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