]> git.ipfire.org Git - thirdparty/linux.git/blob - drivers/block/xen-blkback/xenbus.c
projects / thirdparty / linux.git / blob
? search:


e34219ea2b058c47d7464be4ce64d18c81d0e3f2
[thirdparty/linux.git] / drivers / block / xen-blkback / xenbus.c
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 <linux/module.h>
12 #include <linux/kthread.h>
13 #include <linux/pagemap.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_handle)
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 = sync_blockdev(blkif->vbd.bdev_handle->bdev);
103 if (err) {
104 xenbus_dev_error(blkif->be->dev, err, "block flush");
105 return;
106 }
107 invalidate_inode_pages2(
108 blkif->vbd.bdev_handle->bdev->bd_inode->i_mapping);
109
110 for (i = 0; i < blkif->nr_rings; i++) {
111 ring = &blkif->rings[i];
112 ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
113 if (IS_ERR(ring->xenblkd)) {
114 err = PTR_ERR(ring->xenblkd);
115 ring->xenblkd = NULL;
116 xenbus_dev_fatal(blkif->be->dev, err,
117 "start %s-%d xenblkd", name, i);
118 goto out;
119 }
120 }
121 return;
122
123 out:
124 while (--i >= 0) {
125 ring = &blkif->rings[i];
126 kthread_stop(ring->xenblkd);
127 }
128 return;
129 }
130
131 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
132 {
133 unsigned int r;
134
135 blkif->rings = kcalloc(blkif->nr_rings, sizeof(struct xen_blkif_ring),
136 GFP_KERNEL);
137 if (!blkif->rings)
138 return -ENOMEM;
139
140 for (r = 0; r < blkif->nr_rings; r++) {
141 struct xen_blkif_ring *ring = &blkif->rings[r];
142
143 spin_lock_init(&ring->blk_ring_lock);
144 init_waitqueue_head(&ring->wq);
145 INIT_LIST_HEAD(&ring->pending_free);
146 INIT_LIST_HEAD(&ring->persistent_purge_list);
147 INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
148 gnttab_page_cache_init(&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 /* Enable the persistent grants feature. */
162 static bool feature_persistent = true;
163 module_param(feature_persistent, bool, 0644);
164 MODULE_PARM_DESC(feature_persistent, "Enables the persistent grants feature");
165
166 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
167 {
168 struct xen_blkif *blkif;
169
170 BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
171
172 blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
173 if (!blkif)
174 return ERR_PTR(-ENOMEM);
175
176 blkif->domid = domid;
177 atomic_set(&blkif->refcnt, 1);
178 init_completion(&blkif->drain_complete);
179
180 /*
181 * Because freeing back to the cache may be deferred, it is not
182 * safe to unload the module (and hence destroy the cache) until
183 * this has completed. To prevent premature unloading, take an
184 * extra module reference here and release only when the object
185 * has been freed back to the cache.
186 */
187 __module_get(THIS_MODULE);
188 INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
189
190 return blkif;
191 }
192
193 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
194 unsigned int nr_grefs, unsigned int evtchn)
195 {
196 int err;
197 struct xen_blkif *blkif = ring->blkif;
198 const struct blkif_common_sring *sring_common;
199 RING_IDX rsp_prod, req_prod;
200 unsigned int size;
201
202 /* Already connected through? */
203 if (ring->irq)
204 return 0;
205
206 err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
207 &ring->blk_ring);
208 if (err < 0)
209 return err;
210
211 sring_common = (struct blkif_common_sring *)ring->blk_ring;
212 rsp_prod = READ_ONCE(sring_common->rsp_prod);
213 req_prod = READ_ONCE(sring_common->req_prod);
214
215 switch (blkif->blk_protocol) {
216 case BLKIF_PROTOCOL_NATIVE:
217 {
218 struct blkif_sring *sring_native =
219 (struct blkif_sring *)ring->blk_ring;
220
221 BACK_RING_ATTACH(&ring->blk_rings.native, sring_native,
222 rsp_prod, XEN_PAGE_SIZE * nr_grefs);
223 size = __RING_SIZE(sring_native, XEN_PAGE_SIZE * nr_grefs);
224 break;
225 }
226 case BLKIF_PROTOCOL_X86_32:
227 {
228 struct blkif_x86_32_sring *sring_x86_32 =
229 (struct blkif_x86_32_sring *)ring->blk_ring;
230
231 BACK_RING_ATTACH(&ring->blk_rings.x86_32, sring_x86_32,
232 rsp_prod, XEN_PAGE_SIZE * nr_grefs);
233 size = __RING_SIZE(sring_x86_32, XEN_PAGE_SIZE * nr_grefs);
234 break;
235 }
236 case BLKIF_PROTOCOL_X86_64:
237 {
238 struct blkif_x86_64_sring *sring_x86_64 =
239 (struct blkif_x86_64_sring *)ring->blk_ring;
240
241 BACK_RING_ATTACH(&ring->blk_rings.x86_64, sring_x86_64,
242 rsp_prod, XEN_PAGE_SIZE * nr_grefs);
243 size = __RING_SIZE(sring_x86_64, XEN_PAGE_SIZE * nr_grefs);
244 break;
245 }
246 default:
247 BUG();
248 }
249
250 err = -EIO;
251 if (req_prod - rsp_prod > size)
252 goto fail;
253
254 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(blkif->be->dev,
255 evtchn, xen_blkif_be_int, 0, "blkif-backend", ring);
256 if (err < 0)
257 goto fail;
258 ring->irq = err;
259
260 return 0;
261
262 fail:
263 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
264 ring->blk_rings.common.sring = NULL;
265 return err;
266 }
267
268 static int xen_blkif_disconnect(struct xen_blkif *blkif)
269 {
270 struct pending_req *req, *n;
271 unsigned int j, r;
272 bool busy = false;
273
274 for (r = 0; r < blkif->nr_rings; r++) {
275 struct xen_blkif_ring *ring = &blkif->rings[r];
276 unsigned int i = 0;
277
278 if (!ring->active)
279 continue;
280
281 if (ring->xenblkd) {
282 kthread_stop(ring->xenblkd);
283 ring->xenblkd = NULL;
284 wake_up(&ring->shutdown_wq);
285 }
286
287 /* The above kthread_stop() guarantees that at this point we
288 * don't have any discard_io or other_io requests. So, checking
289 * for inflight IO is enough.
290 */
291 if (atomic_read(&ring->inflight) > 0) {
292 busy = true;
293 continue;
294 }
295
296 if (ring->irq) {
297 unbind_from_irqhandler(ring->irq, ring);
298 ring->irq = 0;
299 }
300
301 if (ring->blk_rings.common.sring) {
302 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
303 ring->blk_rings.common.sring = NULL;
304 }
305
306 /* Remove all persistent grants and the cache of ballooned pages. */
307 xen_blkbk_free_caches(ring);
308
309 /* Check that there is no request in use */
310 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
311 list_del(&req->free_list);
312
313 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
314 kfree(req->segments[j]);
315
316 for (j = 0; j < MAX_INDIRECT_PAGES; j++)
317 kfree(req->indirect_pages[j]);
318
319 kfree(req);
320 i++;
321 }
322
323 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
324 BUG_ON(!list_empty(&ring->persistent_purge_list));
325 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
326 BUG_ON(ring->free_pages.num_pages != 0);
327 BUG_ON(ring->persistent_gnt_c != 0);
328 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
329 ring->active = false;
330 }
331 if (busy)
332 return -EBUSY;
333
334 blkif->nr_ring_pages = 0;
335 /*
336 * blkif->rings was allocated in connect_ring, so we should free it in
337 * here.
338 */
339 kfree(blkif->rings);
340 blkif->rings = NULL;
341 blkif->nr_rings = 0;
342
343 return 0;
344 }
345
346 static void xen_blkif_free(struct xen_blkif *blkif)
347 {
348 WARN_ON(xen_blkif_disconnect(blkif));
349 xen_vbd_free(&blkif->vbd);
350 kfree(blkif->be->mode);
351 kfree(blkif->be);
352
353 /* Make sure everything is drained before shutting down */
354 kmem_cache_free(xen_blkif_cachep, blkif);
355 module_put(THIS_MODULE);
356 }
357
358 int __init xen_blkif_interface_init(void)
359 {
360 xen_blkif_cachep = kmem_cache_create("blkif_cache",
361 sizeof(struct xen_blkif),
362 0, 0, NULL);
363 if (!xen_blkif_cachep)
364 return -ENOMEM;
365
366 return 0;
367 }
368
369 void xen_blkif_interface_fini(void)
370 {
371 kmem_cache_destroy(xen_blkif_cachep);
372 xen_blkif_cachep = NULL;
373 }
374
375 /*
376 * sysfs interface for VBD I/O requests
377 */
378
379 #define VBD_SHOW_ALLRING(name, format) \
380 static ssize_t show_##name(struct device *_dev, \
381 struct device_attribute *attr, \
382 char *buf) \
383 { \
384 struct xenbus_device *dev = to_xenbus_device(_dev); \
385 struct backend_info *be = dev_get_drvdata(&dev->dev); \
386 struct xen_blkif *blkif = be->blkif; \
387 unsigned int i; \
388 unsigned long long result = 0; \
389 \
390 if (!blkif->rings) \
391 goto out; \
392 \
393 for (i = 0; i < blkif->nr_rings; i++) { \
394 struct xen_blkif_ring *ring = &blkif->rings[i]; \
395 \
396 result += ring->st_##name; \
397 } \
398 \
399 out: \
400 return sprintf(buf, format, result); \
401 } \
402 static DEVICE_ATTR(name, 0444, show_##name, NULL)
403
404 VBD_SHOW_ALLRING(oo_req, "%llu\n");
405 VBD_SHOW_ALLRING(rd_req, "%llu\n");
406 VBD_SHOW_ALLRING(wr_req, "%llu\n");
407 VBD_SHOW_ALLRING(f_req, "%llu\n");
408 VBD_SHOW_ALLRING(ds_req, "%llu\n");
409 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
410 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
411
412 static struct attribute *xen_vbdstat_attrs[] = {
413 &dev_attr_oo_req.attr,
414 &dev_attr_rd_req.attr,
415 &dev_attr_wr_req.attr,
416 &dev_attr_f_req.attr,
417 &dev_attr_ds_req.attr,
418 &dev_attr_rd_sect.attr,
419 &dev_attr_wr_sect.attr,
420 NULL
421 };
422
423 static const struct attribute_group xen_vbdstat_group = {
424 .name = "statistics",
425 .attrs = xen_vbdstat_attrs,
426 };
427
428 #define VBD_SHOW(name, format, args...) \
429 static ssize_t show_##name(struct device *_dev, \
430 struct device_attribute *attr, \
431 char *buf) \
432 { \
433 struct xenbus_device *dev = to_xenbus_device(_dev); \
434 struct backend_info *be = dev_get_drvdata(&dev->dev); \
435 \
436 return sprintf(buf, format, ##args); \
437 } \
438 static DEVICE_ATTR(name, 0444, show_##name, NULL)
439
440 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
441 VBD_SHOW(mode, "%s\n", be->mode);
442
443 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
444 {
445 int error;
446
447 error = device_create_file(&dev->dev, &dev_attr_physical_device);
448 if (error)
449 goto fail1;
450
451 error = device_create_file(&dev->dev, &dev_attr_mode);
452 if (error)
453 goto fail2;
454
455 error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
456 if (error)
457 goto fail3;
458
459 return 0;
460
461 fail3: sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
462 fail2: device_remove_file(&dev->dev, &dev_attr_mode);
463 fail1: device_remove_file(&dev->dev, &dev_attr_physical_device);
464 return error;
465 }
466
467 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
468 {
469 sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
470 device_remove_file(&dev->dev, &dev_attr_mode);
471 device_remove_file(&dev->dev, &dev_attr_physical_device);
472 }
473
474 static void xen_vbd_free(struct xen_vbd *vbd)
475 {
476 if (vbd->bdev_handle)
477 bdev_release(vbd->bdev_handle);
478 vbd->bdev_handle = NULL;
479 }
480
481 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
482 unsigned major, unsigned minor, int readonly,
483 int cdrom)
484 {
485 struct xen_vbd *vbd;
486 struct bdev_handle *bdev_handle;
487
488 vbd = &blkif->vbd;
489 vbd->handle = handle;
490 vbd->readonly = readonly;
491 vbd->type = 0;
492
493 vbd->pdevice = MKDEV(major, minor);
494
495 bdev_handle = bdev_open_by_dev(vbd->pdevice, vbd->readonly ?
496 BLK_OPEN_READ : BLK_OPEN_WRITE, NULL, NULL);
497
498 if (IS_ERR(bdev_handle)) {
499 pr_warn("xen_vbd_create: device %08x could not be opened\n",
500 vbd->pdevice);
501 return -ENOENT;
502 }
503
504 vbd->bdev_handle = bdev_handle;
505 if (vbd->bdev_handle->bdev->bd_disk == NULL) {
506 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
507 vbd->pdevice);
508 xen_vbd_free(vbd);
509 return -ENOENT;
510 }
511 vbd->size = vbd_sz(vbd);
512
513 if (cdrom || disk_to_cdi(vbd->bdev_handle->bdev->bd_disk))
514 vbd->type |= VDISK_CDROM;
515 if (vbd->bdev_handle->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
516 vbd->type |= VDISK_REMOVABLE;
517
518 if (bdev_write_cache(bdev_handle->bdev))
519 vbd->flush_support = true;
520 if (bdev_max_secure_erase_sectors(bdev_handle->bdev))
521 vbd->discard_secure = true;
522
523 pr_debug("Successful creation of handle=%04x (dom=%u)\n",
524 handle, blkif->domid);
525 return 0;
526 }
527
528 static void xen_blkbk_remove(struct xenbus_device *dev)
529 {
530 struct backend_info *be = dev_get_drvdata(&dev->dev);
531
532 pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
533
534 if (be->major || be->minor)
535 xenvbd_sysfs_delif(dev);
536
537 if (be->backend_watch.node) {
538 unregister_xenbus_watch(&be->backend_watch);
539 kfree(be->backend_watch.node);
540 be->backend_watch.node = NULL;
541 }
542
543 dev_set_drvdata(&dev->dev, NULL);
544
545 if (be->blkif) {
546 xen_blkif_disconnect(be->blkif);
547
548 /* Put the reference we set in xen_blkif_alloc(). */
549 xen_blkif_put(be->blkif);
550 }
551 }
552
553 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
554 struct backend_info *be, int state)
555 {
556 struct xenbus_device *dev = be->dev;
557 int err;
558
559 err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
560 "%d", state);
561 if (err)
562 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
563
564 return err;
565 }
566
567 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
568 {
569 struct xenbus_device *dev = be->dev;
570 struct xen_blkif *blkif = be->blkif;
571 int err;
572 int state = 0;
573 struct block_device *bdev = be->blkif->vbd.bdev_handle->bdev;
574
575 if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
576 return;
577
578 if (bdev_max_discard_sectors(bdev)) {
579 err = xenbus_printf(xbt, dev->nodename,
580 "discard-granularity", "%u",
581 bdev_discard_granularity(bdev));
582 if (err) {
583 dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
584 return;
585 }
586 err = xenbus_printf(xbt, dev->nodename,
587 "discard-alignment", "%u",
588 bdev_discard_alignment(bdev));
589 if (err) {
590 dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
591 return;
592 }
593 state = 1;
594 /* Optional. */
595 err = xenbus_printf(xbt, dev->nodename,
596 "discard-secure", "%d",
597 blkif->vbd.discard_secure);
598 if (err) {
599 dev_warn(&dev->dev, "writing discard-secure (%d)", err);
600 return;
601 }
602 }
603 err = xenbus_printf(xbt, dev->nodename, "feature-discard",
604 "%d", state);
605 if (err)
606 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
607 }
608
609 int xen_blkbk_barrier(struct xenbus_transaction xbt,
610 struct backend_info *be, int state)
611 {
612 struct xenbus_device *dev = be->dev;
613 int err;
614
615 err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
616 "%d", state);
617 if (err)
618 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
619
620 return err;
621 }
622
623 /*
624 * Entry point to this code when a new device is created. Allocate the basic
625 * structures, and watch the store waiting for the hotplug scripts to tell us
626 * the device's physical major and minor numbers. Switch to InitWait.
627 */
628 static int xen_blkbk_probe(struct xenbus_device *dev,
629 const struct xenbus_device_id *id)
630 {
631 int err;
632 struct backend_info *be = kzalloc(sizeof(struct backend_info),
633 GFP_KERNEL);
634
635 /* match the pr_debug in xen_blkbk_remove */
636 pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
637
638 if (!be) {
639 xenbus_dev_fatal(dev, -ENOMEM,
640 "allocating backend structure");
641 return -ENOMEM;
642 }
643 be->dev = dev;
644 dev_set_drvdata(&dev->dev, be);
645
646 be->blkif = xen_blkif_alloc(dev->otherend_id);
647 if (IS_ERR(be->blkif)) {
648 err = PTR_ERR(be->blkif);
649 be->blkif = NULL;
650 xenbus_dev_fatal(dev, err, "creating block interface");
651 goto fail;
652 }
653
654 err = xenbus_printf(XBT_NIL, dev->nodename,
655 "feature-max-indirect-segments", "%u",
656 MAX_INDIRECT_SEGMENTS);
657 if (err)
658 dev_warn(&dev->dev,
659 "writing %s/feature-max-indirect-segments (%d)",
660 dev->nodename, err);
661
662 /* Multi-queue: advertise how many queues are supported by us.*/
663 err = xenbus_printf(XBT_NIL, dev->nodename,
664 "multi-queue-max-queues", "%u", xenblk_max_queues);
665 if (err)
666 pr_warn("Error writing multi-queue-max-queues\n");
667
668 /* setup back pointer */
669 be->blkif->be = be;
670
671 err = xenbus_watch_pathfmt(dev, &be->backend_watch, NULL,
672 backend_changed,
673 "%s/%s", dev->nodename, "physical-device");
674 if (err)
675 goto fail;
676
677 err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
678 xen_blkif_max_ring_order);
679 if (err)
680 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
681
682 err = xenbus_switch_state(dev, XenbusStateInitWait);
683 if (err)
684 goto fail;
685
686 return 0;
687
688 fail:
689 pr_warn("%s failed\n", __func__);
690 xen_blkbk_remove(dev);
691 return err;
692 }
693
694 /*
695 * Callback received when the hotplug scripts have placed the physical-device
696 * node. Read it and the mode node, and create a vbd. If the frontend is
697 * ready, connect.
698 */
699 static void backend_changed(struct xenbus_watch *watch,
700 const char *path, const char *token)
701 {
702 int err;
703 unsigned major;
704 unsigned minor;
705 struct backend_info *be
706 = container_of(watch, struct backend_info, backend_watch);
707 struct xenbus_device *dev = be->dev;
708 int cdrom = 0;
709 unsigned long handle;
710 char *device_type;
711
712 pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
713
714 err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
715 &major, &minor);
716 if (XENBUS_EXIST_ERR(err)) {
717 /*
718 * Since this watch will fire once immediately after it is
719 * registered, we expect this. Ignore it, and wait for the
720 * hotplug scripts.
721 */
722 return;
723 }
724 if (err != 2) {
725 xenbus_dev_fatal(dev, err, "reading physical-device");
726 return;
727 }
728
729 if (be->major | be->minor) {
730 if (be->major != major || be->minor != minor)
731 pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
732 be->major, be->minor, major, minor);
733 return;
734 }
735
736 be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
737 if (IS_ERR(be->mode)) {
738 err = PTR_ERR(be->mode);
739 be->mode = NULL;
740 xenbus_dev_fatal(dev, err, "reading mode");
741 return;
742 }
743
744 device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
745 if (!IS_ERR(device_type)) {
746 cdrom = strcmp(device_type, "cdrom") == 0;
747 kfree(device_type);
748 }
749
750 /* Front end dir is a number, which is used as the handle. */
751 err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
752 if (err) {
753 kfree(be->mode);
754 be->mode = NULL;
755 return;
756 }
757
758 be->major = major;
759 be->minor = minor;
760
761 err = xen_vbd_create(be->blkif, handle, major, minor,
762 !strchr(be->mode, 'w'), cdrom);
763
764 if (err)
765 xenbus_dev_fatal(dev, err, "creating vbd structure");
766 else {
767 err = xenvbd_sysfs_addif(dev);
768 if (err) {
769 xen_vbd_free(&be->blkif->vbd);
770 xenbus_dev_fatal(dev, err, "creating sysfs entries");
771 }
772 }
773
774 if (err) {
775 kfree(be->mode);
776 be->mode = NULL;
777 be->major = 0;
778 be->minor = 0;
779 } else {
780 /* We're potentially connected now */
781 xen_update_blkif_status(be->blkif);
782 }
783 }
784
785 /*
786 * Callback received when the frontend's state changes.
787 */
788 static void frontend_changed(struct xenbus_device *dev,
789 enum xenbus_state frontend_state)
790 {
791 struct backend_info *be = dev_get_drvdata(&dev->dev);
792 int err;
793
794 pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
795
796 switch (frontend_state) {
797 case XenbusStateInitialising:
798 if (dev->state == XenbusStateClosed) {
799 pr_info("%s: prepare for reconnect\n", dev->nodename);
800 xenbus_switch_state(dev, XenbusStateInitWait);
801 }
802 break;
803
804 case XenbusStateInitialised:
805 case XenbusStateConnected:
806 /*
807 * Ensure we connect even when two watches fire in
808 * close succession and we miss the intermediate value
809 * of frontend_state.
810 */
811 if (dev->state == XenbusStateConnected)
812 break;
813
814 /*
815 * Enforce precondition before potential leak point.
816 * xen_blkif_disconnect() is idempotent.
817 */
818 err = xen_blkif_disconnect(be->blkif);
819 if (err) {
820 xenbus_dev_fatal(dev, err, "pending I/O");
821 break;
822 }
823
824 err = connect_ring(be);
825 if (err) {
826 /*
827 * Clean up so that memory resources can be used by
828 * other devices. connect_ring reported already error.
829 */
830 xen_blkif_disconnect(be->blkif);
831 break;
832 }
833 xen_update_blkif_status(be->blkif);
834 break;
835
836 case XenbusStateClosing:
837 xenbus_switch_state(dev, XenbusStateClosing);
838 break;
839
840 case XenbusStateClosed:
841 xen_blkif_disconnect(be->blkif);
842 xenbus_switch_state(dev, XenbusStateClosed);
843 if (xenbus_dev_is_online(dev))
844 break;
845 fallthrough;
846 /* if not online */
847 case XenbusStateUnknown:
848 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
849 device_unregister(&dev->dev);
850 break;
851
852 default:
853 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
854 frontend_state);
855 break;
856 }
857 }
858
859 /* Once a memory pressure is detected, squeeze free page pools for a while. */
860 static unsigned int buffer_squeeze_duration_ms = 10;
861 module_param_named(buffer_squeeze_duration_ms,
862 buffer_squeeze_duration_ms, int, 0644);
863 MODULE_PARM_DESC(buffer_squeeze_duration_ms,
864 "Duration in ms to squeeze pages buffer when a memory pressure is detected");
865
866 /*
867 * Callback received when the memory pressure is detected.
868 */
869 static void reclaim_memory(struct xenbus_device *dev)
870 {
871 struct backend_info *be = dev_get_drvdata(&dev->dev);
872
873 if (!be)
874 return;
875 be->blkif->buffer_squeeze_end = jiffies +
876 msecs_to_jiffies(buffer_squeeze_duration_ms);
877 }
878
879 /* ** Connection ** */
880
881 /*
882 * Write the physical details regarding the block device to the store, and
883 * switch to Connected state.
884 */
885 static void connect(struct backend_info *be)
886 {
887 struct xenbus_transaction xbt;
888 int err;
889 struct xenbus_device *dev = be->dev;
890
891 pr_debug("%s %s\n", __func__, dev->otherend);
892
893 /* Supply the information about the device the frontend needs */
894 again:
895 err = xenbus_transaction_start(&xbt);
896 if (err) {
897 xenbus_dev_fatal(dev, err, "starting transaction");
898 return;
899 }
900
901 /* If we can't advertise it is OK. */
902 xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
903
904 xen_blkbk_discard(xbt, be);
905
906 xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
907
908 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
909 be->blkif->vbd.feature_gnt_persistent_parm);
910 if (err) {
911 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
912 dev->nodename);
913 goto abort;
914 }
915
916 err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
917 (unsigned long long)vbd_sz(&be->blkif->vbd));
918 if (err) {
919 xenbus_dev_fatal(dev, err, "writing %s/sectors",
920 dev->nodename);
921 goto abort;
922 }
923
924 /* FIXME: use a typename instead */
925 err = xenbus_printf(xbt, dev->nodename, "info", "%u",
926 be->blkif->vbd.type |
927 (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
928 if (err) {
929 xenbus_dev_fatal(dev, err, "writing %s/info",
930 dev->nodename);
931 goto abort;
932 }
933 err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
934 (unsigned long)bdev_logical_block_size(
935 be->blkif->vbd.bdev_handle->bdev));
936 if (err) {
937 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
938 dev->nodename);
939 goto abort;
940 }
941 err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
942 bdev_physical_block_size(
943 be->blkif->vbd.bdev_handle->bdev));
944 if (err)
945 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
946 dev->nodename);
947
948 err = xenbus_transaction_end(xbt, 0);
949 if (err == -EAGAIN)
950 goto again;
951 if (err)
952 xenbus_dev_fatal(dev, err, "ending transaction");
953
954 err = xenbus_switch_state(dev, XenbusStateConnected);
955 if (err)
956 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
957 dev->nodename);
958
959 return;
960 abort:
961 xenbus_transaction_end(xbt, 1);
962 }
963
964 /*
965 * Each ring may have multi pages, depends on "ring-page-order".
966 */
967 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
968 {
969 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
970 struct pending_req *req, *n;
971 int err, i, j;
972 struct xen_blkif *blkif = ring->blkif;
973 struct xenbus_device *dev = blkif->be->dev;
974 unsigned int nr_grefs, evtchn;
975
976 err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
977 &evtchn);
978 if (err != 1) {
979 err = -EINVAL;
980 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
981 return err;
982 }
983
984 nr_grefs = blkif->nr_ring_pages;
985
986 if (unlikely(!nr_grefs)) {
987 WARN_ON(true);
988 return -EINVAL;
989 }
990
991 for (i = 0; i < nr_grefs; i++) {
992 char ring_ref_name[RINGREF_NAME_LEN];
993
994 if (blkif->multi_ref)
995 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
996 else {
997 WARN_ON(i != 0);
998 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref");
999 }
1000
1001 err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
1002 "%u", &ring_ref[i]);
1003
1004 if (err != 1) {
1005 err = -EINVAL;
1006 xenbus_dev_fatal(dev, err, "reading %s/%s",
1007 dir, ring_ref_name);
1008 return err;
1009 }
1010 }
1011
1012 err = -ENOMEM;
1013 for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
1014 req = kzalloc(sizeof(*req), GFP_KERNEL);
1015 if (!req)
1016 goto fail;
1017 list_add_tail(&req->free_list, &ring->pending_free);
1018 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1019 req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
1020 if (!req->segments[j])
1021 goto fail;
1022 }
1023 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1024 req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
1025 GFP_KERNEL);
1026 if (!req->indirect_pages[j])
1027 goto fail;
1028 }
1029 }
1030
1031 /* Map the shared frame, irq etc. */
1032 err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
1033 if (err) {
1034 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
1035 goto fail;
1036 }
1037
1038 return 0;
1039
1040 fail:
1041 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1042 list_del(&req->free_list);
1043 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1044 if (!req->segments[j])
1045 break;
1046 kfree(req->segments[j]);
1047 }
1048 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1049 if (!req->indirect_pages[j])
1050 break;
1051 kfree(req->indirect_pages[j]);
1052 }
1053 kfree(req);
1054 }
1055 return err;
1056 }
1057
1058 static int connect_ring(struct backend_info *be)
1059 {
1060 struct xenbus_device *dev = be->dev;
1061 struct xen_blkif *blkif = be->blkif;
1062 char protocol[64] = "";
1063 int err, i;
1064 char *xspath;
1065 size_t xspathsize;
1066 const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1067 unsigned int requested_num_queues = 0;
1068 unsigned int ring_page_order;
1069
1070 pr_debug("%s %s\n", __func__, dev->otherend);
1071
1072 blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1073 err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1074 "%63s", protocol);
1075 if (err <= 0)
1076 strcpy(protocol, "unspecified, assuming default");
1077 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1078 blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1079 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1080 blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1081 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1082 blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1083 else {
1084 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1085 return -ENOSYS;
1086 }
1087
1088 blkif->vbd.feature_gnt_persistent_parm = feature_persistent;
1089 blkif->vbd.feature_gnt_persistent =
1090 blkif->vbd.feature_gnt_persistent_parm &&
1091 xenbus_read_unsigned(dev->otherend, "feature-persistent", 0);
1092
1093 blkif->vbd.overflow_max_grants = 0;
1094
1095 /*
1096 * Read the number of hardware queues from frontend.
1097 */
1098 requested_num_queues = xenbus_read_unsigned(dev->otherend,
1099 "multi-queue-num-queues",
1100 1);
1101 if (requested_num_queues > xenblk_max_queues
1102 || requested_num_queues == 0) {
1103 /* Buggy or malicious guest. */
1104 xenbus_dev_fatal(dev, err,
1105 "guest requested %u queues, exceeding the maximum of %u.",
1106 requested_num_queues, xenblk_max_queues);
1107 return -ENOSYS;
1108 }
1109 blkif->nr_rings = requested_num_queues;
1110 if (xen_blkif_alloc_rings(blkif))
1111 return -ENOMEM;
1112
1113 pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1114 blkif->nr_rings, blkif->blk_protocol, protocol,
1115 blkif->vbd.feature_gnt_persistent ? "persistent grants" : "");
1116
1117 err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
1118 &ring_page_order);
1119 if (err != 1) {
1120 blkif->nr_ring_pages = 1;
1121 blkif->multi_ref = false;
1122 } else if (ring_page_order <= xen_blkif_max_ring_order) {
1123 blkif->nr_ring_pages = 1 << ring_page_order;
1124 blkif->multi_ref = true;
1125 } else {
1126 err = -EINVAL;
1127 xenbus_dev_fatal(dev, err,
1128 "requested ring page order %d exceed max:%d",
1129 ring_page_order,
1130 xen_blkif_max_ring_order);
1131 return err;
1132 }
1133
1134 if (blkif->nr_rings == 1)
1135 return read_per_ring_refs(&blkif->rings[0], dev->otherend);
1136 else {
1137 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1138 xspath = kmalloc(xspathsize, GFP_KERNEL);
1139 if (!xspath) {
1140 xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1141 return -ENOMEM;
1142 }
1143
1144 for (i = 0; i < blkif->nr_rings; i++) {
1145 memset(xspath, 0, xspathsize);
1146 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1147 err = read_per_ring_refs(&blkif->rings[i], xspath);
1148 if (err) {
1149 kfree(xspath);
1150 return err;
1151 }
1152 }
1153 kfree(xspath);
1154 }
1155 return 0;
1156 }
1157
1158 static const struct xenbus_device_id xen_blkbk_ids[] = {
1159 { "vbd" },
1160 { "" }
1161 };
1162
1163 static struct xenbus_driver xen_blkbk_driver = {
1164 .ids = xen_blkbk_ids,
1165 .probe = xen_blkbk_probe,
1166 .remove = xen_blkbk_remove,
1167 .otherend_changed = frontend_changed,
1168 .allow_rebind = true,
1169 .reclaim_memory = reclaim_memory,
1170 };
1171
1172 int xen_blkif_xenbus_init(void)
1173 {
1174 return xenbus_register_backend(&xen_blkbk_driver);
1175 }
1176
1177 void xen_blkif_xenbus_fini(void)
1178 {
1179 xenbus_unregister_driver(&xen_blkbk_driver);
1180 }
A mirror of Linus' kernel repository