
399 	if (vbd->bdev->bd_disk == NULL) {
400 		pr_warn("xen_vbd_create: device %08x doesn't exist\n",
401 			vbd->pdevice);
402 		xen_vbd_free(vbd);
403 		return -ENOENT;
404 	}
405 	vbd->size = vbd_sz(vbd);
406 
407 	if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
408 		vbd->type |= VDISK_CDROM;
409 	if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
410 		vbd->type |= VDISK_REMOVABLE;
411 
412 	q = bdev_get_queue(bdev);
413 	if (q && q->flush_flags)
414 		vbd->flush_support = true;
415 
416 	if (q && blk_queue_secdiscard(q))
417 		vbd->discard_secure = true;
418 
419 	pr_debug("Successful creation of handle=%04x (dom=%u)\n",
420 		handle, blkif->domid);
421 	return 0;
422 }
423 static int xen_blkbk_remove(struct xenbus_device *dev)
424 {
425 	struct backend_info *be = dev_get_drvdata(&dev->dev);
426 
427 	pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
428 
429 	if (be->major || be->minor)
430 		xenvbd_sysfs_delif(dev);
431 
432 	if (be->backend_watch.node) {
433 		unregister_xenbus_watch(&be->backend_watch);
434 		kfree(be->backend_watch.node);
435 		be->backend_watch.node = NULL;
436 	}
437 
438 	dev_set_drvdata(&dev->dev, NULL);
439 
440 	if (be->blkif) {
441 		xen_blkif_disconnect(be->blkif);
442 		xen_blkif_put(be->blkif);
443 	}
444 
445 	kfree(be->mode);
446 	kfree(be);
447 	return 0;
448 }
449 
450 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
451 			      struct backend_info *be, int state)
452 {
453 	struct xenbus_device *dev = be->dev;
454 	int err;
455 
456 	err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
457 			    "%d", state);
458 	if (err)
459 		dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
460 
461 	return err;
462 }
463 
464 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
465 {
466 	struct xenbus_device *dev = be->dev;
467 	struct xen_blkif *blkif = be->blkif;
468 	int err;
469 	int state = 0, discard_enable;
470 	struct block_device *bdev = be->blkif->vbd.bdev;
471 	struct request_queue *q = bdev_get_queue(bdev);
472 
473 	err = xenbus_scanf(XBT_NIL, dev->nodename, "discard-enable", "%d",
474 			   &discard_enable);
475 	if (err == 1 && !discard_enable)
476 		return;
477 
478 	if (blk_queue_discard(q)) {
479 		err = xenbus_printf(xbt, dev->nodename,
480 			"discard-granularity", "%u",
481 			q->limits.discard_granularity);
482 		if (err) {
483 			dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
484 			return;
485 		}
486 		err = xenbus_printf(xbt, dev->nodename,
487 			"discard-alignment", "%u",
488 			q->limits.discard_alignment);
489 		if (err) {
490 			dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
491 			return;
492 		}
493 		state = 1;
494 		/* Optional. */
495 		err = xenbus_printf(xbt, dev->nodename,
496 				    "discard-secure", "%d",
497 				    blkif->vbd.discard_secure);
498 		if (err) {
499 			dev_warn(&dev->dev, "writing discard-secure (%d)", err);
500 			return;
501 		}
502 	}
503 	err = xenbus_printf(xbt, dev->nodename, "feature-discard",
504 			    "%d", state);
505 	if (err)
506 		dev_warn(&dev->dev, "writing feature-discard (%d)", err);
507 }
508 int xen_blkbk_barrier(struct xenbus_transaction xbt,
509 		      struct backend_info *be, int state)
510 {
511 	struct xenbus_device *dev = be->dev;
512 	int err;
513 
514 	err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
515 			    "%d", state);
516 	if (err)
517 		dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
518 
519 	return err;
520 }
521 
522 /*
523  * Entry point to this code when a new device is created.  Allocate the basic
524  * structures, and watch the store waiting for the hotplug scripts to tell us
525  * the device's physical major and minor numbers.  Switch to InitWait.
526  */
527 static int xen_blkbk_probe(struct xenbus_device *dev,
528 			   const struct xenbus_device_id *id)
529 {
530 	int err;
531 	struct backend_info *be = kzalloc(sizeof(struct backend_info),
532 					  GFP_KERNEL);
533 
534 	/* match the pr_debug in xen_blkbk_remove */
535 	pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
536 
537 	if (!be) {
538 		xenbus_dev_fatal(dev, -ENOMEM,
539 				 "allocating backend structure");
540 		return -ENOMEM;
541 	}
542 	be->dev = dev;
543 	dev_set_drvdata(&dev->dev, be);
544 
545 	be->blkif = xen_blkif_alloc(dev->otherend_id);
546 	if (IS_ERR(be->blkif)) {
547 		err = PTR_ERR(be->blkif);
548 		be->blkif = NULL;
549 		xenbus_dev_fatal(dev, err, "creating block interface");
550 		goto fail;
551 	}
552 
553 	/* setup back pointer */
554 	be->blkif->be = be;
555 
556 	err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
557 				   "%s/%s", dev->nodename, "physical-device");
558 	if (err)
559 		goto fail;
560 
561 	err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
562 			    xen_blkif_max_ring_order);
563 	if (err)
564 		pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
565 
566 	err = xenbus_switch_state(dev, XenbusStateInitWait);
567 	if (err)
568 		goto fail;
569 
570 	return 0;
571 
572 fail:
573 	pr_warn("%s failed\n", __func__);
574 	xen_blkbk_remove(dev);
575 	return err;
576 }
577 
578 
579 /*
580  * Callback received when the hotplug scripts have placed the physical-device
581  * node.  Read it and the mode node, and create a vbd.  If the frontend is
582  * ready, connect.
583  */
584 static void backend_changed(struct xenbus_watch *watch,
585 			    const char **vec, unsigned int len)
586 {
587 	int err;
588 	unsigned major;
589 	unsigned minor;
590 	struct backend_info *be
591 		= container_of(watch, struct backend_info, backend_watch);
592 	struct xenbus_device *dev = be->dev;
593 	int cdrom = 0;
594 	unsigned long handle;
595 	char *device_type;
596 
597 	pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
598 
599 	err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
600 			   &major, &minor);
601 	if (XENBUS_EXIST_ERR(err)) {
602 		/*
603 		 * Since this watch will fire once immediately after it is
604 		 * registered, we expect this.  Ignore it, and wait for the
605 		 * hotplug scripts.
606 		 */
607 		return;
608 	}
609 	if (err != 2) {
610 		xenbus_dev_fatal(dev, err, "reading physical-device");
611 		return;
612 	}
613 
614 	if (be->major | be->minor) {
615 		if (be->major != major || be->minor != minor)
616 			pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
617 				be->major, be->minor, major, minor);
618 		return;
619 	}
620 
621 	be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
622 	if (IS_ERR(be->mode)) {
623 		err = PTR_ERR(be->mode);
624 		be->mode = NULL;
625 		xenbus_dev_fatal(dev, err, "reading mode");
626 		return;
627 	}
628 
629 	device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
630 	if (!IS_ERR(device_type)) {
631 		cdrom = strcmp(device_type, "cdrom") == 0;
632 		kfree(device_type);
633 	}
634 
635 	/* Front end dir is a number, which is used as the handle. */
636 	err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
637 	if (err)
638 		return;
639 
640 	be->major = major;
641 	be->minor = minor;
642 
643 	err = xen_vbd_create(be->blkif, handle, major, minor,
644 			     !strchr(be->mode, 'w'), cdrom);
645 
646 	if (err)
647 		xenbus_dev_fatal(dev, err, "creating vbd structure");
648 	else {
649 		err = xenvbd_sysfs_addif(dev);
650 		if (err) {
651 			xen_vbd_free(&be->blkif->vbd);
652 			xenbus_dev_fatal(dev, err, "creating sysfs entries");
653 		}
654 	}
655 
656 	if (err) {
657 		kfree(be->mode);
658 		be->mode = NULL;
659 		be->major = 0;
660 		be->minor = 0;
661 	} else {
662 		/* We're potentially connected now */
663 		xen_update_blkif_status(be->blkif);
664 	}
665 }
666 
667 
668 /*
669  * Callback received when the frontend's state changes.
670  */
671 static void frontend_changed(struct xenbus_device *dev,
672 			     enum xenbus_state frontend_state)
673 {
674 	struct backend_info *be = dev_get_drvdata(&dev->dev);
675 	int err;
676 
677 	pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
678 
679 	switch (frontend_state) {
680 	case XenbusStateInitialising:
681 		if (dev->state == XenbusStateClosed) {
682 			pr_info("%s: prepare for reconnect\n", dev->nodename);
683 			xenbus_switch_state(dev, XenbusStateInitWait);
684 		}
685 		break;
686 
687 	case XenbusStateInitialised:
688 	case XenbusStateConnected:
689 		/*
690 		 * Ensure we connect even when two watches fire in
691 		 * close succession and we miss the intermediate value
692 		 * of frontend_state.
693 		 */
694 		if (dev->state == XenbusStateConnected)
695 			break;
696 
697 		/*
698 		 * Enforce precondition before potential leak point.
699 		 * xen_blkif_disconnect() is idempotent.
700 		 */
701 		err = xen_blkif_disconnect(be->blkif);
702 		if (err) {
703 			xenbus_dev_fatal(dev, err, "pending I/O");
704 			break;
705 		}
706 
707 		err = connect_ring(be);
708 		if (err)
709 			break;
710 		xen_update_blkif_status(be->blkif);
711 		break;
712 
713 	case XenbusStateClosing:
714 		xenbus_switch_state(dev, XenbusStateClosing);
715 		break;
716 
717 	case XenbusStateClosed:
718 		xen_blkif_disconnect(be->blkif);
719 		xenbus_switch_state(dev, XenbusStateClosed);
720 		if (xenbus_dev_is_online(dev))
721 			break;
722 		/* fall through if not online */
723 	case XenbusStateUnknown:
724 		/* implies xen_blkif_disconnect() via xen_blkbk_remove() */
725 		device_unregister(&dev->dev);
726 		break;
727 
728 	default:
729 		xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
730 				 frontend_state);
731 		break;
732 	}
733 }
734 
735 
736 /* ** Connection ** */
737 
738 
739 /*
740  * Write the physical details regarding the block device to the store, and
741  * switch to Connected state.
742  */
743 static void connect(struct backend_info *be)
744 {
745 	struct xenbus_transaction xbt;
746 	int err;
747 	struct xenbus_device *dev = be->dev;
748 
749 	pr_debug("%s %s\n", __func__, dev->otherend);
750 
751 	/* Supply the information about the device the frontend needs */
752 again:
753 	err = xenbus_transaction_start(&xbt);
754 	if (err) {
755 		xenbus_dev_fatal(dev, err, "starting transaction");
756 		return;
757 	}
758 
759 	/* If we can't advertise it is OK. */
760 	xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
761 
762 	xen_blkbk_discard(xbt, be);
763 
764 	xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
765 
766 	err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
767 	if (err) {
768 		xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
769 				 dev->nodename);
770 		goto abort;
771 	}
772 	err = xenbus_printf(xbt, dev->nodename, "feature-max-indirect-segments", "%u",
773 			    MAX_INDIRECT_SEGMENTS);
774 	if (err)
775 		dev_warn(&dev->dev, "writing %s/feature-max-indirect-segments (%d)",
776 			 dev->nodename, err);
777 
778 	err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
779 			    (unsigned long long)vbd_sz(&be->blkif->vbd));
780 	if (err) {
781 		xenbus_dev_fatal(dev, err, "writing %s/sectors",
782 				 dev->nodename);
783 		goto abort;
784 	}
785 
786 	/* FIXME: use a typename instead */
787 	err = xenbus_printf(xbt, dev->nodename, "info", "%u",
788 			    be->blkif->vbd.type |
789 			    (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
790 	if (err) {
791 		xenbus_dev_fatal(dev, err, "writing %s/info",
792 				 dev->nodename);
793 		goto abort;
794 	}
795 	err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
796 			    (unsigned long)
797 			    bdev_logical_block_size(be->blkif->vbd.bdev));
798 	if (err) {
799 		xenbus_dev_fatal(dev, err, "writing %s/sector-size",
800 				 dev->nodename);
801 		goto abort;
802 	}
803 	err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
804 			    bdev_physical_block_size(be->blkif->vbd.bdev));
805 	if (err)
806 		xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
807 				 dev->nodename);
808 
809 	err = xenbus_transaction_end(xbt, 0);
810 	if (err == -EAGAIN)
811 		goto again;
812 	if (err)
813 		xenbus_dev_fatal(dev, err, "ending transaction");
814 
815 	err = xenbus_switch_state(dev, XenbusStateConnected);
816 	if (err)
817 		xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
818 				 dev->nodename);
819 
820 	return;
821  abort:
822 	xenbus_transaction_end(xbt, 1);
823 }
824 
825 
826 static int connect_ring(struct backend_info *be)
827 {
828 	struct xenbus_device *dev = be->dev;
829 	unsigned int ring_ref[XENBUS_MAX_RING_PAGES];
830 	unsigned int evtchn, nr_grefs, ring_page_order;
831 	unsigned int pers_grants;
832 	char protocol[64] = "";
833 	struct pending_req *req, *n;
834 	int err, i, j;
835 
836 	pr_debug("%s %s\n", __func__, dev->otherend);
837 
838 	err = xenbus_scanf(XBT_NIL, dev->otherend, "event-channel", "%u",
839 			  &evtchn);
840 	if (err != 1) {
841 		err = -EINVAL;
842 		xenbus_dev_fatal(dev, err, "reading %s/event-channel",
843 				 dev->otherend);
844 		return err;
845 	}
846 	pr_info("event-channel %u\n", evtchn);
847 
848 	err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
849 			  &ring_page_order);
850 	if (err != 1) {
851 		err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref",
852 				  "%u", &ring_ref[0]);
853 		if (err != 1) {
854 			err = -EINVAL;
855 			xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
856 					 dev->otherend);
857 			return err;
858 		}
859 		nr_grefs = 1;
860 		pr_info("%s:using single page: ring-ref %d\n", dev->otherend,
861 			ring_ref[0]);
862 	} else {
863 		unsigned int i;
864 
865 		if (ring_page_order > xen_blkif_max_ring_order) {
866 			err = -EINVAL;
867 			xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
868 					 dev->otherend, ring_page_order,
869 					 xen_blkif_max_ring_order);
870 			return err;
871 		}
872 
873 		nr_grefs = 1 << ring_page_order;
874 		for (i = 0; i < nr_grefs; i++) {
875 			char ring_ref_name[RINGREF_NAME_LEN];
876 
877 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
878 			err = xenbus_scanf(XBT_NIL, dev->otherend, ring_ref_name,
879 					   "%u", &ring_ref[i]);
880 			if (err != 1) {
881 				err = -EINVAL;
882 				xenbus_dev_fatal(dev, err, "reading %s/%s",
883 						 dev->otherend, ring_ref_name);
884 				return err;
885 			}
886 			pr_info("ring-ref%u: %u\n", i, ring_ref[i]);
887 		}
888 	}
889 
890 	be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
891 	err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
892 			    "%63s", protocol, NULL);
893 	if (err)
894 		strcpy(protocol, "unspecified, assuming default");
895 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
896 		be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
897 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
898 		be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
899 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
900 		be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
901 	else {
902 		xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
903 		return -1;
904 	}
905 	err = xenbus_gather(XBT_NIL, dev->otherend,
906 			    "feature-persistent", "%u",
907 			    &pers_grants, NULL);
908 	if (err)
909 		pers_grants = 0;
910 
911 	be->blkif->vbd.feature_gnt_persistent = pers_grants;
912 	be->blkif->vbd.overflow_max_grants = 0;
913 	be->blkif->nr_ring_pages = nr_grefs;
914 
915 	pr_info("ring-pages:%d, event-channel %d, protocol %d (%s) %s\n",
916 		nr_grefs, evtchn, be->blkif->blk_protocol, protocol,
917 		pers_grants ? "persistent grants" : "");
918 
919 	for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
920 		req = kzalloc(sizeof(*req), GFP_KERNEL);
921 		if (!req)
922 			goto fail;
923 		list_add_tail(&req->free_list, &be->blkif->pending_free);
924 		for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
925 			req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
926 			if (!req->segments[j])
927 				goto fail;
928 		}
929 		for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
930 			req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
931 							 GFP_KERNEL);
932 			if (!req->indirect_pages[j])
933 				goto fail;
934 		}
935 	}
936 
937 	/* Map the shared frame, irq etc. */
938 	err = xen_blkif_map(be->blkif, ring_ref, nr_grefs, evtchn);
939 	if (err) {
940 		xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
941 		return err;
942 	}
943 
944 	return 0;
945 
946 fail:
947 	list_for_each_entry_safe(req, n, &be->blkif->pending_free, free_list) {
948 		list_del(&req->free_list);
949 		for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
950 			if (!req->segments[j])
951 				break;
952 			kfree(req->segments[j]);
953 		}
954 		for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
955 			if (!req->indirect_pages[j])
956 				break;
957 			kfree(req->indirect_pages[j]);
958 		}
959 		kfree(req);
960 	}
961 	return -ENOMEM;
962 }
963 
964 static const struct xenbus_device_id xen_blkbk_ids[] = {
965 	{ "vbd" },
966 	{ "" }
967 };
968 
969 static struct xenbus_driver xen_blkbk_driver = {
970 	.ids  = xen_blkbk_ids,
971 	.probe = xen_blkbk_probe,
972 	.remove = xen_blkbk_remove,
973 	.otherend_changed = frontend_changed
974 };
975 
976 int xen_blkif_xenbus_init(void)
977 {
978 	return xenbus_register_backend(&xen_blkbk_driver);
979 }
980