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 /* On the XenBus the max length of 'ring-ref%u'. */
27 #define RINGREF_NAME_LEN (20)
28 
29 struct backend_info {
30 	struct xenbus_device	*dev;
31 	struct xen_blkif	*blkif;
32 	struct xenbus_watch	backend_watch;
33 	unsigned		major;
34 	unsigned		minor;
35 	char			*mode;
36 };
37 
38 static struct kmem_cache *xen_blkif_cachep;
39 static void connect(struct backend_info *);
40 static int connect_ring(struct backend_info *);
41 static void backend_changed(struct xenbus_watch *, const char **,
42 			    unsigned int);
43 static void xen_blkif_free(struct xen_blkif *blkif);
44 static void xen_vbd_free(struct xen_vbd *vbd);
45 
46 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
47 {
48 	return be->dev;
49 }
50 
51 /*
52  * The last request could free the device from softirq context and
53  * xen_blkif_free() can sleep.
54  */
55 static void xen_blkif_deferred_free(struct work_struct *work)
56 {
57 	struct xen_blkif *blkif;
58 
59 	blkif = container_of(work, struct xen_blkif, free_work);
60 	xen_blkif_free(blkif);
61 }
62 
63 static int blkback_name(struct xen_blkif *blkif, char *buf)
64 {
65 	char *devpath, *devname;
66 	struct xenbus_device *dev = blkif->be->dev;
67 
68 	devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
69 	if (IS_ERR(devpath))
70 		return PTR_ERR(devpath);
71 
72 	devname = strstr(devpath, "/dev/");
73 	if (devname != NULL)
74 		devname += strlen("/dev/");
75 	else
76 		devname  = devpath;
77 
78 	snprintf(buf, TASK_COMM_LEN, "%d.%s", blkif->domid, devname);
79 	kfree(devpath);
80 
81 	return 0;
82 }
83 
84 static void xen_update_blkif_status(struct xen_blkif *blkif)
85 {
86 	int err;
87 	char name[TASK_COMM_LEN];
88 	struct xen_blkif_ring *ring;
89 	int i;
90 
91 	/* Not ready to connect? */
92 	if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
93 		return;
94 
95 	/* Already connected? */
96 	if (blkif->be->dev->state == XenbusStateConnected)
97 		return;
98 
99 	/* Attempt to connect: exit if we fail to. */
100 	connect(blkif->be);
101 	if (blkif->be->dev->state != XenbusStateConnected)
102 		return;
103 
104 	err = blkback_name(blkif, name);
105 	if (err) {
106 		xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
107 		return;
108 	}
109 
110 	err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
111 	if (err) {
112 		xenbus_dev_error(blkif->be->dev, err, "block flush");
113 		return;
114 	}
115 	invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
116 
117 	for (i = 0; i < blkif->nr_rings; i++) {
118 		ring = &blkif->rings[i];
119 		ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
120 		if (IS_ERR(ring->xenblkd)) {
121 			err = PTR_ERR(ring->xenblkd);
122 			ring->xenblkd = NULL;
123 			xenbus_dev_fatal(blkif->be->dev, err,
124 					"start %s-%d xenblkd", name, i);
125 			goto out;
126 		}
127 	}
128 	return;
129 
130 out:
131 	while (--i >= 0) {
132 		ring = &blkif->rings[i];
133 		kthread_stop(ring->xenblkd);
134 	}
135 	return;
136 }
137 
138 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
139 {
140 	unsigned int r;
141 
142 	blkif->rings = kzalloc(blkif->nr_rings * sizeof(struct xen_blkif_ring), GFP_KERNEL);
143 	if (!blkif->rings)
144 		return -ENOMEM;
145 
146 	for (r = 0; r < blkif->nr_rings; r++) {
147 		struct xen_blkif_ring *ring = &blkif->rings[r];
148 
149 		spin_lock_init(&ring->blk_ring_lock);
150 		init_waitqueue_head(&ring->wq);
151 		INIT_LIST_HEAD(&ring->pending_free);
152 		INIT_LIST_HEAD(&ring->persistent_purge_list);
153 		INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
154 		spin_lock_init(&ring->free_pages_lock);
155 		INIT_LIST_HEAD(&ring->free_pages);
156 
157 		spin_lock_init(&ring->pending_free_lock);
158 		init_waitqueue_head(&ring->pending_free_wq);
159 		init_waitqueue_head(&ring->shutdown_wq);
160 		ring->blkif = blkif;
161 		ring->st_print = jiffies;
162 		xen_blkif_get(blkif);
163 	}
164 
165 	return 0;
166 }
167 
168 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
169 {
170 	struct xen_blkif *blkif;
171 
172 	BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
173 
174 	blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
175 	if (!blkif)
176 		return ERR_PTR(-ENOMEM);
177 
178 	blkif->domid = domid;
179 	atomic_set(&blkif->refcnt, 1);
180 	init_completion(&blkif->drain_complete);
181 	INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
182 
183 	return blkif;
184 }
185 
186 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
187 			 unsigned int nr_grefs, unsigned int evtchn)
188 {
189 	int err;
190 	struct xen_blkif *blkif = ring->blkif;
191 
192 	/* Already connected through? */
193 	if (ring->irq)
194 		return 0;
195 
196 	err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
197 				     &ring->blk_ring);
198 	if (err < 0)
199 		return err;
200 
201 	switch (blkif->blk_protocol) {
202 	case BLKIF_PROTOCOL_NATIVE:
203 	{
204 		struct blkif_sring *sring;
205 		sring = (struct blkif_sring *)ring->blk_ring;
206 		BACK_RING_INIT(&ring->blk_rings.native, sring,
207 			       XEN_PAGE_SIZE * nr_grefs);
208 		break;
209 	}
210 	case BLKIF_PROTOCOL_X86_32:
211 	{
212 		struct blkif_x86_32_sring *sring_x86_32;
213 		sring_x86_32 = (struct blkif_x86_32_sring *)ring->blk_ring;
214 		BACK_RING_INIT(&ring->blk_rings.x86_32, sring_x86_32,
215 			       XEN_PAGE_SIZE * nr_grefs);
216 		break;
217 	}
218 	case BLKIF_PROTOCOL_X86_64:
219 	{
220 		struct blkif_x86_64_sring *sring_x86_64;
221 		sring_x86_64 = (struct blkif_x86_64_sring *)ring->blk_ring;
222 		BACK_RING_INIT(&ring->blk_rings.x86_64, sring_x86_64,
223 			       XEN_PAGE_SIZE * nr_grefs);
224 		break;
225 	}
226 	default:
227 		BUG();
228 	}
229 
230 	err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
231 						    xen_blkif_be_int, 0,
232 						    "blkif-backend", ring);
233 	if (err < 0) {
234 		xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
235 		ring->blk_rings.common.sring = NULL;
236 		return err;
237 	}
238 	ring->irq = err;
239 
240 	return 0;
241 }
242 
243 static int xen_blkif_disconnect(struct xen_blkif *blkif)
244 {
245 	struct pending_req *req, *n;
246 	unsigned int j, r;
247 
248 	for (r = 0; r < blkif->nr_rings; r++) {
249 		struct xen_blkif_ring *ring = &blkif->rings[r];
250 		unsigned int i = 0;
251 
252 		if (ring->xenblkd) {
253 			kthread_stop(ring->xenblkd);
254 			wake_up(&ring->shutdown_wq);
255 			ring->xenblkd = NULL;
256 		}
257 
258 		/* The above kthread_stop() guarantees that at this point we
259 		 * don't have any discard_io or other_io requests. So, checking
260 		 * for inflight IO is enough.
261 		 */
262 		if (atomic_read(&ring->inflight) > 0)
263 			return -EBUSY;
264 
265 		if (ring->irq) {
266 			unbind_from_irqhandler(ring->irq, ring);
267 			ring->irq = 0;
268 		}
269 
270 		if (ring->blk_rings.common.sring) {
271 			xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
272 			ring->blk_rings.common.sring = NULL;
273 		}
274 
275 		/* Remove all persistent grants and the cache of ballooned pages. */
276 		xen_blkbk_free_caches(ring);
277 
278 		/* Check that there is no request in use */
279 		list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
280 			list_del(&req->free_list);
281 
282 			for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
283 				kfree(req->segments[j]);
284 
285 			for (j = 0; j < MAX_INDIRECT_PAGES; j++)
286 				kfree(req->indirect_pages[j]);
287 
288 			kfree(req);
289 			i++;
290 		}
291 
292 		BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
293 		BUG_ON(!list_empty(&ring->persistent_purge_list));
294 		BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
295 		BUG_ON(!list_empty(&ring->free_pages));
296 		BUG_ON(ring->free_pages_num != 0);
297 		BUG_ON(ring->persistent_gnt_c != 0);
298 		WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
299 		xen_blkif_put(blkif);
300 	}
301 	blkif->nr_ring_pages = 0;
302 	/*
303 	 * blkif->rings was allocated in connect_ring, so we should free it in
304 	 * here.
305 	 */
306 	kfree(blkif->rings);
307 	blkif->rings = NULL;
308 	blkif->nr_rings = 0;
309 
310 	return 0;
311 }
312 
313 static void xen_blkif_free(struct xen_blkif *blkif)
314 {
315 
316 	xen_blkif_disconnect(blkif);
317 	xen_vbd_free(&blkif->vbd);
318 
319 	/* Make sure everything is drained before shutting down */
320 	kmem_cache_free(xen_blkif_cachep, blkif);
321 }
322 
323 int __init xen_blkif_interface_init(void)
324 {
325 	xen_blkif_cachep = kmem_cache_create("blkif_cache",
326 					     sizeof(struct xen_blkif),
327 					     0, 0, NULL);
328 	if (!xen_blkif_cachep)
329 		return -ENOMEM;
330 
331 	return 0;
332 }
333 
334 /*
335  *  sysfs interface for VBD I/O requests
336  */
337 
338 #define VBD_SHOW_ALLRING(name, format)					\
339 	static ssize_t show_##name(struct device *_dev,			\
340 				   struct device_attribute *attr,	\
341 				   char *buf)				\
342 	{								\
343 		struct xenbus_device *dev = to_xenbus_device(_dev);	\
344 		struct backend_info *be = dev_get_drvdata(&dev->dev);	\
345 		struct xen_blkif *blkif = be->blkif;			\
346 		unsigned int i;						\
347 		unsigned long long result = 0;				\
348 									\
349 		if (!blkif->rings)				\
350 			goto out;					\
351 									\
352 		for (i = 0; i < blkif->nr_rings; i++) {		\
353 			struct xen_blkif_ring *ring = &blkif->rings[i];	\
354 									\
355 			result += ring->st_##name;			\
356 		}							\
357 									\
358 out:									\
359 		return sprintf(buf, format, result);			\
360 	}								\
361 	static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
362 
363 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
364 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
365 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
366 VBD_SHOW_ALLRING(f_req,  "%llu\n");
367 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
368 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
369 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
370 
371 static struct attribute *xen_vbdstat_attrs[] = {
372 	&dev_attr_oo_req.attr,
373 	&dev_attr_rd_req.attr,
374 	&dev_attr_wr_req.attr,
375 	&dev_attr_f_req.attr,
376 	&dev_attr_ds_req.attr,
377 	&dev_attr_rd_sect.attr,
378 	&dev_attr_wr_sect.attr,
379 	NULL
380 };
381 
382 static const struct attribute_group xen_vbdstat_group = {
383 	.name = "statistics",
384 	.attrs = xen_vbdstat_attrs,
385 };
386 
387 #define VBD_SHOW(name, format, args...)					\
388 	static ssize_t show_##name(struct device *_dev,			\
389 				   struct device_attribute *attr,	\
390 				   char *buf)				\
391 	{								\
392 		struct xenbus_device *dev = to_xenbus_device(_dev);	\
393 		struct backend_info *be = dev_get_drvdata(&dev->dev);	\
394 									\
395 		return sprintf(buf, format, ##args);			\
396 	}								\
397 	static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
398 
399 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
400 VBD_SHOW(mode, "%s\n", be->mode);
401 
402 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
403 {
404 	int error;
405 
406 	error = device_create_file(&dev->dev, &dev_attr_physical_device);
407 	if (error)
408 		goto fail1;
409 
410 	error = device_create_file(&dev->dev, &dev_attr_mode);
411 	if (error)
412 		goto fail2;
413 
414 	error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
415 	if (error)
416 		goto fail3;
417 
418 	return 0;
419 
420 fail3:	sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
421 fail2:	device_remove_file(&dev->dev, &dev_attr_mode);
422 fail1:	device_remove_file(&dev->dev, &dev_attr_physical_device);
423 	return error;
424 }
425 
426 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
427 {
428 	sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
429 	device_remove_file(&dev->dev, &dev_attr_mode);
430 	device_remove_file(&dev->dev, &dev_attr_physical_device);
431 }
432 
433 
434 static void xen_vbd_free(struct xen_vbd *vbd)
435 {
436 	if (vbd->bdev)
437 		blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
438 	vbd->bdev = NULL;
439 }
440 
441 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
442 			  unsigned major, unsigned minor, int readonly,
443 			  int cdrom)
444 {
445 	struct xen_vbd *vbd;
446 	struct block_device *bdev;
447 	struct request_queue *q;
448 
449 	vbd = &blkif->vbd;
450 	vbd->handle   = handle;
451 	vbd->readonly = readonly;
452 	vbd->type     = 0;
453 
454 	vbd->pdevice  = MKDEV(major, minor);
455 
456 	bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
457 				 FMODE_READ : FMODE_WRITE, NULL);
458 
459 	if (IS_ERR(bdev)) {
460 		pr_warn("xen_vbd_create: device %08x could not be opened\n",
461 			vbd->pdevice);
462 		return -ENOENT;
463 	}
464 
465 	vbd->bdev = bdev;
466 	if (vbd->bdev->bd_disk == NULL) {
467 		pr_warn("xen_vbd_create: device %08x doesn't exist\n",
468 			vbd->pdevice);
469 		xen_vbd_free(vbd);
470 		return -ENOENT;
471 	}
472 	vbd->size = vbd_sz(vbd);
473 
474 	if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
475 		vbd->type |= VDISK_CDROM;
476 	if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
477 		vbd->type |= VDISK_REMOVABLE;
478 
479 	q = bdev_get_queue(bdev);
480 	if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
481 		vbd->flush_support = true;
482 
483 	if (q && blk_queue_secure_erase(q))
484 		vbd->discard_secure = true;
485 
486 	pr_debug("Successful creation of handle=%04x (dom=%u)\n",
487 		handle, blkif->domid);
488 	return 0;
489 }
490 static int xen_blkbk_remove(struct xenbus_device *dev)
491 {
492 	struct backend_info *be = dev_get_drvdata(&dev->dev);
493 
494 	pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
495 
496 	if (be->major || be->minor)
497 		xenvbd_sysfs_delif(dev);
498 
499 	if (be->backend_watch.node) {
500 		unregister_xenbus_watch(&be->backend_watch);
501 		kfree(be->backend_watch.node);
502 		be->backend_watch.node = NULL;
503 	}
504 
505 	dev_set_drvdata(&dev->dev, NULL);
506 
507 	if (be->blkif)
508 		xen_blkif_disconnect(be->blkif);
509 
510 	/* Put the reference we set in xen_blkif_alloc(). */
511 	xen_blkif_put(be->blkif);
512 	kfree(be->mode);
513 	kfree(be);
514 	return 0;
515 }
516 
517 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
518 			      struct backend_info *be, int state)
519 {
520 	struct xenbus_device *dev = be->dev;
521 	int err;
522 
523 	err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
524 			    "%d", state);
525 	if (err)
526 		dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
527 
528 	return err;
529 }
530 
531 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
532 {
533 	struct xenbus_device *dev = be->dev;
534 	struct xen_blkif *blkif = be->blkif;
535 	int err;
536 	int state = 0, discard_enable;
537 	struct block_device *bdev = be->blkif->vbd.bdev;
538 	struct request_queue *q = bdev_get_queue(bdev);
539 
540 	err = xenbus_scanf(XBT_NIL, dev->nodename, "discard-enable", "%d",
541 			   &discard_enable);
542 	if (err == 1 && !discard_enable)
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 **vec, unsigned int len)
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 if not online */
813 	case XenbusStateUnknown:
814 		/* implies xen_blkif_disconnect() via xen_blkbk_remove() */
815 		device_unregister(&dev->dev);
816 		break;
817 
818 	default:
819 		xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
820 				 frontend_state);
821 		break;
822 	}
823 }
824 
825 
826 /* ** Connection ** */
827 
828 
829 /*
830  * Write the physical details regarding the block device to the store, and
831  * switch to Connected state.
832  */
833 static void connect(struct backend_info *be)
834 {
835 	struct xenbus_transaction xbt;
836 	int err;
837 	struct xenbus_device *dev = be->dev;
838 
839 	pr_debug("%s %s\n", __func__, dev->otherend);
840 
841 	/* Supply the information about the device the frontend needs */
842 again:
843 	err = xenbus_transaction_start(&xbt);
844 	if (err) {
845 		xenbus_dev_fatal(dev, err, "starting transaction");
846 		return;
847 	}
848 
849 	/* If we can't advertise it is OK. */
850 	xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
851 
852 	xen_blkbk_discard(xbt, be);
853 
854 	xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
855 
856 	err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
857 	if (err) {
858 		xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
859 				 dev->nodename);
860 		goto abort;
861 	}
862 
863 	err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
864 			    (unsigned long long)vbd_sz(&be->blkif->vbd));
865 	if (err) {
866 		xenbus_dev_fatal(dev, err, "writing %s/sectors",
867 				 dev->nodename);
868 		goto abort;
869 	}
870 
871 	/* FIXME: use a typename instead */
872 	err = xenbus_printf(xbt, dev->nodename, "info", "%u",
873 			    be->blkif->vbd.type |
874 			    (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
875 	if (err) {
876 		xenbus_dev_fatal(dev, err, "writing %s/info",
877 				 dev->nodename);
878 		goto abort;
879 	}
880 	err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
881 			    (unsigned long)
882 			    bdev_logical_block_size(be->blkif->vbd.bdev));
883 	if (err) {
884 		xenbus_dev_fatal(dev, err, "writing %s/sector-size",
885 				 dev->nodename);
886 		goto abort;
887 	}
888 	err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
889 			    bdev_physical_block_size(be->blkif->vbd.bdev));
890 	if (err)
891 		xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
892 				 dev->nodename);
893 
894 	err = xenbus_transaction_end(xbt, 0);
895 	if (err == -EAGAIN)
896 		goto again;
897 	if (err)
898 		xenbus_dev_fatal(dev, err, "ending transaction");
899 
900 	err = xenbus_switch_state(dev, XenbusStateConnected);
901 	if (err)
902 		xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
903 				 dev->nodename);
904 
905 	return;
906  abort:
907 	xenbus_transaction_end(xbt, 1);
908 }
909 
910 /*
911  * Each ring may have multi pages, depends on "ring-page-order".
912  */
913 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
914 {
915 	unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
916 	struct pending_req *req, *n;
917 	int err, i, j;
918 	struct xen_blkif *blkif = ring->blkif;
919 	struct xenbus_device *dev = blkif->be->dev;
920 	unsigned int ring_page_order, nr_grefs, evtchn;
921 
922 	err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
923 			  &evtchn);
924 	if (err != 1) {
925 		err = -EINVAL;
926 		xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
927 		return err;
928 	}
929 
930 	err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
931 			  &ring_page_order);
932 	if (err != 1) {
933 		err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u", &ring_ref[0]);
934 		if (err != 1) {
935 			err = -EINVAL;
936 			xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
937 			return err;
938 		}
939 		nr_grefs = 1;
940 	} else {
941 		unsigned int i;
942 
943 		if (ring_page_order > xen_blkif_max_ring_order) {
944 			err = -EINVAL;
945 			xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
946 					 dir, ring_page_order,
947 					 xen_blkif_max_ring_order);
948 			return err;
949 		}
950 
951 		nr_grefs = 1 << ring_page_order;
952 		for (i = 0; i < nr_grefs; i++) {
953 			char ring_ref_name[RINGREF_NAME_LEN];
954 
955 			snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
956 			err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
957 					   "%u", &ring_ref[i]);
958 			if (err != 1) {
959 				err = -EINVAL;
960 				xenbus_dev_fatal(dev, err, "reading %s/%s",
961 						 dir, ring_ref_name);
962 				return err;
963 			}
964 		}
965 	}
966 	blkif->nr_ring_pages = nr_grefs;
967 
968 	for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
969 		req = kzalloc(sizeof(*req), GFP_KERNEL);
970 		if (!req)
971 			goto fail;
972 		list_add_tail(&req->free_list, &ring->pending_free);
973 		for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
974 			req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
975 			if (!req->segments[j])
976 				goto fail;
977 		}
978 		for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
979 			req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
980 							 GFP_KERNEL);
981 			if (!req->indirect_pages[j])
982 				goto fail;
983 		}
984 	}
985 
986 	/* Map the shared frame, irq etc. */
987 	err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
988 	if (err) {
989 		xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
990 		return err;
991 	}
992 
993 	return 0;
994 
995 fail:
996 	list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
997 		list_del(&req->free_list);
998 		for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
999 			if (!req->segments[j])
1000 				break;
1001 			kfree(req->segments[j]);
1002 		}
1003 		for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1004 			if (!req->indirect_pages[j])
1005 				break;
1006 			kfree(req->indirect_pages[j]);
1007 		}
1008 		kfree(req);
1009 	}
1010 	return -ENOMEM;
1011 
1012 }
1013 
1014 static int connect_ring(struct backend_info *be)
1015 {
1016 	struct xenbus_device *dev = be->dev;
1017 	unsigned int pers_grants;
1018 	char protocol[64] = "";
1019 	int err, i;
1020 	char *xspath;
1021 	size_t xspathsize;
1022 	const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1023 	unsigned int requested_num_queues = 0;
1024 
1025 	pr_debug("%s %s\n", __func__, dev->otherend);
1026 
1027 	be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1028 	err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1029 			   "%63s", protocol);
1030 	if (err <= 0)
1031 		strcpy(protocol, "unspecified, assuming default");
1032 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1033 		be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1034 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1035 		be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1036 	else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1037 		be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1038 	else {
1039 		xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1040 		return -ENOSYS;
1041 	}
1042 	err = xenbus_scanf(XBT_NIL, dev->otherend,
1043 			   "feature-persistent", "%u", &pers_grants);
1044 	if (err <= 0)
1045 		pers_grants = 0;
1046 
1047 	be->blkif->vbd.feature_gnt_persistent = pers_grants;
1048 	be->blkif->vbd.overflow_max_grants = 0;
1049 
1050 	/*
1051 	 * Read the number of hardware queues from frontend.
1052 	 */
1053 	err = xenbus_scanf(XBT_NIL, dev->otherend, "multi-queue-num-queues",
1054 			   "%u", &requested_num_queues);
1055 	if (err < 0) {
1056 		requested_num_queues = 1;
1057 	} else {
1058 		if (requested_num_queues > xenblk_max_queues
1059 		    || requested_num_queues == 0) {
1060 			/* Buggy or malicious guest. */
1061 			xenbus_dev_fatal(dev, err,
1062 					"guest requested %u queues, exceeding the maximum of %u.",
1063 					requested_num_queues, xenblk_max_queues);
1064 			return -ENOSYS;
1065 		}
1066 	}
1067 	be->blkif->nr_rings = requested_num_queues;
1068 	if (xen_blkif_alloc_rings(be->blkif))
1069 		return -ENOMEM;
1070 
1071 	pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1072 		 be->blkif->nr_rings, be->blkif->blk_protocol, protocol,
1073 		 pers_grants ? "persistent grants" : "");
1074 
1075 	if (be->blkif->nr_rings == 1)
1076 		return read_per_ring_refs(&be->blkif->rings[0], dev->otherend);
1077 	else {
1078 		xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1079 		xspath = kmalloc(xspathsize, GFP_KERNEL);
1080 		if (!xspath) {
1081 			xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1082 			return -ENOMEM;
1083 		}
1084 
1085 		for (i = 0; i < be->blkif->nr_rings; i++) {
1086 			memset(xspath, 0, xspathsize);
1087 			snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1088 			err = read_per_ring_refs(&be->blkif->rings[i], xspath);
1089 			if (err) {
1090 				kfree(xspath);
1091 				return err;
1092 			}
1093 		}
1094 		kfree(xspath);
1095 	}
1096 	return 0;
1097 }
1098 
1099 static const struct xenbus_device_id xen_blkbk_ids[] = {
1100 	{ "vbd" },
1101 	{ "" }
1102 };
1103 
1104 static struct xenbus_driver xen_blkbk_driver = {
1105 	.ids  = xen_blkbk_ids,
1106 	.probe = xen_blkbk_probe,
1107 	.remove = xen_blkbk_remove,
1108 	.otherend_changed = frontend_changed
1109 };
1110 
1111 int xen_blkif_xenbus_init(void)
1112 {
1113 	return xenbus_register_backend(&xen_blkbk_driver);
1114 }
1115