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
5 
6 
7 */
8 
9 #define pr_fmt(fmt) "xen-blkback: " fmt
10 
11 #include <stdarg.h>
12 #include <linux/module.h>
13 #include <linux/kthread.h>
14 #include <xen/events.h>
15 #include <xen/grant_table.h>
16 #include "common.h"
17 
18 /* On the XenBus the max length of 'ring-ref%u'. */
19 #define RINGREF_NAME_LEN (20)
20 
21 struct backend_info {
22 	struct xenbus_device	*dev;
23 	struct xen_blkif	*blkif;
24 	struct xenbus_watch	backend_watch;
25 	unsigned		major;
26 	unsigned		minor;
27 	char			*mode;
28 };
29 
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 *,
34 			    const char *);
35 static void xen_blkif_free(struct xen_blkif *blkif);
36 static void xen_vbd_free(struct xen_vbd *vbd);
37 
38 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
39 {
40 	return be->dev;
41 }
42 
43 /*
44  * The last request could free the device from softirq context and
45  * xen_blkif_free() can sleep.
46  */
47 static void xen_blkif_deferred_free(struct work_struct *work)
48 {
49 	struct xen_blkif *blkif;
50 
51 	blkif = container_of(work, struct xen_blkif, free_work);
52 	xen_blkif_free(blkif);
53 }
54 
55 static int blkback_name(struct xen_blkif *blkif, char *buf)
56 {
57 	char *devpath, *devname;
58 	struct xenbus_device *dev = blkif->be->dev;
59 
60 	devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
61 	if (IS_ERR(devpath))
62 		return PTR_ERR(devpath);
63 
64 	devname = strstr(devpath, "/dev/");
65 	if (devname != NULL)
66 		devname += strlen("/dev/");
67 	else
68 		devname  = devpath;
69 
70 	snprintf(buf, TASK_COMM_LEN, "%d.%s", blkif->domid, devname);
71 	kfree(devpath);
72 
73 	return 0;
74 }
75 
76 static void xen_update_blkif_status(struct xen_blkif *blkif)
77 {
78 	int err;
79 	char name[TASK_COMM_LEN];
80 	struct xen_blkif_ring *ring;
81 	int i;
82 
83 	/* Not ready to connect? */
84 	if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
85 		return;
86 
87 	/* Already connected? */
88 	if (blkif->be->dev->state == XenbusStateConnected)
89 		return;
90 
91 	/* Attempt to connect: exit if we fail to. */
92 	connect(blkif->be);
93 	if (blkif->be->dev->state != XenbusStateConnected)
94 		return;
95 
96 	err = blkback_name(blkif, name);
97 	if (err) {
98 		xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
99 		return;
100 	}
101 
102 	err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
103 	if (err) {
104 		xenbus_dev_error(blkif->be->dev, err, "block flush");
105 		return;
106 	}
107 	invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
108 
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);
117 			goto out;
118 		}
119 	}
120 	return;
121 
122 out:
123 	while (--i >= 0) {
124 		ring = &blkif->rings[i];
125 		kthread_stop(ring->xenblkd);
126 	}
127 	return;
128 }
129 
130 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
131 {
132 	unsigned int r;
133 
134 	blkif->rings = kcalloc(blkif->nr_rings, sizeof(struct xen_blkif_ring),
135 			       GFP_KERNEL);
136 	if (!blkif->rings)
137 		return -ENOMEM;
138 
139 	for (r = 0; r < blkif->nr_rings; r++) {
140 		struct xen_blkif_ring *ring = &blkif->rings[r];
141 
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 		spin_lock_init(&ring->free_pages_lock);
148 		INIT_LIST_HEAD(&ring->free_pages);
149 
150 		spin_lock_init(&ring->pending_free_lock);
151 		init_waitqueue_head(&ring->pending_free_wq);
152 		init_waitqueue_head(&ring->shutdown_wq);
153 		ring->blkif = blkif;
154 		ring->st_print = jiffies;
155 		ring->active = true;
156 	}
157 
158 	return 0;
159 }
160 
161 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
162 {
163 	struct xen_blkif *blkif;
164 
165 	BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
166 
167 	blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
168 	if (!blkif)
169 		return ERR_PTR(-ENOMEM);
170 
171 	blkif->domid = domid;
172 	atomic_set(&blkif->refcnt, 1);
173 	init_completion(&blkif->drain_complete);
174 	INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
175 
176 	return blkif;
177 }
178 
179 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
180 			 unsigned int nr_grefs, unsigned int evtchn)
181 {
182 	int err;
183 	struct xen_blkif *blkif = ring->blkif;
184 
185 	/* Already connected through? */
186 	if (ring->irq)
187 		return 0;
188 
189 	err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
190 				     &ring->blk_ring);
191 	if (err < 0)
192 		return err;
193 
194 	switch (blkif->blk_protocol) {
195 	case BLKIF_PROTOCOL_NATIVE:
196 	{
197 		struct blkif_sring *sring;
198 		sring = (struct blkif_sring *)ring->blk_ring;
199 		BACK_RING_INIT(&ring->blk_rings.native, sring,
200 			       XEN_PAGE_SIZE * nr_grefs);
201 		break;
202 	}
203 	case BLKIF_PROTOCOL_X86_32:
204 	{
205 		struct blkif_x86_32_sring *sring_x86_32;
206 		sring_x86_32 = (struct blkif_x86_32_sring *)ring->blk_ring;
207 		BACK_RING_INIT(&ring->blk_rings.x86_32, sring_x86_32,
208 			       XEN_PAGE_SIZE * nr_grefs);
209 		break;
210 	}
211 	case BLKIF_PROTOCOL_X86_64:
212 	{
213 		struct blkif_x86_64_sring *sring_x86_64;
214 		sring_x86_64 = (struct blkif_x86_64_sring *)ring->blk_ring;
215 		BACK_RING_INIT(&ring->blk_rings.x86_64, sring_x86_64,
216 			       XEN_PAGE_SIZE * nr_grefs);
217 		break;
218 	}
219 	default:
220 		BUG();
221 	}
222 
223 	err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
224 						    xen_blkif_be_int, 0,
225 						    "blkif-backend", ring);
226 	if (err < 0) {
227 		xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
228 		ring->blk_rings.common.sring = NULL;
229 		return err;
230 	}
231 	ring->irq = err;
232 
233 	return 0;
234 }
235 
236 static int xen_blkif_disconnect(struct xen_blkif *blkif)
237 {
238 	struct pending_req *req, *n;
239 	unsigned int j, r;
240 	bool busy = false;
241 
242 	for (r = 0; r < blkif->nr_rings; r++) {
243 		struct xen_blkif_ring *ring = &blkif->rings[r];
244 		unsigned int i = 0;
245 
246 		if (!ring->active)
247 			continue;
248 
249 		if (ring->xenblkd) {
250 			kthread_stop(ring->xenblkd);
251 			wake_up(&ring->shutdown_wq);
252 		}
253 
254 		/* The above kthread_stop() guarantees that at this point we
255 		 * don't have any discard_io or other_io requests. So, checking
256 		 * for inflight IO is enough.
257 		 */
258 		if (atomic_read(&ring->inflight) > 0) {
259 			busy = true;
260 			continue;
261 		}
262 
263 		if (ring->irq) {
264 			unbind_from_irqhandler(ring->irq, ring);
265 			ring->irq = 0;
266 		}
267 
268 		if (ring->blk_rings.common.sring) {
269 			xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
270 			ring->blk_rings.common.sring = NULL;
271 		}
272 
273 		/* Remove all persistent grants and the cache of ballooned pages. */
274 		xen_blkbk_free_caches(ring);
275 
276 		/* Check that there is no request in use */
277 		list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
278 			list_del(&req->free_list);
279 
280 			for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
281 				kfree(req->segments[j]);
282 
283 			for (j = 0; j < MAX_INDIRECT_PAGES; j++)
284 				kfree(req->indirect_pages[j]);
285 
286 			kfree(req);
287 			i++;
288 		}
289 
290 		BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
291 		BUG_ON(!list_empty(&ring->persistent_purge_list));
292 		BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
293 		BUG_ON(!list_empty(&ring->free_pages));
294 		BUG_ON(ring->free_pages_num != 0);
295 		BUG_ON(ring->persistent_gnt_c != 0);
296 		WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
297 		ring->active = false;
298 	}
299 	if (busy)
300 		return -EBUSY;
301 
302 	blkif->nr_ring_pages = 0;
303 	/*
304 	 * blkif->rings was allocated in connect_ring, so we should free it in
305 	 * here.
306 	 */
307 	kfree(blkif->rings);
308 	blkif->rings = NULL;
309 	blkif->nr_rings = 0;
310 
311 	return 0;
312 }
313 
314 static void xen_blkif_free(struct xen_blkif *blkif)
315 {
316 	WARN_ON(xen_blkif_disconnect(blkif));
317 	xen_vbd_free(&blkif->vbd);
318 	kfree(blkif->be->mode);
319 	kfree(blkif->be);
320 
321 	/* Make sure everything is drained before shutting down */
322 	kmem_cache_free(xen_blkif_cachep, blkif);
323 }
324 
325 int __init xen_blkif_interface_init(void)
326 {
327 	xen_blkif_cachep = kmem_cache_create("blkif_cache",
328 					     sizeof(struct xen_blkif),
329 					     0, 0, NULL);
330 	if (!xen_blkif_cachep)
331 		return -ENOMEM;
332 
333 	return 0;
334 }
335 
336 /*
337  *  sysfs interface for VBD I/O requests
338  */
339 
340 #define VBD_SHOW_ALLRING(name, format)					\
341 	static ssize_t show_##name(struct device *_dev,			\
342 				   struct device_attribute *attr,	\
343 				   char *buf)				\
344 	{								\
345 		struct xenbus_device *dev = to_xenbus_device(_dev);	\
346 		struct backend_info *be = dev_get_drvdata(&dev->dev);	\
347 		struct xen_blkif *blkif = be->blkif;			\
348 		unsigned int i;						\
349 		unsigned long long result = 0;				\
350 									\
351 		if (!blkif->rings)				\
352 			goto out;					\
353 									\
354 		for (i = 0; i < blkif->nr_rings; i++) {		\
355 			struct xen_blkif_ring *ring = &blkif->rings[i];	\
356 									\
357 			result += ring->st_##name;			\
358 		}							\
359 									\
360 out:									\
361 		return sprintf(buf, format, result);			\
362 	}								\
363 	static DEVICE_ATTR(name, 0444, show_##name, NULL)
364 
365 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
366 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
367 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
368 VBD_SHOW_ALLRING(f_req,  "%llu\n");
369 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
370 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
371 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
372 
373 static struct attribute *xen_vbdstat_attrs[] = {
374 	&dev_attr_oo_req.attr,
375 	&dev_attr_rd_req.attr,
376 	&dev_attr_wr_req.attr,
377 	&dev_attr_f_req.attr,
378 	&dev_attr_ds_req.attr,
379 	&dev_attr_rd_sect.attr,
380 	&dev_attr_wr_sect.attr,
381 	NULL
382 };
383 
384 static const struct attribute_group xen_vbdstat_group = {
385 	.name = "statistics",
386 	.attrs = xen_vbdstat_attrs,
387 };
388 
389 #define VBD_SHOW(name, format, args...)					\
390 	static ssize_t show_##name(struct device *_dev,			\
391 				   struct device_attribute *attr,	\
392 				   char *buf)				\
393 	{								\
394 		struct xenbus_device *dev = to_xenbus_device(_dev);	\
395 		struct backend_info *be = dev_get_drvdata(&dev->dev);	\
396 									\
397 		return sprintf(buf, format, ##args);			\
398 	}								\
399 	static DEVICE_ATTR(name, 0444, show_##name, NULL)
400 
401 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
402 VBD_SHOW(mode, "%s\n", be->mode);
403 
404 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
405 {
406 	int error;
407 
408 	error = device_create_file(&dev->dev, &dev_attr_physical_device);
409 	if (error)
410 		goto fail1;
411 
412 	error = device_create_file(&dev->dev, &dev_attr_mode);
413 	if (error)
414 		goto fail2;
415 
416 	error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
417 	if (error)
418 		goto fail3;
419 
420 	return 0;
421 
422 fail3:	sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
423 fail2:	device_remove_file(&dev->dev, &dev_attr_mode);
424 fail1:	device_remove_file(&dev->dev, &dev_attr_physical_device);
425 	return error;
426 }
427 
428 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
429 {
430 	sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
431 	device_remove_file(&dev->dev, &dev_attr_mode);
432 	device_remove_file(&dev->dev, &dev_attr_physical_device);
433 }
434 
435 
436 static void xen_vbd_free(struct xen_vbd *vbd)
437 {
438 	if (vbd->bdev)
439 		blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
440 	vbd->bdev = NULL;
441 }
442 
443 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
444 			  unsigned major, unsigned minor, int readonly,
445 			  int cdrom)
446 {
447 	struct xen_vbd *vbd;
448 	struct block_device *bdev;
449 	struct request_queue *q;
450 
451 	vbd = &blkif->vbd;
452 	vbd->handle   = handle;
453 	vbd->readonly = readonly;
454 	vbd->type     = 0;
455 
456 	vbd->pdevice  = MKDEV(major, minor);
457 
458 	bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
459 				 FMODE_READ : FMODE_WRITE, NULL);
460 
461 	if (IS_ERR(bdev)) {
462 		pr_warn("xen_vbd_create: device %08x could not be opened\n",
463 			vbd->pdevice);
464 		return -ENOENT;
465 	}
466 
467 	vbd->bdev = bdev;
468 	if (vbd->bdev->bd_disk == NULL) {
469 		pr_warn("xen_vbd_create: device %08x doesn't exist\n",
470 			vbd->pdevice);
471 		xen_vbd_free(vbd);
472 		return -ENOENT;
473 	}
474 	vbd->size = vbd_sz(vbd);
475 
476 	if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
477 		vbd->type |= VDISK_CDROM;
478 	if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
479 		vbd->type |= VDISK_REMOVABLE;
480 
481 	q = bdev_get_queue(bdev);
482 	if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
483 		vbd->flush_support = true;
484 
485 	if (q && blk_queue_secure_erase(q))
486 		vbd->discard_secure = true;
487 
488 	pr_debug("Successful creation of handle=%04x (dom=%u)\n",
489 		handle, blkif->domid);
490 	return 0;
491 }
492 static int xen_blkbk_remove(struct xenbus_device *dev)
493 {
494 	struct backend_info *be = dev_get_drvdata(&dev->dev);
495 
496 	pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
497 
498 	if (be->major || be->minor)
499 		xenvbd_sysfs_delif(dev);
500 
501 	if (be->backend_watch.node) {
502 		unregister_xenbus_watch(&be->backend_watch);
503 		kfree(be->backend_watch.node);
504 		be->backend_watch.node = NULL;
505 	}
506 
507 	dev_set_drvdata(&dev->dev, NULL);
508 
509 	if (be->blkif) {
510 		xen_blkif_disconnect(be->blkif);
511 
512 		/* Put the reference we set in xen_blkif_alloc(). */
513 		xen_blkif_put(be->blkif);
514 	}
515 
516 	return 0;
517 }
518 
519 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
520 			      struct backend_info *be, int state)
521 {
522 	struct xenbus_device *dev = be->dev;
523 	int err;
524 
525 	err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
526 			    "%d", state);
527 	if (err)
528 		dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
529 
530 	return err;
531 }
532 
533 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
534 {
535 	struct xenbus_device *dev = be->dev;
536 	struct xen_blkif *blkif = be->blkif;
537 	int err;
538 	int state = 0;
539 	struct block_device *bdev = be->blkif->vbd.bdev;
540 	struct request_queue *q = bdev_get_queue(bdev);
541 
542 	if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
543 		return;
544 
545 	if (blk_queue_discard(q)) {
546 		err = xenbus_printf(xbt, dev->nodename,
547 			"discard-granularity", "%u",
548 			q->limits.discard_granularity);
549 		if (err) {
550 			dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
551 			return;
552 		}
553 		err = xenbus_printf(xbt, dev->nodename,
554 			"discard-alignment", "%u",
555 			q->limits.discard_alignment);
556 		if (err) {
557 			dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
558 			return;
559 		}
560 		state = 1;
561 		/* Optional. */
562 		err = xenbus_printf(xbt, dev->nodename,
563 				    "discard-secure", "%d",
564 				    blkif->vbd.discard_secure);
565 		if (err) {
566 			dev_warn(&dev->dev, "writing discard-secure (%d)", err);
567 			return;
568 		}
569 	}
570 	err = xenbus_printf(xbt, dev->nodename, "feature-discard",
571 			    "%d", state);
572 	if (err)
573 		dev_warn(&dev->dev, "writing feature-discard (%d)", err);
574 }
575 int xen_blkbk_barrier(struct xenbus_transaction xbt,
576 		      struct backend_info *be, int state)
577 {
578 	struct xenbus_device *dev = be->dev;
579 	int err;
580 
581 	err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
582 			    "%d", state);
583 	if (err)
584 		dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
585 
586 	return err;
587 }
588 
589 /*
590  * Entry point to this code when a new device is created.  Allocate the basic
591  * structures, and watch the store waiting for the hotplug scripts to tell us
592  * the device's physical major and minor numbers.  Switch to InitWait.
593  */
594 static int xen_blkbk_probe(struct xenbus_device *dev,
595 			   const struct xenbus_device_id *id)
596 {
597 	int err;
598 	struct backend_info *be = kzalloc(sizeof(struct backend_info),
599 					  GFP_KERNEL);
600 
601 	/* match the pr_debug in xen_blkbk_remove */
602 	pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
603 
604 	if (!be) {
605 		xenbus_dev_fatal(dev, -ENOMEM,
606 				 "allocating backend structure");
607 		return -ENOMEM;
608 	}
609 	be->dev = dev;
610 	dev_set_drvdata(&dev->dev, be);
611 
612 	be->blkif = xen_blkif_alloc(dev->otherend_id);
613 	if (IS_ERR(be->blkif)) {
614 		err = PTR_ERR(be->blkif);
615 		be->blkif = NULL;
616 		xenbus_dev_fatal(dev, err, "creating block interface");
617 		goto fail;
618 	}
619 
620 	err = xenbus_printf(XBT_NIL, dev->nodename,
621 			    "feature-max-indirect-segments", "%u",
622 			    MAX_INDIRECT_SEGMENTS);
623 	if (err)
624 		dev_warn(&dev->dev,
625 			 "writing %s/feature-max-indirect-segments (%d)",
626 			 dev->nodename, err);
627 
628 	/* Multi-queue: advertise how many queues are supported by us.*/
629 	err = xenbus_printf(XBT_NIL, dev->nodename,
630 			    "multi-queue-max-queues", "%u", xenblk_max_queues);
631 	if (err)
632 		pr_warn("Error writing multi-queue-max-queues\n");
633 
634 	/* setup back pointer */
635 	be->blkif->be = be;
636 
637 	err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
638 				   "%s/%s", dev->nodename, "physical-device");
639 	if (err)
640 		goto fail;
641 
642 	err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
643 			    xen_blkif_max_ring_order);
644 	if (err)
645 		pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
646 
647 	err = xenbus_switch_state(dev, XenbusStateInitWait);
648 	if (err)
649 		goto fail;
650 
651 	return 0;
652 
653 fail:
654 	pr_warn("%s failed\n", __func__);
655 	xen_blkbk_remove(dev);
656 	return err;
657 }
658 
659 
660 /*
661  * Callback received when the hotplug scripts have placed the physical-device
662  * node.  Read it and the mode node, and create a vbd.  If the frontend is
663  * ready, connect.
664  */
665 static void backend_changed(struct xenbus_watch *watch,
666 			    const char *path, const char *token)
667 {
668 	int err;
669 	unsigned major;
670 	unsigned minor;
671 	struct backend_info *be
672 		= container_of(watch, struct backend_info, backend_watch);
673 	struct xenbus_device *dev = be->dev;
674 	int cdrom = 0;
675 	unsigned long handle;
676 	char *device_type;
677 
678 	pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
679 
680 	err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
681 			   &major, &minor);
682 	if (XENBUS_EXIST_ERR(err)) {
683 		/*
684 		 * Since this watch will fire once immediately after it is
685 		 * registered, we expect this.  Ignore it, and wait for the
686 		 * hotplug scripts.
687 		 */
688 		return;
689 	}
690 	if (err != 2) {
691 		xenbus_dev_fatal(dev, err, "reading physical-device");
692 		return;
693 	}
694 
695 	if (be->major | be->minor) {
696 		if (be->major != major || be->minor != minor)
697 			pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
698 				be->major, be->minor, major, minor);
699 		return;
700 	}
701 
702 	be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
703 	if (IS_ERR(be->mode)) {
704 		err = PTR_ERR(be->mode);
705 		be->mode = NULL;
706 		xenbus_dev_fatal(dev, err, "reading mode");
707 		return;
708 	}
709 
710 	device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
711 	if (!IS_ERR(device_type)) {
712 		cdrom = strcmp(device_type, "cdrom") == 0;
713 		kfree(device_type);
714 	}
715 
716 	/* Front end dir is a number, which is used as the handle. */
717 	err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
718 	if (err) {
719 		kfree(be->mode);
720 		be->mode = NULL;
721 		return;
722 	}
723 
724 	be->major = major;
725 	be->minor = minor;
726 
727 	err = xen_vbd_create(be->blkif, handle, major, minor,
728 			     !strchr(be->mode, 'w'), cdrom);
729 
730 	if (err)
731 		xenbus_dev_fatal(dev, err, "creating vbd structure");
732 	else {
733 		err = xenvbd_sysfs_addif(dev);
734 		if (err) {
735 			xen_vbd_free(&be->blkif->vbd);
736 			xenbus_dev_fatal(dev, err, "creating sysfs entries");
737 		}
738 	}
739 
740 	if (err) {
741 		kfree(be->mode);
742 		be->mode = NULL;
743 		be->major = 0;
744 		be->minor = 0;
745 	} else {
746 		/* We're potentially connected now */
747 		xen_update_blkif_status(be->blkif);
748 	}
749 }
750 
751 
752 /*
753  * Callback received when the frontend's state changes.
754  */
755 static void frontend_changed(struct xenbus_device *dev,
756 			     enum xenbus_state frontend_state)
757 {
758 	struct backend_info *be = dev_get_drvdata(&dev->dev);
759 	int err;
760 
761 	pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
762 
763 	switch (frontend_state) {
764 	case XenbusStateInitialising:
765 		if (dev->state == XenbusStateClosed) {
766 			pr_info("%s: prepare for reconnect\n", dev->nodename);
767 			xenbus_switch_state(dev, XenbusStateInitWait);
768 		}
769 		break;
770 
771 	case XenbusStateInitialised:
772 	case XenbusStateConnected:
773 		/*
774 		 * Ensure we connect even when two watches fire in
775 		 * close succession and we miss the intermediate value
776 		 * of frontend_state.
777 		 */
778 		if (dev->state == XenbusStateConnected)
779 			break;
780 
781 		/*
782 		 * Enforce precondition before potential leak point.
783 		 * xen_blkif_disconnect() is idempotent.
784 		 */
785 		err = xen_blkif_disconnect(be->blkif);
786 		if (err) {
787 			xenbus_dev_fatal(dev, err, "pending I/O");
788 			break;
789 		}
790 
791 		err = connect_ring(be);
792 		if (err) {
793 			/*
794 			 * Clean up so that memory resources can be used by
795 			 * other devices. connect_ring reported already error.
796 			 */
797 			xen_blkif_disconnect(be->blkif);
798 			break;
799 		}
800 		xen_update_blkif_status(be->blkif);
801 		break;
802 
803 	case XenbusStateClosing:
804 		xenbus_switch_state(dev, XenbusStateClosing);
805 		break;
806 
807 	case XenbusStateClosed:
808 		xen_blkif_disconnect(be->blkif);
809 		xenbus_switch_state(dev, XenbusStateClosed);
810 		if (xenbus_dev_is_online(dev))
811 			break;
812 		/* fall through */
813 		/* if not online */
814 	case XenbusStateUnknown:
815 		/* implies xen_blkif_disconnect() via xen_blkbk_remove() */
816 		device_unregister(&dev->dev);
817 		break;
818 
819 	default:
820 		xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
821 				 frontend_state);
822 		break;
823 	}
824 }
825 
826 
827 /* ** Connection ** */
828 
829 
830 /*
831  * Write the physical details regarding the block device to the store, and
832  * switch to Connected state.
833  */
834 static void connect(struct backend_info *be)
835 {
836 	struct xenbus_transaction xbt;
837 	int err;
838 	struct xenbus_device *dev = be->dev;
839 
840 	pr_debug("%s %s\n", __func__, dev->otherend);
841 
842 	/* Supply the information about the device the frontend needs */
843 again:
844 	err = xenbus_transaction_start(&xbt);
845 	if (err) {
846 		xenbus_dev_fatal(dev, err, "starting transaction");
847 		return;
848 	}
849 
850 	/* If we can't advertise it is OK. */
851 	xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
852 
853 	xen_blkbk_discard(xbt, be);
854 
855 	xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
856 
857 	err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
858 	if (err) {
859 		xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
860 				 dev->nodename);
861 		goto abort;
862 	}
863 
864 	err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
865 			    (unsigned long long)vbd_sz(&be->blkif->vbd));
866 	if (err) {
867 		xenbus_dev_fatal(dev, err, "writing %s/sectors",
868 				 dev->nodename);
869 		goto abort;
870 	}
871 
872 	/* FIXME: use a typename instead */
873 	err = xenbus_printf(xbt, dev->nodename, "info", "%u",
874 			    be->blkif->vbd.type |
875 			    (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
876 	if (err) {
877 		xenbus_dev_fatal(dev, err, "writing %s/info",
878 				 dev->nodename);
879 		goto abort;
880 	}
881 	err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
882 			    (unsigned long)
883 			    bdev_logical_block_size(be->blkif->vbd.bdev));
884 	if (err) {
885 		xenbus_dev_fatal(dev, err, "writing %s/sector-size",
886 				 dev->nodename);
887 		goto abort;
888 	}
889 	err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
890 			    bdev_physical_block_size(be->blkif->vbd.bdev));
891 	if (err)
892 		xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
893 				 dev->nodename);
894 
895 	err = xenbus_transaction_end(xbt, 0);
896 	if (err == -EAGAIN)
897 		goto again;
898 	if (err)
899 		xenbus_dev_fatal(dev, err, "ending transaction");
900 
901 	err = xenbus_switch_state(dev, XenbusStateConnected);
902 	if (err)
903 		xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
904 				 dev->nodename);
905 
906 	return;
907  abort:
908 	xenbus_transaction_end(xbt, 1);
909 }
910 
911 /*
912  * Each ring may have multi pages, depends on "ring-page-order".
913  */
914 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
915 {
916 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
917 	struct pending_req *req, *n;
918 	int err, i, j;
919 	struct xen_blkif *blkif = ring->blkif;
920 	struct xenbus_device *dev = blkif->be->dev;
921 	unsigned int nr_grefs, evtchn;
922 
923 	err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
924 			  &evtchn);
925 	if (err != 1) {
926 		err = -EINVAL;
927 		xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
928 		return err;
929 	}
930 
931 	nr_grefs = blkif->nr_ring_pages;
932 
933 	if (unlikely(!nr_grefs)) {
934 		WARN_ON(true);
935 		return -EINVAL;
936 	}
937 
938 	for (i = 0; i < nr_grefs; i++) {
939 		char ring_ref_name[RINGREF_NAME_LEN];
940 
941 		snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
942 		err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
943 				   "%u", &ring_ref[i]);
944 
945 		if (err != 1) {
946 			if (nr_grefs == 1)
947 				break;
948 
949 			err = -EINVAL;
950 			xenbus_dev_fatal(dev, err, "reading %s/%s",
951 					 dir, ring_ref_name);
952 			return err;
953 		}
954 	}
955 
956 	if (err != 1) {
957 		WARN_ON(nr_grefs != 1);
958 
959 		err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u",
960 				   &ring_ref[0]);
961 		if (err != 1) {
962 			err = -EINVAL;
963 			xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
964 			return err;
965 		}
966 	}
967 
968 	err = -ENOMEM;
969 	for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
970 		req = kzalloc(sizeof(*req), GFP_KERNEL);
971 		if (!req)
972 			goto fail;
973 		list_add_tail(&req->free_list, &ring->pending_free);
974 		for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
975 			req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
976 			if (!req->segments[j])
977 				goto fail;
978 		}
979 		for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
980 			req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
981 							 GFP_KERNEL);
982 			if (!req->indirect_pages[j])
983 				goto fail;
984 		}
985 	}
986 
987 	/* Map the shared frame, irq etc. */
988 	err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
989 	if (err) {
990 		xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
991 		goto fail;
992 	}
993 
994 	return 0;
995 
996 fail:
997 	list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
998 		list_del(&req->free_list);
999 		for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1000 			if (!req->segments[j])
1001 				break;
1002 			kfree(req->segments[j]);
1003 		}
1004 		for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1005 			if (!req->indirect_pages[j])
1006 				break;
1007 			kfree(req->indirect_pages[j]);
1008 		}
1009 		kfree(req);
1010 	}
1011 	return err;
1012 }
1013 
1014 static int connect_ring(struct backend_info *be)
1015 {
1016 	struct xenbus_device *dev = be->dev;
1017 	struct xen_blkif *blkif = be->blkif;
1018 	unsigned int pers_grants;
1019 	char protocol[64] = "";
1020 	int err, i;
1021 	char *xspath;
1022 	size_t xspathsize;
1023 	const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1024 	unsigned int requested_num_queues = 0;
1025 	unsigned int ring_page_order;
1026 
1027 	pr_debug("%s %s\n", __func__, dev->otherend);
1028 
1029 	blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1030 	err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1031 			   "%63s", protocol);
1032 	if (err <= 0)
1033 		strcpy(protocol, "unspecified, assuming default");
1034 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1035 		blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1036 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1037 		blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1038 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1039 		blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1040 	else {
1041 		xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1042 		return -ENOSYS;
1043 	}
1044 	pers_grants = xenbus_read_unsigned(dev->otherend, "feature-persistent",
1045 					   0);
1046 	blkif->vbd.feature_gnt_persistent = pers_grants;
1047 	blkif->vbd.overflow_max_grants = 0;
1048 
1049 	/*
1050 	 * Read the number of hardware queues from frontend.
1051 	 */
1052 	requested_num_queues = xenbus_read_unsigned(dev->otherend,
1053 						    "multi-queue-num-queues",
1054 						    1);
1055 	if (requested_num_queues > xenblk_max_queues
1056 	    || requested_num_queues == 0) {
1057 		/* Buggy or malicious guest. */
1058 		xenbus_dev_fatal(dev, err,
1059 				"guest requested %u queues, exceeding the maximum of %u.",
1060 				requested_num_queues, xenblk_max_queues);
1061 		return -ENOSYS;
1062 	}
1063 	blkif->nr_rings = requested_num_queues;
1064 	if (xen_blkif_alloc_rings(blkif))
1065 		return -ENOMEM;
1066 
1067 	pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1068 		 blkif->nr_rings, blkif->blk_protocol, protocol,
1069 		 pers_grants ? "persistent grants" : "");
1070 
1071 	ring_page_order = xenbus_read_unsigned(dev->otherend,
1072 					       "ring-page-order", 0);
1073 
1074 	if (ring_page_order > xen_blkif_max_ring_order) {
1075 		err = -EINVAL;
1076 		xenbus_dev_fatal(dev, err,
1077 				 "requested ring page order %d exceed max:%d",
1078 				 ring_page_order,
1079 				 xen_blkif_max_ring_order);
1080 		return err;
1081 	}
1082 
1083 	blkif->nr_ring_pages = 1 << ring_page_order;
1084 
1085 	if (blkif->nr_rings == 1)
1086 		return read_per_ring_refs(&blkif->rings[0], dev->otherend);
1087 	else {
1088 		xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1089 		xspath = kmalloc(xspathsize, GFP_KERNEL);
1090 		if (!xspath) {
1091 			xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1092 			return -ENOMEM;
1093 		}
1094 
1095 		for (i = 0; i < blkif->nr_rings; i++) {
1096 			memset(xspath, 0, xspathsize);
1097 			snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1098 			err = read_per_ring_refs(&blkif->rings[i], xspath);
1099 			if (err) {
1100 				kfree(xspath);
1101 				return err;
1102 			}
1103 		}
1104 		kfree(xspath);
1105 	}
1106 	return 0;
1107 }
1108 
1109 static const struct xenbus_device_id xen_blkbk_ids[] = {
1110 	{ "vbd" },
1111 	{ "" }
1112 };
1113 
1114 static struct xenbus_driver xen_blkbk_driver = {
1115 	.ids  = xen_blkbk_ids,
1116 	.probe = xen_blkbk_probe,
1117 	.remove = xen_blkbk_remove,
1118 	.otherend_changed = frontend_changed
1119 };
1120 
1121 int xen_blkif_xenbus_init(void)
1122 {
1123 	return xenbus_register_backend(&xen_blkbk_driver);
1124 }
1125