OSDN Git Service

Merge tag 'trace-v5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux...
[uclinux-h8/linux.git] / drivers / block / xen-blkback / xenbus.c
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                             const char *);
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 = kcalloc(blkif->nr_rings, sizeof(struct xen_blkif_ring),
143                                GFP_KERNEL);
144         if (!blkif->rings)
145                 return -ENOMEM;
146
147         for (r = 0; r < blkif->nr_rings; r++) {
148                 struct xen_blkif_ring *ring = &blkif->rings[r];
149
150                 spin_lock_init(&ring->blk_ring_lock);
151                 init_waitqueue_head(&ring->wq);
152                 INIT_LIST_HEAD(&ring->pending_free);
153                 INIT_LIST_HEAD(&ring->persistent_purge_list);
154                 INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
155                 spin_lock_init(&ring->free_pages_lock);
156                 INIT_LIST_HEAD(&ring->free_pages);
157
158                 spin_lock_init(&ring->pending_free_lock);
159                 init_waitqueue_head(&ring->pending_free_wq);
160                 init_waitqueue_head(&ring->shutdown_wq);
161                 ring->blkif = blkif;
162                 ring->st_print = jiffies;
163                 ring->active = true;
164         }
165
166         return 0;
167 }
168
169 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
170 {
171         struct xen_blkif *blkif;
172
173         BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
174
175         blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
176         if (!blkif)
177                 return ERR_PTR(-ENOMEM);
178
179         blkif->domid = domid;
180         atomic_set(&blkif->refcnt, 1);
181         init_completion(&blkif->drain_complete);
182         INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
183
184         return blkif;
185 }
186
187 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
188                          unsigned int nr_grefs, unsigned int evtchn)
189 {
190         int err;
191         struct xen_blkif *blkif = ring->blkif;
192
193         /* Already connected through? */
194         if (ring->irq)
195                 return 0;
196
197         err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
198                                      &ring->blk_ring);
199         if (err < 0)
200                 return err;
201
202         switch (blkif->blk_protocol) {
203         case BLKIF_PROTOCOL_NATIVE:
204         {
205                 struct blkif_sring *sring;
206                 sring = (struct blkif_sring *)ring->blk_ring;
207                 BACK_RING_INIT(&ring->blk_rings.native, sring,
208                                XEN_PAGE_SIZE * nr_grefs);
209                 break;
210         }
211         case BLKIF_PROTOCOL_X86_32:
212         {
213                 struct blkif_x86_32_sring *sring_x86_32;
214                 sring_x86_32 = (struct blkif_x86_32_sring *)ring->blk_ring;
215                 BACK_RING_INIT(&ring->blk_rings.x86_32, sring_x86_32,
216                                XEN_PAGE_SIZE * nr_grefs);
217                 break;
218         }
219         case BLKIF_PROTOCOL_X86_64:
220         {
221                 struct blkif_x86_64_sring *sring_x86_64;
222                 sring_x86_64 = (struct blkif_x86_64_sring *)ring->blk_ring;
223                 BACK_RING_INIT(&ring->blk_rings.x86_64, sring_x86_64,
224                                XEN_PAGE_SIZE * nr_grefs);
225                 break;
226         }
227         default:
228                 BUG();
229         }
230
231         err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
232                                                     xen_blkif_be_int, 0,
233                                                     "blkif-backend", ring);
234         if (err < 0) {
235                 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
236                 ring->blk_rings.common.sring = NULL;
237                 return err;
238         }
239         ring->irq = err;
240
241         return 0;
242 }
243
244 static int xen_blkif_disconnect(struct xen_blkif *blkif)
245 {
246         struct pending_req *req, *n;
247         unsigned int j, r;
248         bool busy = false;
249
250         for (r = 0; r < blkif->nr_rings; r++) {
251                 struct xen_blkif_ring *ring = &blkif->rings[r];
252                 unsigned int i = 0;
253
254                 if (!ring->active)
255                         continue;
256
257                 if (ring->xenblkd) {
258                         kthread_stop(ring->xenblkd);
259                         wake_up(&ring->shutdown_wq);
260                 }
261
262                 /* The above kthread_stop() guarantees that at this point we
263                  * don't have any discard_io or other_io requests. So, checking
264                  * for inflight IO is enough.
265                  */
266                 if (atomic_read(&ring->inflight) > 0) {
267                         busy = true;
268                         continue;
269                 }
270
271                 if (ring->irq) {
272                         unbind_from_irqhandler(ring->irq, ring);
273                         ring->irq = 0;
274                 }
275
276                 if (ring->blk_rings.common.sring) {
277                         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
278                         ring->blk_rings.common.sring = NULL;
279                 }
280
281                 /* Remove all persistent grants and the cache of ballooned pages. */
282                 xen_blkbk_free_caches(ring);
283
284                 /* Check that there is no request in use */
285                 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
286                         list_del(&req->free_list);
287
288                         for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
289                                 kfree(req->segments[j]);
290
291                         for (j = 0; j < MAX_INDIRECT_PAGES; j++)
292                                 kfree(req->indirect_pages[j]);
293
294                         kfree(req);
295                         i++;
296                 }
297
298                 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
299                 BUG_ON(!list_empty(&ring->persistent_purge_list));
300                 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
301                 BUG_ON(!list_empty(&ring->free_pages));
302                 BUG_ON(ring->free_pages_num != 0);
303                 BUG_ON(ring->persistent_gnt_c != 0);
304                 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
305                 ring->active = false;
306         }
307         if (busy)
308                 return -EBUSY;
309
310         blkif->nr_ring_pages = 0;
311         /*
312          * blkif->rings was allocated in connect_ring, so we should free it in
313          * here.
314          */
315         kfree(blkif->rings);
316         blkif->rings = NULL;
317         blkif->nr_rings = 0;
318
319         return 0;
320 }
321
322 static void xen_blkif_free(struct xen_blkif *blkif)
323 {
324         WARN_ON(xen_blkif_disconnect(blkif));
325         xen_vbd_free(&blkif->vbd);
326         kfree(blkif->be->mode);
327         kfree(blkif->be);
328
329         /* Make sure everything is drained before shutting down */
330         kmem_cache_free(xen_blkif_cachep, blkif);
331 }
332
333 int __init xen_blkif_interface_init(void)
334 {
335         xen_blkif_cachep = kmem_cache_create("blkif_cache",
336                                              sizeof(struct xen_blkif),
337                                              0, 0, NULL);
338         if (!xen_blkif_cachep)
339                 return -ENOMEM;
340
341         return 0;
342 }
343
344 /*
345  *  sysfs interface for VBD I/O requests
346  */
347
348 #define VBD_SHOW_ALLRING(name, format)                                  \
349         static ssize_t show_##name(struct device *_dev,                 \
350                                    struct device_attribute *attr,       \
351                                    char *buf)                           \
352         {                                                               \
353                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
354                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
355                 struct xen_blkif *blkif = be->blkif;                    \
356                 unsigned int i;                                         \
357                 unsigned long long result = 0;                          \
358                                                                         \
359                 if (!blkif->rings)                              \
360                         goto out;                                       \
361                                                                         \
362                 for (i = 0; i < blkif->nr_rings; i++) {         \
363                         struct xen_blkif_ring *ring = &blkif->rings[i]; \
364                                                                         \
365                         result += ring->st_##name;                      \
366                 }                                                       \
367                                                                         \
368 out:                                                                    \
369                 return sprintf(buf, format, result);                    \
370         }                                                               \
371         static DEVICE_ATTR(name, 0444, show_##name, NULL)
372
373 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
374 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
375 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
376 VBD_SHOW_ALLRING(f_req,  "%llu\n");
377 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
378 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
379 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
380
381 static struct attribute *xen_vbdstat_attrs[] = {
382         &dev_attr_oo_req.attr,
383         &dev_attr_rd_req.attr,
384         &dev_attr_wr_req.attr,
385         &dev_attr_f_req.attr,
386         &dev_attr_ds_req.attr,
387         &dev_attr_rd_sect.attr,
388         &dev_attr_wr_sect.attr,
389         NULL
390 };
391
392 static const struct attribute_group xen_vbdstat_group = {
393         .name = "statistics",
394         .attrs = xen_vbdstat_attrs,
395 };
396
397 #define VBD_SHOW(name, format, args...)                                 \
398         static ssize_t show_##name(struct device *_dev,                 \
399                                    struct device_attribute *attr,       \
400                                    char *buf)                           \
401         {                                                               \
402                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
403                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
404                                                                         \
405                 return sprintf(buf, format, ##args);                    \
406         }                                                               \
407         static DEVICE_ATTR(name, 0444, show_##name, NULL)
408
409 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
410 VBD_SHOW(mode, "%s\n", be->mode);
411
412 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
413 {
414         int error;
415
416         error = device_create_file(&dev->dev, &dev_attr_physical_device);
417         if (error)
418                 goto fail1;
419
420         error = device_create_file(&dev->dev, &dev_attr_mode);
421         if (error)
422                 goto fail2;
423
424         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
425         if (error)
426                 goto fail3;
427
428         return 0;
429
430 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
431 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
432 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
433         return error;
434 }
435
436 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
437 {
438         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
439         device_remove_file(&dev->dev, &dev_attr_mode);
440         device_remove_file(&dev->dev, &dev_attr_physical_device);
441 }
442
443
444 static void xen_vbd_free(struct xen_vbd *vbd)
445 {
446         if (vbd->bdev)
447                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
448         vbd->bdev = NULL;
449 }
450
451 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
452                           unsigned major, unsigned minor, int readonly,
453                           int cdrom)
454 {
455         struct xen_vbd *vbd;
456         struct block_device *bdev;
457         struct request_queue *q;
458
459         vbd = &blkif->vbd;
460         vbd->handle   = handle;
461         vbd->readonly = readonly;
462         vbd->type     = 0;
463
464         vbd->pdevice  = MKDEV(major, minor);
465
466         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
467                                  FMODE_READ : FMODE_WRITE, NULL);
468
469         if (IS_ERR(bdev)) {
470                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
471                         vbd->pdevice);
472                 return -ENOENT;
473         }
474
475         vbd->bdev = bdev;
476         if (vbd->bdev->bd_disk == NULL) {
477                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
478                         vbd->pdevice);
479                 xen_vbd_free(vbd);
480                 return -ENOENT;
481         }
482         vbd->size = vbd_sz(vbd);
483
484         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
485                 vbd->type |= VDISK_CDROM;
486         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
487                 vbd->type |= VDISK_REMOVABLE;
488
489         q = bdev_get_queue(bdev);
490         if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
491                 vbd->flush_support = true;
492
493         if (q && blk_queue_secure_erase(q))
494                 vbd->discard_secure = true;
495
496         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
497                 handle, blkif->domid);
498         return 0;
499 }
500 static int xen_blkbk_remove(struct xenbus_device *dev)
501 {
502         struct backend_info *be = dev_get_drvdata(&dev->dev);
503
504         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
505
506         if (be->major || be->minor)
507                 xenvbd_sysfs_delif(dev);
508
509         if (be->backend_watch.node) {
510                 unregister_xenbus_watch(&be->backend_watch);
511                 kfree(be->backend_watch.node);
512                 be->backend_watch.node = NULL;
513         }
514
515         dev_set_drvdata(&dev->dev, NULL);
516
517         if (be->blkif) {
518                 xen_blkif_disconnect(be->blkif);
519
520                 /* Put the reference we set in xen_blkif_alloc(). */
521                 xen_blkif_put(be->blkif);
522         }
523
524         return 0;
525 }
526
527 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
528                               struct backend_info *be, int state)
529 {
530         struct xenbus_device *dev = be->dev;
531         int err;
532
533         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
534                             "%d", state);
535         if (err)
536                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
537
538         return err;
539 }
540
541 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
542 {
543         struct xenbus_device *dev = be->dev;
544         struct xen_blkif *blkif = be->blkif;
545         int err;
546         int state = 0;
547         struct block_device *bdev = be->blkif->vbd.bdev;
548         struct request_queue *q = bdev_get_queue(bdev);
549
550         if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
551                 return;
552
553         if (blk_queue_discard(q)) {
554                 err = xenbus_printf(xbt, dev->nodename,
555                         "discard-granularity", "%u",
556                         q->limits.discard_granularity);
557                 if (err) {
558                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
559                         return;
560                 }
561                 err = xenbus_printf(xbt, dev->nodename,
562                         "discard-alignment", "%u",
563                         q->limits.discard_alignment);
564                 if (err) {
565                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
566                         return;
567                 }
568                 state = 1;
569                 /* Optional. */
570                 err = xenbus_printf(xbt, dev->nodename,
571                                     "discard-secure", "%d",
572                                     blkif->vbd.discard_secure);
573                 if (err) {
574                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
575                         return;
576                 }
577         }
578         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
579                             "%d", state);
580         if (err)
581                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
582 }
583 int xen_blkbk_barrier(struct xenbus_transaction xbt,
584                       struct backend_info *be, int state)
585 {
586         struct xenbus_device *dev = be->dev;
587         int err;
588
589         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
590                             "%d", state);
591         if (err)
592                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
593
594         return err;
595 }
596
597 /*
598  * Entry point to this code when a new device is created.  Allocate the basic
599  * structures, and watch the store waiting for the hotplug scripts to tell us
600  * the device's physical major and minor numbers.  Switch to InitWait.
601  */
602 static int xen_blkbk_probe(struct xenbus_device *dev,
603                            const struct xenbus_device_id *id)
604 {
605         int err;
606         struct backend_info *be = kzalloc(sizeof(struct backend_info),
607                                           GFP_KERNEL);
608
609         /* match the pr_debug in xen_blkbk_remove */
610         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
611
612         if (!be) {
613                 xenbus_dev_fatal(dev, -ENOMEM,
614                                  "allocating backend structure");
615                 return -ENOMEM;
616         }
617         be->dev = dev;
618         dev_set_drvdata(&dev->dev, be);
619
620         be->blkif = xen_blkif_alloc(dev->otherend_id);
621         if (IS_ERR(be->blkif)) {
622                 err = PTR_ERR(be->blkif);
623                 be->blkif = NULL;
624                 xenbus_dev_fatal(dev, err, "creating block interface");
625                 goto fail;
626         }
627
628         err = xenbus_printf(XBT_NIL, dev->nodename,
629                             "feature-max-indirect-segments", "%u",
630                             MAX_INDIRECT_SEGMENTS);
631         if (err)
632                 dev_warn(&dev->dev,
633                          "writing %s/feature-max-indirect-segments (%d)",
634                          dev->nodename, err);
635
636         /* Multi-queue: advertise how many queues are supported by us.*/
637         err = xenbus_printf(XBT_NIL, dev->nodename,
638                             "multi-queue-max-queues", "%u", xenblk_max_queues);
639         if (err)
640                 pr_warn("Error writing multi-queue-max-queues\n");
641
642         /* setup back pointer */
643         be->blkif->be = be;
644
645         err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
646                                    "%s/%s", dev->nodename, "physical-device");
647         if (err)
648                 goto fail;
649
650         err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
651                             xen_blkif_max_ring_order);
652         if (err)
653                 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
654
655         err = xenbus_switch_state(dev, XenbusStateInitWait);
656         if (err)
657                 goto fail;
658
659         return 0;
660
661 fail:
662         pr_warn("%s failed\n", __func__);
663         xen_blkbk_remove(dev);
664         return err;
665 }
666
667
668 /*
669  * Callback received when the hotplug scripts have placed the physical-device
670  * node.  Read it and the mode node, and create a vbd.  If the frontend is
671  * ready, connect.
672  */
673 static void backend_changed(struct xenbus_watch *watch,
674                             const char *path, const char *token)
675 {
676         int err;
677         unsigned major;
678         unsigned minor;
679         struct backend_info *be
680                 = container_of(watch, struct backend_info, backend_watch);
681         struct xenbus_device *dev = be->dev;
682         int cdrom = 0;
683         unsigned long handle;
684         char *device_type;
685
686         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
687
688         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
689                            &major, &minor);
690         if (XENBUS_EXIST_ERR(err)) {
691                 /*
692                  * Since this watch will fire once immediately after it is
693                  * registered, we expect this.  Ignore it, and wait for the
694                  * hotplug scripts.
695                  */
696                 return;
697         }
698         if (err != 2) {
699                 xenbus_dev_fatal(dev, err, "reading physical-device");
700                 return;
701         }
702
703         if (be->major | be->minor) {
704                 if (be->major != major || be->minor != minor)
705                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
706                                 be->major, be->minor, major, minor);
707                 return;
708         }
709
710         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
711         if (IS_ERR(be->mode)) {
712                 err = PTR_ERR(be->mode);
713                 be->mode = NULL;
714                 xenbus_dev_fatal(dev, err, "reading mode");
715                 return;
716         }
717
718         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
719         if (!IS_ERR(device_type)) {
720                 cdrom = strcmp(device_type, "cdrom") == 0;
721                 kfree(device_type);
722         }
723
724         /* Front end dir is a number, which is used as the handle. */
725         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
726         if (err) {
727                 kfree(be->mode);
728                 be->mode = NULL;
729                 return;
730         }
731
732         be->major = major;
733         be->minor = minor;
734
735         err = xen_vbd_create(be->blkif, handle, major, minor,
736                              !strchr(be->mode, 'w'), cdrom);
737
738         if (err)
739                 xenbus_dev_fatal(dev, err, "creating vbd structure");
740         else {
741                 err = xenvbd_sysfs_addif(dev);
742                 if (err) {
743                         xen_vbd_free(&be->blkif->vbd);
744                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
745                 }
746         }
747
748         if (err) {
749                 kfree(be->mode);
750                 be->mode = NULL;
751                 be->major = 0;
752                 be->minor = 0;
753         } else {
754                 /* We're potentially connected now */
755                 xen_update_blkif_status(be->blkif);
756         }
757 }
758
759
760 /*
761  * Callback received when the frontend's state changes.
762  */
763 static void frontend_changed(struct xenbus_device *dev,
764                              enum xenbus_state frontend_state)
765 {
766         struct backend_info *be = dev_get_drvdata(&dev->dev);
767         int err;
768
769         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
770
771         switch (frontend_state) {
772         case XenbusStateInitialising:
773                 if (dev->state == XenbusStateClosed) {
774                         pr_info("%s: prepare for reconnect\n", dev->nodename);
775                         xenbus_switch_state(dev, XenbusStateInitWait);
776                 }
777                 break;
778
779         case XenbusStateInitialised:
780         case XenbusStateConnected:
781                 /*
782                  * Ensure we connect even when two watches fire in
783                  * close succession and we miss the intermediate value
784                  * of frontend_state.
785                  */
786                 if (dev->state == XenbusStateConnected)
787                         break;
788
789                 /*
790                  * Enforce precondition before potential leak point.
791                  * xen_blkif_disconnect() is idempotent.
792                  */
793                 err = xen_blkif_disconnect(be->blkif);
794                 if (err) {
795                         xenbus_dev_fatal(dev, err, "pending I/O");
796                         break;
797                 }
798
799                 err = connect_ring(be);
800                 if (err) {
801                         /*
802                          * Clean up so that memory resources can be used by
803                          * other devices. connect_ring reported already error.
804                          */
805                         xen_blkif_disconnect(be->blkif);
806                         break;
807                 }
808                 xen_update_blkif_status(be->blkif);
809                 break;
810
811         case XenbusStateClosing:
812                 xenbus_switch_state(dev, XenbusStateClosing);
813                 break;
814
815         case XenbusStateClosed:
816                 xen_blkif_disconnect(be->blkif);
817                 xenbus_switch_state(dev, XenbusStateClosed);
818                 if (xenbus_dev_is_online(dev))
819                         break;
820                 /* fall through */
821                 /* if not online */
822         case XenbusStateUnknown:
823                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
824                 device_unregister(&dev->dev);
825                 break;
826
827         default:
828                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
829                                  frontend_state);
830                 break;
831         }
832 }
833
834
835 /* ** Connection ** */
836
837
838 /*
839  * Write the physical details regarding the block device to the store, and
840  * switch to Connected state.
841  */
842 static void connect(struct backend_info *be)
843 {
844         struct xenbus_transaction xbt;
845         int err;
846         struct xenbus_device *dev = be->dev;
847
848         pr_debug("%s %s\n", __func__, dev->otherend);
849
850         /* Supply the information about the device the frontend needs */
851 again:
852         err = xenbus_transaction_start(&xbt);
853         if (err) {
854                 xenbus_dev_fatal(dev, err, "starting transaction");
855                 return;
856         }
857
858         /* If we can't advertise it is OK. */
859         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
860
861         xen_blkbk_discard(xbt, be);
862
863         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
864
865         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
866         if (err) {
867                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
868                                  dev->nodename);
869                 goto abort;
870         }
871
872         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
873                             (unsigned long long)vbd_sz(&be->blkif->vbd));
874         if (err) {
875                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
876                                  dev->nodename);
877                 goto abort;
878         }
879
880         /* FIXME: use a typename instead */
881         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
882                             be->blkif->vbd.type |
883                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
884         if (err) {
885                 xenbus_dev_fatal(dev, err, "writing %s/info",
886                                  dev->nodename);
887                 goto abort;
888         }
889         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
890                             (unsigned long)
891                             bdev_logical_block_size(be->blkif->vbd.bdev));
892         if (err) {
893                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
894                                  dev->nodename);
895                 goto abort;
896         }
897         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
898                             bdev_physical_block_size(be->blkif->vbd.bdev));
899         if (err)
900                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
901                                  dev->nodename);
902
903         err = xenbus_transaction_end(xbt, 0);
904         if (err == -EAGAIN)
905                 goto again;
906         if (err)
907                 xenbus_dev_fatal(dev, err, "ending transaction");
908
909         err = xenbus_switch_state(dev, XenbusStateConnected);
910         if (err)
911                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
912                                  dev->nodename);
913
914         return;
915  abort:
916         xenbus_transaction_end(xbt, 1);
917 }
918
919 /*
920  * Each ring may have multi pages, depends on "ring-page-order".
921  */
922 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
923 {
924         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
925         struct pending_req *req, *n;
926         int err, i, j;
927         struct xen_blkif *blkif = ring->blkif;
928         struct xenbus_device *dev = blkif->be->dev;
929         unsigned int ring_page_order, nr_grefs, evtchn;
930
931         err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
932                           &evtchn);
933         if (err != 1) {
934                 err = -EINVAL;
935                 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
936                 return err;
937         }
938
939         err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
940                           &ring_page_order);
941         if (err != 1) {
942                 err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u", &ring_ref[0]);
943                 if (err != 1) {
944                         err = -EINVAL;
945                         xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
946                         return err;
947                 }
948                 nr_grefs = 1;
949         } else {
950                 unsigned int i;
951
952                 if (ring_page_order > xen_blkif_max_ring_order) {
953                         err = -EINVAL;
954                         xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
955                                          dir, ring_page_order,
956                                          xen_blkif_max_ring_order);
957                         return err;
958                 }
959
960                 nr_grefs = 1 << ring_page_order;
961                 for (i = 0; i < nr_grefs; i++) {
962                         char ring_ref_name[RINGREF_NAME_LEN];
963
964                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
965                         err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
966                                            "%u", &ring_ref[i]);
967                         if (err != 1) {
968                                 err = -EINVAL;
969                                 xenbus_dev_fatal(dev, err, "reading %s/%s",
970                                                  dir, ring_ref_name);
971                                 return err;
972                         }
973                 }
974         }
975         blkif->nr_ring_pages = nr_grefs;
976
977         for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
978                 req = kzalloc(sizeof(*req), GFP_KERNEL);
979                 if (!req)
980                         goto fail;
981                 list_add_tail(&req->free_list, &ring->pending_free);
982                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
983                         req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
984                         if (!req->segments[j])
985                                 goto fail;
986                 }
987                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
988                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
989                                                          GFP_KERNEL);
990                         if (!req->indirect_pages[j])
991                                 goto fail;
992                 }
993         }
994
995         /* Map the shared frame, irq etc. */
996         err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
997         if (err) {
998                 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
999                 return err;
1000         }
1001
1002         return 0;
1003
1004 fail:
1005         list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1006                 list_del(&req->free_list);
1007                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1008                         if (!req->segments[j])
1009                                 break;
1010                         kfree(req->segments[j]);
1011                 }
1012                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1013                         if (!req->indirect_pages[j])
1014                                 break;
1015                         kfree(req->indirect_pages[j]);
1016                 }
1017                 kfree(req);
1018         }
1019         return -ENOMEM;
1020
1021 }
1022
1023 static int connect_ring(struct backend_info *be)
1024 {
1025         struct xenbus_device *dev = be->dev;
1026         unsigned int pers_grants;
1027         char protocol[64] = "";
1028         int err, i;
1029         char *xspath;
1030         size_t xspathsize;
1031         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1032         unsigned int requested_num_queues = 0;
1033
1034         pr_debug("%s %s\n", __func__, dev->otherend);
1035
1036         be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1037         err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1038                            "%63s", protocol);
1039         if (err <= 0)
1040                 strcpy(protocol, "unspecified, assuming default");
1041         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1042                 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1043         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1044                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1045         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1046                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1047         else {
1048                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1049                 return -ENOSYS;
1050         }
1051         pers_grants = xenbus_read_unsigned(dev->otherend, "feature-persistent",
1052                                            0);
1053         be->blkif->vbd.feature_gnt_persistent = pers_grants;
1054         be->blkif->vbd.overflow_max_grants = 0;
1055
1056         /*
1057          * Read the number of hardware queues from frontend.
1058          */
1059         requested_num_queues = xenbus_read_unsigned(dev->otherend,
1060                                                     "multi-queue-num-queues",
1061                                                     1);
1062         if (requested_num_queues > xenblk_max_queues
1063             || requested_num_queues == 0) {
1064                 /* Buggy or malicious guest. */
1065                 xenbus_dev_fatal(dev, err,
1066                                 "guest requested %u queues, exceeding the maximum of %u.",
1067                                 requested_num_queues, xenblk_max_queues);
1068                 return -ENOSYS;
1069         }
1070         be->blkif->nr_rings = requested_num_queues;
1071         if (xen_blkif_alloc_rings(be->blkif))
1072                 return -ENOMEM;
1073
1074         pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1075                  be->blkif->nr_rings, be->blkif->blk_protocol, protocol,
1076                  pers_grants ? "persistent grants" : "");
1077
1078         if (be->blkif->nr_rings == 1)
1079                 return read_per_ring_refs(&be->blkif->rings[0], dev->otherend);
1080         else {
1081                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1082                 xspath = kmalloc(xspathsize, GFP_KERNEL);
1083                 if (!xspath) {
1084                         xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1085                         return -ENOMEM;
1086                 }
1087
1088                 for (i = 0; i < be->blkif->nr_rings; i++) {
1089                         memset(xspath, 0, xspathsize);
1090                         snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1091                         err = read_per_ring_refs(&be->blkif->rings[i], xspath);
1092                         if (err) {
1093                                 kfree(xspath);
1094                                 return err;
1095                         }
1096                 }
1097                 kfree(xspath);
1098         }
1099         return 0;
1100 }
1101
1102 static const struct xenbus_device_id xen_blkbk_ids[] = {
1103         { "vbd" },
1104         { "" }
1105 };
1106
1107 static struct xenbus_driver xen_blkbk_driver = {
1108         .ids  = xen_blkbk_ids,
1109         .probe = xen_blkbk_probe,
1110         .remove = xen_blkbk_remove,
1111         .otherend_changed = frontend_changed
1112 };
1113
1114 int xen_blkif_xenbus_init(void)
1115 {
1116         return xenbus_register_backend(&xen_blkbk_driver);
1117 }